TW201349119A - Parallel processing system for combining genetic algorithms with the ant colony optimization algorithm to obtain the optimal result - Google Patents

Abstract

A parallel processing system for combining genetic algorithms with the ant colony optimization (ACO) algorithm to obtain the optimal result, comprising: a host computer and a plurality of sub-cluster computers connected to one another via a network; the host computer has a first main program as well as a main thread, the first main program is used for executing the main thread, a resource control thread and a plurality of connection communication threads; each of the sub-cluster computers has a second main program, a listening thread and an ant server thread, where the second main program is used for executing the listening thread and the ant server thread. Through the preceding threads, the host computer is enabled to deliver computation tasks to the sub-cluster computers for computation, and the host computer is responsible for the gene decoding, acquiring parameters for ACO, and deciding the best fit parameters.

Description

Parallel processing system that can be combined with genetic algorithm and ant colony algorithm to obtain optimal results

The invention relates to a solution technique for solving an optimization through a computer system, in particular to a parallel processing system which can combine genetic algorithm and ant colony algorithm to obtain an optimized result.

At present, genetic algorithm and ant colony algorithm are widely used optimization methods, and some people apply it to the solution of combination optimization problems such as Vehicle Routing Problem (VRP). A good result.

However, although the genetic algorithm has the simplicity and versatility of the structure, it is easy to generate modules, expand and apply to various optimization problems, but the computational energy it consumes is higher, that is, the CPU time consumed by it is higher. More, so the solution is less efficient.

As for the ant colony algorithm, although the theoretical framework is based on the shortest path search, it will be more CPU time spent on solving the combination optimization problem related to network optimization, such as the vehicle route problem. Genetic algorithms come prevailing. However, because the ant colony algorithm itself has a lot of parameters to be set, and the parameters often interact with the problem type, scale and route exchange improvement program, it is easy to cause the parameter set value suggested by the previous literature or experiment to be invalid. Therefore, to get better results, researchers often have to invest a lot of time and effort to test various possible combinations of parameters.

For parallel processing techniques of genetic algorithms or ant colony algorithms, there are currently conceptual architectures (eg Ghiani, Guerriero, Laporte, & Musmanno, 2003; Berger, Barkaoui, & Braysy, 2004) or clustered computers (Cluster) PCs/Worksations) (eg Drummond, Ochi, & Vianna, 1999; Doerner, Hartl, Kieshle, Lucka, & Reimann, 2004). However, the former does not consider how to communicate in the actual multi-computer environment, while the latter must install an operating system that supports cluster computers (such as Linux Cluster, Microsoft Windows Computer Cluster Sever 2003). Since China currently uses Windows users mostly, it is necessary to construct a cluster computer in the Unix/Linux environment, and there are many restrictions on the application.

The main object of the present invention is to provide a parallel processing system capable of obtaining an optimization result by combining a genetic algorithm and an ant colony algorithm, which can use a personal computer of a general Windows operating system to construct a parallel processing cluster system, and This system combines genetic algorithms with ant colony algorithms to find optimal results.

In order to achieve the foregoing object, a parallel processing system according to the present invention which can be combined with a genetic algorithm and an ant colony algorithm to obtain an optimized result includes: a plurality of personal computers equipped with Microsoft Windows system (Windows). Defining one of the personal computers as a master computer, and defining the remaining personal computers as sub-cluster computers, the personal computers are connected to each other through a network; the main control computer has a first main program and a main thread, the first a main program is used to execute the main thread, a resource control thread, and a plurality of connection communication threads; each of the sub-cluster computers has a second main program, a listening thread, and an ant server thread. The second main program is configured to execute the listening thread and the ant server thread; wherein, when executing the main thread, establishing and starting a resource control thread, establishing an initial group and an individual, and then searching for each The sub-cluster computer corresponds to a user-side message to determine which sub-cluster computer is currently available without calculation work, and those with high computational efficiency are preferentially selected, and each individual's gene is decoded to obtain an ant. The parameters of the group algorithm, after finding the available sub-cluster computer, deliver the individual's fitness value calculation work, so that the computing work is continuously assigned to each sub-cluster computer that can be used until all the individual's fitness values have been sought Finally, the individuals with the best fitness values are recorded, and then the mating and mutation of the genetic algorithm are performed until the maximum evolution algebra is reached, and finally The individual output with the best fitness value; when executing the resource control thread, the connection parameters are first set and the connection structure is established, so as to find out the sub-cluster computers currently connected with the main control computer, and then Each of the sub-cluster computers establishes a connection communication thread, and then establishes a user-side message for each of the sub-cluster computers, and then updates the corresponding user-side message with the connection communication thread; When the thread communicates with the thread, it waits for the execution result of the ant server thread of the corresponding sub-cluster computer, and waits for the main thread to record the adaptive value and the sub-cluster computer computing performance through the synchronization mechanism, and then updates the user. Ending the message, and transmitting the parameters of the next group of ant colony algorithms to the listening thread; when each of the sub-cluster computers executes the listening thread, first establishing a connection with the host computer Line, then start the ant server thread, then notify the master computer to prepare, and then enter the after-command state, after receiving the parameters of the ant colony algorithm transmitted by the master computer connection thread, The synchronization mechanism allows the ant server thread to perform calculations; when executing the ant server thread, the system first updates its corresponding client message to indicate that it has received the calculation work, and then sets the ant colony algorithm parameters to perform the ant. The group algorithm is used to obtain the best result, and then the pheromones are updated, and then the corresponding client message is updated to indicate that there is no calculation work at present, and the best result is converted into the fitness value, and the fitness value is used as the execution result. Transfer to the connection communication thread corresponding to the host computer.

Preferably, the network is an Ethernet network.

Preferably, when the main thread is executed, the adaptive value of the sub-cluster computer backhaul is obtained, and the operation time of the sub-cluster computer is recorded and sorted to obtain the operation efficiency of each sub-cluster computer, and then When assigning a new computing task to a sub-cluster computer without computational work, it is preferentially assigned to a sub-cluster computer with higher computational efficiency.

Preferably, when the main thread is executed, after each evolution generation records the individual with the best fitness value, it is checked that all the connected sub-cluster computers have computing workers and change the corresponding client information. In order to interrupt the operation, and transmit to the listening thread of each sub-cluster computer, subsequent gene decoding and subsequent actions are performed; after receiving the user information of the interrupt operation, the listening thread forcibly ends the ant servo The thread is executed and the new ant server thread is restarted.

Preferably, the parameters of the ant colony algorithm include initial pheromone concentration, preference degree of pheromone concentration α, degree of preference β of the heuristic function, random selection threshold Q ₀ , pheromone region renewal evaporation rate, And the pheromones globally update the evaporation rate.

In order to explain the structure and features of the present invention in detail, the following description of the preferred embodiments and the accompanying drawings, wherein: FIG. A parallel processing system 10 capable of obtaining an optimized result by combining a genetic algorithm and an ant colony algorithm is mainly composed of a plurality of personal computers 11 in which: the personal computer 11 is equipped with a Microsoft Windows system (Windows). One of the personal computers 11 is defined as a master computer 12, and the remaining personal computers 11 are defined as sub-cluster computers 22, which are connected to each other via the network 99. In this embodiment, the network 99 is an Ethernet network.

The main control computer 12 has a first main program 121 and a main thread 123. The first main program 121 is configured to execute the main thread 123, a resource control thread 125, and a plurality of connection communication threads 127.

Each of the sub-cluster computers 22 has a second main program 221, an listening thread 223, and an ant server thread 225. The second main program 221 is configured to execute the listening thread 223 and the ant server thread. 225.

Wherein, when the main thread 123 is executed, a resource control thread 125 is established and started, and an initial group and an individual are established, and then the search is performed. Looking up a user-end message corresponding to each sub-cluster computer 22 to determine which sub-cluster computer 22 is currently usable without calculation work, and those with high computational efficiency are preferentially selected, and the genes of each individual are decoded. And obtaining the parameters of the ant colony algorithm, finding the usable sub-cluster computer 22, then delivering an individual's fitness value calculation work, so that the computing work is continuously assigned to each sub-cluster computer 22 that can be used until all the individuals The fitness values have been obtained, and finally the individuals with the best fitness values are recorded, and then the mating and mutation of the genetic algorithm are performed until the maximum evolution algebra is reached, and finally the individual with the best fitness value is output. In this embodiment, when the main thread 123 is executed, the adaptive values returned by the sub-cluster computer 22 are obtained, and the operation time of the sub-cluster computer 22 is recorded and sorted to obtain the sub-cluster computer 22 The computational efficiency is then preferentially assigned to the sub-cluster computer 22, which is more computationally efficient, when assigning a new computational task to the sub-cluster computer 22 without computational work, i.e., in accordance with the foregoing ordering.

When the resource control thread 125 is executed, the connection parameters are first set and the connection architecture is established, so as to find out the sub-cluster computers 22 currently connected to the main control computer 12, and then to the sub-cluster computers 22 A connection communication thread 127 is established, and the user terminal information is established for each of the sub-cluster computers 22, and the corresponding communication terminal 127 is used to update the corresponding client information.

When the connection communication thread 127 is executed, the connection relationship shown in the fourth figure waits for the execution result transmitted by the ant server thread 225 of the corresponding sub-cluster computer 22, and waits through the synchronization mechanism. The main thread 123 records the fitness value and the performance of the sub-cluster computer 22, and then updates The client message and the parameters of the next set of ant colony algorithms are transmitted to the listening thread 223.

When the sub-cluster computer 22 executes the listening thread 223, the connection with the main control computer 12 is first established, the ant server thread 225 is restarted, and then the main control computer 12 is notified that the preparation is completed. Then, after entering the post-command state, after receiving the parameters of the ant colony algorithm transmitted from the thread communication thread 127 corresponding to the main control computer 12, the ant server thread 225 performs the calculation through the synchronization mechanism.

When the ant server thread 225 is executed, the corresponding client message is updated to indicate that the computing work has been received, and then the parameters of the ant colony algorithm are set, and the ant colony algorithm is performed to obtain the best result, and then The pheromone update, and then update its corresponding client message indicates that there is currently no calculation work, and the best result is converted into an adaptation value, and the adaptation value is transmitted as an execution result to the connection corresponding to the main control computer 12. Line communication thread 127. The parameters of the ant colony algorithm include initial pheromone concentration, preference degree of pheromone concentration α, preference degree of heuristic function β, random selection threshold Q ₀ , pheromones region renewal evaporation rate, and fee Lomon's global update evaporation rate.

It should be noted that, in the embodiment, when the main thread 123 is executed, after each evolution generation records the individual with the best fitness value, the current sub-cluster computer 22 of all connections may be checked first. The computing worker changes the corresponding client message to the interrupt operation and transmits it to the listening thread 223 of each of the sub-cluster computers 22, after which the subsequent genetic algorithm and its subsequent actions are performed; the listening thread 223 is Received interrupted shipment After the client message is calculated, the ant server thread 225 is forcibly terminated, and the new ant server thread 225 is restarted. In this way, after finding the best fitness value for each evolution generation, all ant server threads 225 are forcibly terminated, so that all the sub-cluster computers 22 stop computing to avoid unnecessary waste of computing resources. However, this action of forcibly ending the calculation work of each sub-cluster computer 22 is not a necessary technical feature of the present case, that is, even if the compulsory termination technique is not provided and other technical features are the same as the case, the patent scope of the case should be Covered.

Next, the operational state of the present invention will be described.

Referring to the first to fourth figures, assuming that the user wants to find the best result of a certain vehicle routing problem, the main control computer 12 and the sub-cluster computers 22 are first turned on, and the main control computer 12 executes the The first main program 121, and the sub-clusters computer 22 execute the second main program.

After the main control computer 12 executes the first main program 121, the main thread 123 is further executed, thereby establishing and starting a resource control thread 125, and establishing an initial group of related data of the vehicle path problem. The group and the individual, then by the relationship between the resource lock and the network connection, determine which sub-cluster computer 22 is currently not computing, and decode the genes of the individual to obtain the parameters of the ant colony algorithm. When a certain sub-cluster computer 22 is found to have no calculation work, the calculation of the fitness value of an individual is performed together with the parameters of the ant colony algorithm, so that the adaptive value calculation work of each individual is continuously assigned to each sub-cluster computer 22 The calculation work is completed by each of the sub-cluster computers 22 until the fitness values of all the individuals have been found.

After that, the mating and mutation of the genetic algorithm is performed until the maximum evolution algebra is reached, and finally the individual with the best fitness value is output, and the best result of the vehicle's desired problem is obtained.

The operation of the resource control thread 125, the connection communication thread 127, the listening thread 223, and the ant server thread 225 are similar to the above description, and are not repeated here.

Accordingly, the present invention can use a personal computer of a Windows system, and one of the personal computers executes the first main program 121, and the remaining personal computers execute the second main program 221 (the users of these computers can still Multiplexing execution of other programs without being affected by execution of the second main program 221 can form the system of the present invention without the use of a Unix/Linux system architecture or a cluster computer. Not only is the cost low, but also easy to use.

It can be seen from the above that the achievable effect of the present invention is that a parallel system of parallel processing can be constructed by using a personal computer of a general Windows operating system, and the system is combined with a genetic algorithm and an ant colony algorithm to obtain an optimized result. .

10‧‧‧Parallel processing system that combines genetic algorithm and ant colony algorithm to obtain optimal results

11‧‧‧Personal Computer

12‧‧‧Master computer

121‧‧‧First main program

123‧‧‧Master Thread

125‧‧‧Resource Control Thread

127‧‧‧Connected communication thread

22‧‧‧Sub-cluster computer

221‧‧‧Second main program

223‧‧‧ Listening to the thread

225‧‧‧ Ant Server Thread

99‧‧‧Network

The first figure is a schematic diagram of the architecture of a preferred embodiment of the present invention.

The second figure is a partial schematic diagram of a preferred embodiment of the present invention, showing the architecture of the host computer.

The third figure is a partial schematic diagram of a preferred embodiment of the present invention, showing the architecture of a sub-cluster computer.

The fourth figure is a detailed schematic diagram of a preferred embodiment of the present invention, showing the interaction between the host computer and the sub-cluster computer when the connection communication thread is executed.

10‧‧‧Parallel processing system that combines genetic algorithm and ant colony algorithm to obtain optimal results

12‧‧‧Master computer

121‧‧‧First main program

123‧‧‧Master Thread

125‧‧‧Resource Control Thread

127‧‧‧Connected communication thread

22‧‧‧Sub-cluster computer

221‧‧‧Second main program

223‧‧‧ Listening to the thread

225‧‧‧ Ant Server Thread

Claims (5)

A parallel processing system capable of combining the genetic algorithm and the ant colony algorithm to obtain an optimized result, comprising: a plurality of personal computers installed with Microsoft Windows system (Windows), defining one of the computers as a master computer, and Defining the remaining personal computers as sub-cluster computers, the personal computers are connected to each other through a network; the main control computer has a first main program and a main thread, and the first main program is used to execute the main thread, a resource control thread and a plurality of connection communication threads; each of the sub-cluster computers has a second main program, a listening thread, and an ant server thread, and the second main program is used to execute the listening thread And the ant server thread; wherein, when executing the main thread, establishing and starting a resource control thread, establishing an initial group and an individual, and then searching for a client message corresponding to each of the child cluster computers To determine which sub-cluster computer is currently available without calculation work, and those with high computational efficiency are selected first, and the base of each individual is Decoding to obtain the parameters of the ant colony algorithm, and after finding the available sub-cluster computer, it delivers an individual's fitness value calculation work, so that the computing work is continuously assigned to each sub-cluster computer that can be used until all the individuals The fitness values have been obtained, and finally the individuals with the best fitness value are recorded, and then the mating and mutation of the genetic algorithm are performed until the maximum evolution algebra is reached, and finally the individual with the best fitness value is output; When the thread is executed, the connection parameters are first set and built. a connection structure to find out the sub-cluster computers currently connected to the main control computer, and then establish a connection communication thread for each sub-cluster computer, and then establish a user for each sub-cluster computer The end message, and then the connection communication thread is used to update the corresponding client message; when executing the connection communication thread, the system first waits for the corresponding ant server computer ant server thread to send it. Execution result, after waiting for the main thread to record the fitness value and the sub-cluster computer computing performance through the synchronization mechanism, update the user terminal message, and transmit the parameters of the next group of ant colony algorithm to the listening thread; When the cluster computer executes the listening thread, it first establishes a connection with the main control computer, then starts the ant server thread, then notifies the main control computer that preparation is completed, and then enters the post-command state, and is connected. After the main control computer connects the ant parameter of the ant colony algorithm transmitted by the thread, the ant server executes the calculation through the synchronization mechanism; the ant server is executed When the thread is online, the corresponding user-side message is updated to indicate that it has received the calculation work, and then the parameters of the ant colony algorithm are set, the ant colony algorithm is performed to obtain the best result, and then the pheromone update is performed, and then the update is performed. The corresponding client message indicates that there is no calculation work at present, and the best result is converted into an adaptation value, and the adaptation value is transmitted as an execution result to the connection communication thread corresponding to the main control computer. According to the patent application scope, the parallel processing system capable of obtaining the optimization result can be obtained by combining the genetic algorithm and the ant colony algorithm, wherein the network is an Ethernet network. Combinable genetic algorithm according to item 1 of the patent application scope A parallel processing system with an ant colony algorithm to obtain an optimized result, wherein: when the main thread is executed, after each evolution generation records the individual with the best fitness value, the current connected line is checked. The cluster computer has a computing worker, and changes its corresponding client message to an interrupt operation, and transmits it to the listening thread of each sub-cluster computer, and then performs subsequent gene decoding and subsequent actions; the listening thread is receiving After interrupting the client message of the operation, the ant server thread is forcibly terminated, and the new ant server thread is restarted. According to the patent application scope, the parallel processing system capable of obtaining the optimization result can be obtained by combining the genetic algorithm and the ant colony algorithm, wherein: when the main thread is executed, the sub-cluster computer is returned. Adapting the values, and recording and sorting the operation time of the sub-cluster computer to obtain the computational efficiency of each sub-cluster computer, and then assigning a new computing task to the sub-cluster computer without calculation work, that is, assigning the operation preferentially A more efficient sub-cluster computer. A parallel processing system capable of obtaining an optimized result according to the genetic algorithm and the ant colony algorithm according to claim 1 of the patent application scope, wherein: the parameter of the ant colony algorithm includes an initial pheromone concentration and a fee The degree of preference of Lomon concentration, the degree of preference β of the heuristic function, the threshold of random selection Q ₀ , the rate of renewal of the pheromone region, and the rate of pheromones renewal.