CN115580564B - Dynamic calling device for communication gateway of Internet of things - Google Patents

Dynamic calling device for communication gateway of Internet of things Download PDF

Info

Publication number
CN115580564B
CN115580564B CN202211397669.XA CN202211397669A CN115580564B CN 115580564 B CN115580564 B CN 115580564B CN 202211397669 A CN202211397669 A CN 202211397669A CN 115580564 B CN115580564 B CN 115580564B
Authority
CN
China
Prior art keywords
feature map
training
scale
gateway
internet
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
Application number
CN202211397669.XA
Other languages
Chinese (zh)
Other versions
CN115580564A (en
Inventor
张云超
王延鹏
石辉
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.)
Shenzhen Airbridge Telecommunication Technologies Co ltd
Original Assignee
Shenzhen Airbridge Telecommunication Technologies Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Airbridge Telecommunication Technologies Co ltd filed Critical Shenzhen Airbridge Telecommunication Technologies Co ltd
Priority to CN202211397669.XA priority Critical patent/CN115580564B/en
Publication of CN115580564A publication Critical patent/CN115580564A/en
Application granted granted Critical
Publication of CN115580564B publication Critical patent/CN115580564B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/16—Threshold monitoring
    • 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
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y10/00—Economic sectors
    • G16Y10/75—Information technology; Communication
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y20/00—Information sensed or collected by the things
    • G16Y20/20—Information sensed or collected by the things relating to the thing itself
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y30/00—IoT infrastructure
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00—IoT characterised by the purpose of the information processing
    • G16Y40/10—Detection; Monitoring
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00—IoT characterised by the purpose of the information processing
    • G16Y40/20—Analytics; Diagnosis
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L12/00—Data switching networks
    • H04L12/66—Arrangements for connecting between networks having differing types of switching systems, e.g. gateways
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0805—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters by checking availability
    • H04L43/0817—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters by checking availability by checking functioning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/12—Protocols specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Abstract

The application relates to the field of internet, and particularly discloses a dynamic calling device for an internet of things communication gateway. Therefore, the communication gateway of the Internet of things can be dynamically called according to actual conditions, so that the stable operation of gateway software and hardware equipment before dynamic calling can be effectively ensured, and the calling efficiency is improved.

Description

Dynamic calling device for communication gateway of Internet of things
Technical Field
The application relates to the field of the Internet, in particular to a dynamic calling device for a communication gateway of the Internet of things.
Background
The gateway is also called an inter-network connector and a protocol converter, is only used for interconnecting networks with different protocols at a higher layer on a transmission layer, is the most complex network interconnection equipment, can be used for interconnecting a wide area network and a local area network, and is a computer system or equipment serving as a conversion task. With the continuous development of gateway technology, gateway devices are increasingly widely applied to communication networks of the internet of things, and the gateway devices begin to enter the public life, so that the life of people is more intelligent and convenient. However, when the current gateway is called, the problems of poor gateway operation stability, low calling efficiency and the like often exist.
Therefore, an optimized dynamic calling device for the communication gateway of the internet of things is desired, which can dynamically call the communication gateway of the internet of things according to actual conditions so as to improve calling efficiency on the basis of ensuring gateway operation stability.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a dynamic calling device for a communication gateway of the Internet of things, which extracts multi-scale implicit associated features among parameter features in state data of the communication gateway by adopting an artificial intelligence algorithm based on deep learning, returns and generates data index values of the communication gateway of the Internet of things based on fusion feature distribution information of the multi-scale implicit associated features among the parameter features, compares the data index values with a threshold value to detect faults, and further determines whether the communication gateway is called. Therefore, the communication gateway of the Internet of things can be dynamically called according to actual conditions, so that the stable operation of gateway software and hardware equipment before dynamic calling can be effectively ensured, and the calling efficiency is improved.
According to an aspect of the application, a dynamic calling device for a communication gateway of the internet of things is provided, which includes:
the gateway state data acquisition unit is used for acquiring gateway state data of the communication gateway to be called;
the matrix construction unit is used for constructing the gateway state data into a gateway state data matrix, wherein values of all positions in the gateway state data matrix are used for expressing parameter characteristics corresponding to jth data in the ith data type;
the multi-scale convolution detection unit is used for enabling the gateway state data matrix to pass through a double-current network model comprising a first convolution neural network and a second convolution neural network so as to obtain a first scale characteristic diagram and a second scale characteristic diagram;
a fusion unit, configured to fuse the first scale feature map and the second scale feature map to obtain a decoded feature map;
the decoding unit is used for enabling the decoding characteristic graph to pass through a decoder to obtain a decoding value used for representing a data index value of the communication gateway of the Internet of things; and
a call control result generation unit for determining whether the internet of things communication gateway can be called based on a comparison between the decoded value and a predetermined threshold value.
In the above dynamic invoking device for the internet of things communication gateway, the first convolutional neural network uses a first convolutional kernel having a first size, and the second convolutional neural network uses a second convolutional kernel having a second size, where the first size is different from the second size.
In the above dynamic invoking apparatus for an internet of things communication gateway, the first convolutional neural network uses a first convolutional kernel with a first hole rate, the second convolutional neural network uses a second convolutional kernel with a second hole rate, and the first convolutional kernel and the second convolutional kernel have the same size.
In the above dynamic calling apparatus for an internet of things communication gateway, the multi-scale convolution detecting unit includes: a first scale convolutional coding subunit, configured to perform convolutional processing of a first convolutional kernel on input data in forward transfer of a layer using each layer of the first convolutional neural network: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein, the output of the last layer of the first convolutional neural network is the first scale feature map, and the input of the first layer of the first convolutional neural network is a gateway state data matrix; and a second scale convolutional coding subunit, configured to perform convolutional processing of a second convolutional kernel on the input data in forward transfer of layers using each layer of the second convolutional neural network: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the second convolutional neural network is the second scale characteristic diagram, and the input of the first layer of the second convolutional neural network is a gateway state data matrix.
In the above dynamic call device for an internet of things communication gateway, the fusion unit is further configured to: fusing the first scale feature map and the second scale feature map according to the following formula to obtain a decoding feature map; wherein the formula is:
F
Figure 100002_DEST_PATH_IMAGE001
wherein
Figure 358423DEST_PATH_IMAGE002
Represents the first scale feature map, and>
Figure 100002_DEST_PATH_IMAGE003
representing the second scale feature map, F denotes the decoded feature map, ->
Figure 948805DEST_PATH_IMAGE004
A weighting parameter, which represents the first scale feature map quantity and the second scale feature map respectively, is selected>
Figure 100002_DEST_PATH_IMAGE005
Indicating a position by bitAnd (4) adding.
In the above dynamic invoking device for an internet of things communication gateway, the decoding unit is further configured to: decoding the decoding characteristic graph by using the decoder to perform decoding regression according to the following formula to obtain a decoding value; wherein the formula is:
Figure 646502DEST_PATH_IMAGE006
wherein->
Figure 100002_DEST_PATH_IMAGE007
Represents the decoded feature map, is selected>
Figure 612797DEST_PATH_IMAGE008
Is the decoded value, is greater than or equal to>
Figure 100002_DEST_PATH_IMAGE009
Is a weight matrix, is->
Figure 737748DEST_PATH_IMAGE010
Representing a matrix multiplication.
In the above dynamic invocation device for the communication gateway of the internet of things, the invocation control result generation unit is configured to determine that the communication gateway of the internet of things can be invoked in response to the decoded value being greater than or equal to the predetermined threshold value.
In the above dynamic calling device for the communication gateway of the internet of things, the device further comprises a training module for training the double-flow network model and the decoder.
In the above-mentioned thing networking communication gateway dynamic call device, the training module includes: the training data acquisition unit is used for acquiring training data, wherein the training data comprises training gateway state data of a communication gateway to be called and a real value of whether the communication gateway of the Internet of things can be called; the training gateway state data matrix is used for acquiring training gateway state data, wherein values of all positions in the training gateway state data matrix are used for expressing parameter characteristics corresponding to jth data in the ith data type; the training multi-scale convolution detection unit is used for enabling the training gateway state data matrix to pass through the double-current network model comprising the first convolution neural network and the second convolution neural network so as to obtain a training first scale feature map and a training second scale feature map; the training fusion unit is used for fusing the training first scale feature map and the training second scale feature map to obtain a training decoding feature map; a training decoding unit, configured to pass the training decoding feature map through the decoder to obtain a decoding loss function value; the intrinsic learning loss unit is used for calculating an intrinsic learning loss function value of a sequence pair sequence response rule based on the distance between a first feature vector obtained after the training of the first scale feature map and a second feature vector obtained after the training of the second scale feature map; and a training unit to compute a weighted sum of the decoding loss function values and the sequence-to-sequence response rule intrinsic learning loss function values as loss function values to train the dual-stream network model and the decoder.
In the above dynamic invoking device for an internet of things communication gateway, the intrinsic learning loss unit is further configured to: calculating the intrinsic learning loss function value of the sequence-to-sequence response rule according to the following formula based on the distance between a first feature vector obtained after the training of the first scale feature map and a second feature vector obtained after the training of the second scale feature map;
wherein the formula is:
Figure 100002_DEST_PATH_IMAGE011
Figure 815425DEST_PATH_IMAGE012
Figure 100002_DEST_PATH_IMAGE013
wherein the content of the first and second substances,
Figure 67546DEST_PATH_IMAGE014
is the first activation feature vector, is asserted>
Figure 100002_DEST_PATH_IMAGE015
Is the second activation characteristic vector, is greater than>
Figure 406123DEST_PATH_IMAGE016
Is the value of the loss of learning function intrinsic to the sequence response rule, in question>
Figure 100002_DEST_PATH_IMAGE017
Is a first feature vector obtained after the expansion of the training first scale feature map, is selected>
Figure 577342DEST_PATH_IMAGE018
Is a second feature vector obtained after the training second scale feature map is expanded, and ^ is greater than or equal to>
Figure 100002_DEST_PATH_IMAGE019
And &>
Figure 752102DEST_PATH_IMAGE020
Is the weight matrix of the decoder for the first eigenvector and the second eigenvector, respectively>
Figure 100002_DEST_PATH_IMAGE021
Represents->
Figure 791602DEST_PATH_IMAGE022
An activation function +>
Figure 100002_DEST_PATH_IMAGE023
Represents->
Figure 469840DEST_PATH_IMAGE024
An activation function +>
Figure 100002_DEST_PATH_IMAGE025
Represents a matrix multiplication,. Sup.>
Figure 936593DEST_PATH_IMAGE026
Representing the euclidean distance between the two vectors.
According to another aspect of the application, a method for using a dynamic calling device of an internet of things communication gateway is provided, which includes:
acquiring gateway state data of a communication gateway to be called;
constructing the gateway state data into a gateway state data matrix, wherein values of all positions in the gateway state data matrix are used for representing parameter characteristics corresponding to jth data in the ith data type;
enabling the gateway state data matrix to pass through a double-flow network model comprising a first convolutional neural network and a second convolutional neural network to obtain a first scale characteristic diagram and a second scale characteristic diagram;
fusing the first scale feature map and the second scale feature map to obtain a decoding feature map;
enabling the decoding characteristic graph to pass through a decoder to obtain a decoding value used for representing a data index value of the communication gateway of the Internet of things; and
determining whether the IOT communication gateway can be invoked based on a comparison between the decoded value and a predetermined threshold.
Compared with the prior art, the dynamic calling device for the communication gateway of the internet of things extracts the multi-scale implicit associated features among the parameter features in the state data of the communication gateway by adopting the artificial intelligence algorithm based on deep learning, returns the fusion feature distribution information of the multi-scale implicit associated features among the parameter features to generate the data index value of the communication gateway of the internet of things, compares the data index value with the threshold value to detect faults, and further determines whether the communication gateway is called. Therefore, the communication gateway of the Internet of things can be dynamically called according to actual conditions, so that the stable operation of gateway software and hardware equipment before dynamic calling can be effectively ensured, and the calling efficiency is improved.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
Fig. 1 is a block diagram of a dynamic invocation device of an internet of things communication gateway according to an embodiment of the present application;
FIG. 2 is an Internet of things according to an embodiment of the application a block diagram of a communication gateway dynamic call device;
fig. 3 is a system architecture diagram of a dynamic invocation device of an internet of things communication gateway according to an embodiment of the present application;
fig. 4 is a block diagram of a multi-scale convolution detection unit in a dynamic calling device of an internet of things communication gateway according to an embodiment of the present application;
fig. 5 is a flowchart of a first convolutional code in a dynamic invoking device of an internet of things communication gateway according to an embodiment of the present application;
fig. 6 is a system architecture diagram of a training module in a dynamic invocation device of an internet of things communication gateway according to an embodiment of the present application;
fig. 7 is a flowchart of a method for using a dynamic invocation device of an internet of things communication gateway according to an embodiment of the present application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only a few embodiments of the present application, and not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.
Overview of a scene
As described above, a gateway is also called an inter-network connector and a protocol converter, and is only used for interconnecting networks with different protocols at a higher layer on a transport layer, and is the most complex network interconnection device, which can be used for both a wide area network interconnection and a local area network interconnection, and is a computer system or device serving as a conversion task. With the continuous development of gateway technology, gateway devices are more and more widely applied to communication networks of the internet of things, and the gateway devices begin to enter the public life, so that the life of people is more intelligent and convenient. However, when the current gateway is called, the problems of poor gateway operation stability, low calling efficiency and the like often exist. Therefore, an optimized dynamic calling device for the communication gateway of the internet of things is desired, which can dynamically call the communication gateway of the internet of things according to actual conditions so as to improve calling efficiency on the basis of ensuring gateway operation stability.
At present, deep learning and neural networks have been widely applied in the fields of computer vision, natural language processing, speech signal processing, and the like. In addition, in the fields of image classification, object detection, semantic segmentation, text translation and the like, also exhibit levels approaching and even exceeding those of humans.
In recent years, deep learning and development of a neural network provide new solutions and schemes for dynamic calling of gateways of internet of things communication.
Specifically, in the technical scheme of the application, it is expected that whether a fault exists in the operation of the gateway is determined in advance by detecting the state of the gateway in advance, and dynamic calling is performed after the fault is detected, so that stable operation of software and hardware equipment of the gateway before dynamic calling can be effectively ensured. More specifically, in the technical scheme of the application, an artificial intelligence algorithm based on deep learning is adopted to extract multi-scale implicit associated features among parameter features in state data of the communication gateway, and fusion feature distribution information of the multi-scale implicit associated features among the parameter features is returned to generate data index values of the communication gateway of the internet of things, and the data index values are compared with a threshold value to detect faults, so that whether the communication gateway is called or not is determined. Therefore, the communication gateway of the Internet of things can be dynamically called according to actual conditions, so that the stable operation of gateway software and hardware equipment before dynamic calling can be effectively ensured, and the calling efficiency is improved.
Specifically, in the technical scheme of the application, in order to obtain an accurate fault detection result of the operation of the communication gateway of the internet of things and perform dynamic calling judgment of the gateway based on the fault detection result, firstly, gateway state data of the communication gateway to be called is obtained. And then, constructing the gateway state data into a gateway state data matrix to integrate the data parameter characteristic information of each data type in the gateway state data of the communication gateway to be called. It should be noted that, here, the value of each position in the gateway status data matrix is used to represent the parameter characteristic corresponding to the jth data in the ith data type.
Then, carrying out implicit association feature mining on each parameter feature in the gateway state data matrix by using a convolutional neural network model with excellent performance in implicit association feature extraction. In particular, considering that different correlation characteristic information exists between various parameter features of various types in different scale spans and different types of gateway state data matrixes, in order to capture more sufficient feature correlation information, convolutional neural networks with convolution kernels of different receptive fields are selected to perform different feature correlation mining. That is, specifically, the gateway state data matrix is passed through a dual-flow network model including a first convolutional neural network and a second convolutional neural network to obtain a first scale feature map and a second scale feature map.
In particular, in one particular example of the present application, the first convolutional neural network uses a first convolutional kernel having a first size, and the second convolutional neural network uses a second convolutional kernel having a second size, the first size being different from the second size. In this way, multi-scale associated features under the receptive fields with different sizes among different parameter feature data in the gateway state data matrix can be extracted. Accordingly, in another specific example of the present application, the first convolutional neural network uses a first convolutional kernel having a first void rate, the second convolutional neural network uses a second convolutional kernel having a second void rate, and the first convolutional kernel and the second convolutional kernel have the same size. In this way, implicit association features focusing on specific different parameter features in the gateway state data matrix can be extracted during feature extraction. In this way, the extraction of the correlation characteristic information among the richer gateway state data is facilitated.
Further, feature distribution information in the first scale feature map and the second scale feature map is fused to obtain a decoding feature map. Accordingly, in a specific example, the feature information fusion of the first scale feature map and the second scale feature map may be performed by a position weighted sum. And then, carrying out decoding regression processing on the decoding characteristic graph through a decoder to obtain a decoding value for representing the data index value of the communication gateway of the Internet of things. Therefore, the fault detection of the gateway can be carried out based on the comparison between the data index value of the communication gateway of the Internet of things and the preset threshold value, and whether the communication gateway of the Internet of things can be called or not can be determined. Specifically, in response to the decoded value being greater than or equal to the predetermined threshold value, it is determined that the internet of things communication gateway can be invoked.
In particular, in the technical solution of the present application, for the first scale feature map and the second scale feature map, the associated features of the gateway state data at different scales are expressed, but although the expression scales are different, it is still desirable that the expressions of the internal feature distributions of the source data state tend to be consistent, so as to improve the fusion effect of the first scale feature map and the second scale feature map.
Based on this, in the technical solution of the present application, an intrinsic learning loss function in the sequence pair sequence consistency rule between the first scale feature map and the second scale feature map is further calculated, and specifically expressed as:
Figure 988863DEST_PATH_IMAGE011
Figure 972999DEST_PATH_IMAGE012
/>
Figure 305410DEST_PATH_IMAGE013
wherein the content of the first and second substances,
Figure 349589DEST_PATH_IMAGE014
is the first activation feature vector, is asserted>
Figure 13789DEST_PATH_IMAGE015
Is the second activation characteristic vector, is greater than>
Figure 801616DEST_PATH_IMAGE016
Is the value of the loss of learning function intrinsic to the sequence response rule, in question>
Figure 375817DEST_PATH_IMAGE017
Is a first feature vector obtained after the expansion of the first scale feature map, is/are based on>
Figure 200685DEST_PATH_IMAGE018
Is a second feature vector obtained after the expansion of the second scale feature map, and->
Figure 227547DEST_PATH_IMAGE019
And &>
Figure 819065DEST_PATH_IMAGE020
Respectively is the decoder for->
Figure 106827DEST_PATH_IMAGE017
And &>
Figure 492809DEST_PATH_IMAGE018
The weight matrix of (2).
Here, the sequence-to-sequence consistency rules intrinsic learning penalty function may obtain enhanced discriminative power between sequences through a decoder's press-and-fire channel attention mechanism for weight matrices of different sequences. In this way, by training the network with this loss function, it is possible to recover the association relationship feature with better distinction between the first scale feature map and the second scale feature map, so as to perform internalized learning (internalized learning) on the association consistency rule between the feature distributions of the first scale feature map and the second scale feature map. Therefore, the consistency of the first scale feature map and the second scale feature map on the expression of the internal feature distribution of the source data state is enhanced, so that the fusion effect of the first scale feature map and the second scale feature map is improved, and the accuracy of decoding regression is improved. Therefore, the communication gateway of the Internet of things can be dynamically called according to actual conditions, so that the stable operation of gateway software and hardware equipment before dynamic calling can be effectively ensured, and the calling efficiency is improved.
Based on this, this application has proposed a thing networking communication gateway developments calling device, and it includes: the gateway state data acquisition unit is used for acquiring gateway state data of the communication gateway to be called; the matrix construction unit is used for constructing the gateway state data into a gateway state data matrix, wherein values of all positions in the gateway state data matrix are used for expressing parameter characteristics corresponding to jth data in the ith data type; the multi-scale convolution detection unit is used for enabling the gateway state data matrix to pass through a double-current network model comprising a first convolution neural network and a second convolution neural network so as to obtain a first scale characteristic diagram and a second scale characteristic diagram; the fusion unit is used for fusing the first scale feature map and the second scale feature map to obtain a decoding feature map; the decoding unit is used for enabling the decoding characteristic graph to pass through a decoder to obtain a decoding value used for representing a data index value of the communication gateway of the Internet of things; and a call control result generation unit for determining whether the internet of things communication gateway can be called based on a comparison between the decoded value and a predetermined threshold.
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary System
Fig. 1 is a block diagram of a dynamic invocation device of an internet of things communication gateway according to an embodiment of the application. As shown in fig. 1, the dynamic invoking device 300 for the internet of things communication gateway according to the embodiment of the present application includes an inference module, where the inference module includes: a gateway status data acquisition unit 310; a matrix construction unit 320; a multi-scale convolution detection unit 330; a fusion unit 340; a decoding unit 350; and, a call control result generation unit 360.
The gateway state data acquisition unit 310 is configured to acquire gateway state data of a communication gateway to be called; the matrix constructing unit 320 is configured to construct the gateway state data into a gateway state data matrix, where a value of each position in the gateway state data matrix is used to represent a parameter feature corresponding to jth data in an ith data type; the multi-scale convolution detection unit 330 is configured to pass the gateway state data matrix through a dual-flow network model including a first convolution neural network and a second convolution neural network to obtain a first scale feature map and a second scale feature map; the fusion unit 340 is configured to fuse the first scale feature map and the second scale feature map to obtain a decoded feature map; the decoding unit 350 is configured to pass the decoded feature map through a decoder to obtain a decoded value of a data index value representing an internet of things communication gateway; and the call control result generation unit 360 is configured to determine whether the internet of things communication gateway can be called based on a comparison between the decoded value and a predetermined threshold.
Fig. 3 is a system architecture diagram of a dynamic invocation device of an internet of things communication gateway according to an embodiment of the present application. As shown in fig. 3, in the system architecture of the dynamic invoking device 300 for the internet of things communication gateway, in the inference process, first, the gateway state data of the communication gateway to be invoked is acquired through the gateway state data acquiring unit 310; then, the matrix constructing unit 320 constructs the gateway state data acquired by the gateway state data acquiring unit 310 into a gateway state data matrix, where values at various positions in the gateway state data matrix are used to represent parameter characteristics corresponding to the jth data in the ith data type; the multi-scale convolution detecting unit 330 passes the gateway state data matrix obtained by the matrix constructing unit 320 through a dual-flow network model including a first convolutional neural network and a second convolutional neural network to obtain a first scale feature map and a second scale feature map; then, the fusion unit 340 fuses the first scale feature map and the second scale feature map obtained by the multi-scale convolution detection unit 330 to obtain a decoded feature map; the decoding unit 350 passes the decoding feature map obtained by the fusion unit 340 through a decoder to obtain a decoding value used for representing a data index value of the communication gateway of the internet of things; further, the call control result generation unit 360 determines whether the internet of things communication gateway can be called based on a comparison between the decoded value obtained by the decoding unit 350 and a predetermined threshold value.
Specifically, in the operation process of the dynamic invoking device 300 for the internet of things communication gateway, the gateway state data acquiring unit 310 is configured to acquire gateway state data of the communication gateway to be invoked. In the technical scheme of the application, in order to obtain an accurate fault detection result of the operation of the communication gateway of the internet of things and perform dynamic calling judgment of the gateway based on the fault detection result, firstly, gateway state data of the communication gateway to be called is obtained through a server, wherein the gateway state data can reflect the gateway state.
Specifically, in the operation process of the internet of things communication gateway dynamic call device 300, the matrix constructing unit 320 is configured to construct the gateway state data into a gateway state data matrix, where values of each position in the gateway state data matrix are used to represent a parameter characteristic corresponding to jth data in an ith data type. Namely, the gateway state data is constructed into a gateway state data matrix to integrate the data parameter characteristic information of each data type in the gateway state data of the communication gateway to be called. It should be noted that, here, the value of each position in the gateway status data matrix is used to represent the parameter characteristic corresponding to the jth data in the ith data type.
Specifically, in the operation process of the internet of things communication gateway dynamic calling apparatus 300, the multi-scale convolution detecting unit 330 is configured to pass the gateway state data matrix through a dual-flow network model including a first convolution neural network and a second convolution neural network to obtain a first scale feature map and a second scale feature map. In the technical scheme of the application, a convolutional neural network model with excellent performance in the aspect of implicit association feature extraction is used for mining the implicit association features of all parameter features in the gateway state data matrix. In particular, considering that different correlation characteristic information exists between different types of parameter features in different scale spans and different kinds of gateway state data matrixes, in order to capture more sufficient feature correlation information, convolutional neural networks with convolutional kernels of different receptive fields are selected to perform different feature correlation mining. That is, specifically, the gateway state data matrix is passed through a dual-flow network model including a first convolutional neural network and a second convolutional neural network to obtain a first scale feature map and a second scale feature map. In particular, in one particular example of the present application, the first convolutional neural network uses a first convolutional kernel having a first size, and the second convolutional neural network uses a second convolutional kernel having a second size, the first size being different from the second size. In this way, multi-scale associated features under the receptive fields with different sizes among different parameter feature data in the gateway state data matrix can be extracted. Accordingly, in another specific example of the present application, the first convolutional neural network uses a first convolutional kernel having a first void rate, the second convolutional neural network uses a second convolutional kernel having a second void rate, and the first convolutional kernel and the second convolutional kernel have the same size. In this way, implicit association features focusing on specific different parameter features in the gateway state data matrix can be extracted during feature extraction. In this way, the extraction of the correlation characteristic information among the richer gateway state data is facilitated.
Fig. 4 is a block diagram of a multi-scale convolution detection unit in a dynamic calling device of an internet of things communication gateway according to an embodiment of the present application. As shown in fig. 4, the multi-scale convolution detecting unit 330 includes: a first scale convolution coding subunit 331, configured to perform convolution processing of a first convolution kernel on input data in forward pass of a layer using each layer of the first convolution neural network: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein, the output of the last layer of the first convolutional neural network is the first scale feature map, and the input of the first layer of the first convolutional neural network is a gateway state data matrix; and a second scale convolutional coding subunit 332, configured to perform convolutional processing of a second convolutional kernel on the input data in forward pass of the layer using each layer of the second convolutional neural network: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the second convolutional neural network is the second scale characteristic diagram, and the input of the first layer of the second convolutional neural network is a gateway state data matrix.
Fig. 5 is a flowchart of a first convolutional code in a dynamic invoking device of an internet of things communication gateway according to an embodiment of the present application. As shown in fig. 5, in the first convolutional neural network coding process, the method includes: performing convolution processing of a first convolution kernel on input data in forward transfer of layers using layers of the first convolution neural network: s210, performing convolution processing on input data to obtain a convolution characteristic diagram; s220, performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and S230, carrying out nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the first convolutional neural network is the first scale characteristic diagram, and the input of the first layer of the first convolutional neural network is a gateway state data matrix.
Specifically, in an operation process of the internet of things communication gateway dynamic calling apparatus 300, the fusion unit 340 is configured to fuse the first scale feature map and the second scale feature map to obtain a decoding feature map. And fusing the feature distribution information in the first scale feature map and the second scale feature map to obtain a decoding feature map. Accordingly, in a specific example, the feature information fusion of the first scale feature map and the second scale feature map may be performed by a position weighted sum. In a specific example of the present application, the fusion unit is further configured to: fusing the first scale feature map and the second scale feature map according to the following formula to obtain a decoding feature map;
wherein the formula is:
F
Figure 6967DEST_PATH_IMAGE001
wherein
Figure 11963DEST_PATH_IMAGE002
Represents the first scale feature map, and>
Figure 29598DEST_PATH_IMAGE003
represents the second scale feature map, F represents the decoded feature map, and ` is `>
Figure 852060DEST_PATH_IMAGE004
A weighting parameter, which represents the first scale feature map quantity and the second scale feature map respectively, is selected>
Figure 978148DEST_PATH_IMAGE005
Indicating a sum by position.
Specifically, during the operation of the internet of things communication gateway dynamic invoking device 300, the decoding unit 350, for passing the decoded feature map through a decoder to obtain a representation of the communication gateway of the internet of thingsA decoded value of a data index value. Namely, the decoding characteristic graph is processed by decoding regression in a decoder to obtain a decoding value used for representing the data index value of the communication gateway of the Internet of things. In a specific example of the present application, the decoding unit is further configured to: decoding the decoding characteristic graph by using the decoder to perform decoding regression according to the following formula to obtain a decoding value; wherein the formula is:
Figure 645890DEST_PATH_IMAGE006
wherein->
Figure 783610DEST_PATH_IMAGE007
Represents the decoded feature map, is selected>
Figure 386761DEST_PATH_IMAGE008
Is the decoded value, is greater than or equal to>
Figure 875511DEST_PATH_IMAGE009
Is a weight matrix, in conjunction with a weighting function>
Figure 205998DEST_PATH_IMAGE010
Representing a matrix multiplication.
Specifically, during the operation of the dynamic invoking device 300 for the internet of things communication gateway, the invoking control result generating unit 360 is configured to determine whether the internet of things communication gateway can be invoked based on the comparison between the decoded value and a predetermined threshold. Namely, the data index value of the communication gateway of the internet of things is generated back and forth based on the fusion feature distribution information of the multi-scale implicit association features among the parameter features, and is compared with the threshold value to detect the fault, so that whether the communication gateway is called or not is determined. Therefore, the communication gateway of the Internet of things can be dynamically called according to actual conditions, so that the stable operation of gateway software and hardware equipment before dynamic calling can be effectively ensured, and the calling efficiency is improved. In a specific example of the present application, it is determined that the internet of things communication gateway may be invoked in response to the decoded value being greater than or equal to the predetermined threshold.
It will be appreciated that the dual-stream network model and the decoder need to be trained before the inference can be made using the neural network model described above. That is to say, in the dynamic calling device for the communication gateway of the internet of things of the application, the dynamic calling device further comprises a training module, wherein the training module is used for training the double-flow network model and the decoder.
Fig. 2 is a block diagram of a dynamic invocation device of an internet of things communication gateway according to an embodiment of the application. As shown in fig. 2, the dynamic invoking device 300 for the internet of things communication gateway according to the embodiment of the present application further includes a training module 400, where the training module includes: a training data acquisition unit 410; a training matrix construction unit 420; training the multi-scale convolution detection unit 430; a training fusion unit 440; a training decoding unit 450; an intrinsic learning loss unit 460; and a training unit 470.
The training data acquisition unit 410 is configured to acquire training data, where the training data includes training gateway state data of a communication gateway to be called and a true value of whether the internet of things communication gateway can be called; the training matrix constructing unit 420 is configured to construct the training gateway state data into a training gateway state data matrix, where a value at each position in the training gateway state data matrix is used to represent a parameter feature corresponding to jth data in an ith data type; the training multi-scale convolution detecting unit 430 is configured to pass the training gateway state data matrix through the dual-flow network model including the first convolutional neural network and the second convolutional neural network to obtain a training first-scale feature map and a training second-scale feature map; the training fusion unit 440 is configured to fuse the training first scale feature map and the training second scale feature map to obtain a training decoding feature map; the training decoding unit 450 is configured to pass the training decoded feature map through the decoder to obtain a decoding loss function value; the intrinsic learning loss unit 460 is configured to calculate a sequence pair sequence response rule intrinsic learning loss function value based on a distance between a first feature vector obtained after the training of the first scale feature map and a second feature vector obtained after the training of the second scale feature map; and the training unit 470, for calculating a weighted sum of the decoding loss function values and the sequence versus sequence response rule intrinsic learning loss function values as loss function values to train the dual-stream network model and the decoder.
Fig. 6 is a system architecture diagram of a training module in a dynamic invocation device of an internet of things communication gateway according to an embodiment of the present application. As shown in fig. 6, in the system architecture of the dynamic invocation device 300 for the internet of things communication gateway, in the training process, training data is first obtained through the training data acquisition unit 410, where the training data includes training gateway state data of the communication gateway to be invoked, and a true value of whether the internet of things communication gateway can be invoked; then, the training matrix constructing unit 420 constructs the training gateway state data acquired by the training data acquiring unit 410 into a training gateway state data matrix, where values at various positions in the training gateway state data matrix are used to represent parameter characteristics corresponding to the jth data in the ith data type; the training multi-scale convolution detecting unit 430 passes the training gateway state data matrix obtained by the training matrix constructing unit 420 through the dual-flow network model including the first convolution neural network and the second convolution neural network to obtain a training first scale feature map and a training second scale feature map; then, the training fusion unit 440 fuses the training first scale feature map and the training second scale feature map obtained by the training multi-scale convolution detecting unit 430 to obtain a training decoding feature map; the training decoding unit 450 passes the training decoding feature map obtained by the training fusion unit 440 through the decoder to obtain a decoding loss function value; the intrinsic learning loss unit 460 calculates a sequence-to-sequence response rule intrinsic learning loss function value based on a distance between a first feature vector obtained by unfolding the training first scale feature map and a second feature vector obtained by unfolding the training second scale feature map; further, the training unit 470 is configured to compute a weighted sum of the decoding loss function values and the sequence versus sequence response rule intrinsic learning loss function values as loss function values to train the dual-stream network model and the decoder.
In particular, in the technical solution of the present application, for the first scale feature map and the second scale feature map, the associated features of the gateway state data at different scales are expressed, but although the expression scales are different, it is still desirable that the expressions of the internal feature distributions of the source data state tend to be consistent, so as to improve the fusion effect of the first scale feature map and the second scale feature map.
Based on this, in the technical solution of the present application, an intrinsic learning loss function in the sequence pair sequence consistency rule between the first scale feature map and the second scale feature map is further calculated, and is specifically expressed as:
Figure 463804DEST_PATH_IMAGE011
Figure 362490DEST_PATH_IMAGE012
Figure 945394DEST_PATH_IMAGE013
wherein the content of the first and second substances,
Figure 954938DEST_PATH_IMAGE014
is a first activation feature vector, based on a first activation signal, is determined>
Figure 332830DEST_PATH_IMAGE015
Is the second activation characteristic vector, is greater than>
Figure 261471DEST_PATH_IMAGE016
Is the value of the loss of learning function intrinsic to the sequence response rule, in question>
Figure 724814DEST_PATH_IMAGE017
Is as followsA first feature vector obtained by unfolding the training first scale feature map is based on the comparison result>
Figure 147836DEST_PATH_IMAGE018
Is a second feature vector obtained after the training second scale feature map is expanded, and ^ is greater than or equal to>
Figure 645813DEST_PATH_IMAGE019
And &>
Figure 620723DEST_PATH_IMAGE020
Is the weight matrix of the decoder for the first and second feature vector, respectively>
Figure 430416DEST_PATH_IMAGE021
Represents->
Figure 47342DEST_PATH_IMAGE022
An activation function +>
Figure 868667DEST_PATH_IMAGE023
Represents->
Figure 421003DEST_PATH_IMAGE024
An activation function +>
Figure 593358DEST_PATH_IMAGE025
Represents a matrix multiplication,. Sup.>
Figure DEST_PATH_IMAGE027
Representing the euclidean distance between the two vectors. Here, the sequence-to-sequence consistency rules intrinsic learning penalty function may obtain enhanced discriminative power between sequences through a decoder's press-and-fire channel attention mechanism for weight matrices of different sequences. In this way, by training the network with the loss function, the recovery of the incidence relation feature with better distinguishability between the first scale feature map and the second scale feature map can be realized, so as to carry out the recovery on the first scale feature map and the second scale feature mapThe association consistency rules between feature distributions of the feature maps are internalized learning (internalized learning). Therefore, the consistency of the first scale feature map and the second scale feature map on the expression of the internal feature distribution of the source data state is enhanced, so that the fusion effect of the first scale feature map and the second scale feature map is improved, and the accuracy of decoding regression is improved. Therefore, the communication gateway of the Internet of things can be dynamically called according to actual conditions, so that the stable operation of gateway software and hardware equipment before dynamic calling can be effectively ensured, and the calling efficiency is improved.
In summary, the dynamic calling device 300 for the internet of things communication gateway according to the embodiment of the present application is set forth, which extracts multi-scale implicit association features among parameter features in state data of the communication gateway by using an artificial intelligence algorithm based on deep learning, and generates data index values of the internet of things communication gateway based on fusion feature distribution information of the multi-scale implicit association features among the parameter features, so as to compare the data index values with a threshold value to perform fault detection, thereby determining whether the communication gateway is called. Therefore, the communication gateway of the Internet of things can be dynamically called according to actual conditions, so that the stable operation of gateway software and hardware equipment before dynamic calling can be effectively ensured, and the calling efficiency is improved.
Exemplary method
Fig. 7 is a flowchart of a method for using a dynamic invocation device of an internet of things communication gateway according to an embodiment of the present application. As shown in fig. 7, a method for using a dynamic invocation device of an internet of things communication gateway according to an embodiment of the present application includes the steps of: s110, acquiring gateway state data of the communication gateway to be called; s120, constructing the gateway state data into a gateway state data matrix, wherein values of all positions in the gateway state data matrix are used for representing parameter characteristics corresponding to jth data in the ith data type; s130, enabling the gateway state data matrix to pass through a double-flow network model comprising a first convolutional neural network and a second convolutional neural network to obtain a first scale characteristic diagram and a second scale characteristic diagram; s140, fusing the first scale feature map and the second scale feature map to obtain a decoding feature map; s150, enabling the decoding characteristic graph to pass through a decoder to obtain a decoding value used for representing a data index value of the Internet of things communication gateway; and S160, determining whether the Internet of things communication gateway can be called or not based on the comparison between the decoding value and a preset threshold value.
In an example, in the method for using the dynamic invocation device of the internet of things communication gateway, the step S130 includes: the first convolutional neural network uses a first convolutional kernel having a first size, and the second convolutional neural network uses a second convolutional kernel having a second size, the first size being different from the second size, wherein the first convolutional neural network uses a first convolutional kernel having a first void rate, the second convolutional neural network uses a second convolutional kernel having a second void rate, and the first convolutional kernel and the second convolutional kernel have the same size. More specifically, the passing the gateway state data matrix through a dual-flow network model including a first convolutional neural network and a second convolutional neural network to obtain a first scale feature map and a second scale feature map includes: performing convolution processing of a first convolution kernel on input data in forward transfer of layers using layers of the first convolution neural network: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein, the output of the last layer of the first convolutional neural network is the first scale feature map, and the input of the first layer of the first convolutional neural network is a gateway state data matrix; and performing convolution processing of a second convolution kernel on the input data in forward transfer of layers using layers of the second convolutional neural network: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the second convolutional neural network is the second scale characteristic diagram, and the input of the first layer of the second convolutional neural network is a gateway state data matrix.
In an example, in the method for using the dynamic invoking device for an internet of things communication gateway, the step S140 includes: fusing the first scale feature map and the second scale feature map according to the following formula to obtain a decoding feature map; wherein the formula is:
F
Figure 76292DEST_PATH_IMAGE001
wherein
Figure 17703DEST_PATH_IMAGE002
Represents the first scale feature map, and>
Figure 475360DEST_PATH_IMAGE003
represents the second scale feature map, F represents the decoded feature map, and ` is `>
Figure 869433DEST_PATH_IMAGE004
A weighting parameter, which represents the first scale feature map quantity and the second scale feature map respectively, is selected>
Figure 828161DEST_PATH_IMAGE005
Indicating a sum by position.
In an example, in the method for using the dynamic invocation device of the internet of things communication gateway, the step S150 includes: decoding the decoding characteristic graph by using the decoder to perform decoding regression according to the following formula to obtain a decoding value; wherein the formula is:
Figure 14292DEST_PATH_IMAGE006
in which>
Figure 33064DEST_PATH_IMAGE007
Represents the decoded feature map, is selected>
Figure 914432DEST_PATH_IMAGE008
Is the decoded value, is greater than or equal to>
Figure 12270DEST_PATH_IMAGE009
Is a weight matrix, is->
Figure 928274DEST_PATH_IMAGE010
Representing a matrix multiplication.
In an example, in the method for using the dynamic invocation device of the internet of things communication gateway, the step S160 includes: in response to the decoded value being greater than or equal to the predetermined threshold, determining that the internet of things communication gateway can be invoked.
In summary, the use method of the dynamic calling device for the internet of things communication gateway according to the embodiment of the application is clarified, the artificial intelligence algorithm based on deep learning is adopted to extract the multi-scale implicit association features among the parameter features in the state data of the communication gateway, the multi-scale implicit association features among the parameter features are fused with feature distribution information to generate the data index value of the internet of things communication gateway, the data index value is compared with the threshold value to detect the fault, and then whether the communication gateway is called or not is determined. Therefore, the communication gateway of the Internet of things can be dynamically called according to actual conditions, so that the stable operation of gateway software and hardware equipment before dynamic calling can be effectively ensured, and the calling efficiency is improved.

Claims (10)

1. The utility model provides a thing networking communication gateway dynamic calling device which characterized in that includes:
the gateway state data acquisition unit is used for acquiring gateway state data of the communication gateway to be called;
the matrix construction unit is used for constructing the gateway state data into a gateway state data matrix, wherein values of all positions in the gateway state data matrix are used for expressing parameter characteristics corresponding to jth data in the ith data type;
the multi-scale convolution detection unit is used for enabling the gateway state data matrix to pass through a double-current network model comprising a first convolution neural network and a second convolution neural network so as to obtain a first scale characteristic diagram and a second scale characteristic diagram;
a fusion unit, configured to fuse the first scale feature map and the second scale feature map to obtain a decoded feature map;
the decoding unit is used for enabling the decoding characteristic graph to pass through a decoder to obtain a decoding value used for representing a data index value of the communication gateway of the Internet of things; and
a call control result generation unit for determining whether the internet of things communication gateway can be called based on a comparison between the decoded value and a predetermined threshold value.
2. The internet-of-things communication gateway dynamic invoking device according to claim 1, wherein the first convolutional neural network uses a first convolutional kernel having a first size, and the second convolutional neural network uses a second convolutional kernel having a second size, and the first size is different from the second size.
3. The internet of things communication gateway dynamic invoking device according to claim 1, wherein the first convolutional neural network uses a first convolutional kernel with a first hole rate, the second convolutional neural network uses a second convolutional kernel with a second hole rate, and the first convolutional kernel and the second convolutional kernel have the same size.
4. The internet of things communication gateway dynamic calling device as claimed in claim 2 or 3, wherein the multi-scale convolution detection unit comprises:
a first scale convolutional coding subunit, configured to perform convolutional processing of a first convolutional kernel on input data in forward transfer of a layer using each layer of the first convolutional neural network: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein, the output of the last layer of the first convolutional neural network is the first scale feature map, and the input of the first layer of the first convolutional neural network is a gateway state data matrix; and
a second scale convolutional coding subunit, configured to perform convolutional processing of a second convolutional kernel on the input data in forward transfer of layers using each layer of the second convolutional neural network: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the second convolutional neural network is the second scale characteristic diagram, and the input of the first layer of the second convolutional neural network is a gateway state data matrix.
5. The internet of things communication gateway dynamic invoking device of claim 4, wherein the fusion unit is further configured to: fusing the first scale feature map and the second scale feature map according to the following formula to obtain a decoding feature map;
wherein the formula is:
F
Figure DEST_PATH_IMAGE001
wherein
Figure 627868DEST_PATH_IMAGE002
Represents the first scale feature map, and>
Figure DEST_PATH_IMAGE003
represents the second scale feature map, F represents the decoded feature map, and ` is `>
Figure 538055DEST_PATH_IMAGE004
A weighting parameter, representing the first scale feature map and the second scale feature map, respectively>
Figure DEST_PATH_IMAGE005
Indicating a sum by position.
6. The device for dynamically invoking the internet of things communication gateway according to claim 5, wherein the decoding unit is further configured to: decoding the decoding characteristic graph by using the decoder to perform decoding regression according to the following formula to obtain a decoding value;
wherein the formula is:
Figure 197706DEST_PATH_IMAGE006
wherein->
Figure DEST_PATH_IMAGE007
Represents the decoded feature map, is selected>
Figure 969484DEST_PATH_IMAGE008
Is the decoded value, is greater than or equal to>
Figure DEST_PATH_IMAGE009
Is a weight matrix, is->
Figure 93298DEST_PATH_IMAGE010
Representing a matrix multiplication.
7. The device for dynamically invoking an internet of things communication gateway according to claim 6, wherein the invocation control result generating unit is configured to determine that the internet of things communication gateway can be invoked in response to the decoded value being greater than or equal to the predetermined threshold.
8. The device for dynamically invoking the IOT communication gateway according to claim 7, further comprising a training module for training said dual-flow network model and said decoder.
9. The internet of things communication gateway dynamic invoking device of claim 8, wherein the training module comprises:
the training data acquisition unit is used for acquiring training data, wherein the training data comprises training gateway state data of a communication gateway to be called and a true value of whether the communication gateway of the Internet of things can be called;
the training gateway state data matrix is used for acquiring training gateway state data, wherein values of all positions in the training gateway state data matrix are used for expressing parameter characteristics corresponding to jth data in the ith data type;
the training multi-scale convolution detection unit is used for enabling the training gateway state data matrix to pass through the double-flow network model containing the first convolution neural network and the second convolution neural network so as to obtain a training first scale characteristic diagram and a training second scale characteristic diagram;
the training fusion unit is used for fusing the training first scale feature map and the training second scale feature map to obtain a training decoding feature map;
a training decoding unit, configured to pass the training decoded feature map through the decoder to obtain a decoding loss function value;
the intrinsic learning loss unit is used for calculating an intrinsic learning loss function value of a sequence pair sequence response rule based on the distance between a first feature vector obtained after the training of the first scale feature map and a second feature vector obtained after the training of the second scale feature map; and
a training unit to compute a weighted sum of the decoding loss function values and the sequence versus sequence response rules intrinsic learning loss function values as loss function values to train the dual-stream network model and the decoder.
10. The internet-of-things communication gateway dynamic invoking device according to claim 9, wherein the intrinsic learning loss unit is further configured to: calculating the intrinsic learning loss function value of the sequence-to-sequence response rule according to the following formula based on the distance between a first feature vector obtained after the training of the first scale feature map and a second feature vector obtained after the training of the second scale feature map;
wherein the formula is:
Figure DEST_PATH_IMAGE011
Figure 721857DEST_PATH_IMAGE012
Figure DEST_PATH_IMAGE013
wherein the content of the first and second substances,
Figure 868804DEST_PATH_IMAGE014
is the first activation feature vector, is asserted>
Figure DEST_PATH_IMAGE015
Is the second activation characteristic vector, is greater than>
Figure 959120DEST_PATH_IMAGE016
Is the value of the loss of learning function intrinsic to the sequence response rule, in question>
Figure DEST_PATH_IMAGE017
Is a first feature vector obtained after the expansion of the training first scale feature map, is selected>
Figure 484910DEST_PATH_IMAGE018
Is a second feature vector obtained after the training second scale feature map is expanded, and ^ is greater than or equal to>
Figure DEST_PATH_IMAGE019
And &>
Figure 471321DEST_PATH_IMAGE020
Is the weight matrix of the decoder for the first eigenvector and the second eigenvector, respectively>
Figure DEST_PATH_IMAGE021
Represents->
Figure 839985DEST_PATH_IMAGE022
The function is activated in such a way that,
Figure DEST_PATH_IMAGE023
represents->
Figure 12953DEST_PATH_IMAGE024
An activation function +>
Figure DEST_PATH_IMAGE025
Represents a matrix multiplication,. Sup.>
Figure 783463DEST_PATH_IMAGE026
Representing the euclidean distance between the two vectors. />
CN202211397669.XA 2022-11-09 2022-11-09 Dynamic calling device for communication gateway of Internet of things Active CN115580564B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211397669.XA CN115580564B (en) 2022-11-09 2022-11-09 Dynamic calling device for communication gateway of Internet of things

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211397669.XA CN115580564B (en) 2022-11-09 2022-11-09 Dynamic calling device for communication gateway of Internet of things

Publications (2)

Publication Number Publication Date
CN115580564A CN115580564A (en) 2023-01-06
CN115580564B true CN115580564B (en) 2023-04-18

Family

ID=84588262

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211397669.XA Active CN115580564B (en) 2022-11-09 2022-11-09 Dynamic calling device for communication gateway of Internet of things

Country Status (1)

Country Link
CN (1) CN115580564B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116178344B (en) * 2023-01-09 2023-10-20 泽田(山东)药业有限公司 Synthesis process of barnidipine hydrochloride

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110348319A (en) * 2019-06-18 2019-10-18 武汉大学 A kind of face method for anti-counterfeit merged based on face depth information and edge image
CN110796105A (en) * 2019-11-04 2020-02-14 中国矿业大学 Remote sensing image semantic segmentation method based on multi-modal data fusion
CN112333706A (en) * 2019-07-16 2021-02-05 中国移动通信集团浙江有限公司 Internet of things equipment anomaly detection method and device, computing equipment and storage medium
CN113870335A (en) * 2021-10-22 2021-12-31 重庆邮电大学 Monocular depth estimation method based on multi-scale feature fusion
WO2022134655A1 (en) * 2020-12-25 2022-06-30 神思电子技术股份有限公司 End-to-end video action detection and positioning system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110348319A (en) * 2019-06-18 2019-10-18 武汉大学 A kind of face method for anti-counterfeit merged based on face depth information and edge image
CN112333706A (en) * 2019-07-16 2021-02-05 中国移动通信集团浙江有限公司 Internet of things equipment anomaly detection method and device, computing equipment and storage medium
CN110796105A (en) * 2019-11-04 2020-02-14 中国矿业大学 Remote sensing image semantic segmentation method based on multi-modal data fusion
WO2022134655A1 (en) * 2020-12-25 2022-06-30 神思电子技术股份有限公司 End-to-end video action detection and positioning system
CN113870335A (en) * 2021-10-22 2021-12-31 重庆邮电大学 Monocular depth estimation method based on multi-scale feature fusion

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
"基于深度学习的关节点行为识别综述";刘云 等;《电子与信息学报》;20210113;第43卷(第6期);全文 *

Also Published As

Publication number Publication date
CN115580564A (en) 2023-01-06

Similar Documents

Publication Publication Date Title
CN113688723B (en) Infrared image pedestrian target detection method based on improved YOLOv5
CN115580564B (en) Dynamic calling device for communication gateway of Internet of things
CN114820871B (en) Font generation method, model training method, device, equipment and medium
CN113361794A (en) Information pushing method and AI (Artificial Intelligence) pushing system based on Internet e-commerce big data
CN112149526B (en) Lane line detection method and system based on long-distance information fusion
CN110647916A (en) Pornographic picture identification method and device based on convolutional neural network
CN111046949A (en) Image classification method, device and equipment
CN112200772A (en) Pox check out test set
CN116189139A (en) Traffic sign detection method based on Transformer
US20210150154A1 (en) Chapter-level text translation method and device
CN112967227B (en) Automatic diabetic retinopathy evaluation system based on focus perception modeling
CN113723426A (en) Image classification method and device based on deep multi-flow neural network
JP2021033994A (en) Text processing method, apparatus, device and computer readable storage medium
CN112861115B (en) Encryption strategy calling method based on block chain security authentication and cloud authentication server
CN112819006B (en) Image processing method and device and electronic equipment
CN113886547A (en) Client real-time conversation switching method and device based on artificial intelligence and electronic equipment
KR20230073608A (en) Node Selection Apparatus and Method for Maximizing Influence using Node Metadata in Network with Unknown Topology
CN114764593A (en) Model training method, model training device and electronic equipment
CN111191035B (en) Method and device for recognizing lung cancer clinical database text entity
CN111368026B (en) Text inclusion analysis method based on word meaning relation and dynamic convolution neural network
CN112765481A (en) Data processing method and device, computer and readable storage medium
Guo et al. An Offline Assistance Tool for Visually Impaired People Based on Image Captioning
CN113781110B (en) User behavior prediction method and system based on multi-factor weighted BI-LSTM learning
CN113806542B (en) Text analysis method and system
CN114219516B (en) Information flow session recommendation method based on big data and deep learning service system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant