CN117371812B

CN117371812B - Aircraft group collaborative decision generation method, system and equipment

Info

Publication number: CN117371812B
Application number: CN202311325343.0A
Authority: CN
Inventors: 李雄; 秦小营; 倪晓升; 张易东; 蒋燕梅; 吕雅丽; 熊宇涵; 冼军; 成诚
Original assignee: Sun Yat Sen University
Current assignee: Sun Yat Sen University
Filing date: 2023-10-12
Publication date: 2024-07-02
Anticipated expiration: 2043-10-12

Abstract

The invention discloses a method, a system and equipment for generating collaborative decisions of an aircraft group, and relates to the technical field of aircrafts. And calculating event pseudo tags by adopting a preset density peak clustering algorithm, and training an initial event extraction model based on an event pseudo tag set to obtain a target event extraction model. And respectively carrying out event extraction and event comparison on the aircraft cluster data set and the ground control station data set through a target event extraction model to obtain a scheme comparison result. And when the scheme comparison result is inconsistent, selecting an optimal instruction sequence according to the corresponding authorization data of the aircraft group and a preset improved multi-agent depth deterministic strategy gradient algorithm to obtain an aircraft group collaborative decision. And when the scheme comparison results are consistent, taking the aircraft group scheme as an aircraft group collaborative decision. And fine-tuning an initial event extraction model based on the event pseudo tag so as to extract the event, rapidly grasping the overall situation of the flight environment, and preparing a corresponding scheme to cope with the emergency.

Description

Aircraft group collaborative decision generation method, system and equipment

Technical Field

The invention relates to the technical field of aircrafts, in particular to a method, a system and equipment for generating cooperative decisions of an aircraft group.

Background

The aircraft has the function of carrying out flight tasks in or out of the atmosphere, including transportation, reconnaissance, military, scientific research and other fields. In transportation, the aircraft can be used for long-distance travel and cargo transportation, so that the distance is greatly shortened, and the transportation efficiency is improved. In scientific research, aircrafts are widely applied to various scientific researches, such as astronomy, meteorology, geology and other fields. The aircraft group is a group formed by a plurality of aircrafts, and specific tasks such as reconnaissance monitoring, rescue searching, logistics transportation and the like are completed through cooperative work. The aircraft clusters may enable more efficient, more accurate operation, and in some cases more powerful functionality and flexibility.

The flight control and stability requirements of the high-speed flying aircraft group are higher, because the multiple high-speed flying aircraft need to effectively ensure cooperative operation while avoiding mutual interference when performing tasks. This makes real-time, efficient communication and collaboration work required by the aircraft clusters to ensure coordination and coordination between the individual aircraft.

However, aircraft may face more complex challenges in aerodynamic environments, such as airflow turbulence, aerodynamic instability, etc., which may lead to difficult, unstable, and even uncontrolled aircraft attitude control. The existing cooperative decision of the aircraft group is usually set in advance, cannot be adjusted according to emergency, so that the normal completion of operation can be influenced by the faults of single aircraft in the aircraft group flying at high speed, and even the normal operation of other aircrafts is interfered, so that huge losses are caused.

Disclosure of Invention

The invention provides a method, a system and equipment for generating an aircraft group collaborative decision, which solve the technical problems that the conventional aircraft group collaborative decision cannot be adjusted according to emergency conditions and the normal completion of operation is easily affected due to the faults of single aircraft in an aircraft group.

The invention provides a method for generating a collaborative decision of an aircraft group, which comprises the following steps:

When emergency data of an aircraft group are received, calculating event pseudo tags corresponding to an aircraft group data set corresponding to the emergency data and a ground control station data set generated by a ground control station by adopting a preset density peak clustering algorithm to obtain an event pseudo tag set;

Training an initial event extraction model based on the event pseudo tag set to obtain a target event extraction model;

Respectively carrying out event extraction and event comparison on the aircraft group data set and the ground control station data set through the target event extraction model to obtain a scheme comparison result;

When the scheme comparison result is inconsistent, selecting an optimal instruction sequence according to the authorization data corresponding to the aircraft group and a preset improved multi-agent depth deterministic strategy gradient algorithm to obtain an aircraft group collaborative decision-making scheme corresponding to the emergency data;

and when the scheme comparison results are consistent, taking the aircraft group scheme in the aircraft group data as an aircraft group collaborative decision scheme corresponding to the emergency data.

Optionally, when the emergency data of the aircraft group is received, calculating an event pseudo tag corresponding to an aircraft group data set corresponding to the emergency data and a ground control station data set generated by a ground control station by adopting a preset density peak clustering algorithm, so as to obtain an event pseudo tag set, which comprises the following steps:

when receiving emergency data of an aircraft group, acquiring an aircraft group data set generated by the aircraft group based on the emergency data and a ground control station data set generated by a ground control station;

Extracting feature vectors of the aircraft group data set and the ground control station data set through a preset jump model to obtain a feature vector set;

Respectively calculating the densities corresponding to each point in the feature vector set by adopting a preset density calculation formula to obtain the density;

the preset density calculation formula is as follows:

Where ρ (x _i) represents the density of the point x _i; k represents the number of neighbors of a point, namely the top k points nearest to the point by taking Euclidean distance as a measurement standard; representing the Euclidean distance between point x _i and point x _j, u represents the dimension of one feature vector set point, i.e., the number of features, g is a subscript used to index the different features; KNN (x _i) represents the set of top k points in the dataset that have the smallest euclidean distance from point x _i; i represents the i-th point x _i in the dataset; j represents the j-th point x _j in the dataset;

substituting the density into a preset relative density calculation formula to calculate the relative density of the points to obtain the relative density;

the preset relative density calculation formula is as follows:

Wherein r _ρ(x_i) represents the relative density of point x _i; ρ (x _i) represents the density of the point x _i; ρ (x _j) represents the density of the point x _j; k represents the number of neighbors of a point, namely the top k points nearest to the point by taking Euclidean distance as a measurement standard; KNN (x _i) represents the set of top k points in the dataset that have the smallest euclidean distance from point x _i; i represents the i-th point x _i in the dataset; j represents the j-th point x _j in the dataset;

selecting points with the relative density larger than the density average value corresponding to the relative density to obtain a plurality of cluster centers;

dividing the feature vector set by adopting a preset domain calculation formula according to the cluster center to obtain a plurality of sets;

the preset domain calculation formula is as follows:

D_mn(x_p)＝{x_q|x_q∈MN(x_p)∨(x_m∈MN(x_p))∧x_q∈MN(x_m))};

Wherein D _mn(x_p) represents the domain of point x _p; MN (x _p) represents the set of all mutual neighbors of point x _p; MN (x _m) represents the set of all mutual neighbors of point x _m; x _p denotes the p-th point in the feature vector set; x _m denotes the mth point in the feature vector set; x _q represents the qth point in the feature vector set;

And setting corresponding labels for the points corresponding to the sets according to the labels corresponding to the cluster centers to obtain event pseudo-label sets.

Optionally, the step of training an initial event extraction model based on the event pseudo tag set to obtain a target event extraction model includes:

inputting the event pseudo tag set into an initial event extraction model to obtain an initial loss function value;

fine-tuning the initial event extraction model according to the initial loss function value to obtain an intermediate event extraction model;

Inputting the event pseudo tag set into the intermediate event extraction model to obtain a target loss function value;

judging whether the target loss function value is a preset function value or not;

if yes, taking the intermediate event extraction model corresponding to the target loss function value as a target event extraction model;

If not, taking the intermediate event extraction model as an initial event extraction model, and jumping to execute the step of inputting the event pseudo tag set into the initial event extraction model to obtain an initial loss function value.

Optionally, the step of performing event extraction and event comparison on the aircraft cluster data set and the ground control station data set through the target event extraction model to obtain a scheme comparison result includes:

respectively carrying out event extraction on the aircraft group data set and the ground control station data set through the target event extraction model to obtain an aircraft group event and a ground event;

comparing the aircraft group event with the ground event according to the trigger word category and the parameter similarity to obtain an event comparison result;

and constructing a scheme comparison result by adopting all event comparison results.

Optionally, when the solution comparison result is inconsistent, selecting an optimal instruction sequence according to the authorization data corresponding to the aircraft group and a preset improved multi-agent depth deterministic strategy gradient algorithm, and obtaining an aircraft group collaborative decision-making solution corresponding to the emergency data, where the method includes the steps of:

When the scheme comparison result is inconsistent, judging whether the authorization data corresponding to the aircraft group is authorization or not;

if yes, determining an aircraft group collaborative decision-making scheme corresponding to the emergency data according to the instruction sequence in the preset time and the aircraft group scheme;

If not, selecting the instruction sequence of the aircraft group data set and the ground control station data set according to a preset improved multi-agent depth deterministic strategy gradient algorithm to obtain a target instruction sequence;

Judging whether the target instruction sequence is a preset sequence or not;

if yes, taking the aircraft group scheme as an aircraft group collaborative decision scheme corresponding to the emergency data;

if not, taking the target instruction sequence as an aircraft group collaborative decision scheme corresponding to the emergency data.

Optionally, the step of determining the aircraft group collaborative decision-making scheme corresponding to the emergency data according to the instruction sequence in the preset time and the aircraft group scheme includes:

Judging whether the aircraft group and the ground control station find an optimal instruction sequence within a preset time;

if yes, taking the optimal instruction sequence as an aircraft group collaborative decision scheme corresponding to the emergency data;

and if not, taking the aircraft group scheme as an aircraft group collaborative decision scheme corresponding to the emergency data.

Optionally, the step of selecting the instruction sequence for the aircraft group data set and the ground control station data set according to the preset improved multi-agent depth deterministic strategy gradient algorithm to obtain a target instruction sequence includes:

constructing an initial decision criterion set by adopting decision criteria corresponding to the aircraft group data set and the ground control station data set;

Selecting a decision criterion with the highest priority in the initial decision criterion set to obtain a priority decision criterion;

according to the priority decision criterion, an instruction acquisition algorithm is adopted to respectively endow the instructions corresponding to the aircraft group data set and the ground control station data set with preset weights, so as to obtain an instruction sequence set;

selecting a plurality of initial instruction sequences meeting a preset score threshold value in the instruction sequence set by adopting a preset improved multi-agent depth deterministic strategy gradient algorithm, and constructing an initial instruction queue;

selecting an initial instruction sequence with the highest score in the initial instruction queue as an intermediate instruction sequence;

And determining a target instruction sequence according to the self constraint condition of the aircraft group and the intermediate instruction sequence.

Optionally, the step of determining a target instruction sequence according to the self-constraint condition of the aircraft group and the intermediate instruction sequence includes:

Judging whether the aircraft group can execute the intermediate instruction sequence under the self constraint condition;

If yes, the intermediate instruction sequence is used as a target instruction sequence;

If not, deleting the intermediate instruction sequence from the initial instruction queue to obtain an intermediate instruction queue;

judging whether the intermediate instruction queue is an empty set or not;

If yes, deleting the priority decision criterion from the initial decision criterion set to obtain a target decision criterion set;

judging whether the target decision criterion set is an empty set or not;

if yes, the preset sequence is used as a target instruction sequence;

if not, taking the target decision criterion set as an initial decision criterion set, and jumping to execute the decision criterion with the highest priority in the initial decision criterion set to obtain a priority decision criterion;

if not, taking the intermediate instruction queue as an initial instruction queue, and jumping to execute the step of selecting the initial instruction sequence with the highest score in the initial instruction queue as an intermediate instruction sequence.

The invention also provides an aircraft group collaborative decision generation system, which comprises:

The system comprises an event pseudo tag set obtaining module, a data processing module and a data processing module, wherein the event pseudo tag set obtaining module is used for calculating an event pseudo tag corresponding to an aircraft cluster data set corresponding to emergency data and a ground control station data set generated by a ground control station by adopting a preset density peak clustering algorithm when the emergency data of the aircraft cluster are received;

The target event extraction model obtaining module is used for training an initial event extraction model based on the event pseudo tag set to obtain a target event extraction model;

The scheme comparison result obtaining module is used for respectively carrying out event extraction and event comparison on the aircraft cluster data set and the ground control station data set through the target event extraction model to obtain a scheme comparison result;

The first acquisition module is used for selecting an optimal instruction sequence according to the authorization data corresponding to the aircraft group and a preset improved multi-agent depth deterministic strategy gradient algorithm when the scheme comparison result is inconsistent, so as to acquire an aircraft group collaborative decision scheme corresponding to the emergency data;

And the second obtaining module is used for taking the aircraft group scheme in the aircraft group data set as the aircraft group collaborative decision scheme corresponding to the emergency data when the scheme comparison results are consistent.

The invention also provides electronic equipment, which comprises a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to enable the processor to execute the steps for realizing the collaborative decision-making method based on the aircraft group according to any one of the above steps.

From the above technical scheme, the invention has the following advantages:

According to the method, an event pseudo tag set is obtained by calculating an aircraft cluster data set corresponding to emergency data and an event pseudo tag corresponding to a ground control station data set generated by a ground control station through a preset density peak clustering algorithm. And training an initial event extraction model based on the event pseudo tag set to obtain a target event extraction model. And respectively carrying out event extraction and event comparison on the aircraft cluster data set and the ground control station data set through a target event extraction model to obtain a scheme comparison result. And when the scheme comparison result is inconsistent, selecting an optimal instruction sequence according to the authorization data corresponding to the aircraft group and a preset improved multi-agent depth deterministic strategy gradient algorithm to obtain an aircraft group collaborative decision scheme corresponding to the emergency data. And when the scheme comparison results are consistent, taking the aircraft group scheme in the aircraft group data as an aircraft group collaborative decision scheme corresponding to the emergency data. The technical problem that the conventional collaborative decision of the aircraft group cannot be adjusted according to emergency conditions, and the normal completion of operation is affected easily due to the faults of single aircraft in the aircraft group is solved. And fine-tuning an initial event extraction model based on the event pseudo tag so as to extract the event, rapidly grasping the overall situation of the flight environment, and preparing a corresponding scheme to cope with the emergency. The improved multi-agent depth deterministic strategy gradient algorithm is adopted to search an instruction sequence, so that the aircraft group avoids threat, completes tasks and successfully copes with emergency by maximum force.

Drawings

In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained from these drawings without inventive faculty for a person skilled in the art.

Fig. 1 is a flowchart illustrating steps of an aircraft group collaborative decision-making method according to a first embodiment of the present invention;

fig. 2 is a flowchart of steps of an aircraft group collaborative decision-making method according to a second embodiment of the present invention;

Fig. 3 is a CNN-based event extraction model framework according to a second embodiment of the present invention;

FIG. 4 is a flow chart of a fine tuning process of an event extraction model according to a second embodiment of the present invention;

FIG. 5 is a block diagram of a second embodiment of the present invention for executing an instruction fetch algorithm;

Fig. 6 is a flow chart of an aircraft group collaborative decision-making method according to a second embodiment of the present invention;

Fig. 7 is a schematic diagram of an emergency system to which the method for generating a collaborative decision of an aircraft group according to the second embodiment of the present invention is applied;

Fig. 8 is a block diagram of an aircraft group collaborative decision-making system according to a third embodiment of the present invention.

Detailed Description

The embodiment of the invention provides a method, a system and equipment for generating an aircraft group collaborative decision, which are used for solving the technical problems that the conventional aircraft group collaborative decision cannot be adjusted according to emergency conditions and the normal completion of operation is easily affected due to the faults of single aircraft in an aircraft group.

In order to make the objects, features and advantages of the present invention more comprehensible, the technical solutions in the embodiments of the present invention are described in detail below with reference to the accompanying drawings, and it is apparent that the embodiments described below are only some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Referring to fig. 1, fig. 1 is a flowchart illustrating steps of an aircraft group collaborative decision generation method according to an embodiment of the present invention.

The first embodiment of the invention provides a collaborative decision generation method for an aircraft group, which comprises the following steps:

And 101, when the emergency data of the aircraft group is received, calculating an event pseudo tag corresponding to the aircraft group data set corresponding to the emergency data and the ground control station data set generated by the ground control station by adopting a preset density peak clustering algorithm to obtain an event pseudo tag set.

In embodiments of the present invention, an aircraft group may be considered as a plurality of agents that, when they encounter an emergency, will generate a new sequence of instructions in order to address the emergency. These new command sequences are often inconsistent with the old command sequences assigned to the ground control station prior to departure of the aircraft fleet. In addition, the old instruction sequence is called ground control station scheme, and the new instruction sequence is called aircraft group scheme. The aircraft fleet data set includes an aircraft fleet solution and aircraft fleet data. The ground control station data set includes a ground control station scheme and ground control station data.

When the emergency data of the aircraft group is received, an aircraft group data set generated by the aircraft group based on the emergency data and a ground control station data set generated by the ground control station are acquired. And extracting the characteristic vectors of the aircraft group data set and the ground control station data set through a preset jump word model to obtain a characteristic vector set, wherein the preset jump word model is a Skip-gram model. And respectively calculating the densities corresponding to each point in the feature vector set by adopting a preset density calculation formula to obtain the density. Substituting the density into a preset relative density calculation formula to calculate the relative density of the points, and obtaining the relative density. And selecting points with the relative density larger than the density average value corresponding to the relative density to obtain a plurality of cluster centers. And dividing the feature vector set by adopting a preset domain calculation formula according to the cluster center to obtain a plurality of sets. And setting corresponding labels for the points corresponding to the sets according to the labels corresponding to the cluster centers to obtain an event pseudo-label set.

And 102, training an initial event extraction model based on the event pseudo tag set to obtain a target event extraction model.

In the embodiment of the invention, the initial loss function value is obtained by inputting the event pseudo tag set into the initial event extraction model. And fine-tuning the initial event extraction model according to the initial loss function value to obtain an intermediate event extraction model. And inputting the event pseudo tag set into an intermediate event extraction model to obtain a target loss function value. And judging whether the target loss function value is a preset function value or not. If yes, taking the intermediate event extraction model corresponding to the target loss function value as a target event extraction model. If not, taking the intermediate event extraction model as an initial event extraction model, and jumping to execute the step of inputting the event pseudo tag set into the initial event extraction model to obtain an initial loss function value.

And 103, respectively carrying out event extraction and event comparison on the aircraft cluster data set and the ground control station data set through a target event extraction model to obtain a scheme comparison result.

In the embodiment of the invention, the target event extraction model is used for respectively carrying out event extraction on the aircraft group data set and the ground control station data set to obtain the aircraft group event and the ground event. And comparing the aircraft group event with the ground event according to the trigger word category and the parameter similarity to obtain an event comparison result. And (5) adopting all event comparison results to construct a scheme comparison result.

And 104, when the scheme comparison result is inconsistent, selecting an optimal instruction sequence according to the authorization data corresponding to the aircraft group and a preset improved multi-agent depth deterministic strategy gradient algorithm to obtain an aircraft group collaborative decision scheme corresponding to the emergency data.

In the embodiment of the invention, when the scheme comparison result is inconsistent, whether the authorization data corresponding to the aircraft group is authorization is judged. If yes, determining an aircraft group collaborative decision-making scheme corresponding to the emergency data according to the instruction sequence and the aircraft group scheme in the preset time. If not, selecting the instruction sequence of the aircraft cluster data set and the ground control station data set according to a preset improved multi-agent depth deterministic strategy gradient algorithm to obtain a target instruction sequence. And judging whether the target instruction sequence is a preset sequence or not. If yes, taking the aircraft group scheme as an aircraft group collaborative decision scheme corresponding to the emergency data. If not, taking the target instruction sequence as an aircraft group collaborative decision scheme corresponding to the emergency data.

And 105, when the scheme comparison results are consistent, taking the aircraft group scheme in the aircraft group data as an aircraft group collaborative decision scheme corresponding to the emergency data.

In the embodiment of the invention, when more than 80% of the events in the ground control station scheme and the aircraft swarm scheme are the same, the ground control station scheme and the aircraft swarm scheme are considered to be consistent. And when the two schemes are consistent, executing an aircraft swarm scheme, namely taking the aircraft swarm scheme in the aircraft swarm data as an aircraft swarm collaborative decision scheme corresponding to the emergency data.

In the embodiment of the invention, the event pseudo tag set is obtained by calculating the aircraft cluster data set corresponding to the emergency data and the event pseudo tag corresponding to the ground control station data set generated by the ground control station by adopting a preset density peak clustering algorithm. And training an initial event extraction model based on the event pseudo tag set to obtain a target event extraction model. And respectively carrying out event extraction and event comparison on the aircraft cluster data set and the ground control station data set through a target event extraction model to obtain a scheme comparison result. And when the scheme comparison result is inconsistent, selecting an optimal instruction sequence according to the authorization data corresponding to the aircraft group and a preset improved multi-agent depth deterministic strategy gradient algorithm to obtain an aircraft group collaborative decision scheme corresponding to the emergency data. And when the scheme comparison results are consistent, taking the aircraft group scheme in the aircraft group data as an aircraft group collaborative decision scheme corresponding to the emergency data. The technical problem that the conventional collaborative decision of the aircraft group cannot be adjusted according to emergency conditions, and the normal completion of operation is affected easily due to the faults of single aircraft in the aircraft group is solved. And fine-tuning an initial event extraction model based on the event pseudo tag so as to extract the event, rapidly grasping the overall situation of the flight environment, and preparing a corresponding scheme to cope with the emergency. The improved multi-agent depth deterministic strategy gradient algorithm is adopted to search an instruction sequence, so that the aircraft group avoids threat, completes tasks and successfully copes with emergency by maximum force.

Referring to fig. 2, fig. 2 is a flowchart illustrating steps of an aircraft group collaborative decision generation method according to a second embodiment of the present invention.

The second embodiment of the invention provides another method for generating collaborative decisions of an aircraft group, which comprises the following steps:

step 201, when emergency data of an aircraft group is received, calculating an event pseudo tag corresponding to an aircraft group data set corresponding to the emergency data and a ground control station data set generated by a ground control station by adopting a preset density peak clustering algorithm, and obtaining an event pseudo tag set.

Further, step 201 may include the following sub-steps S11-S17:

S11, when the emergency data of the aircraft group are received, acquiring an aircraft group data set generated by the aircraft group based on the emergency data and a ground control station data set generated by the ground control station.

And S12, extracting the characteristic vectors of the aircraft group data set and the ground control station data set through a preset jump model to obtain a characteristic vector set.

And S13, respectively calculating the densities corresponding to each point in the feature vector set by adopting a preset density calculation formula to obtain the density.

S14, substituting the density into a preset relative density calculation formula to calculate the relative density of the points, and obtaining the relative density.

S15, selecting points with the relative density larger than the density average value corresponding to the relative density, and obtaining a plurality of cluster centers.

S16, dividing the feature vector set by adopting a preset domain calculation formula according to the cluster center to obtain a plurality of sets.

S17, setting corresponding labels for the points corresponding to the sets according to the labels corresponding to the cluster centers, and obtaining an event pseudo label set.

In the embodiment of the invention, in order to generate more accurate event pseudo tags to improve the accuracy of event extraction, a density peak clustering algorithm needs to be improved. The density peak clustering algorithm is a simple and effective clustering analysis method. But have the following limitations: the density peak clustering algorithm requires a user to manually select a cluster center in the execution process, and the operation has certain subjectivity. The distribution strategy of the density peak clustering algorithm is to distribute one point to the cluster where the point is more dense than it and closest to it. Such allocation strategies often allocate most of the points in a less dense cluster to a high density cluster, i.e. domino effect. The invention provides a method for automatically determining the cluster center and a preset density peak clustering algorithm of a new point allocation strategy aiming at the limitation.

To automatically determine cluster centers, the density of points, the relative density of points, needs to be calculated. For the calculation of the density, for any point x _i, the smaller its average distance to its K-nearest neighbor (K-Nearest Neighbors, KNN), the greater its density. Therefore, the density calculation formula of the point x _i, that is, the preset density calculation formula is:

Where ρ (x _i) represents the density of the point x _i; k represents the number of neighbors of a point, namely the top k points nearest to the point by taking Euclidean distance as a measurement standard; Representing the Euclidean distance between point x _i and point x _j, u represents the dimension of one feature vector set point, i.e., the number of features, g is a subscript used to index the different features; KNN (x _i) represents the set of top k points in the dataset that have the smallest euclidean distance from point x _i; i represents the i-th point x _i in the dataset; j denotes the j-th point x _j in the dataset, let a total of v points in the dataset, then i=1, 2,. -%, v; j=1, 2,..v. The data sets herein refer to both aircraft cluster data sets and ground control station data sets.

The calculation mode of the relative density of the points, namely a preset relative density calculation formula is as follows:

wherein r _ρ(x_i) represents the relative density of point x _i; ρ (x _i) represents the density of the point x _i; ρ (x _j) represents the density of the point x _j; k represents the number of neighbors of a point, namely the top k points nearest to the point by taking Euclidean distance as a measurement standard; KNN (x _i) represents the set of top k points in the dataset that have the smallest euclidean distance from point x _i; i represents the i-th point x _i in the dataset; j represents the j-th point x _j in the dataset.

The method for automatically determining the cluster center comprises the following steps: normalization is performed on the relative density r _ρ and the high-density nearest neighbor distance delta of the points, and the results after normalization are still represented by r _ρ and delta. Let v ₁＝r_ρ delta denote the result of ordering v ₁ from large to small by v ₂. For point 1 in v ₂, if its relative density is greater than the density average of the relative densities of all points and its high density nearest neighbor distance is greater than the average of the high density nearest neighbors of all points, then this point is the cluster center. The next point in v ₂ is continuously examined as to whether the cluster is the center or not until the number of cluster centers reaches the set value. The number of cluster centers is a parameter, and needs to be set in advance.

The new point allocation strategy is based on the domain of the point, and the domain calculation formula of the point, namely the preset domain calculation formula is as follows:

D_mn(x_p)＝{x_q|x_q∈MN(x_p)∨(x_m∈MN(x_p))∧x_q∈MN(x_m))};

Wherein D _mn(x_p) represents the domain of point x _p; MN (x _p) represents the set of all mutual neighbors of point x _p; If x _i∈KNN(x_j) and x _j∈KNN(x_i), then x _j is the mutual adjacency of x _i, denoted as x _j∈MN(x_i). Where x _i、x_j each represents any point in the dataset. The meaning of MN (x _m)、MN(x_p) is similar to that of MN (x _i). x _p represents the p-th point in the dataset, assuming a total of v points in the dataset, then p=1, 2. x _m represents any one point in MN (x _p). x _q denotes any one point in D _mn(x_p). D _mn(x_p) represents the domain of x _p, which essentially is a set of points that meet some condition, either the mutual adjacency of x _p or the mutual adjacency of points in the mutual adjacency of x _p. That is, if x _q∈MN(x_p), then x _q is in the set of MN (x _p); if it is But x _q∈MN(x_m) and x _m∈MN(x_p), then x _q is in this set D _mn(x_p).

New point allocation policy: for non-cluster center points, a clustering process is performed within the domain. Starting from the point x _i where the density is greatest, if the high density nearest neighbor x _j of x _i is in D _mn(x_i), x _i is assigned to the cluster where x _j is located. If the high-density nearest neighbor x _j of x _i is not in D _mn(x_i), the point closest to x _i and already allocated in D _mn(x_i) is denoted by x _k, and x _j is allocated to the cluster where x _k is located, so as to obtain a set corresponding to the center of the cluster.

Ground control station schemes, ground control station data, aircraft fleet schemes, and aircraft fleet data are referred to as correlation data. The specific process of obtaining the event pseudo tag is as follows: the feature vector of the related data can be extracted according to Skip-gram model, i.e. preset Skip model. These feature vectors are input into an improved density peak clustering algorithm, i.e., a preset density peak clustering algorithm. The improved density peak clustering algorithm calculates the similarity between the feature vectors according to the Euclidean distance, and then the clustering process is developed. The improved density peak clustering algorithm outputs a number of cluster centers, and for any one cluster center, the improved density peak clustering algorithm outputs all points that are in the same cluster. Where a point represents an event. The number of cluster centers is the number of event types. For any cluster center, a label is set for it, and for points in the same cluster as this cluster center, their labels are the same as the label of the cluster center. Assuming there are currently A, B, C event types, then the improved density peak clustering algorithm will output 3 cluster centers and which points are in the same cluster as which cluster center. The label of the cluster center corresponding to the type a event is set to "a", the label of the cluster center corresponding to the type B event is set to "B", and the label of the cluster center corresponding to the type C event is set to "C". "A", "B", "C" are event tags herein also referred to as event pseudo tags. In the supervised learning algorithm, most of the labels are manually marked, relatively accurate and true labels, so the labels are called event pseudo labels. The improved density peak clustering algorithm in the patent belongs to an unsupervised learning algorithm. In the unsupervised learning algorithm, no label is marked by people, only labels generated by the algorithm are used, and in most cases, the labels are not as accurate as the labels marked by people and have wrong labels, such as the labels of the type A event and the type B event, so the labels generated by the unsupervised learning algorithm are called event pseudo labels.

And 202, training an initial event extraction model based on the event pseudo tag set to obtain a target event extraction model.

Further, step 202 may comprise the following sub-steps S21-S26:

S21, inputting the event pseudo tag set into an initial event extraction model to obtain an initial loss function value.

S22, fine-tuning the initial event extraction model according to the initial loss function value to obtain an intermediate event extraction model.

S23, inputting the event pseudo tag set into an intermediate event extraction model to obtain a target loss function value.

S24, judging whether the target loss function value is a preset function value, if so, executing the step S25, and if not, executing the step S26.

S25, taking the intermediate event extraction model corresponding to the target loss function value as a target event extraction model.

S26, taking the intermediate event extraction model as an initial event extraction model, and jumping to execute the step of inputting the event pseudo tag set into the initial event extraction model to obtain an initial loss function value.

In the embodiment of the present invention, as shown in fig. 3, the initial event extraction model is a CNN-based event extraction model, and the event extraction modeling is set as a multi-classification task. Setting the sentence length as sl, cutting when the sentence length is greater than sl, and filling when the sentence length is less than sl. The chinese sentence in the dataset is partitioned into token (i.e., a basic unit of text) using jieba toolkit (i.e., the barking toolkit). The Skip-gram model (which is a word embedding model) is used to obtain word embedding of the token, and the dimension is assumed to be d. Assume that the number of event trigger word categories is n ₁ and the number of event parameter role categories is n ₂. Features of the semantic units of a specific length are extracted through the convolution layer, and sentence features are extracted through the multi-layer perceptron. When convolutional neural networks are applied to sentences, features of a length of semantic units are captured by sizing the filters. If the filter size is 3×d, features of semantic units of length 3 are captured. The patent adopts filters with the sizes of 2×d, 3×d, 4×d and 5×d to extract the features of semantic units with the corresponding lengths. The number of filters of different sizes was 128. Thus, sentences are encoded by the convolutional layer as vectors of length 128×4=512. And connecting word embedding of all token in a sentence to obtain a vector with the length of sl. The number of neurons of a first layer of the multi-layer perceptron is set as sl, the number of neurons of a subsequent layer is sequentially halved, the number of neurons of an output layer is greater than or equal to 2.5 (n ₁+n₂), and the number of neurons of the output layer is assumed to be n ₀. And connecting semantic unit features acquired from the convolution layer with sentence features acquired from the multi-layer perceptron to obtain vectors with lengths of 512+n ₀, and inputting the vectors into the convolution layer to obtain a classification result. Based on the classification result, a loss function is calculated, and a back propagation error is obtained, thereby determining model parameters.

As shown in fig. 4, a modified density peak clustering algorithm is applied to the correlation data to obtain event pseudo tags and event parameter role pseudo tags. And inputting the pseudo tag into the initial event extraction model to obtain a loss function value, so that the error is reversely propagated to fine-tune the initial event extraction model, and an intermediate event extraction model is obtained. And inputting the event pseudo tag set into a CNN-based event extraction model, namely an intermediate event extraction model, and obtaining a target loss function value. The preset function value means that the objective loss function value is not decreased any more. And stopping the fine tuning process when the target loss function value no longer reduces the convergence of the model, and taking the intermediate event extraction model corresponding to the target loss function value as the target event extraction model. Otherwise, taking the intermediate event extraction model as an initial event extraction model, and jumping to execute the step of inputting the event pseudo tag set into the initial event extraction model to obtain an initial loss function value.

And 203, respectively carrying out event extraction and event comparison on the aircraft cluster data set and the ground control station data set through a target event extraction model to obtain a scheme comparison result.

Further, step 203 may comprise the following sub-steps S31-S33:

S31, respectively carrying out event extraction on the aircraft cluster data set and the ground control station data set through a target event extraction model to obtain an aircraft cluster event and a ground event.

S32, comparing the aircraft group event with the ground event according to the trigger word category and the parameter similarity to obtain an event comparison result.

S33, adopting all event comparison results, and constructing scheme comparison results.

In the embodiment of the present invention, the purposes of event extraction on related data are two: firstly, judging whether the aircraft group scheme is consistent with the ground control station scheme or not based on an event extraction result: and secondly, inputting event extraction results into an improved multi-agent depth deterministic strategy gradient algorithm to obtain an optimal instruction sequence. And extracting the event on the related data by adopting the finely-adjusted event extraction model based on the CNN, namely the target event extraction model. The ground control station scheme and the aircraft fleet scheme are each comprised of a series of instructions. The instructions are in one-to-one correspondence with the events. An event in the ground control station scheme is denoted by e _i, an event in the aircraft group scheme is denoted by e _j, and if the trigger words of e _i and e _j belong to the same category, and more than 50% of the parameters of e _i and e _j are the same, then e _i and e _j belong to the same event. A ground control station scenario is considered consistent with an aircraft fleet scenario when more than 80% of the events in the ground control station scenario and the aircraft fleet scenario are identical. When the two schemes agree, the aircraft group scheme is executed and the algorithm is ended, i.e. step 210 is executed, otherwise step 204 is executed.

Step 204, if the comparison result of the schemes is inconsistent, determining whether the authorization data corresponding to the aircraft group is authorization, if yes, executing step 205, and if not, executing step 206.

In the embodiment of the invention, in order to cooperate with the practicability of the decision algorithm, an authorization module is added to determine whether to authorize the aircraft group to act on its own. Step 205 is performed when authorized, and step 206 is performed when unauthorized.

Step 205, determining an aircraft group collaborative decision-making scheme corresponding to the emergency data according to the instruction sequence and the aircraft group scheme in the preset time.

Further, step 205 may include the following substeps S41-S43:

s41, judging whether the aircraft group and the ground control station find the optimal instruction sequence within the preset time, if so, executing the step S42, and if not, executing the step S43.

S42, taking the optimal instruction sequence as an aircraft group collaborative decision scheme corresponding to the emergency data.

S43, taking the aircraft group scheme as an aircraft group collaborative decision scheme corresponding to the emergency data.

In the embodiment of the invention, the preset time refers to a time interval set based on actual needs. And executing the instruction sequence if the aircraft group or the ground control station outputs the optimal instruction sequence within the preset time. And if the optimal instruction sequence is not output, taking the aircraft group scheme as the optimal instruction sequence and executing.

And 206, selecting an instruction sequence according to a preset improved multi-agent depth deterministic strategy gradient algorithm to obtain a target instruction sequence.

Further, step 206 may include the following substeps S51-S56:

s51, constructing an initial decision criterion set by adopting decision criteria corresponding to the aircraft cluster data set and the ground control station data set.

S52, selecting a decision criterion with the highest priority in the initial decision criterion set to obtain a priority decision criterion.

And S53, respectively endowing the instructions corresponding to the aircraft group data set and the ground control station data set with preset weights by adopting an instruction acquisition algorithm according to a priority decision criterion, so as to obtain an instruction sequence set.

S54, selecting a plurality of initial instruction sequences meeting a preset score threshold value in an instruction sequence set by adopting a preset improved multi-agent depth deterministic strategy gradient algorithm, and constructing an initial instruction queue.

S55, selecting an initial instruction sequence with the highest score in the initial instruction queue as an intermediate instruction sequence.

S56, determining a target instruction sequence according to the self constraint condition and the intermediate instruction sequence of the aircraft group.

Further, step S56 may include the following sub-steps S561-S569:

S561, judging whether the aircraft group can execute the intermediate instruction sequence under the self constraint condition, if so, executing step S562, and if not, executing step S563.

S562, taking the intermediate instruction sequence as a target instruction sequence.

S563, deleting the intermediate instruction sequence from the initial instruction queue to obtain an intermediate instruction queue.

S564, judging whether the intermediate instruction queue is empty, if yes, executing step S565, and if not, executing step S569.

S565, deleting the priority decision criterion from the initial decision criterion set to obtain a target decision criterion set.

S566, judging whether the target decision criterion set is an empty set, if so, executing step S567, and if not, executing step S568.

S567, taking the preset sequence as a target instruction sequence.

S568, taking the target decision criterion set as an initial decision criterion set, and jumping to execute the step of selecting the decision criterion with the highest priority in the initial decision criterion set to obtain the priority decision criterion.

S569, taking the intermediate instruction queue as an initial instruction queue, and jumping to execute the step of selecting the initial instruction sequence with the highest score in the initial instruction queue as the intermediate instruction sequence.

In the embodiment of the invention, the aircraft group and the ground control station respectively determine instruction weights according to different decision criteria, a plurality of candidate instruction sequences are determined by using an improved multi-agent depth deterministic strategy gradient algorithm, namely a preset improved multi-agent depth deterministic strategy gradient algorithm, and a target instruction sequence is selected by combining the constraint conditions of the aircraft group. Step 206 involves three aspects: instruction acquisition algorithm principle; improving a multi-agent depth deterministic strategy gradient algorithm; candidate instruction sequences are obtained based on an improved multi-agent depth deterministic strategy gradient algorithm.

(1) Instruction fetch algorithm principles. To more quickly complete collaborative decision tasks, resources are allocated more reasonably, and decision criteria are ordered in order of priority from high to low. On the aircraft group, a decision criterion with high priority is input into an instruction acquisition algorithm to find an optimal instruction sequence. On the ground control station, a low priority decision criterion is input to the instruction fetch algorithm to find the optimal instruction sequence. Executing the sequence when the aircraft group can find the optimal instruction sequence by the decision criterion with high priority; otherwise, the system communicates with the ground control station to acquire the optimal instruction sequence searched by the system. And guiding the next stage of action of the aircraft group by the optimal instruction sequence.

Instruction fetch algorithm process. And (5) taking out the decision criterion with the highest priority from the decision criterion set, and giving instruction weight according to the decision criterion. Higher weight is given to instructions related to ground control station schemes and aircraft group schemes, and lower weight is given to instructions related to ground control station data and aircraft group data. I.e. if the agent completes the instructions related to the ground control station scheme, the aircraft group scheme, a high prize is awarded to the agent. If the agent completes the instructions associated with the ground control station data, the aircraft population data, a low prize is awarded to the agent, which is positive. And selecting a plurality of candidate instruction sequences with highest scores from the instructions by adopting an improved multi-agent depth deterministic strategy gradient algorithm, and storing the candidate instruction sequences in a queue. Taking out the instruction sequence with highest score in the queue, and if the aircraft group residual energy can support to complete the instruction sequence, then the instruction sequence is the optimal instruction sequence; otherwise, deleting the instruction sequence from the queue, and continuously inspecting other instruction sequences in the queue. If the remaining energy of the aircraft group can not support any instruction sequence in the completion queue, replacing a decision criterion until an optimal instruction sequence is found. The above process is shown in fig. 5.

The higher weight is given to the instructions related to the ground control station scheme and the aircraft group scheme, and the lower weight is given to the instructions related to the ground control station data and the aircraft group data: the ground control station is a plan made after careful consideration, and the aircraft group plan is a plan made after consideration of an emergency. The instructions involved in the ground control station scheme and the aircraft group scheme have a high degree of trustworthiness. On the other hand, the ground control station data reflection information is mostly global information of the flight environment, and the aircraft group data reflection information is mostly local information of the flight environment. And the combination of the global information and the local information can reflect the flight environment of the aircraft group more exactly. Thus, not only are instructions associated with ground control station schemes and aircraft fleet schemes referenceable, but instructions associated with ground control station data and aircraft fleet data also have some referenceability.

(2) The improved multi-agent depth deterministic strategy gradient algorithm is preset. The decision-making scenario of the present invention is a host aircraft population and a target equipment population. The functions of all aircraft groups on own are consistent. In order to improve the generalization capability of the collaborative decision-making method, the invention sets the scene as a game process between two equipment groups. Similar to football matches, the team works against each other, the team works in cooperation, and different roles in the team play different roles. Different equipment corresponds to different roles within the team. The multi-agent depth deterministic strategy gradient algorithm is a multi-agent depth reinforcement learning algorithm which solves the problem of inter-team countermeasures on intra-team cooperation. However, in the multi-agent depth deterministic strategy gradient algorithm, each agent has a global evaluator, and the input space of the Critic network (i.e. the value network) increases with the increase of the number of agents, so that the learning period of the agents is long, and the multi-agent depth deterministic strategy gradient algorithm is only suitable for scenes with a small number of agents. Aiming at the problems, the invention improves the multi-agent depth deterministic strategy gradient algorithm, and then adopts the improved multi-agent depth deterministic strategy gradient algorithm to search a plurality of candidate instruction sequences aiming at a determined decision criterion.

An improved multi-agent depth deterministic strategy gradient algorithm. The central controller refers to N value networks, and the strategy area refers to N strategy networks equipped for N intelligent agents. In order to make the logic relationship among multiple agents clearer, a fusion layer is added between the central controller and the strategy interval, the fusion layer is composed of two pseudo-agents, and the central controller only needs to be responsible for the two pseudo-agents. The first pseudo-agent integrates the actions, observations and rewards of all agents of the own party into a set of actions, observations and rewards. The actions, observations and rewards of all agents in the second pseudo-agent integration target are a set of actions, observations and rewards. And then uploading the two groups of actions, observations and rewards to a central controller. The central controller is only aimed at two game parties, and only the competition relationship between two pseudo-intelligent agents is considered, so that the problem that a large number of intelligent agents are difficult to process by a multi-intelligent-agent depth deterministic strategy gradient algorithm is solved to a certain extent. A single pseudo-agent is directed to all real agents in the gaming party, which only need to consider the partnership. In the whole, the multi-agent depth deterministic strategy gradient algorithm inserted into the fusion layer not only relieves the current situation that the complexity of the algorithm increases along with the increase of the input space, but also can process the complex multi-agent relationship of cooperation and competition together. In this multi-agent game we ultimately focus on whether the own side can win only, and not on which agent wins the most prize. The entire multi-agent gaming scenario is thus divided into 2 views, one view facing the own, i.e. the first pseudo-agent. The other view faces the equipment cluster at the target. Observations captured by all agents on the own side are not exactly identical, and thus the observations are summed and input to the 1 st pseudo-agent. The same is true for the fusion of actions and rewards of all agents on the own side. The central controller evaluates the two pseudo-agents. The two pseudo-intelligent agents respectively help the true intelligent agents represented by the two pseudo-intelligent agents to improve the self strategy, and for a plurality of true intelligent agents corresponding to one pseudo-intelligent agent, the targets of the two pseudo-intelligent agents are consistent, namely winning, and the two pseudo-intelligent agents are in a complete cooperative relationship, and only the improvement opinion acquired from the central controller by the pseudo-intelligent agents is matched.

(3) Candidate instruction sequences are obtained based on an improved multi-agent depth deterministic strategy gradient algorithm. The decision scene is set as a group countermeasure scene of both equipment. The temporary setting equipment has three functions of reconnaissance, burst prevention and striking, and can be added with functions continuously. A＝{a₁,a₂,...,a_i,...,a_n1}、B＝{b₁,b₂,...,b_i,...,b_n2}、C＝{c₁,c₂,...,c_i,...,c_n3} represents a collection of equipment with reconnaissance, burst prevention, or striking functions on the own side, respectively. The ith equipment with reconnaissance, burst prevention, or striking functions is denoted by a _i、b_i、c_i, respectively. n ₁、n₂、n₃ represents the number of kinds of equipment having reconnaissance, burst prevention, and striking functions, respectively. A. B, C may be intersected as an apparatus may have one or more functions. Let X＝{x₁,x₂,...,x_i,...,x_n4}、Y＝{y₁,y₂,...,y_i,...,y_n5}、Z＝{z₁,z₂,...,z_i,...,z_n6} denote the set of equipment in the target with reconnaissance, burst prevention, or striking functions, respectively. The ith equipment with reconnaissance, burst prevention, or striking functions is denoted by x _i、y_i、z_i, respectively. n ₄、n₅、n₆ represents the number of kinds of various kinds of equipment having reconnaissance, burst prevention, and striking functions, respectively. X, Y, Z may be intersected as an apparatus may have one or more functions. For any one of the sets A, B, C, X, Y, Z, there are no repeating elements. The number of i-th equipment with reconnaissance function on the own side is denoted by a _ij, and b _ij、c_ij has a similar meaning. The number of ith equipment with reconnaissance function in the target is denoted by x _ij, and y _ij、z_ij has a similar meaning. The number of times that the i-th equipment with striking function of the own side can perform striking is denoted by c '_i, and z' _i has a similar meaning. For various types of equipment, the number of times that the reconnaissance and defense functions can be implemented is not limited, but after being defeated by other agents, the service is terminated, and the reconnaissance, strike or defense functions cannot be implemented.

An action space is set. For any agent, the actions that it can take come from the set a= { hit, scout, burst, rest, motion }, which is not required to have the ability to perform all actions in a.

A state space is set. The state space s= { kind of detected target, number of detected targets, kind of own equipment, number of own equipment, actions being performed by each equipment }.

A bonus function is set. When the own a _i th device detects a target, the own a _i th device adds r ₁ to the corresponding agent prize for the detected target, which is subtracted by r ₁ from the corresponding agent prize for the detected target. When the own b _i equipment burst succeeds, the own b _i equipment adds r ₂ to the corresponding agent prize, and fails to prevent the agent corresponding to the target of the b _i burst, and the prize is reduced by r ₂. When the own c _i equipment hits the target, the own c _i equipment adds r ₃ to the corresponding agent prize of the hit target, and the corresponding agent of the hit target subtracts r ₃ from the prize of the hit target, and the hit target is then out of service. By a certain time, if the sum of rewards obtained by the own equipment is larger than the sum of rewards obtained by the equipment at the target, the own is winning, otherwise the own is not winning.

The process of two-party agent game is a process that different agents respectively execute a series of actions in a time period and a space. There are currently two decision schemes, one is an aircraft swarm scheme and one is a ground control station scheme. For the two-party agent gaming process, the two decision schemes are two priori information. And respectively inputting the two priori information into an improved multi-agent depth deterministic strategy gradient algorithm, and observing own winning conditions. And then inputting the two priori information into an improved multi-agent depth deterministic strategy gradient algorithm at the same time, cutting out the conflict part when the two priori information conflict, and observing the winning condition of me. Finally, the first instruction sequences with the best prizes and winning own winning are selected.

Step 207, determining whether the target instruction sequence is a preset sequence, if yes, executing step 208, and if not, executing step 209.

In the embodiment of the present invention, the preset sequence refers to that the target instruction sequence is not the optimal instruction sequence, and the preset instruction sequence may be an empty set. And judging whether the target instruction sequence is an optimal instruction sequence or not according to the instruction sequence selection of the aircraft group data set and the ground control station data set according to a preset improved multi-agent depth deterministic strategy gradient algorithm.

And step 208, taking the aircraft group scheme as an aircraft group collaborative decision scheme corresponding to the emergency data.

In the embodiment of the invention, in extreme cases, the aircraft group and the ground control station do not find the optimal instruction sequence, and the aircraft group scheme is used as the optimal instruction sequence. A series of events can be extracted from the fleet solution, which events correspond one-to-one to the instructions, so that the fleet solution is essentially a sequence of instructions.

Step 209, taking the target instruction sequence as an aircraft group collaborative decision scheme corresponding to the emergency data.

In an embodiment of the invention, because the aircraft clusters find an optimal instruction sequence based on a high priority decision criterion, this instruction sequence is executed when the aircraft clusters can find the optimal instruction sequence. And when the aircraft group cannot find the optimal instruction sequence, executing the optimal instruction sequence found by the ground control station.

And 210, when the scheme comparison results are consistent, taking the aircraft group scheme in the aircraft group data as an aircraft group collaborative decision scheme corresponding to the emergency data.

In the embodiment of the present invention, the implementation process of step 210 is similar to that of step 105, and will not be repeated here.

In the embodiment of the present invention, as shown in fig. 6, a ground control station scheme, ground control station data, an aircraft group scheme, and aircraft group data are input, and an optimal instruction sequence is output. Specifically, a density peak clustering algorithm is improved, an event pseudo tag is generated, an existing dynamic multi-pooling convolutional neural network model is finely tuned, event extraction is carried out on related data, and the event is mapped into an instruction. And judging whether the ground control station scheme is consistent with the aircraft group scheme or not according to the event extraction result. If the two schemes are not consistent, a determination is made as to whether the aircraft group is authorized to act on its own. If the unauthorized aircraft group can act by itself, the aircraft group and the ground control station respectively call an instruction acquisition algorithm to acquire an optimal instruction sequence. If the aircraft group finds the optimal instruction sequence I ₁ according to the high-priority decision criterion, I ₁ is used for guiding the aircraft group to act. If the aircraft group does not find the optimal command sequence, and the ground control station finds the optimal command sequence I ₂, I ₂ is used for guiding the aircraft group to act. And if the ground control station does not find the optimal instruction sequence, taking the aircraft group scheme as the optimal instruction sequence. If the authorized aircraft group can act by itself, if I ₁ or I ₂ has been found, then I ₁ or I ₂ is executed within a set time; otherwise, taking the aircraft group scheme as an optimal instruction sequence. When the aircraft group scheme is consistent with the ground control station scheme, the aircraft group scheme is taken as an optimal instruction sequence.

Aiming at the characteristics of complexity, uncertainty and few data samples of an actual scene when an aircraft group encounters an emergency, detailed researches are carried out on the types of the emergency, the processing of data, the mode of collaborative decision, the allocation and utilization of computing resources and the like. The event pseudo tag is generated through a preset density peak clustering algorithm, the problem of no event tag is solved, the event extraction model based on CNN is finely adjusted based on the event pseudo tag, then the event is extracted, and the overall situation of the flight environment is rapidly grasped. The ground control station determines whether the aircraft group can move by itself, and whether the aircraft group can move by itself or not, a corresponding scheme is made to cope with the emergency. The decision criteria of high processing priority of the aircraft group and the decision criteria of low processing priority of the ground control station are made reasonable use of, and the calculation resources of the aircraft group and the ground control station are utilized. The instruction sequence is closely related to the flight route and the execution task of the aircraft group, and the improved multi-agent depth deterministic strategy gradient algorithm is adopted to search the instruction sequence, so that the aircraft group avoids threat, completes the task and successfully copes with the emergency by the maximum force. According to the collaborative decision-making method, the emergency situation faced by the aircraft group can be dealt with, the accuracy and the high efficiency of grasping the flight scene can be improved, and scientific decision-making basis can be provided for the actions of the aircraft in the next stage.

As shown in fig. 7, the embodiment of the invention further includes an emergency system applying the method for generating the collaborative decisions of the aircraft group provided by the embodiment of the invention, and the emergency system includes a decision processing module, a data processing module, a flight control module, a sensor module, a database module and a communication module. The hardware facilities of the emergency system comprise three components of an aircraft group, a wireless communication link system and a ground control station. Wherein the fleet of aircraft typically carries a payload, and equipment carried for performing a particular mission. A wireless communication link system for communicating data between the fleet of aircraft and the ground control station. The ground control station controls the aircraft population and performs assignment and coordination of tasks.

The decision processor module is configured to perform an improved density peak clustering algorithm. The data processing module provides computational resources for the decision processor for analyzing the aircraft population flight data to detect quality of flight or to generate a decision scheme. The flight control module is used for a flight management and control system and is responsible for starting a collaborative decision algorithm, executing a calculation result of the decision processor module and transmitting the result to the ground control station through the communication module. The sensor module is used for acquiring data such as speed, altitude, attitude, acceleration, angular rate and the like in the flight process of the aircraft group. The database module is used for storing data acquired by the sensor module, an execution result of the decision processor module, command and capture data of the ground control station, various data transmitted by the aircraft group and data required for maintaining normal flight of the aircraft group.

And the communication module is used for transmitting various data of the aircraft group to the ground control station, and is used for receiving various data transmitted to the aircraft group by the ground control station and transmitting various data transmitted to the ground control station by the aircraft group. The control module is used for conveying control instructions and remotely controlling the aircraft group, interacting with the data processing module, analyzing flight data and generating decision instructions.

Further, the emergency system also comprises a man-machine interaction interface module, wherein the man-machine interaction interface is used for generating a visual interface for an operator to communicate instructions or analyze flight data through the ground control station. The communication module is a wireless communication link and is also used to communicate data between the fleet of aircraft and the ground control station. The sensor module stores the data into the database module, and when the flight control module finds that the data is abnormal, the sensor module indicates that an emergency occurs. The flight control module starts a decision processor module, and the decision processor module executes a collaborative decision algorithm. When the calculation resources of the aircraft group are insufficient, the aircraft group communicates with the ground control station through the wireless communication link system, and the collaborative decision algorithm is jointly completed with the help of the ground control station. And outputting an optimal instruction sequence by the collaborative decision algorithm, and guiding the next-stage action of the aircraft group by the optimal instruction sequence.

Referring to fig. 8, fig. 8 is a block diagram illustrating a collaborative decision-making system for an aircraft group according to a third embodiment of the present invention.

The third embodiment of the invention provides an aircraft group collaborative decision generation system, which comprises:

The event pseudo tag set obtaining module 801 is configured to calculate, when emergency data of an aircraft group is received, an event pseudo tag corresponding to an aircraft group data set corresponding to the emergency data and a ground control station data set generated by a ground control station by using a preset density peak clustering algorithm, thereby obtaining an event pseudo tag set.

The target event extraction model obtaining module 802 is configured to train the initial event extraction model based on the event pseudo tag set to obtain a target event extraction model.

The scheme comparison result obtaining module 803 is configured to obtain a scheme comparison result by respectively performing event extraction and event comparison on the aircraft cluster data set and the ground control station data set through the target event extraction model.

The first obtaining module 804 of the aircraft group collaborative decision is configured to, when the comparison result of the schemes is inconsistent, perform optimal instruction sequence selection according to the authorization data corresponding to the aircraft group and a preset improved multi-agent depth deterministic strategy gradient algorithm, and obtain an aircraft group collaborative decision scheme corresponding to the emergency data.

And a second obtaining module 805 of the aircraft group collaborative decision, configured to use the aircraft group plan in the aircraft group data set as the aircraft group collaborative decision plan corresponding to the emergency data when the plan comparison results are consistent.

Optionally, the event pseudo tag set obtaining module 801 includes:

And the aircraft cluster data set and ground control station data set acquisition module is used for acquiring the aircraft cluster data set generated by the aircraft cluster based on the emergency data and the ground control station data set generated by the ground control station when the emergency data of the aircraft cluster are received.

The characteristic vector set obtaining module is used for extracting characteristic vectors of the aircraft group data set and the ground control station data set through a preset jump model to obtain a characteristic vector set.

The density obtaining module is used for calculating the density corresponding to each point in the feature vector set by adopting a preset density calculation formula to obtain the density.

The preset density calculation formula is:

Where ρ (x _i) represents the density of the point x _i; k represents the number of neighbors of a point, namely the top k points nearest to the point by taking Euclidean distance as a measurement standard; Representing the Euclidean distance between point x _i and point x _j, u represents the dimension of one feature vector set point, i.e., the number of features, g is a subscript used to index the different features; KNN (x _i) represents the set of top k points in the dataset that have the smallest euclidean distance from point x _i; i represents the i-th point x _i in the dataset; j represents the j-th point x _j in the dataset.

The relative density obtaining module is used for substituting the density into a preset relative density calculation formula to calculate the relative density of the points, so as to obtain the relative density.

The preset relative density calculation formula is as follows:

The cluster center obtaining module is used for selecting points with the relative density being larger than the density average value corresponding to the relative density to obtain a plurality of cluster centers.

The set obtaining module is used for dividing the feature vector set by adopting a preset domain calculation formula according to the cluster center to obtain a plurality of sets.

The preset domain calculation formula is:

D_mn(x_p)＝{x_q|x_q∈MN(x_p)∨(x_m∈MN(x_p))∧x_q∈MN(x_m))};

Wherein D _mn(x_p) represents the domain of point x _p; MN (x _p) represents the set of all mutual neighbors of point x _p; MN (x _m) represents the set of all mutual neighbors of point x _m; x _p denotes the p-th point in the feature vector set; x _m denotes the mth point in the feature vector set; x _q denotes the qth point in the feature vector set.

The event pseudo tag set obtaining submodule is used for respectively setting corresponding tags for points corresponding to each set according to the tags corresponding to the cluster center to obtain the event pseudo tag set.

Optionally, the target event extraction model obtaining module 802 includes:

The initial loss function value obtaining module is used for inputting the event pseudo tag set into the initial event extraction model to obtain an initial loss function value.

The intermediate event extraction model obtaining module is used for fine-tuning the initial event extraction model according to the initial loss function value to obtain an intermediate event extraction model.

The objective loss function value obtaining module is used for inputting the event pseudo tag set into the intermediate event extraction model to obtain the objective loss function value.

And the objective loss function value judging module is used for judging whether the objective loss function value is a preset function value or not.

And the target event extraction model is used for taking the intermediate event extraction model corresponding to the target loss function value as a target event extraction model if the target event extraction model is the target event extraction model.

And the jump execution module is used for taking the intermediate event extraction model as an initial event extraction model if not, and jumping to execute the step of inputting the event pseudo tag set into the initial event extraction model to obtain an initial loss function value.

Optionally, the solution comparison result obtaining module 803 includes:

And the aircraft group event and ground event obtaining module is used for respectively carrying out event extraction on the aircraft group data set and the ground control station data set through the target event extraction model to obtain the aircraft group event and the ground event.

The event comparison result obtaining module is used for comparing the aircraft group event with the ground event according to the trigger word category and the parameter similarity to obtain an event comparison result.

The scheme comparison result is obtained as a sub-module, and is used for constructing the scheme comparison result by adopting all event comparison results.

Optionally, the aircraft group collaborative decision first obtaining module 804 includes:

and the authorization data judging module is used for judging whether the authorization data corresponding to the aircraft group is authorization or not when the scheme comparison result is inconsistent.

And the aircraft group collaborative decision-making obtains a first sub-module, and is used for determining an aircraft group collaborative decision-making scheme corresponding to the emergency data according to the instruction sequence and the aircraft group scheme in the preset time if the first sub-module is used for the aircraft group collaborative decision-making.

And the target instruction sequence obtaining module is used for selecting the instruction sequence of the aircraft cluster data set and the ground control station data set according to a preset improved multi-agent depth deterministic strategy gradient algorithm if not, so as to obtain the target instruction sequence.

And the target instruction sequence judging module is used for judging whether the target instruction sequence is a preset sequence or not.

And the aircraft group collaborative decision-making obtains a second sub-module, and if yes, the aircraft group scheme is used as an aircraft group collaborative decision-making scheme corresponding to the emergency data.

And the third sub-module is used for taking the target instruction sequence as an aircraft group collaborative decision scheme corresponding to the emergency data if the third sub-module is not used for the aircraft group collaborative decision.

Optionally, the first sub-module for collaborative decision-making of the aircraft group may perform the following steps:

if not, taking the aircraft group scheme as an aircraft group collaborative decision scheme corresponding to the emergency data.

Alternatively, the target instruction sequence obtaining module may perform the steps of:

According to a priority decision criterion, an instruction acquisition algorithm is adopted to respectively endow the instructions corresponding to the aircraft group data set and the ground control station data set with preset weights, so as to obtain an instruction sequence set;

selecting a plurality of initial instruction sequences meeting a preset score threshold value in an instruction sequence set by adopting a preset improved multi-agent depth deterministic strategy gradient algorithm, and constructing an initial instruction queue;

Selecting an initial instruction sequence with the highest score in an initial instruction queue as an intermediate instruction sequence;

and determining a target instruction sequence according to the self constraint condition and the intermediate instruction sequence of the aircraft group.

Optionally, the target instruction sequence obtaining module may further perform the following steps:

If yes, taking the intermediate instruction sequence as a target instruction sequence;

judging whether the intermediate instruction queue is an empty set or not;

judging whether the target decision criterion set is an empty set or not;

if yes, taking the preset sequence as a target instruction sequence;

If not, taking the target decision criterion set as an initial decision criterion set, and jumping to execute the step of selecting the decision criterion with the highest priority in the initial decision criterion set to obtain the priority decision criterion;

If not, taking the intermediate instruction queue as an initial instruction queue, and jumping to execute the step of selecting the initial instruction sequence with the highest score in the initial instruction queue as the intermediate instruction sequence.

The embodiment of the invention also provides electronic equipment, which comprises: a memory and a processor, the memory storing a computer program; the computer program, when executed by a processor, causes the processor to perform the aircraft-based collaborative decision-making method of any of the embodiments described above.

The memory may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read only memory), an EPROM, a hard disk, or a ROM. The memory has memory space for program code to perform any of the method steps described above. For example, the memory space for the program code may include individual program code for implementing the various steps in the above method, respectively. The program code can be read from or written to one or more computer program products. These computer program products comprise a program code carrier such as a hard disk, a Compact Disc (CD), a memory card or a floppy disk. The program code may be compressed, for example, in a suitable form. The codes, when executed by a computing processing device, cause the computing processing device to perform the steps in the aircraft-based group collaborative decision-making method described above.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.

In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of elements is merely a logical functional division, and there may be additional divisions of actual implementation, e.g., multiple elements or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in essence or a part contributing to the prior art or all or part of the technical solution in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An aircraft group collaborative decision-making method, comprising:

When the scheme comparison results are consistent, taking the aircraft group scheme in the aircraft group data as an aircraft group collaborative decision scheme corresponding to the emergency data;

When the emergency data of the aircraft group is received, calculating an event pseudo tag corresponding to the aircraft group data set corresponding to the emergency data and a ground control station data set generated by a ground control station by adopting a preset density peak clustering algorithm, and obtaining an event pseudo tag set, wherein the event pseudo tag set comprises the following steps:

the preset density calculation formula is as follows:

；

Wherein, Representation pointsIs a density of (3); Representing the number of neighbors of a point, namely the top k points nearest to the point by taking Euclidean distance as a measurement standard; representation points Sum pointThe Euclidean distance between the two points, u represents the dimension of a feature vector concentration point, namely the number of the features, and g is a subscript used for indexing different features; Representing and points in a dataset A set of top k points with minimum Euclidean distance; i represents the i-th point in the dataset; J represents the j-th point in the dataset；

the preset relative density calculation formula is as follows:

；

Wherein, Representation pointsIs a relative density of (2); representation points Is a density of (3); representation points Is a density of (3); Representing the number of neighbors of a point, namely the top k points nearest to the point by taking Euclidean distance as a measurement standard; Representing and points in a dataset A set of top k points with minimum Euclidean distance; i represents the i-th point in the dataset; J represents the j-th point in the dataset；

the preset domain calculation formula is as follows:

；

In the method, in the process of the invention, Representation pointsIs a domain of (2); representation points All mutually adjacent sets; representation points All mutually adjacent sets; representing the p-th point in the feature vector set; Representing the mth point in the feature vector set; representing the q-th point in the feature vector set;

setting corresponding labels for the points corresponding to the sets according to the labels corresponding to the cluster centers to obtain event pseudo-label sets;

The step of training an initial event extraction model based on the event pseudo tag set to obtain a target event extraction model comprises the following steps:

if not, taking the intermediate event extraction model as an initial event extraction model, and jumping to execute the step of inputting the event pseudo tag set into the initial event extraction model to obtain an initial loss function value;

the step of obtaining a scheme comparison result by respectively carrying out event extraction and event comparison on the aircraft cluster data set and the ground control station data set through the target event extraction model comprises the following steps:

adopting all event comparison results to construct scheme comparison results;

When the scheme comparison result is inconsistent, selecting an optimal instruction sequence according to the authorization data corresponding to the aircraft group and a preset improved multi-agent depth deterministic strategy gradient algorithm, and obtaining an aircraft group collaborative decision scheme corresponding to the emergency data, wherein the method comprises the following steps of:

Judging whether the target instruction sequence is a preset sequence or not;

If not, taking the target instruction sequence as an aircraft group collaborative decision scheme corresponding to the emergency data;

the step of selecting the instruction sequence of the aircraft group data set and the ground control station data set according to a preset improved multi-agent depth deterministic strategy gradient algorithm to obtain a target instruction sequence comprises the following steps:

Selecting a plurality of initial instruction sequences which meet a preset score threshold value in the instruction sequence set by adopting a preset improved multi-agent depth deterministic strategy gradient algorithm, and constructing an initial instruction queue, wherein the preset improved multi-agent depth deterministic strategy gradient algorithm refers to N value networks by a central controller, refers to N strategy networks equipped for N agents by a strategy area, and adds a fusion layer between the central controller and the strategy area, wherein the fusion layer is composed of two pseudo-agents, and the actions, the observations and the rewards of all agents on the own side of the first pseudo-agent integration are a group of actions, observations and rewards; the actions, observations and rewards of all agents in the second pseudo-agent integration target are a group of actions, observations and rewards;

2. The method for generating an aircraft group collaborative decision according to claim 1, wherein the step of determining an aircraft group collaborative decision corresponding to the emergency data according to the instruction sequence within a preset time and the aircraft group plan includes:

3. The aircraft group collaborative decision-making method of claim 1, wherein the step of determining a target instruction sequence based on the aircraft group's own constraints and the intermediate instruction sequence comprises:

judging whether the intermediate instruction queue is an empty set or not;

judging whether the target decision criterion set is an empty set or not;

if yes, the preset sequence is used as a target instruction sequence;

4. An aircraft swarm collaborative decision-making system, comprising:

the second obtaining module of the cooperative decision of the aircraft clusters is used for taking the aircraft cluster schemes in the aircraft cluster data as the cooperative decision schemes of the aircraft clusters corresponding to the emergency data when the scheme comparison results are consistent;

the event pseudo tag set obtaining module includes:

The system comprises an aircraft cluster data set and ground control station data set acquisition module, a control station data set acquisition module and a control station data set acquisition module, wherein the aircraft cluster data set acquisition module is used for acquiring an aircraft cluster data set generated by the aircraft cluster based on the emergency data and a ground control station data set generated by a ground control station when the emergency data of the aircraft cluster are received;

The characteristic vector set obtaining module is used for extracting characteristic vectors of the aircraft group data set and the ground control station data set through a preset jump model to obtain a characteristic vector set;

The density obtaining module is used for respectively calculating the densities corresponding to each point in the feature vector set by adopting a preset density calculation formula to obtain the density;

the preset density calculation formula is as follows:

；

The relative density obtaining module is used for substituting the density into a preset relative density calculation formula to calculate the relative density of the points so as to obtain the relative density;

the preset relative density calculation formula is as follows:

；

The cluster center obtaining module is used for selecting points with the relative density larger than the density average value corresponding to the relative density to obtain a plurality of cluster centers;

The set obtaining module is used for dividing the feature vector set by adopting a preset domain calculation formula according to the cluster center to obtain a plurality of sets;

the preset domain calculation formula is as follows:

；

The event pseudo tag set obtaining submodule is used for respectively setting corresponding tags for points corresponding to the sets according to the tags corresponding to the cluster centers to obtain event pseudo tag sets;

the target event extraction model obtaining module comprises:

the initial loss function value obtaining module is used for inputting the event pseudo tag set into an initial event extraction model to obtain an initial loss function value;

The intermediate event extraction model obtaining module is used for finely adjusting the initial event extraction model according to the initial loss function value to obtain an intermediate event extraction model;

The objective loss function value obtaining module is used for inputting the event pseudo tag set into the intermediate event extraction model to obtain an objective loss function value;

The objective loss function value judging module is used for judging whether the objective loss function value is a preset function value or not;

the sub-module is used for taking the intermediate event extraction model corresponding to the target loss function value as a target event extraction model if the target event extraction model is obtained;

The jump execution module is used for taking the intermediate event extraction model as an initial event extraction model if not, and jumping to execute the step of inputting the event pseudo tag set into the initial event extraction model to obtain an initial loss function value;

the scheme comparison result obtaining module comprises:

The aircraft group event and ground event obtaining module is used for respectively carrying out event extraction on the aircraft group data set and the ground control station data set through the target event extraction model to obtain an aircraft group event and a ground event;

the event comparison result obtaining module is used for comparing the aircraft group event with the ground event according to the trigger word category and the parameter similarity to obtain an event comparison result;

The scheme comparison result is obtained as a sub-module, and is used for constructing a scheme comparison result by adopting all the event comparison results;

the first obtaining module of the aircraft group collaborative decision-making comprises:

The authorization data judging module is used for judging whether the authorization data corresponding to the aircraft group is authorization or not when the scheme comparison result is inconsistent;

the aircraft group collaborative decision-making obtains a first sub-module, and is used for determining an aircraft group collaborative decision-making scheme corresponding to the emergency data according to the instruction sequence in the preset time and the aircraft group scheme if yes;

the target instruction sequence obtaining module is used for selecting the instruction sequence of the aircraft group data set and the ground control station data set according to a preset improved multi-agent depth deterministic strategy gradient algorithm if not, so as to obtain a target instruction sequence;

the target instruction sequence judging module is used for judging whether the target instruction sequence is a preset sequence or not;

The aircraft group collaborative decision-making obtains a second sub-module, and if yes, the aircraft group scheme is used as an aircraft group collaborative decision-making scheme corresponding to the emergency data;

The third sub-module is used for taking the target instruction sequence as an aircraft group collaborative decision scheme corresponding to the emergency data if the third sub-module is not used for obtaining the aircraft group collaborative decision;

the target instruction sequence obtaining module executes the following steps:

5. An electronic device comprising a memory and a processor, the memory having stored therein a computer program which, when executed by the processor, causes the processor to perform the steps of the aircraft population collaborative decision-making method of any one of claims 1 to 3.