Task allocation method and system based on extreme learning machine and salp swarm optimization algorithm
Through the extreme learning machine collaborative optimization algorithm, combined with multi-stage allocation and correlation merge, the problem of poor matching of tasks and workers in crowdsourcing task allocation is solved, the accuracy and efficiency of task allocation is achieved, and users are improved Satisfaction.
Patent Information
- Application Number
- CN202411320580.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-23
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-23
AI Technical Summary
In crowdsourcing task allocation, it is difficult for the existing technology to accurately match the characteristics of tasks and workers, resulting in low task allocation efficiency and low user satisfaction, affecting the attractiveness and sustainable development of the platform.
The extreme learning machine collaborative sea squirt group optimization algorithm is adopted to obtain the properties of workers and tasks, build position, time and preference models, combine Spearman's correlation coefficient and local aggregation method to perform multi-stage task allocation to ensure accurate task docking To the most suitable worker.
It achieves the accuracy and efficiency of task allocation, maximizes the expectation of completing the number of tasks, ensures that tasks can be accurately connected to the most suitable workers, and improves user satisfaction and platform attractiveness.
Smart Images

Figure CN118863468B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of crowdsourcing task allocation, and specifically relates to a task allocation method and system for an extreme learning machine collaborative salp swarm optimization algorithm. Background Art
[0002] In terms of crowdsourcing task allocation, in order to give full play to the advantages of the crowdsourcing model, the crowdsourcing platform needs to accurately grasp the fit between task characteristics and worker capabilities and achieve intelligent matching of tasks. The effectiveness of task allocation is closely linked to user satisfaction, and has become one of the key indicators for measuring the attractiveness and sustainable development of crowdsourcing platforms. User satisfaction is directly related to whether they will continue to choose and recommend the crowdsourcing application in the future, which in turn has a profound impact on the prosperity and development of the entire crowdsourcing ecosystem. Therefore, how to optimize the task allocation mechanism and ensure that tasks can be accurately matched to the most suitable workers has become an important problem that needs to be overcome in the crowdsourcing field. Summary of the invention
[0003] In order to solve the problems raised by the background technology, the present invention provides a task allocation method and system of an extreme learning machine collaborative salp swarm optimization algorithm.
[0004] The technical solution of the present invention is as follows:
[0005] The present invention provides a task allocation method for an extreme learning machine-assisted salp swarm optimization algorithm, comprising the following steps:
[0006] S1: Get the properties of workers and tasks;
[0007] Worker attributes, including: worker location, worker activity area, worker movement speed, worker preferences, and worker task execution history;
[0008] Task attributes, including: task location, task start time, task end time, and task category;
[0009] S2: Based on the attributes of workers and tasks, construct the objective function, specifically:
[0010] S21: Based on the attributes of workers and tasks, build location models, time models, and preference models as constraints;
[0011] S22: construct the objective function according to the constraints;
[0012] S3: Based on the objective function, the first stage allocation is performed, specifically:
[0013] S31: After initializing the parameters of the salp swarm optimization algorithm, several subpopulations are initialized;
[0014] S32: Calculate the fitness value according to the objective function, sort the salps according to the fitness value, and record the position information of each salp;
[0015] S33: After the sorted salps are updated by the salp group optimization algorithm, S32 is executed until the number of iterations reaches a first preset number of iterations, and the first stage allocation result is obtained;
[0016] S4: Based on the results of the first-stage allocation, the extreme learning machine is trained to perform the second-stage allocation, specifically:
[0017] S41: Based on the allocation results of the first stage, the Spearman correlation coefficients between several sub-populations are calculated, and the sub-populations are merged according to the Spearman correlation coefficients;
[0018] S42: Based on the allocation result of the first stage, train an extreme learning machine to obtain a trained extreme learning machine;
[0019] S43: Based on the merged sub-population, the fitness value is calculated according to the objective function, the salps individuals are sorted according to the fitness value, and the position information of each salp individual is recorded;
[0020] S44: after the sorted salps are updated by the salp swarm optimization algorithm or the trained extreme learning machine according to the first probability, S43 is executed until the number of iterations reaches a second preset number of iterations, and a second-stage allocation result is obtained;
[0021] S5: Based on the second stage allocation result, according to the second probability, the trained extreme learning machine or local aggregation method is used to update and obtain the third allocation result, which is specifically:
[0022] S51: Based on the second stage allocation result, the trained extreme learning machine or local aggregation method is updated according to the second probability;
[0023] S52: Calculate the fitness value according to the objective function, sort the salps according to the fitness value, record the position information of each salp, and execute S51 until the number of iterations reaches the third preset number of iterations to obtain the maximum fitness value. The position of the salp group individual corresponding to the maximum fitness value is used as the third stage allocation result.
[0024] In step S21, a location model is constructed based on the attributes of the worker and the task, specifically:
[0025] According to the formula: , build a location model;
[0026] In the formula, Indicates workers In time period Internal satisfaction task The probability of the position of Is a worker and tasks The distance between It is the largest activity area for workers.
[0027] In step S21, a time model is constructed based on the attributes of the worker and the task, specifically:
[0028] According to the formula: , construct a time model;
[0029] In the formula, Indicates the time period in which the task is located; For workers and tasks The distance between For workers Movement speed; For the task The start time of For the task The end time of .
[0030] In step S21, a preference model is constructed based on the attributes of workers and tasks, specifically:
[0031] According to the formula: ,
[0032] , build a preference model;
[0033] In the formula, Indicates workers In time period Internal satisfaction task Task category Probability of preference; For workers In time period The total number of tasks executed within For workers The task execution history; Indicates workers Execution task category frequency; is the category of the task to be performed; For the task The time of start; For the task End time.
[0034] In step S22, an objective function is constructed according to the constraint conditions, specifically:
[0035] According to the formula: ,
[0036] , construct the objective function;
[0037] In the formula, represents the expectation of maximizing the number of completed tasks; is the total number of tasks; is the total number of workers; Indicates workers Complete the task probability; If If established, The value is 1; Indicates workers In time period Internal satisfaction task The probability of the position of Indicates workers In time period Internal satisfaction task Task category probability.
[0038] In step S41, based on the allocation result of the first stage, the Spearman correlation coefficient between several sub-populations is calculated, specifically:
[0039] According to the formula: , calculate the Spearman correlation coefficient between several subpopulations;
[0040] In the formula, is the Spearman correlation coefficient; is the difference in fitness ranking of individual salps in the two subpopulations; is the number of salps individuals in the subpopulation; is the number of subpopulations.
[0041] In step S51, the updating is performed by the local aggregation method, specifically:
[0042] According to the formula: ,
[0043] , to update;
[0044] In the formula, After the update individuals in a subpopulation; For the The best individual in a subpopulation; For the Suboptimal individuals in a subpopulation; To adjust the parameters of the aggregation scope, according to the formula: ,get;
[0045] In the formula, is the current iteration number, The third preset number of iterations.
[0046] The first probability in step S44 or the second probability in step S5 is a random number between [0, 1].
[0047] The present invention also provides a task allocation system for an extreme learning machine-assisted salp swarm optimization algorithm, comprising:
[0048] Get module: used to get the properties of workers and tasks;
[0049] Worker attributes, including: worker location, worker activity area, worker movement speed, worker preferences, and worker task execution history;
[0050] Task attributes, including: task location, task start time, task end time, and task category;
[0051] Objective function building module: Based on the attributes of workers and tasks, the objective function is built, specifically:
[0052] Based on the attributes of workers and tasks, we build location models, time models, and preference models as constraints;
[0053] Construct the objective function according to the constraints;
[0054] The first stage allocation module: Based on the objective function, the first stage allocation is carried out, specifically:
[0055] After initializing the parameters of the salp swarm optimization algorithm, several subpopulations are initialized;
[0056] The fitness value is calculated according to the objective function, the salps are sorted according to the fitness value, and the position information of each salp is recorded;
[0057] After the sorted salps are updated by the salp group optimization algorithm, the fitness values are calculated and sorted until the number of iterations reaches a first preset number of iterations, and the first stage allocation result is obtained;
[0058] Second-stage allocation module: Based on the first-stage allocation results, the extreme learning machine is trained to perform the second-stage allocation, specifically:
[0059] Based on the allocation results of the first stage, the Spearman correlation coefficients between several sub-populations are calculated, and the sub-populations are merged according to the Spearman correlation coefficients;
[0060] Based on the allocation result of the first stage, an extreme learning machine is trained to obtain a trained extreme learning machine;
[0061] Based on the merged sub-population, the fitness value is calculated according to the objective function, the salps are sorted according to the fitness value, and the position information of each salp is recorded;
[0062] According to the first probability, the sorted salps are updated by the salps swarm optimization algorithm or the trained extreme learning machine, and the fitness values are calculated and sorted until the number of iterations reaches a second preset number of iterations, thereby obtaining the second stage allocation result;
[0063] The third stage allocation module: based on the second stage allocation result, according to the second probability, the trained extreme learning machine or local aggregation method is used to update the third allocation result, which is specifically:
[0064] Based on the second stage allocation result, the trained extreme learning machine or local aggregation method is updated according to the second probability;
[0065] The fitness value is calculated according to the objective function, and the salps are sorted according to the fitness value. The position information of each salp is recorded and updated according to the second probability by the trained extreme learning machine or local aggregation method until the number of iterations reaches the third preset number of iterations to obtain the maximum fitness value. The position of the salp group individual corresponding to the maximum fitness value is used as the allocation result of the third stage.
[0066] Beneficial effects: The present invention constructs an objective function based on the attributes of workers and tasks, and sequentially performs three-stage task allocation to obtain a crowdsourcing task allocation plan that maximizes the objective function, wherein, in the first stage allocation, the multidimensional crowdsourcing task allocation optimization problem is converted into a one-dimensional optimization problem, and the first stage allocation result is obtained by the salp swarm optimization algorithm; in the second stage allocation, the dimensions with high correlation are merged into a population, which is updated by the salp swarm optimization algorithm or the trained extreme learning machine to obtain the second stage allocation result; in the third stage allocation, the third stage allocation result is updated by the trained extreme learning machine or the local aggregation method to obtain the third stage allocation result, that is, the optimal task allocation plan, which ensures that the tasks can be accurately connected to the most suitable workers and maximizes the expectation of completing the number of tasks. BRIEF DESCRIPTION OF THE DRAWINGS
[0067] Figure 1 It is a flowchart of a task allocation method of an extreme learning machine collaborative salp swarm optimization algorithm in an embodiment. DETAILED DESCRIPTION
[0068] The following examples are intended to illustrate the present invention rather than to further limit the present invention.
[0069] The present invention provides a task allocation method for an extreme learning machine collaborative salp swarm optimization algorithm, such as Figure 1 As shown, the following steps are included:
[0070] S1: Get the properties of workers and tasks;
[0071] The attributes of a worker include: the worker's location, the worker's activity area, the worker's movement speed, the worker's preferences, and the worker's task execution history.
[0072] The specific expressions are as follows:
[0073] represents the existing pool of workers, Indicates A worker, ,and . Workers use It indicates that, Is a worker location, It is the activity area of the workers. is the worker's movement speed.
[0074] Regarding the worker's task execution history, given a worker , which has been executed tasks, then the worker's task execution history is a task collection, which is express, .
[0075] The attributes of a task include: the location of the task, the start time of the task, the end time of the task, and the category of the task.
[0076] The specific expressions are as follows:
[0077] Represents the existing task pool, Indicates tasks, ,and . For tasks It indicates that, It's a task location, is the start time of the task, is the end time of the task, Is the type of task.
[0078] S2: Based on the attributes of workers and tasks, construct the objective function, specifically:
[0079] S21: Based on the attributes of workers and tasks, build location models, time models, and preference models as constraints.
[0080] Preferably, a location model is constructed based on the attributes of workers and tasks. In time period Internal satisfaction task The probability of the position , for workers and tasks The distance between workers The ratio of the maximum active area is:
[0081] According to the formula: , build a location model;
[0082] In the formula, Indicates workers In time period Internal satisfaction task The probability of the position of Is a worker and tasks The distance between It is the largest activity area for workers.
[0083] Preferably, a time model is constructed based on the attributes of workers and tasks to satisfy the workers' In time period Satisfy the arrival task Time on task Before the end, specifically:
[0084] According to the formula: , construct a time model;
[0085] In the formula, Indicates the time period in which the task is located; For workers and tasks The distance between For workers Movement speed; For the task The start time of For the task The end time of .
[0086] Preferably, a preference model is constructed based on the attributes of workers and tasks. In time period Internal satisfaction task Task category Probability of preference , first calculate the workers Execution task category The frequency of this preference is the probability that the worker In time period The execution belongs to the category The number of tasks and their time periods The ratio of the total number of tasks executed within a period is:
[0087] According to the formula: ,
[0088] , build a preference model;
[0089] In the formula, Indicates workers In time period Internal satisfaction task Task category Probability of preference; For workers In time period The total number of tasks executed within For workers The task execution history; Indicates workers Execution task category frequency; is the category of the task to be performed; For the task The time of start; For the task End time.
[0090] Since space-time tasks have limitations such as short life cycle, remote location, and single type, it is necessary to maximize the expected number of completed space-time tasks under these limitations.
[0091] S22: Construct an objective function based on the constraints.
[0092] Given a set of spatiotemporal tasks to be assigned T and a group of workers W The goal is to consider the location, preferences and time of each worker and find a task assignment set A so that A maximizes the expected number of completed tasks while satisfying the constraints.
[0093] In order to optimize the task allocation mechanism and ensure that tasks can be accurately assigned to the most suitable workers, the objective function is the expected function that maximizes the number of crowdsourcing tasks completed, specifically:
[0094] According to the formula: ,
[0095] , construct the objective function;
[0096] In the formula, represents the expectation of maximizing the number of completed tasks; is the total number of tasks; is the total number of workers; Indicates workers Complete the task The probability of If If established, The value is 1; Indicates workers In time period Internal satisfaction task The probability of the position of Indicates workers In time period Internal satisfaction task Task category probability.
[0097] S3: Based on the objective function, the first stage allocation is performed, specifically:
[0098] S31: After initializing the parameters of the salp swarm optimization algorithm, several subpopulations are initialized.
[0099] against The crowdsourcing task allocation optimization problem of tasks is initialized. The subpopulation consists of 1-dimensional vectors, each subpopulation represents a task of the crowdsourcing task allocation problem, and each individual dimension is 1. Each subpopulation attempts to optimize a single component of the solution vector, which is essentially a one-dimensional optimization problem.
[0100] S32: Calculate the fitness value according to the objective function, sort the salps according to the fitness value, and record the position information of each salp.
[0101] Specifically, the salps in the subpopulations were combined into dimensional vector, each A dimensional vector as a The candidate solutions of the crowdsourcing task allocation problem are selected, the fitness value is calculated through the objective function, the individuals in each subpopulation are sorted according to the fitness value, and the position information of each subpopulation is recorded.
[0102] Since the fitness value of the objective function can only be calculated after all components are combined, this means that improvements in two dimensions (two steps forward) will negate the potential good value of a single dimension (one step backward). nThe advantage of using a set of one-dimensional vectors is that only one dimension is modified at a time, resulting in the desired fine-grained search, effectively preventing the "two steps forward, one step back" situation. Since many combinations are formed using different individuals from different subpopulations, the diversity of allocation schemes is also increased.
[0103] S33: After the sorted salps are updated by the salp group optimization algorithm, S32 is executed until the number of iterations reaches a first preset number of iterations to obtain the first stage allocation result.
[0104] Specifically, the salp swarm optimization algorithm is used to generate new sub-population individuals. The salp swarm optimization algorithm divides the salp individuals in the sub-population into two roles: leaders and followers. The leader individual update formula is expressed as:
[0105] ;
[0106] In the formula, Indicates A leader in a subpopulation; The location of the food source The first dimension, The optimal individual position in each subpopulation; , is a random number between [0,1]; It is The upper limit of the value of the dimension; It is The lower limit of the value of the dimension; The exploration and exploitation capabilities of the optimization algorithm can be balanced and expressed as:
[0107] ;
[0108] In the formula, is the current iteration number in the first phase allocation, is the first preset number of iterations.
[0109] The follower individual update formula is expressed as:
[0110] ;
[0111] In the formula, Indicates The first Followers, .
[0112] S4: Based on the results of the first-stage allocation, the extreme learning machine is trained to perform the second-stage allocation, specifically:
[0113] S41: Based on the allocation results of the first stage, the Spearman correlation coefficients between several sub-populations are calculated, and the sub-populations are merged according to the Spearman correlation coefficients.
[0114] Consider that if some dimensions are correlated, the correlated dimensions should be grouped in the same subpopulation, because independent changes made by different subpopulations will have an adverse effect on the correlated dimensions.
[0115] According to the historical data allocated in the first stage, the Spearman correlation coefficient between the dimensions is calculated, and the one-dimensional sub-populations with high correlation are merged into a multidimensional sub-population. The individuals in the merged sub-population are multidimensional individuals.
[0116] Preferably, based on the first-stage allocation results, the Spearman correlation coefficients between the subpopulations are calculated, specifically:
[0117] According to the formula: , calculate the Spearman correlation coefficient between several subpopulations;
[0118] In the formula, is the Spearman correlation coefficient; is the difference in fitness ranking of individual salps in the two subpopulations; is the number of salps individuals in the subpopulation; is the number of subpopulations.
[0119] The present invention uses the Spearman correlation coefficient to help understand the correlation between dimensions, merge sub-populations of related dimensions into the same group, and facilitate the joint evolution of related dimensions.
[0120] S42: Based on the allocation result of the first stage, train an extreme learning machine to obtain a trained extreme learning machine.
[0121] Based on the historical data allocated in the first stage, that is, the historical positions and optimal positions of individuals in the sub-populations, an extreme learning machine is trained for each sub-population. By using the extreme learning machine to mine the historical data of the evolutionary process, we can obtain knowledge that helps the salp swarm optimization algorithm generate effective evolution and predict the optimal individual positions of each sub-population, that is, the optimal allocation plan for the crowdsourcing task.
[0122] S43: Based on the merged sub-population, the fitness value is calculated according to the objective function, the salps individuals are sorted according to the fitness value, and the position information of each salp individual is recorded.
[0123] S44: According to the first probability, the sorted salps are updated by the salp swarm optimization algorithm or the trained extreme learning machine, and then S43 is executed until the number of iterations reaches a second preset number of iterations to obtain the second stage allocation result.
[0124] The first probability is a random number between 0 and 1, and the individual positions of each subpopulation are updated probabilistically using the salp swarm optimization algorithm or the trained extreme learning machine. For example, when the random number is less than 0.5, the trained extreme learning machine is used to update the individual positions in each merged subpopulation; when the random number is not less than 0.5, the salp swarm optimization algorithm is used to update the individual positions. n dimensional vector, i.e. n Dimensional candidate solutions, calculate the fitness value of the candidate solutions, and record the historical position, optimal position and fitness value of each individual in the subpopulation.
[0125] According to the no free lunch theory, no method can be effective in solving any problem. Probabilistically using the salp swarm optimization algorithm or the trained extreme learning machine to update the individual positions of each subpopulation can increase population diversity and improve the possibility of finding the global optimal position.
[0126] S5: Based on the second stage allocation result, according to the second probability, the trained extreme learning machine or local aggregation method is used to update and obtain the third allocation result, which is specifically:
[0127] S51: Based on the second stage allocation result, the trained extreme learning machine or local clustering method is updated according to the second probability.
[0128] Similarly, the second probability is obtained by a random number between [0,1] and is updated probabilistically using the trained extreme learning machine or local aggregation method. For example, when the random number is less than 0.5, the trained extreme learning machine is used to update the individual position; when the random number is not less than 0.5, the local aggregation method is used to update the individual position. n Dimensional candidate solutions, calculate the fitness value of the candidate solutions, and record the historical position, optimal position and fitness value of each individual in the subpopulation.
[0129] The local aggregation method has good development capabilities, and the trained extreme learning machine has the ability to predict the optimal position of an individual. Probabilistically using the trained extreme learning machine or local aggregation method to update the individual position helps to deeply develop near the global optimal position, thereby finding the global optimal solution and obtaining the optimal task allocation plan.
[0130] Preferably, in step S51, the updating is performed by a local aggregation method, specifically:
[0131] According to the formula: ,
[0132] , to update;
[0133] In the formula, After the update individuals in a subpopulation; For the The best individual in a subpopulation; For the Suboptimal individuals in a subpopulation; To adjust the parameters of the aggregation scope, according to the formula: ,get;
[0134] In the formula, is the current iteration number, The third preset number of iterations.
[0135] S52: Calculate the fitness value according to the objective function, sort the salps according to the fitness value, record the position information of each salp, and execute S51 until the number of iterations reaches the third preset number of iterations to obtain the maximum fitness value. The position of the salp group individual corresponding to the maximum fitness value is used as the third stage allocation result.
[0136] The present invention constructs an objective function based on the attributes of workers and tasks, and sequentially performs three-stage task allocation to obtain a crowdsourcing task allocation scheme that maximizes the objective function, wherein in the first stage allocation, the multidimensional crowdsourcing task allocation optimization problem is converted into a one-dimensional optimization problem, and the first stage allocation result is obtained by a salp swarm optimization algorithm; in the second stage allocation, dimensions with high correlation are merged into a population, which is updated by a salp swarm optimization algorithm or a trained extreme learning machine to obtain a second stage allocation result; in the third stage allocation, the third stage allocation result, i.e., the optimal task allocation scheme, is obtained by updating by a trained extreme learning machine or a local aggregation method to ensure that tasks can be accurately assigned to the most suitable workers and maximize the expectation of completing the number of tasks.
[0137] The present invention also provides a task allocation system for an extreme learning machine-assisted salp swarm optimization algorithm, comprising:
[0138] Get module: used to get the properties of workers and tasks;
[0139] Worker attributes, including: worker location, worker activity area, worker movement speed, worker preferences, and worker task execution history;
[0140] Task attributes, including: task location, task start time, task end time, and task category;
[0141] Objective function building module: Based on the attributes of workers and tasks, the objective function is built, specifically:
[0142] Based on the attributes of workers and tasks, we build location models, time models, and preference models as constraints;
[0143] Construct the objective function according to the constraints;
[0144] The first stage allocation module: Based on the objective function, the first stage allocation is carried out, specifically:
[0145] After initializing the parameters of the salp swarm optimization algorithm, several subpopulations are initialized;
[0146] The fitness value is calculated according to the objective function, the salps are sorted according to the fitness value, and the position information of each salp is recorded;
[0147] After the sorted salps are updated by the salp group optimization algorithm, the fitness values are calculated and sorted until the number of iterations reaches a first preset number of iterations, and the first stage allocation result is obtained;
[0148] Second-stage allocation module: Based on the first-stage allocation results, the extreme learning machine is trained to perform the second-stage allocation, specifically:
[0149] Based on the allocation results of the first stage, the Spearman correlation coefficients between several sub-populations are calculated, and the sub-populations are merged according to the Spearman correlation coefficients;
[0150] Based on the allocation result of the first stage, an extreme learning machine is trained to obtain a trained extreme learning machine;
[0151] Based on the merged sub-population, the fitness value is calculated according to the objective function, the salps are sorted according to the fitness value, and the position information of each salp is recorded;
[0152] According to the first probability, the sorted salps are updated by the salps swarm optimization algorithm or the trained extreme learning machine, and the fitness values are calculated and sorted until the number of iterations reaches a second preset number of iterations, thereby obtaining the second stage allocation result;
[0153] The third stage allocation module: based on the second stage allocation result, according to the second probability, the trained extreme learning machine or local aggregation method is used to update the third allocation result, which is specifically:
[0154] Based on the second stage allocation result, the trained extreme learning machine or local aggregation method is updated according to the second probability;
[0155] The fitness value is calculated according to the objective function, and the salps are sorted according to the fitness value. The position information of each salp is recorded and updated according to the second probability by the trained extreme learning machine or local aggregation method until the number of iterations reaches the third preset number of iterations to obtain the maximum fitness value. The position of the salp group individual corresponding to the maximum fitness value is used as the allocation result of the third stage.
Claims
1. A task allocation method for an extreme learning machine-assisted salp swarm optimization algorithm, characterized in that: The following steps are involved: S1: Get the properties of workers and tasks; Worker attributes, including: worker location, worker activity area, worker movement speed, worker preferences, and worker task execution history; Task attributes, including: task location, task start time, task end time, and task category; S2: Based on the attributes of workers and tasks, construct the objective function, specifically: S21: Based on the attributes of workers and tasks, build location models, time models, and preference models as constraints; S22: Construct the objective function according to the constraints, specifically: According to the formula: , , construct the objective function; In the formula, represents the expectation of maximizing the number of completed tasks; is the total number of tasks; is the total number of workers; Indicates workers Complete the task probability; If If established, The value is 1; is the location model, representing the worker In time period Internal satisfaction task The probability of the position of is the time model, representing the time period in which the task is located; is the preference model, representing the worker In time period Internal satisfaction task Task category probability; S3: Based on the objective function, the first stage allocation is performed, specifically: S31: After initializing the parameters of the salp swarm optimization algorithm, several subpopulations are initialized; S32: Calculate the fitness value according to the objective function, sort the salps according to the fitness value, and record the position information of each salp; S33: After the sorted salps are updated by the salp group optimization algorithm, S32 is executed until the number of iterations reaches a first preset number of iterations, and the first stage allocation result is obtained; S4: Based on the results of the first-stage allocation, the extreme learning machine is trained to perform the second-stage allocation, specifically: S41: Based on the allocation results of the first stage, the Spearman correlation coefficients between several sub-populations are calculated, and the sub-populations are merged according to the Spearman correlation coefficients; S42: Based on the allocation result of the first stage, train an extreme learning machine to obtain a trained extreme learning machine; S43: Based on the merged sub-population, the fitness value is calculated according to the objective function, the salps individuals are sorted according to the fitness value, and the position information of each salp individual is recorded; S44: after the sorted salps are updated by the salp swarm optimization algorithm or the trained extreme learning machine according to the first probability, S43 is executed until the number of iterations reaches a second preset number of iterations, and a second-stage allocation result is obtained; S5: Based on the second stage allocation result, according to the second probability, the trained extreme learning machine or local aggregation method is used to update and obtain the third allocation result, which is specifically: S51: Based on the second stage allocation result, the trained extreme learning machine or local aggregation method is updated according to the second probability; Updated by local aggregation method, specifically: According to the formula: , , to update; In the formula, After the update individuals in a subpopulation; For the The best individual in a subpopulation; For the Suboptimal individuals in a subpopulation; To adjust the parameters of the aggregation scope, according to the formula: ,get; In the formula, is the current iteration number, The third preset number of iterations; S52: Calculate the fitness value according to the objective function, sort the salps according to the fitness value, record the position information of each salp, and execute S51 until the number of iterations reaches the third preset number of iterations to obtain the maximum fitness value. The position of the salp group individual corresponding to the maximum fitness value is used as the third stage allocation result.
2. The task allocation method of the extreme learning machine collaborative salp swarm optimization algorithm according to claim 1, characterized in that: In step S21, a location model is constructed based on the attributes of the worker and the task, specifically: According to the formula: , build a location model; In the formula, Indicates workers In time period Internal satisfaction task The probability of the position of Is a worker and tasks The distance between It is the largest activity area for workers.
3. The task allocation method of the extreme learning machine collaborative salp swarm optimization algorithm according to claim 1, characterized in that: In step S21, a time model is constructed based on the attributes of the worker and the task, specifically: According to the formula: , construct a time model; In the formula, Indicates the time period in which the task is located; For workers and tasks The distance between For workers Movement speed; For the task The start time of For the task The end time of .
4. The task allocation method of the extreme learning machine collaborative salp swarm optimization algorithm according to claim 1, characterized in that: In step S21, a preference model is constructed based on the attributes of workers and tasks, specifically: According to the formula: , , build a preference model; In the formula, Indicates workers In time period Internal satisfaction task Task category Probability of preference; For workers In time period The total number of tasks executed within For workers The task execution history; Indicates workers Execution task category frequency; is the category of the task to be performed; For the task The time of start; For the task End time.
5. The task allocation method of the extreme learning machine collaborative salp swarm optimization algorithm according to claim 1, characterized in that: In step S41, based on the allocation result of the first stage, the Spearman correlation coefficient between several sub-populations is calculated, specifically: According to the formula: , calculate the Spearman correlation coefficient between several subpopulations; In the formula, is the Spearman correlation coefficient; is the difference in fitness ranking of individual salps in the two subpopulations; is the number of salps individuals in the subpopulation; is the number of subpopulations.
6. The task allocation method of the extreme learning machine collaborative salp swarm optimization algorithm according to claim 1, characterized in that: The first probability in step S44 or the second probability in step S5 is a random number between [0, 1].
7. A task allocation system based on extreme learning machine and salp swarm optimization algorithm, characterized in that: include: Get module: used to get the properties of workers and tasks; Worker attributes, including: worker location, worker activity area, worker movement speed, worker preferences, and worker task execution history; Task attributes, including: task location, task start time, task end time, and task category; Objective function building module: Based on the attributes of workers and tasks, the objective function is built, specifically: Based on the attributes of workers and tasks, we build location models, time models, and preference models as constraints; According to the constraints, the objective function is constructed, which is as follows: According to the formula: , , construct the objective function; In the formula, represents the expectation of maximizing the number of completed tasks; is the total number of tasks; is the total number of workers; Indicates workers Complete the task probability; If If established, The value is 1; is the location model, representing the worker In time period Internal satisfaction task The probability of the position of is the time model, representing the time period in which the task is located; is the preference model, representing the worker In time period Internal satisfaction task Task category probability; The first stage allocation module: Based on the objective function, the first stage allocation is carried out, specifically: After initializing the parameters of the salp swarm optimization algorithm, several subpopulations are initialized; The fitness value is calculated according to the objective function, the salps are sorted according to the fitness value, and the position information of each salp is recorded; After the sorted salps are updated by the salp group optimization algorithm, the fitness values are calculated and sorted until the number of iterations reaches a first preset number of iterations, and the first stage allocation result is obtained; Second-stage allocation module: Based on the first-stage allocation results, the extreme learning machine is trained to perform the second-stage allocation, specifically: Based on the allocation results of the first stage, the Spearman correlation coefficients between several sub-populations are calculated, and the sub-populations are merged according to the Spearman correlation coefficients; Based on the allocation result of the first stage, an extreme learning machine is trained to obtain a trained extreme learning machine; Based on the merged sub-population, the fitness value is calculated according to the objective function, the salps are sorted according to the fitness value, and the position information of each salp is recorded; According to the first probability, the sorted salps are updated by the salps swarm optimization algorithm or the trained extreme learning machine, and the fitness values are calculated and sorted until the number of iterations reaches a second preset number of iterations, thereby obtaining the second stage allocation result; The third stage allocation module: based on the second stage allocation result, according to the second probability, the trained extreme learning machine or local aggregation method is used to update the third allocation result, which is specifically: Based on the second stage allocation result, the trained extreme learning machine or local aggregation method is updated according to the second probability; Updated by local aggregation method, specifically: According to the formula: , , to update; In the formula, After the update individuals in a subpopulation; For the The best individual in a subpopulation; For the Suboptimal individuals in a subpopulation; To adjust the parameters of the aggregation scope, according to the formula: ,get; In the formula, is the current iteration number, The third preset number of iterations; The fitness value is calculated according to the objective function, and the salps are sorted according to the fitness value. The position information of each salp is recorded and updated according to the second probability by the trained extreme learning machine or local aggregation method until the number of iterations reaches the third preset number of iterations to obtain the maximum fitness value. The position of the salp group individual corresponding to the maximum fitness value is used as the allocation result of the third stage.
Citation Information
Patent Citations
Cloud task scheduling method and device equipment based on Salpa group, equipment and storage medium
CN111274020A
Cloud task scheduling method, device and equipment based on sea squirt group and storage medium
CN111367661A
