CN116362413A - Unmanned platform collaboration system toughness measurement method based on neural network verification - Google Patents

Unmanned platform collaboration system toughness measurement method based on neural network verification Download PDF

Info

Publication number
CN116362413A
CN116362413A CN202310409814.XA CN202310409814A CN116362413A CN 116362413 A CN116362413 A CN 116362413A CN 202310409814 A CN202310409814 A CN 202310409814A CN 116362413 A CN116362413 A CN 116362413A
Authority
CN
China
Prior art keywords
vector
input
training
toughness
unmanned platform
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.)
Pending
Application number
CN202310409814.XA
Other languages
Chinese (zh)
Inventor
张涛
朱凯
白玉琦
陆耿
贾庆山
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.)
Tsinghua University
Original Assignee
Tsinghua University
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 Tsinghua University filed Critical Tsinghua University
Priority to CN202310409814.XA priority Critical patent/CN116362413A/en
Publication of CN116362413A publication Critical patent/CN116362413A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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
    • G06N3/048—Activation functions
    • 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
    • G06N3/0499—Feedforward 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/10—Office automation; Time management
    • G06Q10/101—Collaborative creation, e.g. joint development of products or services
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Abstract

The invention discloses an unmanned platform collaboration system toughness measurement method based on neural network verification, which comprises the following steps: a diversified unmanned platform cooperative scheme is oriented, and parameter combinations and sample data sets of all possible conditions in the performance evaluation process can be traversed are obtained; constructing an artificial neuron network, and training a performance prediction model by adopting a random gradient descent mode; based on a neural network verification method, linear approximation is carried out on the upper and lower boundaries of the interval according to the neuron activation region, and boundary propagation is carried out, so that the efficiency prediction global lower boundary corresponding to the given input disturbance vector is obtained; according to the characteristics of system efficiency influence factors, score-level order-increasing disturbance vectors are designed, and the hierarchical measurement of the toughness of the unmanned platform collaboration system is given. The method fully utilizes mass data, provides an analysis boundary which can be efficiently calculated within polynomial time complexity, and reduces the calculated amount and the test cost.

Description

Unmanned platform collaboration system toughness measurement method based on neural network verification
Technical Field
The invention relates to the technical field of efficiency evaluation and artificial intelligence, in particular to an unmanned platform collaboration system toughness measurement method based on neural network verification.
Background
In future society, unmanned and multi-platform systematic joint actions become key development directions of autonomous unmanned systems, and especially when complex cooperative tasks such as emergency rescue, urban security and the like are oriented, the cooperative operation of heterogeneous unmanned platforms can fully exert the system efficiency. However, due to the significant differences of the heterogeneous unmanned platform in terms of behavior mode, intelligent level and the like, great difficulties are brought to the cooperative organization of the heterogeneous unmanned platform. Not only the cooperative efficiency of the unmanned platform system is improved, but also the robustness under the unknown random interference condition is ensured, and the unmanned system with high efficiency and high toughness is constructed. The toughness index of a given synergistic scheme is rapidly and accurately evaluated, more comprehensive guidance is provided for system optimization, the robustness and the overall efficiency of the system are improved, and the method has important research value.
Currently, research on the aspect of toughness measurement of an unmanned platform cooperative system is lacking, and one feasible method is based on Monte Carlo theory, and the system efficiency variance and the worst result under the multi-dimensional disturbance factor are estimated through simulation deduction and serve as one measure of toughness. However, the method has low efficiency, various disturbance factors are required to be repeatedly added for each technical scheme to carry out deduction, and the toughness of a large number of system schemes is difficult to evaluate in a short time.
On the other hand, a large amount of data is accumulated in unmanned platform collaborative task simulation and training, mass data is processed in an off-line mode through an artificial intelligence method, online tedious calculation of a traditional method is avoided, wide attention of researchers is brought, and an intelligent evaluation means is provided for combat effectiveness. The neural network verification theory is mainly used for verifying the robustness of the neural network to input against sample attack, and the neural network is introduced into the efficiency evaluation of the unmanned platform cooperative system, so that a new direction is provided for the measurement of the toughness of the system.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems in the related art to some extent.
Therefore, the invention aims to provide an unmanned platform collaboration system toughness measurement method based on neural network verification, aiming at different collaboration tasks, by modifying the input disturbance range of a performance prediction neural network, a lower bound of performance output is given, and toughness grades are evaluated.
The invention further aims at providing an unmanned platform collaboration system toughness measuring device based on neural network verification.
In order to achieve the above purpose, in one aspect, the present invention provides a method for measuring toughness of an unmanned platform collaboration system based on neural network verification, including:
obtaining sample data for training and testing based on an unmanned platform collaboration scheme; wherein the sample data includes a feature input vector and a performance output vector;
inputting a characteristic input vector in sample data for training to an MLP efficacy prediction model for model training to output a model prediction output vector, and testing the efficacy prediction model by using sample data for testing according to a true value comparison result of the model prediction output vector and the efficacy output vector for training to obtain a trained MLP efficacy prediction model according to a model test result;
based on model parameters of the trained MLP efficacy prediction model, carrying out limit propagation on the final characteristic input vector under a given disturbance condition to obtain a limit propagation result;
and carrying out toughness measurement on the unmanned platform cooperative scheme based on the limit transmission result to carry out grade measurement on the toughness measurement so as to obtain a toughness measurement grade measurement result.
In addition, the unmanned platform collaboration system toughness measurement method based on neural network verification according to the embodiment of the invention can also have the following additional technical characteristics:
further, in one embodiment of the invention, the feature input vector includes a quantitative parameter characterization based on system efficacy influencing factors; wherein the system effectiveness influencing factors comprise a plurality of communication capacity, protection capacity, node number, node type and formation combination forms;
the efficiency output vector is a target efficiency value.
Further, in one embodiment of the present invention, the MLP performance prediction model is constructed, and feature input vectors for training are set
Figure BDA0004182893820000021
m layers of neurons, each layer of neurons has the number of n k And adopts a ReLU activation function sigma (y) =max (y, 0), and the weight matrix is expressed as +.>
Figure BDA0004182893820000022
The bias vector is denoted +.>
Figure BDA0004182893820000023
Make->
Figure BDA0004182893820000024
Representing the mapping operator from the input layer to the kth layer, there is then the following relationship: phi (phi) o (x)＝x，φ k (x)＝σ(W (k) φ k-1 (x)+b (k) )，for k∈[m-1]，φ m (x)＝W (m) φ m-1 (x)+b (m) The performance output vector for training is f (x) =φ m (x)。
Further, in an embodiment of the present invention, the performing, based on the model parameters of the trained MLP efficacy prediction model, the boundary propagation under the given disturbance condition on the final feature input vector obtains a boundary propagation result, including:
s31, preset at l p Input vector to final feature under norm
Figure BDA0004182893820000031
Applying a bounded perturbation of +.>
Figure BDA00041828938200000312
I.e.
Figure BDA0004182893820000032
Wherein p is E [1, ] infinity]；
S32, each neuron has an upper bound and a lower bound before being activated, and the upper bound and the lower bound of the kth layer of the r-th neuron are respectively expressed as
Figure BDA0004182893820000033
And->
Figure BDA0004182893820000034
Actual value is +.>
Figure BDA0004182893820000035
Representation->
Figure BDA0004182893820000036
S33, for the upper and lower boundaries of the input layer, adopting l ∞ The norm is relaxed to give [ l ] (0) ，u (0) ]＝[x 0 -∈，x 0 +∈]；
S34, for the upper and lower output bounds of the k E [ m ] th layer, classifying based on the situation of the previous layer:
case 1, if
Figure BDA0004182893820000037
The linear region of the ReLU function is activated, < ->
Figure BDA0004182893820000038
Case 2, if
Figure BDA0004182893820000039
The suppression area of the ReLU function is activated, < ->
Figure BDA00041828938200000310
Case 3, if
Figure BDA00041828938200000311
A linear approximation is made between the upper and lower boundary regions of the activation region,
Figure BDA0004182893820000041
s35, defining a diagonal matrix
Figure BDA0004182893820000042
Wherein the method comprises the steps of
Figure BDA0004182893820000043
Definition of the definition
Figure BDA0004182893820000044
Figure BDA0004182893820000045
Definition A (m-1) ＝W (m) D (m-1) ，A (k-1 )＝A (k) W (k) D (k-1) (k∈[m-1])；
S36, recursively using S34 and S35, and obtaining a j-th output interval lower bound value for each characteristic input vector x as follows:
Figure BDA0004182893820000046
further, in an embodiment of the present invention, the ranking the toughness metrics of the unmanned platform collaboration solution based on the boundary propagation results to obtain toughness metric ranking results includes:
s41, given a certain input disturbance vector
Figure BDA0004182893820000047
Figure BDA0004182893820000048
The global performance lower bound is the minimum of all possible input quantities corresponding to the lower bound:
Figure BDA0004182893820000051
s42, based on the system efficiency influence factors, a hierarchical disturbance vector E for multi-dimensional input parameter combination (A) ，∈ (B) ，...，∈ (F) Wherein = [ -e) 1 ，∈ 1 ，∈ 3 ，……]The disturbance degree increases outwards in sequence; e is calculated by S41 respectively (A) ，∈ (B) ，...，∈ (F) The corresponding lower bound of output efficiency up to E (α) And if the lower performance bound of the unmanned platform cooperative scheme is lower than a preset threshold, finally obtaining the toughness measurement grade of the unmanned platform cooperative scheme as alpha grade.
In order to achieve the above objective, another aspect of the present invention provides an unmanned platform collaboration system toughness measurement device based on neural network verification, including:
the data acquisition module is used for acquiring sample data for training and testing based on an unmanned platform cooperative scheme; wherein the sample data includes a feature input vector and a performance output vector;
the model training module is used for inputting the characteristic input vector in the sample data for training to the MLP efficacy prediction model for model training so as to output a model prediction output vector, and testing the efficacy prediction model by using the sample data for testing according to the true value comparison result of the model prediction output vector and the efficacy output vector for training so as to obtain a trained MLP efficacy prediction model according to the model test result;
the limit propagation module is used for carrying out limit propagation on the final characteristic input vector under a given disturbance condition based on the model parameters of the trained MLP efficacy prediction model to obtain a limit propagation result;
and the toughness grading module is used for grading the toughness measurement of the unmanned platform cooperative scheme based on the limit transmission result so as to obtain a toughness measurement grade measurement result.
According to the method and the device for measuring the toughness of the unmanned platform collaboration system based on the neural network verification, provided by the embodiment of the invention, the hierarchical order-increasing disturbance vector of the score hierarchy can be designed according to the characteristics of the system efficiency influence factors, the hierarchical measurement of the toughness of the unmanned platform collaboration system is given, mass data is fully utilized, an analysis boundary capable of being efficiently calculated within polynomial time complexity is provided, and the calculation amount and the test cost are reduced.
The beneficial effects of the invention are as follows:
1) The method fully utilizes mass data of simulation and test tests based on an artificial intelligence method, fully learns the influence of various factors on the final system efficiency in offline calculation, and encodes the potential influence into the weight and bias parameters of the neural network;
2) According to the invention, through the linear approximation of the ReLU network, the analysis boundary of the efficiency output value which can be efficiently calculated within the polynomial time complexity is provided, and the approximate lower boundary of all possible efficiency in a given disturbance range can be traversed by only one calculation, so that the calculation amount and the test cost required by the traditional method for evaluating the toughness of the unmanned platform cooperative system are reduced.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:
FIG. 1 is a flow chart of an unmanned platform collaboration system toughness measurement method based on neural network verification according to an embodiment of the invention;
FIG. 2 is a schematic diagram of activation function limit approximation according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of boundary propagation according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of input disturbance levels according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of toughness metric ratings according to an embodiment of the present invention;
fig. 6 is a block diagram of an unmanned platform collaboration system toughness measurement apparatus based on neural network verification according to an embodiment of the present invention.
Detailed Description
It should be noted that, without conflict, the embodiments of the present invention and features of the embodiments may be combined with each other. The invention will be described in detail below with reference to the drawings in connection with embodiments.
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
The following describes a method and a device for measuring toughness of an unmanned platform collaboration system based on neural network verification according to an embodiment of the invention with reference to the accompanying drawings.
Fig. 1 is a flowchart of an unmanned platform collaboration system toughness measurement method based on neural network verification according to an embodiment of the invention.
As shown in fig. 1, the method includes, but is not limited to, the steps of:
s1, obtaining sample data for training and testing based on an unmanned platform cooperative scheme; the sample data includes a feature input vector and a performance output vector.
It can be understood that the invention is oriented to diversified unmanned platform cooperative schemes such as multi-type formation combinations, multi-node parameter configuration, complex decision logic and the like, obtains parameter combinations capable of traversing all possible situations in the performance evaluation process, and provides training sample data and test sample data containing general rules for the performance evaluation model construction.
In an embodiment of the invention, the sample data is composed of a combination of a feature input vector and a performance output vector. The input vector comprises quantitative parameter characterization of system performance influence factors such as connectivity, protection capability, node number, node type, formation combination form and the like; the output vector is a target efficiency value and is obtained through expert scoring, setting experience weight or simulation deduction, semi-physical test and other modes.
S2, inputting the characteristic input vector in the sample data for training to the MLP efficacy prediction model for model training to output to obtain a model prediction output vector, and testing the efficacy prediction model by using the sample data for testing according to a true value comparison result of the model prediction output vector and the efficacy output vector for training to obtain a trained MLP efficacy prediction model according to a model test result.
Specifically, the MLP efficacy prediction model is constructed, and the feature input vector for training is set
Figure BDA0004182893820000081
m layers of neurons, each layer of neurons has the number of n k And adopts a ReLU activation function sigma (y) =max (y, 0), and the weight matrix is expressed as
Figure BDA0004182893820000082
The bias vector is denoted +.>
Figure BDA0004182893820000083
Make->
Figure BDA0004182893820000084
Representing the mapping from the input layer to the kth layerThe shooting operator has the following relation: phi (phi) 0 (x)＝x，φ k (x)＝σ(W (k) φ k-1 (x)+b (k) )，for k∈[m-1]，φ m (x)＝W (m) φ m-1 (x)+b (m) The performance output vector for training is f (x) =φ m (x)。
Further, a neural network model is trained by using a random gradient descent algorithm, so that a model prediction output vector obtained under a given sample input vector can be as close as possible to a sample output vector true value, a network with the highest precision on a test data set can be selected, and after full training of massive data, the network can better predict system efficiency, so that a trained MLP efficiency prediction model can be obtained according to a model test result.
S3, based on model parameters of the trained MLP efficacy prediction model, carrying out limit propagation on the final characteristic input vector under a given disturbance condition to obtain a limit propagation result.
Specifically, after obtaining a fully trained MLP performance prediction model, using its network weights W and bias b, for each particular unmanned platform collaboration scenario (i.e., a particular system feature input vector
Figure BDA0004182893820000085
) The boundary propagation for a given disturbance condition is achieved as follows. Comprising the following steps:
s31, assume that the input vector is under lp norm
Figure BDA0004182893820000086
Applying a bounded perturbation of +.>
Figure BDA0004182893820000087
I.e.
Figure BDA0004182893820000088
Wherein p is [1, ];
s32, each neuron has an upper bound and a lower bound before being activated, and the upper bound and the lower bound of the kth layer of the r-th neuron are respectively expressed as
Figure BDA0004182893820000091
And->
Figure BDA0004182893820000092
Actual value is +.>
Figure BDA0004182893820000093
Representation->
Figure BDA0004182893820000094
S33, for the upper and lower boundaries of the input layer, adopting l ∞ The norm is relaxed and [ l ] is easily obtained (0) ，u (0) ]＝[x 0 -∈，x 0 +∈]；
S34, for the upper and lower output bounds of the k E [ m ] th layer, classification processing can be performed based on the situation of the previous layer:
case 1, if
Figure BDA0004182893820000095
The linear region of the ReLU function is activated,
Figure BDA0004182893820000096
case 2, if
Figure BDA0004182893820000097
The suppression area of the ReLU function is activated, < ->
Figure BDA0004182893820000098
Case 3, if
Figure BDA0004182893820000099
It is necessary to perform a linear approximation of the upper and lower boundary regions of the activation region, and FIG. 2 shows the principle of approximation of the boundary of the activation function, +.>
Figure BDA00041828938200000912
S35, further, to facilitate unified representation and recursive computation, defining a diagonal matrix
Figure BDA00041828938200000910
Wherein the method comprises the steps of
Figure BDA00041828938200000911
Definition of the definition
Figure BDA0004182893820000101
Figure BDA0004182893820000102
Definition A (m-1) ＝W (m) D (m-1) ，A (k-1 )＝A (k) W (k) D (k-1) (k∈[m-1])。
S36, recursively using S34 and S35, for each input vector x, obtaining a j-th output interval lower bound value as follows:
Figure BDA0004182893820000103
fig. 3 gives a schematic diagram of the boundary propagation, where the output is only one-dimensional (i.e. j=1).
And S4, carrying out toughness measurement on the unmanned platform cooperative scheme based on the limit transmission result, and carrying out grade measurement on the toughness measurement to obtain a toughness measurement grade measurement result.
Specifically, this step performs a toughness grading measure, comprising:
s41, given a certain input disturbance vector
Figure BDA0004182893820000104
Figure BDA0004182893820000105
The global performance lower bound is the corresponding lower bound of all possible input quantitiesThe minimum value of (3):
Figure BDA0004182893820000111
wherein, based on dual norm theory
Figure BDA0004182893820000112
Figure BDA0004182893820000113
And->
Figure BDA0004182893820000114
S42, according to the characteristics of the system efficiency influence factors, FIG. 4 shows a manually defined hierarchical disturbance vector E for multi-dimensional input parameter combinations (A) ，∈ (B) ，...，∈ (F) Wherein = [ -e) 1 ，∈ 1 ，∈ 3 ，……]The disturbance degree increases outwards in sequence; then, as shown in FIG. 5, S41 is used to calculate E (A) ，∈ (B) ，…，∈ (F) The corresponding lower bound of output efficiency up to E (α) The lower performance bound of the battle plan is below a set threshold (e.g., winning probability is below 80%), and the toughness metric rating of the battle plan is ultimately alpha.
According to the neural network verification-based unmanned platform collaboration system toughness measurement method, the calculation cost is reduced, the approximate lower bounds of all possible effectiveness in a given disturbance range can be traversed, and the calculation amount and the test cost required by the traditional method for evaluating the unmanned platform collaboration system toughness are reduced.
In order to implement the above embodiment, as shown in fig. 6, an unmanned platform collaboration system toughness measurement apparatus 10 based on neural network verification is further provided in this embodiment, where the apparatus 10 includes a data acquisition module 100, a model training module 200, a limit propagation module 300, and a toughness classification module 400.
A data acquisition module 100 for obtaining sample data for training and testing based on an unmanned platform collaboration scheme; the sample data comprises a characteristic input vector and a performance output vector;
the model training module 200 is configured to input a feature input vector in sample data for training to an MLP efficacy prediction model for model training to output a model prediction output vector, and test the efficacy prediction model by using sample data for testing according to a true value comparison result of the model prediction output vector and the efficacy output vector for training to obtain a trained MLP efficacy prediction model according to a model test result;
the limit propagation module 300 is configured to perform limit propagation on the final feature input vector under a given disturbance condition based on model parameters of the trained MLP efficacy prediction model to obtain a limit propagation result;
the toughness grading module 400 is configured to grade the toughness metric of the unmanned platform collaboration scheme based on the boundary propagation result, so as to obtain a toughness metric grade measurement result.
Further, the characteristic input vector comprises quantitative parameter characterization based on system efficiency influence factors; the system efficiency influence factors comprise a plurality of communication capacity, protection capacity, node number, node type and formation combination forms;
the performance output vector is the target performance value.
Further, the model training module 200 is further configured to construct an MLP efficacy prediction model, and set a feature input vector for training
Figure BDA0004182893820000121
m layers of neurons, each layer of neurons has the number of n k And adopts a ReLU activation function sigma (y) =max (y, 0), and the weight matrix is expressed as +.>
Figure BDA0004182893820000122
The bias vector is denoted +.>
Figure BDA0004182893820000123
Make->
Figure BDA0004182893820000124
Representing the mapping operator from the input layer to the kth layer, there is then the following relationship: phi (phi) o (x)＝x，φ k (x)＝σ(W (k) φ k-1 (x)+b (k) )，for k∈[m-1]，φ m (x)＝W (m) φ m-1 (x)+b (m) The performance output vector for training is f (x) =φ m (x)。
Further, the limit propagation module 300 includes:
an initial disturbance subunit, for presetting at l p Input vector to final feature under norm
Figure BDA0004182893820000131
Applying a bounded perturbation of +.>
Figure BDA0004182893820000132
I.e. < ->
Figure BDA0004182893820000133
Wherein p is [1, ];
a neuron subunit for each neuron, before being activated, having an upper bound and a lower bound, the upper and lower bounds of the kth layer, r-th neuron, respectively, being represented as
Figure BDA0004182893820000134
And->
Figure BDA0004182893820000135
Actual value is +.>
Figure BDA0004182893820000136
Representation of
Figure BDA0004182893820000137
A relaxation subunit for adopting l for upper and lower boundaries of the input layer ∞ The norm is relaxed to give [ l ] (0) ，u (0) ]＝[x 0 -∈，x 0 +∈]；
The classifying processing subunit is used for classifying the upper and lower output bounds of the k-th E [ m ] layer based on the condition of the previous layer:
case 1, if
Figure BDA0004182893820000138
The linear region of the ReLU function is activated, < ->
Figure BDA0004182893820000139
Case 2, if
Figure BDA00041828938200001310
The suppression area of the ReLU function is activated, < ->
Figure BDA00041828938200001311
Case 3, if
Figure BDA00041828938200001312
A linear approximation is made between the upper and lower boundary regions of the activation region,
Figure BDA0004182893820000141
matrix defining subunit for defining diagonal matrix
Figure BDA0004182893820000142
Wherein the method comprises the steps of
Figure BDA0004182893820000143
Definition of the definition
Figure BDA0004182893820000144
Figure BDA0004182893820000145
Definition A (m-1) ＝W (m) D (m-1) ，A (k-1) ＝A (k) W (k) D (k-1) (k∈[m-1])；
The recursion subunit is configured to recursively utilize the classification processing subunit and the matrix defining subunit, and input the vector x to each feature to obtain a j-th output interval lower bound value as follows:
Figure BDA0004182893820000146
further, the toughness classification module 400 includes:
an input disturbance subunit for giving a certain input disturbance vector
Figure BDA0004182893820000147
Figure BDA0004182893820000148
The global performance lower bound is the minimum of all possible input quantities corresponding to the lower bound:
Figure BDA0004182893820000151
a toughness measurement subunit, configured to, based on the system performance influencing factor, use a hierarchical disturbance vector e for a multi-dimensional input parameter combination (A) ，∈ (B) ，…，∈ (F) Wherein = [ -e) 1 ，∈ 1 ，∈ 3 ，……]The disturbance degree increases outwards in sequence; e is calculated by using the input scrambling subunit respectively (A) ，∈ (B) ，…，∈ (F) The corresponding lower bound of output efficiency up to E (α) If the lower performance bound of the unmanned platform cooperative scheme is lower than a preset threshold, the toughness measurement grade of the unmanned platform cooperative scheme is finally obtained α A stage.
According to the unmanned platform collaboration system toughness measurement device based on neural network verification, the calculation cost is reduced, the approximate lower bounds of all possible effectiveness in a given disturbance range can be traversed, and the calculation amount and the test cost required by the traditional method for evaluating the unmanned platform collaboration system toughness are reduced.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present invention, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.

Claims (10)

1. The unmanned platform collaboration system toughness measurement method based on neural network verification is characterized by comprising the following steps of:
obtaining sample data for training and testing based on an unmanned platform collaboration scheme; wherein the sample data includes a feature input vector and a performance output vector;
inputting a characteristic input vector in sample data for training to an MLP efficacy prediction model for model training to output a model prediction output vector, and testing the efficacy prediction model by using sample data for testing according to a true value comparison result of the model prediction output vector and the efficacy output vector for training to obtain a trained MLP efficacy prediction model according to a model test result;
based on model parameters of the trained MLP efficacy prediction model, carrying out limit propagation on the final characteristic input vector under a given disturbance condition to obtain a limit propagation result;
and carrying out toughness measurement on the unmanned platform cooperative scheme based on the limit transmission result to carry out grade measurement on the toughness measurement so as to obtain a toughness measurement grade measurement result.
2. The method of claim 1, wherein the feature input vector comprises a quantitative parametric characterization based on system performance impact factors; wherein the system effectiveness influencing factors comprise a plurality of communication capacity, protection capacity, node number, node type and formation combination forms;
the efficiency output vector is a target efficiency value.
3. The method of claim 1, wherein constructing the MLP efficacy prediction model sets feature input vectors for training
Figure FDA0004182893810000011
m layers of neurons, each layer of neurons has the number of n k And adopts a ReLU activation function sigma (y) =max (y, 0), and the weight matrix is expressed as +.>
Figure FDA0004182893810000012
The bias vector is denoted +.>
Figure FDA0004182893810000013
Make->
Figure FDA0004182893810000014
Representing the mapping operator from the input layer to the kth layer, there is then the following relationship: phi (phi) 0 (x)＝x，φ k (x)＝σ(W (k) φ k-1 (x)+b (k) )，for k∈[m-1]，φ m (x)＝W (m) φ m-1 (x)+b (m) The performance output vector for training is f (x) =φ m (x)。
4. A method according to claim 3, wherein said performing boundary propagation on the final feature input vector under given perturbation conditions based on model parameters of the trained MLP efficacy prediction model yields boundary propagation results, comprising:
s31, preset at l p Input vector to final feature under norm
Figure FDA0004182893810000021
Applying a bounded perturbation of +.>
Figure FDA0004182893810000022
I.e.
Figure FDA0004182893810000023
Wherein p is [1, ];
s32, each neuron has an upper bound and a lower bound before being activated, and the upper bound and the lower bound of the kth layer of the neurons are respectively expressed as
Figure FDA0004182893810000024
And->
Figure FDA0004182893810000025
Actual value is +.>
Figure FDA0004182893810000026
Representation->
Figure FDA0004182893810000027
S33, for the upper and lower boundaries of the input layer, adopting l ∞ The norm is relaxed to give [ l ] (0) ，ｕ (０) ]＝[x 0 -∈，x 0 +∈]；；
S34, for the kth ε [ m ]]The upper and lower boundaries of the layer output are classified based on the previous layer condition: case 1, if
Figure FDA0004182893810000028
The linear region of the ReLU function is activated, < ->
Figure FDA0004182893810000029
Case 2, if
Figure FDA00041828938100000210
The suppression area of the ReLU function is activated, < ->
Figure FDA00041828938100000211
Case 3, if
Figure FDA0004182893810000031
A linear approximation is made between the upper and lower boundary regions of the activation region,
Figure FDA0004182893810000032
s35, defining a diagonal matrix
Figure FDA0004182893810000033
Wherein the method comprises the steps of
Figure FDA0004182893810000034
Definition of the definition
Figure FDA0004182893810000035
Figure FDA0004182893810000036
Definition A (m-1) ＝W (m) D (m-1 )，A (k-1) ＝A (k) W (k) D (k-1) (k∈[m-1])；
S36, recursively using S34 and S35, and obtaining a j-th output interval lower bound value for each characteristic input vector x as follows:
Figure FDA0004182893810000037
5. the method of claim 4, wherein ranking the toughness metrics for the unmanned platform collaboration scenario based on the boundary propagation results to obtain toughness metric ranking results comprises:
s41, given a certain input disturbance vector
Figure FDA0004182893810000038
Figure FDA0004182893810000041
The global performance lower bound is the minimum of all possible input quantities corresponding to the lower bound:
Figure FDA0004182893810000042
s42, based on the system efficiency influence factors, a hierarchical disturbance vector E for multi-dimensional input parameter combination (A) ，∈ (B) ，...，∈ (F) Wherein = [ -e) 1 ，∈ 1 ，∈ 3 ......]The disturbance degree increases outwards in sequence; e is calculated by S41 respectively (A) ，∈ (B) ，...，∈ (F) The corresponding lower bound of output efficiency up to E (α) The lower performance bound of the (E) is lower than a preset threshold value, and finally the toughness measurement and the like of the unmanned platform cooperative scheme are obtainedThe stage is the alpha stage.
6. Unmanned platform cooperative system toughness measurement device based on neural network verifies, its characterized in that includes:
the data acquisition module is used for acquiring sample data for training and testing based on an unmanned platform cooperative scheme; wherein the sample data includes a feature input vector and a performance output vector;
the model training module is used for inputting the characteristic input vector in the sample data for training to the MLP efficacy prediction model for model training so as to output a model prediction output vector, and testing the efficacy prediction model by using the sample data for testing according to the true value comparison result of the model prediction output vector and the efficacy output vector for training so as to obtain a trained MLP efficacy prediction model according to the model test result;
the limit propagation module is used for carrying out limit propagation on the final characteristic input vector under a given disturbance condition based on the model parameters of the trained MLP efficacy prediction model to obtain a limit propagation result;
and the toughness grading module is used for grading the toughness measurement of the unmanned platform cooperative scheme based on the limit transmission result so as to obtain a toughness measurement grade measurement result.
7. The apparatus of claim 6, wherein the feature input vector comprises a quantitative parametric characterization based on system performance impact factors; wherein the system effectiveness influencing factors comprise a plurality of communication capacity, protection capacity, node number, node type and formation combination forms;
the efficiency output vector is a target efficiency value.
8. The apparatus of claim 7, wherein the model training module is further configured to construct the MLP efficacy prediction model, set feature input vectors for training
Figure FDA0004182893810000051
m layers of neurons, each layer of neurons has the number of n k And adopts a ReLU activation function sigma (y) =max (y, 0), and the weight matrix is expressed as +.>
Figure FDA0004182893810000052
The bias vector is expressed as
Figure FDA0004182893810000053
Make->
Figure FDA0004182893810000054
Representing the mapping operator from the input layer to the kth layer, there is then the following relationship: phi (phi) 0 (x)＝x，φ k (x)＝σ(W (k) φ k-1 (x)+b (k) )，for k∈[m-1]，φ m (x)＝W (m) φ m-1 (x)+b (m) The performance output vector for training is f (x) =φ m (x)。
9. The apparatus of claim 8, wherein the limit propagation module comprises:
an initial disturbance subunit, for presetting at l p Input vector to final feature under norm
Figure FDA0004182893810000061
Applying a bounded disturbance as
Figure FDA0004182893810000062
I.e. < ->
Figure FDA0004182893810000063
Wherein p is [1, ];
a neuron subunit for each neuron, before being activated, having an upper bound and a lower bound, the upper and lower bounds of the kth layer of neurons being represented as
Figure FDA0004182893810000064
And->
Figure FDA0004182893810000065
Actual value is +.>
Figure FDA0004182893810000066
Representation->
Figure FDA0004182893810000067
A relaxation subunit for adopting l for upper and lower boundaries of the input layer ∞ The norm is relaxed to give [ l ] (0) ，u (0) ]＝[x 0 -∈，x 0 +∈]；
The classifying processing subunit is used for classifying the upper and lower output bounds of the k-th E [ m ] layer based on the condition of the previous layer:
case 1, if
Figure FDA0004182893810000068
The linear region of the ReLU function is activated, < ->
Figure FDA0004182893810000069
Case 2, if
Figure FDA00041828938100000610
The suppression area of the ReLU function is activated, < ->
Figure FDA00041828938100000611
Case 3, if
Figure FDA00041828938100000612
A linear approximation is made between the upper and lower boundary regions of the activation region,
Figure FDA00041828938100000613
matrix defining subunit for defining diagonal matrix
Figure FDA00041828938100000614
Wherein the method comprises the steps of
Figure FDA0004182893810000071
Definition of the definition
Figure FDA0004182893810000072
Figure FDA0004182893810000073
Definition A (m-1) ＝W (m) D (m-1) ，A (k-1) ＝A (k) W (k) D (k-1) (k∈[m-1])；
The recursion subunit is configured to recursively utilize the classification processing subunit and the matrix defining subunit, and input the vector x to each feature to obtain a j-th output interval lower bound value as follows:
Figure FDA0004182893810000074
10. the apparatus of claim 9, wherein the toughness classification module comprises:
an input disturbance subunit for giving a certain input disturbance vector
Figure FDA0004182893810000075
Figure FDA0004182893810000076
The global performance lower bound is the lowest of all possible input quantitiesSmall value:
Figure FDA0004182893810000081
a toughness measurement subunit, configured to, based on the system performance influencing factor, use a hierarchical disturbance vector e for a multi-dimensional input parameter combination (A) ，∈ (B) ，...，∈ (F) Wherein = [ -e) 1 ，∈ 2 ，∈ 3 ，......]The disturbance degree increases outwards in sequence; e is calculated by S41 respectively (A) ，∈ (B) ，...，∈ (F) The corresponding lower bound of output efficiency up to E (α) And if the lower performance bound of the unmanned platform cooperative scheme is lower than a preset threshold, finally obtaining the toughness measurement grade of the unmanned platform cooperative scheme as alpha grade.
CN202310409814.XA 2023-04-17 2023-04-17 Unmanned platform collaboration system toughness measurement method based on neural network verification Pending CN116362413A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310409814.XA CN116362413A (en) 2023-04-17 2023-04-17 Unmanned platform collaboration system toughness measurement method based on neural network verification

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310409814.XA CN116362413A (en) 2023-04-17 2023-04-17 Unmanned platform collaboration system toughness measurement method based on neural network verification

Publications (1)

Publication Number Publication Date
CN116362413A true CN116362413A (en) 2023-06-30

Family

ID=86939231

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310409814.XA Pending CN116362413A (en) 2023-04-17 2023-04-17 Unmanned platform collaboration system toughness measurement method based on neural network verification

Country Status (1)

Country Link
CN (1) CN116362413A (en)

Similar Documents

Publication Publication Date Title
CN108095716B (en) Electrocardiosignal detection method based on confidence rule base and deep neural network
CN111062508B (en) Method for evaluating real-time running state of wind turbine generator based on big data technology
KR102238248B1 (en) Battery diagnostic methods using machine learning
Zhao et al. A fuzzy system for concrete bridge damage diagnosis
Lu et al. Integrating dynamic Bayesian network and physics-based modeling for risk analysis of a time-dependent power distribution system during hurricanes
Lahdelma et al. Stochastic multicriteria acceptability analysis (SMAA)
Alban et al. Neural networks to predict dropout at the universities
Chen Multiple criteria choice modeling using the grounds of T‐spherical fuzzy REGIME analysis
CN108830407B (en) Sensor distribution optimization method in structure health monitoring under multi-working condition
CN116362413A (en) Unmanned platform collaboration system toughness measurement method based on neural network verification
Zhao et al. Bayesian network based imprecise probability estimation method for wind power ramp events
Patton et al. A genetic algorithm approach to focused software usage testing
Owolabi et al. Development of priority index assessment model for road pavements in Nigeria
Gata et al. The Feasibility of Credit Using C4. 5 Algorithm Based on Particle Swarm Optimization Prediction
Minghua et al. Rough Set Neural Network Evaluation Method for UAV Reconnaissance Effectiveness
Pianosi et al. ANN-based representation of parametric and residual uncertainty of models
Green et al. Modeling wildfire using evolutionary cellular automata
CN116151128B (en) Equipment system contribution rate assessment method
Morfidis et al. Approach to prediction of R/C buildings' seismic damage as pattern recognition problem using artificial neural networks
CN115374392B (en) Method for acquiring object parameters of target event, electronic equipment and storage medium
CN117332923B (en) Weighting method and system for netlike index system
CN115618743B (en) State evaluation method and state evaluation system of sighting telescope system
Santamaria et al. Impact of unstructured factors on concrete through fuzzy models
CN117650949B (en) Network attack interception method and system based on RPA robot data analysis
Liang et al. Modeling of Neuro-Fuzzy System with Optimization Algorithm as a Support in System Boundary Capability Online Assessment

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