Historical character activation computing power resource scheduling method and system based on improved artificial bee colony algorithm

Abstract

The invention provides a historical character activation computing power resource scheduling method and system based on an improved artificial bee colony algorithm. The method comprises the following steps: constructing a task set and a resource set according to a task type needing to be executed in a historical figure activation process and a computing power resource capability available for scheduling; randomly initializing a nectar source set; wherein each nectar source in the nectar source set represents a task scheduling scheme; selecting a nectar source from the nectar source set; 4, for the current nectar source, updating the nectar source by adopting a multi-dimensional search strategy to obtain a new nectar source; the multi-dimensional search strategy comprises the following steps: selecting an excavable dimension from a current nectar source, and disturbing the selected dimension; the fitness of the new nectar source is calculated by adopting a pre-constructed resource scheduling cost function, if the fitness is improved, the new nectar source is adopted, and the disturbed dimension is marked as an excavable dimension; and otherwise, next iteration is carried out until a stop condition is met, and the current latest nectar source is output as the optimal task scheduling scheme.

Description

Technical Field

[0001] This invention relates to the field of historical figure revitalization technology, and particularly to an improved artificial bee colony algorithm for scheduling computing resources for historical figure revitalization. Background Technology

[0002] With the widespread application of artificial intelligence in the field of historical figure animation, higher demands are being placed on the underlying hardware computing capabilities. Especially in tasks involving multimodal reconstruction, speech synthesis, and behavior simulation, high-performance computing has become a crucial foundation for supporting high-speed dynamic modeling. Current character and scene animation tasks often involve multi-source input and concurrent modeling requirements, involving heterogeneous computing resources including CPUs, GPUs, and DCUs. Their uneven physical distribution leads to low overall utilization of computing resources. In large-scale, heterogeneous hybrid computing systems for dynamic characters and modeling, how to efficiently schedule computing tasks and ensure maximum resource utilization has become a critical problem that urgently needs to be solved.

[0003] However, current methods for character and animation modeling in heterogeneous resource environments suffer from low computational task scheduling efficiency. Traditional scheduling methods mostly rely on a single, customized scheduling strategy, making it difficult to dynamically adjust the allocation of computational resources based on computational tasks, and thus failing to effectively address issues such as uneven resource distribution and excessive data transmission latency.

[0004] In real-world historical figure re-enactment scenarios, the system needs to perform concurrent modeling tasks on multimodal data such as images, speech, and poses, including 3D facial reconstruction, speech synthesis, and motion fitting. These tasks rely on different computing resources; for example, image reconstruction primarily uses GPUs for tensor computation and graphics rendering, speech synthesis may involve DSPs or dedicated speech processors, and motion simulation relies more on CPUs for control reasoning and logical judgment. How to accurately match and schedule tasks based on task type and resource capabilities is the key bottleneck to the efficient computing performance of such re-enactment systems.

[0005] Therefore, achieving low-latency and highly adaptable task scheduling on heterogeneous platforms has become a key requirement for the historical figure revitalization system. Summary of the Invention

[0006] To address the issue of low computational task scheduling efficiency in heterogeneous resource environments for character and animation modeling, this invention provides a method and system for scheduling historical character animation computational resources based on an improved artificial bee colony algorithm. This invention optimizes the artificial bee colony algorithm, combines it with a multi-dimensional search strategy, schedules heterogeneous computing resources in a simulated environment, and effectively enhances the global search capability of the scheduling through a fitness evaluation mechanism, significantly accelerating the convergence speed, ultimately achieving the goal of improving the speed of character and animation modeling.

[0007] In a first aspect, the present invention provides a method for scheduling computing resources for the activation of historical figures based on an improved artificial bee colony algorithm, comprising:

[0008] Step 1: Based on the types of tasks that need to be performed during the revitalization of historical figures and the available computing resources, construct a task set and a resource set;

[0009] Step 2: Randomly initialize the honey source set; wherein, each honey source in the honey source set represents a task scheduling scheme, and one dimension of each honey source corresponds to the allocation decision between a certain task in the task set and a certain resource in the resource set;

[0010] Step 3: Select nectar sources from the nectar source collection;

[0011] Step 4: For the current nectar source, use a multi-dimensional search strategy to update the nectar source and obtain a new nectar source; the multi-dimensional search strategy includes: selecting exploitable dimensions from the current nectar source and perturbing the selected dimensions;

[0012] Step 5: Calculate the fitness of the new honey source using the pre-built resource scheduling cost function. If the fitness is improved, adopt the new honey source and mark the perturbation dimension as a mineable dimension.

[0013] Step 6: Repeat steps 3 to 5 until the stopping condition is met, and output the latest honey source as the optimal task scheduling scheme.

[0014] Furthermore, the process of constructing the scheduling cost function includes:

[0015] The performance loss caused by scheduling and executing tasks on different types of resources and the degree of resource matching are comprehensively considered to determine the loss index; the determined loss index is weighted and combined to obtain the total scheduling loss; the resource waste rate is calculated; the total scheduling loss and the resource waste rate are weighted and combined to obtain the scheduling cost function.

[0016] Furthermore, the loss indicators include: computing power loss, data transmission time loss, network bandwidth loss, resource type mismatch loss, other losses, and resource waste rate; wherein, the computing power loss is used to measure whether the current computing power of the resources can meet the task requirements; the data transmission time loss is used to indicate the time cost required to transmit task input data from the data source to the computing node; the network bandwidth loss is used to indicate the network bandwidth load of the resources; the resource type mismatch loss is used to penalize scheduling behavior that does not match the task type and resource capacity; the other losses include task switching overhead and scheduling latency; and the resource waste rate is used to indicate the proportion of resources that are not effectively utilized.

[0017] Furthermore, the determined loss indicators are weighted and combined to obtain the total scheduling loss, specifically including:

[0018]

[0019] Where Loss represents the total scheduling loss, and w1, w2, w3, w4, and w5 are adjustable weights;

[0020] Correspondingly, the values ​​of each loss index are calculated according to the following formulas:

[0021]

[0022] in, Represents task T i With resource C j The computing power loss between them, Comp(T) i ) for task T i Required computation, Power(C) j ) is for resource C j Currently available computing power;

[0023]

[0024] in, Represents task T i With resource C j Data transmission time loss between Size(T) i ) for task T i Data volume, BW(C j ) is for resource C j The current bandwidth;

[0025]

[0026] in, Indicates resource C j Network bandwidth loss, Load(C j ) is for resource C j Network transmission requirements, BWmax(C j ) is for resource C j The maximum theoretical bandwidth;

[0027]

[0028] in, Represents task T i With resource C j Resource type mismatch loss To preset the penalty value, type(T) i ) represents task T i The type, cap(C j ) represents resource C jAbility tags, type(T) i )~cap(C j ) represents task T i With resource C j They match.

[0029] Furthermore, the resource scheduling cost function Cost is:

[0030] Cost = Loss + w6·Loss RU

[0031] Where w6 is an adjustable weight; Loss RU Indicates the rate of resource waste. R use R represents the amount of computing resources actually used by the current task. ava This indicates the total available resource capacity of the current resource node.

[0032] Secondly, this invention provides a historical figure activation computing resource scheduling system based on an improved artificial bee colony algorithm, comprising:

[0033] The scheduling information construction module is used to construct task sets and resource sets based on the types of tasks to be executed during the revitalization of historical figures and the available computing resources.

[0034] An initialization module is used to randomly initialize a honey source set; wherein, each honey source in the honey source set represents a task scheduling scheme, and one dimension of each honey source corresponds to the allocation decision between a certain task in the task set and a certain resource in the resource set;

[0035] An iterative search module is used to select honey sources from the honey source set; for the current honey source, a multi-dimensional search strategy is used to update the honey source to obtain a new honey source; the fitness of the new honey source is calculated using a pre-built scheduling cost function; if the fitness is improved, the new honey source is adopted and the perturbation dimension is marked as a mineable dimension; otherwise, the next iteration is performed until the stopping condition is reached; the latest honey source is output as the optimal task scheduling scheme; wherein, the multi-dimensional search strategy includes: selecting a mineable dimension from the current honey source and perturbing the selected dimension.

[0036] Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in the first aspect.

[0037] Fourthly, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0038] The beneficial effects of this invention are as follows:

[0039] (1) Improve the utilization rate of heterogeneous resources

[0040] This invention utilizes an artificial bee colony algorithm based on an optimization strategy to achieve dynamic matching and scheduling between tasks and heterogeneous resources (such as CPUs, GPUs, and FPGAs) according to the task's computational type and resource capability tags. The scheduling method designed in this invention introduces a pairing mechanism between task type and resource capability into the designed resource scheduling cost function and uses "resource type mismatch loss" to penalize unreasonable allocations, thereby effectively avoiding resource mismatch problems. Through the optimized scheduling mechanism, the overall resource utilization rate in heterogeneous computing environments can be significantly improved, while reducing resource idle rate and redundant scheduling overhead.

[0041] (2) Optimize the allocation of computing resources and reduce task scheduling costs.

[0042] This invention quantifies the execution cost of each task scheduling scheme by constructing a comprehensive resource scheduling cost function, Cost. This cost function covers multiple factors such as computing power loss, data transmission latency, network bandwidth load, and resource type penalties. Using this cost function as a fitness function, it guides the updating of honey source schemes. Through this dynamic, feedback-driven scheduling optimization strategy, the overall system operating cost is effectively reduced and the comprehensive scheduling efficiency of heterogeneous platforms is improved.

[0043] (3) The artificial bee colony algorithm based on the optimization strategy has better global search capabilities.

[0044] This invention introduces a multi-dimensional search strategy with a "temporary update buffer (temp)" mechanism into the traditional artificial bee colony algorithm, enhancing the algorithm's global search capability. By searching and optimizing task scheduling schemes across multiple dimensions, the trap of local optima can be avoided, thus finding a better task scheduling scheme. Compared with traditional algorithms, the improved artificial bee colony algorithm of this invention, through the design of "dimensional selection + local update + adaptive reset," possesses stronger global search capability and convergence stability in the resource scheduling search process, making it particularly suitable for large-scale historical figure re-enactment task scheduling problems under heterogeneous resources. Attached Figure Description

[0045] Figure 1 A flowchart illustrating the historical figure activation computing resource scheduling method based on the improved artificial bee colony algorithm provided in this embodiment of the invention;

[0046] Figure 2 A flowchart of the first iteration provided for an embodiment of the present invention;

[0047] Figure 3The flowchart of the second iteration provided in the embodiment of the present invention;

[0048] Figure 4 The function convergence curve provided in the embodiments of the present invention;

[0049] Figure 5 This is a schematic diagram of the structure of a historical figure activation computing resource scheduling system based on an improved artificial bee colony algorithm provided in an embodiment of the present invention;

[0050] Figure 6 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0052] In real-world historical figure re-enactment scenarios, the system needs to perform concurrent modeling tasks on multimodal data such as images, speech, and poses, including 3D facial reconstruction, speech synthesis, and motion fitting. These tasks rely on different computing resources; for example, image reconstruction primarily uses GPUs for tensor computation and graphics rendering, speech synthesis may involve DSPs or dedicated speech processors, and motion simulation relies more on CPUs for control reasoning and logical judgment. How to accurately match and schedule tasks based on task type and resource capabilities is the key bottleneck to the efficient computing performance of such re-enactment systems.

[0053] Therefore, achieving low-latency and highly adaptable task scheduling on heterogeneous platforms has become a key requirement for historical figure re-enactment systems. The artificial bee colony algorithm, due to its strong global search capability and parallel adaptability, has good optimization potential in dynamic and complex resource environments and is suitable for resource allocation problems in historical figure modeling tasks.

[0054] In response to the above problems, such as Figure 1 As shown, this embodiment of the invention provides a method for scheduling computing resources for historical figure activation based on an improved artificial bee colony algorithm, including the following steps:

[0055] S101: Based on the types of tasks that need to be performed during the revitalization of historical figures and the available computing resources, construct a task set and a resource set;

[0056] Specifically, this embodiment of the invention models three typical tasks in the revitalization of historical figures: (1) image reconstruction tasks, such as 3D facial modeling and video coloring, which mainly rely on GPU resources; (2) behavior-driven tasks, such as posture simulation and motion fitting, which mainly rely on CPU resources; and (3) audio generation tasks, such as speech synthesis and lip-syncing, which may use DSP or other acceleration modules (such as DCU). Specifically, the task set (Task) represents the computational tasks to be scheduled, defined as Task = {T1, T2, T3....T n}, where T i Let i be the i-th computational task. The resource set (Cluster) represents heterogeneous computing resources, defined as Cluster = {C1, C2, C3...C}. m}, where C j This is the j-th computing resource.

[0057] In this embodiment, a task-resource allocation matrix (TC) is used to represent the allocation relationship between computing tasks and computing resources:

[0058]

[0059] Among them, TC ij The value can be 0 or 1, TC ij =1 indicates task T i Assigned to resource C j Up; TC ij =0 indicates task T i Not allocated to resource C j Therefore, the goal of this invention is to find an optimal resource allocation scheme through a scheduling algorithm to maximize the utilization of computing resources while ensuring that the computational requirements of the task are met.

[0060] This embodiment treats the allocation of tasks and resources as an optimization problem, with each possible scheduling scheme corresponding to a solution, denoted as X. i Among them, X i Let represent the i-th task scheduling scheme (i.e., a specific allocation scheme of tasks and resources). All possible solutions constitute the solution space, where SN represents the number of solutions.

[0061] Based on the aforementioned optimization problem, this embodiment designs an improved artificial bee colony algorithm as the scheduling algorithm to solve the optimization problem (as shown in steps S102 to S105). That is, by searching and optimizing in the solution space based on the improved artificial bee colony algorithm, an optimal task scheduling scheme X is finally found. * This minimizes the cost of using computing resources while improving resource utilization.

[0062] S102: Randomly initialize the honey source set; wherein, each honey source in the honey source set represents a task scheduling scheme, and one dimension of each honey source corresponds to the allocation decision between a certain task in the task set and a certain resource in the resource set;

[0063] Specifically, each honey source is represented as a vector X = [x1, x2, ..., x...]. D ], where D is the dimension of task scheduling, i.e., the number of tasks to be scheduled in the system, x i ∈{1,2,...,m} represents task T i Allocated to resources In other words, each dimension of the honey source represents a specific allocation decision between a certain task and a certain computing resource.

[0064] S103: Select a nectar source from the nectar source set;

[0065] S104: For the current honey source, a multi-dimensional search strategy is used to update the honey source and obtain a new honey source; the multi-dimensional search strategy includes: selecting exploitable dimensions from the current honey source and perturbing the selected dimensions; the perturbation refers to changing a certain task T i Resource allocation from C xi Replace with another resource C j .

[0066] For example, in one iteration, for the i-th honey source X i The d-th dimension x id (d = 1, 2, ..., D), in the current nectar source X i A search was conducted nearby, and a new nectar source was identified as X′. i =(x′) i1 ,x′ i2 ,...,x′ ij ,...,x′ iD It should be noted that if the new value x′ found in the search... id Beyond the limits of this dimension [x idmin ,x idmax If x′ > 0, then the new value should be adjusted accordingly, specifically: if x′ > 0. id <x idmin Then x′ id =x idmin If x′ id >x idmax Then x′ id =x idmax .

[0067] S105: Calculate the fitness of the new honey source using a pre-built resource scheduling cost function. If the fitness is improved, the new honey source is adopted, and the perturbation dimension is marked as a mineable dimension.

[0068] Specifically, the fitness of each nectar source is calculated through a resource scheduling cost function, which is used to measure the quality of the current scheduling scheme. If the updated scheduling scheme causes the value of the resource scheduling cost function to decrease, it indicates an improvement in fitness.

[0069] In actual applications, after each iteration, check whether there are updated dimensions for each nectar source. If certain dimensions have been optimized, mark these dimensions as to-be-mined dimensions and store them in the temporary storage area temp for use in the next iteration. For example, given the nectar source X before update and the nectar source X' after update, and the resource scheduling cost function Cost. If a certain dimension d satisfies Cost(X') < Cost(X) after update, it means the new方案 is better. Then perform the following operations: deposit d into the temporary storage area temp and set the flag bit flag = 1 to mark that this nectar source still has the potential for optimization in this round of iteration; at the same time, append this dimension d to the dimension record array Dim(i, Num) and update the counter Num += 1 to facilitate subsequent statistics of the number of dimensions that this nectar source can be continuously optimized.

[0070] In addition, if a dimension fails to be updated in multiple iterations, increase the number of non-updated times trial. If trial reaches the set threshold, trigger a full-dimensional re-search of this nectar source to jump out of the local optimum trap. Specifically, after the search of each dimension, judge whether there is an effective update according to the flag bit flag. If flag = 1, it means at least one dimension has been optimized, and reset the number of attempts trial(i) of the non-updated nectar source to 0; otherwise, increase the number of attempts. If trial(i) reaches the preset maximum number of attempts Limit, re-execute the search of each dimension of this nectar source.

[0071]

[0072] S106: Repeat steps S103 to S105 until the stop condition is reached, and output the current latest nectar source as the optimal task scheduling scheme.

[0073] Specifically, after multiple iterations, the dimensions in the temporary storage area temp will gradually decrease and finally converge to an optimal solution. At this time, it is considered that the stop condition has been reached, and the optimal solution at this time is used as the optimal scheduling scheme of the current task set on heterogeneous resources.

[0074] Figure 2 ​This demonstrates the initial update operation based on an improved artificial bee colony algorithm. In the initial iteration, the algorithm updates and searches for each dimension of the nectar source (i.e., the specific allocation of tasks and resources). If a better solution is found in a certain dimension, the update result for that dimension is stored and used in subsequent iterations. This process searches for optimal solutions by exploring local regions in the solution space; if no better solution is found in some dimensions, the algorithm increments the unupdated count by 1 and performs a full search of all dimensions again in the next iteration. This avoids premature convergence to local optima while ensuring the comprehensiveness of the search.

[0075] Figure 3 This demonstrates the second iteration of the artificial bee colony algorithm based on an optimization strategy, where only the dimensions for which better solutions were found in the previous iteration are updated. If better solutions are found in dimensions 2, 5, and 7, they are stored in `temp` and optimized again in the next iteration. The next iteration will only update these dimensions, not all dimensions; if no better solutions are found in dimensions 2, 5, and 7, the algorithm increments the count of the unupdated honey source by 1 and searches all dimensions again in the next iteration.

[0076] This invention proposes a resource scheduling method for historical figure activation based on an improved artificial bee colony algorithm. By optimizing the artificial bee colony algorithm and combining it with a multi-dimensional search strategy, heterogeneous computing resources are scheduled in a simulated environment. Furthermore, the global search capability of the scheduling is effectively improved through a fitness evaluation mechanism, significantly accelerating the convergence speed and ultimately achieving the goal of improving the speed of character and activation modeling.

[0077] The following example, a "Historical Figure Restoration and Interaction System," will further illustrate the purpose of this invention. Specifically, in practical applications, this system includes the following typical concurrent tasks:

[0078] (1) Human Image Reconstruction Task T1: Generate 3D head model from historical photos, mainly relying on GPU with large video memory, and calling ResNet-CNN rendering module;

[0079] (2) Speech style synthesis task T2: Synthesize character voices based on historical texts, call the Tacotron2 model, and require a DCU that supports FP16 operations;

[0080] (3) Pose Fitting Driven Task T3: Match the movement trajectory of people in old images, call the OpenPose module, and mainly rely on the CPU for logical reasoning and control judgment.

[0081] According to the present invention, in the process of scheduling information modeling, the task set can be defined as Task = {T1, T2, T3}, and the resource set can be defined as Cluster = {C1(GPU), C2(DCU), C3(CPU)}. This is achieved by constructing a task-resource allocation matrix TC. ij And calculate the scheduling cost function Cost based on the computing power, bandwidth, and load characteristics of heterogeneous resources.

[0082] Taking a certain initial scheme as an example, if T1→C2, T2→C3, T3→C1, resources and tasks are severely mismatched, and scheduling costs are high. After optimization and iteration by improving the artificial bee colony algorithm, it can converge to the optimal scheme T1→C1, T2→C2, T3→C3, achieving precise matching of resource capabilities and task types, thereby significantly reducing scheduling losses and improving the operating efficiency of the activation system.

[0083] In one embodiment, to further optimize the scheduling scheme of computational tasks, this embodiment of the invention establishes a heterogeneous resource scheduling cost model for historical figure revitalization tasks. This model comprehensively considers the performance loss and resource matching degree brought about by scheduling and executing tasks on different resources, achieving dynamic quantification of scheduling costs. The model construction process is as follows: A loss index is determined by comprehensively considering the performance loss and resource matching degree brought about by scheduling and executing tasks on different types of resources; the determined loss index is weighted and combined to obtain the total scheduling loss; the resource waste rate is calculated; and the total scheduling loss and the resource waste rate are weighted and combined to obtain the scheduling cost function.

[0084] In this embodiment, the loss indicators include: computing power loss, data transmission time loss, network bandwidth loss, resource type mismatch loss, other losses, and resource waste rate; wherein, the computing power loss is used to measure whether the current resource computing power can meet the task requirements; the data transmission time loss is used to indicate the time cost required to transmit task input data from the data source to the computing node; the network bandwidth loss is used to indicate the network bandwidth load of the resources; the resource type mismatch loss is used to penalize scheduling behavior that does not match the task type and resource capacity; the other losses include task switching overhead and scheduling latency; and the resource waste rate is used to indicate the proportion of resources that are not effectively utilized.

[0085] The calculation formulas for the above loss indicators are as follows:

[0086]

[0087] in, Represents task T i With resource C j The computing power loss between them, Comp(T) i ) for task T iRequired computation, Power(C) j ) is for resource C j Currently available computing power;

[0088]

[0089] in, Represents task T i With resource C j Data transmission time loss between Size(T) i ) for task T i Data volume, BW(C j ) is for resource C j The current bandwidth;

[0090]

[0091] in, Indicates resource C j Network bandwidth loss, Load(C j ) is for resource C j Network transmission requirements, BWmax(C j ) is for resource C j The maximum theoretical bandwidth;

[0092]

[0093] in, Represents task T i With resource C j Resource type mismatch loss For a preset penalty value (e.g., 0.5~1), type(T) i ) represents task T i Types, such as images, voice, and actions; cap(C j ) represents resource C j Capability tags, such as GPU / CPU / DCU, etc.; type(T i )~cap(C j ) represents task T i With resource C j They match.

[0094] Furthermore, based on the above, the determined loss indicators are weighted and combined to obtain the total scheduling loss, specifically including:

[0095]

[0096] Where Loss represents the total scheduling loss, and w1, w2, w3, w4, and w5 are adjustable weights.

[0097] Furthermore, the resource scheduling cost function Cost is:

[0098] Cost = Loss + w6·Loss RU

[0099] Where w6 is an adjustable weight; Loss RU Indicates the rate of resource waste. R use R represents the amount of computing resources actually used by the current task. ava This represents the total available resource capacity of the current resource node. The resource waste rate reflects the proportion of resources that are not effectively utilized and serves as the dual term of the resource utilization rate.

[0100] In this embodiment, the resource scheduling cost function is uniformly oriented towards minimization, serving as the input to the fitness function of the artificial bee colony algorithm, guiding the task-resource evolution towards higher resource utilization and lower performance loss. Based on this cost model, this invention can achieve efficient coordinated scheduling of multiple types of heterogeneous computing resources in historical figure revitalization tasks.

[0101] For historical figure re-enactment systems, multiple heterogeneous tasks such as image modeling, behavior simulation, and speech generation (e.g., voice-over synthesis) typically need to be executed concurrently. These tasks have varying dependencies on resources such as GPUs, CPUs, and DCUs, and there are data dependencies and synchronization constraints between tasks. The historical figure re-enactment computing resource scheduling method based on an improved artificial bee colony algorithm provided in this invention can dynamically adjust the allocation scheme according to task type and resource characteristics, improve the global search capability and resource utilization of the scheduling, and accelerate the speed of figure re-enactment generation.

[0102] This invention's method is particularly applicable to historical figure revitalization systems whose core tasks include "figure restoration," "multimodal synthesis," and "virtual interaction." For example, virtual exhibition systems in smart museums, historical education video reconstruction platforms, and interactive applications of cultural IP all face the problem of low efficiency in scheduling heterogeneous resources. Based on the optimized scheduling mechanism of this invention, such bottlenecks can be effectively solved.

[0103] In one embodiment, partial pseudocode of the heterogeneous resource scheduling algorithm based on the multi-dimensional search strategy artificial bee colony algorithm designed in this invention is given, as shown in Algorithm 1.

[0104]

[0105]

[0106] To verify the proposed Improved Artificial Bee Colony (IABC) algorithm's good optimization performance and robustness in the computational resource scheduling task of historical figure activation scenarios, a series of standard function optimization experiments were designed. Although the test functions used in this experiment are classical mathematical functions, their characteristics are analogous to typical computational challenges in the historical figure activation process, including high-dimensional task complexity, nonlinear perturbations in resource scheduling paths, and local optima traps in multi-task interactions. Specifically:

[0107] (1) In the task of revitalizing historical figures, it is necessary to deal with the historical interaction relationships between a large number of different figures. Their behavior is constrained by multiple dimensions such as time, events, and space, which can easily form a high-dimensional scheduling space.

[0108] (2) The records of various figures are subject to nonlinear disturbances such as disordered time sequence, conflicting versions, and missing information, which makes it easy for the behavior inference path to fall into local optima.

[0109] (3) During the activation process, computing resources need to be allocated reasonably to support the efficient completion of complex tasks such as virtual reconstruction, character behavior synthesis, and interactive modeling.

[0110] Therefore, four typical functions—Sphere, Rosenbrock, Rastrigin, and Griewank—were selected to map the aforementioned challenge scenarios, thereby indirectly evaluating the global search capability, convergence speed, and solution stability of the IABC algorithm in the historical figure activation scheduling task. To facilitate the explanation of the correspondence between the above functions and the scheduling problem, Table 1 provides the definition of each test function and its mapping significance in the historical figure activation task.

[0111] Table 1 Standard Test Functions

[0112]

[0113] The experiment was conducted using the MATLAB software platform on a computer with a CPU of 3.3 GHz and 2 GB of memory, focusing on the following aspects:

[0114] (a) By varying the population size N and the number of iterations G, and using the optimization results as the evaluation index, the impact of the values ​​of algorithm parameters N and G on the performance of IABC was tested.

[0115] (b) Specify the number of algorithm iterations and compare the performance of IABC and ABC using solution accuracy and convergence speed as evaluation metrics. Here, IABC represents the improved artificial bee colony algorithm proposed in this invention (the same applies below), and ABC represents the unimproved artificial bee colony algorithm (the same applies below).

[0116] Four test functions were solved using IABC and ABC respectively, with G = 3000 iterations, N = 30 population size, and Limit = 100 food source update threshold. Ten experiments were conducted with D values ​​of 10, 30, 50, and 80 for each function. The mean, standard deviation, and average number of iterations to the optimal value were recorded. The experimental results are shown in Table 2, where M represents the mean, S represents the standard deviation, and I represents the average number of iterations.

[0117] Table 2 Performance Comparison of IABC and ABC Algorithms

[0118]

[0119]

[0120] Under the above parameter settings, the iterative process of the four functions with a dimension of 30 is as follows: Figure 4 As shown. By Figure 4 As can be seen from the data in Table 2, for the four test functions, the improved Artificial Bee Colony Algorithm (IABC) significantly outperforms the standard ABC algorithm in both optimization accuracy and convergence speed, demonstrating stronger global search capability and convergence stability.

[0121] For unimodal functions like Sphere and Rosenbrock, the advantage of IABC is particularly evident at high dimensions. When the Sphere function reaches 80 dimensions, the ABC algorithm failed to converge to the optimal value after 3000 iterations, while the IABC algorithm reached final convergence after just over 400 iterations. Furthermore, when the Rosenbrock function exceeds 30 dimensions, the IABC algorithm significantly surpasses the ABC algorithm in terms of solution accuracy.

[0122] For the multimodal function Ratrigin, the IABC algorithm can find the optimal value of 0 under different dimension settings, and it has a significant advantage over the ABC algorithm in terms of the number of iterations required to find the optimal value. For the Griewank function, when the dimension is less than 30, it is a complex multimodal function, and IABC has found the optimal value of 0 in more than 300 iterations. When the dimension is higher than or equal to 30 and it becomes a unimodal function, the optimal value of 0 is found in about 150 iterations.

[0123] Although standard test functions are not directly derived from character modeling tasks, they have good structural isomorphism with scheduling problems in multimodal tasks in terms of optimization structure and data perturbation simulation, providing theoretical basis and methodological verification for the migration of practical systems.

[0124] In summary, the IABC algorithm proposed in this invention demonstrates excellent global search capability and convergence stability in standard function tests. Each test function, in terms of structure and characteristics, maps to the typical computational challenges in historical figure revitalization tasks, providing concrete model support for scheduling strategy design. Therefore, the experimental results can be transferred and applied to resource scheduling optimization in practical systems. In complex systems with concurrent execution of multimodal tasks and heterogeneous resource constraints, the IABC algorithm exhibits good task adaptability, effectively reducing system scheduling costs and improving overall scheduling efficiency, providing stable and efficient computational support for historical figure revitalization systems.

[0125] Based on the same inventive concept, such as Figure 5 As shown, this embodiment of the invention provides a historical figure activation computing resource scheduling system based on an improved artificial bee colony algorithm, including a scheduling information construction module, an initialization module, and an iterative search module.

[0126] The scheduling information construction module is used to construct a task set and a resource set based on the task types to be executed during the historical figure activation process and the available computing power resources. The initialization module is used to randomly initialize the honey source set. Each honey source in the honey source set represents a task scheduling scheme, and one dimension of each honey source corresponds to the allocation decision between a certain task in the task set and a certain resource in the resource set. The iterative search module is used to select honey sources from the honey source set. For the current honey source, a multi-dimensional search strategy is used to update the honey source and obtain a new honey source. The fitness of the new honey source is calculated using a pre-constructed scheduling cost function. If the fitness is improved, the new honey source is adopted, and the perturbed dimension is marked as a mineable dimension. Otherwise, the next iteration is performed until the stopping condition is met. The latest honey source is output as the optimal task scheduling scheme. The multi-dimensional search strategy includes: selecting a mineable dimension from the current honey source and perturbing the selected dimension.

[0127] It should be noted that the historical figure activation computing resource scheduling system based on the improved artificial bee colony algorithm provided in this embodiment of the invention is for implementing the above method. Its specific functions can be referred to in the above method embodiments, and will not be repeated here.

[0128] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6As shown, the electronic device may include: a processor 601, a communication interface 602, a memory 603, and a communication bus 604, wherein the processor 601, the communication interface 602, and the memory 603 communicate with each other through the communication bus 604. Processor 601 can call logical instructions in memory 603 to execute a historical figure activation computing resource scheduling method based on an improved artificial bee colony algorithm. This method includes: Step 1: Constructing a task set and a resource set based on the task types to be executed during the historical figure activation process and the available computing resources; Step 2: Randomly initializing a honey source set; wherein each honey source in the honey source set represents a task scheduling scheme, and one dimension in each honey source corresponds to the allocation decision between a task in the task set and a resource in the resource set; Step 3: Selecting a honey source from the honey source set; Step 4: Updating the current honey source using a multi-dimensional search strategy to obtain a new honey source; the multi-dimensional search strategy includes: selecting a mineable dimension from the current honey source and perturbing the selected dimension; Step 5: Calculating the fitness of the new honey source using a pre-constructed resource scheduling cost function; if the fitness increases, the new honey source is adopted, and the perturbed dimension is marked as a mineable dimension; otherwise, the next iteration is performed; Step 6: Repeating steps 3 to 5 until a stopping condition is met, and outputting the latest honey source as the optimal task scheduling scheme.

[0129] Furthermore, when the logical instructions in the aforementioned memory 603 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0130] This invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer can execute the historical figure activation computing resource scheduling method based on the improved artificial bee colony algorithm provided in the above-described method embodiments.

[0131] This invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the historical figure activation computing resource scheduling method based on the improved artificial bee colony algorithm provided in the above-described method embodiments.

[0132] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for scheduling computing resources for the activation of historical figures based on an improved artificial bee colony algorithm, characterized in that: include: Step 1: Based on the types of tasks that need to be performed during the revitalization of historical figures and the available computing resources, construct a task set and a resource set; Step 2: Randomly initialize the honey source set; wherein, each honey source in the honey source set represents a task scheduling scheme, and one dimension of each honey source corresponds to the allocation decision between a certain task in the task set and a certain resource in the resource set; Step 3: Select nectar sources from the nectar source collection; Step 4: For the current nectar source, use a multi-dimensional search strategy to update the nectar source and obtain a new nectar source; the multi-dimensional search strategy includes: selecting exploitable dimensions from the current nectar source and perturbing the selected dimensions; Step 5: Calculate the fitness of the new honey source using the pre-built resource scheduling cost function. If the fitness is improved, adopt the new honey source and mark the perturbation dimension as a mineable dimension. Step 6: Repeat steps 3 to 5 until the stopping condition is met, and output the latest honey source as the optimal task scheduling scheme.

2. The method for scheduling computing resources for historical figure activation based on the improved artificial bee colony algorithm according to claim 1, characterized in that, The process of constructing the scheduling cost function includes: The performance loss caused by scheduling and executing tasks on different types of resources and the degree of resource matching are comprehensively considered to determine the loss index; the determined loss index is weighted and combined to obtain the total scheduling loss; the resource waste rate is calculated; the total scheduling loss and the resource waste rate are weighted and combined to obtain the scheduling cost function.

3. The method for scheduling computing resources for historical figure activation based on the improved artificial bee colony algorithm according to claim 1, characterized in that, The loss metrics include: computing power loss, data transmission time loss, network bandwidth loss, resource type mismatch loss, other losses, and resource waste rate. The computing power loss measures whether the current computing power can meet the task requirements. The data transmission time loss indicates the time required to transmit task input data from the data source to the computing node. The network bandwidth loss indicates the network bandwidth load of the resources. The resource type mismatch loss penalizes scheduling behavior that does not match the task type and resource capacity. Other losses include task switching overhead and scheduling latency. The resource waste rate indicates the proportion of resources that are not effectively utilized.

4. The method for allocating computing resources for historical figure activation in the improved artificial bee colony algorithm according to claim 3, characterized in that, The total scheduling loss is obtained by weighting and combining the determined loss indicators, specifically including: Where Loss represents the total scheduling loss, and w1, w2, w3, w4, and w5 are adjustable weights; Correspondingly, the values ​​of each loss index are calculated according to the following formulas: in, Indicates task T i With resource C j The computing power loss between them, Comp(T) i ) for task T i Required computation, Power(C) j ) for resource C j Currently available computing power; in, Indicates task T i With resource C j Data transmission time loss between Size(T) i ) for task T i Data volume, BW(C j ) for resource C j The current bandwidth; in, Indicates resource C j Network bandwidth loss, Load(C j ) for resource C j Network transmission requirements, BWmax(C j ) for resource C j The maximum theoretical bandwidth; in, Indicates task T i With resource C j Resource type mismatch loss To preset the penalty value, type(T) i ) represents task T i The type, cap(C j ) represents resource C j Ability tags, type(T) i )~cap(C j ) represents task T i With resource C j They match.

5. The method for allocating computing resources for historical figure activation based on the improved artificial bee colony algorithm according to claim 4, characterized in that, The resource scheduling cost function Cost is: Cost=Loss+w6·Loss RU Where w6 is an adjustable weight; Loss RU Indicates the rate of resource waste. R use R represents the amount of computing resources actually used by the current task. ava This indicates the total available resource capacity of the current resource node.

6. A historical figure activation computing resource scheduling system based on an improved artificial bee colony algorithm, characterized in that: include: The scheduling information construction module is used to construct task sets and resource sets based on the types of tasks to be executed during the revitalization of historical figures and the available computing resources. An initialization module is used to randomly initialize a honey source set; wherein, each honey source in the honey source set represents a task scheduling scheme, and one dimension of each honey source corresponds to the allocation decision between a certain task in the task set and a certain resource in the resource set; An iterative search module is used to select honey sources from the honey source set; for the current honey source, a multi-dimensional search strategy is used to update the honey source to obtain a new honey source; the fitness of the new honey source is calculated using a pre-built scheduling cost function; if the fitness is improved, the new honey source is adopted and the perturbation dimension is marked as a mineable dimension; otherwise, the next iteration is performed until the stopping condition is reached; the latest honey source is output as the optimal task scheduling scheme; wherein, the multi-dimensional search strategy includes: selecting a mineable dimension from the current honey source and perturbing the selected dimension.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.

Images