CN112989063B - 3D modeling method and system based on knowledge graph - Google Patents
3D modeling method and system based on knowledge graph Download PDFInfo
- Publication number
- CN112989063B CN112989063B CN202110261587.1A CN202110261587A CN112989063B CN 112989063 B CN112989063 B CN 112989063B CN 202110261587 A CN202110261587 A CN 202110261587A CN 112989063 B CN112989063 B CN 112989063B
- Authority
- CN
- China
- Prior art keywords
- output
- layer
- model
- hidden layer
- network
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/31—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention relates to a 3D modeling method based on a knowledge graph, which comprises the following steps of establishing a data hotspot library and carrying out modeling management on 3D model information; establishing a 3D model association network to obtain the degree of relationship among all devices in a 3D model system; establishing a 3D model matching network to obtain a data hotspot corresponding to the 3D model; and comparing the real equipment with the obtained characteristic information of the 3D model, and analyzing and evaluating the 3D model matching network. The method of the invention carries out hierarchical management on the 3D model in the knowledge graph, and can quickly inquire the 3D model in the knowledge graph through the received data hotspot information; by adopting the 3D model association network, the association degree among all the devices in the system is quickly obtained, so that the 3D model obtained in the model matching process is more similar to the real device.
Description
Technical Field
The invention relates to the field of computers and intelligent computing, in particular to a 3D modeling method based on a knowledge graph.
Background
Along with the integration of information technology and manufacturing industry, a 3D model system can realize interaction and integration between a real system and the information system, linkage capacity and running state of the real system can be visually displayed through the 3D model system, the existing 3D modeling and data visualization technology adopts 3D tools such as three.js, unity and the like, and an application interface is developed by combining with an actual application scene to display visualized data, so that the development period is too long, when one of an object model and data changes, the model needs to be revised again, the model is inevitably re-developed correspondingly, and the 3D modeling and the data seriously depend on developers. Therefore, a 3D modeling method capable of enhancing model multiplexing and having automatic data hotspot matching capability is needed.
Disclosure of Invention
The technical problem to be solved by the application is to provide a 3D modeling method which can strengthen model multiplexing and has automatic data hotspot matching capability.
In order to solve the above technical problems, the 3D modeling method based on a knowledge-graph of the present invention comprises the following steps,
s1, establishing a data hotspot library, and carrying out modeling management on 3D model information;
s2, establishing a 3D model association network to obtain the degree of relationship among all devices in the 3D model system;
s3, establishing a 3D model matching network to obtain a data hotspot corresponding to the 3D model;
and S4, analyzing and evaluating the 3D model matching network by comparing the real equipment with the obtained characteristic information of the 3D model.
Further, the step S1 includes using a data hotspot library for hierarchical management of the 3D models in the knowledge graph, classifying the 3D models in the knowledge graph into M types according to device types, each device type being represented by data M, M being greater than or equal to 1 and less than or equal to M, and recording the data hotspot library asN represents the total number of models with the device type m;
also, the same applies toThe data set comprises static attributes, dynamic attributes, alarm attributes and control attributes of the equipment; spro denotes the set of static attribute parameters, spro = [ Spro = [ 1 ,spro 2 ,...,spro R ]R represents the number of static attributes; dpro represents a set of dynamic attribute parameters, dpro = [ Dpro = 1 ,dpro 2 ,...,dpro S ]S represents the number of dynamic attributes; alarm represents an Alarm attribute, if the device has no Alarm function, then Alarm =0, and if the device has an Alarm function, then Alarm =1; ctrl represents a control attribute, and ctrl =1 if control is involved, and ctrl =0 if control is not involved.
The 3D models in the knowledge graph are managed in a grading mode through the method, the 3D models in the knowledge graph correspond to data in the data hotspot library one by one, and the 3D models in the knowledge graph can be rapidly inquired through the received data hotspot information.
Further, in the step S2, the following steps are included,
digital information set, CH, corresponding to the real equipment is obtained by digitally encoding the characteristic information of the real equipment k =[ch k0 ,ch k1 ,ch k2 ,...,ch kN ]In which ch k0 Indicates the device type of device k, ch kn Representing a digital code corresponding to the characteristic information of the device k, wherein N is the quantity of the characteristic information; the device type of each device in the real system can be obtained through the digital information of the real devices, and the equipment is recorded with EQTY = [ eq ] 1 ,eqty 2 ,...eqty K ]Wherein the equality k =ch k0 K is more than or equal to 1 and less than or equal to K, and K represents the number of equipment in a real system;
establishing a 3D model association network, which comprises an input layer, three hidden layers and an output layer; wherein the dimension of the input layer is 1, which is the device type equation of the device k k The dimensionality of an output layer is K, the output information is the association degree between the equipment K and other equipment of the system, the higher the association degree between the two equipment is, the closer the running relation between the equipment is, and a mutual matched equipment model is preferentially selected in the 3D model matching process;
the dimension of the first hidden layer is I, and the weight between the first hidden layer and the input layer is IThe activation function is f 1 (x) The dimension of the second hidden layer is K, and the weight between the second hidden layer and the input layer is KThe activation function is f 2 (x) While device type data EQTY = [ eq = 1 ,eqty 2 ,...eqty K ]Participating in calculation, the dimension of the third hidden layer is I, and the weight value between the third hidden layer and the second hidden layer isThe activation function is f 3 (x) The weight between the third hidden layer and the output layer isThe activation function is f 4 (x) Setting a threshold [ epsilon ] between the third hidden layer and the output layer min ,ε max ]If the output information of the third hidden layer exceeds the threshold value, f 'obtained by linearly converting the output information of the third hidden layer' out3 Sending back the participation operation of the first hidden layer;
obtaining a first hidden layer output ofThe output of the second hidden layer isThe output of the third hidden layer isIf it isThen the output result of the third hidden layer is linearly converted to obtainThen f 'is prepared' out3 Returning to the first hidden layer to participate in the operation. If f out3 ∈[ε min ,ε max ]Then the output of the third hidden layer is sent to the output layer for calculation, and the final output result isRepresenting the degree of relationship between device k and other devices of the system.
By adopting the 3D model association network obtained by the invention, the association degree among all the devices in the system can be quickly obtained, so that the 3D model system obtained in the model matching process can run more stably and is more similar to the real devices.
Further, in the above step, the output information RELAT of the output layer is finally output k =[relat 1 ,relat 2 ,...,relat K ]And sample information RELAT' k =[relat′ 1 ,relat′ 2 ,...,relat′ K ]Comparing to test the network training effect, and setting a threshold value epsilon according to the actual situation to orderAnd comparing the result with a threshold value to judge whether the training is finished.
Further, in the step S3, the following steps are included,
the 3D model matching network includes three input layers, three output layers and five hidden layers. Wherein the dimension of the first input layer is N, and the input is digital information corresponding to the equipment, namely CH k =[ch k1 ,ch k2 ,...,ch kN ]The output of the first output layer is the alarm attribute and the control attribute of the equipment; the output of the second output layer is the static attribute of the equipment; the output of the third output layer is the dynamic properties of the device. The combination of the three output layers is the final output of the network, and is a data hotspot of the matched 3D model.
Specifically, the dimension of the first hidden layer is 2, and the activation function is g 1 (x) The weight between the first input layer and the first hidden layer isThe output of the first hidden layer passes a threshold functionThe judgment is made, and the threshold value is set to [0,1 ]]If the condition is met, entering a first output layer, and if the condition is not met, participating an adjustment parameter rho (rho is more than 0 and less than 1) in the operation of the first hidden layer, wherein the dimensionality of the first output layer is 2, the Alarm attribute Alarm and the control attribute ctrl of the equipment are represented, and the output result is 0 or 1;
the output of the first hidden layer isThe output result of the first output layer isWherein sigma is a threshold parameter set by the first output layer;
combining the input information of the first input layer and the output information of the first output layer as the input of the second input layer, the dimension of the second input layer is N +2, the dimension of the second hidden layer is K, and the activation function is g 2 (x) The weight between the second input layer and the second hidden layer isOutput result device relation RELAT of simultaneous 3D model correlation network k =[relat 1 ,relat 2 ,...,relat K ]Participating in the operation of the second hidden layer, setting a third hidden layer to perform convergence stabilization on the network in order to prevent the introduction of the equipment relation from causing large fluctuation on the network, wherein the dimensionality of the third hidden layer is U, and the activation function is g 3 (x) And the weight between the second hidden layer and the first hidden layer isThe dimension of the second output layer is R, and the weight between the third hidden layer and the second output layer isRepresenting the static property of the device and the output result is Spro k =[spro k1 ,spro k2 ,...,spro kR ];
Input on the second input layer is recorded asThe output of the second hidden layer isThe output result of the second output layer is
Taking the input information of the first input layer, the information of the first output layer and the information of the second output layer as the input of the third input layer, the dimensionality of the third input layer is N + R +2, the dimensionality of the fourth hidden layer is K, and the activation function is g 4 (x) The weight between the third input layer and the fourth hidden layer isOutput result device relation RELAT of simultaneous 3D model correlation network k =[relat 1 ,relat 2 ,...,relat K ]Participating in the operation of the fourth hidden layer, setting the fifth hidden layer to perform convergence stabilization on the network in order to prevent the introduction of the equipment relation degree from causing large fluctuation on the network, wherein the dimensionality of the fifth hidden layer is V, and the activation function is g 5 (x) And the weight between the fourth hidden layer and the fourth hidden layer isThe dimension of the third output layer is S, and the preset value between the fifth hidden layer and the third output layer is SThe dynamic property of the equipment is represented, and the output result is Dpro k =[dpro k1 ,dpro k2 ,...,dpro kS ]。
The input of the third input layer is recorded as The output of the fourth hidden layer isThe output result of the third output layer is
Combining the output results of the three output layers, i.e. RST k =[Spro k ,Dpro k ,Alarm,ctrl]Matching the final output result of the network for the 3D model; training is accomplished by performing a large sample set learning on the 3D model matching network.
The 3D model matching network has high convergence rate, can quickly select the optimal 3D model from millions of models, and has high matching degree.
Further, in the step S4, the following steps are included,
output result RST obtained by matching 3D model with network k =[Spro k ,Dpro k ,Alarm,ctrl]Comparing with the data hot spot database to obtain the data hot spotsBy data hot-spotsObtaining a 3D model in a knowledge graph;
establishing a model characteristic information base, and carrying out characteristic information management on the 3D models in the knowledge graph, so that each 3D model corresponds to a characteristic information set
Comparing the feature information of the 3D model obtained through the 3D model matching network with the feature digital information of the real equipment, wherein the method comprises the following steps:
setting threshold [0, gamma ]]If, ifThe calculation result of the 3D model matching network is not suitable for the system, and the network needs to be retrained; the threshold value adopts 0 as a lower limit, so that the overall performance of the selected 3D model is not lower than that of real equipment, and system model errors caused by insufficient performance of the 3D model are prevented;
and if the analysis result does not meet the requirement, the 3D model matching network reselects the sample set or adds the sample set, and the network is trained and updated through the new sample set to obtain a more appropriate network matching algorithm.
By adopting the analysis method, the applicability of the 3D model matching result can be effectively judged, and the phenomenon that the matching result does not meet the actual requirement due to incomplete sample set of the 3D model matching network is avoided.
The invention also relates to an application of the knowledge graph-based 3D modeling method in machine learning.
The invention also relates to a system for operating the knowledge-graph-based 3D modeling method.
The invention has at least the following beneficial effects:
(1) And the 3D model in the knowledge graph is managed in a grading way, and the 3D model in the knowledge graph can be quickly inquired through the received data hot spot information.
(2) By adopting the 3D model association network, the association degree among all the devices in the system is quickly obtained, so that the 3D model obtained in the model matching process is more similar to the real device.
(3) The 3D model matching network is adopted, the convergence speed is high, the optimal 3D model can be selected from millions of models quickly, and the matching degree is high.
(4) The 3D model matching result is analyzed, the applicability of the 3D model matching result can be effectively judged, and the situation that the matching result does not meet the actual requirement due to incomplete sample set of the 3D model matching network is avoided.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below.
FIG. 1 is a flow chart of a method of knowledge-graph based 3D modeling of the present invention.
Fig. 2 is a diagram of a 3D model association network structure of the present invention.
Fig. 3 is a diagram of a 3D model matching network architecture of the present invention.
Detailed Description
For a better understanding of the present invention, reference will now be made in detail to the accompanying drawings and specific examples.
The invention discloses a 3D modeling method based on a knowledge graph, which refers to FIG. 1, and comprises the following specific processing procedures:
s1, establishing a data hotspot library, and carrying out modeling hierarchical management on the 3D model information.
In order to match the best 3D model from the knowledge graph, the invention provides a data hotspot library for carrying out hierarchical management on the 3D model in the knowledge graph, the 3D model is classified firstly, the 3D model in the knowledge graph is divided into M types according to equipment types, each equipment type is represented by data M, M is more than or equal to 1 and less than or equal to M, and the data hotspot library is marked asN represents the total number of models with device type m. The 3D models of the knowledge spectrogram are divided according to the equipment types, so that the range of the matched models can be quickly reduced from a ten million-order 3D model library.
The same applies toThe data set comprises static attributes, dynamic attributes, alarm attributes and control attributes of the equipment. Spro denotes the set of static attribute parameters, spro = [ Spro = [ 1 ,spro 2 ,...,spro R ]R represents the number of static attributes; dpro represents a set of dynamic attribute parameters, dpro =[dpro 1 ,dpro 2 ,...,dpro S ]S represents the number of dynamic attributes; the Alarm attribute is represented by Alarm, if the equipment has no Alarm function, then Alarm =0, and if the equipment has the Alarm function, then Alarm =1; ctrl represents a control attribute, and ctrl =1 if the control is involved, and ctrl =0 if the control is not involved.
By the method, the 3D models in the knowledge graph are managed in a grading mode, the 3D models in the knowledge graph correspond to the data in the data hot spot database one by one, and the 3D models in the knowledge graph can be quickly inquired through the received data hot spot information.
And S2, establishing a 3D model association network to obtain the degree of relationship among all devices in the 3D model system.
Digital information set, CH, corresponding to the real equipment is obtained by digitally encoding the characteristic information of the real equipment k =[ch k0 ,ch k1 ,ch k2 ,...,ch kN ]Wherein ch k0 Indicates the device type of device k, ch kn And representing the digital code corresponding to the characteristic information of the equipment k, wherein N is the quantity of the characteristic information. The equipment type of each equipment in a real system can be obtained through the digital information of the real equipment, and the EQTY = [ eq ] is recorded 1 ,eqty 2 ,...eqty K ]Wherein the equality k =ch k0 K is more than or equal to 1 and less than or equal to K, and K represents the number of equipment in a real system.
Referring to fig. 2, a 3D model correlation network is established, which includes an input layer, three hidden layers and an output layer. Wherein the dimension of the input layer is 1, which is the device type equation of the device k k The dimensionality of the output layer is K, the output information is the association degree between the device K and other devices of the system, the higher the association degree between the two devices is, the closer the operation relation between the devices is, and the matched device models are preferentially selected in the 3D model matching process.
The dimension of the first hidden layer is I, and the weight between the first hidden layer and the input layer is IThe activation function is f 1 (x) The dimension of the second hidden layer is K,the weight between the input layer and the input layer isThe activation function is f 2 (x) While device type data EQTY = [ eq = 1 ,eqty 2 ,...,eqty K ]Participating in calculation, the dimension of the third hidden layer is I, and the weight value between the third hidden layer and the second hidden layer isActivation function of f 3 (x) The weight between the third hidden layer and the output layer isA threshold value epsilon is set between the third hidden layer and the output layer min ,ε max ]If the output information of the third hidden layer exceeds the threshold value, f 'obtained by linearly converting the output information of the third hidden layer' out3 The first hidden layer participation operation is sent back.
Obtaining a first hidden layer output ofThe output of the second hidden layer isThe output of the third hidden layer isIf it isThe output result of the third hidden layer is linearly converted to obtainThen, f 'is' out3 Returning to the first hidden layer to participate in the operation. If f out3 ∈[ε min ,ε max ]Then the output of the third hidden layer is sent to the output layer for calculation, and the final output result isRepresenting the degree of relationship between device k and other devices of the system.
As an embodiment of the present invention, let the activation function of the first hidden layerActivation function of the second hidden layerActivation function of third hidden layerThe output of the first hidden layer isThe output of the second hidden layer isThe output of the third hidden layer isThe output of the output layer is
Finally, the output information RELAT of the output layer k =[relat 1 ,relat 2 ,...,relat K ]And sample information RELAT' k =[relat′ 1 ,relat′ 2 ,...,relat′ K ]Comparing to test the network training effect, setting a threshold value epsilon according to the actual situation, and makingAnd comparing the result with a threshold value to judge whether the training is finished.
By adopting the 3D model association network, the association degree among all the devices in the system can be quickly obtained, so that the 3D model system obtained in the model matching process can run more stably and is more similar to the real devices.
And S3, establishing a 3D model matching network to obtain data hotspots corresponding to the 3D model.
Referring to fig. 3, a 3D model matching network is established, and data hotspots of the most suitable 3D model are obtained by analyzing digital information of the device. The 3D model matching network includes three input layers, three output layers and five hidden layers. The dimension of the first input layer is N, and the input is digital information corresponding to equipment, namely CH k =[ch k1 ,ch k2 ,...,ch kN ]The output of the first output layer is the alarm attribute and the control attribute of the equipment; the output of the second output layer is the static attribute of the equipment; the output of the third output layer is the dynamic properties of the device. The combination of the three output layers is the final output of the network, and is a data hotspot of the matched 3D model.
The dimension of the first hidden layer is 2, and the activation function is g 1 (x) The weight between the first input layer and the first hidden layer isThe output of the first hidden layer is determined by a threshold function, setting the threshold to 0,1]If the condition is met, entering a first output layer, and if the condition is not met, participating the adjustment parameter rho (rho is more than 0 and less than 1) in the first hidden layer operation, wherein the dimensionality of the first output layer is 2, the Alarm attribute Alarm and the control attribute ctrl of the equipment are represented, and the output result is 0 or 1.
The output of the first hidden layer isThe output result of the first output layer isWhere σ is a threshold parameter set by the first output layer.
As an embodiment of our invention, let the activation function of the first hidden layerThen there is
Combining the input information of the first input layer and the output information of the first output layer as the input of the second input layer, the dimension of the second input layer is N +2, the dimension of the second hidden layer is K, and the activation function is g 2 (x) The weight between the second input layer and the second hidden layer isDevice relationship RELAT for output results of simultaneous 3D model correlation network k =[relat 1 ,relat 2 ,...,relat K ]Participating in the operation of the second hidden layer, setting a third hidden layer to perform convergence stabilization on the network in order to prevent the introduction of the equipment relation from causing large fluctuation on the network, wherein the dimensionality of the third hidden layer is U, and the activation function is g 3 (x) And the weight between the second hidden layer and the second hidden layer isThe dimension of the second output layer is R, and the weight between the third hidden layer and the second output layer isRepresenting the static property of the device and the output result is Spro k =[spro k1 ,spro k2 ,...,spro kR ]。
Input on the second input layer is recorded asThe output of the second hidden layer isThe output result of the second output layer is
As an embodiment of the present invention, let the activation function of the second hidden layerActivation function g of the third hidden layer 3 (x)=exp(-x/e -x ) Then, the output of the second hidden layer and the output of the second output layer are:
taking the input information of the first input layer and the information of the first output layer and the second output layer as the input of the third input layer, the dimension of the third input layer is N + R +2, the dimension of the fourth hidden layer is K, and the activation function is g 4 (x) The weight between the third input layer and the fourth hidden layer isDevice relationship RELAT for output results of simultaneous 3D model correlation network k =[relat 1 ,relat 2 ,...,relat K ]Participating in the operation of the fourth hidden layer, setting the fifth hidden layer to perform convergence stabilization on the network in order to prevent the introduction of the equipment relationship degree from causing great fluctuation on the network, wherein the dimensionality of the fifth hidden layer is V, and the activation function is g 5 (x) And the weight between the fourth hidden layer and the fourth hidden layer isThe dimension of the third output layer is S, and the front position between the fifth hidden layer and the third output layer is SThe dynamic property of the equipment is represented, and the output result is Dpro k =[dpro k1 ,dpro k2 ,...,dpro kS ]。
The input of the third input layer is recorded as The output of the fourth hidden layer isThe output result of the third output layer is
As an embodiment of the present invention, let the activation function of the fourth hidden layerActivation function of fifth hidden layerThe output of the fourth hidden layer and the output of the third output layer are:
combining the output results of the three output layers, i.e. RST k =[Spro k ,Dpro k ,Alarm,ctrl]And matching the final output result of the network for the 3D model. Training is accomplished by performing a large sample set learning on the 3D model matching network.
The 3D model matching network has high convergence rate, can quickly select the optimal 3D model from millions of models, and has high matching degree.
And S4, evaluating and analyzing the 3D model matching network by comparing the real equipment with the obtained characteristic information of the 3D model.
Output result RST obtained by matching 3D model with network k =[Spro k ,Dpro k ,Alarm,ctrl]Comparing with the data hotspot library to obtain data hotspotsBy data hot spotsA 3D model in the knowledge-graph is obtained.
Establishing a model characteristic information base, and carrying out characteristic information management on the 3D models in the knowledge graph, so that each 3D model corresponds to a characteristic information set
Comparing the feature information of the 3D model obtained through the 3D model matching network with the feature digital information of the real equipment, wherein the method comprises the following steps:
setting threshold [0, gamma ]]If at allThe 3D model matching network computation results are not applicable to the system and the network needs to be retrained. The threshold value adopts 0 as the lower limit, so that the overall performance of the selected 3D model is not lower than that of real equipment, and system model errors caused by insufficient performance of the 3D model are prevented.
And if the analysis result does not meet the requirement, the 3D model matching network reselects the sample set or adds the sample set, and the network is trained and updated through the new sample set to obtain a more appropriate network matching algorithm.
By adopting the analysis method, the applicability of the 3D model matching result can be effectively judged, and the condition that the matching result does not meet the actual requirement due to incomplete sample set of the 3D model matching network is avoided.
In conclusion, the 3D modeling method based on the knowledge graph is realized.
The following detailed description will be provided with reference to the drawings in the present embodiment, so that how to apply the technical means to solve the technical problems and achieve the technical effects can be fully understood and implemented. It should be noted that, as long as there is no conflict, the features in the embodiments of the present invention may be combined with each other, and the formed technical solutions are within the scope of the present invention.
Claims (7)
1. A3D modeling method based on knowledge graph is characterized in that: comprises the following steps of (a) carrying out,
s1, establishing a data hotspot library, and carrying out modeling management on 3D model information;
s2, establishing a 3D model association network to obtain the degree of relationship among all devices in the 3D model system;
s3, establishing a 3D model matching network to obtain a data hotspot corresponding to the 3D model;
s4, analyzing and evaluating the 3D model matching network by comparing the real equipment with the obtained characteristic information of the 3D model;
in the step S2, the method comprises the following steps:
digital information set, CH, corresponding to the real equipment is obtained by digitally encoding the characteristic information of the real equipment k =[ch k0 ,ch k1 ,ch k2 ,...,ch kN ]Wherein ch k0 Indicates the device type of device k, ch kn Representing digital codes corresponding to the characteristic information of the equipment k, wherein N is the quantity of the characteristic information; the device type of each device in the real system can be obtained through the digital information of the real devices, and the equipment is recorded with EQTY = [ eq ] 1 ,eqty 2 ,...eqty K ]Wherein the equality k =ch k0 K is more than or equal to 1 and less than or equal to K, and K represents the number of equipment in a real system;
establishing a 3D model association network, which comprises an input layer, three hidden layers and an output layer; wherein the dimension of the input layer is 1, which is the device type equation of the device k k The dimensionality of the output layer is K, the output information is the association degree between the device K and other devices of the system, the higher the association degree between the two devices is, the closer the operation relation between the devices is, and the matched device models are preferentially selected in the 3D model matching process;
the dimension of the first hidden layer is I, and the weight between the first hidden layer and the input layer is IThe activation function is f 1 (x) The dimension of the second hidden layer is K, and the weight between the second hidden layer and the input layer is KThe activation function is f 2 (x) While device type data EQTY = [ eq = 1 ,eqty 2 ,...eqty K ]Participating in calculation, the dimension of the third hidden layer is I, and the weight value between the third hidden layer and the second hidden layer isThe activation function is f 3 (x) The weight between the third hidden layer and the output layer isThe activation function is f 4 (x) Setting a threshold [ epsilon ] between the third hidden layer and the output layer min ,ε max ]If the output information of the third hidden layer exceeds the threshold value, f 'obtained by linearly converting the output information of the third hidden layer' out3 Sending back the first hidden layer participation operation;
obtaining a first hidden layer output ofThe output of the second hidden layer isThe output of the third hidden layer isIf it isThen to the firstThe output result of the three hidden layers is linearly converted to obtainThen, f 'is' out3 Returning to the first hidden layer to participate in the operation; if f out3 ∈[ε min ,ε max ]Then the output of the third hidden layer is sent to the output layer for calculation, and the final output result isRepresenting the degree of relationship between the device k and other devices of the system;
in the step S3, the following steps are included:
the 3D model matching network comprises three input layers, three output layers and five hidden layers; wherein the dimension of the first input layer is N, and the input is digital information corresponding to the equipment, namely CH k =[ch k1 ,ch k2 ,...,ch kN ]The output of the first output layer is the alarm attribute and the control attribute of the equipment; the output of the second output layer is the static attribute of the equipment; the output of the third output layer is the dynamic attribute of the equipment; the combination of the three output layers is the final output of the network, and is a data hotspot of the matched 3D model.
2. The knowledge-graph based 3D modeling method according to claim 1, characterized in that: in the step S1, the following steps are included,
using a data hotspot library for carrying out hierarchical management on the 3D models in the knowledge graph, classifying the 3D models, dividing the 3D models in the knowledge graph into M types according to equipment types, wherein each equipment type is represented by data M, M is more than or equal to 1 and less than or equal to M, and recording the data hotspot library asN represents the total number of models with the equipment type m;
also, the same applies toThe data set comprises static attributes, dynamic attributes, alarm attributes and control attributes of the equipment; spro denotes the set of static attribute parameters, spro = [ Spro = [ 1 ,spro 2 ,...,spro R ]R represents the number of static attributes; dpro represents a set of dynamic attribute parameters, dpro = [ Dpro = 1 ,dpro 2 ,...,dpro S ]S represents the number of dynamic attributes; the Alarm attribute is represented by Alarm, if the equipment has no Alarm function, then Alarm =0, and if the equipment has the Alarm function, then Alarm =1; ctrl represents a control attribute, and ctrl =1 if control is involved, and ctrl =0 if control is not involved.
3. The knowledge-graph based 3D modeling method according to claim 1, characterized in that: finally, the output information RELAT of the output layer k =[relat 1 ,relat 2 ,...,relat K ]And sample information RELAT' k =[relat′ 1 ,relat′ 2 ,...,relat′ K ]Comparing to test the network training effect, setting a threshold value epsilon according to the actual situation, and makingAnd comparing the result with a threshold value to judge whether the training is finished.
4. The knowledge-graph based 3D modeling method according to claim 1, characterized in that: the first hidden layer has dimension 2 and the activation function is g 1 (x) The weight between the first input layer and the first hidden layer isThe output of the first hidden layer is determined by a threshold function, setting the threshold to [0,1 ]]If the condition is met, entering a first output layer, and if the condition is not met, participating an adjustment parameter rho (rho is more than 0 and less than 1) in the operation of the first hidden layer, wherein the dimensionality of the first output layer is 2, the Alarm attribute Alarm and the control attribute ctrl of the equipment are represented, and the output result is 0 or 1;
the output of the first hidden layer isThe output result of the first output layer isWherein sigma is a threshold parameter set by the first output layer;
combining the input information of the first input layer and the output information of the first output layer as the input of the second input layer, the dimension of the second input layer is N +2, the dimension of the second hidden layer is K, and the activation function is g 2 (x) The weight between the second input layer and the second hidden layer isOutput result device relation RELAT of simultaneous 3D model correlation network k =[relat 1 ,relat 2 ,...,relat K ]Participating in the operation of the second hidden layer, setting a third hidden layer to perform convergence stabilization on the network in order to prevent the introduction of the equipment relation degree from causing large fluctuation on the network, wherein the dimensionality of the third hidden layer is U, and the activation function is g 3 (x) And the weight between the second hidden layer and the first hidden layer isThe dimension of the second output layer is R, and the weight between the third hidden layer and the second output layer isRepresenting the static properties of the device with the output being Spro k =[spro k1 ,spro k2 ,...,spro kR ];
The input of the second input layer is recorded asThe output of the second hidden layer isThe output result of the second output layer is
Taking the input information of the first input layer, the information of the first output layer and the information of the second output layer as the input of the third input layer, the dimension of the third input layer is N + R +2, the dimension of the fourth hidden layer is K, and the activation function is g 4 (x) The weight between the third input layer and the fourth hidden layer isOutput result device relation RELAT of simultaneous 3D model correlation network k =[relat 1 ,relat 2 ,...,relat K ]Participating in the operation of the fourth hidden layer, setting the fifth hidden layer to perform convergence stabilization on the network in order to prevent the introduction of the equipment relation degree from causing large fluctuation on the network, wherein the dimensionality of the fifth hidden layer is V, and the activation function is g 5 (x) And the weight between the fourth hidden layer and the fourth hidden layer isThe dimension of the third output layer is S, and the preset value between the fifth hidden layer and the third output layer is SThe dynamic property of the equipment is represented, and the output result is Dpro k =[dpro k1 ,dpro k2 ,...,dpro kS ];
The input of the third input layer is recorded as(w is not less than 1 and not more than N + R + 2), the output of the fourth hidden layer isThe output result of the third output layer is
Combining the output results of the three output layers, i.e. RST k =[Spro k ,Dpro k ,Alarm,ctrl]Matching the final output result of the network for the 3D model; training is accomplished by performing a large sample set learning on the 3D model matching network.
5. The knowledge-graph based 3D modeling method according to claim 4, characterized in that: in the step S4, the following steps are included,
output result RST obtained by matching 3D model with network k =[Spro k ,Dpro k ,Alarm,ctrl]Comparing with the data hotspot library to obtain data hotspotsBy data hot-spotsObtaining a 3D model in a knowledge graph;
establishing a model characteristic information base, and carrying out characteristic information management on the 3D models in the knowledge graph, so that each 3D model corresponds to a characteristic information set
Comparing the feature information of the 3D model obtained through the 3D model matching network with the feature digital information of the real equipment, wherein the method comprises the following steps:
set threshold [0, γ ]]If, ifThe calculation result of the 3D model matching network is not suitable for the system, and the network needs to be retrained; the threshold value adopts 0 as the lower limit, so that the overall performance of the selected 3D model is not lower than that of real equipment, and system model errors caused by insufficient performance of the 3D model are prevented;
and if the analysis result does not meet the requirement, the 3D model matching network reselects the sample set or adds the sample set, and the network is trained and updated through the new sample set to obtain a more appropriate network matching algorithm.
6. Use of a knowledge-graph based 3D modeling method according to any of claims 1-5 in machine learning.
7. A system for operating the knowledge-graph based 3D modeling method of any one of claims 1-5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110261587.1A CN112989063B (en) | 2021-03-10 | 2021-03-10 | 3D modeling method and system based on knowledge graph |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110261587.1A CN112989063B (en) | 2021-03-10 | 2021-03-10 | 3D modeling method and system based on knowledge graph |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112989063A CN112989063A (en) | 2021-06-18 |
CN112989063B true CN112989063B (en) | 2023-02-10 |
Family
ID=76334859
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110261587.1A Active CN112989063B (en) | 2021-03-10 | 2021-03-10 | 3D modeling method and system based on knowledge graph |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112989063B (en) |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103885788A (en) * | 2014-04-14 | 2014-06-25 | 焦点科技股份有限公司 | Dynamic WEB 3D virtual reality scene construction method and system based on model componentization |
CN104915522A (en) * | 2015-07-01 | 2015-09-16 | 华东理工大学 | Mixed modeling method and system based on combination of process priors and data-driven model |
CN105468704A (en) * | 2015-11-18 | 2016-04-06 | 中国传媒大学 | Quick ideas generation method for stage artistic scene design |
CN108280873A (en) * | 2018-01-05 | 2018-07-13 | 上海户美信息科技有限公司 | Model space position capture and hot spot automatically generate processing system |
CN108596329A (en) * | 2018-05-11 | 2018-09-28 | 北方民族大学 | Threedimensional model sorting technique based on end-to-end Deep integrating learning network |
CN108960288A (en) * | 2018-06-07 | 2018-12-07 | 山东师范大学 | Threedimensional model classification method and system based on convolutional neural networks |
CN109213884A (en) * | 2018-11-26 | 2019-01-15 | 北方民族大学 | A kind of cross-module state search method based on Sketch Searching threedimensional model |
CN109544703A (en) * | 2018-12-29 | 2019-03-29 | 杭州长宽数字科技有限公司 | It is a kind of to be easily achieved interactive data center Web3D model loading method |
CN109961436A (en) * | 2019-04-04 | 2019-07-02 | 北京大学口腔医学院 | A kind of median plane construction method based on artificial nerve network model |
CN109992670A (en) * | 2019-04-04 | 2019-07-09 | 西安交通大学 | A kind of map completion method of knowledge based map neighbour structure |
CN110599592A (en) * | 2019-09-12 | 2019-12-20 | 北京工商大学 | Three-dimensional indoor scene reconstruction method based on text |
CN111666313A (en) * | 2020-05-25 | 2020-09-15 | 中科星图股份有限公司 | Correlation construction and multi-user data matching method based on multi-source heterogeneous remote sensing data |
CN112073474A (en) * | 2020-08-19 | 2020-12-11 | 深圳市国鑫恒运信息安全有限公司 | Js-based intelligent data center management method and system |
-
2021
- 2021-03-10 CN CN202110261587.1A patent/CN112989063B/en active Active
Patent Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103885788A (en) * | 2014-04-14 | 2014-06-25 | 焦点科技股份有限公司 | Dynamic WEB 3D virtual reality scene construction method and system based on model componentization |
CN104915522A (en) * | 2015-07-01 | 2015-09-16 | 华东理工大学 | Mixed modeling method and system based on combination of process priors and data-driven model |
CN105468704A (en) * | 2015-11-18 | 2016-04-06 | 中国传媒大学 | Quick ideas generation method for stage artistic scene design |
CN108280873A (en) * | 2018-01-05 | 2018-07-13 | 上海户美信息科技有限公司 | Model space position capture and hot spot automatically generate processing system |
CN108596329A (en) * | 2018-05-11 | 2018-09-28 | 北方民族大学 | Threedimensional model sorting technique based on end-to-end Deep integrating learning network |
CN108960288A (en) * | 2018-06-07 | 2018-12-07 | 山东师范大学 | Threedimensional model classification method and system based on convolutional neural networks |
CN109213884A (en) * | 2018-11-26 | 2019-01-15 | 北方民族大学 | A kind of cross-module state search method based on Sketch Searching threedimensional model |
CN109544703A (en) * | 2018-12-29 | 2019-03-29 | 杭州长宽数字科技有限公司 | It is a kind of to be easily achieved interactive data center Web3D model loading method |
CN109961436A (en) * | 2019-04-04 | 2019-07-02 | 北京大学口腔医学院 | A kind of median plane construction method based on artificial nerve network model |
CN109992670A (en) * | 2019-04-04 | 2019-07-09 | 西安交通大学 | A kind of map completion method of knowledge based map neighbour structure |
CN110599592A (en) * | 2019-09-12 | 2019-12-20 | 北京工商大学 | Three-dimensional indoor scene reconstruction method based on text |
CN111666313A (en) * | 2020-05-25 | 2020-09-15 | 中科星图股份有限公司 | Correlation construction and multi-user data matching method based on multi-source heterogeneous remote sensing data |
CN112073474A (en) * | 2020-08-19 | 2020-12-11 | 深圳市国鑫恒运信息安全有限公司 | Js-based intelligent data center management method and system |
Also Published As
Publication number | Publication date |
---|---|
CN112989063A (en) | 2021-06-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11176715B2 (en) | Method and system for color representation generation | |
Sedlmair et al. | Visual parameter space analysis: A conceptual framework | |
Rieck et al. | Persistent homology for the evaluation of dimensionality reduction schemes | |
Cartone et al. | Does economic convergence hold? A spatial quantile analysis on European regions | |
CN109598279A (en) | Based on the zero sample learning method for generating network from coding confrontation | |
JP2023523029A (en) | Image recognition model generation method, apparatus, computer equipment and storage medium | |
Zheng et al. | A review of QoE research progress in metaverse | |
Roewer-Despres et al. | Towards finite element simulation using deep learning | |
CN111709914A (en) | Non-reference image quality evaluation method based on HVS characteristics | |
CN108241898A (en) | A kind of resource allocation methods and device | |
CN112862669A (en) | Training method, generating method, device and equipment of image generation model | |
CN113642666A (en) | Active enhanced soft measurement method based on sample expansion and screening | |
US20080004855A1 (en) | Design support apparatus, design support method, and design support program | |
CN111159241B (en) | Click conversion estimation method and device | |
CN114546365B (en) | Flow visualization modeling method, server, computer system and medium | |
Wang | Cluster analysis using a validated self‐organizing method: cases of problem identification | |
Siripornpitak et al. | Spatial up-sampling of HRTF sets using generative adversarial networks: A pilot study | |
CN112989063B (en) | 3D modeling method and system based on knowledge graph | |
Zhu et al. | Arrow’s impossibility theorem inspired subjective image quality assessment approach | |
Meygret et al. | Segmentation of optical flow and 3D data for the interpretation of mobile objects | |
Steinbach et al. | Machine learning state-of-the-art with uncertainties | |
CN111429175B (en) | Method for predicting click conversion under sparse characteristic scene | |
CN110457329A (en) | A kind of method and device for realizing personalized recommendation | |
Fatima et al. | Grey is the new RGB: How good is GAN-based image colorization for image compression? | |
Kamel et al. | A New Approach for Model Selection with Two Qualitative Regressors |
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 |