CN116822347A - Aerial action plan construction method based on dynamic influence network and discrete artificial bee colony - Google Patents

Abstract

The invention discloses an aerial action plan construction method based on a dynamic influence network and a discrete artificial bee colony, which comprises the following steps: quantitatively describing the actions of the my party, the actions of the other party, the expected effects and the intermediate effects; step two: constructing an aerial action sequence model based on the factors of the action of the my party, the action of the opposite party, the expected effect and the intermediate effect; step three: designing a dynamic influence network probability propagation mechanism according to a dynamic influence network-based action plan optimization mathematical model; step four: according to the air action sequence model and a dynamic influence network probability propagation mechanism, designing an air action sequence of the swarm unmanned aerial vehicle; the method constructs the overhead action planning problem by adopting causality modeling and optimization problem modeling, solves the model by adopting an intelligent optimization method, introduces a high-efficiency encoding and decoding technology in the method, effectively improves the searching efficiency, and has the characteristics of strong feasibility and robustness.

Description

Aerial action plan construction method based on dynamic influence network and discrete artificial bee colony

Technical Field

The invention relates to the technical field of air coordination, in particular to an air action plan construction method based on a dynamic influence network and discrete artificial bee colony.

Background

At present, the problems of multiple air action participation forces, various action patterns, changeable action environments, and difficult large-scale air action planning under a complex action background are not solved effectively for a long time; in the air action planning process, the influence of the actions of the opposite party on the overall action situation is considered except that the action strategies and the action effects generated by the action strategies are considered; in addition, if the action effect of the actions of the my party and the actions of the other party cannot be accurately measured, the credibility of the aerial action plan is greatly affected;

the aerial action plan is preferably to consider complex time sequence and logic relation between different actions for realizing specific action targets, and efficiently generate an action sequence which enables the final action effect to be optimal by combining action resource constraint; in the process, the complexity of diversity, hierarchy, coupling and the like in the air action causality model needs to be fully considered, and a perfect plan is formulated for the air action by combining the characteristics of evolution of the action stage and backward propagation of the action effect;

in the air action plan optimization method, due to the existence of high-intensity action countermeasures, the availability of the original my action plan can be reduced due to the occurrence of a large amount of unknown opponent actions, and the traditional (class) classical method and decision theory method are not applicable any more, so that the following main optimization methods exist at present:

(1) The hierarchical task network method comprises the following steps: the method can more efficiently acquire and utilize domain knowledge, improves action plan generation efficiency, and is suitable for workflow modeling, task decomposition and other types of problems;

(2) Based on a case reasoning method: according to the method, multiplexing and correcting are carried out in the new air action plan optimizing process according to the previous air action similar cases, so that an air action plan is rapidly generated;

(3) The comprehensive evaluation method comprises the following steps: according to a limited number of selectable air action plan sets, an index system for evaluating the advantages and disadvantages of different air action plans is constructed, the utility value of the air action plan is calculated on the basis of index weighting, and the air action plan with the highest utility value is selected;

in the methods, the hierarchical task network method only considers actions per se, so that the causal relationship between different actions and effects thereof is difficult to measure, and preference information of decision makers is difficult to embody in the planning process; the method has higher requirements on the completeness of the case library based on the case reasoning method, and has lower case retrieval efficiency, and the retrieval time of the method increases exponentially along with the increase of the scale of the case library; the comprehensive evaluation method does not generate an optional aerial action plan, but selects based on a certain rule in the existing plan, so that the applicable scene is limited;

Currently, patents related to aerial action plan preference focus primarily on action plan generation, adaptive analysis, plan representation, and plan rationality interpretation:

in the patent literature of Nanjing university applied for it, a method for generating a combined combat plan (publication number: CN114741861A, application number: CN 202210339300.7), a combined hitting plan generation model comprising data binding, constraint condition setting and objective function construction is proposed, and a CP-SAT solver is adopted for constraint planning solving. Finally, reasoning is carried out on the variable value and the optimized value generated by solving, so that the problem of low searching efficiency caused by complex coding and decoding mechanisms in the current method is effectively solved; of course, the method has higher requirements on the constructed model, and the number and the types of constraint conditions can directly influence the solving effect;

in the patent literature "fight plan adaptability analysis method based on simulation experiment and storage medium" filed by Beijing-ing, such as science and technology, inc. (publication No. CN108647414B, publication No. CN 201810391433.2), a method for analyzing action plan adaptability by means of simulation experiment is proposed, the method establishes simulation design and generates desired elements according to fight scene design, fight both sides action plan and battlefield event set, and under the premise of extracting experimental factors based on the desired elements, large sample exploratory simulation experiment considering system internal interactivity and external antagonism is carried out, and the action plan is continuously adjusted and optimized depending on simulation experiment results, thereby effectively avoiding the influence of uncertainty factors on the action plan. The method is essentially an evaluation method of the action plan, and the exact action plan of the opposite party is difficult to obtain in actual application, so that the applicability of the method is affected to a certain extent;

In the patent literature, "formation fight plan representation method and system based on event ontology" (publication No. CN109670672a, application No. CN 201811371803.2), it is proposed to represent formation action plans by event ontology technology: firstly, acquiring an offshore battlefield event and forming an event sequence, and characterizing the event based on an event body; subsequently, categorizing the characterized events, thereby categorizing the offshore battlefield event into a plurality of event classes; finally, based on analysis and reasoning of event class relations, the fusion of the marine battlefield event ontology is realized. The method can effectively promote the information transfer rate among all action elements and ensure that the combat cycle is completed rapidly. The method can effectively characterize the formation action plan, but also does not relate to a specific generation process of the action plan;

in summary, most of the existing action planning optimization methods have certain application limitations, and the characteristics of high air action antagonism, dynamic property and timeliness are further combined, so that the requirements of air action planning optimization reliability and robustness are met;

based on the above, a new air operation plan construction method is needed to be designed, so as to solve the problems that in the prior art, causal relationship in the existing military task plan modeling optimization field is various, hierarchical, coupling and dynamic, uncertainty exists in optimization problem model parameters, and an operation plan with high reliability and robustness cannot be generated well.

Disclosure of Invention

Aiming at the problems, the invention aims to provide an air action plan construction method based on a dynamic influence network and discrete artificial bee colony.

In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:

an aerial action plan construction method based on dynamic influence network and discrete artificial bee colony comprises

Step one: quantitatively describing the actions of the my party, the actions of the other party, the expected effects and the intermediate effects;

step two: constructing an aerial action sequence model based on the factors of the My action, the other party action, the expected effect and the intermediate effect on the basis of quantitative description of the My action, the other party action, the expected effect and the intermediate effect;

step three: designing a dynamic influence network probability propagation mechanism according to a dynamic influence network-based action plan optimization mathematical model;

step four: and designing the air action sequence of the swarm unmanned aerial vehicle according to the air action sequence model and the dynamic influence network probability propagation mechanism.

Preferably, the quantitative description of the my actions in the first step includes:

step111 set the my action set asDividing the action phase into T+1 and starting T ₀ ～t ₁ The external environment of the phase is CE (t ₀ ) Start t ₀ ～t ₁ No action is taken at stage;

wherein ,|S_A The I is the number of actions of the swarm unmanned aerial vehicle;

Step112.P _k-1 (A _i )＝1,1≤i≤|S _A i denotes at t _k-1 ～t _k Stage (2.ltoreq.k.ltoreq.T+1) takes action A _i ，P _k-1 (A _i )＝0,1≤i≤|S _A The I indicates that no action A will be taken _i The method comprises the steps of carrying out a first treatment on the surface of the If the resource and rule constraints are not considered, action A _i The policy combination at all t+1 phases is 2 ^T A plurality of; due to the existence of corresponding constraints, A _i The number of viable strategies at each stage is far from 2 ^T A plurality of;

step113 set up combat action A _i Action policy set at all T+1 phases isIn the T+1 combat phases, all combat actions constitute a feasible movement space +.>

wherein ,ψ_i ＜2 ^T ，Any element->From sequences of length TIs composed of->

Preferably, the process of constructing the air action sequence model based on the My action, the other party action, the expected effect and the intermediate effect influence factors in the step II comprises

Step21. Expressing causal relationship influence intensities among the my actions, the opponent actions, the expected effects and the intermediate effects by using CAST parameters, and establishing a causal relationship static model based on an influence network according to probability propagation from a root node to a leaf node;

Step22. Introducing the expected effect and the intermediate effect to have backward influence relation based on static modeling, and establishing an action plan optimal mathematical model based on a dynamic influence network.

Preferably, the establishing process of the influence network-based causal relationship static model in Step21 comprises

Step211. Let the causal relationship static model based on the influence network be four-element in= { V, E, CAST, BP },

wherein v= { a, B, C, D }, represents influencing the node set in the network, the values have binarycity, a represents the quantitative description of the actions of my party, B represents the quantitative description of the actions of the other party, C represents the quantitative description of the intermediate effects of the actions, D represents the quantitative description of the desired effects of the actions;

e= { (a, C), (B, C), (C, D) } represents a causal relationship between nodes;

CAST representation pairInfluence causal relationship in the net affects the set of intensities, for directed edges (a, C), the impact intensity is CAST _(A,C) E { (h, g) |h is more than or equal to-1, g is less than or equal to 1}; wherein h represents the influence degree of the parent node of 1 on the child node of 1, g represents the influence degree of the parent node of 0 on the child node of 1; if the relation is the promotion relation, the corresponding directed edge E E is provided with an arrow; if the two types of the inhibition relation are adopted, the corresponding directed edge E E is provided with a round head;

BP represents the prior probability or reference probability affecting the value of each node in the network;

step212 if event C _n Subject action setInfluence, set->Is of dimension |S _A Binary random vector of I, +.>The ith component in (a) takes on the value x _i If action A _i And then x is _i =1; conversely, x _i =0; defining event A from a qualitative perspective _i For event C _n The influence value h (A _i )：

(1) Given A _i ，C _n The conditional probability of occurrence is P (C _n |A _i ) The impact value is defined from a quantitative point of view:

wherein ,P(C_n ) For event C _n A reference probability of occurrence;

(2) Let h (A) _i )∈[-1,1]Expanding P (C) by linear interpolation _n |A _i ) Is given A _i ，C _n Conditional probability of occurrence P (C _n |A _i ) The definition is as follows:

(3) In the formulae (1) to (3), h (A) _i ) The value of (1) is related to the CAST value, i.e. if x _i ＝1，h(A _i )＝h _i The method comprises the steps of carrying out a first treatment on the surface of the If x _i ＝0，h(A _i )＝g _i 。

Preferably, the process of creating the dynamic impact net-based action plan preference mathematical model described in Step22 includes:

step221 introducing self-loop mechanism in calculation of causal influence intensity in influence network, expected effect and intermediate effect have backward influence relation, parametersStage t of representation _k-1 Stage t _k The probability that the corresponding node of the transfer takes on a value of 1 is +.>Establishment;

step222. Taking action resources as constraint conditions, taking the probability of achieving the expected effect under a specific action sequence as an optimization target, and establishing an action plan optimal mathematical model based on a dynamic influence network:

wherein ,R(t_k ) At t _k-1 ～t _k Stage combat resource consumption, R ₀ Is a combat resource threshold.

Preferably, the design process of the dynamic influence network probability propagation mechanism in the third step includes step31. According to the key parameter values in the mathematical model of the action plan optimization of the dynamic influence network given by the expert, the expert opinions of the same key parameter are sequenced, kendall synergetic coefficient inspection amount is obtained according to the sequencing result, and compared with the judgment threshold value under the corresponding significance level, so as to determine the expert group with the strongest consistency, and the opinions of the expert group are clustered to obtain the key parameter values subjected to consistency inspection;

step32, according to the probability change condition of the parent node, the child nodes are updated in sequence from top to bottom, the probability is generated according to a probability propagation algorithm, and the design of a probability propagation mechanism of the dynamic influence network is completed.

Preferably, the key parameter value process of the consistency check described in Step31 comprises

Step311. Set expert set as|S _Z Expert number, expert Z _o ∈S _Z (1≤o≤|S _Z I) to give an influence intensity h _o (Qh _o ,Ch _o) and g_o (Qg _o ,Cg _o )，

wherein ,Qh_o and Qg_o For expert Z _o Authority of Qh _o ＝Qg _o ，Ch _o and Cg_o Expert z respectively _o The h and g values given;

According to expert Z _o Giving a H-value vector H _o ＝(Ch _o,1 ,Ch _o,2 ,…,Ch _o,|V| ) The corresponding ascending number vector of the construction is:

R _o ＝(r _o,1 ,r _o,2 ,…,r _o,|V| ) (5)

wherein ,r_o,v (V is not less than 1 and not more than |V|) is Ch _o,v At H _o A sequence number in ascending order;

step312 establish hypothesis J ₀ Set S _Z The middle expert disagrees about influencing the intensity assignment; alternative hypothesis J ₁ Set S _Z The middle expert has the same opinion about the influence intensity assignment; let significance level α=0.05; according to equation (6), calculate expert set S _Z Kendall co-ordination coefficient check quantity Kendall (S) _Z )：

If Kendall (S) _Z )＜K _α Then consider to be hypothesis J ₀ Establishment; otherwise, consider hypothesis J ₁ Establishment;

step312 the consistency degree eta of the expert opinion of the collection Z is obtained by consistency test of the expert opinion _Z ：

Step313 for consistency expert opinion aggregation, a group of expert groups S needs to be found _Z' So that S _Z' The middle experts agree and have the highest expert group authority, with:

wherein the first constraint represents S _Z' The opinion of the middle specialist must be consistent;is S _Z' The expert group authority of (2) is calculated as follows:

the process of solving formula (8) is as follows:

(1) Initializing expert setsLet count flag count=1;

(2) Judgment expert set S _Z Opinion agreement of (2)Whether or not equal to 1, if not, S _Z The expert with minimum similarity of the medium opinion is moved to the collection S _Y Looping execution until ++ >Or S _Z Only one expert remains, and the expert opinion similarity calculation method is shown in the formula (10):

step314. ConvergenceAnd count=count+1, and reconstruct S _Z ＝S _Y The method comprises the steps of carrying out a first treatment on the surface of the Comparison->… expert group authority and let ∈ ->

Calculating by adopting a formula (11) to obtain an influence intensity h value result fused with multiple expert knowledge;

step315 similarly, the impact strength g value, the prior probability and the reference probability fused with multiple expert knowledge can be obtained through steps Step311-Step315 described above.

Preferably, step32 describes a probability generation process based on a probability propagation algorithm comprising

Step321, for a specific stage, dividing nodes into different layers according to the output and input conditions of all nodes in the current influence network;

step322, judging whether to enter the next stage, if yes, updating the hierarchy where all nodes are located and the prior probability of the root node;

step323 calculating the conditional probability of the child node based on the influence intensity values

Let A _i The influence intensity of (2) takes a value of h _i All influence intensities are combined into setPolymerization of positively influenced intensities to form M _pos ：

Polymerization of negative influence intensity is carried out to generate M _neg ：

Will M _pos and M_neg Merging to generate the overall influence intensity M:

calculating the corresponding conditional probability

According to the full probability formula, calculating the realization probability P (C _n )：

Step324 probability updating all hierarchical nodes according to the steps Step321-Step323, and calculating objective function value P { D after all stages are completed _m (t _T+1 )}。

Preferably, the design process of the aerial action sequence of the bee colony unmanned aerial vehicle in the step four comprises

Step41, coding the bee colony by adopting a real number coding mode, and determining the adopted basic strategy number according to the integer information of individual elements to decode;

step42, designing a fitness function;

step43, adopting a discrete artificial bee colony algorithm comprising an initialization stage, an employment stage, an observation stage and a reconnaissance stage to perform iterative optimization, and jumping out of the local optimization through a periodic checking mechanism to finally obtain an aerial action plan with high reliability and robustness.

Preferably, the iterative optimization process using the discrete artificial bee colony algorithm in Step43 comprises

Step Step431, setting the number of initialized honey sources, the number of employed bees and the number of observed bees to N respectively _fs 、N _eb 、N _ob Satisfy N _fs ＝N _eb ＝N _ob An unevended counter three with an initial value of 0 corresponding to each honey source; the initial honey source matrix X is randomly generated in a feasible interval, and the row number of the matrix is the number N of the honey sources _fs The column number of the matrix is the optional action number |S _A I, as shown in formula (18), is the m-th row and s-th column element X in matrix X _ms Is a random generation formula of:

wherein , and />Respectively represent the upper and lower bounds, rand of the s-th column variable in the matrix X ₁ A random number having a value within a (0, 1) interval;

step432 after the initial honey source matrix is generated, hiring bees to find better honey sources nearby the honey sources according to the honey source positions in the memory, and updating the formula as follows:

x _ms '＝x _ms +rand ₂ ×(x _ms -x _m's ) (19)

wherein ,x_ms ' is an element of newly generated honey source, x _m's For any other honey source corresponding position element, rand ₂ Is a random number with a value in the interval (0, 1); if x _ms ' beyondIs x _ms ' is the nearest boundary value, and there are:

calculating the fitness value of the new honey source, and if the fitness value is superior to the old honey source, replacing; if the current value is inferior to the old honey source, keeping the old honey source unchanged, and adding 1 to the real value of the non-evolution iteration algebraic counter of the honey source;

step step433 after all employment bees complete the search, exchanging honey source position information with observation bees, and calculating following probability by the observation bees according to the corresponding honey source position information:

if p _m Random generation number rand larger than value in (0, 1) interval ₃ Then the observed bees are observed according to formula (19) Updating the honey source position; similarly, the adaptation values corresponding to the new honey source and the old honey source need to be compared, and if the adaptation values are better, the new honey source and the old honey source are replaced; if the latter is more optimal, the value of the non-evolution counter three is increased by 1;

Step Step434 when the solution quality of a honey source is in the real _limit None of the updates has evolved, and the employment bee corresponding to the honey source is changed into a scout bee, the original honey source is abandoned, the new honey source position is randomly generated again according to the formula (18), and the corresponding trial value is set to 0.

The beneficial effects of the invention are as follows: the invention discloses an air action plan construction method based on a dynamic influence network and a discrete artificial bee colony, which is improved compared with the prior art in that:

1. the invention designs an aerial action plan construction method based on a dynamic influence network and discrete artificial bee colony, which constructs an aerial action plan problem by adopting causality modeling and optimization problem modeling, solves a model by adopting an intelligent optimization method, introduces a high-efficiency coding and decoding technology in the method, and improves the searching efficiency; by adopting the key parameter determining method based on the group decision method, the scientific rationality of the key parameter value is improved; the dynamic modeling of the optimization of the aerial action plan is realized by adopting an influence network and an influence relation backward propagation technology, so that the Markov characteristic of the influence relation is better represented; the robustness of the air action plan is improved by adopting a signal-to-noise ratio value optimization method for realizing probability based on the expected effect; the method overcomes the defects of the existing air action plan optimization method, improves the effectiveness and the robustness of the air action plan construction, and has the advantages of strong feasibility and robustness;

2. The method adopts a dynamic modeling mechanism, aims at the backward transfer characteristics of the expected effect and the intermediate effect, fully reflects the Markov characteristic of the air action effect, adopts a group decision method to carry out consistency check in the process of determining the key parameters of the causal relationship model, can ensure the scientific and reasonable value of the key parameters of the model, and solves the problems of low reliability and consistency of the value of the key parameters of the existing causal relationship modeling;

3. the method aims at the problem of optimization of the aerial action plan containing action resource constraint, adopts a series of technical means such as coding and decoding mode determination, fitness function design, iterative optimizing mechanism design and the like, and solves various problems of insufficient robustness, low generation efficiency and the like of a plan generation result caused by the influence of uncertain factors in the existing aerial action plan.

Drawings

FIG. 1 is a diagram of an aerial action plan preferred basic framework in the present invention;

FIG. 2 is a diagram of a static model of an impact network in the present invention;

FIG. 3 is a diagram of a dynamic model of a dynamic impact network in the present invention;

FIG. 4 is a graph of an offshore island attack mission causal relationship model in the present invention;

FIG. 5 is S in the present invention ₁ A lower objective function change diagram;

FIG. 6 is S in the present invention ₂ A lower objective function change diagram;

FIG. 7 is a graph of the results of optimization under the probability of occurrence of 20 sets of opponent actions in the present invention;

fig. 8 is a graph of the result of optimization under the probability of occurrence of a specific counterpart action in the present invention.

Detailed Description

In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below with reference to the accompanying drawings and examples.

Example 1: 1-8, a method for constructing an aerial action plan based on a dynamic impact network and discrete artificial bee colony, comprising

Step one: quantitative description of My action, party action, desired Effect and intermediate Effect

Step11. Quantitatively describe the actions of the opponent for the purpose of the actions, which are determined by a combination of multiple experts:

step12. Quantitatively describe the opponent action, which is the hostile action the opponent takes against the my action, with the aim of thwarting the achievement of the my action goal:

step13. Quantitatively describe the desired effect of the action, which is the final action objective to be achieved by my, and the kinds and amounts of the desired effects are different for different tasks to be performed:

The expected effect of the air actions is integrated into wherein ,|S_D I is the number of desired effects;

step14. description of intermediate effects of actions, which are staged effects achieved to achieve the final action objective, are the links between my actions, opponent actions and desired effects:

the intermediate effects of the air actions are collected as

wherein ,|S_C I is the number of intermediate effects;

through the steps Step11-Step14, the quantitative description of the actions of the my, the actions of the other party, the expected effects and the intermediate effects is completed;

step two: constructing an air action sequence model based on the factors of the My action, the Party action, the expected effect and the intermediate effect on the basis of quantitative description of the My action, the Party action, the expected effect and the intermediate effect

Step21. Expressing causal relationship influence intensities among the my actions, the opponent actions, the expected effects and the intermediate effects by using CAST parameters, and establishing a causal relationship static model based on an influence network according to probability propagation from a root node to a leaf node;

step22, introducing a backward influence relation between the expected effect and the intermediate effect on the basis of static modeling, and establishing an action plan optimal mathematical model based on a dynamic influence network;

Step three: dynamic impact network probability propagation mechanism designed according to action plan optimization mathematical model based on dynamic impact network

Step31, ranking the expert opinions of the same key parameters according to the key parameter values in the action plan optimization mathematical model of the dynamic influence network given by the expert, obtaining Kendall synergetic coefficient test quantity according to the ranking result, comparing the Kendall synergetic coefficient test quantity with a judgment threshold value under the corresponding significance level, thereby determining the expert group with the strongest consistency, gathering the opinions of the expert group, and obtaining the key parameter values subjected to consistency test;

step32, according to the probability change condition of the parent node, updating the probability of the child node from top to bottom in sequence, and according to the probability propagation algorithm, generating the probability, and completing the design of a probability propagation mechanism of the dynamic influence network;

step four: according to the air action sequence model and the dynamic influence network probability propagation mechanism, designing the air action sequence of the swarm unmanned aerial vehicle

Step41, firstly, adopting a real number type coding mode to code the bee colony, and determining the adopted basic strategy number according to the integer information of individual elements to decode;

step42, designing a fitness function;

step43, adopting a discrete artificial bee colony algorithm comprising an initialization stage, an employment stage, an observation stage and a reconnaissance stage to perform iterative optimization, and jumping out of the local optimization through a periodic checking mechanism to finally obtain an aerial action plan with high reliability and robustness.

Preferably, the process of quantitatively describing the my actions in Step11 includes:

step111 set the my action set asAssume that the action phase can be divided into T+1 and start T ₀ ～t ₁ The external environment of the phase is CE (t ₀ ) Start t ₀ ～t ₁ No action is taken at stage;

wherein ,|S_A The I is the number of actions of the swarm unmanned aerial vehicle;

Step112.P _k-1 (A _i )＝1,1≤i≤|S _A i denotes at t _k-1 ～t _k (2≤k≤Action A is taken during T+1) stage _i And P is _k-1 (A _i )＝0,1≤i≤|S _A The I indicates that no action A will be taken _i The method comprises the steps of carrying out a first treatment on the surface of the Thus, if resource and rule constraints are not considered, action A _i The policy combination at all t+1 phases is 2 ^T A plurality of; and due to the existence of corresponding constraint, A _i The number of viable strategies at each stage is far from 2 ^T A plurality of;

step113 set up combat action A _i Action policy set at all T+1 phases isIn the T+1 combat phases, all combat actions constitute a feasible movement space +.>

wherein ,ψ_i ＜2 ^T ，Any element->From sequences of length TIs composed of->

Preferably, the process of quantitatively describing the actions of the opposite party in Step12 includes:

step121. Set the set of possible actions of the opposite party asBecause of uncertainty of the action of the opposite side, the occurrence of the action of the opposite side has certain random characteristics, but the expert can still give the occurrence probability of the action of the opposite side at a certain stage according to the historical data and self experience;

wherein ,|S_B The I is the action quantity of the other party;

step122 set at t _k-1 ～t _k Stage (2.ltoreq.k.ltoreq.T+1), external event B _l Probability of occurrence P of (2) _k-1 (B _l ) Generally, the value range isWithin which probability interval P is assumed _k-1 (B _l ) Is subject to uniform distribution, and completes quantitative description of actions of the opposite party.

Preferably, the causal relationship static model set up principle described in Step21 is as follows: the causal relationship influence intensity among the actions of the my party, the actions of the opposite party, the expected effect and the intermediate effect is expressed by using CAST parameters, and the probability of realizing the expected effect is generated through probability propagation from the root node to the leaf node, so that the construction of a causal relationship static model based on an influence network is realized:

step211. Let the influence network based causal relationship static model be characterized as four tuples in= { V, E, CAST, BP },

wherein v= { a, B, C, D }, represents a set of nodes in the impact network, the value has a binary property, i.e. the value is 0 or 1, a represents a quantitative description of my actions, B represents a quantitative description of the actions of the other party, C represents a quantitative description of intermediate effects of the actions, D represents a quantitative description of desired effects of the actions;

e= { (a, C), (B, C), (C, D) } represents a causal relationship between nodes, described by a directional edge with an arrow or rounded head;

CAST representation pairInfluence causal relationship in the net affects the set of intensities, for directed edges (a, C), the impact intensity is CAST _(A,C) E { (h, g) |h is more than or equal to-1, g is less than or equal to 1}; wherein h represents the influence degree of the parent node of 1 on the child node of 1, g represents the influence degree of the parent node of 0 on the child node of 1; if the relation is the promotion relation, the corresponding directed edge E E is provided with an arrow; if the two types of the inhibition relation are adopted, the corresponding directed edge E E is provided with a round head;

as shown in FIG. 2, P (A ₁ =1) =0.7 means for node a ₁ In other words, the value of the prior probability is 1The rate is 0.7; c (C) ₁ (0.5) represents the node C ₁ The reference probability of the value of 1 is 0.5; a is that ₁ To C ₁ CAST parameter (0.90, -0.33), representing A ₁ Is 1 to C ₁ An impact strength value of 1 of 0.90, A ₁ Is 0 to C ₁ An impact strength value of-0.33 for 1, A ₁ For C ₁ Is a promoting relationship; b (B) ₁ To C ₁ CAST parameter (-0.67,0.67), representing B ₁ Is 1 to C ₁ An impact strength value of-0.67 for 1, B ₁ Is 0 to C ₁ An impact strength value of 1 of 0.67, A ₁ For C ₁ Is a suppression relationship;

BP represents the prior probability or reference probability affecting the value of each node in the network, namely the probability that the value of the corresponding node is 1 under the influence of no external causality; wherein, the root node corresponds to the prior probability and the leaf node corresponds to the reference probability;

Step212 with event C _n Subject action setInfluence is exemplified by->Is of dimension |S _A Binary random vector of I, +.>The ith component in (a) takes on the value x _i If action A _i And then x is _i =1; conversely, x _i =0; defining event A from a qualitative perspective _i For event C _n The influence value h (A _i )：

(1) Given A _i ，C _n The conditional probability of occurrence is P (C _n |A _i ) The impact value is defined from a quantitative point of view:

wherein ,P(C_n ) For event C _n A reference probability of occurrence;

(2) Let h (A) _i )∈[-1,1]Expanding P (C) by linear interpolation _n |A _i ) Is given A _i ，C _n Conditional probability of occurrence P (C _n |A _i ) The definition is as follows:

/>

(3) In the formulae (1) to (3), h (A) _i ) The value of (1) is related to the CAST value, i.e. if x _i =1, then h (a _i )＝h _i The method comprises the steps of carrying out a first treatment on the surface of the If x _i =0, h (a _i )＝g _i And constructing a causal relationship static model based on the influence network.

Preferably, the process of establishing a dynamic impact network-based action plan preference mathematical model described in Step22 includes:

step221 introducing a self-loop mechanism in the calculation of the influence intensity of the causal relation in the influence network, namely, the expected effect and the intermediate effect realization probability of a certain stage are not only related to the action sequence of the current stage, but also influenced by the expected effect and the intermediate effect realization probability of the last stage, namely, the action sequence of the last stage;

The desired effect and the intermediate effect have a backward influence relationship, parametersStage t of representation _k-1 Stage t _k The probability that the corresponding node of the transfer takes on a value of 1, i.e. there is +.>This is true as shown in fig. 3;

step222. Thus, taking action resources as constraint conditions, taking the expected effect realization probability under a specific action sequence as an optimization target, and establishing an action plan optimal mathematical model based on a dynamic influence network as follows:

wherein ,R(t_k ) At t _k-1 ～t _k Stage combat resource consumption, R ₀ Is a combat resource threshold.

Preferably, the key parameter in Step31 is based on the following principle: according to the method, key parameter values in dynamic influence network are given out by different experts, expert opinions of the same key parameter are sequenced, kendall synergetic coefficient test quantity is obtained according to the sequencing result, and compared with a judgment threshold value under the corresponding significance level, so that an expert group with the strongest consistency is determined, the opinions of the expert group are clustered, and the key parameter values after consistency test are obtained, wherein the method comprises the following specific processes of

Step311. Set expert set as|S _Z Expert Z is the number of experts _o ∈S _Z (1≤o≤|S _Z I) to give an influence intensity h _o (Qh _o ,Ch _o) and g_o (Qg _o ,Cg _o )，

wherein ,Qh_o and Qg_o For expert Z _o Authority of (1), generally Qh _o ＝Qg _o ，Ch _o and Cg_o Expert z respectively _o The h and g values given;

taking the value of h as an example, according to expert Z _o Giving a H-value vector H _o ＝(Ch _o,1 ,Ch _o,2 ,…,Ch _o,|V| ) The corresponding ascending number vector of the construction is:

R _o ＝(r _o,1 ,r _o,2 ,…,r _o,|V| ) (5)

wherein ,r_o,v (V is not less than 1 and not more than |V|) is Ch _o,v At H _o A sequence number in ascending order;

step312 establish hypothesis J ₀ Set S _Z The middle expert disagrees about influencing the intensity assignment; alternative hypothesis J ₁ : set S _Z The middle expert has the same opinion about the influence intensity assignment; let significance level α=0.05; then according to equation (6) the expert set S is calculated _Z Kendall co-ordination coefficient check quantity Kendall (S) _Z ) The method comprises the following steps:

if Kendall (S) _Z )＜K _α Then consider to be hypothesis J ₀ Establishment; otherwise, consider hypothesis J ₁ Establishment;

step312 the consistency η of the expert opinion of the collection Z can be obtained by checking the consistency of the expert opinion _Z The method comprises the following steps:

step313 for consistency expert opinion aggregation, a group of such expert groups S needs to be found _Z' So that S _Z' The middle experts agree and have the highest expert group authority, namely:

wherein the first constraint represents S _Z' The opinion of the middle specialist must be consistent; mu (mu) _SZ' Is S _Z' The expert group authority of (2) is calculated as follows:

the process of solving formula (8) is as follows:

(1) First, an expert set is initializedLet count flag count=1;

(2) Then, judge expert set S _Z Opinion agreement of (2)Whether or not equal to 1, if not, S _Z The expert with minimum similarity of the medium opinion is moved to the collection S _Y Looping execution until ++>Or S _Z Only one expert remains, and the expert opinion similarity calculation method is shown in the formula (10):

step314. ConvergenceAnd count=count+1, and reconstruct S _Z ＝S _Y The method comprises the steps of carrying out a first treatment on the surface of the Comparison->… expert group authority and let ∈ ->Through the above process, an expert group which accords with the consistency principle and enables the authority degree to be maximum is determined, and the influence intensity h value result fused with multiple expert knowledge is obtained by adopting the formula (11);

step315, similarly, the influence strength g value, the prior probability and the reference probability which are fused with multiple expert knowledge can be obtained through calculation through the steps, and the value of the key parameter is completed.

Preferably, the principle of implementing the probability generating process according to the probability propagation algorithm in Step32 is as follows: along with the continuous progress of actions, the update of the child node realization probability is needed to be sequentially performed from top to bottom according to the realization probability change condition of the parent node, and the method specifically comprises the following steps:

step321, firstly, for a specific stage, dividing nodes into different layers according to the output and input conditions of all nodes in the current influence network;

the root node has the highest level, the middle node has the middle level and the leaf node has the lowest level;

Step322, judging whether to enter the next stage, if yes, updating the hierarchy where all nodes are located and the prior probability of the root node;

step323 again, calculating the conditional probability of the child node based on the influence intensity value to calculate the conditional probabilityFor example, A _i The influence intensity of (2) takes a value of h _i All influence intensities are combined into setPolymerization of positively influenced intensities to form M _pos ：/>

Polymerization of negative influence intensity is carried out to generate M _neg ：

Will M _pos and M_neg Merging to generate the overall influence intensity M:

calculating the corresponding conditional probability

According to the full probability formula, calculating the realization probability P (C _n )：

Step324 the probability update is performed on all hierarchical nodes according to the steps Step321-Step323, and the objective function value P { D may be calculated after all stages are completed _m (t _T+1 )}。

Preferably, the encoding and decoding process described in Step41 includes

Step411 for action A _i In the T+1 stages, the feasible movement strategy set is determined by expert, and is thatThe plurality of combat actions form an overall action strategy;

step412 using a length of |S _A Real coding scheme of i: the value ranges of the individual elements at different positions are different, for example, the value range of the ith individual element is (1, ψ) _i +1), wherein ψ _i For action A _i The number of basic strategies in the system;

step413 determining the basic strategy number according to the integer information of individual elements, and if the 3 rd individual element has a value of 4.1536, it represents A ₃ Actual use of policy 4, i.e

Preferably, the process of designing the fitness function described in Step42 includes

The probability of occurrence of the opposite action is randomly generated by adopting a Monte Carlo method, the random times are set as R, and R P { D } can be calculated _m (t _T+1 )|CE(t ₀ ),Ψ _S Mean μ and variance σ of the values ² And is based on mu and sigma ² The signal-to-noise value is calculated according to equation (17):

preferably, the iterative optimization process using the discrete artificial bee colony algorithm in Step43 includes:

the discrete artificial bee colony algorithm flow mainly comprises 4 stages, namely an initialization stage, a employment stage, an observation stage and a reconnaissance stage;

step Step431, setting the number of initialized honey sources, the number of employed bees and the number of observed bees to N respectively _fs 、N _eb 、N _ob Satisfy N _fs ＝N _eb ＝N _ob An unevended counter three with an initial value of 0 corresponding to each honey source; the initial honey source matrix X is randomly generated in a feasible interval, and the row number of the matrix is the number N of the honey sources _fs The column number of the matrix is the optional action number |S _A I, as shown in formula (18), is the m-th row and s-th column element X in matrix X _ms Is a random generation formula of:

wherein , and />Respectively represent the upper and lower bounds, rand of the s-th column variable in the matrix X ₁ A random number having a value within a (0, 1) interval;

step432 after the initial honey source matrix is generated, hiring bees to find better honey sources near the honey sources according to the honey source positions in the memory, and updating the formula:

x _ms '＝x _ms +rand ₂ ×(x _ms -x _m's ) (19)

wherein ,x_ms ' is an element of newly generated honey source, x _m's For any other honey source corresponding position element, rand ₂ Is a random number with a value in the interval (0, 1); if x _ms ' beyondIs x _ms ' as recently asBoundary values, namely:

calculating the fitness value of the new honey source, and if the fitness value is superior to the old honey source, replacing; if the current value is inferior to the old honey source, keeping the old honey source unchanged, and adding 1 to the real value of the non-evolution iteration algebraic counter of the honey source;

step step433 after all employment bees complete the search, exchanging honey source position information with observation bees, and calculating following probability by the observation bees according to the corresponding honey source position information:

if p _m Random generation number rand larger than value in (0, 1) interval ₃ Then the observing bees update the honey source position according to the formula (19); similarly, the adaptation values corresponding to the new honey source and the old honey source need to be compared, and if the adaptation values are better, the new honey source and the old honey source are replaced; if the latter is more optimal, the value of the non-evolution counter three is increased by 1;

Step Step434 when the solution quality of a honey source is in the real _limit None of the updates has evolved, and the employment bee corresponding to the honey source is changed into a scout bee, the original honey source is abandoned, the new honey source position is randomly generated again according to the formula (18), and the corresponding trial value is set to 0.

Example 2: in order to verify the effectiveness of the method for constructing an air action plan based on a dynamic impact network and a discrete artificial bee colony according to embodiment 1 of the present invention, the effect of embodiment 1 is verified by this embodiment as follows:

1. first, according to fig. 4, a typical aerial mission scenario is determined as performing an offshore island attack mission; as shown in tables 1 to 3, the my action set, the opponent action set, and the intermediate effect and desired effect set are respectively:

table 1: my action set

/>

Table 2: opponent action set

Action category	Description of action
		B 1	Implementing interception for air formation of other party
B 2	Opponent surface vessel formation implementation air interception
		B 3	Air defense action of ground defense system
B 4	The other party performs the electronic interference action
		B 5	Out-island air formation for augmentation

Table 3: intermediate effects and desired effect sets

Action category	Action Effect description
		D 1	My completion of offshore island attack tasks
D 2	Fight loss condition of my in task execution
		C 1	Attack effects in the form of air-to-air
C 2	Attack effects in the form of air-to-sea
		C 3	Effect of attack on preamble non-predetermined targets
C 4	Air staging effect before attacking a predetermined target
		C 5	Effects of attack on predetermined targets
C 6	Air gathering effect of back voyage after task

2. All simulation experiments were run in MATLAB2014a environment on hardware of 4G memory, 3.06GHz CPU, as follows: the offshore island attack task is executed and can be divided into 6 stages; t is t ₀ ～t ₁ The stage of intercepting the aerial formation of the opponent; t is t ₁ ～t ₂ Pressing the sea surface naval vessel of the other party; t is t ₂ ～t ₃ The attack on the fixed or mobile non-preset target of the preamble; t is t ₃ ～t ₄ After aerial refueling, aerial aggregation is carried out; t is t ₄ ～t ₅ Stage, attack the other side to fix or move the preset target; t is t ₅ ～t ₆ At the stage, after the amplified air formation outside the opposite island is intercepted, the back voyage is carried out; as shown in table 4, the occurrence probability of the opponent action at each stage is shown;

table 4: probability of occurrence of opponent action

As shown in table 5, to give optional action strategies for each stage of heating according to the experience of the expert in the field, taking into account the air action resources and rule constraints;

table 5: optional strategies for different actions at various stages of combat

And (3) a step of: key parameter determination verification

Number of assumed experts |s _Z With i=6, when the significance level α=0.05, the threshold of Kendall synergy coefficient in degree of freedom 24 is K via table lookup _α = 36.4151; if to theQh _o ＝Qg _o And authority vectors of 6 experts are (0.25,0.15,0.1,0.15,0.2,0.15); as shown in table 6, the expert is given a value that affects the intensity values (i.e., h-value and g-value);

table 6: the h value and the g value given by each expert take the values

Expert	h 1	h 2	h 3	h 4	h 5	h 6	h 7	h 8	h 9	h 10
											z 1	0.95	-0.98	0.04	0.06	-0.98	0.13	-0.22	-0.40	0.65	0.62
z 2	0.72	-0.22	0.66	0.53	-0.07	0.76	-0.01	-0.72	0.75	0.26
											z 3	0.63	-0.97	0.12	0.58	-0.17	0.04	-0.70	-0.94	0.21	0.40
z 4	0.31	-0.06	0.92	0.89	-0.74	0.83	-0.08	-0.40	0.48	0.85
											z 5	0.02	-0.50	0.39	0.55	-0.04	0.86	-0.06	-0.03	0.75	0.49
z 6	0.57	-0.26	0.71	0.94	-0.43	0.69	-0.06	-0.17	0.51	0.38
											Expert	h 11	h 12	h 13	h 14	h 15	h 16	h 17	h 18	h 19	h 20
z 1	0.44	0.36	0.84	-0.39	0.09	0.02	0.18	0.75	0.16	0.08
											z 2	0.87	0.07	0.86	-0.48	0.08	0.89	0.91	0.86	0.48	0.97
z 3	0.53	0.35	0.95	-0.13	0.78	0.98	0.16	0.69	0.95	0.13
											z 4	0.69	0.98	0.77	-0.07	0.35	0.08	0.09	0.74	0.26	0.70
z 5	0.41	0.74	0.96	-0.96	0.06	0.53	0.06	0.10	0.55	0.41
											z 6	0.33	0.87	0.09	-0.87	0.48	0.85	0.85	0.37	0.07	0.73
Expert	h 21	h 22	h 23	h 24	h 25	g 1	g 2	g 3	g 4	g 5
											z 1	0.09	0.05	-0.18	-0.05	0.43	-0.70	0.05	-0.07	-0.61	0.02
z 2	0.82	0.98	-0.85	-0.56	0.04	-0.91	0.31	-0.93	-0.64	0.96
											z 3	0.46	0.36	-0.92	-0.92	0.26	-0.76	0.22	-0.97	-0.86	0.96
z 4	0.92	0.01	-0.29	-0.07	0.31	-0.44	0.66	-0.48	-0.02	0.08
											z 5	0.14	0.73	-0.50	-0.97	0.56	-0.61	0.93	-0.19	-0.94	0.59
z 6	0.32	0.60	-0.41	-0.51	0.43	-0.66	0.36	-0.78	-0.95	0.55
											Expert	g 6	g 7	g 8	g 9	g 10	g 11	g 12	g 13	g 14	g 15
z 1	-0.30	0.97	0.65	0.08	0.12	0.99	0.55	0.80	0.92	0.20
											z 2	-0.66	0.30	0.35	0.94	0.98	0.11	0.12	0.02	0.11	0.58
z 3	-0.82	0.65	0.04	0.53	0.91	0.10	0.83	0.58	0.17	0.62
											z 4	-0.61	0.93	0.85	0.91	0.45	0.97	0.99	0.76	0.97	0.17
z 5	-0.65	0.23	0.04	0.36	0.64	0.18	0.95	0.01	0.48	0.51
											z 6	-0.11	0.36	0.39	0.91	0.65	0.13	0.33	0.67	0.06	0.80
Expert	g 16	g 17	g 18	g 19	g 20	g 21	g 22	g 23	g 24	g 25
											z 1	0.17	0.21	0.09	0.10	0.07	0.38	0.05	0.08	0.75	0.10
z 2	0.32	0.35	0.94	0.74	0.25	0.52	0.75	0.09	0.93	0.85
											z 3	0.47	0.84	0.99	0.78	0.06	0.98	0.37	0.95	0.19	0.01
z 4	0.82	0.82	0.58	0.81	0.30	0.96	0.26	0.80	0.99	0.76
											z 5	0.03	0.09	0.89	0.23	0.60	0.64	0.78	0.98	0.95	0.31
z 6	0.90	0.78	0.93	0.36	0.84	0.92	0.34	0.94	0.27	0.62

The method is available by adopting Kendall synergetic coefficient test method, and when the consistency test is carried out on the h value to generate, the expert group with the highest authority is { Z } ₂ ,Z ₄ ,Z ₅ ,Z ₆ The final h values generated were 0.38, -0.28, 0.65, 0.71, -0.30, 0.79, -0.05, -0.31, 0.63, 0.50, 0.56, 0.67, 0.69, -0.62, 0.23, 0.58, 0.44, 0.49, 0.35, 0.68, 0.52, 0.59, -0.51, -0.56, 0.35, respectively;

when the consistency check is performed on the g value, the expert group with the highest authority is { Z } ₂ ,Z ₃ ,Z ₅ ,Z ₆ The final g values generated were-0.72, 0.51, -0.66, -0.85, 0.73, -0.55, 0.35, 0.20, -0.67, -0.77, -0.14, -0.57, -0.27, 0.23, -0.62, -0.40, -0.45, -0.93, -0.48, -0.74, -0.59, 0.74, 0.65, -0.47, respectively;

and II: action plan generation validation

Make the other party act B _l The occurrence probability isSolving the model by adopting a discrete artificial bee colony algorithm; generating action plan at random->And an optimal action plan generated by using a discrete artificial bee colony algorithm +.>Next, for F ₁ (S)＝P{D ₁ (t _T+1 )＝1|CE(t ₀ ),Ψ _S }、F ₂ (S)＝P{D ₂ (t _T+1 )＝1|CE(t ₀ ),Ψ _S} and F₃ (S)＝P{D ₁ (t _T+1 )＝1,D ₂ (t _T+1 )＝0|CE(t ₀ ),Ψ _S Analyzing the change condition of the three kinds of objective functions; as shown in FIG. 5 and FIG. 6, S is respectively ₁ and S₂ The change condition of the objective function in different combat stages is carried out;

as can be seen from FIGS. 5 and 6, F ₃ (S ₁ )＝0.1340，F ₃ (S ₂ ) = 0.3416; thus, action plan S ₂ The effect is better than S ₁ The method comprises the steps of carrying out a first treatment on the surface of the Through the simulation, the effectiveness of the discrete artificial bee colony algorithm in solving the optimization problem of the aerial action plan is proved;

thirdly,: action plan generation superiority verification

Firstly, taking the probability of occurrence of the opposite side action as a random value in a certain interval, determining the probability of occurrence of the opposite side action of H groups by an H-time Monte Carlo method, and respectively running F for G times by a discrete artificial bee colony algorithm, a discrete firefly algorithm and a discrete particle swarm algorithm under each group of probabilities ₃ The (S) value is taken as a fitness function, h=g=20;

then, taking the occurrence probability of the opposite side action as a certain value, respectively adopting a discrete artificial bee colony algorithm, a discrete firefly algorithm and a discrete particle swarm algorithm to run G times, and comparing the distribution condition of the solving results of the algorithms to obtain G=100; as shown in fig. 7, the fitness value of each comparison algorithm under the occurrence probability of 20 pairs of actions is shown; as shown in fig. 8, the distribution of the solving results of multiple experiments under the occurrence probability of the action of a specific counterpart is shown;

As can be seen from FIG. 7, compared with the discrete firefly algorithm and the discrete particle swarm algorithm, the discrete artificial bee colony algorithm has the advantages that the optimal solution can be obtained 17 times, and the effect is better; as can be seen from the figure 8, the solution optimality and stability of the discrete artificial bee colony algorithm are better, and the bad value is less;

therefore, it can be seen that in the embodiment 1 of the present invention, causality modeling and optimization problem modeling are adopted to construct an overhead action planning problem, and an intelligent optimization method is adopted to solve a model, and a high-efficiency coding and decoding technology is introduced in the method, so that the search efficiency is improved; the key parameter determination method based on the group decision method is adopted, so that the scientific rationality of the key parameter value is improved; the influence network and the influence relation back propagation technology are adopted to realize the preferable dynamic modeling of the aerial action plan, and the Markov characteristic of the influence relation is better represented; the signal-to-noise ratio value optimization method based on the expected effect probability is adopted, so that the robustness of the air action plan is improved; the scheme of the embodiment 1 of the invention overcomes the defects of the prior air action plan optimization method and improves the effectiveness and the robustness of air action plan construction.

The foregoing has shown and described the basic principles, principal features and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (10)

1. The aerial action plan construction method based on the dynamic influence network and the discrete artificial bee colony is characterized by comprising the following steps of: comprising

Step one: quantitatively describing the actions of the my party, the actions of the other party, the expected effects and the intermediate effects;

step two: constructing an aerial action sequence model based on the factors of the My action, the other party action, the expected effect and the intermediate effect on the basis of quantitative description of the My action, the other party action, the expected effect and the intermediate effect;

step three: designing a dynamic influence network probability propagation mechanism according to a dynamic influence network-based action plan optimization mathematical model;

step four: and designing the air action sequence of the swarm unmanned aerial vehicle according to the air action sequence model and the dynamic influence network probability propagation mechanism.

2. The method for constructing an air action plan based on a dynamic influence network and discrete artificial bee colony as claimed in claim 1, wherein: the process of quantitatively describing the my actions in the first step comprises the following steps:

step111 set the my action set asWill actThe phase is divided into T+1, and T is started ₀ ～t ₁ The external environment of the phase is CE (t ₀ ) Start t ₀ ～t ₁ No action is taken at stage;

wherein ,|S_A The I is the number of actions of the swarm unmanned aerial vehicle;

Step112.P _k-1 (A _i )＝1,1≤i≤|S _A I denotes at t _k-1 ～t _k Stage (2.ltoreq.k.ltoreq.T+1) takes action A _i ，P _k-1 (A _i )＝0,1≤i≤|S _A The I indicates that no action A will be taken _i The method comprises the steps of carrying out a first treatment on the surface of the If the resource and rule constraints are not considered, action A _i The policy combination at all t+1 phases is 2 ^T A plurality of; due to the existence of corresponding constraints, A _i The number of viable strategies at each stage is far from 2 ^T A plurality of;

step113 set up combat action A _i Action policy set at all T+1 phases isIn the T+1 combat phases, all combat actions constitute a feasible movement space +.>

wherein ,ψ_i ＜2 ^T ，Any element->From the length T sequence->Is composed of->

3. The method for constructing an air action plan based on a dynamic influence network and discrete artificial bee colony as claimed in claim 1, wherein: the process of constructing the aerial action sequence model based on the factors of the My action, the opposite action, the expected effect and the intermediate effect comprises

Step21. Expressing causal relationship influence intensities among the my actions, the opponent actions, the expected effects and the intermediate effects by using CAST parameters, and establishing a causal relationship static model based on an influence network according to probability propagation from a root node to a leaf node;

step22. Introducing the expected effect and the intermediate effect to have backward influence relation based on static modeling, and establishing an action plan optimal mathematical model based on a dynamic influence network.

4. A method of constructing an airborne action plan based on a dynamic impact network and discrete artificial bee colony as claimed in claim 3, wherein: the process for establishing the influence network-based causal relationship static model in Step21 comprises

Step211. Let the causal relationship static model based on the influence network be four-element in= { V, E, CAST, BP },

wherein v= { a, B, C, D }, represents influencing the node set in the network, the values have binarycity, a represents the quantitative description of the actions of my party, B represents the quantitative description of the actions of the other party, C represents the quantitative description of the intermediate effects of the actions, D represents the quantitative description of the desired effects of the actions;

e= { (a, C), (B, C), (C, D) } represents a causal relationship between nodes;

CAST representation pairInfluence causal relationship in the net affects the set of intensities, for directed edges (a, C), the impact intensity is CAST _(A,C) E { (h, g) |h is more than or equal to-1, g is less than or equal to 1}; wherein h represents the influence degree of the parent node of 1 on the child node of 1, g represents the influence degree of the parent node of 0 on the child node of 1; if the relation is the promotion relation, the corresponding directed edge E E is provided with an arrow; if the two types of the inhibition relation are adopted, the corresponding directed edge E E is provided with a round head;

BP represents the prior probability or reference probability affecting the value of each node in the network;

Step212 if event C _n Subject action setInfluence, set->Is of dimension |S _A Binary random vector of I, +.>The ith component in (a) takes on the value x _i If action A _i And then x is _i =1; conversely, x _i =0; defining event A from a qualitative perspective _i For event C _n The influence value h (A _i )：

(1) Given A _i ，C _n The conditional probability of occurrence is P (C _n |A _i ) The impact value is defined from a quantitative point of view:

wherein ,P(C_n ) For event C _n A reference probability of occurrence;

(2) Let h (A) _i )∈[-1,1]Expanding P (C) by linear interpolation _n |A _i ) Is given A _i ，C _n Conditional probability of occurrence P (C _n |A _i ) The definition is as follows:

(3) In the formulae (1) to (3), h (A) _i ) The value of (1) is related to the CAST value, i.e. if x _i ＝1，h(A _i )＝h _i The method comprises the steps of carrying out a first treatment on the surface of the If x _i ＝0，h(A _i )＝g _i 。

5. A method of constructing an airborne action plan based on a dynamic impact network and discrete artificial bee colony as claimed in claim 3, wherein: the process of establishing the dynamic impact net-based action plan preference mathematical model described in Step22 comprises:

step221 introducing self-loop mechanism in calculation of causal influence intensity in influence network, expected effect and intermediate effect have backward influence relation, parametersStage t of representation _k-1 Stage t _k The probability that the corresponding node of the transfer takes on a value of 1 is +.>Establishment;

step222. Taking action resources as constraint conditions, taking the probability of achieving the expected effect under a specific action sequence as an optimization target, and establishing an action plan optimal mathematical model based on a dynamic influence network:

wherein ,R(t_k ) At t _k-1 ～t _k Stage combat resource consumption, R ₀ Is a combat resource threshold.

6. The method for constructing an air action plan based on a dynamic influence network and discrete artificial bee colony as claimed in claim 1, wherein: the design process of the dynamic influence network probability propagation mechanism in the third step comprises

Step31, ranking the expert opinions of the same key parameters according to the key parameter values in the action plan optimization mathematical model of the dynamic influence network given by the expert, obtaining Kendall synergetic coefficient test quantity according to the ranking result, comparing the Kendall synergetic coefficient test quantity with a judgment threshold value under the corresponding significance level, thereby determining the expert group with the strongest consistency, gathering the opinions of the expert group, and obtaining the key parameter values subjected to consistency test;

step32, according to the probability change condition of the parent node, the child nodes are updated in sequence from top to bottom, the probability is generated according to a probability propagation algorithm, and the design of a probability propagation mechanism of the dynamic influence network is completed.

7. The method for constructing an air action plan based on a dynamic influence network and discrete artificial bee colony as claimed in claim 6, wherein: the key parameter value process of the consistency check in Step31 comprises

Step311. Set expert set as|S _Z Expert number, expert Z _o ∈S _Z (1≤o≤|S _Z I) to give an influence intensity h _o (Qh _o ,Ch _o) and g_o (Qg _o ,Cg _o )，

wherein ,Qh_o and Qg_o For expert Z _o Authority of Qh _o ＝Qg _o ，Ch _o and Cg_o Expert z respectively _o The h and g values given;

according to expert Z _o Giving a vector of h valuesThe corresponding ascending number vector of the construction is:

R _o ＝(r _o,1 ,r _o,2 ,…,r _o,|V| ) (5)

wherein ,r_o,v (V is not less than 1 and not more than |V|) is Ch _o,v At H _o A sequence number in ascending order;

step312 establish hypothesis J ₀ Set S _Z The middle expert disagrees about influencing the intensity assignment; alternative hypothesis J ₁ Set S _Z The middle expert has the same opinion about the influence intensity assignment; make the significanceLevel α=0.05; according to equation (6), calculate expert set S _Z Kendall co-ordination coefficient check quantity Kendall (S) _Z )：

If Kendall (S) _Z )＜K _α Then consider to be hypothesis J ₀ Establishment; otherwise, consider hypothesis J ₁ Establishment;

step312 the consistency degree eta of the expert opinion of the collection Z is obtained by consistency test of the expert opinion _Z ：

Step313 for consistency expert opinion aggregation, a group of expert groups S needs to be found _Z' So that S _Z' The middle experts agree and have the highest expert group authority, with:

wherein the first constraint represents S _Z' The opinion of the middle specialist must be consistent;is S _Z' The expert group authority of (2) is calculated as follows:

the process of solving formula (8) is as follows:

(1) Initializing expert setsLet count flag count=1;

(2) Judgment expert set S _Z Opinion of (2)Consistency degreeWhether or not equal to 1, if not, S _Z The expert with minimum similarity of the medium opinion is moved to the collection S _Y Looping execution until ++>Or S _Z Only one expert remains, and the expert opinion similarity calculation method is shown in the formula (10):

step314. ConvergenceAnd count=count+1, and reconstruct S _Z ＝S _Y The method comprises the steps of carrying out a first treatment on the surface of the Comparison->Expert group authority of (2) and let +.>

Calculating by adopting a formula (11) to obtain an influence intensity h value result fused with multiple expert knowledge;

step315 similarly, the impact strength g value, the prior probability and the reference probability fused with multiple expert knowledge can be obtained through steps Step311-Step315 described above.

8. The method for constructing an air action plan based on a dynamic influence network and discrete artificial bee colony as claimed in claim 6, wherein: the probability generation process according to Step32 includes

Step321, for a specific stage, dividing nodes into different layers according to the output and input conditions of all nodes in the current influence network;

step322, judging whether to enter the next stage, if yes, updating the hierarchy where all nodes are located and the prior probability of the root node;

Step323 calculating the conditional probability of the child node based on the influence intensity valuesLet A _i The influence intensity of (2) takes a value of h _i All influence intensities are combined to be +.>Polymerization of positively influenced intensities to form M _pos ：

Polymerization of negative influence intensity is carried out to generate M _neg ：

Will M _pos and M_neg Merging to generate the overall influence intensity M:

calculating the corresponding conditional probability

According to the full probability formula, calculating the realization probability P (C _n )：

Step324 probability updating all hierarchical nodes according to the steps Step321-Step323, and calculating objective function value P { D after all stages are completed _m (t _T+1 )}。

9. The method for constructing an air action plan based on a dynamic influence network and discrete artificial bee colony as claimed in claim 1, wherein: the design process of the air action sequence of the bee colony unmanned aerial vehicle comprises the following steps of

Step41, coding the bee colony by adopting a real number coding mode, and determining the adopted basic strategy number according to the integer information of individual elements to decode;

step42, designing a fitness function;

step43, adopting a discrete artificial bee colony algorithm comprising an initialization stage, an employment stage, an observation stage and a reconnaissance stage to perform iterative optimization, and jumping out of the local optimization through a periodic checking mechanism to finally obtain an aerial action plan with high reliability and robustness.

10. The method for constructing an air action plan based on a dynamic influence network and discrete artificial bee colony as claimed in claim 9, wherein: the iterative optimization process using the discrete artificial bee colony algorithm in Step43 comprises

Step Step431, setting the number of initialized honey sources, the number of employed bees and the number of observed bees to N respectively _fs 、N _eb 、N _ob Satisfy N _fs ＝N _eb ＝N _ob An unevended counter three with an initial value of 0 corresponding to each honey source; the initial honey source matrix X is randomly generated in a feasible interval, and the row number of the matrix is the number N of the honey sources _fs The column number of the matrix is the optional action number |S _A I, as shown in formula (18), is the m-th row and s-th column element X in matrix X _ms Is a random generation formula of:

wherein , and />Respectively represent the upper and lower bounds, rand of the s-th column variable in the matrix X ₁ A random number having a value within a (0, 1) interval;

step432 after the initial honey source matrix is generated, hiring bees to find better honey sources nearby the honey sources according to the honey source positions in the memory, and updating the formula as follows:

x _ms '＝x _ms +rand ₂ ×(x _ms -x _m's ) (19)

wherein ,x_ms ' is an element of newly generated honey source, x _m's For any other honey source corresponding position element, rand ₂ Is a random number with a value in the interval (0, 1); if x _ms ' beyondIs x _ms ' is the nearest boundary value, and there are:

Calculating the fitness value of the new honey source, and if the fitness value is superior to the old honey source, replacing; if the current value is inferior to the old honey source, keeping the old honey source unchanged, and adding 1 to the real value of the non-evolution iteration algebraic counter of the honey source;

step step433 after all employment bees complete the search, exchanging honey source position information with observation bees, and calculating following probability by the observation bees according to the corresponding honey source position information:

if p _m Random generation number rand larger than value in (0, 1) interval ₃ Then the observing bees update the honey source position according to the formula (19); similarly, the adaptation values corresponding to the new honey source and the old honey source need to be compared, and if the adaptation values are better, the new honey source and the old honey source are replaced; if the latter is more optimal, the value of the non-evolution counter three is increased by 1;

step Step434 when the solution quality of a honey source is in the real _limit None of the updates has evolved, and the employment bee corresponding to the honey source is changed into a scout bee, the original honey source is abandoned, the new honey source position is randomly generated again according to the formula (18), and the corresponding trial value is set to 0.