CN111352799A - Inspection method and device - Google Patents

Inspection method and device Download PDF

Info

Publication number
CN111352799A
CN111352799A CN202010103868.XA CN202010103868A CN111352799A CN 111352799 A CN111352799 A CN 111352799A CN 202010103868 A CN202010103868 A CN 202010103868A CN 111352799 A CN111352799 A CN 111352799A
Authority
CN
China
Prior art keywords
model
training
mobile
time slice
mobile device
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN202010103868.XA
Other languages
Chinese (zh)
Inventor
杨洁
何东杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Unionpay Co Ltd
Original Assignee
China Unionpay Co Ltd
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 China Unionpay Co Ltd filed Critical China Unionpay Co Ltd
Priority to CN202010103868.XA priority Critical patent/CN111352799A/en
Publication of CN111352799A publication Critical patent/CN111352799A/en
Priority to PCT/CN2020/137411 priority patent/WO2021164404A1/en
Priority to TW109147101A priority patent/TWI770749B/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3058—Monitoring arrangements for monitoring environmental properties or parameters of the computing system or of the computing system component, e.g. monitoring of power, currents, temperature, humidity, position, vibrations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3065—Monitoring arrangements determined by the means or processing involved in reporting the monitored data
    • G06F11/3072—Monitoring arrangements determined by the means or processing involved in reporting the monitored data where the reporting involves data filtering, e.g. pattern matching, time or event triggered, adaptive or policy-based reporting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G07—CHECKING-DEVICES
    • G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
    • G07C1/00—Registering, indicating or recording the time of events or elapsed time, e.g. time-recorders for work people
    • G07C1/20—Checking timed patrols, e.g. of watchman
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)
  • Testing Or Measuring Of Semiconductors Or The Like (AREA)
  • Transplanting Machines (AREA)
  • Indicating And Signalling Devices For Elevators (AREA)

Abstract

The invention discloses a polling method and a polling device, wherein the method comprises the following steps: the method comprises the steps that model training instructions are sent to mobile devices by the model management devices, the mobile devices collect training data when traveling in the machine room according to corresponding routing inspection routes according to the model training instructions, intermediate models are obtained through training based on local models and the training data and reported to the model management devices, the model management devices obtain recognition models through training based on model parameters of the intermediate models, and the recognition models are used for determining the health state of the machine room. The training process of the intermediate model is placed on the mobile equipment side to be executed, so that the mobile equipment can only report the model parameters of the intermediate model to the model management equipment, and does not need to report the whole amount of training data, thereby saving communication overhead and improving inspection efficiency.

Description

Inspection method and device
Technical Field
The invention relates to the technical field of image processing, in particular to a polling method and a polling device.
Background
An Internet Data Center (IDC) room is a standardized room environment established on the basis of Internet communication lines and bandwidth resources, and the IDC room can accommodate various types of devices, such as servers, monitoring devices, management devices, security devices, and the like. In actual operation, inspection of the IDC room is generally required, such as collecting signal lamp images of equipment to identify the state of signal lamps of the equipment, collecting equipment images to identify the type of the equipment, collecting odor data and/or temperature data to analyze the room environment. Therefore, when equipment in the IDC machine room is patrolled and examined, how to effectively identify the data of patrolling and examining is very important for maintaining the safety of the machine room and troubleshooting in time.
In a current implementation manner, each mobile device collects training data from a machine room and reports the training data to a server for centralized training, and after the server trains and obtains a recognition model, the recognition model is issued to each mobile device, so that each mobile device uses the recognition model to recognize the collected inspection data, and inspection of the machine room is completed. However, this method has the following problems: training data (such as images, smell or temperature) generally corresponds to a large data volume, and since the method needs each mobile device to report the training data to the server, there is a large communication overhead, resulting in low polling efficiency.
In summary, there is a need for a polling method to solve the technical problems of large communication overhead and low polling efficiency caused by the centralized training of the recognition model by the server in the prior art.
Disclosure of Invention
The invention provides a routing inspection method and a routing inspection device, which are used for solving the technical problems of high communication overhead and low routing inspection efficiency caused by the adoption of a server to intensively train an identification model in the prior art.
In a first aspect, the present invention provides a polling method, which is applied to a model management device, and the method includes:
sending a model training instruction to each mobile device; the model training instruction is used for acquiring training data when each mobile device travels in the machine room according to respective routing inspection routes, and an intermediate model is obtained based on a local model and the training data; further, model parameters of the intermediate model reported by each mobile device are received, and an identification model is obtained based on model parameter training of the intermediate model reported by each mobile device, wherein the identification model is used for determining the health state of the machine room.
In the invention, the training process of the intermediate model is carried out on the mobile equipment side, so that the mobile equipment can report the model parameters of the intermediate model to the model management equipment only without reporting the full amount of training data, and the model parameters have smaller data volume relative to the training data, therefore, the mode can save communication overhead and improve inspection efficiency; in addition, the health state of the machine room is determined by using the recognition model in the inspection process of the mobile equipment, so that the combined operation of model training and model recognition is realized, and the inspection efficiency can be improved.
In a possible implementation manner, the identification model is an identification model corresponding to a current time slice; in a specific implementation, before sending the model training instruction to each mobile device, the method further includes: and selecting the mobile equipment corresponding to the current time slice from all the mobile equipment in the machine room as each piece of mobile equipment.
In the implementation mode, the model training is decomposed into a plurality of time slices for execution, and part of the mobile equipment is selected from each time slice for executing the training, so that the training data and the mobile equipment can be effectively distributed and uniformly utilized, the problem that parameter iteration cannot be converged due to overlarge data during model training is avoided, and the accuracy of model identification is improved.
In a possible implementation manner, the model training instruction is used for acquiring training data when each mobile device circularly travels in a machine room according to respective routing inspection routes, training based on a local model and the training data to obtain an intermediate model and reporting the intermediate model; the receiving the model parameters of the intermediate model reported by each mobile device, and training based on the model parameters of the intermediate model reported by each mobile device to obtain the recognition model includes: each time a set number of model parameters are received, constructing and obtaining comprehensive model parameters at least based on the set number of model parameters, and sending the comprehensive model parameters to each mobile device; the comprehensive model parameters are used for updating local models by the mobile devices; and when the model corresponding to a certain comprehensive model parameter meets the end condition of the current time slice, sending a model training end instruction to each mobile device, wherein the model corresponding to the comprehensive model parameter is the identification model.
In the implementation mode, model comprehensive operations are executed for multiple times in each time slice, the model comprehensive operations use the set number of model parameters to obtain comprehensive model parameters each time, model parameters of all mobile devices are not used for synthesis, the comprehensive model parameters can be obtained by using as many model parameters as possible, the fault conditions of part of the mobile devices are compatible, and smooth model training is ensured.
In one possible implementation, the model management device is a server; after obtaining a recognition model based on the model parameter training of the intermediate model reported by each mobile device, further judging whether the recognition model meets the end condition of the model training, if not, selecting the mobile device corresponding to the next time slice from all the mobile devices of the machine room, and sending a model training instruction to the mobile device corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
In the implementation mode, the server is used as the model management device, so that the pressure of the mobile device can be effectively reduced while the model training process of each mobile device is uniformly managed, and the efficiency of training the middle model of the mobile device is improved; and by reselecting the mobile equipment for executing training when each time slice is started, the mobile equipment can be effectively distributed, the difference of different mobile equipment is fully considered, and the accuracy of the identification model is improved.
In one possible implementation, the model management device is a mobile device; after obtaining the recognition model based on the model parameter training of the intermediate model reported by each mobile device, judging whether the recognition model meets the end condition of the model training, if not, selecting a certain mobile device as a model management device corresponding to the next time slice through communication with other mobile devices, and sending a model starting instruction to the model management device corresponding to the next time slice; the model starting instruction is used for the model management equipment corresponding to the next time slice to select other mobile equipment corresponding to the next time slice from other mobile equipment and send a model training instruction to the other mobile equipment corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
In the implementation mode, the mobile equipment is used as the model management equipment, so that a server does not need to be additionally arranged, and the cost of model training can be reduced; and by reselecting the mobile equipment for executing training when each time slice is started, the mobile equipment can be effectively distributed, the difference of different mobile equipment is fully considered, and the accuracy of the identification model is improved.
In a second aspect, the present invention provides a polling method applied to a mobile device, where the method includes:
receiving a model training instruction sent by model management equipment, acquiring training data when a corresponding inspection route travels in a machine room according to the model training instruction, training on the basis of a local model and the training data to obtain an intermediate model corresponding to the mobile equipment, and reporting model parameters of the intermediate model to the model management equipment; the model parameters of the intermediate model reported by each mobile device are used for training the model management device to obtain an identification model, and the identification model is used for determining the health state of the machine room.
In a possible implementation manner, the acquiring training data when traveling in a machine room according to a routing inspection route according to the model training instruction, obtaining an intermediate model corresponding to the mobile device based on a local model and the training data training, and reporting model parameters of the intermediate model to the model management device includes: according to the model training instruction, circularly executing: acquiring training data when a routing inspection route advances in a machine room, training based on a local model and the training data to obtain an intermediate model corresponding to the mobile equipment, and reporting model parameters of the intermediate model to the model management equipment; in each cycle, after reporting the model parameters of the intermediate model to the model management device, the integrated model parameters sent by the model management device are also received, and the local model is updated by using the integrated model parameters.
In a possible implementation manner, after the local model is updated by using the comprehensive model parameters, data to be identified is collected when the inspection route travels in the machine room, and the updated local model is used for identifying the data to be identified so as to determine the health state of the machine room.
In the implementation mode, training, optimizing and identifying operations are jointly executed in the inspection process, so that the effects of training the model, optimizing the model and identifying the model can be achieved, the training and identifying efficiency is improved, and the identifying effect can be improved by identifying the model after real-time optimization.
In a third aspect, the present invention provides an inspection device, comprising:
the receiving and sending module is used for sending model training instructions to each mobile device; the model training instruction is used for acquiring training data when each mobile device travels in the machine room according to respective routing inspection routes, and an intermediate model is obtained based on a local model and the training data; receiving the model parameters of the intermediate model reported by each mobile device;
and the training module is used for training based on the model parameters of the intermediate model reported by each mobile device to obtain an identification model, and the identification model is used for determining the health state of the machine room.
In a possible implementation manner, the identification model is an identification model corresponding to a current time slice; before the transceiver module sends the model training instructions to each mobile device, the training module is further configured to: and selecting the mobile equipment corresponding to the current time slice from all the mobile equipment in the machine room as each piece of mobile equipment.
In a possible implementation manner, the model training instruction is used for acquiring training data when each mobile device circularly travels in a machine room according to respective routing inspection routes, training based on a local model and the training data to obtain an intermediate model and reporting the intermediate model; correspondingly, the training module is specifically configured to: every time a set number of model parameters are received, constructing at least based on the set number of model parameters to obtain a comprehensive model parameter;
the transceiver module is specifically configured to: sending the comprehensive model parameters to each mobile device; the comprehensive model parameters are used for updating local models by the mobile devices; and when the model corresponding to a certain comprehensive model parameter meets the end condition of the current time slice, sending a model training end instruction to each mobile device, wherein the model corresponding to the comprehensive model parameter is the identification model.
In one possible implementation, the apparatus is a server; after the training module trains and obtains a recognition model based on the model parameters of the intermediate model reported by each mobile device, the training module also judges whether the recognition model meets the end condition of the model training, if not, the training module selects the mobile device corresponding to the next time slice from all the mobile devices of the machine room and sends a model training instruction to the mobile device corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
In one possible implementation, the apparatus is a mobile device; after the training module trains and obtains a recognition model based on the model parameters of the intermediate model reported by each mobile device, the training module also judges whether the recognition model meets the end condition of the model training, if not, a certain mobile device is selected as a model management device corresponding to the next time slice through communication with other mobile devices, and a model starting instruction is sent to the model management device corresponding to the next time slice; the model starting instruction is used for the model management equipment corresponding to the next time slice to select other mobile equipment corresponding to the next time slice from other mobile equipment and send a model training instruction to the other mobile equipment corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
In a fourth aspect, the present invention provides an inspection device, the inspection device comprising:
the receiving and sending module is used for receiving a model training instruction sent by the model management equipment;
the training module is used for acquiring training data when the mobile equipment travels in a machine room according to the model training instruction and a corresponding routing inspection route, and training based on a local model and the training data to obtain an intermediate model corresponding to the mobile equipment;
the transceiver module is further configured to report the model parameters of the intermediate model to the model management device; the model parameters of the intermediate model reported by each mobile device are used for training the model management device to obtain an identification model, and the identification model is used for determining the health state of the machine room.
In a possible implementation manner, the training module is specifically configured to: according to the model training instruction, circularly executing: acquiring training data when the mobile equipment travels in a machine room according to a routing inspection route, and training based on a local model and the training data to obtain an intermediate model corresponding to the mobile equipment; correspondingly, the transceiver module is specifically configured to: according to the model training instruction, circularly executing: reporting the model parameters of the intermediate model to the model management equipment; wherein, after reporting the model parameters of the intermediate model to the model management device in each cycle, the transceiver module is further configured to: receiving the comprehensive model parameters sent by the model management equipment; the training module is further configured to: updating the local model using the integrated model parameters.
In one possible implementation, the apparatus further includes a recognition module, after the training module updates the local model using the integrated model parameters, the recognition module is configured to: and acquiring data to be identified when the routing inspection route travels in the machine room, and identifying the data to be identified by using the updated local model so as to determine the health state of the machine room.
In a fifth aspect, the present invention provides a computing device, comprising at least one processor and at least one memory, wherein the memory stores a computer program, and when the program is executed by the processor, the program causes the processor to execute the inspection method according to any of the first or second aspects.
In a sixth aspect, the present invention provides a computer-readable storage medium storing a computer program executable by a computing device, the program causing the computing device to perform the inspection method according to any one of the first or second aspects when the program runs on the computing device.
These and other implementations of the invention will be more readily understood from the following description of the embodiments.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
Fig. 1 is a schematic structural diagram of an IDC machine room according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a system architecture of an inspection system according to an embodiment of the present invention;
fig. 3 is a schematic flow chart corresponding to the inspection method according to the embodiment of the present invention;
FIG. 4 is a schematic view of an interaction flow corresponding to the inspection method when the server is used as the model management device;
FIG. 5 is a schematic diagram of an interaction flow corresponding to the inspection method when the mobile device is used as the model management device;
fig. 6 is a schematic structural diagram of an inspection device according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of another inspection device according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of a computing device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail with reference to the accompanying drawings, and it is apparent that the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic structural diagram of an IDC room according to an embodiment of the present invention, as shown in fig. 1, at least one row of cabinets, such as a cabinet 101 to a cabinet 106, may be arranged in the IDC room, the cabinets 101 to 104 may be arranged in parallel, the cabinet 105 and the cabinet 106 may be arranged in parallel, and each row of cabinets may be provided with multiple devices, such as a server device, a data acquisition device, a monitoring device, a temperature control device, and the like.
In the embodiment of the present invention, the cabinet may have a single-layer structure, and multiple devices may be placed in parallel on the single-layer structure, or the cabinet may also have a multi-layer structure, where multiple devices are placed in the multi-layer structure, and one or more devices may be placed in parallel on each layer of the multi-layer structure, which is not limited specifically.
Based on the IDC room illustrated in fig. 1, fig. 2 is a schematic diagram of a system architecture of an inspection system according to an embodiment of the present invention, as shown in fig. 2, the system architecture includes a model management device 110 and at least one mobile device, such as a mobile device 121, a mobile device 122, and a mobile device 123. The model management device 110 may be connected to any mobile device, for example, through a wired connection or a wireless connection, which is not limited specifically.
In the embodiment of the invention, each mobile device can be deployed in the same IDC machine room or different IDC machine rooms, if one mobile device is deployed in each IDC machine room, the mobile device can be responsible for polling the whole IDC machine room, if a plurality of mobile devices are deployed in each IDC machine room, each mobile device can be responsible for polling one area in the IDC machine room, and the plurality of mobile devices jointly complete the polling operation on the whole IDC machine room.
It should be noted that the inspection method in the embodiment of the present invention may be used to inspect one IDC room, and may also be used to inspect a plurality of IDC rooms, which is not limited specifically.
Based on the system architecture illustrated in fig. 2, fig. 3 is a schematic flow diagram corresponding to a polling method provided in an embodiment of the present invention, where the method includes:
step 301, the model management device sends a model training instruction to each mobile device.
In a possible implementation manner, the target recognition model may be obtained through training in multiple time slices, and in any time slice, the model management device 110 may first select a mobile device corresponding to the time slice from all mobile devices in the machine room, and then send a model training instruction to the mobile device corresponding to the time slice; accordingly, the mobile device that receives the model training instruction may perform model training in conjunction with the model management device 110 to update the recognition model corresponding to the previous time slice to obtain the recognition model corresponding to the time slice, while the mobile device that does not receive the model training instruction may only perform recognition operation without participating in model training in the time slice.
Each mobile device in step 301 may be a mobile device corresponding to any time slice.
In a specific implementation, the manner of selecting the mobile device corresponding to any time slice may be multiple, for example, a fixed number or a random number of mobile devices may be randomly selected as the mobile device corresponding to any time slice, or a fixed number or a random number of mobile devices may be selected in turn as the mobile device corresponding to any time slice, or a fixed number or a random number of mobile devices with stronger processing capability may be selected as the mobile device corresponding to any time slice, which is not limited specifically.
In one example, a set percentage of mobile devices may be selected from the total number of mobile devices as the mobile devices for any time slice. The set proportion may be set by a person skilled in the art according to experience, for example, may be set to 50% to 80%, so as to avoid overfitting caused by too much training data while keeping most of the features of the training data, and improve the accuracy of the recognition model.
In the implementation mode, by setting a plurality of time slices and selecting part of the mobile devices from each time slice to participate in model training, training data and the mobile devices can be effectively distributed and uniformly utilized, the problem that parameter iteration cannot be converged due to overlarge data during model training can be avoided by uniform distribution of the training data, the capacity of model management equipment for dealing with mobile device faults can be improved by uniform distribution of the mobile devices, and the usability of a training recognition model is improved.
And 302, the mobile equipment acquires training data when the mobile equipment travels in the machine room according to the model training instruction and the routing inspection route, and obtains an intermediate model corresponding to the mobile equipment based on the local model and the training data.
In one possible implementation, the model training instructions sent by the model management device 110 are used to instruct the mobile device to cyclically perform the following operations: training data are collected when the inspection route travels in the machine room, an intermediate model corresponding to the mobile device is obtained based on the local model and the training data, and the intermediate model is reported to the model management device 110. Therefore, in any time slice, for any mobile device receiving the model training instruction, the mobile device may continuously acquire training data, and after a fixed amount of training data is acquired, the acquired training data may be used to train the local model, so as to obtain an intermediate model corresponding to the mobile device, and report the intermediate model to the model management device 110. The fixed number may be 1, or may be any integer greater than 1, and is not limited.
The training data may be set by a person skilled in the art as needed, for example, the training data may be any one or more of an image of a machine room device, an image of a signal lamp, temperature information, and odor information, and is not limited.
In the embodiment of the present invention, the manner of optimizing the local model may be: the method comprises the steps of firstly predicting each training data by using a local model to obtain a prediction label of each training data, then determining the number of training data with correct prediction labels and the number of training data with wrong prediction labels from each training data according to the matching degree of the prediction label of each training data and a real label to calculate a loss function, and finally adjusting model parameters of the local model according to the loss function to obtain an intermediate model.
Correspondingly, when training data is collected, the mobile device may collect the training data according to a fixed frequency, or collect continuous data first, then intercept the training data from the continuous data according to the fixed frequency, and use the training data in each fixed time period in the intercepted training data as primary training data without limitation. For example, when the training data is an apparatus image and the fixed time period is 1 minute, if the fixed frequency is 6 milliseconds each time, the mobile apparatus may capture an apparatus image every 6 milliseconds, and then capture 600 apparatus images captured in each 1 minute time period as one training data, or may record an apparatus video with a duration of 1 minute (or longer) first, and then intercept one apparatus image every 6 milliseconds from the apparatus video, and select 600 consecutive apparatus images from the intercepted apparatus images as one training data each time.
It should be noted that the fixed number is only an exemplary illustration, and does not limit the present solution, and in a specific implementation, the collected training data may be used to train the local model after each training data collection in a fixed time period, or the collected training data may be used to train the local model after each training data collection in a random number or in a random time period, and the specific implementation is not limited.
In the embodiment of the present invention, the local model of the mobile device may be any one of an initial model, an identification model corresponding to a previous time slice, and a model corresponding to a comprehensive model parameter. In the starting stage of model training of a first time slice, the local model in the mobile equipment is an initial model, in the starting stage of model training of other time slices, the local model in the mobile equipment is a recognition model corresponding to the previous time slice, and in the executing stage of model training of any time slice, the local model of the mobile equipment is a model corresponding to the comprehensive model parameters.
In specific implementation, the initial model may be obtained by training the model management device 110 or any mobile device, taking the example of training the initial model by the model management device 110 as an example, the model management device 110 may first obtain initial training data, then use the initial training data to train to obtain the initial model, and then send model parameters of the initial model to all the mobile devices in the machine room. The initial training data may be obtained in various ways, such as downloading through a network, acquiring from a machine room before the first time slice, and obtaining from a third-party model management device, without limitation.
Correspondingly, for any mobile device which does not receive the model training instruction, the mobile device may be in a waiting state without participating in the recognition model training corresponding to the time slice, and in the next time slice, if the model training instruction is received, the mobile device may participate in the recognition model training corresponding to the next time slice, and if the model training instruction is not received, the mobile device may continue to be in the waiting state.
Step 303, the mobile device reports the model parameters of the intermediate model to the model management device.
Here, the model parameters of the intermediate model may include gradient and loss values, and may also include other information, without limitation.
In specific implementation, the mobile device may compress the model parameters of the intermediate model first, and then report the compressed packet to the model management device 110, so as to reduce communication overhead, reduce time consumed by data transmission, and improve inspection efficiency; or, the mobile device may further perform an encryption operation on the compressed packet, and report the encrypted compressed packet to the model management device 110, so as to improve the security of the data during transmission.
And 304, training the model management device based on the model parameters of the intermediate model corresponding to each mobile device to obtain a recognition model.
In this embodiment of the present invention, the model management device 110 may be a server or a mobile device, when the model management device 110 is a server, the model management device may directly calculate to obtain the integrated model parameters using the model parameters of each intermediate model sent by other mobile devices corresponding to the current time slice, and when the model management device 110 is a mobile device, the model management device 110 may send a model training instruction to other mobile devices corresponding to the current time slice, and may train to obtain the intermediate model corresponding to the model management device 110 using training data, so as to calculate to obtain the integrated model parameters based on the model parameters of other mobile devices corresponding to the current time slice and the model parameters of the intermediate model corresponding to the model management device 110.
In a specific implementation, for any time slice, the model management device 110 may train to obtain the recognition model corresponding to the time slice in multiple ways, for example, after receiving the model parameters sent by all or most of the mobile devices corresponding to the time slice, the recognition model corresponding to the time slice may be obtained by using the model parameters corresponding to all or most of the mobile devices, or the mobile device corresponding to the time slice may be instructed to perform model training alone multiple times to obtain multiple model parameters corresponding to each mobile device, and then the recognition model corresponding to the time slice is obtained by using each model parameter corresponding to each mobile device, which is not limited.
In a possible implementation manner, the model management device 110 may train to obtain the recognition model corresponding to any time slice by using the following method:
step a, the model management device 110 calculates the comprehensive model parameters according to the set number of model parameters when receiving the set number of model parameters.
In specific implementation, the model training instruction is used to instruct each mobile device to cyclically execute operations of acquiring training data, training an intermediate model, and reporting model parameters of the intermediate model, so that the model management device 110 may continuously receive the model parameters sent by each mobile device, and when a set number of model parameters are received, the model management device 110 may calculate a comprehensive model parameter according to the received set number of model parameters. Here, the set number may be set by a person skilled in the art according to experience, for example, the set number may be set to be slightly smaller than the total number of the mobile devices corresponding to the time slice, so that while the comprehensive model parameters are obtained by using as many model parameters as possible, the situation of a fault of a part of the mobile devices is compatible, and smooth model training is ensured.
In specific implementation, there may be a plurality of manners for obtaining the comprehensive model parameters through calculation, for example, an average parameter of a set number of model parameters (and model parameters of the model management device 110) may be used as the comprehensive model parameter, or a weighted average parameter of a set number of model parameters (and model parameters of the model management device 110) may be used as the comprehensive model parameter, or a model parameter that does not meet requirements may be first screened from a set number of model parameters (and model parameters of the model management device 110), and then an average parameter or a weighted average parameter of model parameters that meet requirements may be used as the comprehensive model parameter, which is not limited.
And b, the model management equipment 110 issues the comprehensive model parameters to all the mobile equipment in the machine room.
In one example, any mobile device in the machine room can also collect data to be identified in the process of traveling, and identify the data to be identified by using the local model, so as to complete the identification operation on the machine room in the process of routing inspection. The identification operation may be triggered by an identification instruction, or may be executed according to a set period, which is not limited.
Correspondingly, the model management device 110 may also issue the integrated model parameters to all the mobile devices in the machine room during the training process of each time slice, and accordingly, after any mobile device in the machine room (the mobile device that does not receive the model training instruction or the mobile device that receives the model training instruction) receives the integrated model parameters, the local model may be updated by using the integrated model parameters. Therefore, the mobile equipment can use the updated model to identify the data to be identified according to any data to be identified which is acquired subsequently, so that the combined operation of training, optimizing and identifying can be performed in the inspection process, the effects of training the model, optimizing the model and identifying the data can be realized, the training and identifying efficiency can be improved, and the identification can be realized by using the model which is optimized in real time.
Step c, the model management device 110 determines whether the comprehensive model parameters meet the end conditions corresponding to the current time slice, and if so, executes step d1If not, executing step d2。
The ending condition corresponding to any time slice may be any one or more of that the model parameter is not received in a set time period, the number of times of model training is greater than or equal to the set number of times, and the time of model training is greater than or equal to the set time.
Step d1The model management device 110 determines that the model training of the current time slice is finished, the integrated model parameter is the model parameter of the recognition model corresponding to the current time slice, and the model corresponding to the integrated model parameter is the recognition model corresponding to the current time slice, so the model management device 110 can send a model training finishing instruction to each mobile device corresponding to the current time slice.
Step d2The model management device 110 determines that the model training for the current time slice has not been completed, the integrated model parameter is not the recognition model parameter corresponding to the current time slice, and the model corresponding to the integrated model parameter is not the recognition model corresponding to the current time slice, so the model management device 110 may not perform special processing. Since each mobile device corresponding to the current time slice repeatedly performs the operations of acquiring training data, training the intermediate model, and reporting the model parameters, the model management device 110 may repeatedly perform steps a to d1Or from step a to step d2。
In the embodiment of the present invention, after determining the recognition model corresponding to the current time slice, the model management device 110 may further obtain a recognition effect of each mobile device using the recognition model corresponding to the current time slice, and determine whether the recognition effect meets an end condition of model training, if so, the recognition model corresponding to the current time slice may be used as a target recognition model, and a model training end instruction is sent to each mobile model corresponding to the current time slice; if not, model training for the next time slice can be started.
In an example, when a new inspection area or an inspection machine room is added, the model management device 110 may establish a communication connection with a mobile device in the new inspection area or the inspection machine room, and then issue the latest model parameter stored in the model management device 110 to the newly accessed mobile device, so that the newly accessed mobile device can be identified by using the latest model parameter. Correspondingly, when the next time slice is started, the model management device 110 may select the mobile device corresponding to the next time slice from all the devices (including the newly accessed mobile device) again, and perform the model training of the next time slice in a combined manner, so as to quickly implement the inspection operation on the newly inspected area, and improve the flexibility of the collaborative training process.
In this embodiment of the present invention, the model management device 110 may be a server or a mobile device, and when the model management device 110 is different, the mode of starting model training for the next time slice is also different, specifically:
when the model management device 110 is a server, the model management device 110 may be communicatively coupled to each mobile device in the computer room, and may not be communicatively coupled between the mobile devices in the computer room. In a specific implementation, if the model management device 110 determines to start the model training for the next time slice, it may select some mobile devices from all the mobile devices as the mobile devices corresponding to the next time slice, and then train in combination with the mobile devices corresponding to the next time slice to obtain the recognition model corresponding to the next time slice.
When the model management device 110 is a mobile device, any two mobile devices in the computer room may be communicatively connected such that each mobile device in the computer room forms a decentralized distributed cluster. In specific implementation, if it is determined that model training for the next time slice is started, all mobile devices may select a certain mobile device as a model management device corresponding to the next time slice through communication interaction, select some mobile devices from other mobile devices as other mobile devices corresponding to the next time slice through the model management device corresponding to the next time slice, and train with the model management device corresponding to the next time slice in combination with the other mobile devices corresponding to the next time slice to obtain an identification model corresponding to the next time slice.
For example, the mobile device with the highest computation capability may be used as the model management device corresponding to the next time slice, or the mobile device without the model management device may be selected in turn or randomly as the model management device corresponding to the next time slice, without limitation.
For convenience of understanding, specific implementation processes of the inspection method when the server is used as the model management device and the mobile device is used as the model management device are respectively described below.
Fig. 4 is an interaction flow diagram corresponding to the inspection method provided in the embodiment of the present invention, where the model management device is a server, and as shown in fig. 4, the method includes:
step 401, the model management device uses the initial training data to train to obtain an initial model.
Step 402, the model management device issues the model parameters of the initial model to all the mobile devices in the machine room.
Step 403, the model management device selects a mobile device corresponding to the current time slice from all mobile devices in the machine room.
Step 404, the model management device sends a model training instruction to the mobile device corresponding to the current time slice.
Step 405, for any mobile device corresponding to the current time slice (i.e. any mobile device that receives the model training instruction), the mobile device performs the following operations according to the model training instruction in a loop: and acquiring training data when the mobile equipment travels in the machine room according to the routing inspection route corresponding to the mobile equipment, and performing model training based on the local model and the training data to obtain an intermediate model corresponding to the mobile equipment.
The local model may be any one or more of an initial model, a recognition model corresponding to a previous time slice, and a model corresponding to the integrated model parameters, if the current time slice is the first time slice and the training is the first training in the first time slice, the local model is the initial model, if the current time slice is any time slice after the first time slice and the training is the first training in any time slice, the local model is the recognition model corresponding to the previous time slice, and if the current time slice is not the first training in any time slice, the local model is the model corresponding to the integrated model parameters.
Accordingly, for any mobile device that is not the current time slice (i.e., any mobile device that has not received the model training instruction), the mobile device does not participate in the model training of the current time slice, but may perform the identification operation, for example, when the mobile device travels along the corresponding patrol route in the machine room, the data to be identified may be collected, and then the data to be identified may be identified using the local model, or may perform the model update operation, for example, when the mobile device travels along the corresponding patrol route in the machine room, the mobile device may also simultaneously collect the training data, update the local model of the mobile device using the training data, and perform the identification operation on the data to be identified using the local model.
And 406, circularly reporting the model parameters of the intermediate model to the model management device by any mobile device corresponding to the current time slice according to the model training instruction.
Step 407, the model management device calculates to obtain the comprehensive model parameters according to the set number of model parameters every time the model management device receives the set number of model parameters.
And step 408, the model management device issues the comprehensive model parameters to all the mobile devices in the machine room.
Step 409, after any mobile device in the machine room receives the comprehensive model parameters sent by the model management device, updating the local model by using the comprehensive model parameters, so as to identify the data to be identified collected on the routing inspection route by using the updated local model. Any mobile device in the computer room may be any mobile device corresponding to the current time slice, and may also be any other mobile device (i.e., any mobile device that has not received the model training instruction).
In step 410, the model management device determines whether the model corresponding to the integrated model parameter meets the end condition of the current time slice, if yes, step 411 is executed, and if no, step 414 is executed.
The ending condition of the current time slice may be that the training times are greater than or equal to a preset training time, the training duration is greater than or equal to a preset training duration, the model effect corresponding to the comprehensive model parameter meets a preset model effect, and any one or more of the model parameters sent by the mobile device are not received within a set duration, which is not limited.
In step 411, the model management device sends a model training end instruction to the mobile device corresponding to the current time slice.
In step 412, the model management device determines that the model corresponding to the integrated model parameter is the recognition model corresponding to the current time slice, and determines whether the recognition model corresponding to the current time slice meets the end condition of the model training, if so, step 413 is executed, and if not, step 414 is executed.
The ending condition of the model training may be any one or any plurality of conditions, without limitation, in which the model effect satisfies the preset effect, the model training time is longer than or equal to the preset time, and the number of time slices is longer than or equal to the preset number of time slices.
For example, if the ending condition of the model training is that the model effect satisfies the preset effect, the model management device may obtain the recognition effect of each mobile device (the mobile device corresponding to the current time slice, or other mobile devices) that executes the recognition operation using the recognition model corresponding to the current time slice, then obtain the comprehensive recognition effect according to the recognition effect of each mobile device, if the comprehensive recognition effect does not reach the preset effect, determine that the recognition model corresponding to the current time slice does not satisfy the ending condition of the model training, and if the comprehensive recognition effect reaches the preset effect, determine that the recognition model corresponding to the current time slice satisfies the ending condition of the model training.
And 413, the model management equipment takes the recognition model corresponding to the current time slice as a target recognition model, and finishes model training.
In step 414, the model management device determines that the model training of the current time slice is finished, takes the next time slice as the current time slice, and executes step 403 to start the model training of the next time slice.
In the embodiment of the invention, the server is used as the model management equipment, so that the pressure of the mobile equipment can be reduced while the model training process of each mobile equipment is uniformly managed, and the efficiency of the mobile equipment in training the intermediate model is improved; and by reselecting the mobile equipment for executing training when each time slice is started, the mobile equipment can be effectively distributed, the difference of different mobile equipment is fully considered, and the accuracy of the identification model is improved.
Fig. 5 is an interaction flow diagram corresponding to another inspection method provided in an embodiment of the present invention, where the model management device refers to a mobile device, and as shown in fig. 5, the method includes:
step 501, any mobile device trains to obtain an initial model by using initial training data.
At step 502, any mobile device synchronizes the model parameters of the initial model to other mobile devices (mobile devices other than any mobile device) in the computer room.
Step 503, performing communication interaction on each mobile device in the computer room, and selecting a certain mobile device from each mobile device as a model management device corresponding to the current time slice.
Here, the selection manner may be multiple, for example, any two mobile devices may be set to send respective resource occupation statuses to each other, the mobile device with the lowest resource occupancy rate is identified by each mobile device, so as to select the mobile device with the strongest processing capability as the model management device, or a number may be set for each mobile device, each mobile device stores a correspondence between the number of another mobile device and an Internet Protocol (IP) address, when a certain mobile device completes model training for a corresponding time slice, the correspondence may be queried to determine the IP address of the mobile device with the next number, and then an instruction is sent to the IP address of the mobile device with the next number, where the instruction is used to instruct the mobile device with the next number to start model training for the next time slice, and so on.
In step 504, the model management device corresponding to the current time slice selects other mobile devices corresponding to the current time slice from other mobile devices (mobile devices other than the model management device corresponding to the current time slice).
Step 505, the model management device corresponding to the current time slice sends a model training instruction to other mobile devices corresponding to the current time slice.
Step 506, for any other mobile device corresponding to the current time slice, after receiving the model training instruction, the mobile device executes the following steps in a loop according to the model training instruction: and acquiring training data when the mobile equipment travels in the machine room according to the corresponding routing inspection route, and performing model training based on the local model and the training data to obtain an intermediate model corresponding to the mobile equipment.
Correspondingly, the model management device corresponding to the current time slice also executes the following steps: and acquiring training data when the corresponding inspection route travels in the machine room, and performing model training based on the local model and the training data to obtain an intermediate model corresponding to the model management equipment corresponding to the current time slice.
Step 507, any other mobile device corresponding to the current time slice executes in a cycle: and synchronizing the model parameters of the intermediate model to the model management equipment corresponding to the current time slice.
And step 508, calculating to obtain comprehensive model parameters according to the model parameters of the intermediate model obtained by training of the model management equipment corresponding to the current time slice and the model parameters of the set number every time the model management equipment corresponding to the current time slice receives the set number of model parameters.
Step 509, the model management device corresponding to the current time slice issues the integrated model parameters to each mobile device in the machine room.
And step 510, updating a local model by any mobile device in the machine room by using the comprehensive model parameters, and identifying the data to be identified collected on the corresponding routing inspection route by using the updated local model to determine the health state of the machine room.
In step 511, the model management device corresponding to the current time slice determines whether the model corresponding to the integrated model parameter meets the end condition of the current time slice, if so, step 512 is executed, and if not, step 514 is executed.
Step 512, the model management device corresponding to the current time slice sends a model training end instruction to other mobile devices corresponding to the current time slice.
Step 513, the model management device corresponding to the current time slice determines that the model corresponding to the integrated model parameter is the recognition model corresponding to the current time slice, and determines whether the recognition model corresponding to the current time slice meets the end condition of the model training, if so, step 514 is executed, and if not, step 515 is executed.
And step 514, the model management device corresponding to the current time slice takes the identification model corresponding to the current time slice as the target identification model.
Step 515, the model management device corresponding to the current time slice determines that the model training of the current time slice is finished, and executes step 503 with the next time slice as the current time slice.
For example, the model management device corresponding to the current time slice may send an interactive instruction to other mobile devices (mobile devices except the model management device corresponding to the current time slice), where the interactive instruction is used for any mobile device to send respective resource occupation conditions to the other mobile devices, and the mobile device with the lowest resource occupation condition is identified by each mobile device, so as to select the mobile device with the strongest processing capability as the model management device; alternatively, the model management device corresponding to the current time slice may send a start instruction to the IP address of the mobile device with the next number, where the start instruction is used to instruct the mobile device with the next number to start model training for the next time slice, and the like, without limitation.
In the embodiment of the invention, the mobile equipment is used as the model management equipment, so that a server does not need to be additionally arranged, and the cost of model training can be reduced; and by reselecting the mobile equipment for executing training when each time slice is started, the mobile equipment can be effectively distributed, the difference of different mobile equipment is fully considered, and the accuracy of the identification model is improved.
It should be noted that the step numbers in fig. 4 and fig. 5 are only one example of an execution flow, and do not form a limitation on the execution sequence of each step, for example, step 409 or step 510 may occur at any time in the flow, and is not limited.
In the above embodiment of the present invention, the model management device sends the model training instruction to each mobile device; and correspondingly, after the model management equipment receives the model parameters of the intermediate model reported by each mobile equipment, the model management equipment trains and obtains a recognition model based on the model parameters of the intermediate model reported by each mobile equipment, and the recognition model is used for determining the health state of the machine room. In the embodiment of the invention, the training process of the intermediate model is executed by the mobile equipment, so that the mobile equipment can only report the model parameters of the intermediate model to the model management equipment without reporting the whole amount of training data, and the model parameters have smaller data volume relative to the training data, therefore, the mode can save communication overhead and improve inspection efficiency; in addition, the health state of the machine room is determined by using the recognition model in the inspection process of the mobile equipment, so that the combined operation of model training and model recognition is realized, and the inspection efficiency can be improved.
Aiming at the method flow, the embodiment of the invention also provides the inspection device, and the specific content of the inspection device can be implemented by referring to the method.
Fig. 6 is a schematic structural diagram of an inspection device according to an embodiment of the present invention, including:
a transceiver module 601, configured to send a model training instruction to each mobile device; the model training instruction is used for acquiring training data when each mobile device travels in the machine room according to respective routing inspection routes, and an intermediate model is obtained based on a local model and the training data; receiving the model parameters of the intermediate model reported by each mobile device;
a training module 602, configured to train to obtain an identification model based on the model parameters of the intermediate model reported by each mobile device, where the identification model is used to determine a health state of the machine room.
Optionally, the identification model is an identification model corresponding to the current time slice;
before the transceiver module 601 sends the model training instructions to each mobile device, the training module 602 is further configured to:
and selecting the mobile equipment corresponding to the current time slice from all the mobile equipment in the machine room as each piece of mobile equipment.
Optionally, the model training instruction is used for acquiring training data when each mobile device circularly travels in the machine room according to respective routing inspection routes, training based on a local model and the training data to obtain an intermediate model and reporting the intermediate model;
the training module 602 is specifically configured to: every time a set number of model parameters are received, constructing at least based on the set number of model parameters to obtain a comprehensive model parameter;
the transceiver module 601 is specifically configured to: sending the comprehensive model parameters to each mobile device; the comprehensive model parameters are used for updating local models by the mobile devices; and when the model corresponding to a certain comprehensive model parameter meets the end condition of the current time slice, sending a model training end instruction to each mobile device, wherein the model corresponding to the comprehensive model parameter is the identification model.
Optionally, the apparatus is a server;
after the training module 602 obtains the recognition model based on the model parameters of the intermediate model reported by each mobile device, the training module is further configured to:
judging whether the recognition model meets the end condition of the model training, if not, selecting the mobile equipment corresponding to the next time slice from all the mobile equipment of the machine room, and sending a model training instruction to the mobile equipment corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
Optionally, the apparatus is a mobile device;
after the training module 602 obtains the recognition model based on the model parameters of the intermediate model reported by each mobile device, the training module is further configured to:
judging whether the recognition model meets the end condition of the model training, if not, selecting a certain mobile device as a model management device corresponding to the next time slice through communication with other mobile devices, and sending a model starting instruction to the model management device corresponding to the next time slice; the model starting instruction is used for the model management equipment corresponding to the next time slice to select other mobile equipment corresponding to the next time slice from other mobile equipment and send a model training instruction to the other mobile equipment corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
Fig. 7 is a schematic structural diagram of an inspection device according to an embodiment of the present invention, including:
a transceiver module 701, configured to receive a model training instruction sent by a model management device;
the training module 702 is configured to acquire training data when the mobile device travels in the machine room according to the model training instruction and according to the corresponding routing inspection route, and train based on a local model and the training data to obtain an intermediate model corresponding to the mobile device;
the transceiver module 701 is further configured to report the model parameters of the intermediate model to the model management device; the model parameters of the intermediate model reported by each mobile device are used for training the model management device to obtain an identification model, and the identification model is used for determining the health state of the machine room.
Optionally, the training module 702 is specifically configured to: according to the model training instruction, circularly executing: acquiring training data when the mobile equipment travels in a machine room according to a routing inspection route, and training based on a local model and the training data to obtain an intermediate model corresponding to the mobile equipment;
the transceiver module 701 is specifically configured to: according to the model training instruction, circularly executing: reporting the model parameters of the intermediate model to the model management equipment;
wherein, after reporting the model parameters of the intermediate model to the model management device in each cycle, the transceiver module 701 is further configured to: receiving the comprehensive model parameters sent by the model management equipment;
the training module 702 is further configured to: updating the local model using the integrated model parameters.
Optionally, the apparatus further includes a recognition module 703, after the training module 702 updates the local model using the comprehensive model parameters, the recognition module 703 is configured to:
and acquiring data to be identified when the routing inspection route travels in the machine room, and identifying the data to be identified by using the updated local model so as to determine the health state of the machine room.
From the above, it can be seen that: in the above embodiment of the present invention, the model management device sends the model training instruction to each mobile device; and correspondingly, after the model management equipment receives the model parameters of the intermediate model reported by each mobile equipment, the model management equipment trains and obtains a recognition model based on the model parameters of the intermediate model reported by each mobile equipment, and the recognition model is used for determining the health state of the machine room. In the embodiment of the invention, the training process of the intermediate model is executed by the mobile equipment, so that the mobile equipment can only report the model parameters of the intermediate model to the model management equipment without reporting the whole amount of training data, and the model parameters have smaller data volume relative to the training data, therefore, the mode can save communication overhead and improve inspection efficiency; in addition, the health state of the machine room is determined by using the recognition model in the inspection process of the mobile equipment, so that the combined operation of model training and model recognition is realized, and the inspection efficiency can be improved.
Based on the same technical concept, the embodiment of the present invention provides a computing device, as shown in fig. 8, including at least one processor 801 and a memory 802 connected to the at least one processor, where the specific connection medium between the processor 801 and the memory 802 is not limited in the embodiment of the present invention, and the processor 801 and the memory 802 are connected through a bus in fig. 8 as an example. The bus may be divided into an address bus, a data bus, a control bus, etc.
In the embodiment of the present invention, the memory 802 stores instructions executable by the at least one processor 801, and the at least one processor 801 may execute the inspection method according to any of the above steps by executing the instructions stored in the memory 802.
The processor 801 is a control center of the computing device, and may connect various parts of the computing device by using various interfaces and lines, and implement data processing by executing or executing instructions stored in the memory 802 and calling up data stored in the memory 802. Optionally, the processor 801 may include one or more processing units, and the processor 801 may integrate an application processor and a modem processor, where the application processor mainly processes an operating system, a user interface, an application program, and the like, and the modem processor mainly processes an issued instruction. It will be appreciated that the modem processor described above may not be integrated into the processor 801. In some embodiments, the processor 801 and the memory 802 may be implemented on the same chip, or in some embodiments, they may be implemented separately on separate chips.
The processor 801 may be a general-purpose processor, such as a Central Processing Unit (CPU), a digital signal processor, an Application Specific Integrated Circuit (ASIC), a field programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof, configured to implement or perform the methods, steps, and logic blocks disclosed in embodiments of the present invention. A general purpose processor may be a microprocessor or any conventional processor or the like. The steps of the method disclosed in connection with the inspection embodiment may be directly implemented by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor.
Memory 802, which is a non-volatile computer-readable storage medium, may be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. The Memory 802 may include at least one type of storage medium, and may include, for example, a flash Memory, a hard disk, a multimedia card, a card-type Memory, a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Programmable Read Only Memory (PROM), a Read Only Memory (ROM), a charged Erasable Programmable Read Only Memory (EEPROM), a magnetic Memory, a magnetic disk, an optical disk, and so on. The memory 802 is any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to such. The memory 802 of embodiments of the present invention may also be circuitry or any other device capable of performing a storage function to store program instructions and/or data.
Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium storing a computer program executable by a computing device, wherein when the program runs on the computing device, the computer program causes the computing device to execute the inspection method described in any of fig. 3 to 5.
It should be apparent to those skilled in the art that embodiments of the present invention may be provided as a method, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims (18)

1. A routing inspection method is applied to model management equipment, and the method comprises the following steps:
sending a model training instruction to each mobile device; the model training instruction is used for acquiring training data when each mobile device travels in the machine room according to respective routing inspection routes, and an intermediate model is obtained based on a local model and the training data;
receiving the model parameters of the intermediate model reported by each mobile device;
and training based on the model parameters of the intermediate model reported by each mobile device to obtain an identification model, wherein the identification model is used for determining the health state of the machine room.
2. The method of claim 1, wherein the recognition model is a recognition model corresponding to a current time slice;
before sending the model training instruction to each mobile device, the method further includes:
and selecting the mobile equipment corresponding to the current time slice from all the mobile equipment in the machine room as each piece of mobile equipment.
3. The method of claim 2, wherein the model training instructions are used for acquiring training data when the mobile devices circularly travel in the machine room according to respective routing inspection routes, training based on a local model and the training data to obtain an intermediate model and reporting the intermediate model;
the receiving the model parameters of the intermediate model reported by each mobile device, and training based on the model parameters of the intermediate model reported by each mobile device to obtain the recognition model includes:
each time a set number of model parameters are received, constructing and obtaining comprehensive model parameters at least based on the set number of model parameters, and sending the comprehensive model parameters to each mobile device; the comprehensive model parameters are used for updating local models by the mobile devices;
and when the model corresponding to a certain comprehensive model parameter meets the end condition of the current time slice, sending a model training end instruction to each mobile device, wherein the model corresponding to the comprehensive model parameter is the identification model.
4. The method according to claim 2 or 3, wherein the model management device is a server;
after the model parameters based on the intermediate models reported by the mobile devices are trained to obtain the recognition models, the method further comprises the following steps:
judging whether the recognition model meets the end condition of the model training, if not, selecting the mobile equipment corresponding to the next time slice from all the mobile equipment of the machine room, and sending a model training instruction to the mobile equipment corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
5. A method according to claim 2 or 3, wherein the model management device is a mobile device;
after the model parameters based on the intermediate models reported by the mobile devices are trained to obtain the recognition models, the method further comprises the following steps:
judging whether the recognition model meets the end condition of the model training, if not, selecting a certain mobile device as a model management device corresponding to the next time slice through communication with other mobile devices, and sending a model starting instruction to the model management device corresponding to the next time slice; the model starting instruction is used for the model management equipment corresponding to the next time slice to select other mobile equipment corresponding to the next time slice from other mobile equipment and send a model training instruction to the other mobile equipment corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
6. A polling method is applied to mobile equipment, and comprises the following steps:
receiving a model training instruction sent by model management equipment;
acquiring training data when the mobile equipment travels in a machine room according to the model training instruction and a corresponding routing inspection route, and training based on a local model and the training data to obtain an intermediate model corresponding to the mobile equipment;
reporting the model parameters of the intermediate model to the model management equipment; the model parameters of the intermediate model reported by each mobile device are used for training the model management device to obtain an identification model, and the identification model is used for determining the health state of the machine room.
7. The method according to claim 6, wherein the collecting training data when traveling in a machine room according to the routing inspection route according to the model training instruction, obtaining an intermediate model corresponding to the mobile device based on a local model and the training data training, and reporting model parameters of the intermediate model to the model management device comprises:
according to the model training instruction, circularly executing: acquiring training data when a routing inspection route advances in a machine room, training based on a local model and the training data to obtain an intermediate model corresponding to the mobile equipment, and reporting model parameters of the intermediate model to the model management equipment;
wherein, after reporting the model parameters of the intermediate model to the model management device in each cycle, the method further comprises:
receiving the comprehensive model parameters sent by the model management equipment;
updating the local model using the integrated model parameters.
8. The method of claim 7, wherein after updating the local model using the integrated model parameters, further comprising:
and acquiring data to be identified when the routing inspection route travels in the machine room, and identifying the data to be identified by using the updated local model so as to determine the health state of the machine room.
9. An inspection device, the device comprising:
the receiving and sending module is used for sending model training instructions to each mobile device; the model training instruction is used for acquiring training data when each mobile device travels in the machine room according to respective routing inspection routes, and an intermediate model is obtained based on a local model and the training data; receiving the model parameters of the intermediate model reported by each mobile device;
and the training module is used for training based on the model parameters of the intermediate model reported by each mobile device to obtain an identification model, and the identification model is used for determining the health state of the machine room.
10. The apparatus of claim 9, wherein the recognition model is a recognition model corresponding to a current time slice;
before the transceiver module sends the model training instructions to each mobile device, the training module is further configured to:
and selecting the mobile equipment corresponding to the current time slice from all the mobile equipment in the machine room as each piece of mobile equipment.
11. The apparatus of claim 10, wherein the model training instructions are used for the mobile devices to collect training data while circulating to travel in the machine room according to respective routing inspection routes, and to train to obtain an intermediate model based on a local model and the training data and to report;
the training module is specifically configured to: every time a set number of model parameters are received, constructing at least based on the set number of model parameters to obtain a comprehensive model parameter;
the transceiver module is specifically configured to: sending the comprehensive model parameters to each mobile device; the comprehensive model parameters are used for updating local models by the mobile devices; and when the model corresponding to a certain comprehensive model parameter meets the end condition of the current time slice, sending a model training end instruction to each mobile device, wherein the model corresponding to the comprehensive model parameter is the identification model.
12. The apparatus according to claim 10 or 11, wherein the apparatus is a server;
after the training module obtains the recognition model based on the model parameters of the intermediate model reported by each mobile device, the training module is further configured to:
judging whether the recognition model meets the end condition of the model training, if not, selecting the mobile equipment corresponding to the next time slice from all the mobile equipment of the machine room, and sending a model training instruction to the mobile equipment corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
13. The apparatus according to claim 10 or 11, wherein the apparatus is a mobile device;
after the training module obtains the recognition model based on the model parameters of the intermediate model reported by each mobile device, the training module is further configured to:
judging whether the recognition model meets the end condition of the model training, if not, selecting a certain mobile device as a model management device corresponding to the next time slice through communication with other mobile devices, and sending a model starting instruction to the model management device corresponding to the next time slice; the model starting instruction is used for the model management equipment corresponding to the next time slice to select other mobile equipment corresponding to the next time slice from other mobile equipment and send a model training instruction to the other mobile equipment corresponding to the next time slice; and if so, taking the recognition model as a target recognition model.
14. An inspection device, the device comprising:
the receiving and sending module is used for receiving a model training instruction sent by the model management equipment;
the training module is used for acquiring training data when the mobile equipment travels in a machine room according to the model training instruction and a corresponding routing inspection route, and training based on a local model and the training data to obtain an intermediate model corresponding to the mobile equipment;
the transceiver module is further configured to report the model parameters of the intermediate model to the model management device; the model parameters of the intermediate model reported by each mobile device are used for training the model management device to obtain an identification model, and the identification model is used for determining the health state of the machine room.
15. The apparatus of claim 14,
the training module is specifically configured to: according to the model training instruction, circularly executing: acquiring training data when the mobile equipment travels in a machine room according to a routing inspection route, and training based on a local model and the training data to obtain an intermediate model corresponding to the mobile equipment;
the transceiver module is specifically configured to: according to the model training instruction, circularly executing: reporting the model parameters of the intermediate model to the model management equipment;
wherein, after reporting the model parameters of the intermediate model to the model management device in each cycle, the transceiver module is further configured to: receiving the comprehensive model parameters sent by the model management equipment;
the training module is further configured to: updating the local model using the integrated model parameters.
16. The apparatus of claim 15, further comprising a recognition module, wherein after the training module updates the local model using the integrated model parameters, the recognition module is configured to:
and acquiring data to be identified when the routing inspection route travels in the machine room, and identifying the data to be identified by using the updated local model so as to determine the health state of the machine room.
17. A computing device comprising at least one processor and at least one memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the method of any of claims 1 to 8.
18. A computer-readable storage medium storing a computer program executable by a computing device, the program, when run on the computing device, causing the computing device to perform the method of any of claims 1 to 8.
CN202010103868.XA 2020-02-20 2020-02-20 Inspection method and device Pending CN111352799A (en)

Priority Applications (3)

Application Number Priority Date Filing Date Title
CN202010103868.XA CN111352799A (en) 2020-02-20 2020-02-20 Inspection method and device
PCT/CN2020/137411 WO2021164404A1 (en) 2020-02-20 2020-12-17 Inspection method and apparatus
TW109147101A TWI770749B (en) 2020-02-20 2020-12-31 Inspection method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010103868.XA CN111352799A (en) 2020-02-20 2020-02-20 Inspection method and device

Publications (1)

Publication Number Publication Date
CN111352799A true CN111352799A (en) 2020-06-30

Family

ID=71192396

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010103868.XA Pending CN111352799A (en) 2020-02-20 2020-02-20 Inspection method and device

Country Status (3)

Country Link
CN (1) CN111352799A (en)
TW (1) TWI770749B (en)
WO (1) WO2021164404A1 (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021164404A1 (en) * 2020-02-20 2021-08-26 中国银联股份有限公司 Inspection method and apparatus
CN116841301A (en) * 2023-09-01 2023-10-03 杭州义益钛迪信息技术有限公司 Inspection robot inspection model training method, device, equipment and medium
WO2023185711A1 (en) * 2022-03-28 2023-10-05 华为技术有限公司 Communication method and apparatus used for training machine learning model

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114434456A (en) * 2022-01-19 2022-05-06 山东新一代信息产业技术研究院有限公司 Machine room inspection robot and inspection method thereof

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180089587A1 (en) * 2016-09-26 2018-03-29 Google Inc. Systems and Methods for Communication Efficient Distributed Mean Estimation
CN109711556A (en) * 2018-12-24 2019-05-03 中国南方电网有限责任公司 Machine patrols data processing method, device, net grade server and provincial server
CN110309847A (en) * 2019-04-26 2019-10-08 深圳前海微众银行股份有限公司 A kind of model compression method and device
CN110490738A (en) * 2019-08-06 2019-11-22 深圳前海微众银行股份有限公司 A kind of federal learning method of mixing and framework

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI587152B (en) * 2016-10-26 2017-06-11 先智雲端數據股份有限公司 Method for extending life expectancy of disks in cloud-based service system and system using the same
CN110334633A (en) * 2019-06-27 2019-10-15 北京御航智能科技有限公司 Identification inspection data and the method, apparatus and storage medium for updating identification model
CN110298185A (en) * 2019-06-28 2019-10-01 北京金山安全软件有限公司 Model training method and device, electronic equipment and storage medium
CN110442457A (en) * 2019-08-12 2019-11-12 北京大学深圳研究生院 Model training method, device and server based on federation's study
CN110766169A (en) * 2019-10-31 2020-02-07 深圳前海微众银行股份有限公司 Transfer training optimization method and device for reinforcement learning, terminal and storage medium
CN111352799A (en) * 2020-02-20 2020-06-30 中国银联股份有限公司 Inspection method and device

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180089587A1 (en) * 2016-09-26 2018-03-29 Google Inc. Systems and Methods for Communication Efficient Distributed Mean Estimation
CN109711556A (en) * 2018-12-24 2019-05-03 中国南方电网有限责任公司 Machine patrols data processing method, device, net grade server and provincial server
CN110309847A (en) * 2019-04-26 2019-10-08 深圳前海微众银行股份有限公司 A kind of model compression method and device
CN110490738A (en) * 2019-08-06 2019-11-22 深圳前海微众银行股份有限公司 A kind of federal learning method of mixing and framework

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021164404A1 (en) * 2020-02-20 2021-08-26 中国银联股份有限公司 Inspection method and apparatus
WO2023185711A1 (en) * 2022-03-28 2023-10-05 华为技术有限公司 Communication method and apparatus used for training machine learning model
CN116841301A (en) * 2023-09-01 2023-10-03 杭州义益钛迪信息技术有限公司 Inspection robot inspection model training method, device, equipment and medium
CN116841301B (en) * 2023-09-01 2024-01-09 杭州义益钛迪信息技术有限公司 Inspection robot inspection model training method, device, equipment and medium

Also Published As

Publication number Publication date
TWI770749B (en) 2022-07-11
WO2021164404A1 (en) 2021-08-26
TW202132989A (en) 2021-09-01

Similar Documents

Publication Publication Date Title
CN111352799A (en) Inspection method and device
CN108768877B (en) Distribution method and device of burst traffic and proxy server
CN110784929B (en) Access resource allocation method, device, equipment and system
CN107925612A (en) Network monitoring system, network monitoring method and program
CN111612153A (en) Method and device for training model
CN112929187B (en) Network slice management method, device and system
CN108319974B (en) Data processing method, data processing device, storage medium and electronic device
CN111753875A (en) Power information system operation trend analysis method and device and storage medium
CN111966289A (en) Partition optimization method and system based on Kafka cluster
CN107943697A (en) Problem distribution method, device, system, server and computer-readable storage medium
CN106020777A (en) Data processing method, device and system
CN113869521A (en) Method, device, computing equipment and storage medium for constructing prediction model
CN107004003B (en) Model parameter fusion method and device
CN110855481B (en) Data acquisition system and method
CN107786390B (en) Method and device for correcting networking nodes
CN109462510B (en) CDN node quality evaluation method and device
CN115134582A (en) Video quality evaluation method and device
CN109587520A (en) A kind of method and apparatus of positioning video traffic failure
CN115037665B (en) Equipment testing method and device
CN107517474B (en) Network analysis optimization method and device
CN110825466A (en) Program jamming processing method and jamming processing device
CN109831335A (en) A kind of data monitoring method, monitor terminal, storage medium and data monitoring system
CN108989116A (en) Big data collects and surveys system and method
CN115225500A (en) Network slice allocation method and device
CN111078466B (en) Sensor data loss analysis method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
REG Reference to a national code

Ref country code: HK

Ref legal event code: DE

Ref document number: 40032060

Country of ref document: HK