CN115658259A - Component scheduling method based on load balancing strategy and improved ant colony algorithm in multiple industrial networks - Google Patents

Abstract

In today's complex and varied industrial network environment, the efficiency of industrial task completion depends on collaboration and interaction between multiple networks. The invention provides a component allocation method based on load balancing and ant colony algorithm in multiple industrial networks, which realizes the search and scheduling of different subnet components by improving the ant colony algorithm and solves the problem of overload in the multiple networks by a load balancing strategy. The method simulates the scheduling process of the intra-network and inter-network components by crawling of ants according to the business logic relation of specific tasks. Through multiple iterations, an optimal path for related component scheduling is planned, and self-assembly of intra-network and inter-network components is achieved. On the premise of not influencing the execution of a plurality of target tasks, the method can effectively reduce the scheduling time and cost, improve the utilization rate of the components and improve the parallel execution efficiency of a plurality of industrial software systems under the micro-service framework.

Description

Component scheduling method based on load balancing strategy and improved ant colony algorithm in multiple industrial networks

Technical Field

The invention relates to searching and scheduling related components in each sub-network according to the business logic and flow of specific tasks in a multi-industrial network environment. And planning an optimal path for scheduling the components based on a load balancing strategy and an improved ant colony algorithm, and realizing scheduling and self-assembly of the components in the network and among the networks according to the path. The method gives consideration to efficiency and cost, can effectively reduce scheduling time and cost on the premise of not influencing a plurality of target tasks, and improves the parallel execution efficiency of a plurality of industrial software systems under a micro-service framework.

Background

Existing industrial software systems are typically based on a single network architecture. Although data sharing can be performed to some extent between different industrial networks, complete integrated cooperation between industrial software systems in different network environments is not achieved. With the development of industrial technology and enterprise information intelligence, the current industrial software system is simultaneously constrained and influenced by multiple networks of information flow, control flow, business flow and the like. In addition, a complex industrial task puts an urgent need on cooperation among different industrial software systems in multiple industrial networks, and the traditional industrial software systems with a single network architecture have the problems of poor cooperative optimization effect, excessive local load and the like due to lack of effective integration and association cooperation. Therefore, how to solve the problem of scheduling and assembling components based on complex industrial tasks through the cooperation and interaction of components among different software systems in multiple industrial networks becomes a key point and a difficulty for improving the parallel execution efficiency of industrial software systems.

In recent years, with the rapid development of computer technology, micro services are taken as a service-oriented idea, and the core significance of the micro services lies in that a large-scale single application system is subjected to module splitting, fine-grained services are used, a series of software service units which can be independently designed, asynchronously developed, dispersedly deployed and respectively operated and maintained are used for replacing the original large-scale single application, the services respectively use independent databases, strong relevant relations are not established, the services are mutually cooperated, and a highly coupled relation is not established. The integration mode of the multiple networks constructed under the micro-service model is independent, the services are divided according to the granularity, and the service deployment is scattered. The task scheduling process spans a plurality of different component networks, each component network provides different component services, the logic is clear and flexible to expand, and all aspects are very suitable for the industrial software system which is highly developed nowadays.

When a local load in an industrial system is too heavy, in consideration of task completion of the whole system, a component which is too heavy in load in the network needs to be subjected to task deployment, and tasks which need to be completed but are not completed because of the too heavy load are distributed to other components which are lighter in load or idle. However, due to the difference of the scheduling algorithms used by different systems, the scheduling order and scheme are different, and the execution efficiency of different components is different due to the use frequency of the components. Therefore, in the conventional industrial system, the local overload is easy to occur, and if the local overload is not solved in time, the task scheduling is continuously executed, so that the component load is further increased, and the execution of a plurality of target tasks is influenced.

In general, when the load is not uniform in an industrial system, tasks of heavy load are allocated mainly by a manual method. However, the occurrence of the overload situation cannot be observed in time at each time and can be processed in time, and the overload situation is often found after the overload situation affects the tasks. The method is mainly characterized in that under a micro-service framework, an assembly scheduling method based on load balancing and an improved ant colony algorithm is used, the assembly process of intra-network and inter-network assemblies is simulated by the crawling process of ants, and specific parameters are applied to related assemblies. The heuristic function of the ants depends on the parameters, the crawling of the ants is equivalent to the scheduling of the components, when all the ants complete crawling, the obtained final path is the component assembly path meeting the task requirement, and the components are self-assembled according to the assembly sequence. According to the method, a load balancing strategy is introduced in the path searching process, so that even if the load is uneven, the situation tends to be balanced under the iteration of scheduling for a plurality of times due to the existence of the strategy.

Disclosure of Invention

The technical problem is as follows: the invention aims to provide a component scheduling method based on load balancing and ant colony algorithm in a multiple industrial network, which realizes task allocation of intra-network and inter-network components and solves the problems in the background technology. The method mainly uses a load balancing strategy to improve the traditional ant colony algorithm, and uses the crawling of ants to simulate the scheduling of intra-network and inter-network components so as to meet the requirements of specific working conditions and working tasks. And the final path crawled after the ants iterate for a plurality of times is the optimal path for scheduling the components. The method can relieve the problem of overload of component scheduling in the industrial network to a certain extent, and can also find the optimal scheduling path meeting the task in a short time. The utilization rate of each component in the industrial network environment is improved to a certain extent, and the parallel execution efficiency of a plurality of industrial software systems is improved to a certain extent.

The technical scheme is as follows: in a complete multi-industrial network environment, each component is responsible for a considerable number of tasks, and the completion of each industrial task is not independent of the ordered scheduling of the components. However, the performance of the assembly is not constant. With the change of the scheduling environment such as the use time, the call intensity, etc., the load condition of the component is aggravated, and the performance of the component is also affected to a certain extent. At this time, the tasks processed by the components need to be allocated, and part of the tasks to be processed is allocated to the components with lighter loads or idle.

In order to ensure that the dispatching can be completed orderly and efficiently among the assemblies, the invention is mainly realized by the following steps: firstly, initializing the number m of ants, pheromone heuristic factor alpha, heuristic function factor beta, path length L, pheromone volatilization factor rho, pheromone constant Q, maximum iteration number T and pheromone concentration tau _ij Constructing a solution space G =for component scheduling<V,E>The components are analogized to ants, m ants are placed on nodes in the network, and the roulette method is adopted for each step of path selection of the ants. According to the cost N of the currently scheduled component, the assembly of different component designs is time T, a load balancing coefficient X is defined, and the average load value X _ AVE of i nodes is recorded _ij Calculating the load variance X _ s of the i nodes, wherein when the load variance X _ s is larger, the node is the group at the momentThe larger the task amount of the component is, the longer the component is scheduled, and therefore, the larger the load of the network is represented. When the load variance X _ s is smaller, the task quantity of the component is smaller, the component scheduling time is shorter, the load of the network is smaller, and the probability that ants crawl to the next node is calculated according to the data

Wherein i and j are respectively the starting point and the next node of a certain ant crawling in the execution process.

Represents the probability of an ant crawling from i node to j node, eta _ij Represents the inverse of the path distance between two points i, j. allowed _k Representing nodes that have not been visited. P under constraint _ij May change as the load balancing value changes. When the load value X _ s of a certain node is smaller (namely the load condition of the node is smaller), the probability P that the ant selects the node is smaller _ij The larger will be. When the load value X _ s of a certain node is larger (namely the load condition of the node is heavier), the probability P that the ant selects the node is higher _ij It is reduced.

The crawling of ants selects the next node according to the result of roulette algorithm, and the pheromone concentration tau is determined after the ants move every time _ij The update is performed, and after a period of crawling, the system returns to the original component, so that a scheduling path is completed, and local update of the pheromone is completed. After all ants complete one period of crawling, global pheromone updating is carried out on the network. Due to the influence of pheromones, all ants converge on one path in the process of several iterations, and the obtained assembly path is the optimal path for component deployment and covers the searching range of intra-network and cross-network. The path isThe method is obtained under the condition of ensuring that each node in the whole scheduling system is in load balance. The stress on the part of the relevant components is relieved to some extent.

Has the advantages that:

1) The invention fully utilizes the characteristic of an ant colony algorithm, simulates the scheduling process of the component to the foraging process of ants, and completes the search of the path through heuristic information. The method not only widens the search range of the scheduling scheme, covers the search space in the network and across the network, but also can accurately find out the optimal scheduling scheme, optimizes the structure of the scheduling mode to a certain extent, improves the scheduling efficiency and reduces the time required by component scheduling.

2) The invention improves the safety and stability of the assembly system, is a scheduling scheme based on load balancing, so that the condition of overweight load of individual nodes does not exist, the overall stability and the local optimal condition of scheduling are obviously improved, the utilization rate of the assembly in a plurality of task cycles is improved, and the self-assembly of the assembly in a service combination mode is conveniently met.

3) The invention balances the load condition of each component in the network, acts the component scheduling process of the whole industrial network by introducing a load balancing strategy, avoids the negative influence caused by over-heavy load of part of components, ensures that the scheduling task can be carried out in a better environment, and improves the parallel execution efficiency of a plurality of industrial software systems to a certain extent.

Drawings

FIG. 1 is a diagram illustrating the scheduling of industrial task components in a conventional single network.

FIG. 2 is a diagram illustrating multi-network task-driven component scheduling in a microservice model.

FIG. 3 is a schematic diagram of a method for scheduling multiple network components based on load balancing and ant colony optimization

Detailed Description

In order to perform component scheduling normally in a multi-industrial network, it is necessary to detect the performance of each component in the network, the load condition of the component, and the like.

In a multi-industrial network composed of a micro-service architecture, a multi-industrial network is composed of a plurality of industrial networks, and each industrial network is composed of components having the same characteristics. When there is a series of work tasks to schedule related components, its efficiency of completion depends on the collaboration and interaction between multiple industrial networks. The method improves the traditional ant colony algorithm by using a load balancing strategy, and simulates the scheduling process of intra-network and inter-network components by using the crawling of ants. And each component completes the allocation and assembly in the network and across the network through a series of scheduling instructions according to specific task requirements, thereby completing corresponding tasks. When the scheduled component load condition far exceeds the rest average load value, the method reduces the probability that the component is scheduled each iteration. After a number of iterations, the load condition of the component will be close to the average load value. When the load condition of the scheduled component is better, the method searches out an optimal path meeting the scheduling requirement for the task, and can complete the self-assembly of the intra-network and inter-network components according to the path.

Initialization of the parameters: defining a K x K order matrix A by setting n different assemblies _o The number of components required for the product is represented, and a K x K order matrix A is defined _n To indicate the number of nodes in the component library at a certain time, if the matrix O = A _n -A _o If there is a negative number in the characteristic value of (2), the system gives feedback indicating that the task cannot be completed. If the characteristic value of the matrix O does not have a negative number, namely the assembly requirements of various components are met, the system provides a signal capable of searching an assembly path. Initialization of pheromone concentration tau _ij ＝τ ₀ i, j ∈ K. The number m of initialized ants, pheromone heuristic factor alpha, heuristic function factor beta, path length L, pheromone volatilization factor rho, pheromone constant Q, maximum iteration number T and pheromone concentration tau _ij . Pheromone elicitation factors, elicitation function factors, pheromone volatilization factors, pheromone constants, maximum iteration times and pheromone concentration are used as important characteristic factors in the ant colony algorithm to determine the execution process of the ant colony algorithm, and the result is set to be too high or too smallWith serious consequences. Therefore, the parameters should be initialized and the relevant parameter setting should be performed according to the actual situation.

Construct the component scheduling space: the total number m of ants is set. K industrial networks are defined, and each component in each industrial network is selected as a node of the network. And randomly placing m ants on each node in the K industrial networks, and selecting the next node by the ants according to a roulette algorithm each time of crawling. When ants reach the i node at the time t, the cost N of the component is calculated first _t Time T required for the component to be scheduled _t Load factor X of the assembly:

X＝η ₁ ×N _t +η ₂ ×T _t ，η ₁ 、η ₂ a weight representing the amount of component availability and assembly time.

The load mean value and the load variance are obtained through the values

Average load value of i nodes

Because the variance can well reflect the fluctuation degree of a system, the load condition of the whole network can be more intuitively reflected.

So the load variance of i nodes

Therefore, the load condition of the current node is evaluated, and the network scheduling structure is adjusted according to the load condition. The load factor is usually used to analyze the load condition of the network, and in the present invention, because the load factor is composed of the component cost and the time occupied by the component assembly, which are the target factors of the component scheduling, the load factor becomes the heuristic function in the ant colony algorithm at the same time, and is one of the determining factors for the next node selection.

The ant colony algorithm is a unique heuristic algorithm, and the heuristic function determines the execution direction of the algorithm. Therefore, the heuristic function of the algorithm is defined as follows:

wherein X _i And the load value of the ith node is represented, and X _ AVE represents the average load value of the i nodes, wherein the average load value refers to the average of the load balance values of the i nodes. After the heuristic function is constructed, the probability of reaching each subsequent node can be calculated according to the function. Calculating the selection probability of the subsequent nodes according to the pheromone concentration tau, the pheromone heuristic factor alpha, the heuristic function X _ s and the heuristic function factor beta:

it can be observed from the above equation that the less the cost of the components and the time spent in assembly, the lower the complex equalization coefficients will be, and the easier it will be to become the nodes in the final scheduling sequence. When the load factor of a component is too high, the probability that it is selected as the next node is lower, and after several iterations, the load condition is gradually reduced. On the contrary, when the load factor of a component is too low, the probability that the component will be selected as the next node gradually increases, and the utilization rate of the component is greatly improved.

And each time iteration is completed, all ants finish path searching equivalently, a shortest path meeting the scheduling requirement is selected from the paths, and the pheromone is locally updated. The pheromone update rule is as follows:

τ _ij (t+1)＝τ _ij (t)*(1-ρ)+△τ _ij ，0<ρ<1，△τ _ij ＝Q/L

wherein rho represents pheromone volatilization coefficient, Q represents pheromone constant (sum of information amount released by single ant), L represents path length of ant walking, and delta tau _ij Indicating the change value of the pheromone.

The ants can release the pheromone on the paths traveled by the ants, but the pheromone concentration can volatilize in rho proportion along with the change of time, so that the pheromone concentration on the scheduling path which is accorded with the task is higher along with the increase of the iteration times, and the pheromone concentration on the rest of more complex paths can be gradually reduced according to the volatilization of the pheromone. And the crawling of the ants always crawls according to the path with the highest pheromone concentration, so the final crawling route of the whole ant colony is the scheduling path required by the task.

When all ants complete the path search, the optimal path in this iteration can be obtained, and the global pheromone is updated at this time:

wherein D is a constant, L ^best Representing the optimal path length in this iteration.

And after the global pheromone is updated, checking whether the iteration times reach a set maximum value, and if not, outputting the optimal path. Otherwise, the problem exists in the iterative process, and the path search needs to be carried out again.

Claims (9)

1. The patent refers to the field of 'transmission of digital information'. The method is mainly used for realizing path search of scheduling of different subnet components by improving an ant colony algorithm under a complex and changeable industrial network environment, and solving the problem of overload existing in the multiple industrial network through a load balancing strategy. Therefore, an optimal path meeting the task requirement is obtained, and self-assembly of intra-network and inter-network components is completed according to the path. The method can solve the problems of low component scheduling efficiency, heavy load and the like in the complex industrial environment.

2. The method of claim 1, wherein the multiple industrial networks comprise K industrial networks, and each industrial network selects a component as a node of the network. Using K x K order matrix A _o To indicate the number of components required for the finished product,continuing to define a K x K order matrix A _n To indicate the number of nodes in the component library at a certain time, if the matrix O = A _n -A _o If there is a negative number in the characteristic value of (2), the system gives feedback indicating that the task cannot be completed. And if the characteristic values of the matrix O do not have negative numbers, the assembly requirements of various components are met.

3. The method as claimed in claim 2, wherein the ant number m, pheromone elicitation factor α, elicitation function factor β, path length L, pheromone volatilization factor ρ, pheromone constant Q, maximum iteration number T, and pheromone concentration τ are initialized _ij . The value of m should be set to 1.5 times the number of components. When the pheromone concentration is too large, pheromones on each path tend to be average, the positive feedback effect is weakened, so that the convergence speed is reduced, and when the pheromone concentration is too small, the pheromone concentration of some paths which are not searched is reduced to 0, so that premature convergence is caused, and the global optimality of the solution is reduced. The constant of pheromone is selected according to the practical situation of industrial environment [10, 1000%]If the size is too large, the search range of the ant colony is reduced, premature convergence is easy, and the colony falls into local optimum. When the pheromone constant Q is obtained, the difference of the content of the pheromone on each path is small, and the pheromone is easy to fall into a chaotic state. Maximum number of iterations Tselect [200, 500]The method can avoid the operation from being actually too long, and can ensure enough optional paths. The pheromone factor alpha reflects the relative importance degree of the quantity of pheromones accumulated on the path in the ant movement process in guidance ant colony search, and the value range is usually [1,4]This range may avoid reduced randomness or premature convergence of the ant colony algorithm. The heuristic function factor beta reflects the relative importance degree of heuristic information in the process of guiding ant searching, and the strength of the action of the prior and deterministic factors in the ant colony optimization process. The value range is [0,5]This range can avoid convergence too fast or falling into a purely random search. The pheromone volatilization factor reflects the level of disappearance of the pheromone, whereas 1-p, in contrast, reflects the level of maintenance of the pheromone. The value range is usually [0.2,0.5]In the meantime. When the concentration is too high, the pheromone volatilizes faster, easily resulting in the elimination of a better pathway. When the concentration is too low, the pheromone on the scheduling path containsThe quantity difference is small, and the algorithm convergence speed is low.

4. The method as claimed in claim 3, wherein the crawling of ants is used to simulate the intra-network and cross-network scheduling process of the components: m ants are randomly placed on each node in the K industrial networks, and each time the ants crawl, the ants select the next node according to a roulette algorithm. Heuristic functions are used to represent the visibility of ants from one node to another. When the distance between two locations is short, the probability that a path having a higher pheromone concentration should be selected is higher. From the probability calculation formula, it can be seen that the pheromone factor α is an index of pheromone concentration, and the heuristic function factor β is an index of the heuristic function. Determines the pheromone concentration and the degree to which the transfer is expected to contribute to the likelihood of ant k going from i to j.

5. The method as claimed in claim 4, wherein the cost N of the device is calculated when ants reach the i-node at time t _t Time T required for the component to be scheduled _t The load factor X of the assembly.

X＝η ₁ ×N _t +η ₂ ×T _t ，η ₁ 、η ₂ A weight representing the amount of component availability and assembly time.

The load mean value and the load variance are obtained through the values

Average load value of i nodes

The variance can well reflect the fluctuation degree of a system, and the load condition of the whole network can be more intuitively reflected.

So that the load variance of the i nodes

6. The method of claim 5, wherein the method comprises: the load factor can be used to evaluate the load condition of the current node, and the network scheduling structure is adjusted according to the load condition. The load factor is usually used to analyze the load condition of the network, and in the present invention, because the load factor is composed of the component cost and the time occupied by the component assembly, which are the target factors of the component scheduling, the load factor becomes the heuristic function in the ant colony algorithm at the same time, and is one of the determining factors for the next node selection.

7. The method of claim 5, wherein the method comprises: the ant colony algorithm is a unique heuristic algorithm, and the heuristic function determines the execution direction of the algorithm. Therefore, the heuristic function of the algorithm is defined as follows:

wherein X _i And the load value of the ith node is represented, and X _ AVE represents the average load value of the i nodes, wherein the average load value refers to the average of the load balance values of the i nodes. After constructing the heuristic function, the probability of reaching each subsequent node can be calculated from the function. Calculating the selection probability of the subsequent nodes according to the pheromone concentration tau, the pheromone heuristic factor alpha, the heuristic function X _ s and the heuristic function factor beta:

8. the method of claim 7, further comprising: it can be observed from the next node probability selection formula that the less the cost of the components and the time spent in assembly, the lower the complex equalization coefficients will be, and the easier it will be to become the nodes in the final scheduling sequence. When the load factor of a component is too high, the probability that it is selected as the next node is lower, and after several iterations, the load condition gradually decreases. Conversely, when the load factor of a component is too low, the probability that the component will be selected as the next node gradually increases, and the utilization rate of the component is greatly improved.

9. The method of claim 8, wherein the method comprises: every time an iteration is completed, all ants complete a path search equivalently, and a shortest path meeting the scheduling requirement is selected from the paths. One iteration means that all nodes are completed by m fingers, namely m search paths exist. And comparing all paths, selecting the path with the shortest length, making the visual result of the iteration, and updating the pheromone. And comparing the length of the current shortest path with the length of the past shortest path, adding 1 to the iteration times, and then judging whether the current iteration times are equal to the set iteration times. If equal, stopping iteration, otherwise, performing next iteration. The local update of the pheromone is carried out, and the pheromone update rule is as follows:

τ _ij (t+1)＝τ _ij (t)*(1-ρ)+△τ _ij ，0<ρ<1，△τ _ij ＝Q/L

where ρ represents pheromone volatilization coefficient, Q represents pheromone constant (sum of information amount released by single ant), L represents path length of ant walking, and Δ τ _ij Indicating the change value of the pheromone.

The ants can release pheromones on the paths traveled by the ants, but the pheromone concentration can volatilize in rho proportion along with the change of time, so that the pheromone concentration on the scheduling path met by the task is higher along with the increase of iteration times, and finally the crawling route of the whole ant colony is the scheduling path required by the task.

After all ants complete the path search, the optimal path in the iteration can be obtained, and the global pheromone is updated at the moment:

wherein D is a constant, L ^best Representing the optimal path length in this iteration.

Images