Summary of the invention
Based on this, it is necessary to which in view of the above technical problems, providing a kind of can both can use each provincial machine and patrol data
It is trained and data processing side can be patrolled to avoid the machine for concentrating the machine of each point of province to patrol network transmission bottleneck problem caused by data
Method, device, net grade server, provincial server and readable storage medium storing program for executing.
In a first aspect, a kind of machine patrols data processing method, which comprises
Net grade server obtains the parameter for each provincial machine learning submodel that each provincial server is sent;The each province
The parameter of the machine learning submodel of grade is that each provincial server patrols data training sample according to each provincial machine respectively, to default
Initial machine learning model be trained;It includes: to learn mould as initial machine that the machine, which patrols data training sample,
The machine of type input patrols image data and the accident analysis data as the output of initial machine learning model;
The net grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains described
The target component of initial machine learning model;It is related that the target component to multiple provincial machines patrols data training sample;
The target component is distributed to each provincial server by the net grade server, to indicate each provincial server by institute
It states target component and substitutes into each provincial machine learning submodel, each provincial machine learning submodel after being upgraded.
In one of the embodiments, the net grade server to the parameter of each provincial machine learning submodel into
Row Combined Treatment obtains the target component of the initial machine learning model, comprising:
The net grade server is obtained according to the parameter and preset each provincial weight of each provincial machine learning submodel
The target component of the initial machine learning model.
In one of the embodiments, the net grade server to the parameter of each provincial machine learning submodel into
Row Combined Treatment obtains the target component of the initial machine learning model, comprising:
First parameter of at least one the first provincial machine learning submodel is distributed at least by the net grade server
One the second provincial corresponding second provincial server;
The net grade server obtains the first parameter after the optimization that the described second provincial server returns, as described first
The target component of beginning machine learning model;The first parameter after the optimization is the described second provincial server according to described second
Provincial machine patrols data training sample, is trained to obtain to the corresponding machine learning submodel of first parameter before optimization
's.
The net grade server obtains each provincial engineering that each provincial server is sent in one of the embodiments,
Practise the parameter of submodel, comprising:
The net grade server is directed to the upgrade command of each provincial machine learning submodel to each provincial server distribution;
The net grade server receives the machine learning submodel that each provincial server is sent for the upgrade command
Parameter.
The net grade server obtains each provincial engineering that each provincial server is sent in one of the embodiments,
Practise the parameter of submodel, further includes:
The net grade server distributes encrypted public key to each provincial server;
The net grade server receives the machine learning submodel that each provincial server is sent for the upgrade command
Parameter, comprising:
The net grade server receives the machine learning submodel that each provincial server is sent for the upgrade command
Encryption parameter, and the encryption parameter is decrypted according to encryption key corresponding with the encrypted public key, it obtains each provincial
Machine learning submodel parameter.
In one of the embodiments, the method also includes:
The net grade server issues analysis instruction to each provincial server and machine patrols data test sample;The analysis refers to
Order, which is used to indicate each provincial server, patrols data test sample according to the machine and tests machine learning submodel;
The net grade server receives the test result for the machine learning submodel that each provincial server returns.
Second aspect, a kind of machine patrol data processing method, which comprises
Provincial server patrols data training sample according to provincial machine, instructs to preset initial machine study submodel
Practice, obtains the parameter of the provincial machine learning submodel;It includes: as initial machine that the machine, which patrols data training sample,
The machine for practising mode input patrols image data and the accident analysis data as the output of initial machine learning model;
The parameter of the machine learning submodel is uploaded to net grade server by the provincial server, to indicate the net
Grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains the initial machine learning model
Target component;It is related that the target component to multiple provincial machines patrols data training sample;
The provincial server receives the target component of net grade server distribution, and by the target component generation
Enter the provincial machine learning submodel, the machine learning submodel after being upgraded.
In one of the embodiments, the method also includes:
The provincial server obtains the corresponding machine of fault ticket processing business and patrols image data and failure analysis result;
The corresponding machine of the fault ticket processing business is patrolled image data and failure analysis result by the provincial server,
Data training sample is patrolled as new machine, trains the machine learning submodel again.
The third aspect, a kind of machine patrol data processing equipment, and described device includes:
Provincial model parameter obtains module, obtains each provincial machine that each provincial server is sent for netting grade server
Learn the parameter of submodel;The parameter of each provincial machine learning submodel is each provincial server respectively according to each provincial
Machine patrol data training sample, preset initial machine learning model is trained;The machine patrols data training sample
It originally include: the failure patrolling image data as the machine of initial machine learning model input and being exported as initial machine learning model
Analyze data;
Target component computing module, for the net grade server to the parameter of each provincial machine learning submodel
Combined Treatment is carried out, the target component of the initial machine learning model is obtained;The target component is patrolled with multiple provincial machines
Data training sample is related;
The target component is distributed to each provincial server for the net grade server by target component sending module,
It is each provincial after being upgraded to indicate that the target component is substituted into each provincial machine learning submodel by each provincial server
Machine learning submodel.
Fourth aspect, a kind of machine patrol data processing equipment, and described device includes:
Provincial model training module patrols data training sample according to provincial machine for provincial server, to preset first
Beginning machine learning submodel is trained, and obtains the parameter of the provincial machine learning submodel;The machine patrols data training
Sample includes: the event patrolling image data as the machine of initial machine learning model input and exporting as initial machine learning model
Barrier analysis data;
Provincial model parameter sending module uploads the parameter of the machine learning submodel for the provincial server
Net grade server is given, to indicate that the net grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel,
Obtain the target component of the initial machine learning model;The target component patrols data training sample phase with multiple provincial machines
It closes;
Provincial model upgraded module receives the target ginseng of the net grade server distribution for the provincial server
Number, and the target component is substituted into the provincial machine learning submodel, the machine learning submodel after being upgraded.
5th aspect, a kind of net grade server, including memory and processor, the memory are stored with computer journey
Sequence, the processor perform the steps of when executing the computer program
Net grade server obtains the parameter for each provincial machine learning submodel that each provincial server is sent;The each province
The parameter of the machine learning submodel of grade is that each provincial server patrols data training sample according to each provincial machine respectively, to default
Initial machine learning model be trained;It includes: to learn mould as initial machine that the machine, which patrols data training sample,
The machine of type input patrols image data and the accident analysis data as the output of initial machine learning model;
The net grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains described
The target component of initial machine learning model;It is related that the target component to multiple provincial machines patrols data training sample;
The target component is distributed to each provincial server by the net grade server, to indicate each provincial server by institute
It states target component and substitutes into each provincial machine learning submodel, each provincial machine learning submodel after being upgraded.
6th aspect, a kind of provincial server, including memory and processor, the memory are stored with computer journey
Sequence, the processor perform the steps of when executing the computer program
Provincial server patrols data training sample according to provincial machine, instructs to preset initial machine study submodel
Practice, obtains the parameter of the provincial machine learning submodel;It includes: as initial machine that the machine, which patrols data training sample,
The machine for practising mode input patrols image data and the accident analysis data as the output of initial machine learning model;
The parameter of the machine learning submodel is uploaded to net grade server by the provincial server, to indicate the net
Grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains the initial machine learning model
Target component;It is related that the target component to multiple provincial machines patrols data training sample;
The provincial server receives the target component of net grade server distribution, and by the target component generation
Enter the provincial machine learning submodel, the machine learning submodel after being upgraded.
7th aspect, a kind of readable storage medium storing program for executing are stored thereon with computer program, and the computer program is by processor
It is performed the steps of when execution
Net grade server obtains the parameter for each provincial machine learning submodel that each provincial server is sent;The each province
The parameter of the machine learning submodel of grade is that each provincial server patrols data training sample according to each provincial machine respectively, to default
Initial machine learning model be trained;It includes: to learn mould as initial machine that the machine, which patrols data training sample,
The machine of type input patrols image data and the accident analysis data as the output of initial machine learning model;
The net grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains described
The target component of initial machine learning model;It is related that the target component to multiple provincial machines patrols data training sample;
The target component is distributed to each provincial server by the net grade server, to indicate each provincial server by institute
It states target component and substitutes into each provincial machine learning submodel, each provincial machine learning submodel after being upgraded.
Eighth aspect, a kind of readable storage medium storing program for executing are stored thereon with computer program, and the computer program is by processor
It is performed the steps of when execution
Provincial server patrols data training sample according to provincial machine, instructs to preset initial machine study submodel
Practice, obtains the parameter of the provincial machine learning submodel;It includes: as initial machine that the machine, which patrols data training sample,
The machine for practising mode input patrols image data and the accident analysis data as the output of initial machine learning model;
The parameter of the machine learning submodel is uploaded to net grade server by the provincial server, to indicate the net
Grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains the initial machine learning model
Target component;It is related that the target component to multiple provincial machines patrols data training sample;
The provincial server receives the target component of net grade server distribution, and by the target component generation
Enter the provincial machine learning submodel, the machine learning submodel after being upgraded.
Above-mentioned machine patrols data processing method, device, net grade server, provincial server and readable storage medium storing program for executing, although respectively
The case where it is different that provincial machine patrols data training sample, but feature is similar, meets " laterally federal study ", net level processor can
Target component is obtained to pass through the parameter for obtaining each provincial machine learning submodel and carry out Combined Treatment, and issues target ginseng
Each provincial server is counted to, is equivalent to and data training sample in a disguised form is patrolled to virtually shared initial machine by each provincial machine
Learning model is trained to obtain target component, there is no need to concentrate each provincial machine patrol the machines such as data training sample patrol data into
Row training, can also be co-stretched machine learning, play data scale advantage, realize the common of each provincial machine learning submodel
Growth improves the accuracy of each provincial machine learning submodel and comprehensive.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, not
For limiting the application.
Machine provided by the present application patrols data processing method, can be applied in application environment as shown in Figure 1.Wherein, net
Grade server 11 and multiple provincial servers 12 are communicated by network;Each provincial server is patrolled according to each provincial machine respectively
Data training sample is trained preset initial machine learning model, obtains the ginseng of each provincial machine learning submodel
It counts and uploads net grade server;Net grade server is received and is carried out at joint to the parameter of each provincial machine learning submodel
Reason, obtains target component, and target component is distributed to each provincial server;Each provincial server can be according to the target component
Each provincial machine learning submodel is updated, realizes federal study.Wherein, net grade server and provincial server can be used respectively
The server cluster of independent server either multiple servers composition is realized.
In one embodiment, as shown in Fig. 2, providing a kind of machine patrols data processing method, it is applied to Fig. 1 in this way
In net grade server for be illustrated, comprising the following steps:
S201, net grade server obtain the parameter for each provincial machine learning submodel that each provincial server is sent;Institute
The parameter for stating each provincial machine learning submodel is that each provincial server patrols data training sample according to each provincial machine respectively,
Preset initial machine learning model is trained;It includes: as initial machine that the machine, which patrols data training sample,
The machine of learning model input patrols image data and the accident analysis data as the output of initial machine learning model.
Net grade in the present embodiment can correspond to certain level-one administrative center in operation flow, provincial to correspond to operation flow
In with respect to net grade next stage administrative center;Such as electrical network business, net grade can correspond to national processing center, and each
The provincial processing center that can correspond to each point of province.For each provincial, provincial server can be stored in advance or to obtain this provincial
Machine patrol image data and patrol the accident analysis data of image data for each machine, and using machine patrol image data as input, will
Machine patrols the accident analysis data of image data as corresponding output, is trained to initial machine learning model, after being trained
Initial machine learning model, i.e. the provincial machine learning submodel, and obtain the parameter of the machine learning submodel;Example
Such as, for neural-network learning model, which can be the weighting parameter etc. of each neuron of each layer.Wherein, machine patrols
Image data can be image data, the video data etc. of unmanned plane shooting when carrying out inspection to transmission line of electricity;Accident analysis
Data can patrol image data to machine by a variety of analysis modes such as manual analysis, equipment analysis and be analyzed to obtain, can
To include simple evaluation parameter, such as " failure ", " normal ", " suspected malfunctions " etc., also may include to fault zone (such as certain
A insulator) mark, failure cause (such as aging) mark.
Above-mentioned initial machine learning model can be, but not limited to be neural-network learning model, Bayesian model, recurrence
Model etc..Each provincial initial machine learning model can be the same model;It or is same model, each parameter of model
Initial value can be different, but the configuration data of model is identical, such as the total number of each parameter is identical, the physics meaning of relevant parameter
Justice is identical: for example, each provincial initial machine learning model is neural network model, wherein each neural network model number of plies
Identical, the number of the neuron of same layer is identical, and the connection relationship of the neuron between each layer is identical.In one embodiment,
Above-mentioned net grade server can distribute above-mentioned modeling instruction to each provincial server, and the modeling instruction includes the initial machine
The configuration data of learning model, modeling instruction be used to indicate each provincial server according to the configuration data establish it is described just
Beginning machine learning model.
S202, the net grade server carry out Combined Treatment to the parameter of each provincial machine learning submodel, obtain
To the target component of the initial machine learning model;The target component patrols data training sample phase with multiple provincial machines
It closes.
For example, k1j,k2j,...,kij,...,kMj, j=1 ..., N are the parameter of N number of provincial machine learning submodel,
Net grade server can carry out Combined Treatment to the parameter of each provincial machine learning submodel, obtain target component K1,
K2,...,Ki,...,KM, for example, KiIt can be kijAverage value, median etc..It is understood that because above-mentioned each provincial
The parameter of machine learning submodel to patrol data training sample to each provincial machine respectively related, therefore above-mentioned target component and more
A provincial machine patrols data training sample correlation, is equivalent to and in a disguised form patrols data training sample to virtual total by each provincial machine
Some initial machine learning models are trained, the parameter of the machine learning model after being trained, i.e., above-mentioned target component;Respectively
Point province's machine patrols data and is retained in local, it is not necessary to upload to net grade server;Each point of province's server is equivalent to through net grade server
Realize data exchange.That is, the machine of the present embodiment, which patrols data processing method, is equivalent to the sea for both having given full play to the whole network
Amount machine patrols data, has carried out depth training to each provincial machine learning submodel analyzed for data, and adapt to net and save
Between network channel current resources situation.
The target component is distributed to each provincial server by S203, the net grade server, to indicate each provincial service
The target component is substituted into each provincial machine learning submodel by device, each provincial machine learning submodule after being upgraded
Type.
Above-mentioned target component is sent respectively to each provincial server by network connection by net grade server;Each provincial service
Device receives above-mentioned target component, updates each provincial machine learning submodel according to above-mentioned target component respectively;For example, by above-mentioned
The parameter current of machine learning submodel replaces with above-mentioned target component, can also be according to the current of above-mentioned machine learning submodel
Parameter and above-mentioned target component obtain the specific objective parameter for above-mentioned machine learning submodel, and above-mentioned machine learning is sub
The parameter current of model replaces with the specific objective parameter, which can working as above-mentioned machine learning submodel
The value that the average value or weighted sum of preceding parameter and above-mentioned target component are handled, weight can be default for experience, and
Weights sum is 1.It should be noted that the target component is distributed at least one provincial server by net grade server, and
It is not necessarily all provincial servers.
Optionally, the machine that net grade server can execute the present embodiment according to predetermined period patrols data processing method, realizes
To periodically updating for each provincial machine learning submodel, the timeliness of each provincial machine learning submodel is improved.
It is patrolled in data processing method in the machine of the present embodiment, although each provincial machine patrols data training sample difference,
The case where feature is similar, meets " laterally federal study ", net level processor can be by obtaining each provincial machine learning submodule
The parameter of type simultaneously carries out Combined Treatment and obtains target component, and issues target component to each provincial server, is equivalent in a disguised form
It patrols data training sample by each provincial machine virtually shared initial machine learning model is trained to obtain target component,
There is no need to concentrate each provincial machine to patrol the machines such as data training sample and patrol data to be trained, it can be also co-stretched machine learning,
Data scale advantage is played, realizes the common growth of each provincial machine learning submodel, improves each provincial machine learning
The accuracy of model and comprehensive.
Optionally, what is involved is net grade servers to pass through a kind of realization side of Combined Treatment acquisition target component for the present embodiment
Formula can specifically include: the net grade server is according to the parameter of each provincial machine learning submodel and preset each provincial
Weight obtains the target component of the initial machine learning model.
Wherein, each provincial weight can be experience preset value, be also possible to each provincial in other models of electrical network business
Weight is also possible to other way acquisition, for example, the available above-mentioned each provincial machine learning submodel of net grade server
The corresponding machine of parameter patrol the number of data training sample, and the number of data training sample is patrolled according to above-mentioned each provincial machine
Training sample specific gravity of the mesh in the total number that machine patrols data training sample, calculates each provincial weight, and simplest mode is
Using each provincial training sample specific gravity as each provincial weight, or each provincial training sample specific gravity convert
To each provincial weight.The machine of the present embodiment patrols data processing method and provides a kind of simple and effective Combined Treatment acquisition target
A kind of implementation of parameter.
Optionally, referring to shown in Fig. 3, what is involved is net grade servers to obtain target component by Combined Treatment for the present embodiment
Another implementation, can specifically include:
S301, the net grade server distribute the first parameter of at least one the first provincial machine learning submodel
Give at least one second provincial corresponding second provincial server.
Optionally, the parameter of multiple provincial machine learning submodels can be distributed to except described more by net grade server
It is a it is provincial except other provincial corresponding provincial servers.
S302, the net grade server obtain the first parameter after the optimization that the described second provincial server returns, as
The target component of the initial machine learning model;The first parameter after the optimization is the described second provincial server according to institute
It states the second provincial machine and patrols data training sample, the corresponding machine learning submodel of first parameter before optimization is instructed
It gets.
Illustratively, the of the A that the provincial server of the available A of net grade server is sent provincial machine learning submodel
One parameter, which is that the provincial server of A according to the provincial machine of A patrols data training sample, to preset initial machine
Practise what model was trained;Then above-mentioned first parameter is sent to the provincial server of B by net grade server;The provincial service of B
Device can patrol data using the initial parameter of the above-mentioned first parameter machine learning submodel provincial as B, and according to the provincial machine of B
The training sample machine learning submodel provincial to B is trained, the first parameter after being optimized, and returns to net grade service
Device;Net grade server can be distributed to each provincial server as each using the first parameter after above-mentioned optimization as target component
The parameter of provincial machine learning submodel;It is understood that optimization after the first parameter be equivalent to it is provincial at least through A
Machine patrol data training sample and the provincial machine of B patrols the training of data training sample, i.e., without concentrating A provincial and B provincial machine
It patrols the machines such as data training sample and patrols data and be trained, can also be co-stretched machine learning, play data scale advantage.
It is understood that above-mentioned net grade server can be directed to N number of provincial server respectively, in order will to tandem
The parameter of first provincial machine learning submodel is distributed to second provincial server, by second provincial server foundation
Second provincial machine patrols data training sample and is trained optimization, the parameter after being optimized;Net grade server again will optimization
Parameter afterwards is distributed to the provincial server of third, patrols data training according to the provincial machine of third by the provincial server of third
Sample carries out training optimization again, obtains the parameter of suboptimization again, in this way, all provincial servers of traversal, obtain final optimization pass
Target component afterwards, the corresponding machine learning model of the target component be equivalent to have passed through N number of provincial machine patrol data training sample
This training can be also co-stretched there is no need to concentrate each provincial machine to patrol the machines such as data training sample and patrol data to be trained
Machine learning plays data scale advantage.
In addition, net grade server can obtain X provincial machines from X provincial servers in order to save the processing time
Learn the parameter of submodel, and is distributed to other Y provincial servers;Then after obtaining the optimization that Y provincial servers return
Parameter, then remaining Z provincial servers are handed down to, until traversing all provincial servers, and multiple to what is finally obtained
Final optimization pass parameter is averaged, and target component is obtained, wherein X, Y, Z etc. are all larger than or are equal to 1.It is understood that net grade
Multiple parameters can be handed down to the same provincial server and optimized by server.Illustratively, net grade server is respectively with 5
A provincial server (mark is respectively 1~5) connection, the ginseng that the available provincial server 1,2 of net grade server is sent respectively
Number 1, parameter 2;Then parameter 1 is distributed to provincial server 3 to optimize, the parameter 1 after returned a optimization;It will
Parameter 2 is distributed to provincial server 4 and optimizes, the parameter 2 after returned a optimization;After net grade server will optimize
Parameter 1 and optimization after parameter 2 be distributed to provincial server 5 and optimize, two returned final optimization pass parameter is led to
It crosses and averages to obtain target component.
In short, the machine of the present embodiment patrol data processing method at least can be in a disguised form provincial and second is provincial by first
Machine patrols data training sample and is trained to obtain target component to virtually shared initial machine learning model, without concentrating the
One is provincial and the second provincial machine patrols the machines such as data training sample and patrols data and is trained.
Optionally, referring to shown in Fig. 4, what is involved is net grade servers to obtain each provincial machine learning submodule for the present embodiment
The process of type, can specifically include:
S401, the net grade server are directed to the upgrading of each provincial machine learning submodel to each provincial server distribution
Instruction.
Above-mentioned upgrade command can be a kind of agreement instruction of net grade server and provincial server, and provincial server can be with
It identifies above-mentioned upgrade command, then data training sample can be patrolled to provincial engineering according to the new machine in preset time period
It practises submodel to be trained, and the parameter of trained machine learning submodel is returned into net grade server, or can be direct
Obtain the parameter of trained machine learning submodel and return in preset time period.
S402, the net grade server receive each provincial server and are directed to the machine learning submodule that the upgrade command is sent
The parameter of type.
The net grade server of the present embodiment can receive each provincial of each provincial server transmission after distributing upgrade command
The parameter of machine learning submodel, it can realize that the unified of each provincial machine learning submodel rises in the whole network by upgrade command
Grade updates.
In one embodiment, the net grade server obtains each provincial machine learning that each provincial server is sent
The parameter of submodel, further includes: the net grade server distributes encrypted public key to each provincial server;The net grade server connects
Receive the parameter for the machine learning submodel that each provincial server is sent for the upgrade command, comprising: the net grade server
The encryption parameter for the machine learning submodel that each provincial server is sent for the upgrade command is received, and is added according to described
The encryption parameter is decrypted in key corresponding encryption key in Migong, obtains the parameter of each provincial machine learning submodel.
That is, above-mentioned net grade server can also distribute encrypted public key while distributing upgrade command, the encrypted public key is used
The parameter of each provincial machine learning submodel is encrypted using above-mentioned encrypted public key in each provincial server of instruction;Therefore
The parameter that above-mentioned net grade server receives be it is encrypted, net grade server can by corresponding with above-mentioned encrypted public key plus
Close private key is decrypted to obtain the parameter of each provincial machine learning submodel.The machine of the present embodiment patrols data processing method can be
It is co-stretched machine learning under the premise of protection data-privacy, meets the new demand of " privacy, safety and satisfaction supervision ".
Optionally, referring to Figure 5, what is involved is net grade servers to each provincial machine learning submodel for the present embodiment
The process tested, can specifically include:
S501, the net grade server issues analysis instruction to each provincial server and machine patrols data test sample;It is described
Analysis instruction is used to indicate each provincial server and patrols data test sample to the progress of machine learning submodel according to the machine
Test.
It is understood that provincial server can patrol data training sample to the engineering after upgrading using newest machine
Submodel is practised to be trained again;And net the accuracy etc. that grade server needs periodically to obtain each provincial machine learning submodel
Performance at the appropriate time upgrades each provincial machine learning submodel with realizing monitoring, for example, when at least half of
When the accuracy of provincial machine learning submodel is lower than pre-set level.
It may include patrolling image measurement data and as standard as the machine that standard inputs that above-mentioned machine, which patrols data test sample,
The accident analysis test data that image measurement data are patrolled for the machine of output;Illustratively, each provincial server can incite somebody to action
Above-mentioned machine patrols input of the image measurement data as provincial machine learning submodel, and the analysis exported will be as a result, and will be described
Analysis result and the accident analysis test data are matched, if matching, tests correct number and add 1, count final survey
It tries correct number and machine patrols the ratio of data test number of samples, as test result.
S502, the net grade server receive the test result for the machine learning submodel that each provincial server returns.
Above-mentioned test result can characterize the performance of above-mentioned each provincial machine learning submodel, such as accuracy;Net grade
Server can know the performance of each machine learning submodel according to the test result of each machine learning submodel, realize unified
Monitoring management.
In one embodiment, as shown in fig. 6, providing a kind of machine patrols data processing method, it is applied to Fig. 1 in this way
In provincial server for be illustrated, comprising the following steps:
S601, provincial server patrol data training sample according to provincial machine, learn submodel to preset initial machine
It is trained, obtains the parameter of the provincial machine learning submodel;It includes: as initial that the machine, which patrols data training sample,
The machine of machine learning model input patrols image data and the accident analysis data as the output of initial machine learning model;
The parameter of the machine learning submodel is uploaded to net grade server by S602, the provincial server, with instruction
The net grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains the initial machine study
The target component of model;It is related that the target component to multiple provincial machines patrols data training sample;
S603, the provincial server receive the target component of net grade server distribution, and by the target
Parameter substitutes into the provincial machine learning submodel, the machine learning submodel after being upgraded.
The technical solution of above-mentioned S601-S603 illustrates the explanation for being referred to above-mentioned S201-S203, here no longer
It repeats.
Optionally, above-mentioned each provincial server can patrol the number that the machines such as image data and accident analysis data patrol data to machine
Unified arrangement is carried out according to standard, unified data management classification, naming rule, storage organization etc. is defined, further promotes data
The efficiency and accuracy of analysis;The metadata that machine can also be patrolled data by above-mentioned each provincial server carries out unified arrangement and real-time
It stores to net grade server, to net the situation that grade server monitoring the whole network machine patrols data, the data of group knitmesh grade analyze work,
Wherein, metadata is mainly to describe the information of data attribute, has the function of designation date storage location, flag data etc..
Optionally, referring to shown in Fig. 7, what is involved is provincial servers according to the number of fault ticket processing business for the present embodiment
According to process trained again is carried out to provincial machine learning submodel, can specifically include:
S701, the provincial server obtain the corresponding machine of fault ticket processing business and patrol image data and accident analysis knot
Fruit.
Above-mentioned fault ticket processing business be usually when patrolling image data progress manual analysis to machine or after equipment analysis,
Manual analysis is carried out again or carries out the business such as reaffirming to fault in-situ, may include that the fault ticket processing business is corresponding
Machine patrol image data and failure analysis result.
The corresponding machine of the fault ticket processing business is patrolled image data and accident analysis by S702, the provincial server
As a result, patrolling data training sample as new machine, the machine learning submodel is trained again.
In the present embodiment, above-mentioned fault ticket processing business can be can be used as a kind of new more needle by provincial server
To the training sample of property, the machine learning submodel is trained again, improves the accuracy of machine learning model.
It is patrolled in data processing method in the machine of the present embodiment, although each provincial machine patrols data training sample difference,
The case where feature is similar, meets " laterally federal study ", provincial server can patrol data training sample according to provincial machine, right
Preset initial machine study submodel is trained, and is obtained the parameter of the provincial machine learning submodel and is given net grade
Server, net level processor can carry out Combined Treatment to the parameter of each provincial machine learning submodel and obtain target component,
And target component is issued to each provincial server, it is equivalent to and data training sample in a disguised form is patrolled to virtual total by each provincial machine
Some initial machine learning models are trained to obtain target component, and there is no need to concentrate each provincial machine to patrol data training sample
Equal machines patrol data and are trained, and can also be co-stretched machine learning, play data scale advantage, realize each provincial machine learning
The common growth of submodel improves the accuracy of each provincial machine learning submodel and comprehensive.
Referring to shown in Fig. 8 a, data processing method is patrolled to the machine of the present embodiment from another point of view and is illustrated, net grade service
Device can patrol data to the machine of the whole network and carry out unified management (metadata is centrally stored, physical file dispersion storage), regular basis
The whole network machine patrols the acquisition situation of data and the unified progress of research situation of machine learning techniques, and the machine in network-wide basis is organized to patrol
Data processing model (i.e. each provincial machine learning submodel) upgrading, can specifically include:
S801, net grade server can distribute computations to each provincial server;The computations may include upgrading
Instruction, is used to indicate the parameter that each provincial server returns to each provincial trained machine learning submodel;The computations
It can also include encrypted public key, be used to indicate each provincial server and the parameter for the machine learning submodel that needs return is added
It is close;
S802, each provincial net grade server can patrol data training sample according to each provincial machine respectively, to preset first
Beginning machine learning model is trained, and obtains the parameter of each provincial machine learning submodel, i.e. intermediate result, and using encryption
Public key returns to net grade server after encrypting the parameter of machine learning submodel;
S803, net grade server can carry out Combined Treatment to the parameter of each provincial machine learning submodel, obtain institute
The target component for stating initial machine learning model, is equivalent to the parameter of virtually shared conjunctive model, and by the target component
It is distributed to each provincial server, to indicate that each provincial server updates each provincial machine learning submodule according to the target component
The parameter of type.
Referring to shown in Fig. 8 b, provincial user can be according to the provincial analysis demand analyzed and the machine of initiating be needed to patrol data, specifically
Ground, the machine that can be analysed to patrol picture number and are sent to provincial server, and the machine that provincial server can be analysed to patrols image
Data are input in provincial machine learning submodel, exported as a result, and by output result feed back to above-mentioned provincial user,
The output result is that the machine of input patrols the failure analysis result of image data.Provincial server can also be by above-mentioned output result certainly
It is dynamic to be synchronized to net grade server and make a backup store.
Referring to shown in Fig. 8 c, net grade user can analyze the analysis demand for needing the machine of initiating to patrol data according to net grade, specifically
Ground can be managed according to net grade and execution each province's machine is required to patrol situation across comparison etc., such as need to obtain the defeated of each provincial same day
Line fault situation;It is stored in each provincial since machine patrols data dispersion, net grade server can be distributed to each provincial server
Analysis instruction, which, which is used to indicate each provincial server, patrols image data to the machine on the same day and analyzes;Each provincial clothes
The machine on business device available each provincial same day patrols image data, and is input in each provincial machine learning submodel, obtains each
The machine on the provincial same day patrols the failure analysis result of image data, can count and be analyzed for transmission line malfunction rate etc. as a result, uploading
Result is analyzed to net grade server;Net grade server receives each provincial analysis of above-mentioned each provincial server transmission as a result, can
To carry out across comparison to each provincial transmission line malfunction rate, comparing result is obtained, and above-mentioned comparing result is fed back into net
Grade user.
It should be understood that although each step in the flow chart of Fig. 2-7 is successively shown according to the instruction of arrow,
These steps are not that the inevitable sequence according to arrow instruction successively executes.Unless expressly stating otherwise herein, these steps
Execution there is no stringent sequences to limit, these steps can execute in other order.Moreover, at least one in Fig. 2-7
Part steps may include that perhaps these sub-steps of multiple stages or stage are not necessarily in synchronization to multiple sub-steps
Completion is executed, but can be executed at different times, the execution sequence in these sub-steps or stage is also not necessarily successively
It carries out, but can be at least part of the sub-step or stage of other steps or other steps in turn or alternately
It executes.
In one embodiment, as shown in figure 9, providing a kind of machine patrols data processing equipment, it is applied to net grade server,
It may include: that provincial model parameter obtains module 91, target component computing module 92 and target component sending module 93, in which:
Provincial model parameter obtains module 91, obtains each provincial machine that each provincial server is sent for netting grade server
The parameter of device study submodel;The parameter of each provincial machine learning submodel is each provincial server respectively according to each province
The machine of grade patrols data training sample, is trained to preset initial machine learning model;The machine patrols data training
Sample includes: the event patrolling image data as the machine of initial machine learning model input and exporting as initial machine learning model
Barrier analysis data;
Target component computing module 92, for the net grade server to the ginseng of each provincial machine learning submodel
Number carries out Combined Treatment, obtains the target component of the initial machine learning model;The target component and multiple provincial machines
It is related to patrol data training sample;
The target component is distributed to each provincial service for the net grade server by target component sending module 93
Device, it is each after being upgraded to indicate that the target component is substituted into each provincial machine learning submodel by each provincial server
Provincial machine learning submodel.
Optionally, in one embodiment, referring to Fig.1 shown in 0, on the basis of above-mentioned Fig. 9, the target component meter
Calculating module 92 may include:
Target component computing unit 921, the net grade server according to the parameter of each provincial machine learning submodel and
Preset each provincial weight, obtains the target component of the initial machine learning model.
Optionally, in another embodiment, the target component computing module 92 may include:
Parameter distribution optimization unit, for the net grade server by least one the first provincial machine learning submodel
The first parameter, be distributed at least one second provincial corresponding second provincial server;
Target component acquiring unit, after the optimization that the described second provincial server returns is obtained for the net grade server
The first parameter, the target component as the initial machine learning model;The first parameter after the optimization is described second
Provincial server patrols data training sample according to the described second provincial machine, to the corresponding machine of first parameter before optimization
Study submodel is trained.
Optionally, referring to Fig.1 shown in 0, the provincial model parameter obtains module 91 and may include:
Upgrade command transmission unit 911, for the net grade server to each provincial server distribution for each provincial
The upgrade command of machine learning submodel;
Provincial model parameter acquiring unit 912 receives each provincial server for the net grade server and is directed to the liter
The parameter for the machine learning submodel that grade instruction is sent.
Optionally, referring to Fig.1 shown in 0, the provincial model parameter, which obtains module 91, to include:
Public key transmission unit 913 distributes encrypted public key to each provincial server for the net grade server;
The provincial model parameter acquiring unit 912 specifically can be used for the net grade server and receive each provincial server
For the encryption parameter for the machine learning submodel that the upgrade command is sent, and according to encryption corresponding with the encrypted public key
The encryption parameter is decrypted in private key, obtains the parameter of each provincial machine learning submodel.
Optionally, referring to Fig.1 shown in 0, the machine, which patrols data processing equipment, to include:
Test sample sending module 94 issues analysis instruction to each provincial server for the net grade server and machine patrols
Data test sample;The analysis instruction is used to indicate each provincial server and patrols data test sample to machine according to the machine
Device study submodel is tested;
Test result receiving module 95 receives machine learning that each provincial server returns for the net grade server
The test result of model.
It is patrolled in data processing equipment in the machine of the present embodiment, although each provincial machine patrols data training sample difference,
The case where feature is similar, meets " laterally federal study ", net level processor can be by obtaining each provincial machine learning submodule
The parameter of type simultaneously carries out Combined Treatment and obtains target component, and issues target component to each provincial server, is equivalent in a disguised form
It patrols data training sample by each provincial machine virtually shared initial machine learning model is trained to obtain target component,
There is no need to concentrate each provincial machine to patrol the machines such as data training sample and patrol data to be trained, it can be also co-stretched machine learning,
Data scale advantage is played, realizes the common growth of each provincial machine learning submodel, improves each provincial machine learning
The accuracy of model and comprehensive.
In one embodiment, as shown in figure 11, it provides a kind of machine and patrols data processing equipment, be applied to provincial service
Device, may include: provincial model training module 111, provincial model parameter sending module 112 and provincial model upgraded module 113,
Wherein:
Provincial model training module 111 patrols data training sample according to provincial machine for provincial server, to preset
Initial machine study submodel is trained, and obtains the parameter of the provincial machine learning submodel;The machine patrols data instruction
Practicing sample includes: that the machine inputted as initial machine learning model patrols image data and exports as initial machine learning model
Accident analysis data;
Provincial model parameter sending module 112, for the provincial server by the parameter of the machine learning submodel
It is uploaded to net grade server, to indicate that the net grade server carries out at joint the parameter of each provincial machine learning submodel
Reason, obtains the target component of the initial machine learning model;The target component and multiple provincial machines patrol data training sample
This correlation;
Provincial model upgraded module 113 receives the mesh of the net grade server distribution for the provincial server
Parameter is marked, and the target component is substituted into the provincial machine learning submodel, the machine learning submodule after being upgraded
Type.
Optionally, in one embodiment, referring to Fig.1 shown in 2, on the basis of above-mentioned Figure 11, the machine patrols data
Processing unit can also include:
It is corresponding to obtain fault ticket processing business for the provincial server for fault ticket data acquisition module 114
Machine patrols image data and failure analysis result;
Fault ticket data training module 115, it is for the provincial server that the fault ticket processing business is corresponding
Machine patrol image data and failure analysis result, patrol data training sample as new machine, train machine learning again
Model.
It is patrolled in data processing equipment in the machine of the present embodiment, although each provincial machine patrols data training sample difference,
The case where feature is similar, meets " laterally federal study ", provincial server can patrol data training sample according to provincial machine, right
Preset initial machine study submodel is trained, and is obtained the parameter of the provincial machine learning submodel and is given net grade
Server, net level processor can carry out Combined Treatment to the parameter of each provincial machine learning submodel and obtain target component,
And target component is issued to each provincial server, it is equivalent to and data training sample in a disguised form is patrolled to virtual total by each provincial machine
Some initial machine learning models are trained to obtain target component, and there is no need to concentrate each provincial machine to patrol data training sample
Equal machines patrol data and are trained, and can also be co-stretched machine learning, play data scale advantage, realize each provincial machine learning
The common growth of submodel improves the accuracy of each provincial machine learning submodel and comprehensive.
The specific restriction for patrolling data processing equipment about machine may refer to the limit that data processing method is patrolled above for machine
Fixed, details are not described herein.Above-mentioned machine patrol the modules in data processing equipment can fully or partially through software, hardware and its
Combination is to realize.Above-mentioned each module can be embedded in the form of hardware or independently of in the processor in computer equipment, can also be with
It is stored in the memory in computer equipment in a software form, in order to which processor calls the above modules of execution corresponding
Operation.
In one embodiment, a kind of net grade server is provided, internal structure chart can be as shown in figure 13.The net grade
Server includes processor, memory, network interface and the database connected by system bus.Wherein, the net grade server
Processor for provide calculate and control ability.The memory of the net grade server includes non-volatile memory medium, memory
Reservoir.The non-volatile memory medium is stored with operating system, computer program and database.The built-in storage is non-volatile
The operation of operating system and computer program in storage medium provides environment.The database of the net grade server is each for storing
The parameter of provincial machine learning submodel, each provincial machine patrol metadata of data etc..The network interface of the net grade server
For being communicated with external terminal by network connection.To realize that a kind of machine patrols data when the computer program is executed by processor
Processing method.
It will be understood by those skilled in the art that structure shown in Figure 13, only part relevant to application scheme
The block diagram of structure does not constitute the restriction for the net grade server being applied thereon to application scheme, specific net grade service
Device may include perhaps combining certain components or with different component layouts than more or fewer components as shown in the figure.
In one embodiment, a kind of net grade server, including memory and processor are provided, is stored in memory
Computer program, the processor perform the steps of when executing computer program
Net grade server obtains the parameter for each provincial machine learning submodel that each provincial server is sent;The each province
The parameter of the machine learning submodel of grade is that each provincial server patrols data training sample according to each provincial machine respectively, to default
Initial machine learning model be trained;It includes: to learn mould as initial machine that the machine, which patrols data training sample,
The machine of type input patrols image data and the accident analysis data as the output of initial machine learning model;
The net grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains described
The target component of initial machine learning model;It is related that the target component to multiple provincial machines patrols data training sample;
The target component is distributed to each provincial server by the net grade server, to indicate each provincial server by institute
It states target component and substitutes into each provincial machine learning submodel, each provincial machine learning submodel after being upgraded.
In one embodiment, a kind of readable storage medium storing program for executing is provided, computer program, computer program are stored thereon with
It is performed the steps of when being executed by processor
Net grade server obtains the parameter for each provincial machine learning submodel that each provincial server is sent;The each province
The parameter of the machine learning submodel of grade is that each provincial server patrols data training sample according to each provincial machine respectively, to default
Initial machine learning model be trained;It includes: to learn mould as initial machine that the machine, which patrols data training sample,
The machine of type input patrols image data and the accident analysis data as the output of initial machine learning model;
The net grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains described
The target component of initial machine learning model;It is related that the target component to multiple provincial machines patrols data training sample;
The target component is distributed to each provincial server by the net grade server, to indicate each provincial server by institute
It states target component and substitutes into each provincial machine learning submodel, each provincial machine learning submodel after being upgraded.
In one embodiment, a kind of provincial server, including memory and processor are provided, is stored in memory
Computer program, the processor perform the steps of when executing computer program
Provincial server patrols data training sample according to provincial machine, instructs to preset initial machine study submodel
Practice, obtains the parameter of the provincial machine learning submodel;It includes: as initial machine that the machine, which patrols data training sample,
The machine for practising mode input patrols image data and the accident analysis data as the output of initial machine learning model;
The parameter of the machine learning submodel is uploaded to net grade server by the provincial server, to indicate the net
Grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains the initial machine learning model
Target component;It is related that the target component to multiple provincial machines patrols data training sample;
The provincial server receives the target component of net grade server distribution, and by the target component generation
Enter the provincial machine learning submodel, the machine learning submodel after being upgraded.
The internal structure chart of above-mentioned provincial server is referred to the description of above-mentioned net grade server, and which is not described herein again.
In one embodiment, a kind of readable storage medium storing program for executing is provided, computer program, computer program are stored thereon with
It is performed the steps of when being executed by processor
Provincial server patrols data training sample according to provincial machine, instructs to preset initial machine study submodel
Practice, obtains the parameter of the provincial machine learning submodel;It includes: as initial machine that the machine, which patrols data training sample,
The machine for practising mode input patrols image data and the accident analysis data as the output of initial machine learning model;
The parameter of the machine learning submodel is uploaded to net grade server by the provincial server, to indicate the net
Grade server carries out Combined Treatment to the parameter of each provincial machine learning submodel, obtains the initial machine learning model
Target component;It is related that the target component to multiple provincial machines patrols data training sample;
The provincial server receives the target component of net grade server distribution, and by the target component generation
Enter the provincial machine learning submodel, the machine learning submodel after being upgraded.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein,
To any reference of memory, storage, database or other media used in each embodiment provided herein,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of above embodiments can be combined arbitrarily, for simplicity of description, not to above-described embodiment
In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance
Shield all should be considered as described in this specification.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the concept of this application, various modifications and improvements can be made, these belong to the protection of the application
Range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.