CN109670600A - Decision-making technique and system based on cloud platform - Google Patents
Decision-making technique and system based on cloud platform Download PDFInfo
- Publication number
- CN109670600A CN109670600A CN201811536784.4A CN201811536784A CN109670600A CN 109670600 A CN109670600 A CN 109670600A CN 201811536784 A CN201811536784 A CN 201811536784A CN 109670600 A CN109670600 A CN 109670600A
- Authority
- CN
- China
- Prior art keywords
- engine
- initial model
- decision
- simulated
- environment
- 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.)
- Granted
Links
Abstract
The invention belongs to field of artificial intelligence, it discloses a kind of decision-making technique and system based on cloud platform.Method includes receiving simulated training request；Configure hardware resource, initialization simulation engine, decision engine and study engine；Decision engine obtains acting on the instruction in simulated environment according to the simulated environment state and initial model that are generated by simulation engine, and circulation executes repeatedly, obtains simulated training collection；Learn engine according to simulated training collection training initial model, obtains updated initial model；Terminal device initialization requests are sent, terminal device initialization requests include hardware resource and updated initial model, and are used to make terminal equipment configuration hardware resource and act on the instruction of true environment according to the state output of true environment.System includes the first dispatching device, simulation engine, decision engine, study engine and the second dispatching device.The present invention forms complete decision closed loop through the above scheme, has reached and has met expected decision purpose.
Description
Technical field
The invention belongs to field of artificial intelligence, in particular to a kind of decision-making technique and system based on cloud platform.
Background technique
The rule that existing decision system is defined based on the mankind can not be adjusted according to the variation of environment.
Existing intelligence system is based only on artificial intelligence, can not calibrate to the result of machine learning.Also, it is most of
Intelligence system concentrates on study side, can not provide decision and support end to end.There are also a small number of systems end to end to be in scientific research
Stage can not adapt to the production environment that extensive and high availability requires, also not to various deployment forms and study end or certainly
Plan end equipment provides support.
Summary of the invention
To solve the above-mentioned problems, one aspect of the present invention provides a kind of decision-making technique based on cloud platform comprising: it connects
It receives and receives simulated training request, simulated training request includes: hardware resource, true environment, initial model and nitrification enhancement；
Configure the hardware resource, initialization simulation engine, decision engine and study engine；The decision engine is according to by the simulation
The simulated environment state and the initial model that engine generates are instructed, which acts in the simulated environment, circulation
It executes repeatedly, obtains simulated training collection, it includes: the state and instruction of the simulated environment that each sample is concentrated in the simulated training；
The study engine obtains updated initial model according to the simulated training collection training initial model；Send terminal
Equipment initialization requests, the terminal device initialization requests include: hardware resource and updated initial model, and for making
Terminal equipment configuration hardware resource and the instruction that true environment is acted on according to the state output of true environment.
In decision-making technique as described above, it is preferable that the simulated training request further include: rule；Accordingly, described
Decision engine is instructed according to the simulated environment state, initial model and the rule that are generated by the simulation engine；Accordingly, institute
Stating terminal device initialization requests includes: hardware resource, updated initial model and rule.
In decision-making technique as described above, it is preferable that the decision engine is according to the mould generated by the simulation engine
Quasi- ambient condition, initial model and rule are instructed, and are specifically included: the decision engine first given birth to by the simulation engine by basis
At simulated environment state and initial model obtain initial order, further according to described in the rules modification modify instruct, will correct
As a result the instruction as output；Or the decision engine first judges that the state of the simulated environment is to meet model prediction, still
Meet regular prediction, meet model prediction if being judged as, is referred to according to the state and initial model that are generated by simulated environment
It enables, meets regular prediction if being judged as, instructed according to the state and rule that are generated by simulated environment.
On the other hand a kind of decision system based on cloud platform is provided comprising: the first scheduler subsystem, for receiving
Simulated training request, simulated training request includes: hardware resource, true environment, initial model and nitrification enhancement, and is configured
Hardware resource and initialization simulation engine, decision engine and study engine；The simulation engine is for generating simulated environment with mould
Quasi- true environment；The decision engine is described for being instructed according to the state and the initial model of the simulated environment
In the simulated environment, circulation executes repeatedly instruction execution, obtains simulated training collection；The study engine is used for according to
The simulated training collection training initial model, obtains updated initial model；Second dispatching device is for sending terminal
Equipment initialization requests, the terminal device initialization requests include: hardware resource and updated initial model and are used to make
Terminal equipment configuration hardware resource and the instruction that true environment is acted on according to the state output of true environment.
In decision system as described above, it is preferable that the received simulated training request of the first scheduler subsystem is also
It include: rule；Accordingly, the decision engine is used for according to simulated environment state, the introductory die generated by the simulation engine
Type and rule are instructed；After the terminal device initialization requests that second dispatching device is sent include: hardware resource, update
Initial model and rule.
In decision system as described above, it is preferable that the decision engine is used for basis and is generated by the simulation engine
Simulated environment state, initial model and rule instructed, be specifically used for: the decision engine is first used for according to by the mould
The simulated environment state and initial model that quasi- engine generates obtain initial order, refer to further according to modification described in the rules modification
It enables, using correction result as the instruction of output；Or the decision engine is for first judging that the state of the simulated environment is to meet
Model prediction still meets regular prediction, meets model prediction if being judged as, according to the state generated by simulated environment and just
Beginning model is instructed, and meets regular prediction if being judged as, is instructed according to the state and rule that are generated by simulated environment.
Another aspect of the invention provides a kind of decision-making technique based on cloud platform comprising: receive simulated training
Request, simulated training request includes: hardware resource, true environment, initial model and nitrification enhancement；Configure the hardware money
Source, initialization simulation engine, decision engine and study engine；The decision engine is according to the simulation generated by the simulation engine
Ambient condition and the initial model are instructed, which acts in the simulated environment, and circulation executes repeatedly, obtain mould
Quasi- training set, it includes: the state and instruction of the simulated environment that each sample is concentrated in the simulated training；The study engine according to
The simulated training collection training initial model, obtains updated initial model；Send terminal device initialization requests, institute
Stating terminal device initialization requests includes: hardware resource；Hardware resource is configured, institute is returned to the true environment state that will acquire
It states decision engine and executes the instruction of the decision engine output in true environment；Circulation executes repeatedly in true environment,
True training set is obtained, each sample includes: the state and instruction of true environment in the true training set；The study engine root
According to the updated initial model of the true training set training, updated initial model again is obtained；At the beginning of sending terminal device
Beginningization request, terminal device initialization requests include: hardware resource and updated initial model again, and are used to that terminal to be made to set
Standby configuration hardware resource and the instruction that true environment is acted on according to the state output of true environment.
Further aspect of the present invention provides a kind of decision system based on cloud platform comprising: the first scheduler subsystem is used
In receiving simulated training request, the simulated training request includes: hardware resource, true environment, initial model and intensified learning
Algorithm, and configure hardware resource and initialization simulation engine, decision engine and study engine；The simulation engine is for generating mould
Near-ring border is to simulate true environment；The decision engine according to the state and the initial model of the simulated environment for obtaining
The instruction in the simulated environment is acted on, circulation executes repeatedly, obtains simulated training collection；The study engine is used for according to institute
The simulated training collection training initial model is stated, updated initial model is obtained；Second dispatching device is for sending end
End equipment initialization requests, the terminal device initialization requests include: hardware resource so that terminal equipment configuration hardware resource
And the true environment state that will acquire returns to the decision engine and executes the decision engine output in true environment
Instruction；The decision engine is also used to be instructed according to updated initial model and the state output of true environment, in true ring
Circulation executes repeatedly in border, obtains true training set, each sample includes: the state of true environment and refers in the true training set
It enables；The study engine is also used to be obtained updated again according to the updated initial model of the true training set training
Initial model；Second dispatching device is also used to send terminal device initialization requests, the terminal device initialization requests
It include: hardware resource and updated initial model again, and for making terminal equipment configuration hardware resource and according to true ring
The state output in border acts on the instruction of true environment.
Bring of the embodiment of the present invention has the beneficial effect that:
Complete decision closed loop is formd, has reached and has met expected decision purpose.Based on cloud technology, public affairs can be provided
There are cloud, private clound, mixed cloud etc. to dispose form, while supporting various heterogeneous terminals equipment.The primary production scale for supporting super large,
Meet requirement of the production environment to availability simultaneously.
Detailed description of the invention
Fig. 1 is a kind of flow diagram for decision-making technique based on cloud platform that one embodiment of the invention provides；
Fig. 2 be another embodiment of the present invention provides a kind of decision-making technique based on cloud platform flow diagram；
Fig. 3 is a kind of structural schematic diagram for decision system based on cloud platform that one embodiment of the invention provides.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to embodiment party of the present invention
Formula is described in further detail.
Referring to Fig. 1, the embodiment of the invention provides a kind of decision-making techniques based on cloud platform comprising:
Step 101, simulated training request is received, simulated training request includes: hardware resource, true environment, initial model
And nitrification enhancement.
Specifically, simulated training request can be sent by web console or terminal device by user, can also passed through
The script finished in advance sends simulated training request.The request includes: hardware resource, true environment, initial model and extensive chemical
Practise algorithm.Hardware resource is user based on the hardware configuration for needing service scale to be offered and selecting comprising but be not limited to:
CPU quantity, amount of memory, GPU quantity, disk size, machine quantity.True environment is to apply ring when executing this decision-making technique
Border (or scene), such as 1) city brain field: traffic, electric power, communication, the energy.2) sphere of life: smart home, intelligence
Assistant's reason, automatic Pilot, intelligent translation etc..3) entertainment field: game, music synthesis, dance & art sound effect control, intelligent body performance
(such as unmanned aerial vehicle group).4) industrial circle: Mechanical course etc..5) information industry field: recommendation and searching order etc..6) education neck
Domain.7) military field.Due to directlying adopt Live Environmental Training, economic cost and time cost are very big, influence efficiency, therefore
It needs to generate simulated environment according to true environment.Initial model can be the model that user voluntarily takes in, and can also be user's root
The model provided according to the cloud platform that the actual demand of oneself selects.The initial model is used for defeated according to following simulated environment states
It instructs out, which acts on simulated environment.Preferably, initial model is neural network framework.Nitrification enhancement is training
Algorithm used when the initial model.Cloud platform can be the deployment form such as public cloud, private clound, mixed cloud.
Step 102, hardware resource, initialization simulation engine, decision engine and study engine are configured.
Hardware resource is configured according to the demand that the received simulated training request of step 101 is user, such as configures entity services
Device or cluster resource or Cloud Server example resource.Simulation engine generates simulated environment according to true environment, that is, completes environment
Initialization.In practical application, the implementation of simulation engine, which can be, uses illusory series, unity game engine etc.；It can be with
It is using some modeling tools, such as: Simulink；For bigger scene, it is also possible to using distributed simulation engine,
Such as: SpatialOS and KBEngine.Decision engine is completed to initialize according to initial model.Study engine is calculated according to intensified learning
Method completes initialization.In practical application, learn engine implementation can by TensorFlow, Pytorch, MXNet,
Caffee, MPI, Parameter Server etc. are realized.
Step 103, decision engine is instructed according to the state and initial model of simulated environment, and the instruction execution is in simulation
In environment, circulation is executed repeatedly, obtains simulated training collection, simulated training concentrates each sample to include: the state of simulated environment and refer to
It enables.
Decision engine is made prediction using state of the initial model to received simulated environment, is instructed, which makees
For simulated environment, the state of simulated environment can be made to change therewith, then repeat the step, largely simulated
Training sample (or being simulated environment sample) under environment, each sample includes: the state of simulated environment, instruction, by a large amount of samples
This collection is collectively referred to as simulated training collection.Duplicate number can be determined that such as duplicate number is reached by preset frequency threshold value
To frequency threshold value, then stop executing the step, duplicate number is number of repetition to be determined at this time；It can also be by presetting
Duration threshold value determine, such as start the clock from receiving request, after repeating n times, decision duration reaches duration threshold value, then
Stop executing the step, N is number of repetition to be determined at this time, and the present embodiment is to this without limiting.
Step 104, study engine obtains updated initial model according to simulated training collection training initial model.
Learn engine and be based on nitrification enhancement, using analog sample collection training initial model, reaches convergence or reach pre-
If after time threshold, updating the parameter of initial model, the updated model of parameter is referred to as updated initial model, by it
Output is to decision engine, and decision engine is as decision model.
Step 105, terminal device initialization requests are sent, terminal device initialization requests include: hardware resource and update
Initial model afterwards, the request are acted on for making terminal equipment configuration hardware resource, and according to the state output of true environment
The instruction of true environment.
Decision model of the updated initial model as terminal device, terminal device utilize its state to true environment
It makes a policy, the decision is for making terminal device execute corresponding behavior in true environment.Terminal device is that a variety of isomeries are whole
One of end equipment, can be with are as follows: mobile phone, tablet computer, mobile phone, automobile and Intelligent hardware so support that isomery is whole
Indifference deployment and operation are realized, using convenient in end.Since decision model is deployed on terminal device, can terminal be set
It is standby to make a policy in real time, it is not weak by signal of communication between terminal device and cloud platform or transmission delay is influenced.At it
In his embodiment, based on the considerations of terminal device hardware resource limiting factor, decision model can be deployed in cloud platform, at this time
Terminal device initialization requests include: hardware resource, and the state for the true environment that terminal device will acquire returns to decision engine,
Instruction is returned to terminal device according to the state of true environment by decision engine, and terminal device executes this in true environment and refers to
It enables.
In order to further improve the foundation of prediction (or output order), simulated training request further include: rule passes through at this time
Using the combination output order of rule and initial model, achieve the purpose that hybrid predicting, rule is preset condition.Practical application
In, according to different application scenarios, rule can be adaptively adjusted.The method of hybrid predicting may is that first with first
Beginning model is made prediction result (i.e. output order) according to the state of simulated environment, then corrects prediction result according to rule.Also
It may is that and judge that the state of simulated environment is to meet model prediction, still meet regular prediction, to meet model pre- if being judged as
It surveys, is then handled to obtain prediction result using state of the initial model to simulated environment；Meet regular prediction if being judged as,
It is handled using state of the rule to simulated environment, obtains prediction result.It may is that again and tear the state of simulated environment open
Point, multiple sub- states are resolved into, the method predicted every sub- state can be any one of aforementioned two methods.
The embodiment of the present invention forms complete decision closed loop through the above scheme, has reached and has met expected decision mesh
's.Based on cloud technology, the deployment form such as public cloud, private clound, mixed cloud can be provided, while various heterogeneous terminals being supported to set
It is standby.The primary production scale for supporting super large, while meeting requirement of the production environment to availability.
Referring to fig. 2, in order to further complete decision model, another embodiment of the present invention provides a kind of based on cloud platform
Decision-making technique comprising following steps:
Step 201, simulated training request is received, simulated training request includes: hardware resource, true environment, initial model
And nitrification enhancement.
Step 202, hardware resource, initialization simulation engine, decision engine and study engine are configured.Cloud platform configures hardware
Resource.
Step 203, decision engine obtains the finger executed in simulated environment according to the state and initial model of simulated environment
It enables, circulation executes repeatedly, obtains simulated training collection, it includes: the state and instruction of simulated environment that each sample is concentrated in simulated training.
Step 204, study engine obtains updated initial model according to simulated training collection training initial model.
Step 205, terminal device initialization requests are sent, terminal device initialization requests include: hardware resource.Terminal is set
Standby configuration hardware resource.
Step 206, hardware resource is configured, decision engine is returned to and in true environment with the true environment state that will acquire
The middle instruction for executing decision engine output.
Step 207, circulation executes repeatedly in true environment, obtains true training set, each sample packet in true training set
It includes: the state and instruction of true environment.
Step 208, study engine obtains updated first again according to the updated initial model of true training set training
Beginning model.Decision model of the updated initial model as terminal device again.
Step 209, terminal device initialization requests are sent, terminal device initialization requests include: hardware resource and decision
Model, the request act on true environment for making terminal equipment configuration hardware resource, and according to the state output of true environment
Instruction.
Associated description about step 201~209 can be found in the related content of step 101~105 of above-described embodiment, this
Place no longer repeats one by one.
The embodiment of the present invention forms complete decision closed loop through the above scheme, has reached and has met expected decision mesh
's.Based on cloud technology, the deployment form such as public cloud, private clound, mixed cloud can be provided, while various heterogeneous terminals being supported to set
It is standby.The primary production scale for supporting super large, while meeting requirement of the production environment to availability.
Referring to Fig. 3, the embodiment of the invention provides a kind of decision systems based on cloud platform, are used to execute above-mentioned implementation
The method of example comprising: the first dispatching device 10, simulation engine 20, decision engine 30, study engine 40 and the second dispatching device
50。
Wherein, for the first dispatching device 10 for receiving simulated training request, it includes: hardware resource that simulated training, which is requested, true
Real environment, initial model and nitrification enhancement, and configure hardware resource and initialize simulation engine, decision engine and study and draw
It holds up.Simulation engine 20 is for generating simulated environment to simulate true environment.Decision engine 30 is used for the state according to simulated environment
It is instructed with initial model, in simulated environment, circulation executes repeatedly the instruction execution, obtains simulated training collection.Study is drawn
40 are held up according to simulated training collection training initial model, obtains updated initial model.Second dispatching device 50 is for sending end
End equipment initialization requests, terminal device initialization requests include: hardware resource and updated initial model, which is used for
Make terminal equipment configuration hardware resource, and acts on the instruction of true environment according to the state output of true environment.
The embodiment of the present invention forms complete decision closed loop through the above scheme, has reached and has met expected decision mesh
's.Based on cloud technology, the deployment form such as public cloud, private clound, mixed cloud can be provided, while various heterogeneous terminals being supported to set
It is standby.The primary production scale for supporting super large, while meeting requirement of the production environment to availability.
The embodiment of the invention provides a kind of decision systems based on cloud platform, are used to execute the side of above-described embodiment
Method comprising: the first dispatching device 10, simulation engine 20, decision engine 30, study engine 40 and the second dispatching device 50.
Wherein, for the first dispatching device 10 for receiving simulated training request, it includes: hardware resource that simulated training, which is requested, true
Real environment, initial model and nitrification enhancement, and configure hardware resource and initialize simulation engine, decision engine and study and draw
It holds up.Simulation engine 20 is for generating simulated environment to simulate true environment.Decision engine 30 is used for the state according to simulated environment
It is instructed with initial model, which executes in simulated environment, and decision engine circulation executes repeatedly, obtains simulated training
Collection.Learn engine 40 to be used to obtain updated initial model according to simulated training collection training initial model.Second dispatching device
50 for sending terminal device initialization requests, and terminal device initialization requests include: hardware resource.The request is for making terminal
Device configuration hardware resource and the true environment state that will acquire return to decision engine 30 and execute decision in true environment
The instruction that engine 30 exports.Decision engine 30 is also used to be instructed according to true environment state output, and the circulation of decision engine 30 executes
Repeatedly, true training set is obtained, each sample includes: the state and instruction of true environment in true training set.Learn engine 40 also
For obtaining updated initial model again according to the updated initial model of true training set training.Second dispatching device
50 are also used to send terminal device initialization requests, and terminal device initialization requests include: hardware resource and updated again
Initial model, the request act on really for making terminal equipment configuration hardware resource, and according to the state output of true environment
The instruction of environment.
The embodiment of the present invention forms complete decision closed loop through the above scheme, has reached and has met expected decision mesh
's.Based on cloud technology, the deployment form such as public cloud, private clound, mixed cloud can be provided, while various heterogeneous terminals being supported to set
It is standby.The primary production scale for supporting super large, while meeting requirement of the production environment to availability.
It should be understood that decision system provided by the above embodiment is in decision, only drawing with above-mentioned each functional module
Divide and be illustrated, in practical application, can according to need and be completed by different functional modules above-mentioned function distribution, i.e.,
The internal structure of system is divided into different functional modules, to complete all or part of the functions described above, such as by
One dispatching device and the second dispatching device are divided into adjustment device.In addition, decision system provided by the above embodiment and decision-making party
It is owned by France in same design, specific implementation process is detailed in embodiment of the method, and which is not described herein again.
As known by the technical knowledge, the present invention can pass through the embodiment party of other essence without departing from its spirit or essential feature
Case is realized.Therefore, embodiment disclosed above, in all respects are merely illustrative, not the only.Institute
Have within the scope of the present invention or is included in the invention in the change being equal in the scope of the present invention.
Claims (8)
1. a kind of decision-making technique based on cloud platform, which is characterized in that the decision-making technique includes:
Simulated training request is received, simulated training request includes: hardware resource, true environment, initial model and extensive chemical
Practise algorithm；
Configure the hardware resource, initialization simulation engine, decision engine and study engine；
The decision engine is acted on according to the simulated environment state and the initial model that are generated by the simulation engine
Instruction in the simulated environment, circulation execute repeatedly, obtain simulated training collection, and each sample of the simulated training concentration includes:
The state and instruction of the simulated environment；
The study engine obtains updated initial model according to the simulated training collection training initial model；
Send terminal device initialization requests, the terminal device initialization requests include: hardware resource and updated initial
Model, and be used to make terminal equipment configuration hardware resource and act on the finger of true environment according to the state output of true environment
It enables.
2. decision-making technique according to claim 1, which is characterized in that the simulated training request further include: rule；
Accordingly, the decision engine according to generated by the simulation engine simulated environment state, initial model and regular
To instruction；
Accordingly, the terminal device initialization requests include: hardware resource, updated initial model and rule.
3. decision-making technique according to claim 2, which is characterized in that the decision engine is given birth to according to by the simulation engine
At simulated environment state, initial model and rule instructed, specifically include:
The decision engine first obtains initial order according to the simulated environment state and initial model that are generated by the simulation engine,
Further according to instruction is modified described in the rules modification, using correction result as the instruction of output；Or
The decision engine first judges that the state of the simulated environment is to meet model prediction, still meets regular prediction, if sentencing
Break to meet model prediction, is then instructed according to the state and initial model that are generated by simulated environment, meet rule if being judged as
It then predicts, is then instructed according to the state and rule that are generated by simulated environment.
4. a kind of decision system based on cloud platform, which is characterized in that the decision system includes:
First dispatching device, for receiving simulated training request, simulated training request includes: hardware resource, true environment, initial
Model and nitrification enhancement, and configure hardware resource and initialization simulation engine, decision engine and study engine；
The simulation engine is for generating simulated environment to simulate true environment；
The decision engine according to the state and the initial model of the simulated environment for obtaining acting on the analog loop
Instruction in border, circulation execute repeatedly, obtain simulated training collection；
The study engine is used to obtain updated initial model according to the simulated training collection training initial model；
For second dispatching device for sending terminal device initialization requests, the terminal device initialization requests include: hard
Part resource and updated initial model and for making terminal equipment configuration hardware resource and the state output according to true environment
Act on the instruction of true environment.
5. decision system according to claim 4, which is characterized in that the received simulated training of the first scheduler subsystem
Request further include: rule；
Accordingly, the decision engine is used for according to simulated environment state, initial model and the rule generated by the simulation engine
Then instructed；
The terminal device initialization requests that second dispatching device is sent include: hardware resource, updated initial model and
Rule.
6. decision system according to claim 5, which is characterized in that the decision engine is used to draw according to by the simulation
Simulated environment state, initial model and the rule for holding up generation are instructed, and are specifically used for:
The decision engine is first used to be obtained initially according to the simulated environment state and initial model that are generated by the simulation engine
Instruction is instructed further according to modifying described in the rules modification, using correction result as the instruction of output；Or
The decision engine still meets regular prediction for first judging that the state of the simulated environment is to meet model prediction,
Meet model prediction if being judged as, is instructed according to the state and initial model that are generated by simulated environment, if being judged as symbol
It normally predicts, is then instructed according to the state and rule that are generated by simulated environment.
7. a kind of decision-making technique based on cloud platform, which is characterized in that the decision-making technique includes:
Simulated training request is received, simulated training request includes: that hardware resource, true environment, initial model and intensified learning are calculated
Method；
Configure the hardware resource, initialization simulation engine, decision engine and study engine；
The decision engine is acted on according to the simulated environment state and the initial model that are generated by the simulation engine
Instruction in the simulated environment, circulation execute repeatedly, obtain simulated training collection, and each sample of the simulated training concentration includes:
The state and instruction of the simulated environment；
The study engine obtains updated initial model according to the simulated training collection training initial model；
Terminal device initialization requests are sent, the terminal device initialization requests include: hardware resource；
Hardware resource is configured, the decision engine is returned to the true environment state that will acquire and executes institute in true environment
State the instruction of decision engine output；
Circulation executes repeatedly in true environment, obtains true training set, each sample includes: true ring in the true training set
The state and instruction in border；
The study engine obtains updated introductory die again according to the updated initial model of the true training set training
Type；
Terminal device initialization requests are sent, the terminal device initialization requests include: hardware resource and updated again
Initial model, and for making terminal equipment configuration hardware resource and acting on true environment according to the state output of true environment
Instruction.
8. a kind of decision system based on cloud platform, which is characterized in that the decision system includes:
First scheduler subsystem, for receiving simulated training request, the simulated training request includes: hardware resource, true ring
Border, initial model and nitrification enhancement, and configure hardware resource and initialization simulation engine, decision engine and study engine；
The simulation engine is for generating simulated environment to simulate true environment；
The decision engine according to the state and the initial model of the simulated environment for obtaining acting on the analog loop
Instruction in border, circulation execute repeatedly, obtain simulated training collection；
The study engine is used to obtain updated initial model according to the simulated training collection training initial model；
For second dispatching device for sending terminal device initialization requests, the terminal device initialization requests include: hard
Part resource is so that the terminal equipment configuration hardware resource and true environment state that will acquire returns to the decision engine and true
The instruction of the decision engine output is executed in real environment；
The decision engine is also used to be instructed according to updated initial model and the state output of true environment, in true environment
Middle circulation executes repeatedly, obtains true training set, each sample includes: the state of true environment and refers in the true training set
It enables；
The study engine is also used to be obtained updated again according to the updated initial model of the true training set training
Initial model；
Second dispatching device is also used to send terminal device initialization requests, and the terminal device initialization requests include:
Hardware resource and again updated initial model, and for making terminal equipment configuration hardware resource and the shape according to true environment
State output action is in the instruction of true environment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811536784.4A CN109670600B (en) | 2018-12-14 | 2018-12-14 | Decision-making technique and system based on cloud platform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811536784.4A CN109670600B (en) | 2018-12-14 | 2018-12-14 | Decision-making technique and system based on cloud platform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109670600A true CN109670600A (en) | 2019-04-23 |
| CN109670600B CN109670600B (en) | 2019-10-25 |
Family
ID=66144807
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811536784.4A Active CN109670600B (en) | 2018-12-14 | 2018-12-14 | Decision-making technique and system based on cloud platform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109670600B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111882072A (en) * | 2020-07-09 | 2020-11-03 | 北京华如科技股份有限公司 | Intelligent model automatic course training method for playing chess with rules |
| CN112667594A (en) * | 2021-01-14 | 2021-04-16 | 北京智源人工智能研究院 | Heterogeneous computing platform based on hybrid cloud resources and model training method |
| CN112766508A (en) * | 2021-04-12 | 2021-05-07 | 北京一流科技有限公司 | Distributed data processing system and method thereof |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170109667A1 (en) * | 2011-05-08 | 2017-04-20 | Panaya Ltd. | Automaton-Based Identification of Executions of a Business Process |
| CN106600668A (en) * | 2016-12-12 | 2017-04-26 | 中国科学院自动化研究所 | Animation generation method used for carrying out interaction with virtual role, apparatus and electronic equipment |
| CN107506830A (en) * | 2017-06-20 | 2017-12-22 | 同济大学 | Towards the artificial intelligence training platform of intelligent automobile programmed decision-making module |
| CN108427992A (en) * | 2018-03-16 | 2018-08-21 | 济南飞象信息科技有限公司 | A kind of machine learning training system and method based on edge cloud computing |
-
2018
- 2018-12-14 CN CN201811536784.4A patent/CN109670600B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170109667A1 (en) * | 2011-05-08 | 2017-04-20 | Panaya Ltd. | Automaton-Based Identification of Executions of a Business Process |
| CN106600668A (en) * | 2016-12-12 | 2017-04-26 | 中国科学院自动化研究所 | Animation generation method used for carrying out interaction with virtual role, apparatus and electronic equipment |
| CN107506830A (en) * | 2017-06-20 | 2017-12-22 | 同济大学 | Towards the artificial intelligence training platform of intelligent automobile programmed decision-making module |
| CN108427992A (en) * | 2018-03-16 | 2018-08-21 | 济南飞象信息科技有限公司 | A kind of machine learning training system and method based on edge cloud computing |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111882072A (en) * | 2020-07-09 | 2020-11-03 | 北京华如科技股份有限公司 | Intelligent model automatic course training method for playing chess with rules |
| CN111882072B (en) * | 2020-07-09 | 2023-11-14 | 北京华如科技股份有限公司 | Intelligent model automatic course training method for playing chess with rules |
| CN112667594A (en) * | 2021-01-14 | 2021-04-16 | 北京智源人工智能研究院 | Heterogeneous computing platform based on hybrid cloud resources and model training method |
| CN112766508A (en) * | 2021-04-12 | 2021-05-07 | 北京一流科技有限公司 | Distributed data processing system and method thereof |
| CN112766508B (en) * | 2021-04-12 | 2022-04-08 | 北京一流科技有限公司 | Distributed data processing system and method thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109670600B (en) | 2019-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109670600B (en) | Decision-making technique and system based on cloud platform | |
| Chen et al. | DNNOff: offloading DNN-based intelligent IoT applications in mobile edge computing | |
| WO2020259502A1 (en) | Method and device for generating neural network model, and computer-readable storage medium | |
| CN110310628B (en) | Method, device and equipment for optimizing wake-up model and storage medium | |
| Chen et al. | An adaptive offloading framework for android applications in mobile edge computing | |
| Song et al. | Networking systems of AI: On the convergence of computing and communications | |
| CN108961888B (en) | Personalized reconstruction knowledge system learning system and method | |
| Liu et al. | The SOLIDS 6G mobile network architecture: driving forces, features, and functional topology | |
| CN110428046A (en) | Acquisition methods and device, the storage medium of neural network structure | |
| Guo et al. | An energy sensitive computation offloading strategy in cloud robotic network based on ga | |
| Chen et al. | Deep-broad learning system for traffic flow prediction toward 5G cellular wireless network | |
| CN113778871A (en) | Mock testing method, device, equipment and storage medium | |
| CN115277689A (en) | Yun Bianwang network communication optimization method and system based on distributed federal learning | |
| Peng et al. | A cloud load forecasting model with nonlinear changes using whale optimization algorithm hybrid strategy | |
| Boutekkouk | Embedded systems codesign under artificial intelligence perspective: a review | |
| Han et al. | The method of allocating resources for ideological and political education in universities based on iot technology | |
| CN110012021B (en) | Self-adaptive computing migration method under mobile edge computing | |
| CN112199154A (en) | Distributed collaborative sampling central optimization-based reinforcement learning training system and method | |
| CN111368998A (en) | Spark cluster-based model training method, device, equipment and storage medium | |
| WO2023143080A1 (en) | Data processing method and related device | |
| JP2022136234A (en) | Federated learning method and apparatus, electronic apparatus, storage medium, and computer program | |
| Kölsch et al. | Co-Simulating the Internet of Things in a Smart Grid use case scenario | |
| Santos et al. | Power systems simulation using ontologies to enable the interoperability of multi-agent systems | |
| CN105357243B (en) | A kind of method and apparatus with internet of things equipment remote interaction | |
| CN114564882A (en) | Construction and application of edge deep learning simulator based on discrete events |
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 | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20190423 Assignee: Baiyang Times (Beijing) Technology Co., Ltd. Assignor: Qiyuan World (Beijing) Information Technology Service Co., Ltd. Contract record no.: X2019990000075 Denomination of invention: A cloud platform-based decision-making method and system License type: Common License Record date: 20190904 |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |