CN112131782A - Multi-loop intelligent factory edge side digital twin scene coupling device - Google Patents

Multi-loop intelligent factory edge side digital twin scene coupling device Download PDF

Info

Publication number
CN112131782A
CN112131782A CN202010881839.6A CN202010881839A CN112131782A CN 112131782 A CN112131782 A CN 112131782A CN 202010881839 A CN202010881839 A CN 202010881839A CN 112131782 A CN112131782 A CN 112131782A
Authority
CN
China
Prior art keywords
scene
production
simulation
module
data
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
Application number
CN202010881839.6A
Other languages
Chinese (zh)
Other versions
CN112131782B (en
Inventor
金炫智
应天裕
冯毅萍
潘戈
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202010881839.6A priority Critical patent/CN112131782B/en
Publication of CN112131782A publication Critical patent/CN112131782A/en
Application granted granted Critical
Publication of CN112131782B publication Critical patent/CN112131782B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Geometry (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a multi-loop intelligent factory edge side digital twin scene coupling device, which comprises a virtual space and a physical space of a production line edge side digital twin system, a decision support tool, a digitization tool and an online dynamic scene database, wherein the virtual space and the physical space form a plurality of loops; the virtual space comprises a Conditional WGAN dynamic scene generation method, a scene generator and a simulation model for generating an antagonistic network based on Wasserstein; the physical space comprises a physical system; the circuit comprises: a digital twin simulation decision loop, a scene simulation loop and a dynamic scene generation loop. The coupling device can be embedded into an intelligent algorithm system at the edge side of an intelligent factory as a service for coupling generation of a digital twin production scene, and can also provide reliable original data and training samples for other experiments and researches to carry out targeted research.

Description

Multi-loop intelligent factory edge side digital twin scene coupling device
Technical Field
The invention relates to the field of edge-side digital twin of an intelligent factory production line, in particular to a multi-loop intelligent factory edge-side digital twin scene coupling device.
Background
In the large context of industrial 4.0 and intelligent plants, production enterprises want to simulate and analyze data in industrial production processes and achieve fault early warning and optimization of process parameters by using digital twin technology. With the gradual improvement of a technical equipment system And a communication standard system of an automation Control system of a large project, a modern digital factory can record mass production Data through a Data Acquisition And monitoring Control System (SCADA), a Human Machine Interface (HMI) And the like in the production process.
Any data collection and transmission is expensive, and although the development of information electronic technology has prompted the emergence of a large number of intelligent and high-precision micro sensors, the development of internet of things technology, industrial internet technology and edge-side computing has increased the scale of data transmission and reduced the cost of data transmission, data collection under limited time and limited resources still remains a huge problem in the industry.
In the actual production data acquisition process, conditions occurring in the production process need to be analyzed according to different types of data, and the specific result of the optimized scene is verified through simulation, so that the benefit of the actual production process is improved, and the faults caused in the production process are reduced. The static scene sample can well simulate the distribution characteristic and the coupling characteristic of the scene characteristic in the production process, the dynamic scene sample visually shows the change of the scene variable along with the time, and the time distribution rule is recorded.
Chinese patent publication No. CN110471279A discloses a method for generating static scenes in industrial production based on vine-copulas, which comprises the following steps: the method comprises the steps of collecting existing scene data from an initial scene database module, preprocessing the scene data through a scene clustering module, a feature standardization module, a feature dimension reduction module and the like, constructing a scene generation model through a model fitting module, a goodness inspection module and a binary copula model database module by adopting a vine copula method, and generating a required scene through a scene sampling module and a reflection module.
Shishuihua is reported in the literature "intelligent manufacturing, digital twin and strategic scene modeling" (journal of Beijing university of transportation (social science edition) 2019, volume 18, No. 2, pages 69-77), and indicates that in the digital twin, the construction of the scene can more conveniently simulate or reconstruct the complex deduction process of enterprise strategic decisions, thereby enriching and deepening the comprehensiveness and uncertainty cognition of the enterprise strategic process.
Chinese patent publication No. CN111027195A discloses a simulation scene generation method for the field of intelligent driving, which includes the following steps: acquiring sensing data of a plurality of moments acquired by an acquisition device in a preset scene, wherein the preset scene comprises at least one obstacle; converting the sensing data into structured data, wherein the structured data comprises track information of each obstacle, and the track information of one obstacle comprises running state information of the obstacle at each moment; and adjusting the track information of at least one obstacle in the structured data to obtain the structured data corresponding to the simulation scene.
Chinese patent publication No. CN110998663A discloses an image generation method of a simulation scene, which includes the following steps: obtaining semantic segmentation information and example segmentation information of scene white mould; receiving instance text information of a scene white mould; the example text information is editable information and is used for describing the attribute of the example; and generating an image of the simulation scene based on the semantic segmentation information, the example text information and the pre-trained generation confrontation network.
Chinese patent publication No. CN111190689A discloses a digital twin system simulation method and apparatus. The method comprises the following steps: acquiring at least two events to be simulated in the digital twin system; determining a mutually different probability guarantee value; distributing simulation priorities to the at least two events to be simulated based on the different probability guarantee values; and simulating the at least two events to be simulated based on the simulation priority to generate a simulation result.
In summary, the problems that the simulation scene is single, the scene occurrence data type is single, the data island of multi-scene data cannot be coupled and cooperated and the like exist in the application of the industrial production digital twin scene coupling device comprising the generation and simulation of multiple scenes in the industrial field at present, and the cooperative simulation requirement of the industrial scene digital twin on the multi-source heterogeneous scene data in the intelligent manufacturing production mode cannot be met. How to reasonably and effectively carry out digital twin scene coupling through scene generation is a problem which needs to be solved urgently.
Disclosure of Invention
The invention provides a multi-loop intelligent factory edge side digital twin scene coupling device which can be embedded into an intelligent factory edge side intelligent algorithm system as a service for coupling generation of digital twin production scenes, and can also provide reliable original data and training samples for other experiments and researches for targeted research.
The specific technical scheme of the invention is as follows:
a multi-loop intelligent factory edge side digital twin scene coupling device comprises a virtual space and a physical space of a production line edge side digital twin system, a decision support tool, a digital tool and an online dynamic scene database, wherein the virtual space and the physical space form a plurality of loops; the virtual space comprises a Conditional WGAN dynamic scene generation method, a scene generator and a simulation model for generating an antagonistic network based on Wasserstein; the physical space comprises a physical system;
the circuit comprises:
the digital twin simulation decision loop is used for connecting a virtual space and a physical space of the digital twin system at the edge of the production line to form a mutual coupling mapping relation;
the scene simulation loop is used for collecting actual production data of a physical space or simulation data of a virtual space, assisting a user in configuring a simulation scene through a scene generator and quickly generating a simulation sample of the physical production scene;
and the dynamic scene generation circuit learns scene data in the scene database by adopting a dynamic scene generation method, provides a dynamic virtual scene which accords with the statistical rule of historical scene samples for the scene generator, and expands the online dynamic scene database.
The digital twin simulation decision loop is a first loop: in the virtual space, the simulation results of different scenes quickly generate decision variables through a decision support tool to act on a physical system; the information of the physical system forms digital twin scene digital resources through digital tools such as a sensor network, RFID, virtual reality equipment and the like, and the digital twin scene digital resources are stored in an online dynamic scene database and used for generating a dynamic virtual scene and optimizing a simulation model.
The scene simulation loop is a second loop.
The core of the dynamic scene generation loop is a Conditional WGAN dynamic scene generation method for generating an antagonistic network based on Wasserstein, and the dynamic scene generation loop is a support loop of a dynamic scene generator, provides a large number of dynamic virtual scene samples for edge side intelligent calculation, and forms a third loop of the device.
Forming a scheduling scheme according to a simulation result of the simulation model; the decision support tool constructs a production instruction according to the scheduling scheme and issues the production instruction to a physical system; the physical system produces according to the production instruction, and collects the produced production data into an online dynamic scene database; and generating production data in the online dynamic scene database to a simulation module for simulation to form a digital twin simulation decision loop.
The digital twin simulation decision loop comprises a fault monitoring module, a production statistics module, a rescheduling module, an instruction issuing module, a plan scheduling optimization module and an online dynamic scene database;
the fault monitoring module receives actual production execution data and state data from the PCS layer interface, monitors whether production deviates from expectation or not, adjusts production through the rescheduling module and issues a rescheduling instruction; the production statistical module records the production result and the production efficiency, and the scheduling scheme is adjusted through the plan scheduling optimization module; the instruction issuing module issues a production instruction according to the rescheduling instruction and the adjusted scheduling scheme; meanwhile, the fault monitoring module and the production statistical module upload the monitoring statistical result to the prior dynamic scene database.
The method comprises the following steps that a scene generator establishes a scene model according to parameter configuration (such as inventory condition, order condition and yield model) in an online dynamic scene database, and a simulation model carries out scheduling calculation according to the scene model to obtain a scheduling scheme; and the scheduling scheme is transmitted back to the online dynamic scene database to form a scene simulation loop.
The scene model takes a scheduling period as a time unit; the scene model is as follows:
Figure BDA0002653211270000041
wherein, wvu，nThe binary variable represents whether the production unit u is produced at the event point n; bu，nRepresenting the total material receiving capacity of the device u in the time range from the beginning to the end of the event point n; sts，nRepresents the inventory of material s at the end of event point n; tsu，n，Tfu，nRespectively representing the starting time and the ending time of the production task in the unit u at the event point n;
Figure BDA0002653211270000042
representing the inventory of the product s at the end of the scheduling period t; opttRepresents an optimization objective, i.e. overall production cost;
Figure BDA0002653211270000043
representing the initial inventory of the material s in the scheduling period t; rhopu，sRepresents the material ratio of the material s in the unit u product; rhocu，sRepresenting the material ratio of the material s in the unit u raw material;
Figure BDA0002653211270000044
indicating a demand gap for product s at scheduling period t.
In the context of the scene model,
Figure BDA0002653211270000045
for input, the static parameters to be configured are input in a scheduling link in the simulation system;
Figure BDA0002653211270000046
opttfor output, belong to in the simulation systemAnd the scheduling result of the t-th scheduling period output in the scheduling link.
The simulation model solves the problem of mixed positive linear optimization (MILP) of planned scheduling through a CPLEX solver to perform optimization simulation, and a scheduling scheme of production scheduling is obtained.
The scene simulation loop comprises a scene generator, a parameter configuration module, a plan scheduling optimization simulation module, a production process simulation module, a rescheduling module, a production statistical module and an online dynamic scene database;
extracting preprocessed data from an online dynamic scene database to a scene generator, configuring parameters through a parameter configuration module according to scene requirements to be responded, optimizing a scheduling scheme through a plan scheduling optimization module according to the configured parameters, performing production simulation through a production process simulation module according to the scheduling scheme, and intervening a simulation process through a re-scheduling module and a production statistical module; and the rescheduling instruction generated by the rescheduling module and the production process data recorded by the production statistical module are transmitted back to the online dynamic scene database.
The dynamic scene generating circuit comprises a scene generator and an online dynamic scene database; the scene generator comprises a white noise-based scene generator and an antagonistic resolver used for training the scene generator; the scene generator generates a scene simulation sample by adopting a Conditional WGAN dynamic scene generation method for generating an antagonistic network based on Wasserstein.
The Conditional WGAN dynamic scene generation method for generating the countermeasure network based on Wasserstein comprises the following steps:
acquiring time series scene data in an actual production scene and historical scene data to construct a real scene sample database;
(2) adding label information of samples into an input layer and a hidden layer of a network, and constructing a Conditional WGAN dynamic scene generation model for generating an antagonistic network based on Wasserstein;
(3) selecting real scene sample data from a real scene sample database as training set data based on a zero-sum game thought, training a Conditional WGAN dynamic scene generation model, and continuously correcting the model, thereby obtaining a scene generator;
(4) and generating time series dynamic scene sample data containing real scene characteristics according to the scene generator.
The time series dynamic scene sample data needs to contain different characteristics contained in training set data, the characteristics of the data need to present consistency, but the scene representation of the data needs to present diversity.
In the step (1), the time series scene data comprises scene features and corresponding time; the scene characteristics comprise equipment parameters and inventory data.
In the step (2), when a conditional wgan dynamic scene generation model for generating an anti-network based on Wasserstein is constructed:
(a) setting generator web learning rate alphag(αg∈(0，0.1]Default value is 0.001), the resolver net learning rate αd(αd∈(0，0.1]Default value is 0.001), number of samples (batch size) M (M is from 10% M, 20% M) selected in a single training]M is the total number of samples), the ratio of the number of iterations of the network of resolvers to the network of generators ncritic(ncritic∈[2，20]Default value is 5), let Lipschitz constant K be 1, and set the gradient penalty weight λ (λ ∈ [0, 1)), and Adam learner parameter β1，β2(β1，β2∈[0，1))；
(b) After setting the parameters, the generator network G (·; ω) is generatedG) And a network of resolvers D (·; omegaD) An initialization operation is carried out, in which the generator network G (·; omegaG) And a network of resolvers D (·; omegaD) All are multilayer networks composed of full connection layers, convolution layers and deconvolution layers; the active Units of the two networks respectively adopt a Rectified Linear unit (ReLU) and a leakage Rectified Linear unit (LReLU).
In the step (3), the real scene sample data selected from the real scene sample database is adopted to train the Conditional WGAN dynamic scene generation model, and the method comprises the following steps:
(3-1) randomly extracting m pieces of data from real scene sample data
Figure BDA0002653211270000061
Wherein P isXFor the unknown distribution recorded in the real scene, which is difficult to model, { x, y } is a sample pair;
(3-2) randomly sampling m pieces of noise sample data from Gaussian distribution
Figure BDA0002653211270000062
Wherein P isZIs a Gaussian distribution, { z, y } is a sample pair;
(3-3) setting a gradient penalty term to make the resolver network satisfy Lipschitz continuity;
(3-3-1) sampling a random number from the [0, 1] uniform distribution;
(3-3-2) constructing interpolation samples
Figure BDA0002653211270000063
Wherein x and z are each independently in accordance with PXAnd PZA sample of the distribution;
(3-3-3) construction of gradient penalty term
Figure BDA0002653211270000064
Wherein
Figure BDA0002653211270000065
Is composed of
Figure BDA0002653211270000066
And
Figure BDA0002653211270000067
wasserstein distance of;
(3-4) updating the resolver network parameter ω(D)；
(3-5) repeating (3-1) to (3-4) ncriticSecondly;
(3-6) randomly sampling m pieces of noise sample data from Gaussian distribution
Figure BDA0002653211270000068
(3-7) updating Generator network parameters ω(G)；
(3-8) repeating (3-1) to (3-7) until the parameter ω(D)And (6) converging.
In order to enable the discriminator network to meet the Lipschitz continuous condition, the step (3-3) and the step (3-4) add a gradient penalty term in the loss function of the discriminator network, compared with the original WGAN which adopts the step of cutting the network parameters of the discriminator network to meet the Lipschitz condition, the method of the invention can avoid the gradient disappearance phenomenon caused by the binaryzation of the network parameters of the discriminator network, and can obviously improve the training speed of the network.
In the steps (3-3) and (3-5), the adaptive Adam optimization algorithm is adopted to iteratively update the network weight, and the learning rate can be automatically adjusted by calculating the first-order moment estimation and the second-order moment estimation of the gradient. The scene generator and the antagonism discriminator are initialized by adopting an Xavier initialization method.
To further improve the stability of the training, Layer Normalization and Batch Normalization are employed before the hidden layers of the generator network and the antipodal discriminator network, respectively, to normalize the hidden Layer input data.
Iteration number ratio n of a network of resolvers to a network of generatorscriticSet to 5, the generator network parameters are updated once every 5 times the resolver network parameters are updated, and in order to get the initialized resolver network fully trained, the generator network is updated for the first time 25 times after the resolver network parameters are updated.
Compared with the prior art, the invention has the beneficial effects that:
aiming at large-scale complex scenes in the industrial field, the invention can generate static and dynamic coupling scenes according to different characteristics and different time and space of the scenes, thereby meeting the requirements of application modes of industrial production enterprises on scene simulation at the present stage and perception, analysis, decision and execution integration of manufacturing enterprises in the intelligent manufacturing era. In the aspect of a dynamic scene generation loop, different from the traditional time sequence prediction method, the method provided by the invention can provide more sample independence and sample diversity of scenes, and performs coupling cooperation of multi-source heterogeneous scene data through a scene coupling device on the basis, so that various complicated coupling scenes in the industrial field can be met.
Drawings
FIG. 1 is a schematic structural diagram of a multi-loop intelligent factory edge side digital twin scene coupling device;
FIG. 2 is a schematic view of a production flow of a comprehensive production line of a petrochemical enterprise;
FIG. 3 is a schematic diagram of an implementation flow of a digital twin simulation decision loop;
FIG. 4 is a schematic diagram of an implementation flow of a scene simulation loop;
FIG. 5 is a schematic diagram of a dynamic scene generation circuit;
FIG. 6 is a schematic diagram of a dynamic scene generation method model training process based on Conditional WGAN.
Detailed Description
The invention will be described in further detail below with reference to the drawings and examples, which are intended to facilitate the understanding of the invention without limiting it in any way.
As shown in fig. 1, the present invention provides a multi-loop intelligent factory edge-side digital twin scene coupling device, which comprises the following loops:
1. and the digital twin simulation decision loop is used for connecting the virtual space and the physical space of the digital twin system at the edge of the production line to form a mutual coupling mapping relation to form a first loop of the device. In the virtual space, the simulation results of different scenes quickly generate decision variables through a decision tool to act on a physical system. The information of the physical system forms digital twin scene digital resources through digital tools such as a sensor network, RFID, virtual reality equipment and the like, and the digital twin scene digital resources are stored in an online dynamic scene database and used for scene generation and model optimization.
2. And the scene simulation loop is used for collecting actual production data or simulation data, assisting a user in configuring a simulation scene through a scene generator of a virtual space, and quickly generating a simulation sample of the physical production scene. Constituting the second circuit of the device.
3. A dynamic scene generating circuit is a Conditional WGAN dynamic scene generating method for generating an antagonistic network based on Wasserstein, and is a supporting circuit of a dynamic scene generator. The scene generation method provides a dynamic virtual scene which accords with the statistical rule of historical scene samples for the scene generator through the learning of scene data in the online dynamic scene database, and expands the online dynamic scene database. And a large number of dynamic scene samples are provided for the edge-side intelligent calculation, and a third loop of the device is formed.
In this embodiment, a flow of an integrated production line of a petrochemical enterprise is taken as a prototype, and a flow chart is shown in fig. 2, and a loop flow description is performed on the digital twin scene coupling device on the edge side of the multi-loop intelligent plant based on a production scheduling workflow. The petrochemical enterprise production line takes crude oil as a main raw material, and is further processed by nine production units of raw material pretreatment, hydrofining, normal line hydrogenation, delayed coking, hydroisomerization, product refining, continuous reforming, aromatic hydrocarbon extraction, tail oil preflow and the like to produce various refining products such as saturated liquefied petroleum gas, naphtha, diesel oil, gasoline, kerosene, aromatic hydrocarbon and the like, wherein part of the refining products are supplied to ethylene branch companies in enterprises as production raw materials, and the rest refining products are used as export sales.
The first loop mainly comprises the steps that a plurality of scheduling schemes are formed from a simulation model according to a simulation result, and a decision support tool constructs and issues production instructions to a physical system according to the scheduling schemes. In the process of producing according to the instruction, the physical system collects production data and the like into the scene database through the data collection tool, and then resends the data to the simulation module through some modes so as to make a further optimized scheduling scheme.
Detailed description of an embodiment thereof with reference to fig. 3, the planning, scheduling, simulation module performs a scenario model sce based on the parameter configuration (e.g., inventory status, order status, and yield model) in the scenario databasetIs established and based thereonTo perform scheduling calculations. The model is a scene model with a scheduling period as a time unit:
Figure BDA0002653211270000081
wherein, wvu，nThe binary variable represents whether the production unit u is produced at the event point n; bu，nRepresenting the total material receiving capacity of the device u in the time range from the beginning to the end of the event point n; sts，nRepresents the inventory of material s at the end of event point n; tsu，n，Tfu，nRespectively representing the starting time and the ending time of the production task in the unit u at the event point n;
Figure BDA0002653211270000091
representing the inventory of the product s at the end of the scheduling period t; opttRepresents an optimization objective, i.e. overall production cost;
Figure BDA0002653211270000092
representing the initial inventory of the material s in the scheduling period t; rhopu，sRepresents the material ratio of the material s in the unit u product; rhocu，sRepresenting the material ratio of the material s in the unit u raw material;
Figure BDA0002653211270000093
indicating a demand gap for product s at scheduling period t.
In this model, the model is shown as,
Figure BDA0002653211270000094
static parameters to be configured, which are input in a scheduling link in a simulation system, are input for a scheduling model;
Figure BDA0002653211270000095
opttand the scheduling result belongs to the t-th scheduling period output by the scheduling link in the simulation system for scheduling model output.
Solving planned schedules through CPLEX solverAfter optimization simulation is carried out by mixing a positive linear optimization (MILP) problem, an obtained production scheduling scheme (the scheduling period is set to be 24 hours) is transmitted to an instruction issuing module, and the instruction issuing module mainly decomposes the production scheduling scheme. Because a modeling method based on batch processing is adopted in the scheduling stage, the scheduling scheme is given in
Figure BDA0002653211270000096
To
Figure BDA0002653211270000097
Total processing amount b of production unit u in time period of (1)u，nBy means of a yield model
Figure BDA0002653211270000098
Can obtain the side lines of the production unit
Figure BDA0002653211270000099
To
Figure BDA00026532112700000910
Total flow of (a). To drive the advance in unit time to match the dynamic process of the PCS level, the total flow in each side line can be decomposed again into flow control amount in unit time to be transmitted to the PCS level interface. The instruction issuing module makes a scheduling scheme according to the production instruction frequency (unit time is set to be 30 seconds), and in the actual production scene, the instruction controls the siding flow of 9 production units in unit time.
In the production process, the fault monitoring module can monitor the production deviation generated in the production process, the production counting module can record the production result, the production efficiency and the like, and the two modules can correct and operate according to the deviation caused by uncertainty in the actual production process and respectively transmit the records to the instruction issuing module and the plan scheduling optimization simulation module so as to intervene in the production process. For the fault monitoring module, the fault monitoring module mainly refers to that production personnel receive actual production execution data and state data from a PCS layer interface and monitor whether production is performed or notDeviating from expectations to adjust production and issue rescheduling instructions. Due to a number of factors, there is often hysteresis in the handling of the monitored problem. There are many types of faults that can be detected during the monitoring phase, such as the maximum capacity of a production unit in a single lot due to a production unit fault
Figure BDA0002653211270000101
Or material loss due to pipe leakage, can result in
Figure BDA0002653211270000102
And
Figure BDA0002653211270000103
changes in the location of the display, etc. When the production monitoring link monitors the fluctuation of production, the production can be adjusted by issuing the instruction again. Calculated by rescheduling algorithm
Figure BDA0002653211270000104
Figure BDA0002653211270000105
And
Figure BDA0002653211270000106
and regenerating the instruction sequence by using a dynamic scene generation method. For the production statistics module, statistics personnel perform bottom-up production statistics based on actual changes of raw material consumption, product yield, inventory change and the like reflected by the sensors, form assessment reports of shift data, day data, week data, month data and the like, and further adjust production. The flow control and flow monitoring are carried out in the simulation of the command issuing and production monitoring stage. Meanwhile, the two modules can simultaneously store the information into a real-time dynamic scene database so as to facilitate the extraction of the scene in the process of further optimizing the scheduling.
The second loop is shown in FIG. 4, and mainly extracts the preprocessed data from the scene database to the scene generator according to the resultIn the embodiment, a scene simulation of two scheduling cycles at a time is taken as an example. Product demand for the first scheduling cycle
Figure BDA0002653211270000107
And
Figure BDA0002653211270000108
given manually according to experience, the user can select the specific type of the user,
Figure BDA0002653211270000109
and
Figure BDA00026532112700001010
it needs to be obtained after the simulation of the first scheduling period is completed. While two scheduling cycle yield models
Figure BDA00026532112700001011
And
Figure BDA00026532112700001012
are all obtained by a static scene generation method. Firstly, relevant flow measurement and control points are selected from a system database, the distribution of the flow measurement and control points is shown in a table 1, and flow time sequence data of the flow measurement and control points for one month is collected. Because the yield model represents the relatively stable mutual relation among the lateral line flows of the production units, the time sequence data is further processed, the average flow of each measurement and control point in each hour is respectively calculated, and finally 372 groups of 30-dimensional sample data in total are obtained and are put into a scene database. And taking the 372 groups of samples as training samples, using a vine-copula-based static scene generation method to perform dimension reduction and standardization of the scene samples, creating a scene model and generating the scene samples, generating to obtain 1000 groups of generated samples, and putting the generated samples into a scene database.
TABLE 1 refinery System measurement and control Point distribution
Figure BDA00026532112700001013
Figure BDA0002653211270000111
Yield model for calculating any scheduling period by randomly selecting a group of generation samples
Figure BDA0002653211270000112
Figure BDA0002653211270000113
Taking the raw material pretreatment unit as an example, if F01-F06 in a group of randomly generated samples are 74944.37kg/h, 92155.5kg/h, 120137.2kg/h, 126953.3kg/h, 10617.81kg/h and 113717kg/h respectively, the samples can pass through
Figure BDA0002653211270000114
The formula of s ═ F01, F02, F03, F04, F05 and F06 respectively obtains that all six side stream flow ratios of the normal line oil, the straight-run diesel oil, the vacuum wax oil, the vacuum residuum, the liquefied gas and the straight-run naphtha on the output side of the raw material pretreatment unit are respectively 0.139, 0.171, 0.223, 0.236, 0.020 and 0.211.
For the sidings not included in table 1, the actual production system does not record the flow rate due to the limitation of cost and other factors, so the generated scene sample does not include the scene feature, and the calculation cannot be directly performed through the generated sample, which can be indirectly calculated through the material balance of the production unit and the siding flow rate ratio normalization constraint. For example, the flow rate of the crude oil side line at the input side of the raw material pretreatment unit is the sum of F01-F06 based on the material balance constraint of the production unit, and the flow rate of the crude oil side line at the input side of the raw material pretreatment unit is the only side line at the input side of the raw material pretreatment unit based on the side line flow rate ratio normalization constraint, so that the side line flow rate ratio at the input side is obtained
Figure BDA0002653211270000115
Figure BDA0002653211270000116
And
Figure BDA0002653211270000117
together, the yield models of the raw material pretreatment unit are formed, and the yield models of other production units can be obtained by the same method.
And then, establishing and solving a simulation model in the first loop according to the parameters, performing instruction decomposition simulation and production simulation on the solved result, intervening the simulation process according to a rescheduling module and a data statistical unit, and transmitting all data generated by the rescheduling instruction and the production process back to a scene database.
The flow of the third loop is shown in fig. 5, which includes the sampling of scene samples, the white noise-based scene sample generator, the antithetic resolver used for training the generator, and so on. The method comprises the steps of collecting time sequence data of a constant line oil flow measuring and controlling point, preprocessing the data in a real-time dynamic database, carrying out sampling construction samples of time sequence samples (for example, time sequence data of 05: 05-06: 05 is one sample, time sequence data of 05: 10-06: 10 is the next sample) which are often 1h on constant line oil outlet data at a sampling interval of 5min, wherein each sample is a time sequence of 120 sample points, and placing the samples into a scene database. These samples are then used as training samples. Of these, 80% was used as a training sample and the remaining 20% was used as a proof. Through continuous confrontation of a generator and a discriminator which take white noise as metadata, training and correction are carried out according to a discrimination result, and finally generated scene samples which meet requirements and are generated by the generator in a scene generation model are output to the outside of the system through a scene sample output step to form simulated scene samples. The goal of this simulated scene sample is to generate time series samples with similar patterns and the same length as the samples, which finally occurs to yield 5000 scene samples.
The generator and resolver antagonistic training process is shown in fig. 6, where both the generator network and the resolver network are initialized with Xavier. The same number of samples is selected for a single training to reduce training oscillations while the gradient descent is accurate. To further improve the stability of the training, in the generator netThe hidden layers of the net and the network of resolvers have been preceded by Layer Normalization and Batch Normalization, respectively, to normalize the hidden Layer input data. Iteration number ratio n of a network of resolvers to a network of generatorscriticSet to 5, the generator network parameters are updated once every 5 times the resolver network parameters are updated, and in order to get the initialized resolver network fully trained, the generator network is updated for the first time 25 times after the resolver network parameters are updated. After a number of update iterations, D (x) and D (g (z)) tend to be equal, and the loss function of the network of resolvers fluctuates around "0", at which point the generator can produce the required time series of samples.
The Conditional WGAN dynamic scene generation method for generating the countermeasure network based on Wasserstein comprises the following steps:
acquiring time sequence data in an actual production scene and historical data to provide original data for training for scene sample generation;
(2) adding label information of samples into an input layer and a hidden layer of a network, and constructing a Conditional WGAN scene generation model for generating an antagonistic network based on Wasserstein;
(3) training a Conditional WGAN scene generation model through real sample data selected from a real scene sample database based on a zero-sum game thought, and continuously correcting the model, so as to obtain a dynamic scene simulation sample generation model;
(4) and generating time sequence dynamic scene sample data containing real scene characteristics according to the trained dynamic scene simulation sample generation model. The time series dynamic scene sample data needs to contain different characteristics contained in training set data, the characteristics of the data need to present consistency, but the scene representation of the data needs to present diversity.
In the step (1), when acquiring time series data in an actual production scene and a historical scene of an industrial production process, the collected scene data includes a certain amount of scene characteristics such as equipment parameters and inventory data, and the time corresponding to the collected scene data is recorded.
Step (2) when a Conditional WGAN scene occurrence model for generating the countermeasure network based on Wasserstein is constructed:
(a) it is necessary to set the generator net learning rate αgNetwork learning rate of the resolver αdNumber of samples (batch size) m selected in a single training, ratio of number of iterations of the network of resolvers to the network of generators ncriticLet Lipschitz constant K be 1, and set gradient penalty weight λ, and Adam learner parameter β1，β2。
(b) After setting the parameters, the generator network G (·; ω) is generatedG) And a network of resolvers D (·; omegaD) An initialization operation is carried out, in which the generator network G (·; omegaG) And a network of resolvers D (·; omegaD) All are multilayer networks composed of fully-connected layers, convolutional layers and deconvolution layers. The active Units of the two networks respectively adopt a Rectified Linear unit (ReLU) and a leakage Rectified Linear unit (LReLU).
The training of the Conditional WGAN scene occurrence model by the real sample data selected from the real scene sample database in the step (3) can be summarized as follows:
(3-1) randomly extracting m pieces of data from the disordered historical sample data
Figure BDA0002653211270000131
Wherein P isXFor the unknown distribution recorded in the real scene, which is difficult to model, { x, y } is a sample pair;
(3-2) randomly sampling m pieces of noise sample data from Gaussian distribution
Figure BDA0002653211270000132
Wherein P isZIs a Gaussian distribution, { z, y } is a sample pair;
(3-3) setting a gradient penalty term to make the discriminator meet Lipschitz continuity;
(3-3-1) sampling a random number from the [0, 1] uniform distribution;
(3-3-2) constructing interpolation samples
Figure BDA0002653211270000133
Wherein x and z are each independently PXAnd PZA sample of the distribution;
(3-3-3) construction of gradient penalty term
Figure BDA0002653211270000134
Wherein
Figure BDA0002653211270000135
Is composed of
Figure BDA0002653211270000136
And
Figure BDA0002653211270000137
wasserstein distance of. (ii) a
(3-4) updating the resolver network parameter ω(D)；
(3-5) repeating (3-1) to (3-4) ncriticSecondly;
(3-6) randomly sampling m pieces of noise sample data from Gaussian distribution
Figure BDA0002653211270000141
(3-7) updating Generator network parameters ω(G)；
(3-8) repeating (3-1) to (3-7) until the parameter ω(D)And (6) converging.
In order to enable the discriminator network to meet the Lipschitz continuous condition, the steps (3-3) and (3-4) add a gradient penalty term in the loss function of the discriminator network, compared with the method for clipping the network parameters of the discriminator network to meet the Lipschitz condition adopted by the native WGAN, the method can avoid the gradient vanishing phenomenon caused by the binarization of the network parameters of the discriminator network, and can obviously improve the training speed of the network.
In the steps (3-3) and (3-5), the adaptive Adam optimization algorithm is adopted to iteratively update the network weight, and the learning rate can be automatically adjusted by calculating the first-order moment estimation and the second-order moment estimation of the gradient.
The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only specific embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.

Claims (10)

1. A multi-loop intelligent factory edge side digital twin scene coupling device is characterized by comprising a virtual space and a physical space of a production line edge side digital twin system, a decision support tool, a digitization tool and an online dynamic scene database, wherein the virtual space and the physical space form a plurality of loops; the virtual space comprises a Conditional WGAN dynamic scene generation method, a scene generator and a simulation model for generating an antagonistic network based on Wasserstein; the physical space comprises a physical system;
the circuit comprises:
the digital twin simulation decision loop is used for connecting a virtual space and a physical space of the digital twin system at the edge of the production line to form a mutual coupling mapping relation;
the scene simulation loop is used for collecting actual production data of a physical space or simulation data of a virtual space, assisting a user in configuring a simulation scene through a scene generator and quickly generating a simulation sample of the physical production scene;
and the dynamic scene generation circuit learns scene data in the scene database by adopting a dynamic scene generation method, provides a dynamic virtual scene which accords with the statistical rule of historical scene samples for the scene generator, and expands the online dynamic scene database.
2. The multi-loop intelligent factory edge side digital twin scene coupling device as claimed in claim 1, wherein a scheduling scheme is formed according to a simulation result of a simulation model; the decision support tool constructs a production instruction according to the scheduling scheme and issues the production instruction to a physical system; the physical system produces according to the production instruction, and collects the produced production data into an online dynamic scene database; and generating production data in the online dynamic scene database to a simulation module for simulation to form a digital twin simulation decision loop.
3. The multi-loop intelligent factory edge side digital twin scene coupling device as claimed in claim 1, wherein the digital twin simulation decision loop comprises a fault monitoring module, a production statistics module, a rescheduling module, an instruction issuing module, a plan scheduling optimization module and an online dynamic scene database;
the fault monitoring module receives actual production execution data and state data from the PCS layer interface, monitors whether production deviates from expectation or not, adjusts production through the rescheduling module and issues a rescheduling instruction; the production statistical module records the production result and the production efficiency, and the scheduling scheme is adjusted through the plan scheduling optimization module; the instruction issuing module issues a production instruction according to the rescheduling instruction and the adjusted scheduling scheme; meanwhile, the fault monitoring module and the production statistical module upload the monitoring statistical result to the prior dynamic scene database.
4. The device for coupling the digital twin scenes at the edge of the multi-loop intelligent factory according to claim 1, wherein the scene generator establishes a scene model according to parameter configuration in an online dynamic scene database, and the simulation model performs scheduling calculation according to the scene model to obtain a scheduling scheme; and the scheduling scheme is transmitted back to the online dynamic scene database to form a scene simulation loop.
5. The multi-loop intelligent plant edge-side digital twin scene coupling device as claimed in claim 4, wherein the scene model is in a scheduling cycle as a time unit; the scene model is as follows:
Figure FDA0002653211260000021
wherein, wvu，nThe binary variable represents whether the production unit u is produced at the event point n; bu，nRepresenting the total material receiving capacity of the device u in the time range from the beginning to the end of the event point n; sts，nRepresents the inventory of material s at the end of event point n; tsu，n，Tfu，nRespectively representing the starting time and the ending time of the production task in the unit u at the event point n;
Figure FDA0002653211260000022
representing the inventory of the product s at the end of the scheduling period t; opttRepresents an optimization objective, i.e. overall production cost;
Figure FDA0002653211260000023
representing the initial inventory of the material s in the scheduling period t; rhopu，sRepresents the material ratio of the material s in the unit u product; rhocu，sRepresenting the material ratio of the material s in the unit u raw material;
Figure FDA0002653211260000024
indicating a demand gap for product s at scheduling period t.
6. The multi-loop intelligent factory edge side digital twin scene coupling device as claimed in claim 1, wherein the scene simulation loop comprises a scene generator, a parameter configuration module, a plan scheduling optimization simulation module, a production process simulation module, a rescheduling module, a production statistics module and an online dynamic scene database;
extracting preprocessed data from an online dynamic scene database to a scene generator, configuring parameters through a parameter configuration module according to scene requirements to be responded, optimizing a scheduling scheme through a plan scheduling optimization module according to the configured parameters, performing production simulation through a production process simulation module according to the scheduling scheme, and intervening a simulation process through a re-scheduling module and a production statistical module; and the rescheduling instruction generated by the rescheduling module and the production process data recorded by the production statistical module are transmitted back to the online dynamic scene database.
7. The apparatus of claim 1, wherein the dynamic scene generation method is a Conditional WGAN dynamic scene generation method based on Wasserstein generation countermeasure network.
8. The apparatus as claimed in claim 7, wherein the scene generator comprises a white noise-based scene generator and an antagonistic resolver for training the scene generator.
9. The device for coupling digital twin scenes at the edge of a multi-loop intelligent factory according to claim 8, wherein the method for generating the Conditional WGAN dynamic scene of the countermeasure network based on Wasserstein comprises the following steps:
acquiring time series scene data in an actual production scene and historical scene data to construct a real scene sample database;
(2) adding label information of samples into an input layer and a hidden layer of a network, and constructing a Conditional WGAN dynamic scene generation model for generating an antagonistic network based on Wasserstein;
(3) selecting real scene sample data from a real scene sample database as training set data based on a zero-sum game thought, training a Conditional WGAN dynamic scene generation model, and continuously correcting the model, thereby obtaining a scene generator;
(4) and generating time series dynamic scene sample data containing real scene characteristics according to the scene generator.
10. The multi-loop intelligent factory edge-side digital twin scene coupling device according to claim 9, wherein step (3) comprises:
(3-1) randomly extracting m pieces of data from real scene sample data
Figure FDA0002653211260000031
Wherein P isXFor the unknown distribution recorded in the real scene, which is difficult to model, { x, y } is a sample pair;
(3-2) randomly sampling m pieces of noise sample data from Gaussian distribution
Figure FDA0002653211260000032
Wherein P isZIs a Gaussian distribution, { z, y } is a sample pair;
(3-3) setting a gradient penalty term to make the resolver network satisfy Lipschitz continuity;
(3-3-1) sampling a random number from the [0, 1] uniform distribution;
(3-3-2) constructing interpolation samples
Figure FDA0002653211260000033
Wherein x and z are each independently in accordance with PXAnd PZA sample of the distribution;
(3-3-3) construction of gradient penalty term
Figure FDA0002653211260000034
Wherein
Figure FDA0002653211260000035
Is composed of
Figure FDA0002653211260000036
And
Figure FDA0002653211260000037
wasserstein distance of;
(3-4) updating the resolver network parameter ω(G)；
(3-5) repeating (3-1) to (3-4) ncriticSecondly;
(3-6) randomly sampling m pieces of noise sample data from Gaussian distribution
Figure FDA0002653211260000041
(3-7) renewNetwork parameter omega of generator(G)；
(3-8) repeating (3-1) to (3-7) until the parameter ω(D)And (6) converging.
CN202010881839.6A 2020-08-27 2020-08-27 Multi-loop intelligent factory edge side digital twin scene coupling device Active CN112131782B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010881839.6A CN112131782B (en) 2020-08-27 2020-08-27 Multi-loop intelligent factory edge side digital twin scene coupling device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010881839.6A CN112131782B (en) 2020-08-27 2020-08-27 Multi-loop intelligent factory edge side digital twin scene coupling device

Publications (2)

Publication Number Publication Date
CN112131782A true CN112131782A (en) 2020-12-25
CN112131782B CN112131782B (en) 2024-05-14

Family

ID=73847573

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010881839.6A Active CN112131782B (en) 2020-08-27 2020-08-27 Multi-loop intelligent factory edge side digital twin scene coupling device

Country Status (1)

Country Link
CN (1) CN112131782B (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112762100A (en) * 2021-01-14 2021-05-07 哈尔滨理工大学 Bearing full-life-cycle monitoring method based on digital twinning
CN112859789A (en) * 2021-01-29 2021-05-28 重庆邮电大学 Method and system for constructing data center digital twin body based on CFD
CN113219922A (en) * 2021-05-11 2021-08-06 浙江大学 Integrated control method based on edge side data under endogenous uncertainty
CN113741910A (en) * 2021-09-01 2021-12-03 上海商汤智能科技有限公司 Scene interaction method and device, electronic equipment and storage medium
CN114418177A (en) * 2021-12-13 2022-04-29 中国科学院合肥物质科学研究院 New product material distribution prediction method for digital twin workshop based on generation of countermeasure network
CN115148064A (en) * 2022-06-23 2022-10-04 广东三向智能科技股份有限公司 Intelligent factory simulation device of virtual simulation teaching system
CN115828719A (en) * 2022-07-17 2023-03-21 东南大学溧阳研究院 Power system subsynchronous oscillation data generation method based on generation countermeasure network
WO2023045463A1 (en) * 2021-09-22 2023-03-30 西安热工研究院有限公司 Online monitoring and early warning system and method for power station pipeline based on digital twin technology
CN117130351A (en) * 2023-09-18 2023-11-28 上海勘测设计研究院有限公司 New energy station area joint control protection system based on digital twin technology

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180131907A1 (en) * 2016-11-08 2018-05-10 Rockwell Automation Technologies, Inc. Virtual reality and augmented reality for industrial automation
CN108445777A (en) * 2017-01-14 2018-08-24 费德姆技术公司 The data conversion of fictitious assets
WO2019076233A1 (en) * 2017-10-17 2019-04-25 广东工业大学 Quick customization design method and system for intelligent workshop
US20190163215A1 (en) * 2017-11-27 2019-05-30 General Electric Company Building energy modeling tool systems and methods
CN110471279A (en) * 2019-07-25 2019-11-19 浙江大学 A kind of industrial production simulated scenario generator and scene method for generation based on vine-copulas
CN111161410A (en) * 2019-12-30 2020-05-15 中国矿业大学(北京) Mine digital twinning model and construction method thereof
WO2020138603A1 (en) * 2018-12-27 2020-07-02 부산대학교 산학협력단 Digital twin system and method for optimizing control scenario of mechanical device
US20200265329A1 (en) * 2019-02-14 2020-08-20 Rockwell Automation Technologies, Inc. Ai extensions and intelligent model validation for an industrial digital twin

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180131907A1 (en) * 2016-11-08 2018-05-10 Rockwell Automation Technologies, Inc. Virtual reality and augmented reality for industrial automation
CN108445777A (en) * 2017-01-14 2018-08-24 费德姆技术公司 The data conversion of fictitious assets
WO2019076233A1 (en) * 2017-10-17 2019-04-25 广东工业大学 Quick customization design method and system for intelligent workshop
US20190163215A1 (en) * 2017-11-27 2019-05-30 General Electric Company Building energy modeling tool systems and methods
WO2020138603A1 (en) * 2018-12-27 2020-07-02 부산대학교 산학협력단 Digital twin system and method for optimizing control scenario of mechanical device
US20200265329A1 (en) * 2019-02-14 2020-08-20 Rockwell Automation Technologies, Inc. Ai extensions and intelligent model validation for an industrial digital twin
CN110471279A (en) * 2019-07-25 2019-11-19 浙江大学 A kind of industrial production simulated scenario generator and scene method for generation based on vine-copulas
CN111161410A (en) * 2019-12-30 2020-05-15 中国矿业大学(北京) Mine digital twinning model and construction method thereof

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
林润泽;王行健;冯毅萍;赵久强;: "基于数字孪生的智能装配机械臂实验***", 实验室研究与探索, no. 12 *

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112762100A (en) * 2021-01-14 2021-05-07 哈尔滨理工大学 Bearing full-life-cycle monitoring method based on digital twinning
CN112859789B (en) * 2021-01-29 2024-02-06 重庆邮电大学 CFD-based method and system for constructing digital twin body of data center
CN112859789A (en) * 2021-01-29 2021-05-28 重庆邮电大学 Method and system for constructing data center digital twin body based on CFD
CN113219922A (en) * 2021-05-11 2021-08-06 浙江大学 Integrated control method based on edge side data under endogenous uncertainty
CN113219922B (en) * 2021-05-11 2022-05-03 浙江大学 Integrated control method based on edge side data under endogenous uncertainty
CN113741910A (en) * 2021-09-01 2021-12-03 上海商汤智能科技有限公司 Scene interaction method and device, electronic equipment and storage medium
WO2023045463A1 (en) * 2021-09-22 2023-03-30 西安热工研究院有限公司 Online monitoring and early warning system and method for power station pipeline based on digital twin technology
CN114418177A (en) * 2021-12-13 2022-04-29 中国科学院合肥物质科学研究院 New product material distribution prediction method for digital twin workshop based on generation of countermeasure network
CN114418177B (en) * 2021-12-13 2024-03-29 中国科学院合肥物质科学研究院 New product material distribution prediction method based on digital twin workshops for generating countermeasure network
CN115148064A (en) * 2022-06-23 2022-10-04 广东三向智能科技股份有限公司 Intelligent factory simulation device of virtual simulation teaching system
CN115828719B (en) * 2022-07-17 2023-10-03 东南大学溧阳研究院 Method for generating subsynchronous oscillation data of power system based on generation countermeasure network
CN115828719A (en) * 2022-07-17 2023-03-21 东南大学溧阳研究院 Power system subsynchronous oscillation data generation method based on generation countermeasure network
CN117130351A (en) * 2023-09-18 2023-11-28 上海勘测设计研究院有限公司 New energy station area joint control protection system based on digital twin technology
CN117130351B (en) * 2023-09-18 2024-03-19 上海勘测设计研究院有限公司 New energy station area joint control protection system based on digital twin technology

Also Published As

Publication number Publication date
CN112131782B (en) 2024-05-14

Similar Documents

Publication Publication Date Title
CN112131782A (en) Multi-loop intelligent factory edge side digital twin scene coupling device
CN107977737B (en) Distribution transformation load prediction method based on mxnet frame deep neural network
CN105302096B (en) Intelligent factory scheduling method
CN106326585B (en) Prediction analysis method and device based on Bayesian Network Inference
CN109726503A (en) Missing data complementing method and device
CN110516901A (en) Customer value hierarchical mode building system and client's layered approach based on big data
CN103310286A (en) Product order prediction method and device with time series characteristics
CN103902816A (en) Electrification detection data processing method based on data mining technology
DE10223725A1 (en) Process control system for use in petroleum process plant, implements software routine by receiving process data, process control data and process performance data, for performing function within plant
CN104407589A (en) Workshop manufacturing process-oriented active sensing and anomaly analysis method of real-time generating performance
CN111127246A (en) Intelligent prediction method for transmission line engineering cost
CN112084385B (en) Part-process-equipment association relationship topological view generation method based on digital twinning
CN107918831A (en) BIM Schedule managements method and its system based on browser
Raikov et al. Big data refining on the base of cognitive modeling
CN105260835A (en) Method for modeling, analyzing, and self-optimizing multi-source business big data
CN112330050A (en) Power system load prediction method considering multiple features based on double-layer XGboost
CN105893669A (en) Global simulation performance predication method based on data digging
CN108038616A (en) A kind of work transmission line analysis of prices method
Ungermann et al. Data analytics for manufacturing systems–a data-driven approach for process optimization
CN115438726A (en) Device life and fault type prediction method and system based on digital twin technology
CN113627784A (en) Enterprise asset management intelligent decision-making system based on industrial internet
CN106292591A (en) A kind of equipment big data platform monitoring diagnosis system
CN110941646A (en) Project progress control system based on 10KV project three-rate data mining
Friederich et al. A Framework for Validating Data-Driven Discrete-Event Simulation Models of Cyber-Physical Production Systems
CN115118578A (en) SCADA system based on WEB

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
GR01 Patent grant
GR01 Patent grant