CN110084295B - Multi-agent system grouping surrounding control method and control system - Google Patents
Multi-agent system grouping surrounding control method and control system Download PDFInfo
- Publication number
- CN110084295B CN110084295B CN201910318251.7A CN201910318251A CN110084295B CN 110084295 B CN110084295 B CN 110084295B CN 201910318251 A CN201910318251 A CN 201910318251A CN 110084295 B CN110084295 B CN 110084295B
- Authority
- CN
- China
- Prior art keywords
- agent
- frame
- network
- group
- intelligent
- 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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- H04L41/06—Management of faults, events, alarms or notifications
- H04L41/0677—Localisation of faults
-
- 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
- H04L41/12—Discovery or management of network topologies
-
- 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
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- 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
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the technical field of computer application, and discloses a multi-agent system grouping surrounding control method and a control system, which are used for detecting through a sensor network with a known topological structure to determine the topological structure of the multi-agent system; combining relatively close and relatively stationary agents into a group to simplify the control model; the model adopts a centralized control method, and the intelligent agent group with the largest intelligent agent number in the intelligent agent group is used as a central control intelligent agent group of the control model; coding the grouped agent groups; an agent code in the agent group; and determining the intelligent agent group and the intelligent agent according to the codes, and performing corresponding control operation. The invention uses the sensor network with known topological structure to build and correct the model, which saves time and has high control speed; the method adopts a grouping mode, so that the whole distributed processing and the local centralized processing are realized; the real-time coding mode is adopted for control, so that the precision is high and the positioning is accurate.
Description
Technical Field
The invention belongs to the technical field of computer application, and particularly relates to a multi-agent system grouping surrounding control method and a control system.
Background
The multi-agent surrounding target control algorithm is a control algorithm for realizing a group of agents surrounding targets in a detection range through a distributed network and local information. The generation of intelligent concepts originates from earlier human research in the field of artificial intelligence, where human beings no longer need to deal with complex dangerous tasks personally, but where "agents" are used instead of humans to accomplish these tasks are an important reason that artificial intelligence theory has been proposed. A physical entity such as a sensor, an actuator, a robot, a biological individual, a drone, etc. may be considered as an agent. Along with the development of the intelligent agent theory, the multi-intelligent agent system theory is also proposed, namely, a group of intelligent agents with sensor, calculation, execution and communication capabilities are related to form a network system through communication and the like. For a single agent, it can only obtain limited information, so it is difficult to solve the distributed problem. For multi-agent systems, the ability to solve such problems far exceeds that of a single agent through the interaction between agents. Compared with a single intelligent agent system, the multi-intelligent agent system has more advantages, which are mainly represented by the following aspects: higher efficiency, better system performance, stronger robustness and fault tolerance, lower price, simple use and operation, etc.
To sum up, the prior art has the following problems:
in the existing network system of a single intelligent agent, limited information is acquired, and the distributed problem cannot be solved. The existing multi-agent control method is low in convergence speed, cannot realize rapid control, is low in control precision and inaccurate in positioning in actual application.
Disclosure of Invention
Aiming at the problems existing in the prior art, the invention provides a multi-agent system grouping surrounding control method, which belongs to the energy-saving efficient and technical improvement inventions.
The multi-agent system grouping and surrounding control method provided by the invention comprises the following steps:
step one, detecting through a sensor network with a known topological structure, and determining the topological structure of the multi-agent system;
combining the relatively static intelligent agents with relatively close positions into a group to simplify a control model;
thirdly, the model adopts a centralized control method, and the intelligent agent group with the largest intelligent agent number in the intelligent agent group is used as a central control intelligent agent group of the control model;
step four, the grouped agent groups are encoded;
fifthly, encoding the agents in the agent groups;
step six, determining the intelligent agent group and the intelligent agent according to the codes, and performing corresponding control operation.
In the fourth step, the coding mode is: and placing all the agent groups in a two-dimensional vector space, establishing a Laplace matrix, wherein the boundary of the aggregation space is the boundary of the matrix, and establishing a space rectangular coordinate system by using the lower left corner boundary, wherein the coordinates of the agent groups in the matrix are the agent group codes.
In the fifth step, the encoding priorities of the agents in the group are as follows: high, low, up, down, left and right, and adopts relative position coding by taking the intelligent agent with the strongest calculation force as the center.
Further, the second step combines relatively close and relatively stationary agents into a group to simplify the control model,
selecting an agent evaluation model, calculating a predictive value of each frame of a relatively static agent by comparing an agent with a relatively static agent at an original relatively near position, and marking the obtained frame fraction as a vector X and the total frame number of the agent as N;
the window length of the sliding window is winLen, the sliding window processing is carried out on the obtained frame level quality fraction, namely the frame level fraction of the n-th frame after processing is the average value of the frame level fractions of the [ n-winlen+1, n ] frames, and the frame level fraction after the sliding window processing is marked as a vector WX;
sequencing WX from small to large, marking the sequenced result as WX', taking the average value of the worst p% frames as the quality measurement value of the whole agent topological structure sequence, sequencing, and taking the minimum p% frame average value as the final measurement result;
and sequentially carrying out sliding window treatment on all frame fractions calculated by the agent evaluation model, namely:
wherein, winLen represents window length during sliding window filtering, which is a parameter to be adjusted, X (t) represents quality fraction of the t-th frame, and WX (n) represents quality fraction of the n-th frame after sliding window processing;
and fusing the predicted frame fraction by using a time domain information fusion method based on the inter-frame association and the worst time slot, and finally predicting the fraction:
wherein p% is a parameter to be adjusted, N is the total frame number of the topological structure of the intelligent agent, WX' (t) represents the t th frame fraction after sequencing from small to large, OM winPooling The final evaluation result of the quality of the agent.
Further, the final evaluation result of the quality of the obtained intelligent agent is processed uniformly, and a uniform control model is constructed.
Further, the final evaluation result of the quality of the obtained intelligent agent is subjected to unified processing including data selection, data preprocessing and data transformation.
Further, the data transformation comprises data transformation dimension reduction, wherein the data transformation dimension reduction is a feature extracted from initial features of data through cluster analysis, and the data dimension is reduced.
Further, the process of constructing the unified control model includes: selecting reference alarm information parameter vector and establishing reference sequence X 0 ，
X 0 ＝{X 0 (k)|k＝1,2,…,n}＝(X 0 (1),X 0 (2),…,X 0 (n))
Wherein k represents time of day, X 0 Representing alarm information, wherein n represents the feature dimension of the parameter vector of the alarm information;
secondly, assuming that m comparison fault alarm information data exist, a comparison sequence X is established i
X i ＝{X i (k)|k＝1,2,…,n}＝(X i (1),X i (2),…,X i (n))i＝1,2,…,m
Then, a comparison sequence X is established i For reference number series X 0 Correlation coefficient ζ at time k i (k)
Wherein w is 1 The corresponding weight of each parameter is adjusted and decided according to the network attribute of the user; where ρ is the resolution factor and, ρ is e [0 ], ++ infinity a) is provided; the larger ρ isThe greater the resolution; the smaller the ρ, the smaller the resolution;
finally, calculate the comparison sequence X i For reference number series X 0 Is related to the degree of association of (2)
Further, establishing a unified control model further comprises establishing a mapping relation between the intelligent agent alarm information and the faults between networks through association analysis, establishing a fault positioning model, and performing network fault positioning through a BP neural network; the method comprises the following steps:
first, an m-dimensional alert vector Q is obtained n ＝(s 1 ,s 2 ,s 3 …s m ) And n-dimensional fault vector O n ＝(p 1 ,p 2 ,p 3 …p m ) The system is enabled to have a parallel structure and parallel processing capability by simultaneously inputting the system through a plurality of network nodes, and the input is dynamically processed in real time;
secondly, the BP network gives a value in a specified range to each connected weight, and simultaneously, a threshold value is specified to each neuron node;
thirdly, providing a group input alarm sample machine target result to a network, and calculating the connection weight, the threshold value and the input and output values of each hidden layer unit of the neural network node;
then, positive layer errors: calculating the error of the output layer unit by using the target vector and the network actual output value, correcting the connection weight and the threshold value by combining the output of each unit of the hidden layer, and carrying out reverse error propagation correction;
finally, training sample vectors and training the heterogeneous network system until the complete part of samples are trained, inputting network operation and maintenance fault alarm information into a trained BP network to locate network faults.
It is another object of the present invention to provide a multi-intelligent system packet enclosure control system implementing the multi-intelligent system packet enclosure control method.
In summary, the invention has the advantages and positive effects that:
the invention uses the sensor network with known topological structure to build and correct the model, which saves time and has high control speed; the invention adopts a grouping mode, thus realizing the overall distributed processing and the local centralized processing; the real-time coding mode is adopted for control, so that the precision is high and the positioning is accurate. The central control intelligent body is trained by adopting a neural network, so that the convergence rate is improved.
The method comprises the steps of combining relatively nearer and relatively stationary intelligent agents into a group to simplify a control model, selecting an intelligent agent evaluation model, calculating a predictive value of each frame of the relatively stationary intelligent agent by comparing the original nearer intelligent agent with the relatively stationary intelligent agent, marking the obtained frame fraction as a vector X, and marking the total frame number of the intelligent agent as N; the window length of the sliding window is winLen, the sliding window processing is carried out on the obtained frame level quality fraction, namely the frame level fraction of the n-th frame after processing is the average value of the frame level fractions of the [ n-winlen+1, n ] frames, and the frame level fraction after the sliding window processing is marked as a vector WX; sequencing WX from small to large, marking the sequenced result as WX', taking the average value of the worst p% frames as the quality measurement value of the whole agent topological structure sequence, sequencing, and taking the minimum p% frame average value as the final measurement result; and sequentially carrying out sliding window treatment on all frame fractions calculated by the agent evaluation model, namely:
The invention considers the connection between frames, and uses the sliding window mean value to process the data of each frame, so that the estimation accuracy is greatly improved.
By optimizing network parameters, the network self-healing is realized, and the intelligent operation and maintenance of the network are realized. The invention adopts the BP neural network method in the data mining to mine the network operation and maintenance data and the alarm information, combines the user perception information to intelligently locate the network faults, accurately judges the network faults, improves the network quality, reduces the operation cost, improves the system performance, and ensures the high efficiency, safety and stability of the network operation.
Drawings
Fig. 1 is a flowchart of a multi-agent system packet enclosure control method provided by the invention.
Detailed Description
For a further understanding of the invention, its features and advantages, reference is now made to the following examples, which are illustrated in the accompanying drawings.
The existing single intelligent agent acquires limited information, and the distributed problem cannot be solved. The existing multi-agent control method is low in convergence speed, cannot realize rapid control, is low in control precision and inaccurate in positioning in actual application.
In order to solve the above technical problems, the structure of the present invention will be described in detail with reference to the accompanying drawings.
As shown in fig. 1, the multi-agent system packet enclosure control method provided by the embodiment of the invention includes:
s101, detecting through a sensor network with a known topological structure, and determining the topological structure of the multi-agent system.
S102, combining relatively close-positioned and relatively static intelligent agents into a group to simplify a control model.
S103, the model adopts a centralized control method, takes the agent group with the largest number of agents in the agent group as a central control agent group of the control model, and trains the central control agent group by utilizing the BP neural network.
S104, coding the grouped agent groups.
S105, encoding the agents in the agent group.
S106, the trained central control agent group controls the corresponding agent group and agent through agent group and agent coding.
In step S103, the training method based on the BP neural network provided by the embodiment of the present invention specifically includes:
(1) A 3-layer neural network of one input layer, one hidden layer and one output layer is formed.
(2) The adjustment formulas of the weight and the threshold value of the basic BP neural network are as follows:
w kj (t+1)＝w kj (t)+αδ k H j
w ji (t+1)＝w ji (t)+αδ j I i
θ k (t+1)＝θ k (t)+βδ k
θ j (t+1)＝θ j (t)+βδ j
in the above formula, H j Is the output of the hidden layer node. I j A signal input for the input node i. w (w) kj (t) and w kj And (t+1) is the connection weight of the hidden node j and the output layer node k during the training of the front and the back two times respectively. w (w) ji (t) and w ji And (t+1) is the connection weight of the input node i and the hidden node j during the training of the front and back two times respectively. θ k And theta j The thresholds at output node k and hidden node j, respectively. Alpha and beta are learning parameters, respectively. Delta k And delta j The error signals of the output layer node k and the hidden node j are respectively calculated as follows:
δ k ＝(T k -O k )O k (1-O k )
wherein T is k For the target output value of the sample at the output node, O k And H j The actual output values of the samples at the output node and hidden node of the network are respectively.
Wherein the hidden layer output calculation formula is:
wherein M is the number of input nodes. f is an S-type activation function:
the output layer output uses linear weighted summation:
in the formula, s is the number of hidden nodes.
In step S104, the coding manner provided in the embodiment of the present invention specifically includes:
and placing all the agent groups in a two-dimensional vector space, establishing a Laplace matrix, wherein the boundary of the aggregation space is the boundary of the matrix, and establishing a space rectangular coordinate system by using the lower left corner boundary, wherein the coordinates of the agent groups in the matrix are the agent group codes.
In step S105, the agent code provided in the embodiment of the present invention specifically includes:
the coding priority of the agent in the group is as follows: high, low, up, down, left and right, and adopts relative position coding by taking the intelligent agent with the strongest calculation force as the center.
In the embodiment of the present invention, step S102 combines relatively close-located and relatively stationary agents into a group to simplify the control model,
selecting an agent evaluation model, calculating a predictive value of each frame of a relatively static agent by comparing an agent with a relatively static agent at a relatively near position, and marking the obtained frame fraction as a vector X and the total frame number of the agent as N.
And the window length of the sliding window is winLen, the sliding window processing is carried out on the obtained frame level quality fraction, namely the frame level fraction of the n-th frame after processing is the average value of the frame level fractions of the [ n-winlen+1, n ] frames, and the frame level fraction after the sliding window processing is marked as a vector WX.
And sequencing WX from small to large, marking the sequenced result as WX', taking the average value of the worst p% frames as the quality measurement value of the whole agent topological structure sequence, sequencing, and taking the minimum p% frame average value as the final measurement result.
And sequentially carrying out sliding window treatment on all frame fractions calculated by the agent evaluation model, namely:
wherein, winLen represents window length during sliding window filtering, which is a parameter to be adjusted, X (t) represents quality fraction of the t-th frame, and WX (n) represents quality fraction of the n-th frame after sliding window processing.
And fusing the predicted frame fraction by using a time domain information fusion method based on the inter-frame association and the worst time slot, and finally predicting the fraction:
wherein p% is a parameter to be adjusted, N is the total frame number of the topological structure of the intelligent agent, WX' (t) represents the t th frame fraction after sequencing from small to large, OM winPooling The final evaluation result of the quality of the agent.
And uniformly processing the obtained final evaluation result of the quality of the intelligent agent, and constructing a uniform control model.
And carrying out unified processing on the obtained final evaluation result of the quality of the intelligent agent, wherein the unified processing comprises data selection, data preprocessing and data transformation. The data transformation comprises data transformation dimension reduction, wherein the data transformation dimension reduction is to cut down the data dimension of the features extracted from the initial features of the data through cluster analysis.
The process for constructing the unified control model comprises the following steps: selecting reference alarm information parameter vector and establishing reference sequence X 0 ，
X 0 ＝{X 0 (k)|k＝1,2,…,n}＝(X 0 (1),X 0 (2),…,X 0 (n))
Wherein k represents time of day, X 0 Representing alarm information, wherein n represents the feature dimension of the parameter vector of the alarm information;
secondly, assuming that m comparison fault alarm information data exist, a comparison sequence X is established i
X i ＝{X i (k)|k＝1,2,…,n}＝(X i (1),X i (2),…,X i (n))i＝1,2,…,m
Then, a comparison sequence X is established i For reference number series X 0 Correlation coefficient ζ at time k i (k)
Wherein w is 1 The corresponding weight of each parameter is adjusted and decided according to the network attribute of the user; where ρ is the resolution factor and, ρ is e [0 ], ++ infinity a) is provided; the larger the ρ, the greater the resolution; the smaller the ρ, the smaller the resolution;
finally, calculate the comparison sequence X i For reference number series X 0 Is related to the degree of association of (2)
Establishing a unified control model further comprises the steps of establishing a mapping relation between intelligent agent alarm information and faults between networks through association analysis, establishing a fault positioning model, and performing network fault positioning through a BP neural network; the method comprises the following steps:
first, an m-dimensional alert vector is obtainedQ n ＝(s 1 ,s 2 ,s 3 …s m ) And n-dimensional fault vector O n ＝(p 1 ,p 2 ,p 3 …p m ) The system is enabled to have a parallel structure and parallel processing capability by simultaneously inputting the system through a plurality of network nodes, and the input is dynamically processed in real time;
secondly, the BP network gives a value in a specified range to each connected weight, and simultaneously, a threshold value is specified to each neuron node;
thirdly, providing a group input alarm sample machine target result to a network, and calculating the connection weight, the threshold value and the input and output values of each hidden layer unit of the neural network node;
then, positive layer errors: calculating the error of the output layer unit by using the target vector and the network actual output value, correcting the connection weight and the threshold value by combining the output of each unit of the hidden layer, and carrying out reverse error propagation correction;
finally, training sample vectors and training the heterogeneous network system until the complete part of samples are trained, inputting network operation and maintenance fault alarm information into a trained BP network to locate network faults.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the invention in any way, but any simple modification, equivalent variation and modification of the above embodiments according to the technical principles of the present invention are within the scope of the technical solutions of the present invention.
Claims (8)
1. A multi-agent system packet enclosure control method, characterized in that the multi-agent system packet enclosure control method comprises:
step one, detecting through a sensor network with a known topological structure, and determining the topological structure of the multi-agent system;
combining the relatively static intelligent agents with relatively close positions into a group to simplify a control model;
step three, the control model adopts a centralized control method, and the intelligent agent group with the largest intelligent agent number in the intelligent agent group is used as the central control intelligent agent group of the control model;
step four, the grouped agent groups are encoded;
fifthly, performing agent coding in the agent grouping;
step six, determining the intelligent agent group and the intelligent agent according to the codes, and performing corresponding control operation;
in the fourth step, the coding mode is: placing all the intelligent agent groups in a two-dimensional vector space, establishing a Laplace matrix, wherein the boundary of the aggregation space is the boundary of the matrix, and establishing a space rectangular coordinate system by using the lower left corner boundary, wherein the coordinates of the intelligent agent groups in the matrix are the intelligent agent group codes;
in the fifth step, the encoding priorities of the agents in the group are as follows: high, low, up, down, left and right, and adopts relative position coding by taking the intelligent agent with the strongest calculation force as the center.
2. The multi-agent system packet enclosure control method of claim 1, wherein step two combines relatively close-located and relatively stationary agents into a group to simplify the control model,
selecting an agent evaluation model, calculating a predictive value of each frame of a relatively static agent by comparing an agent with a relatively static agent at an original relatively near position, and marking the obtained frame fraction as a vector X and the total frame number of the agent as N;
the window length of the sliding window is winLen, the sliding window processing is carried out on the obtained frame level quality fraction, namely the frame level fraction of the n-th frame after processing is the average value of the frame level fractions of the [ n-winlen+1, n ] frames, and the frame level fraction after the sliding window processing is marked as a vector WX;
sequencing WX from small to large, marking the sequenced result as WX', taking the average value of the worst p% frames as the quality measurement value of the whole agent topological structure sequence, sequencing, and taking the minimum p% frame average value as the final measurement result;
and sequentially carrying out sliding window treatment on all frame fractions calculated by the agent evaluation model, namely:
wherein, winLen represents window length during sliding window filtering, which is a parameter to be adjusted, X (t) represents quality fraction of the t-th frame, and WX (n) represents quality fraction of the n-th frame after sliding window processing;
and fusing the predicted frame fraction by using a time domain information fusion method based on the inter-frame association and the worst time slot, and finally predicting the fraction:
wherein p% is a parameter to be adjusted, N is the total frame number of the topological structure of the intelligent agent, WX' (t) represents the t th frame fraction after sequencing from small to large, OM winPooling The final evaluation result of the quality of the agent.
3. The multi-agent system packet enclosure control method of claim 2,
and uniformly processing the obtained final evaluation result of the quality of the intelligent agent, and constructing a uniform control model.
4. The multi-agent system packet enclosure control method of claim 3,
and carrying out unified processing on the obtained final evaluation result of the quality of the intelligent agent, wherein the unified processing comprises data selection, data preprocessing and data transformation.
5. The multi-agent system packet enclosure control method of claim 4,
the data transformation comprises data transformation dimension reduction, wherein the data transformation dimension reduction is to cut down the data dimension of the features extracted from the initial features of the data through cluster analysis.
6. The multi-agent system packet enclosure control method of claim 3,
the process for constructing the unified control model comprises the following steps: selecting reference alarm information parameter vector and establishing reference sequence X 0 ，
X 0 ＝{X 0 (k)|k＝1，2，…，n}＝(X 0 (1)，X 0 (2)，…，X 0 (n))
Wherein k represents time of day, X 0 Representing alarm information, wherein n represents the feature dimension of the parameter vector of the alarm information;
secondly, assuming that m comparison fault alarm information data exist, a comparison sequence X is established i
X i ＝{X i (k)|k＝1，2，…，n}＝(X i (1)，X i (2)，…，X i (n))i＝1，2，…，m
Then, a comparison sequence X is established i For reference number series X 0 Correlation coefficient ζ at time k i (k)
Where ρ is the resolution factor and, ρ is e [0 ], ++ infinity a) is provided; the larger the ρ, the greater the resolution; the smaller the ρ, the smaller the resolution;
finally, calculate the comparison sequence X i For reference number series X 0 Is related to the degree of association of (2)
7. The multi-agent system grouping and surrounding control method of claim 3, wherein constructing a unified control model further comprises establishing a mapping relation between agent alarm information and faults between networks through association analysis, constructing a fault location model, and performing network fault location through a BP neural network; the method comprises the following steps:
first, an m-dimensional alert vector Q is obtained n ＝(s 1 ，s 2 ，s 3 …s m ) And n-dimensional fault vector O n ＝(p 1 ，p 2 ，p 3 …p m ) The system is enabled to have a parallel structure and parallel processing capability by simultaneously inputting the system through a plurality of network nodes, and the input is dynamically processed in real time;
secondly, the BP network gives a value in a specified range to each connected weight, and simultaneously, a threshold value is specified to each neuron node;
thirdly, providing a group input alarm sample machine target result to a network, and calculating the connection weight, the threshold value and the input and output values of each hidden layer unit of the neural network node;
then, positive layer errors: calculating the error of the output layer unit by using the target vector and the network actual output value, correcting the connection weight and the threshold value by combining the output of each unit of the hidden layer, and carrying out reverse error propagation correction;
finally, training sample vectors and training the heterogeneous network system until the complete part of samples are trained, inputting network operation and maintenance fault alarm information into a trained BP network to locate network faults.
8. A multi-agent system packet enclosure control system implementing the multi-agent system packet enclosure control method of claim 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910318251.7A CN110084295B (en) | 2019-04-19 | 2019-04-19 | Multi-agent system grouping surrounding control method and control system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910318251.7A CN110084295B (en) | 2019-04-19 | 2019-04-19 | Multi-agent system grouping surrounding control method and control system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110084295A CN110084295A (en) | 2019-08-02 |
| CN110084295B true CN110084295B (en) | 2023-05-05 |
Family
ID=67415747
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910318251.7A Active CN110084295B (en) | 2019-04-19 | 2019-04-19 | Multi-agent system grouping surrounding control method and control system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110084295B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110503195A (en) * | 2019-08-14 | 2019-11-26 | 北京中科寒武纪科技有限公司 | The method and its Related product of task are executed using artificial intelligence process device |
| CN111766783B (en) * | 2020-06-29 | 2021-08-24 | 北京航空航天大学 | Cluster system-oriented formation enclosure tracking method capable of converging in limited time |
| CN113146624B (en) * | 2021-03-25 | 2022-04-29 | 重庆大学 | Multi-agent control method based on maximum angle aggregation strategy |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7777741B2 (en) * | 2005-03-04 | 2010-08-17 | Aiseek Ltd. | Techniques for accurately determining visibility of objects from multiple viewpoints in a three-dimensional (3D) environment |
| CN104469833B (en) * | 2015-01-08 | 2018-08-21 | 西安电子科技大学 | A kind of heterogeneous network operation management method based on user's perception |
| CN105429185B (en) * | 2015-12-08 | 2018-02-27 | 华南理工大学 | A kind of economic load dispatching method with the harmonious property of robust |
| CN105589333B (en) * | 2016-02-03 | 2018-09-07 | 电子科技大学 | Control method is surrounded in multi-agent system grouping |
| CN105979266B (en) * | 2016-05-06 | 2019-01-29 | 西安电子科技大学 | It is a kind of based on intra-frame trunk and the worst time-domain information fusion method of time slot |
| CN106502090B (en) * | 2016-09-29 | 2019-03-01 | 电子科技大学 | The positioning of multi-agent system and around control method in a kind of three-dimensional space |
| CN108092833A (en) * | 2018-02-28 | 2018-05-29 | 哈尔滨理工大学 | A kind of multiple agent uniformity method for diagnosing faults |
| CN108681320A (en) * | 2018-05-11 | 2018-10-19 | 北京理工大学 | A kind of distributed multi agent real-time fault detection method based on regional cooperative |
| CN109299305A (en) * | 2018-10-30 | 2019-02-01 | 湖北工业大学 | A kind of spatial image searching system based on multi-feature fusion and search method |
-
2019
- 2019-04-19 CN CN201910318251.7A patent/CN110084295B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110084295A (en) | 2019-08-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107396322B (en) | Indoor positioning method based on path matching and coding-decoding cyclic neural network | |
| CN110084295B (en) | Multi-agent system grouping surrounding control method and control system | |
| CN110223517B (en) | Short-term traffic flow prediction method based on space-time correlation | |
| CN112560656B (en) | Pedestrian multi-target tracking method combining attention mechanism end-to-end training | |
| CN110488611B (en) | Bionic robot fish motion control method, controller and bionic robot fish | |
| CN112200165A (en) | Model training method, human body posture estimation method, device, equipment and medium | |
| CN111612243A (en) | Traffic speed prediction method, system and storage medium | |
| CN114022847A (en) | Intelligent agent trajectory prediction method, system, equipment and storage medium | |
| Kim et al. | Structural recurrent neural network for traffic speed prediction | |
| Su et al. | Uncertainty quantification of collaborative detection for self-driving | |
| CN109918995B (en) | Crowd abnormity detection method based on deep learning | |
| CN117077727B (en) | Track prediction method based on space-time attention mechanism and neural ordinary differential equation | |
| CN113362368A (en) | Crowd trajectory prediction method based on multi-level space-time diagram neural network | |
| Kim et al. | Scalable learning with a structural recurrent neural network for short-term traffic prediction | |
| CN113298191A (en) | User behavior identification method based on personalized semi-supervised online federal learning | |
| CN111488552B (en) | Close-proximity multi-target tracking method based on Gaussian mixture probability hypothesis density | |
| CN104298650B (en) | Multi-method fusion based Kalman filtering quantization method | |
| Niwa et al. | Spatio-temporal graph localization networks for image-based navigation | |
| CN104318072B (en) | QKF-MMF (Quantitative Kalman Filtering-Multi Method Fusion) based multi-sensor quantitative fusion method | |
| CN117029838B (en) | Navigation control method and system for underwater robot | |
| CN117314956A (en) | Interactive pedestrian track prediction method based on graphic neural network | |
| CN112149896A (en) | Attention mechanism-based mechanical equipment multi-working-condition fault prediction method | |
| CN115393955A (en) | Gesture recognition method and system based on BVP and WiTransformer by utilizing WiFi | |
| CN115205750A (en) | Motion real-time counting method and system based on deep learning model | |
| Postnikov et al. | Conditioned Human Trajectory Prediction using Iterative Attention Blocks |
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 |