Multi-target game-oriented resource iterative allocation method

Abstract

The invention discloses a multi-target game-oriented resource iterative allocation method. The method comprises the steps of obtaining a to-be-gamed task information set and to-be-allocated resource information; the to-be-allocated resource information comprises the number of to-be-allocated resources, the position information of the to-be-allocated resources, the bandwidth of the to-be-allocated resources and the coverage area; performing priority evaluation processing on the to-be-gamed task information to obtain a weighting coefficient vector; constructing an initial resource allocation model by using the to-be-gamed task information, the to-be-allocated resource information and the weighting coefficient vector; solving the initial resource allocation model to obtain an initial resource allocation matrix; obtaining a resource historical allocation data set; constructing a task channel transfer model by using the resource historical allocation data set; and processing the initial resource allocation information by using the task channel transfer model to obtain resource allocation scheme information.

Description

Technical Field

[0001] The present invention relates to the field of swarm intelligence, and in particular to a resource iterative allocation method for multi-objective games. Background Art

[0002] At present, cluster intelligence is one of the important development directions of unmanned systems in the future. Unmanned clusters that form scale effects can solve the problems of single functions, poor reliability, and low intelligence of existing single systems, and can complete more complex and difficult tasks and work. In this context, the game of multi-objective cluster systems has become a difficult problem. Since multi-objective clusters have advantages such as intelligent collaboration and adaptive optimization, effective game play has become a technical problem.

[0003] For the game problem of multi-objective cluster system, how to optimize the allocation of existing game resources and form the allocation strategy of game resources quickly and accurately is an urgent problem to be solved. Summary of the invention

[0004] The present invention mainly solves the problem of how to optimize the allocation of existing game resources and quickly and accurately form a game resource allocation strategy for multi-objective cluster systems. The present invention discloses a resource iterative allocation method for multi-objective games.

[0005] In a first aspect, an embodiment of the present application discloses a resource iterative allocation method for a multi-objective game, comprising:

[0006] S1, obtaining a set of information on tasks to be played and information on resources to be allocated; the set of information on tasks to be played includes information on several tasks to be played; the information on tasks to be played includes information on task target locations, task time spent, task bandwidth, task coverage area, and task priority of the tasks to be played; the information on resources to be allocated includes the number of resources to be allocated, location information of resources to be allocated, bandwidth of resources to be allocated, and coverage area;

[0007] S2, performing priority evaluation processing on the task information to be played to obtain a weighted coefficient vector;

[0008] S3, constructing an initial resource allocation model using the information of the game tasks to be gambled, the information of the resources to be allocated and the weighted coefficient vector;

[0009] S4, solving the initial resource allocation model to obtain an initial resource allocation matrix;

[0010] S5, obtaining a resource historical allocation data set;

[0011] S6, constructing a task channel transfer model using the resource historical allocation data set;

[0012] S7, using the task channel transfer model, processing the initial resource allocation information to obtain resource allocation solution information.

[0013] The priority evaluation process is performed on the task information to be played to obtain a weighted coefficient vector, including:

[0014] S21, normalize the priority value of each task to be played, and obtain the priority factor ρ of the task to be played 1i ; For the priority P of the i-th task to be played i , whose priority factor is ρ 1i =P i / (P max -P min ), where P max and P min They represent the upper and lower limits of the priority value range respectively;

[0015] S22, according to the task target location information of each task to be played, calculate the average distance between the task target and the resource to be allocated; use the preset N1 distance ranges to determine the distance range e to which the average distance belongs t , t=1,2,…,N1; each distance range has a corresponding distance coefficient; determine the distance range e to which the average distance belongs t The corresponding distance coefficient τ t , is the distance factor ρ of the game task to be played 2i ρ 2i represents the distance factor of the i-th task to be played;

[0016] S23, divide the task bandwidth information of the task to be played into N groups from high to low. b Each level has a corresponding bandwidth requirement coefficient. b The bandwidth requirement coefficient of each level is recorded as where n b =1,2,…,N b , According to the task bandwidth information B of the i-th task to be played i , determine the grade corresponding to the bandwidth demand and obtain its bandwidth demand coefficient ω i , determine the bandwidth requirement coefficient ω i is the bandwidth factor ρ of the i-th task to be played 3i ;

[0017] S24, calculate the median t0 of the task time information of all the tasks to be played; for each task time information of the task to be played, calculate the absolute value of the difference between the task time information and the median t0, and determine the absolute value as the time factor ρ of the task to be played 4i ρ 4i represents the time factor of the i-th task to be played;

[0018] S25, calculate and obtain the mode value s0 of the task coverage area information of all tasks to be played; for each task coverage area information of the task to be played, calculate and obtain the quotient of the task coverage area information and the mode value s0, and determine the quotient as the area factor ρ of the task to be played 5i ρ 5i represents the area factor of the i-th task to be played;

[0019] S26, calculating the weighted coefficient vector for all the task information to be played;

[0020] The calculation expression of the weighted coefficient vector is:

[0021]

[0022] Among them, m is the total number of tasks to be played, η j is the weighting coefficient of the j-th target attribute, and is also the j-th element of the weighting coefficient vector.

[0023] The expression of the initial resource allocation model is:

[0024]

[0025]

[0026] Among them, F(x) is the overall objective function of cluster game, f1(x) is the task priority sub-function, P ij represents the priority of the jth resource to complete the i-th task to be negotiated, x ij Indicates whether the jth resource participates in completing the ith pending game task. If so, then x ij is 1, otherwise x ij is 0; x is the resource allocation matrix to be solved, whose elements are x ij , i=1,2,…,m, j=1,2,…,n,D ij represents the distance between the jth resource and the task target of the i-th task to be negotiated, B ij Indicates the task bandwidth information of the jth resource participating in the ith task to be negotiated, t ij represents the time it takes for the jth resource to participate in the ith task to be gambled, s ijIndicates the coverage area that can be achieved by the j-th resource participating in the i-th task to be played, sr i Represents the task coverage area information of the i-th task to be played.

[0027] The initial resource allocation model is solved to obtain an initial resource allocation matrix, including:

[0028] S41, constructing and initializing a feature element set and the number of iterations, and constructing a parent feature element set using the initialized feature element set; the parent feature element set includes a plurality of parent feature elements; the parent feature element includes a corresponding feature vector and a step size; the feature vector is obtained by concatenating row vectors of a resource allocation matrix;

[0029] S42, using an element update operation to update the step size of the parent feature element;

[0030] S43, splitting the parent generation feature element set to obtain feature element sub-sets;

[0031] S44, performing a hierarchical operation on each feature element subset according to the step size of the parent feature element to obtain a corresponding hierarchical feature element subset;

[0032] S45, performing a feature interaction operation on hierarchical feature element subsets of the same level that belong to different feature element subsets to obtain a descendant feature element set;

[0033] S46, calculating a game performance index for the set of descendant characteristic elements to obtain a game performance index;

[0034] S47, using the game performance indicator, performing a migration operation on the offspring characteristic element set to obtain a migrated offspring characteristic element set and a characteristic element with the best fitness;

[0035] S48, performing an iteration stop judgment process on the game performance index and the number of iterations to obtain an iteration stop judgment result;

[0036] If the result of the iteration stop judgment is yes, the initial resource allocation information is constructed by using the fitness optimal characteristic element, and the solution process of the initial resource allocation model is completed;

[0037] If the result of the iteration stop judgment is no, increase the number of iterations, use the migrated offspring feature element set to construct the parent feature element set, and execute step S42.

[0038] The step size of the parent feature element is updated by using an element update operation, including:

[0039] S421, dynamically randomize the feature vector of the parent feature element to obtain a randomized feature vector; the expression p1 of the dynamically randomized feature vector is:

[0040] p1=α1p0+α2β(t),

[0041] Among them, p0 is the characteristic vector value of the parent characteristic element before randomization, [α1, α2] is the Gaussian noise vector, β(t) is the dynamic weight at time t; the expression of β(t) is:

[0042]

[0043] Wherein, ti represents the number of iterations, T0 and T1 represent the first turning iteration number and the second turning iteration number respectively, β0, β1 and β2 represent the first weight value, the second weight value and the third weight value respectively;

[0044] S422, for each parent feature element, establish a corresponding adversarial sample set; the adversarial sample set includes a plurality of adversarial samples; the adversarial sample includes a corresponding feature vector and a step length;

[0045] S423, performing robust index judgment processing on the parent feature element and its corresponding adversarial sample set to obtain a robust index judgment result;

[0046] If the robust index judgment result is greater than 0, the step size of the parent feature element is reduced by 1;

[0047] If the robust index judgment result is less than 0, the step size of the parent feature element is increased by 1.

[0048] For each parent feature element, a corresponding adversarial sample set is established, including:

[0049] S4221, searching for a set of neighboring feature elements whose distance to the parent feature element is less than a set distance threshold according to the distance between the feature vectors of the parent feature element;

[0050] S4222, performing feature vector ratio weighting processing on the parent feature element and the neighbor feature element set to construct a feature vector of the adversarial sample; the step length of the adversarial sample is randomly generated;

[0051] S4223, merging the adversarial samples obtained in step S4222 to obtain an adversarial sample set;

[0052] The feature vector ratio weighting process includes: establishing a number of related adversarial samples a ij , which is generated in a proportional weighted manner, namely:

[0053] a ij=a1×p i +a2×p ij ,

[0054] Among them, a1 and a1 are proportional weighting coefficients, the sum of the two is 1, p i is the characteristic element of the i-th parent generation, p ij is the jth neighbor feature element of the i-th parent feature element.

[0055] The robust index judgment processing is performed on the parent feature element and its corresponding adversarial sample set to obtain a robust index judgment result, including:

[0056] S4231, calculate the robustness index of the parent feature element and the adversarial sample, and its calculation expression is:

[0057]

[0058] in, Represents the parent feature element p i The robustness index, F(p i ) represents the parent feature element p i The total objective function value of the cluster game, F(a ij ) represents the parent feature element p i The total objective function value of the cluster game of the jth adversarial sample, m1 is the number of adversarial samples contained in the adversarial sample set, and α is the adversarial sample penalty weight, which is used to control the penalty weight of the adversarial sample;

[0059] S4232, determine whether the robustness index is greater than 0, and obtain a robustness index determination result.

[0060] The step of constructing a task channel transfer model by utilizing the resource historical allocation data set includes:

[0061] S61, extracting target information corresponding to the resource execution task from the resource historical allocation data set;

[0062] S62, using the target information, establishing a task node of a task channel transfer model; the task node of the task channel transfer model includes the target information;

[0063] S63, performing time association on the task nodes according to the task execution time corresponding to the task nodes to form a task execution path;

[0064] S64, calculating the simultaneous occurrence probability of adjacent task nodes according to the number of occurrences of the task node in the task execution path, and obtaining the connection probability of the adjacent task nodes;

[0065] S65, integrating the target information of the task node and the connection probability of the adjacent task nodes to establish execution path probability statistical information;

[0066] S66, performing data generalization processing on the execution path probability statistical information to obtain a task path transfer matrix;

[0067] S67, using the task path transfer matrix to establish a task channel transfer model.

[0068] The step of integrating the target information of the task node and the connection probability of the adjacent task nodes to establish execution path probability statistical information includes:

[0069] The execution path probability statistical information is represented in a table form; the target information of the task node is used as the row directory and column directory of the table, and the connection probability of adjacent task nodes is written into the spaces of the table corresponding to the adjacent task nodes.

[0070] The data generalization processing is performed on the execution path probability statistical information to obtain a task path transfer matrix, including:

[0071] Let the connection probability from the i-th task node to the j-th task node be e i,j In the execution path probability statistics information, the number of occurrences of the node path connecting the i-th task node to the j-th task node is counted to obtain the number of occurrences of the node path C i,j ; Use the number of occurrences of all node paths connected to the jth task node to construct the path set H of the jth task node j , whose expression is:

[0072] H j =[C 1,j ,C 2,j ,...,C N,j ],

[0073] The connection probability e in the execution path probability statistics i,j Perform data generalization processing to obtain the task path transfer probability H of the i-th task node connecting to the j-th task node i,j , the data generalization process is expressed as:

[0074]

[0075] The task path transfer probability between all task nodes is calculated, and the task path transfer probability between all task nodes is formed to form a task path transfer matrix.

[0076] The beneficial effects of the present invention are:

[0077] The present invention is aimed at the resource allocation problem of multi-objective cluster system game, comprehensively considers factors such as resource game time, communication bandwidth, coverage radius, priority, target distance, etc., and constructs an initial resource allocation model by evaluating various game factors; the initial resource allocation model is solved to obtain an initial resource allocation matrix, thereby ensuring the accuracy of the initial resource allocation plan.

[0078] Based on the historical game resource allocation data, a game resource path transfer model is constructed, and the game resource path transfer model is used to solve the game resource allocation scheme at the subsequent moment. The present invention realizes fast and accurate game resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a flow chart for implementing the method of the present invention. DETAILED DESCRIPTION

[0080] In order to better understand the content of the present invention, an embodiment is given here.

[0081] Figure 1 It is a flow chart for implementing the method of the present invention.

[0082] Aiming at the problem of how to optimize the allocation of existing game resources and quickly and accurately form a game resource allocation strategy for a game oriented to a multi-objective cluster system, the present invention discloses a resource iterative allocation method for a multi-objective game.

[0083] In a first aspect, an embodiment of the present application discloses a resource iterative allocation method for a multi-objective game, comprising:

[0084] S1, obtaining a set of information on tasks to be played and information on resources to be allocated; the set of information on tasks to be played includes information on several tasks to be played; the information on tasks to be played includes information on task target locations, task time spent, task bandwidth, task coverage area, and task priority of the tasks to be played; the information on resources to be allocated includes the number of resources to be allocated, location information of resources to be allocated, bandwidth of resources to be allocated, and coverage area;

[0085] S2, performing priority evaluation processing on the task information to be played to obtain a weighted coefficient vector;

[0086] S3, constructing an initial resource allocation model using the information of the game tasks to be gambled, the information of the resources to be allocated and the weighted coefficient vector;

[0087] S4, solving the initial resource allocation model to obtain an initial resource allocation matrix;

[0088] S5, obtaining a resource historical allocation data set;

[0089] S6, constructing a task channel transfer model using the resource historical allocation data set;

[0090] S7, using the task channel transfer model, processing the initial resource allocation information to obtain resource allocation solution information.

[0091] The priority evaluation process is performed on the task information to be played to obtain a weighted coefficient vector, including:

[0092] S21, normalize the priority value of each task to be played, and obtain the priority factor ρ of the task to be played 1i ; For the priority P of the i-th task to be played i , whose priority factor is ρ 1i =P i / (P max -P min ), where P max and P min They represent the upper and lower limits of the priority value range respectively;

[0093] S22, according to the task target location information of each task to be played, calculate the average distance between the task target and the resource to be allocated; use the preset N1 distance ranges to determine the distance range e to which the average distance belongs t , t=1,2,…,N1; each distance range has a corresponding distance coefficient; determine the distance range e to which the average distance belongs t The corresponding distance coefficient τ t , is the distance factor ρ of the game task to be played 2i ρ 2i represents the distance factor of the i-th task to be played;

[0094] S23, divide the task bandwidth information of the task to be played into N groups from high to low. b Each level has a corresponding bandwidth requirement coefficient. b The bandwidth requirement coefficient of each level is recorded as in According to the task bandwidth information B of the i-th task to be played i , determine the grade corresponding to the bandwidth demand and obtain its bandwidth demand coefficient ω i , determine the bandwidth requirement coefficient ω i is the bandwidth factor ρ of the i-th task to be played 3i ;

[0095] S24, calculate the median t0 of the task time information of all the tasks to be played; for each task time information of the task to be played, calculate the absolute value of the difference between the task time information and the median t0, and determine the absolute value as the time factor ρ of the task to be played 4i ρ 4i represents the time factor of the i-th task to be played;

[0096] S25, calculate and obtain the mode value s0 of the task coverage area information of all tasks to be played; for each task coverage area information of the task to be played, calculate and obtain the quotient of the task coverage area information and the mode value s0, and determine the quotient as the area factor ρ of the task to be played 5i ρ 5i represents the area factor of the i-th task to be played;

[0097] S26, calculating the weighted coefficient vector for all the task information to be played;

[0098] The calculation expression of the weighted coefficient vector is:

[0099]

[0100] Among them, m is the total number of tasks to be played, η j is the weighting coefficient of the j-th target attribute, and is also the j-th element of the weighting coefficient vector.

[0101] The expression of the initial resource allocation model is:

[0102]

[0103] Among them, F(x) is the overall objective function of cluster game, f1(x) is the task priority sub-function, P ij Indicates the priority of the jth resource to complete the i-th task to be played. If participating, then x ij is 1, otherwise x ij is 0; x is the resource allocation matrix to be solved, whose elements are x ij , i=1,2,…,m, j=1,2,…,n,D ij represents the distance between the jth resource and the task target of the i-th task to be negotiated, B ij Indicates the task bandwidth information of the jth resource participating in the ith task to be negotiated, t ij represents the time it takes for the jth resource to participate in the ith task to be gambled, s ij Indicates the coverage area that can be achieved by the j-th resource participating in the i-th task to be played, sr i Indicates the task coverage area information of the i-th task to be played;

[0104] The initial resource allocation model is solved to obtain an initial resource allocation matrix, including:

[0105] S41, constructing and initializing a feature element set and the number of iterations, and constructing a parent feature element set using the initialized feature element set; the parent feature element set includes a plurality of parent feature elements; the parent feature element includes a corresponding feature vector and a step size; the feature vector is obtained by concatenating row vectors of a resource allocation matrix;

[0106] The eigenvector is obtained by concatenating all row vectors of the resource allocation matrix in sequence according to row numbers or column numbers;

[0107] The use of the initialized feature element set to construct the parent feature element set may be to use the initialized feature element set as the parent feature element set; the initialization iteration number is to set the iteration number to 1. The concatenation is to merge multiple vectors into a row vector without changing the order and value of the elements in each vector.

[0108] S42, using an element update operation to update the step size of the parent feature element;

[0109] S43, splitting the parent generation feature element set to obtain feature element sub-sets;

[0110] The splitting of the parent generation characteristic element set may be a random splitting or a uniform splitting or a splitting according to a preset ratio to obtain a characteristic element subset;

[0111] S44, performing a hierarchical operation on each feature element subset according to the step size of the parent feature element to obtain a corresponding hierarchical feature element subset;

[0112] S45, performing a feature interaction operation on hierarchical feature element subsets of the same level belonging to different feature element subsets to obtain a descendant feature element set;

[0113] S46, calculating a game performance index for the set of descendant characteristic elements to obtain a game performance index;

[0114] S47, using the game performance indicator, performing a migration operation on the offspring characteristic element set to obtain a migrated offspring characteristic element set and a characteristic element with the best fitness;

[0115] S48, performing an iteration stop judgment process on the game performance index and the number of iterations to obtain an iteration stop judgment result;

[0116] If the result of the iteration stop judgment is yes, the initial resource allocation information is constructed by using the fitness optimal characteristic element, and the solution process of the initial resource allocation model is completed;

[0117] If the result of the iteration stop judgment is no, increase the number of iterations, use the migrated offspring feature element set to construct the parent feature element set, and execute step S42.

[0118] The initial resource allocation matrix is ​​constructed by using the optimal characteristic element of fitness, which is to divide the optimal characteristic element of fitness into m elements as a vector, obtain n vectors, and construct the initial resource allocation matrix by taking all vectors as column vectors; the element of the jth column and the ith row of the initial resource allocation matrix indicates whether the jth resource participates in completing the ith task to be gambled. If so, x ij is 1, otherwise x ij is 0;

[0119] The step size of the parent feature element is updated by using an element update operation, including:

[0120] S421, dynamically randomize the feature vector of the parent feature element to obtain a randomized feature vector; the expression p1 of the dynamically randomized feature vector is:

[0121] p1=α1p0+α2β(t),

[0122] Among them, p0 is the feature vector value of the parent feature element before randomization, [α1, α2] is the Gaussian noise vector, β(t) is the dynamic weight at time t, which is used to balance the objective function in the feature element evaluation and the penalty weight of the adversarial sample; the expression of β(t) is:

[0123]

[0124] Among them, ti represents the number of iterations, T0 and T1 represent the first turning point iteration number and the second turning point iteration number respectively, β0, β1 and β2 represent the first weight value, the second weight value and the third weight value respectively, which are used to increase the diversity of feature elements and improve the global search capability of the solution method.

[0125] S422, for each parent feature element, establish a corresponding adversarial sample set; the adversarial sample set includes a plurality of adversarial samples; the adversarial sample includes a corresponding feature vector and a step length;

[0126] For each parent feature element, a corresponding adversarial sample set is established, including:

[0127] S4221, searching for a set of neighboring feature elements whose distance to the parent feature element is less than a set distance threshold according to the distance between the feature vectors of the parent feature element;

[0128] S4222, performing feature vector ratio weighting processing on the parent feature element and the neighbor feature element set to construct a feature vector of the adversarial sample; the step length of the adversarial sample is randomly generated;

[0129] S4223, merging the adversarial samples obtained in step S4222 to obtain an adversarial sample set;

[0130] The feature vector ratio weighting process includes: establishing a number of related adversarial samples a ij , which is generated in a proportional weighted manner, namely:

[0131] a ij =a1×p i +a2×p ij ,

[0132] Among them, a1 and a1 are proportional weighting coefficients, the sum of the two is 1, p i is the characteristic element of the i-th parent generation, p ij is the jth neighbor feature element of the i-th parent feature element.

[0133] S423, performing robust index judgment processing on the parent feature element and its corresponding adversarial sample set to obtain a robust index judgment result;

[0134] If the robust index judgment result is greater than 0, the step size of the parent feature element is reduced by 1;

[0135] If the robust index judgment result is less than 0, the step size of the parent feature element is increased by 1;

[0136] The robust index judgment processing is performed on the parent feature element and its corresponding adversarial sample set to obtain a robust index judgment result, including:

[0137] S4231, calculate the robustness index of the parent feature element and the adversarial sample, and its calculation expression is:

[0138]

[0139] in, Represents the parent feature element p i The robustness index, F(p i ) represents the parent feature element p i The total objective function value of the cluster game, F(a ij ) represents the parent feature element p iThe total objective function value of the cluster game of the jth adversarial sample, m1 is the number of adversarial samples contained in the adversarial sample set, and α is the adversarial sample penalty weight, which is used to control the penalty weight of the adversarial sample.

[0140] S4232, determining whether the robustness index is greater than 0, and obtaining a robustness index determination result;

[0141] The step size of the parent feature element is used to perform a hierarchical operation on each feature element subset to obtain a corresponding hierarchical feature element subset, including:

[0142] S441, setting the step value range corresponding to each level;

[0143] S442, matching the step length of the parent feature element with a value range to obtain a step length value range corresponding to the parent feature element;

[0144] S443, dividing the parent generation feature element into a level corresponding to the step value range according to the step value range corresponding to the parent generation feature element;

[0145] S444, merging the parent feature elements belonging to the same level in the feature element subset to obtain a hierarchical feature element subset of the feature element subset;

[0146] The step S44 may be that the basis for stratification is the step value of the characteristic element, and the parent characteristic elements satisfying the step value <3 are divided into one layer, the parent characteristic elements satisfying 3≤step value <10 are divided into two layers, and the parent characteristic elements satisfying 10≤step value are divided into three layers; the levels include one layer, two layers and three layers.

[0147] The feature interaction operation in step S45 includes:

[0148] The hierarchical feature element subsets in different feature element subsets at the same level interact with each other to increase the diversity of feature elements and global search capability. Specifically, for each hierarchical feature element subset P i , in its neighborhood, randomly select several subsets of hierarchical feature elements to construct subset P i The neighboring subset N(P i ), perform feature interaction operations. Hierarchical feature element subset P i The neighborhood of is the hierarchical feature element subset P i The distances of the parent feature elements in the sub-sets of the hierarchical feature elements are all smaller than the preset distances of the parent feature elements in the sub-sets of the hierarchical feature elements. The feature interaction operation includes an overlap operation and an exchange operation.

[0149] The overlapping operation is performed on the hierarchical feature element subset P i A variable is randomly selected from the feature vector of the parent feature element in as the crossover point, and the hierarchical feature element subset P is i The parent feature element and the adjacent subset N(P i ) are exchanged at the intersection to generate the feature vectors of the descendant feature elements after the overlapping operation. The specific implementation method is as follows: Let the hierarchical feature element subset P i The characteristic vector of the parent characteristic element is the same as the adjacent subset N(P i ) are P1 and P2, respectively, and their feature vector lengths are both L. Randomly select the intersection point i∈1,2,…,L-1, and then we can get two overlapping parent feature elements C1 and C2, whose feature vectors are derived from the intersection of the parent feature elements:

[0150] C1=[P1[1],P1[2],…,P1[i],P2[i+1],P2[i+2],…,P2[L]]

[0151] C2=[P2[1],P2[2],…,P2[i],P1[i+1],P1[i+2],…,P1[L]]

[0152] Among them, P1[j] represents the j-th value of the characteristic vector of the parent characteristic element P1, that is, the j-th gene value, and P2[j] represents the j-th gene value of the parent characteristic element P2.

[0153] The interchange operation is to replace the hierarchical feature element subset P after the overlap operation i The characteristic vector of the parent characteristic element is randomly selected and the vector value is changed to generate the characteristic vector of the new offspring characteristic element. The specific implementation method is as follows: suppose a parent characteristic element is P, its characteristic vector length is L, randomly select one or more mutation sites i∈1,2,…,L, and randomly generate a new gene value g, then a new offspring characteristic element C can be obtained, whose characteristic vector is derived from the mutation of the parent characteristic element:

[0154] C=[P[1],P[2],…,P[i-1],g,P[i+1],…,P[L]]

[0155] Among them, P[j] represents the j-th gene value of the characteristic vector of the parent characteristic element P.

[0156] The generated offspring feature elements are used to replace the corresponding parent feature elements in the parent feature element set to obtain the offspring feature element set. Correspondingly, the generated offspring feature elements are used to replace the corresponding parent feature elements in the parent feature element subset to obtain the offspring feature element subset; the generated offspring feature elements are used to replace the corresponding parent feature elements in the parent hierarchical feature element subset to obtain the offspring hierarchical feature element subset.

[0157] The calculation expression of the game performance index is:

[0158] U M (x1)=F(x1)+D(x1)

[0159] Among them, U M () is the game performance index calculation function, F() is the total objective function of cluster game, D() is the crowding distance index function, x1 is the initial resource allocation matrix corresponding to the characteristic vector of the offspring characteristic element; the crowding distance index, its calculation expression is:

[0160]

[0161] Wherein, N(x1) represents the set of k descendant feature elements in the descendant feature element set that are closest to the descendant feature element corresponding to x1, y is the descendant feature element in the set N(x1), k is a preset feature element number parameter, and D(y) represents the Euclidean distance between the descendant feature element corresponding to x1 and the descendant feature element y.

[0162] The step of utilizing the game performance index to perform a migration operation on the offspring feature element set to obtain a migrated offspring feature element set and a feature element with the best fitness includes:

[0163] S471, according to the game performance index of the offspring feature element, for each offspring feature element subset, respectively perform a game performance index search operation to obtain the offspring feature element with the best game performance index and the offspring feature element with the game performance index lower than the preset index threshold; for the offspring feature element subset, the offspring feature element with the best game performance index and the offspring feature element with the game performance index lower than the preset index threshold are respectively used as the best feature element and the inferior feature element of the offspring feature element subset;

[0164] S472, in each descendant feature element subset, migrate the best feature element of the descendant feature element subset to a hierarchical feature element subset of a higher level, and increase the step size of the best feature element according to the step size value range of the migrated hierarchical feature element subset; the descendant feature element subset is the descendant feature element subset;

[0165] S473, in each descendant feature element subset, migrate the inferior feature elements of the descendant feature element subset to a hierarchical feature element subset of a lower level, and reduce the step size of the optimal feature element according to the step size value range of the migrated hierarchical feature element subset;

[0166] In particular, if the optimal feature element is already in the highest level hierarchical feature element subset, or the inferior feature element is already in the lowest level hierarchical feature element subset, no migration operation is performed on the feature element.

[0167] S474, constructing a migration descendant characteristic element set by using all descendant characteristic element subsets that have completed migration;

[0168] The step S474 may be to use all the descendant feature element subsets that have completed migration as the migrated descendant feature element set;

[0169] S475, searching for a descendant characteristic element with the best gaming performance index in the set of migrated descendant characteristic elements, and taking the descendant characteristic element as the characteristic element with the best fitness.

[0170] The performing iteration stop judgment processing on the game performance index and the number of iterations to obtain an iteration stop judgment result includes:

[0171] Determine whether the game performance index of the fitness optimal characteristic element is better than the preset preferred index threshold, and whether the number of iterations is greater than the preset number of iterations threshold. If any of the judgment conditions is yes, determine that the iteration stop judgment result is yes; if the above two judgment conditions are both no, determine that the iteration stop judgment result is no.

[0172] The first to fifth target attributes are task priority, task target location, task bandwidth, task time, and task coverage area;

[0173] The mode value is the difference between the maximum value and the minimum value.

[0174] The resource historical allocation data set includes the execution task time, execution task information and corresponding target information of each resource in the multi-objective game task for which resource allocation has been completed and the game has been completed;

[0175] The execution task information includes the task type, whether to participate in the task, etc.;

[0176] The step of constructing a task channel transfer model by utilizing the resource historical allocation data set includes:

[0177] S61, extracting target information corresponding to the resource execution task from the resource historical allocation data set;

[0178] S62, using the target information, establishing a task node of a task channel transfer model; the task node of the task channel transfer model includes the target information;

[0179] S63, performing time association on the task nodes according to the task execution time corresponding to the task nodes to form a task execution path;

[0180] S64, calculating the simultaneous occurrence probability of adjacent task nodes according to the number of occurrences of the task node in the task execution path, and obtaining the connection probability of the adjacent task nodes;

[0181] S65, integrating the target information of the task node and the connection probability of the adjacent task nodes to establish execution path probability statistical information;

[0182] S66, performing data generalization processing on the execution path probability statistical information to obtain a task path transfer matrix;

[0183] S67, using the task path transfer matrix to establish a task channel transfer model.

[0184] The step of integrating the target information of the task node and the connection probability of the adjacent task nodes to establish execution path probability statistical information includes:

[0185] The execution path probability statistical information is represented in a table form; the target information of the task node is used as the row directory and column directory of the table, and the connection probability of adjacent task nodes is written into the spaces of the table corresponding to the adjacent task nodes.

[0186] The data generalization processing is performed on the execution path probability statistical information to obtain a task path transfer matrix, including:

[0187] Let the connection probability from the i-th task node to the j-th task node be e i,j In the execution path probability statistics information, the number of occurrences of the node path connecting the i-th task node to the j-th task node is counted to obtain the number of occurrences of the node path C i,j ; Use the number of occurrences of all node paths connected to the jth task node to construct the path set H of the jth task node j , whose expression is:

[0188] H j =[C 1,j ,C 2,j ,...,C N,j ],

[0189] The connection probability e in the execution path probability statistics i,j Perform data generalization processing to obtain the task path transfer probability H of the i-th task node connecting to the j-th task node i,j , the data generalization process is expressed as:

[0190]

[0191] The task path transfer probability between all task nodes is calculated, and the task path transfer probability between all task nodes is formed to form a task path transfer matrix.

[0192] The row numbers and column numbers of the task path transfer matrix elements respectively represent the target information of the current task node and the target information of the task node to be transferred.

[0193] The task channel transfer model includes:

[0194] The input of the task channel transfer model is the initial resource allocation matrix;

[0195] The task transfer probability matrix is ​​obtained by multiplying the initial resource allocation matrix with the task path transfer matrix; for each column vector in the task transfer probability matrix, the row number corresponding to the maximum value element is determined, which is the number of the subsequent game task to be participated in by the resource corresponding to the column vector;

[0196] After obtaining the serial numbers of the tasks to be played that all resources will participate in later, the serial numbers of the tasks to be played that all resources will participate in later are used as the subsequent resource allocation information;

[0197] The subsequent resource allocation information and the initial resource allocation matrix are used to construct the resource allocation scheme information.

[0198] For example, the element with row number 4 in the 5th column vector of the task transfer probability matrix is ​​the element with the largest value in the 5th column vector, and determines the 5th resource, which subsequently participates in the 4th task to be gambled.

[0199] The step of processing the initial resource allocation matrix by using the task channel transfer model to obtain resource allocation scheme information includes:

[0200] The initial resource allocation matrix is ​​input into the task channel transfer model, and the task channel transfer model is used to process the initial resource allocation matrix to obtain subsequent resource allocation information. The initial resource allocation matrix and the subsequent resource allocation information are used as resource allocation scheme information to complete the iterative allocation of cluster resources.

[0201] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A resource iterative allocation method for multi-objective games, characterized in that: include: S1, obtaining a set of information on tasks to be played and information on resources to be allocated; The set of information on pending game tasks includes information on a number of pending game tasks; The task information to be played includes the task target location information, task time information, task bandwidth information, task coverage area information, and task priority information of the task to be played; the resource information to be allocated includes the number of resources to be allocated, the location information of the resources to be allocated, the bandwidth and coverage area of ​​the resources to be allocated; S2, performing priority evaluation processing on the task information to be played to obtain a weighted coefficient vector; S3, constructing an initial resource allocation model using the information of the game tasks to be gambled, the information of the resources to be allocated and the weighted coefficient vector; S4, solving the initial resource allocation model to obtain an initial resource allocation matrix; S5, obtaining a resource historical allocation data set; S6, constructing a task channel transfer model using the resource historical allocation data set; S7, using the task channel transfer model, processing the initial resource allocation information to obtain resource allocation solution information.

2. The resource iterative allocation method for multi-objective games according to claim 1, characterized in that: The priority evaluation process is performed on the task information to be played to obtain a weighted coefficient vector, including: S21, normalize the priority value of each task to be played, and obtain the priority factor ρ of the task to be played 1i ; For the priority P of the i-th task to be played i , whose priority factor is ρ 1i =P i / (P max -P min ), where P max and P min They represent the upper and lower limits of the priority value range respectively; S22, according to the task target location information of each task to be played, calculate the average distance between the task target and the resource to be allocated; use the preset N1 distance ranges to determine the distance range e to which the average distance belongs t , t=1,2,...,N1; each distance range has a corresponding distance coefficient; determine the distance range e to which the average distance belongs t The corresponding distance coefficient τ t , is the distance factor ρ of the game task to be played 2i ρ 2i represents the distance factor of the i-th task to be played; S23, divide the task bandwidth information of the task to be played into N groups from high to low. b Each level has a corresponding bandwidth requirement coefficient. b The bandwidth requirement coefficient of each level is recorded as where n b =1,2,…,N b , According to the task bandwidth information B of the i-th task to be played i , determine the grade corresponding to the bandwidth demand and obtain its bandwidth demand coefficient ω i , determine the bandwidth requirement coefficient ω i is the bandwidth factor ρ of the i-th task to be played 3i ; S24, calculate the median t0 of the task time information of all the tasks to be played; for each task time information of the task to be played, calculate the absolute value of the difference between the task time information and the median t0, and determine the absolute value as the time factor ρ of the task to be played 4i ρ 4i represents the time factor of the i-th task to be played; S25, calculate and obtain the mode value s0 of the task coverage area information of all tasks to be played; for each task coverage area information of the task to be played, calculate and obtain the quotient of the task coverage area information and the mode value s0, and determine the quotient as the area factor ρ of the task to be played 5i ρ 5i represents the area factor of the i-th task to be played; S26, calculating the weighted coefficient vector for all the task information to be played; The calculation expression of the weight coefficient vector is: Among them, m is the total number of tasks to be played, η j is the weighting coefficient of the j-th target attribute, and is also the j-th element of the weighting coefficient vector.

3. The iterative resource allocation method for multi-objective games according to claim 2, characterized in that: The expression of the initial resource allocation model is: Among them, F(x) is the overall objective function of cluster game, f1(x) is the task priority sub-function, P ij represents the priority of the jth resource to complete the i-th task to be negotiated, x ij Indicates whether the jth resource participates in completing the ith pending game task. If so, then x ij is 1, otherwise x ij is 0; x is the resource allocation matrix to be solved, whose elements are x ij , i=1,2,…,m, j=1,2,…,n,D ij represents the distance between the jth resource and the task target of the i-th task to be negotiated, B ij Indicates the task bandwidth information of the jth resource participating in the ith task to be negotiated, t ij represents the time it takes for the jth resource to participate in the ith task to be gambled, s ij Indicates the coverage area that can be achieved by the j-th resource participating in the i-th task to be played, sr i Represents the task coverage area information of the i-th task to be played.

4. The iterative resource allocation method for multi-objective games as claimed in claim 3, characterized in that: The initial resource allocation model is solved to obtain an initial resource allocation matrix, including: S41, constructing and initializing a feature element set and the number of iterations, and constructing a parent feature element set using the initialized feature element set; the parent feature element set includes a plurality of parent feature elements; the parent feature element includes a corresponding feature vector and a step size; the feature vector is obtained by concatenating row vectors of a resource allocation matrix; S42, using an element update operation to update the step size of the parent feature element; S43, splitting the parent generation feature element set to obtain feature element sub-sets; S44, performing a hierarchical operation on each feature element subset according to the step size of the parent feature element to obtain a corresponding hierarchical feature element subset; S45, performing a feature interaction operation on hierarchical feature element subsets of the same level belonging to different feature element subsets to obtain a descendant feature element set; S46, calculating a game performance index for the set of descendant characteristic elements to obtain a game performance index; S47, using the game performance indicator, performing a migration operation on the offspring characteristic element set to obtain a migrated offspring characteristic element set and a characteristic element with the best fitness; S48, performing an iteration stop judgment process on the game performance index and the number of iterations to obtain an iteration stop judgment result; If the result of the iteration stop judgment is yes, the initial resource allocation information is constructed by using the fitness optimal characteristic element, and the solution process of the initial resource allocation model is completed; If the result of the iteration stop judgment is no, increase the number of iterations, use the migrated offspring feature element set to construct a parent feature element set, and execute step S42.

5. The resource iterative allocation method for multi-objective games as claimed in claim 4, characterized in that: The step size of the parent feature element is updated by using an element update operation, including: S421, dynamically randomize the feature vector of the parent feature element to obtain a randomized feature vector; the expression p1 of the dynamically randomized feature vector is: p1=α1p0+α2β(t), Among them, p0 is the characteristic vector value of the parent characteristic element before randomization, [α1, α2] is the Gaussian noise vector, β(t) is the dynamic weight at time t; the expression of β(t) is: Wherein, ti represents the number of iterations, T0 and T1 represent the first turning iteration number and the second turning iteration number respectively, β0, β1 and β2 represent the first weight value, the second weight value and the third weight value respectively; S422, for each parent feature element, establish a corresponding adversarial sample set; the adversarial sample set includes a plurality of adversarial samples; the adversarial sample includes a corresponding feature vector and a step length; S423, performing robust index judgment processing on the parent feature element and its corresponding adversarial sample set to obtain a robust index judgment result; If the robust index judgment result is greater than 0, the step size of the parent feature element is reduced by 1; If the robust index judgment result is less than 0, the step size of the parent feature element is increased by 1.

6. The iterative resource allocation method for multi-objective games as claimed in claim 5, characterized in that: For each parent feature element, a corresponding adversarial sample set is established, including: S4221, searching for a set of neighboring feature elements whose distance to the parent feature element is less than a set distance threshold according to the distance between the feature vectors of the parent feature element; S4222, performing feature vector ratio weighting processing on the parent feature element and the neighbor feature element set to construct a feature vector of the adversarial sample; the step length of the adversarial sample is randomly generated; S4223, merging the adversarial samples obtained in step S4222 to obtain an adversarial sample set; The feature vector ratio weighting process includes: establishing a number of related adversarial samples a ij , which is generated in a proportional weighted manner, namely: a ij =a1×p i +a2×p ij , Among them, a1 and a1 are proportional weighting coefficients, the sum of the two is 1, p i is the characteristic element of the i-th parent generation, p ij is the jth neighbor feature element of the i-th parent feature element.

7. The iterative resource allocation method for multi-objective games according to claim 6, characterized in that: The robust index judgment processing is performed on the parent feature element and its corresponding adversarial sample set to obtain a robust index judgment result, including: S4231, calculate the robustness index of the parent feature element and the adversarial sample, and its calculation expression is: in, Represents the parent feature element p i The robustness index, F(p i ) represents the parent feature element p i The total objective function value of the cluster game, F(a ij ) represents the parent feature element p i The total objective function value of the cluster game of the jth adversarial sample, m1 is the number of adversarial samples contained in the adversarial sample set, and α is the adversarial sample penalty weight, which is used to control the penalty weight of the adversarial sample; S4232, determine whether the robustness index is greater than 0, and obtain a robustness index determination result.

8. The iterative resource allocation method for multi-objective games as claimed in claim 3, characterized in that: The step of constructing a task channel transfer model by utilizing the resource historical allocation data set includes: S61, extracting target information corresponding to the resource execution task from the resource historical allocation data set; S62, using the target information, establishing a task node of a task channel transfer model; the task node of the task channel transfer model includes the target information; S63, performing time association on the task nodes according to the task execution time corresponding to the task nodes to form a task execution path; S64, calculating the simultaneous occurrence probability of adjacent task nodes according to the number of occurrences of the task node in the task execution path, and obtaining the connection probability of the adjacent task nodes; S65, integrating the target information of the task node and the connection probability of the adjacent task nodes to establish execution path probability statistical information; S66, performing data generalization processing on the execution path probability statistical information to obtain a task path transfer matrix; S67, using the task path transfer matrix to establish a task channel transfer model.

9. The iterative resource allocation method for multi-objective games as claimed in claim 8, characterized in that: The step of integrating the target information of the task node and the connection probability of the adjacent task nodes to establish execution path probability statistical information includes: The execution path probability statistical information is represented in a table form; the target information of the task node is used as the row directory and column directory of the table, and the connection probability of adjacent task nodes is written into the spaces of the table corresponding to the adjacent task nodes.

10. The iterative resource allocation method for multi-objective games according to claim 8, characterized in that: The data generalization processing is performed on the execution path probability statistical information to obtain a task path transfer matrix, including: Let the connection probability from the i-th task node to the j-th task node be e i,j In the execution path probability statistics information, the number of occurrences of the node path connecting the i-th task node to the j-th task node is counted to obtain the number of occurrences of the node path C i,j ; Use the number of occurrences of all node paths connected to the jth task node to construct the path set H of the jth task node j , whose expression is: H j =[C 1,j ,C 2,j ,...,C N,j ], The connection probability e in the execution path probability statistics i,j Perform data generalization processing to obtain the task path transfer probability H of the i-th task node connecting to the j-th task node i,j , the data generalization process is expressed as: The task path transfer probability between all task nodes is calculated, and the task path transfer probability between all task nodes is formed to form a task path transfer matrix.

Images