WO2023128956A1 - Job sorting device and algorithm based on fpga-based hybrid heartistic search algorithms - Google Patents

Abstract

This invention; It is related to the job sequencing device and algorithm based on FPGA- based hybrid heuristic search algorithms, which provides rapid and easy access to near- optimal solutions by handling heuristic optimization algorithms in a hybrid system. FPGA motherboard (1) that runs the hybrid optimization model by including search algorithms such as Genetic Algorithm (GA), Simulation Annealing (SA), Particle Swarm Optimization (PSO) and Random Search (RA), handheld terminal (2) containing FPGA motherboard (1), FPGA motherboard (1) receives the jobs to be sorted (4) from the database (3) from the server computer (5) and after the FPGA motherboard (1) performs the sorting process, the sorted jobs (6) are sent back to the server computer (5) to the machines (7), includes the transmitted ethernet connection (8).

Description

JOB SORTING DEVICE AND ALGORITHM BASED ON FPGA-BASED

HYBRID HEARTISTIC SEARCH ALGORITHMS

Technical Field:

This invention; It is about the job sorting device and algorithm based on FPGA-based hybrid heuristic search algorithms, which provides fast and easy access to near-optimal solutions by considering heuristic optimization algorithms in a hybrid system.

State of the Art:

As it is known today, search algorithms; It is a general name of algorithms used to search for various target data such as information, words, strings. Each search algorithm has a different solution finding mechanism. It’s algorithms that can find the best solution in a shorter time, even if there are millions of alternative solutions in the search space. The primary purpose of these algorithms is to provide a method that makes it possible to reach the target data to be achieved in the simplest way, in the shortest time and in the most effective way, and to solve the problem. Another purpose is to offer the most appropriate solution or the closest solution to the problems. Searches using classical search techniques cannot be expected to be as effective as search algorithms. Because it is not possible to find the data sought in a solution space with an incalculable number of solution alternatives by examining one by one. As the size of the data structures to be searched increases, the time spent for the process will increase accordingly. However, due to heuristic search algorithms, the best or near-best solutions can be reached easily and quickly in a solution space with an incalculable number of solution alternatives.

In current systems, the jobs are ordered in a certain order according to certain qualifications, and it is determined how the multiple jobs to be done will be done. The ranking of the works is done by considering certain criteria such as the degree of importance and urgency. Jobs that need to be done in any workplace need a job sequence like this. Because ordering the work to be done in a certain order by giving priority to certain criteria brings along a systematic progress. Otherwise, because the work that needs to be done is not presented in a certain order, it may not be completed on time, be delayed, overlooked, or forgotten.

Today, it is not even possible to make job sequences manually in workplaces with many business lines. Because this will cause a large amount of time wasted. At the same time, this situation requires an extra workforce and requires only one personnel to devote their time and energy to this. In the current techniques, where job sequences are automatically performed with the help of computers, the duration of the sequence increases depending on the workload. In systems operated using conventional computers in current techniques, different operations cannot be performed simultaneously, and fast results cannot be obtained. In current techniques, job rankings are performed using various search algorithms. Sequencing jobs through a handheld terminal with several different search algorithms in a hybrid way contributes to the simpler execution of transactions in shorter times.

Application TR201517551 discusses “A System for Determining Job Ranking”. The present invention; It relates to a system for sorting the jobs in a data warehouse according to various criteria. Here, job sequencing is done by considering the total working time, dependency and start times of the jobs, parallel working requirements.

"Multiple Job Scheduling Method of GPU Cluster Based on Genetic Algorithm" is discussed in application CN112905316. The invention discloses a GPU cluster multi-job scheduling method based on a genetic algorithm. The time complexity of searching an optimal solution is reduced through a genetic algorithm, GPU resources are allocated for multiple jobs based on a resource allocation mode of minimizing multi-job completion time, idle resources possibly appearing in the job scheduling process are reused, and finally a multi -job scheduling scheme which is minimum in completion time and high in resource utilization rate is obtained. Compared with the prior art, the method has the advantages that the problems of long multi-job scheduling completion time, low resource utilization rate and the like on the GPU cluster are effectively solved, workers related to deep learning can quickly complete quality verification of different parameter architectures of the model under the guidance of the scheduling scheme, feedback results are obtained, the model is improved, and the method is suitable for large-scale popularization and application. Therefore, the device can be quickly put into the next experiment or production.

The above applications deal with systems based on genetic algorithms, and heuristic search algorithms that are combined to solve the same problem cannot be found in a hybrid structure. The absence of this hybrid algorithm structure in the aforementioned existing techniques, due to this hybrid structure, it is not possible for the algorithms to create a synergy and solve the problems in a much shorter time than the target. Considering the results of a number of sample problems in order to compare the genetic algorithm in the existing techniques and the hybrid solution system, it has been determined that the results obtained with the hybrid solution system are more successful and faster.

In this application numbered CN112905316, there is a GPU cluster in which each node has a graphics processing unit, and fast and high-performance results cannot be obtained due to the use of Field-programmable gate array (FPGA).

As a result, there is a need for a fast, simple, accurate and high-performance new technology that can overcome the disadvantages mentioned above and that performs job sequencing in a much shorter time than expected through a handheld terminal with FPGA circuit in a system based on the use of hybrid heuristic optimization algorithms.

Description of the Invention:

This invention is a job sorting device and algorithm based on FPGA-based hybrid heuristic search algorithms, and its feature is; In a system based on the use of hybrid heuristic optimization algorithms, the tasks in the database of the server computer, which will be sequenced to be done on the machines in a factory in milliseconds, by means of a hand terminal with FPGA circuit, which enables to reach the near-optimal solutions quickly and simply, and provides the possibility of parallel data processing. It is a new technology that enables to perform the sorting process.

The invention in order to realize all the objectives mentioned above and which will emerge from the detailed description below; It makes it possible to reach near-optimal solutions in a very short time as a result of running heuristic search algorithms on this hardware circuit, which has an FPGA hardware circuit that can be programmed to perform one or more logical operations with parallel operation capability.

The FPGA hardware circuit in the handheld terminal of the invention is designed in such a way that it can run the hybrid optimization model. In order to realize the purpose of the invention, after the jobs in the database of the server computer, which will be sorted to be done on the machines in a factory, are transferred to the hand terminal, the sequencing process on the FPGA is performed with the hybrid use of algorithms such as Genetic Algorithm (GA), Simulation Annealing (SA) and Particle Swarm Optimization (PSO). being carried out. After the sorting process is done, the relevant results are transferred to the server computer via ethernet (local network) and transmitted to the machines. In this way, work is completed in a short time in workplaces with many business lines.

In the invention; Search algorithms such as Genetic Algorithm (GA), Simulation Annealing (SA) and Particle Swarm Optimization (PSO) are combined to solve the same problem and work interactively. In this way, the solution is reached in a much shorter time with the hybrid algorithm structure of the invention compared to the algorithms working independently of each other. In addition to the fact that the interactive operation of the algorithms allows problem solving in a very short time, with the integration of the FPGA-based system that provides parallel data processing, the problems that many circuits will come together to solve will be done in milliseconds with a single circuit. The parallel data processing capability of the FPGA means that the processes that need to be done independently of each other are carried out simultaneously. Unlike traditional computers, FPGAs enable different operations to be performed simultaneously, and it is possible to perform high-speed, high-performance operations with minimum power consumption. In the proposed model, which is the subject of the invention; By using Genetic Algorithms (GA), Particle Swarm Optimization (PSO) and Simulation Annealing (SA) and Random Search (RA) approaches, it is aimed to reach a solution faster than the existing search algorithms. Each of the search algorithms has its own specific result finding mechanism. By running the three search algorithms in the invention interactively with the random search method, it is aimed to find a better and faster solution than the solution given by the existing search algorithms. Since Genetic Algorithms (GA), Particle Swarm Optimization (PSO) and Simulation Annealing (SA) are hybrid algorithms that work interactively with each other, whichever algorithm finds a better solution, they will continue to search from that point based on the best solution found in the remaining algorithms. Due to the synergy arising from this interaction, better solutions are reached in a short time.

Description of the Figures:

The invention will be described with reference to the accompanying figures, so that the features of the invention will be more clearly understood and appreciated, but the purpose of this is not to limit the invention to these certain regulations. On the contrary, it is intended to cover all alternatives, changes and equivalences that can be included in the area of the invention defined by the accompanying claims. The details shown should be understood that they are shown only for the purpose of describing the preferred embodiments of the present invention and are presented in order to provide the most convenient and easily understandable description of both the shaping of methods and the rules and conceptual features of the invention. In these drawings.

Figure 1 Schematic view of the system.

Figure 2 Total Weighted Delay-Iteration Number Graph view.

The figures to help understand the present invention are numbered as indicated in the attached image and are given below along with their names. Description of References:

Description of the Invention:

The product subject to the invention; FPGA motherboard (1) that runs the hybrid optimization model by including search algorithms such as Genetic Algorithm (GA), Simulation Annealing (SA), Particle Swarm Optimization (PSO) and Random Search (RA), handheld terminal (2) containing FPGA motherboard (1) , FPGA motherboard (1) receives the jobs to be sorted (4) from the database (3) from the server computer (5) and after the FPGA motherboard (1) performs the sorting process, the sorted jobs (6) are sent back to the server computer (5) to the machines (7), includes the transmitted ethernet connection (8).

The algorithm of the product subject to the invention; Generating initial solutions using search algorithms such as Genetic Algorithm (GA), Random Search (RA), Simulation Annealing (SA), Particle and Swarm Optimization (PSO) on the FPGA motherboard (1) jobs to be sorted (4) from the server computer (5) , transferring the best of the produced initial solutions to the initial population of the Genetic Algorithm (GA) according to the number of uses, Particle and Swarm Optimization (PSO) to the initial solution herd, starting the search for better solutions by taking the single best solution for Simulation Annealing (SA), Genetic Algorithm (GA), Starting to run Random Search (RA), Simulation Annealing (SA), Particle and Swarm Optimization (PSO) methods in a hybrid way, if there is a method that finds a better solution than the initial solution as a result of comparing the solution results, transferring this good solution or solutions to other methods When the best solution starts to repeat for a long time as a result of the end of the search process, the system stops working.

Detailed Description of the Invention:

The basic elements that make up the system that is the subject of the invention; FPGA motherboard (1), handheld terminal (2) database (3) server computer (5) machines (7) and ethemet connection (8).

The invention describes a system based on FPGA, such as PSO, GA, SA, which enables search algorithms to run interactively and find a super-fast solution. In the proposed system, which is the subject of the invention; By using Genetic Algorithms (GA), Particle and Swarm Optimization (PSO) and Annealing Simulation (SA) and Random Search (RA) approaches, it provides solutions faster than existing search algorithms. Here, each search approach such as GA, PSO and SA has a unique result finding mechanism. By running these three search methods and random search methods interactively, it is aimed to find a better and faster solution than the solution given by the existing search algorithms. algoritmasimn

1- Generating initial solutions to be used in search algorithms. Generating these initial solutions according to the situation of the problem, if possible, from other known techniques other than heuristic search algorithms. Or randomly generated.

2- Transferring the best of the produced initial solutions to the initial population of GA and of PSO to the initial solution herd, according to the number of uses. Starting the search for better solutions by getting the single best solution for SA.

3- Begins to run GA, PSO, SA and RA methods. If there is a method that finds a better solution than the initial solution, transferring this good solution or solutions to other methods and starting to look for a solution. For example, if SA found a better solution, take that solution and pass PSO to the solution swarm and GA to the solution population. If GA found the better solution, take the best solution and transfer it to SA and SA start looking for the better solution. RA is not directly related to SA and PSO. If RA finds even the most perfect result, SA and PSO take that result over GA's population.

4- When the search process ends and the same best solution repeats for a long time, the system stops working.

Search algorithms are algorithms that each have different solution finding mechanisms. They are algorithms that can find the best solution in a shorter time, even if there are millions of alternative solutions in the search space. Three of these algorithms are used in the proposed solution system, these are; Genetic Algorithms (GA), Annealing Simulation (SA), Particle and Swarm Optimization (PSO).

Genetic algorithms are based on the principle of natural selection, chromosomes in a particular population cross with each other and mutate. As a result, new solution alternatives are obtained. Then, a new population is created using these solution alternatives and the old population. This process continues until a good solution is found. In annealing simulation, new solutions are obtained based on one solution. This algorithm works by taking the cooling process of metals as principle. Particle swarm optimization, on the other hand, looks for a solution based on the movements of bird flocks. Particle swarm optimization, on the other hand, looks for a solution based on the movements of bird flocks.

The aim of the inventive system is to shorten the solution time by using GA, SA and PSO algorithms interactively. Due to the synergy arising from this interaction, better solutions can be found in a short time. We can compare each search algorithm to runners running a relay race. Later, this system is made FPGA-based and produces solutions in milliseconds super-fast.

Claims

CLAIMS 1-- The invention relates to a job sorting device based on FPGA-based hybrid heuristic search algorithms, and its feature is;

- FPGA motherboard (1), which runs the hybrid optimization model by including search algorithms such as Genetic Algorithm (GA), Simulation Annealing (SA), Particle and Swarm Optimization (PSO) and Random Search (RA),

- Handheld terminal (2) containing the FPGA motherboard (1),

The jobs to be sorted (4) received from the database (3) by the FPGA motherboard (1) are received from the server computer (5) and after the FPGA motherboard (1) is sorted, the sorted jobs (6) are sent back to the machines (7) via the server computer (5). It contains ethemet connection (8). - The invention is the algorithm of the job sorting device based on FPGA-based hybrid heuristic search algorithms, and its feature is;

Generating initial solutions using search algorithms such as Genetic Algorithm (GA), Random Search (RA), Simulation Annealing (SA), Particle and Swarm Optimization (PSO) on the FPGA motherboard (1) by taking the jobs (4) to be sorted from the server computer (5) ,

- transferring the best of the initial solutions produced to the initial population of the Genetic Algorithm (GA) and the Particle and Swarm Optimization (PSO) to the initial solution herd, according to the number of uses,

Starting the search for better solutions by taking the single best solution for Simulation Annealing (SA),

- Hybrid operation of Genetic Algorithm (GA), Random Search (RA), Simulation Annealing (SA), Particle and Swarm Optimization (PSO) methods,

If there is a method that finds a better solution than the initial solution as a result of comparing the solution results, this good solution or the transfer of the solutions to other methods Arama i§leminin bitmesi sonucu en iyi gbziim uzun sure tekrar etrneye

sistemin

durdurmasi

basamaklanm igermesi ile karakterize edilmesidir.