A rapid firepower planning method for multiple types of missiles based on artificial intelligence

Abstract

The present invention discloses a method for rapid firepower planning of multiple types of ammunition based on artificial intelligence, comprising the following steps: S1: establishing a finite element damage simulation cloud computing platform to calculate the damage effects of multiple types of ammunition; S2: matching multiple types of ammunition with targets based on knowledge graph reasoning; S3: establishing an ammunition consumption planning model based on deep learning theory according to the damage effect calculation result of step S1 and the ammunition-target matching result of step S2, and outputting the rapid firepower planning of multiple types of ammunition. In the present invention, a finite element damage simulation cloud computing platform architecture is established for the key technologies in the rapid firepower planning of multiple types of ammunition, a process for realizing ammunition-target matching based on knowledge graph reasoning is proposed, a deep learning model for ammunition consumption planning is constructed, and a system solution for solving the problem of rapid firepower planning of multiple types of ammunition using cloud computing and artificial intelligence technology is proposed.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-type firepower planning, and in particular to a multi-type rapid firepower planning method based on artificial intelligence. Background Art

[0002] In the current combat methods, there is a method of using weapons with different damage mechanisms to attack the same target. By comprehensively using the different damage effects of various types of weapons, the overall optimal combat effect is achieved. This innovative combat method fully explores the comprehensive combat effectiveness of existing weapons. However, theoretical research on combined strikes is still in its infancy. Traditional optimization methods are difficult to solve problems such as dimensionality explosion and unstructured optimization, and it is difficult to meet the needs of modern combat modes for real-time online firepower planning.

[0003] In the multi-type strike mode, warheads with different effects such as penetration, killing, shock waves, and combustion carry out combined strikes on the target. Predicting and evaluating the composite damage effect is the prerequisite for accurate firepower planning. At present, the calculation of composite damage effect is mostly based on finite element modeling. In "Damage Effect and Numerical Simulation of Ultra-High Performance Cement-based Composite Materials under Penetration and Explosion" (Yang Haoruo, Nanjing University of Science and Technology, 2018.03.), the finite element model was used to numerically simulate the coupled damage effect of penetration and explosion based on ANSYS-AUTODYN software. In "Research on Composite Damage of Discrete Rods and Shock Waves to Supersonic Missiles" (Jiang Yingzi et al., Journal of Missiles, Missiles and Guidance, 2015, 35(3): 63-67.), the damage effect of the composite action of discrete rods and shock waves on supersonic missiles was studied based on LS-DYNA software. In the article "Research on the Damage Effectiveness of Combined Effect Warhead on the Top Armor of Tanks and Armored Vehicles" (Li Yupin, North University of China, 2019.05), the damage law of combined effect warhead on the top armor of tanks and armored vehicles with different structural parameters was studied based on LS-DYNA software.

[0004] The finite element method essentially uses the idea of ​​discretization, variational principle, continuum partitioning and piecewise interpolation, and its calculation process is very complicated. This method can accurately model and calculate specific targets, but it is difficult to simulate large system targets and it is difficult to traverse all target types that may be hit in combat operations. The current research on the prediction and evaluation of composite damage effects only stays at the material level and target component level, and cannot support the prediction and evaluation of the damage effects of multiple types of combined strikes at the campaign-level firepower planning level. Summary of the invention

[0005] In view of the above-mentioned problems, the present invention aims to provide a method for rapid firepower planning of multiple types of ammunition based on artificial intelligence, which uses cloud computing, knowledge graph reasoning and deep learning technology to perform rapid firepower planning of multiple types of ammunition, and provides technical support for building an intelligent firepower planning system.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] The method for rapid firepower planning of multiple types of ammunition based on artificial intelligence is characterized by comprising the following steps:

[0008] S1: Establish a finite element damage simulation cloud computing platform to calculate the damage effects of multiple types of missiles;

[0009] S2: Matching of projectiles and targets of multiple types of projectiles based on knowledge graph reasoning;

[0010] S3: Based on deep learning theory, according to the damage effect calculation results of step S1 and the projectile-target matching results of step S2, a bullet consumption planning model is established to output the rapid firepower planning of multiple types of bullets.

[0011] Furthermore, the mathematical description of the multi-type rapid firepower planning is: given m types of bullets, the i-th type of bullet has NM i (i=1,…,m) pieces, the total amount of ammunition is NM pieces, and a combined attack is carried out on n targets. Multi-group MBHL ij represents the firepower allocation plan of the i-th missile to the j-th target, then Where, d ij It represents the amount of ammunition allocated to the i-th type of ammunition to attack the j-th target, which is a positive integer or 0, where i = 1, ..., m; j = 1, ..., n;

[0012] For a given type and quantity of ammunition, the decision variables of the firepower planning problem can be expressed as the multi-tuple MBHL ij The set HL is used to describe, In the multi-tuple matrix HL, there are m×n multi-tuples, each multi-tuple Contains 1 integer variable, and 2×d ij real number variables;

[0013] The overall damage effect of the target in the multi-type firepower planning problem can be expressed as the functional of the attribute information of each target, the attribute information of each type of ammunition, and the mixed firepower allocation scheme HL, max P = f(MB 1 , …, MB n , DZ 1 , …, DZ m , HL); where MB i represents the target parameters required to calculate the damage effect, i = 1, ..., n; DX jRepresents the weapon parameters required to calculate the damage effect, j=1,…,m.

[0014] Further, the finite element damage simulation cloud computing platform described in step S1 includes a firepower planning user layer, a cloud platform application layer, a resource scheduling layer, a model service layer and an infrastructure layer;

[0015] The firepower planning user layer provides an operation interface for firepower planning auxiliary decision-makers, sends damage calculation requests to the cloud computing platform through the firepower planning user layer, and adjusts various parameters of weapons and targets as needed;

[0016] The cloud platform application layer includes model creation, damage simulation and data management. According to the damage calculation requirements of the firepower planning user layer, the finite element analysis service is dynamically combined to respond to user requests, perform task decomposition, calculate the damage effects of multiple types of ammunition, and provide a basis for resource scheduling.

[0017] The resource scheduling layer allocates sufficient virtual resources for implementing finite element-based damage calculations to improve calculation efficiency;

[0018] The model service layer includes finite element meshing, simulation algorithm services and dynamic model services provided by the cloud computing platform, and implements finite element-based damage calculation as a service in the form of functional modules;

[0019] The infrastructure layer includes computing hosts, storage devices, and network devices.

[0020] Furthermore, the specific operation of calculating the damage effects of multiple types of missiles at the cloud platform application layer includes the following steps:

[0021] S11: Divide the hit area Ω into M sub-areas Δσ according to the shape and physical characteristics of the warhead hit area m (m=1,2,…,M), which is a unit;

[0022] S12: number the elements and nodes, determine the relationship between them, and list the natural boundaries and corresponding boundary values;

[0023] S13: According to the number of regional unit nodes and the requirements for calculating approximate solutions, a corresponding interpolation function is selected as a unit shape function, and the solution function of each regional unit is approximated by a linear combination expression of the unit shape function, and then the approximate function is substituted into the integral equation, and the unit area is integrated to obtain an algebraic equation containing undetermined coefficients, which is a unit finite element equation group; the undetermined coefficients include target structural parameters, target material parameters and target functional parameters;

[0024] S14: according to the boundary conditions, an appropriate numerical calculation method in the finite element method is used to solve the problem, that is, to obtain the coefficient value of each node, that is, the value of the undetermined coefficient;

[0025] S15: Warhead hits Δσ m The target damage degree D(m) caused is determined by the damage degree and vulnerability analysis results of each target unit, and the calculation formula is: Where K is the number of target units; S(m, k) is the damage degree of the kth target unit, and A(k) is the impact factor of the kth target unit;

[0026] Among them, the warhead hits Δσ m The calculation expression for the damage degree S(m, k) of the target unit is S(m, k) = S{Z(0, k), Y, C, O}, where Z(0, k) is the damage criterion of the kth target unit in an intact state; Y is the penetration effect parameter; C is the shock wave effect parameter; O is the vibration effect parameter;

[0027] The calculation expression of the influence factor A(k) of the kth target unit is A(k)=[A C (C T , M T ) k , A F (F T , M T ) k ], where A C (C T , M T ) k is the structural impact factor of the kth target unit; A F (F T , M T ) k is the functional impact factor of the kth target unit; C T is the target structure parameter; M T is the target material parameter, F T is the target function parameter.

[0028] Furthermore, the specific operation of performing bullet-target matching based on knowledge graph reasoning in step S2 includes the following steps:

[0029] S21. Knowledge acquisition: Use automated extraction technology to extract target entities, types, characteristics, damage effects, and weapon application principles from structured, semi-structured, and unstructured data from expert experience, combat simulation experiments, field damage tests, and typical combat cases to build a basic knowledge base;

[0030] S22, knowledge fusion: structurally process the basic knowledge base constructed in step S21 through heterogeneous data integration, knowledge importance calculation and verification, and establish a bullet-target matching rule base;

[0031] S23, Knowledge Storage: Adopt graph-based data structure, design flexible and scalable storage mode, realize effective storage of knowledge base and rule base, and realize dynamic management;

[0032] S24, query analysis: predict the user's query intention, classify according to feature words, identify the weapons and target attributes involved in the user's query, generate standard analysis sentences, and analyze the battlefield environment and combat intention of the user's query;

[0033] S25, knowledge reasoning: Based on the knowledge base and rule base in the knowledge graph, use reasoning models and algorithms to infer the answers to user queries and give a ranking of bullet-target matching solutions.

[0034] Furthermore, the specific operations of step S25 include:

[0035] S251: Establish a dynamic key-value memory network model based on the graph structure knowledge base and rule base constructed by the knowledge graph;

[0036] S252: Divide the external input data into two parts: "key" and "value" {(k i , v i )} i＝1 N As the memory of the network, a static key matrix key is used to store all target attributes, battlefield environment and purpose intention information, and a dynamic value matrix value is used to store and update weapon matching results. The key is responsible for addressing and scoring the relevance of memory and problem, while the value is responsible for reading and performing weighted summation of the memory values ​​to obtain the output, thereby realizing the reasoning function.

[0037] Further, the specific operation of implementing the reasoning function in step S252 includes the following steps:

[0038] S2521: Key preprocessing, each triple {s i , p i , o i} in the subject entity and predicate s i +p i Corresponding to k i , object entity o i Corresponding to v i , convert the knowledge base and rule base into the form of (k, v);

[0039] S2522: Select a subset from (k, v), the selection criteria of the subset include: k has common words with the input question, and the common words are not stop words; sort by the number of common words, select the top N items, and construct a corresponding memory unit vector for each input question, and input the memory unit vector m i =Aφ(ki ), output memory cell vector c i =Cφ(v i );

[0040] S2523: Key Addressing: Based on the question, a probability distribution is calculated for all keys in the memory unit, and the memory unit is scored for relevance. The probability distribution is obtained by multiplying the key memory with the input question and then performing Softmax. The size of the probability indicates the degree of relevance.

[0041]

[0042] S2524: ValueReading: Calculate the weighted vector of value based on the probability of key.

[0043]

[0044] S2525: QueryUpdating, adding the input vector o to the vector representation q of the query input question, and then performing R i The matrix linearly maps the input j+1 =R j (q j +o);

[0045] S2526: Use q j+1 Replace the question vector in Key Addressing as the input of the next layer and update it iteratively;

[0046] S2527: Prediction, after H iterations, q H+1 Input into the classifier, predict the answer, and sort the answer ph i =Softmax(q T j+1 AΦ K (k hi )).

[0047] Furthermore, in step S3, based on the deep learning theory, a bullet consumption planning model is established, and the specific operation of outputting the rapid firepower planning of multiple types of bullets includes the following steps:

[0048] S31: Based on expert experience, simulation, exercise tests and actual combat examples, a sample library of target attributes, weapon attributes, battlefield environment, combat intent, projectile-target matching results, and ammunition consumption of various types of ammunition is established;

[0049] S32: Divide the samples in the sample library into a training data set and a test data set, use a deep neural network for learning and training, and establish a multi-type missile strike ammunition planning model based on the missile-target matching results based on knowledge graph reasoning and the damage effect calculation results based on cloud computing technology;

[0050] S33: Input the target attributes, weapon attributes, battlefield environment and combat intentions input by the firepower planning user layer into the ammunition planning model for multi-type strikes established in step S32, and output the ammunition-target matching plan, the ammunition consumption of each type of ammunition and the predicted value of the damage effect.

[0051] The beneficial effects of the present invention are:

[0052] 1. The present invention integrates finite element analysis and cloud computing to calculate the damage effects in the firepower planning of multiple types of ammunition, which can overcome the difficulties faced by finite element calculation of damage effects, such as complex and huge calculation process, high professionalism, high cost of finite element software, and dependence on high-performance computers.

[0053] 2. The bullet-target matching method based on knowledge graph reasoning in the present invention enables the computer to understand the user's intention from a semantic perspective and output a bullet-target matching solution quickly and intelligently.

[0054] 3. The deep learning model of ammunition consumption planning improved in the present invention is an improvement on the traditional heuristic fire planning algorithm represented by genetic algorithm and particle swarm algorithm. It can not only meet the fast calculation requirements of nonlinear problems, but also avoid the dependence of multi-attribute decision-making on expert-set weights, and provide a technical solution for fast and accurate fire planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is the architecture diagram of the finite element damage simulation cloud computing platform in the present invention;

[0056] Figure 2 This is the implementation process of bullet-target matching based on knowledge graph reasoning in the present invention;

[0057] Figure 3 It is the knowledge graph reasoning model in the present invention;

[0058] Figure 4 A deep learning model for the ammunition consumption planning in the present invention;

[0059] Figure 5 This is a curve showing how the target damage effect changes over time in the first embodiment of the present invention.

[0060] Figure 6 This is a scatter diagram of the damage effect in the mixed firepower distribution scheme obtained by the enumeration method in the first embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0062] The method for rapid firepower planning of multiple types of missiles based on artificial intelligence includes the following steps:

[0063] S1: Establish a finite element damage simulation cloud computing platform to calculate the damage effects of multiple types of missiles;

[0064] The typical mathematical description of multi-type rapid firepower planning is: given m types of bullets, the i-th type of bullet has NM i (i=1,…,m) pieces, the total amount of ammunition is NM pieces, and a combined attack is carried out on n targets. Multi-group MBHL ij represents the firepower allocation plan of the i-th missile to the j-th target, then Where, d ij It represents the amount of ammunition allocated to the i-th type of ammunition to attack the j-th target, which is a positive integer or 0, where i = 1, ..., m; j = 1, ..., n;

[0065] For a given type and quantity of ammunition, the decision variables of the firepower planning problem can be expressed as the multi-tuple MBHL ij The set HL is used to describe, In the multi-tuple matrix HL, there are m×n multi-tuples, each multi-tuple Contains 1 integer variable, and 2×d ij real number variables; its variable dimension is very high. If we directly search for the optimal solution on the definition domain of all variables, we will fall into the dilemma of "dimensionality explosion". For example, there are 3 types of bullets, A, B, and C, each with 6 bullets. We carry out combined fire strikes on 8 targets without considering the optimization calculation of the aiming point of each bullet on the target. Firepower distribution scheme.

[0066] The goal of optimizing the firepower planning problem can usually be described as making full use of existing combat resources to pursue the best damage effect on the target. In the mode of multiple types of ammunition hitting multiple targets, the goal is to achieve the best overall damage effect on the target. Its objective function can be expressed as a functional of the attribute information of each target, the attribute information of each type of ammunition, and the mixed firepower allocation scheme HL, max P = f(MB 1 , …, MB n , DZ 1 …, DZ m , HL); where MB i represents the target parameters required to calculate the damage effect, i = 1, ..., n; DX j Represents the weapon parameters required to calculate the damage effect, j=1,…,m.

[0067] Furthermore, the finite element damage simulation cloud computing platform includes a firepower planning user layer, a cloud platform application layer, a resource scheduling layer, a model service layer and an infrastructure layer, as shown in the attached Figure 1 shown.

[0068] The firepower planning user layer provides an operation interface for firepower planning auxiliary decision-makers, sends damage calculation requests to the cloud computing platform through the firepower planning user layer, and adjusts various parameters of weapons and targets as needed;

[0069] The cloud platform application layer includes model creation, damage simulation and data management. According to the damage calculation requirements of the firepower planning user layer, the finite element analysis service is dynamically combined to respond to user requests, perform task decomposition, calculate the damage effects of multiple types of ammunition, and provide a basis for resource scheduling.

[0070] The resource scheduling layer allocates sufficient virtual resources for implementing finite element-based damage calculations to improve calculation efficiency;

[0071] The model service layer includes finite element meshing, simulation algorithm services and dynamic model services provided by the cloud computing platform, and implements finite element-based damage calculation as a service in the form of functional modules;

[0072] The infrastructure layer includes computing hosts, storage devices, and network devices.

[0073] Furthermore, the specific operation of calculating the damage effects of multiple types of missiles at the cloud platform application layer includes the following steps:

[0074] S11: Divide the hit area Ω into M sub-areas Δσ according to the shape and physical characteristics of the warhead hit area m (m=1,2,…,M), which is a unit;

[0075] S12: number the elements and nodes, determine the relationship between them, and list the natural boundaries and corresponding boundary values;

[0076] S13: According to the number of regional unit nodes and the requirements for calculating approximate solutions, a corresponding interpolation function is selected as a unit shape function, and the solution function of each regional unit is approximated by a linear combination expression of the unit shape function, and then the approximate function is substituted into the integral equation, and the unit area is integrated to obtain an algebraic equation containing undetermined coefficients, which is a unit finite element equation group; the undetermined coefficients include target structural parameters, target material parameters and target functional parameters;

[0077] S14: according to the boundary conditions, an appropriate numerical calculation method in the finite element method is used to solve the problem, that is, to obtain the coefficient value of each node, that is, the value of the undetermined coefficient;

[0078] S15: Warhead hits Δσ m The target damage degree D(m) caused is determined by the damage degree and vulnerability analysis results of each target unit, and the calculation formula is: Where K is the number of target units; s(m, k) is the damage degree of the kth target unit, and A(k) is the impact factor of the kth target unit;

[0079] Among them, the warhead hits Δσ m The calculation expression for the damage degree S(m, k) of the target unit is s(m, k) = S{Z(0, k), Y, C, O}, where Z(0, k) is the damage criterion of the kth target unit in an intact state; Y is the penetration effect parameter (including penetration depth, damage area, crater radius, deformation curvature, etc.); C is the shock wave effect parameter (including overpressure peak value, overpressure duration, etc.); O is the vibration effect parameter (including acceleration, vibration frequency, etc.);

[0080] The calculation expression of the influence factor A(k) of the kth target unit is A(k)=[A C (C T , M T ) k , A F (F T , M T ) k ], where A C (C T , M T ) k is the structural impact factor of the kth target unit; A F (F T , M T ) k is the functional impact factor of the kth target unit; C T is the target structure parameter; M T are the target material parameters (including density, elastic modulus, shear modulus, Poisson's ratio, tensile and compressive failure stress, etc.), F T is the target function parameter.

[0081] The integration of finite element analysis and cloud computing can overcome the difficulties faced by finite element calculation of damage effects, such as complex and huge calculation process, high professionalism, high cost of finite element software, and dependence on high-performance computers.

[0082] Further, step S2: performing projectile-target matching of multiple projectile types based on knowledge graph reasoning;

[0083] The knowledge graph was created to meet the demand for knowledge in artificial intelligence. It maps user query requests with entities, attributes, and relationships in the semantic knowledge base. The knowledge graph has the characteristics of logical reasoning, heterogeneous information association, knowledge visualization, and interpretability. Supported by a huge database, it enables computers to understand user intentions from a semantic perspective and intelligently output user query results.

[0084] For the matching of projectiles and targets in firepower planning of multiple types of missiles, the knowledge areas that need to be identified include multi-dimensional knowledge such as single-type damage effect dimension, combined damage effect dimension, target entity dimension, target type dimension, target characteristic dimension, and weapon application dimension. The single-type damage effect dimension identification knowledge includes penetration effect, lethal effect, shock wave effect, and combustion effect. The combined damage effect dimension includes lethal blasting, penetration blasting, and blasting combustion. The target entity dimension includes target type, name, and location. The target characteristic dimension includes target material, composition, structure, and function. The weapon application dimension includes penetrating bombs and explosive bombs, explosive bombs and incendiary bombs, and armor-piercing bombs and incendiary bombs. This knowledge includes application rules or typical case classes, and appears in structured, semi-structured, and unstructured forms. In an environment where the scale of these combat data is growing exponentially, reasoning based on knowledge graphs provides a solution for intelligent firepower planning projectile matching.

[0085] As attached Figure 2 As shown in the figure, the specific operation of bullet-target matching based on knowledge graph reasoning includes the following steps:

[0086] S21. Knowledge acquisition: Use automated extraction technology to extract target entities, types, characteristics, damage effects, and weapon application principles from structured, semi-structured, and unstructured data from expert experience, combat simulation experiments, field damage tests, and typical combat cases to build a basic knowledge base;

[0087] S22, knowledge fusion: structurally process the basic knowledge base constructed in step S21 through heterogeneous data integration, knowledge importance calculation and verification, and establish a bullet-target matching rule base;

[0088] S23, Knowledge Storage: Adopt graph-based data structure, design flexible and scalable storage mode, realize effective storage of knowledge base and rule base, and realize dynamic management;

[0089] S24, query analysis: predict the user's query intention, classify according to feature words, identify the weapons and target attributes involved in the user's query, generate standard analysis sentences, and analyze the battlefield environment and combat intention of the user's query;

[0090] S25. Knowledge reasoning: Based on the knowledge base and rule base in the knowledge graph, use the reasoning model and algorithm to infer the answer to the user's query and give a ranking of the bullet-target matching scheme. The reasoning model is shown in the attached figure. Figure 3 shown.

[0091] Furthermore, the specific operations of knowledge reasoning include:

[0092] S251: Establish a dynamic key-value memory network model based on the graph structure knowledge base and rule base constructed by the knowledge graph;

[0093] S252: Divide the external input data into two parts: "key" and "value" {(k i , v i )} i＝1 N As the memory of the network, a static key matrix key is used to store all target attributes, battlefield environment and purpose intention information, and a dynamic value matrix value is used to store and update weapon matching results. The key is responsible for addressing and scoring the relevance of memory and problem, while the value is responsible for reading and performing weighted summation of the memory values ​​to obtain the output, thereby realizing the reasoning function.

[0094] The specific algorithm process includes:

[0095] S2521: Key preprocessing, each triple {s i , p i , o i} in the subject entity and predicate s i +p i Corresponding to k i , object entity o i Corresponding to v i , convert the knowledge base and rule base into the form of (k, v);

[0096] S2522: Select a subset from (k, v), the selection criteria of the subset include: k has common words with the input question, and the common words are not stop words; sort by the number of common words, select the top N items, and construct a corresponding memory unit vector for each input question, and input the memory unit vector m i =Aφ(k i ), output memory cell vector c i =Cφ(v i );

[0097] S2523: Key Addressing: Based on the question, a probability distribution is calculated for all keys in the memory unit, and the memory unit is scored for relevance. The probability distribution is obtained by multiplying the key memory with the input question and then performing Softmax. The size of the probability indicates the degree of relevance.

[0098]

[0099] S2524: ValueReading: Calculate the weighted vector of value based on the probability of key.

[0100]

[0101] S2525: QueryUpdating, adding the input vector o to the vector representation q of the query input question, and then performing R i The matrix linearly maps the input j+1 =R j (q j +o);

[0102] S2526: Use q j+1 Replace the question vector in Key Addressing as the input of the next layer and update it iteratively;

[0103] S2527: Prediction, after H iterations, q H+1 Input into the classifier, predict the answer, and sort the answer ph i =Softmax(q T j+1 AΦ K (k hi )).

[0104] Furthermore, step S3: based on deep learning theory, according to the damage effect calculation result of step S1 and the projectile-target matching result of step S2, a bullet consumption planning model is established to output the rapid firepower planning of multiple types of bullets.

[0105] After the ammunition-target matching plan is determined, the complete firepower plan also includes the allocation of ammunition of each type of ammunition to the target, that is, ammunition consumption planning. Under the attack of multiple types of ammunition, this problem is a multi-dimensional nonlinear search and multi-attribute decision-making problem. There is no mathematical analytical expression available for calculation, and it is difficult to find the optimal solution. Informationized warfare has put forward higher requirements on the real-time nature of firepower planning. To implement scientific auxiliary decision-making, it is necessary to use artificial intelligence technology and fully tap into a large amount of simulation and simulation data based on the full use of existing experience. Use deep learning methods to learn training samples and map the complex nonlinear relationship between output and input. The ammunition consumption planning model based on deep learning is shown in the attached figure. Figure 4As shown, the specific operation includes the following steps:

[0106] S31: Based on expert experience, simulation, exercise tests and actual combat examples, a sample library of target attributes, weapon attributes, battlefield environment, combat intent, projectile-target matching results, and ammunition consumption of various types of ammunition is established;

[0107] S32: Divide the samples in the sample library into a training data set and a test data set, use a deep neural network for learning and training, and establish a multi-type missile strike ammunition planning model based on the missile-target matching results based on knowledge graph reasoning and the damage effect calculation results based on cloud computing technology;

[0108] S33: Input the target attributes, weapon attributes, battlefield environment and combat intentions input by the firepower planning user layer into the ammunition planning model for multi-type strikes established in step S32, and output the ammunition-target matching plan, the ammunition consumption of each type of ammunition and the predicted value of the damage effect.

[0109] The ammunition consumption planning based on deep learning proposed in this invention is an improvement on the traditional heuristic fire planning algorithm represented by genetic algorithm and particle swarm algorithm. It can not only meet the fast calculation requirements of nonlinear problems, but also avoid the dependence of multi-attribute decision-making on expert-set weights.

[0110] Embodiment 1:

[0111] Assuming that there are 15 missile weapons of 9 different types, firepower planning is implemented for 8 targets to be hit at the airport. The target types and missile types and quantities are shown in Tables 1 and 2 respectively. The objective function of firepower allocation optimization is the optimal overall effect.

[0112] Table 1 Optional strike targets

[0113] Sub-goal number Sub-goal name Sub-goal number Sub-goal name 1 Control Tower 5 Radar Station 2 Oil depot 6 Taxiway 3 Helipad 7 Flight Control Room 4 track 8 Ammunition depot

[0114] Table 2 Bullet type and quantity table

[0115] Bullet Type Number Bullet Type Ammo I Type I Delayed Bomb 3 II Type II Penetrator 2 III Type III explosive round 2 IV Type IV integral projectile 1 V V-shaped penetrator 1 VI Type VI bomb 1 VII Type VII Explosive Bomb 2 VIII Type VIII Penetrator 2 IX Type IX bomb 1

[0116] For any given firepower allocation scheme, as shown in Table 3 below, the overall damage effect of the target can be calculated based on the damage effect calculation model. Figure 5 shown.

[0117] Table 3 Multiple types of ammunition firepower allocation scheme

[0118]

[0119] According to the knowledge graph reasoning model of missile-target matching, the target numbers and rankings suitable for the nine types of missiles are shown in Table 4.

[0120] Table 4 List of projectile types, ammunition quantities and sub-targets suitable for attack by each projectile type

[0121] Bullet Type Number Bullet Type Numbering and ranking of sub-targets suitable for attack I Type I Delayed Bomb 4，3，1，6 II Type II Penetrator 4，3，1，6 III Type III explosive round 2，8 IV Type IV integral projectile 3，2，8，5，7，1 V V-shaped penetrator 4，3，1，6 VI Type VI bomb 3，2，8，5，7，1 VII Type VII Explosive Bomb 3，5，1 VIII Type VIII Penetrator 4，3，1，6 IX Type IX bomb 3，2，8，5，7，1

[0122] Based on the ammunition consumption planning model, a multi-type firepower allocation plan is given under the optimal condition of the overall damage effect of the airport, as shown in Table 5. The overall damage effect of this plan on the target is 0.884.

[0123] Table 5 Optimal firepower allocation scheme for multiple missile types

[0124]

[0125] In order to verify the advancement of the model, 600 mixed firepower allocation schemes are randomly enumerated, and the scheme with the best target damage effect is recorded. The corresponding damage effect is HMJ i Repeat the above process 500 times and calculate the HMJ i (i=1,…,500) is drawn into a scatter plot, as shown in the attached figure. Figure 6 shown.

[0126] Analysis of calculation results:

[0127] As attached Figure 6 As shown in the figure, the average damage effect of the mixed firepower distribution scheme obtained by the enumeration method is 0.423. Among the 500 enumerations, the effect index value exceeded 0.8 22 times, accounting for 4.4%, and exceeded 0.88 2 times, accounting for 0.4%. This shows that under similar calculation scale conditions, the reliability of using the enumeration method to obtain the same effect as the optimization model is 0.4%.

[0128] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A multi-type rapid firepower planning method based on artificial intelligence, characterized in that: The following steps are included: S1: Establish a finite element damage simulation cloud computing platform to calculate the damage effects of multiple types of missiles; S2: Match multiple types of bullets with different targets based on knowledge graph reasoning; use reasoning models and algorithms to infer answers to user queries based on the knowledge base and rule base in the knowledge graph, and give a ranking of bullet-target matching solutions; S3: Based on the deep learning theory, according to the damage effect calculation result of step S1 and the projectile-target matching result of step S2, an ammunition consumption planning model is established, and the target attributes, weapon attributes, battlefield environment and combat intention are input into the ammunition consumption planning model, and the projectile-target matching scheme, the ammunition consumption and damage effect prediction values ​​of each type of ammunition are output, and the rapid firepower planning of multiple types of ammunition is output; The specific operation of calculating the damage effects of multiple types of missiles includes the following steps: S11: Divide the hit area Ω into M sub-areas Δσ according to the shape and physical characteristics of the warhead hit area m (m=1,2,…,M), which is a unit; S12: number the elements and nodes, determine the relationship between them, and list the natural boundaries and corresponding boundary values; S13: According to the number of regional unit nodes and the requirements for calculating approximate solutions, a corresponding interpolation function is selected as a unit shape function, and the solution function of each regional unit is approximated by a linear combination expression of the unit shape function, and then the approximate function is substituted into the integral equation, and the unit area is integrated to obtain an algebraic equation containing undetermined coefficients, which is a unit finite element equation group; the undetermined coefficients include target structural parameters, target material parameters and target functional parameters; S14: according to the boundary conditions, an appropriate numerical calculation method in the finite element method is used to solve the problem, that is, to obtain the coefficient value of each node, that is, the value of the undetermined coefficient; S15: Warhead hits Δσ m The target damage degree D(m) caused is determined by the damage degree and vulnerability analysis results of each target unit, and the calculation formula is: Where K is the number of target units; S(m, k) is the damage degree of the kth target unit, and A(k) is the impact factor of the kth target unit; Among them, the warhead hits Δσ m The calculation expression for the damage degree S(m, k) of the target unit is S(m, k) = S{Z(0, k), Y, C, O}, where Z(0, k) is the damage criterion of the kth target unit in an intact state; Y is the penetration effect parameter; C is the shock wave effect parameter; O is the vibration effect parameter; The calculation expression of the influence factor A(k) of the kth target unit is: In the formula, is the structural impact factor of the kth target unit; A F (F T , M T ) k is the functional impact factor of the kth target unit; C T is the target structure parameter; M T is the target material parameter, F T is the target function parameter.

2. The method for rapid firepower planning of multiple types of ammunition based on artificial intelligence according to claim 1, characterized in that: The mathematical description of the multi-type rapid firepower planning is: given m types of bullets, the i-th type of bullet has NM i (i=1,…,m) pieces, the total amount of ammunition is NM pieces, and a combined attack is carried out on n targets; multi-group MBHL ij represents the firepower allocation plan of the i-th missile to the j-th target, then Where, d ij It represents the amount of ammunition allocated to the i-th type of ammunition to attack the j-th target, which is a positive integer or 0, where i = 1, ..., m; j = 1, ..., n; For a given type and quantity of ammunition, the decision variables of the firepower planning problem can be expressed as the multi-tuple MBHL ij The set HL is used to describe, In the multi-tuple matrix HL, there are m×n multi-tuples, each multi-tuple Contains 1 integer variable, and 2×d ij real number variables; The overall damage effect of the target in the multi-type firepower planning problem can be expressed as the functional of the attribute information of each target, the attribute information of each type of ammunition, and the mixed firepower allocation scheme HL, max P = f(MB1, ..., MB n ,DZ1,…,DZ m , HL); where MB i represents the target parameters required to calculate the damage effect, i = 1, ..., n; DX j Represents the weapon parameters required to calculate the damage effect, j=1,…,m.

3. The method for rapid firepower planning of multiple types of ammunition based on artificial intelligence according to claim 2 is characterized in that: The finite element damage simulation cloud computing platform described in step S1 includes a firepower planning user layer, a cloud platform application layer, a resource scheduling layer, a model service layer and an infrastructure layer; The firepower planning user layer provides an operation interface for firepower planning auxiliary decision-makers, sends damage calculation requests to the cloud computing platform through the firepower planning user layer, and adjusts various parameters of weapons and targets as needed; The cloud platform application layer includes model creation, damage simulation and data management. According to the damage calculation requirements of the firepower planning user layer, the finite element analysis service is dynamically combined to respond to user requests, perform task decomposition, calculate the damage effects of multiple types of ammunition, and provide a basis for resource scheduling. The resource scheduling layer allocates sufficient virtual resources for implementing finite element-based damage calculations to improve calculation efficiency; The model service layer includes finite element meshing, simulation algorithm services and dynamic model services provided by the cloud computing platform, and implements finite element-based damage calculation as a service in the form of functional modules; The infrastructure layer includes computing hosts, storage devices, and network devices.

4. The method for rapid firepower planning of multiple types of ammunition based on artificial intelligence according to claim 2 is characterized in that: The specific operation of performing bullet-target matching based on knowledge graph reasoning in step S2 includes the following steps: S21. Knowledge acquisition: Use automated extraction technology to extract target entities, types, characteristics, damage effects, and weapon application principles from structured, semi-structured, and unstructured data from expert experience, combat simulation experiments, field damage tests, and typical combat cases to build a basic knowledge base; S22, knowledge fusion: structurally process the basic knowledge base constructed in step S21 through heterogeneous data integration, knowledge importance calculation and verification, and establish a bullet-target matching rule base; S23, Knowledge Storage: Adopt graph-based data structure, design flexible and scalable storage mode, realize effective storage of knowledge base and rule base, and realize dynamic management; S24, query analysis: predict the user's query intention, classify according to feature words, identify the weapons and target attributes involved in the user's query, generate standard analysis sentences, and analyze the battlefield environment and combat intention of the user's query; S25, knowledge reasoning: Based on the knowledge base and rule base in the knowledge graph, use reasoning models and algorithms to infer the answers to user queries and give a ranking of bullet-target matching solutions.

5. The method for rapid firepower planning of multiple types of ammunition based on artificial intelligence according to claim 4 is characterized in that: The specific operations of step S25 include: S251: Establish a dynamic key-value memory network model based on the graph structure knowledge base and rule base constructed by the knowledge graph; S252: Divide the external input data into two parts: "key" and "value" {(k i , v i )} i＝1 N As the memory of the network, a static key matrix key is used to store all target attributes, battlefield environment and purpose intention information, and a dynamic value matrix value is used to store and update weapon matching results. The key is responsible for addressing and scoring the relevance of memory and problem, while the value is responsible for reading and performing weighted summation of the memory values ​​to obtain the output, thereby realizing the reasoning function.

6. The method for rapid firepower planning of multiple types of ammunition based on artificial intelligence according to claim 5 is characterized in that: The specific operation of implementing the reasoning function in step S252 includes the following steps: S2521: Key preprocessing, each triple {s i , p i , o i } in the subject entity and the predicate s i +p i Corresponding to k i , object entity o i Corresponding to v i , convert the knowledge base and rule base into the form of (k, v); S2522: Select a subset from (k, v), the selection criteria of the subset include: k has common words with the input question, and the common words are not stop words; sort by the number of common words, select the top N items, and construct a corresponding memory unit vector for each input question, and input the memory unit vector m i =Aφ(k i ), output memory cell vector c i =Cφ(v i ); S2523: Key Addressing: Based on the question, a probability distribution is calculated for all keys in the memory unit, and the memory unit is scored for relevance. The probability distribution is obtained by multiplying the key memory with the input question and then performing Softmax. The size of the probability indicates the degree of relevance. S2524: ValueReading: Calculate the weighted vector of value based on the probability of key. S2525: QueryUpdating, adding the input vector o to the vector representation q of the query input question, and then performing R i The matrix linearly maps the input j+1 =R j (q j +o); S2526: Use q j+1 Replace the question vector in Key Addressing as the input of the next layer and update it iteratively; S2527: Prediction, after H steps of iteration, q H+1 Input into the classifier, predict the answer, and sort the answer 7. The method for rapid firepower planning of multiple types of ammunition based on artificial intelligence according to claim 2 is characterized in that: In step S3, based on the deep learning theory, a bullet consumption planning model is established, and the specific operation of outputting the rapid firepower planning of multiple types of bullets includes the following steps: S31: Based on expert experience, simulation, exercise tests and actual combat examples, a sample library of target attributes, weapon attributes, battlefield environment, combat intent, projectile-target matching results, and ammunition consumption of various types of ammunition is established; S32: Divide the samples in the sample library into a training data set and a test data set, use a deep neural network for learning and training, and establish a multi-type missile strike ammunition planning model based on the missile-target matching results based on knowledge graph reasoning and the damage effect calculation results based on cloud computing technology; S33: Input the target attributes, weapon attributes, battlefield environment and combat intentions input by the firepower planning user layer into the ammunition planning model for multi-type strikes established in step S32, and output the ammunition-target matching plan, the ammunition consumption of each type of ammunition and the predicted value of the damage effect.

Images