CN105701539A - Tend type compact genetic algorithm (GA) based on non-durable elitism strategy - Google Patents
Tend type compact genetic algorithm (GA) based on non-durable elitism strategy Download PDFInfo
- Publication number
- CN105701539A CN105701539A CN201610004867.3A CN201610004867A CN105701539A CN 105701539 A CN105701539 A CN 105701539A CN 201610004867 A CN201610004867 A CN 201610004867A CN 105701539 A CN105701539 A CN 105701539A
- Authority
- CN
- China
- Prior art keywords
- value
- victor
- chromosome
- probability vector
- fitness value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Mathematical Analysis (AREA)
- Algebra (AREA)
- Computational Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Genetics & Genomics (AREA)
- Complex Calculations (AREA)
Abstract
The present invention discloses a tend type compact GA based on a non-durable elitism strategy. The tend type compact GA based on the non-durable elitism strategy comprises the steps of constructing a probability vector P same as a chromosome coding length L, valuating each position of the probability vector as 0.5, and then enabling the probability vector to generate two chromosomes randomly; calculating the fitness values of the generated two chromosomes separately, and comparing the two fitness values to decide a winner and a loser; comparing each position of the generated two chromosomes, and updating the probability vector according to the fitness values; determining whether an algorithm execution result reaches a convergence condition; if the algorithm execution result does not reach a convergence condition, carrying out the mutation operation on the chromosomes; introducing the non-durable elitism strategy and a parameter alpha, and determining the generation numbers of the elitism; if the algorithm execution result reaches the convergence condition, ending the algorithm. According to the present invention, the convergence efficiency of the algorithm is improved, the searching capability is strong, the occupied resource is few, and a good application effect is obtained in the hardware evolution.
Description
Technical field
The invention belongs to Evolvable Hardware technical field, particularly relate to a kind of compact GA method of the trend type based on non-persistent elite retention strategy。
Background technology
After fixing functional hardware and reconfigurable hardware, hardware of future generation will be the hardware that can self-configure and develop, i.e. Evolvable Hardware (EvolvableHardware is called for short EHW), it is to utilize biological developmental pattern to solve problem large-scale, complicated。The thought source of Evolvable Hardware is in the fifties in last century, and the development that the father JohnVonNeumann of computer proposes has the imagination of self-reproduction and self-reparing capability machine。Evolvable Hardware makes it scan in design space due to the fast parallel property of its hardware self, thus realizing design automation, becomes one of study hotspot of Computer Systems Organization and field of electron design automation。Evolvable Hardware refers to and uses evolution algorithmic configuration dynamic reconfigurable circuit in the programmable logic device and finally develop and required logic circuit。Evolvable Hardware can along with the change self structure of the change tread of external environment condition be to realize self-organizing, self adaptation, selfreparing as biology。The research of Evolvable Hardware had important theory significance and practical value。Along with the complexity of hardware system improves constantly, system design difficulty increases, reliability decrease, utilize Evolvable Hardware can meet the hardware adaptability to environment, system is made to adjust its internal structure automatically, in real time, to adapt to the change of interior condition and external environment condition, thus enabling a system to from fault-tolerant operation。Evolvable Hardware technology also will have boundless potential in Embedded System Design, it is possible to well solves the problem of hardware-software partition in Embedded System Design。In a word, Evolvable Hardware all has broad application prospects and huge industry, commercial value in fields such as circuit design, tolerant system, pattern recognition and artificial intelligences。Current Evolvable Hardware technology can only solve fairly simple small-scale circuit, so also facing a lot of problem in its evolution, as tempo of evolution is slow, efficiency of evolution is low, and evolved circuit robustness is low。The key technology of Evolvable Hardware includes PLD and evolution algorithmic。Having randomness due to evolutionary process and evolution number of times is more, thus it requires corresponding device can be configured repeatedly, and the maximum feature of FPGA is online programmable, and therefore FPGA becomes currently more satisfactory and realizes device。The configuration of Evolvable Hardware logic circuit is developed by evolution algorithmic, so the evolution result of evolution algorithmic and evolution efficiency are to whether Evolvable Hardware realizes expectation function important。Genetic algorithm (GA) is a kind of algorithm the most frequently used in evolution algorithmic, it is the random search algorithm of a kind of natural evolution simulated in biosphere, it adopts a series of coding bit string to describe the population of problem candidate solution, but owing to needs preserve substantial amounts of population at individual information, take a large amount of memory space, and computationally intensive when processing challenge, existing many methods and technology improve GA application in FPGA at present。Slow for GA search speed, easily local is precocious, the weakness of local optimal searching ability, at ZhangHui, WuBin, YuZhangguo.Researchofnewgeneticalgorithmsinvolvingmechan ismofsimulatedannealing [J] .JournalofUESTofChina, propose in 2003 to introduce in GA by mechanism of Simulated Annealing, thus combining simulated annealing there is the feature of stronger local search ability and the feature of the more outstanding global optimal control ability of GA, it is effectively improved the operational efficiency of GA and solves quality, but algorithm so can be caused to complicate, and the memory space taken can't be reduced。At FernandoPR, KatkooriS, KeymeulenD, ZebulumR, StoicaA.CustomizableFPGAIPcoreimplementationofageneral-p urposegeneticalgorithmengine [J] .IEEETransEvolComput, proposing to customize an IP kernel in FPGA in 2010 and realize a general genetic algorithm, this genetic algorithm IP kernel can customize according to population quantity, genetic algebra, crossover probability and mutation probability, randomizer seed and fitness function。This method can reduce the logic hardware resource shared by algorithm, reduces memorizer utilization rate, but it is realized by hardware algorithm, does not really improve the performance of algorithm。At MarcoA.Moreno-Armend á riz, NareliCruz-Cort é s, CarlosA.Duchanoy, etal.HardwareimplementationoftheelitistcompactGeneticAlg orithmusingCellularAutomatapseudo-randomnumbergenerator [J] .ComputersandElectricalEngineering, a kind of elite Compact Genetic Algorithms is taught in 2013, it is on the basis of Compact Genetic Algorithms, the elite individuality of every generation to be retained, algorithm is realized by literary composition by hardware, feasibility in hardware is applied in real time is described。But excessive elite retention strategy may result in Premature Convergence, and too high selection pressure makes colony restrain rapidly, sacrifices the multiformity of colony, thus easily it is absorbed in locally optimal solution。
Current Evolvable Hardware technology exists that tempo of evolution is slow, efficiency of evolution is low, evolved circuit robustness is low, the problem of treatable circuit scale and limitednumber。
Summary of the invention
It is an object of the invention to provide a kind of compact GA method of the trend type based on non-persistent elite retention strategy, it is intended to solve current Evolvable Hardware technology and have that tempo of evolution is slow, efficiency of evolution is low, evolved circuit robustness is low, the problem of treatable circuit scale and limitednumber。
The present invention is achieved in that a kind of compact GA method of the trend type based on non-persistent elite retention strategy, and the described compact GA method of the trend type based on non-persistent elite retention strategy includes:
Step one, construct a probability vector P the same with chromosome coding length L, and to make each place value of probability vector be 0.5, then being randomly generated two chromosomes by probability vector, each value of probability vector represents that the corresponding place value of chromosome of generation is the probability of 1；
Step 2, calculate two chromosome fitness values of generation respectively, and compare the two fitness value, the big chromosome of fitness value is as victor, and on the contrary, that fitness value is little is then the vanquished, then each of victor and the vanquished is compared, if two corresponding units values are equal, then continue to compare next bit, otherwise carry out step 3；
Step 3, is reversed in this position of victor, then the individuality after reversion and the fitness value between former individuality are compared, if fitness value becomes big after reversion, judge the value of reversion position, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding place value of update probability by reducing the step-length of 1/N；
Step 4, after each has compared, it is judged that whether probability vector reaches the condition of convergence, reaches, and terminates, it does not have reach, and continues next step, and the condition of convergence is that each in probability vector all converges on 1 or 0；
Step 5, carries out mutation operation to chromosome, judges whether more than 0.5 to each of probability vector, if greater than 0.5, then continues to judge the chromosomal corresponding position of this corresponding victor, is 1 and remains unchanged, be 0 this position of reversing；Otherwise the corresponding position of victor is 1 this position of reversing, and is 0 and remains unchanged。Relatively more consequent new chromosomal fitness value and the chromosomal fitness value of former victor, using chromosome big for fitness value as victor；
Step 6, introduces non-persistent elite retention strategy and parameter alpha, it is judged that the algebraically that elite retains, without more than α, then being produced a new chromosome by probability vector, and turn to step 2, then produced two new chromosomes by probability vector more than α, and turn to step 2。
Further, step one uses probability vector to describe population, and the length of initialization probability vector is L, and it is equal with chromosome length, and in evolutionary process, every generation randomly generates two separate chromosomes according to probability vector value；
Each place value of initialization probability vector is 0.5, and each value of probability vector represents that the corresponding place value of chromosome of generation is the probability of 1。
Further, step 2 calculates two chromosomal fitness values produced by every generation, and these two chromosomes are divided into victor and the vanquished by the size according to fitness value, and that fitness value is big is victor, that fitness value is little is the vanquished, and each of victor and the vanquished is compared。
Further, it is characterised in that step 3 carries out according to the following procedure:
The first step, when each of victor and the vanquished is compared, if the value of victor position corresponding with the vanquished is equal, then continues to compare next bit；
Second step, if victor's value corresponding with the vanquished is unequal, then reverses this position of victor, is 1 and becomes 0, otherwise become 1, and then the new chromosome after this bit reversal and original chromosome carry out the comparison of fitness value；
3rd step, if the new chromosomal fitness value after reversion is more than the fitness value of original chromosome, then substitute original chromosome with new chromosome, and judge the value after reversion, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N；
4th step, if the new chromosomal fitness value after reversion is not more than the fitness value of original chromosome, then retain original chromosome, and judge this value of original chromosome, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N；
5th step, repeat this step until by victor with the vanquished each compared。
Further, step 5 specifically includes:
The first step, if probability vector is not up to the condition of convergence, then carries out mutation operation to chromosome, introduces new mutation operation；
To each of probability vector, second step, judges that whether it is more than 0.5 respectively, if greater than 0.5, then continue to judge to be the value of the chromosomal corresponding position of victor 1 and remain unchanged, be 0 this position of reversing；
3rd step, if it is determined that a certain place value of probability vector is less than 0.5, then continues to judge to be the value of the chromosomal corresponding position of victor 0 and remain unchanged, is 1 this position of reversing；
4th step, relatively more consequent new chromosomal fitness value and the chromosomal fitness value of former victor, using chromosome big for fitness value as victor。
Further, step 6 introduces non-persistent elite retention strategy and parameter alpha, parameter alpha represents the maximum algebraically that elite retains, within α generation, elite individuality is genetic to the next generation, and is produced a new chromosome by probability vector, but more than α generation, elite individuality will be abandoned, and is regenerated two chromosomes by probability vector。
Another object of the present invention is to provide the compact genetic system of trend type of a kind of described trend type compact GA method based on non-persistent elite retention strategy, the described compact genetic system of trend type includes:
Constructing module, for constructing a probability vector P the same with chromosome coding length L, and to make each place value of probability vector be 0.5, then being randomly generated two chromosomes by probability vector, each value of probability vector represents that the corresponding place value of chromosome of generation is the probability of 1；
Computing module, for calculating two chromosome fitness values of generation respectively, and compare the two fitness value, the big chromosome of fitness value is as victor, and on the contrary, that fitness value is little is then the vanquished, then each of victor and the vanquished is compared, if two corresponding units values are equal, then continue to compare next bit；
Reversal block, for being reversed in this position of victor, then the individuality after reversion and the fitness value between former individuality are compared, if fitness value becomes big after reversion, judge the value of reversion position, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding place value of update probability by reducing the step-length of 1/N；
Comparison module, after having compared for each, it is judged that whether probability vector reaches the condition of convergence, reaches, and terminates, it does not have reach, and continues next step, and the condition of convergence is that each in probability vector all converges on 1 or 0；
Mutation operation module, for chromosome is carried out mutation operation, judges whether more than 0.5 to each of probability vector, if greater than 0.5, then continues to judge the chromosomal corresponding position of this corresponding victor, is 1 and remains unchanged, be 0 this position of reversing；Otherwise the corresponding position of victor is 1 this position of reversing, and is 0 and remains unchanged。Relatively more consequent new chromosomal fitness value and the chromosomal fitness value of former victor, using chromosome big for fitness value as victor；
Introduce module, for introducing non-persistent elite retention strategy and parameter alpha, it is judged that the algebraically that elite retains, without more than α, then being produced a new chromosome by probability vector, then produced two new chromosomes by probability vector more than α。
Further, described reversal block farther includes:
Comparing unit, during for each of victor and the vanquished is compared, if the value of victor position corresponding with the vanquished is equal, then continues to compare next bit；
Fitness value comparing unit, if victor's value corresponding with the vanquished is unequal, then reverses this position of victor, is 1 and becomes 0, otherwise become 1, and then the new chromosome after this bit reversal and original chromosome carry out the comparison of fitness value；
Judging unit, if the new chromosomal fitness value after reversion is more than the fitness value of original chromosome, then substitute original chromosome with new chromosome, and judge the value after reversion, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N；
If the new chromosomal fitness value after reversion is not more than the fitness value of original chromosome, then retain original chromosome, and judge this value of original chromosome, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N；
Repeat comparing unit, for repeat to compare until by victor with the vanquished each compared；
Described mutation operation module specifically includes further:
Introduce unit, if probability vector is not up to the condition of convergence, then chromosome is carried out mutation operation, introduce new mutation operation；
To each of probability vector, judging unit, judges that whether it is more than 0.5 respectively respectively, if greater than 0.5, then continues to judge to be the value of the chromosomal corresponding position of victor 1 and remain unchanged, is 0 this position of reversing；
If it is determined that a certain place value of probability vector is less than 0.5, then continues to judge to be the value of the chromosomal corresponding position of victor 0 and remain unchanged, be 1 this position of reversing；
Comparing unit, relatively more consequent new chromosomal fitness value and the chromosomal fitness value of former victor, using chromosome big for fitness value as victor。
Another object of the present invention is to provide a kind of computer system comprising the described compact GA method of trend type based on non-persistent elite retention strategy。
Another object of the present invention is to provide a kind of electric design automation comprising the described compact GA method of trend type based on non-persistent elite retention strategy to control system。
Traditional genetic algorithm adopts coding bit string to describe population, so that when hardware realizes, (wherein L represents chromosome bit string length to the memory space needed with L × N, N represents population number) it is directly proportional, namely the memory element of O (N) magnitude is needed, consume substantial amounts of hardware resource, and it is complex to realize process。Compact Genetic Algorithms (CompactGeneticAlgorithm; CGA) realize while convenient for hardware; but it is appropriate only for the simple problem solving have certain rule; challenge is easy to precocious phenomenon; and often lose defect individual during evolution, particularly in the practical application of hardware lacks enough search capabilities。The compact GA method of trend type based on non-persistent elite retention strategy provided by the invention, genetic algorithm can be solved in hardware develops, take when the big problem of amount of storage processes challenge with CGA the problem that search capability is not enough, be easily absorbed in locally optimal solution, during evolution, both enough selection pressures had been can ensure that, the multiformity of colony can be maintained again, in solving optimal solution with in hardware evolution, there is good execution efficiency。The present invention is based on trend type Compact Genetic Algorithms (the none-persistentelitismCompactGeneticAlgorithmwithTendenc y of non-persistent elite retention strategy, ne-TCGA) improve convergence of algorithm efficiency, evolutionary generation decreases 40%～50% than CGA algorithm, search capability is strong and to take resource few, compared with traditional genetic algorithm, for the chromosome of equal length, it has only to O (log2N) memory element of magnitude, can obtain good application effect in hardware develops。
The present invention compared with prior art, has the following characteristics that
1. the present invention adopts probability vector to describe population, every generation in its evolutionary process is to randomly generate according to probability vector and only produce two separate chromosomes, so internal memory is few shared by the present invention, especially very useful in the application of this limited memory of such as Evolvable Hardware。
2. the present invention has on the basis of clear and definite end condition at CGA algorithm and adds the judgement towards optimal solution trend of the current solution, accelerate the execution speed of algorithm, enhance the search capability of algorithm, improve the extractability to defect individual information；The mutation operation improved, it is possible to obtain better chromosome with maximum probability。Owing to excessive elite retention strategy may result in precocity, too high selection pressure makes colony restrain rapidly, sacrifice the multiformity of colony, thus being absorbed in locally optimal solution, in being introduced into non-persistent elite retention strategy and parameter alpha, α represents the maximum algebraically that elite retains, within α generation, elite individuality can be genetic to the next generation, but more than α generation, it will regenerated two chromosomes by probability vector。
Accompanying drawing explanation
Fig. 1 is the compact GA method flow diagram of the trend type based on non-persistent elite retention strategy that the embodiment of the present invention provides。
Fig. 2 is the flow chart of the embodiment 1 that the embodiment of the present invention provides。
Function used (1) curve chart when Fig. 3 is the checking present invention of embodiment of the present invention offer。
Function used (2) curve chart when Fig. 4 is the checking present invention of embodiment of the present invention offer。
Fig. 5 be the embodiment of the present invention provide with TCGA, CGA algorithm comparison diagram when finding a function (1) maximum。
Fig. 6 be the embodiment of the present invention provide with TCGA, CGA algorithm comparison diagram when finding a function (2) maximum。
The Evolvable Hardware platform figure that Fig. 7 is used when being the checking present invention of embodiment of the present invention offer application in Evolvable Hardware。
The evolution circuit CellArray array that Fig. 8 is designed when being the checking present invention of embodiment of the present invention offer application in Evolvable Hardware。
The internal structure schematic diagram of cell unit in the evolution circuit CellArray array that Fig. 9 is designed when being the checking present invention of embodiment of the present invention offer application in Evolvable Hardware。
Figure 10 be the embodiment of the present invention provide with TCGA, CGA algorithm comparison diagram when evolution one-bit full addres。
Detailed description of the invention
In order to make the purpose of the present invention, technical scheme and advantage clearly understand, below in conjunction with embodiment, the present invention is further elaborated。Should be appreciated that specific embodiment described herein is only in order to explain the present invention, is not intended to limit the present invention。
Below in conjunction with accompanying drawing, the application principle of the present invention is explained in detail。
As it is shown in figure 1, the compact GA method of the trend type based on non-persistent elite retention strategy of the embodiment of the present invention comprises the following steps:
S101: construct a probability vector P the same with chromosome coding length L, and to make each place value of probability vector be 0.5, is then randomly generated two chromosomes by probability vector；
S102: calculating two chromosome fitness values of generation respectively, and compare the two fitness value, that fitness value is big is victor, and that fitness value is little is the vanquished, compares each of victor and the vanquished；
S103: if two corresponding units values are equal, then continue to compare next bit, is otherwise reversed in this position of victor, carry out update probability vector according to its fitness value；
S104: evaluation algorithm performs whether result reaches the condition of convergence；
S105: if not up to the condition of convergence, then carry out mutation operation to chromosome；
S106: introduce non-persistent elite retention strategy and parameter alpha, it is judged that the algebraically that elite retains, is then produced a new chromosome by probability vector without more than α, and continue executing with algorithm, if beyond α, then produced two new chromosomes by probability vector, and continue executing with algorithm。
Below in conjunction with specific embodiment, the application principle of the present invention is further described。
With reference to Fig. 2, the present invention is a kind of compact GA method of the trend type based on non-persistent elite retention strategy, specifically comprises the following steps that
Step 1, constructs a probability vector P and produces two chromosomes according to it。
1.1. initializing evolutionary generation i=1, use probability vector to describe population, construct a probability vector P, the length of initialization probability vector is L, and it is equal with chromosome length。
1.2. each place value of initialization probability vector is 0.5, and each value of probability vector represents that the corresponding place value of chromosome of generation is the probability of 1, and in evolutionary process, every generation randomly generates two separate chromosomes according to probability vector value。
Step 2, calculates and compares two chromosomal fitness values。
Two chromosomal fitness values produced by every generation are calculated according to fitness function, and these two chromosomes are divided into victor and the vanquished by the size according to fitness value, that fitness value is big is victor, that fitness value is little is the vanquished, and victor and the vanquished are compared by turn, if two numbers of corresponding position are equal, then continue to compare next bit, otherwise carrying out step 3, step 3 then compares chromosomal next bit after performing to terminate, until two chromosomal each is all completeer。
Step 3, according to victor's chromosome update probability vector。
3.1. the comparative result provided according to step 2, victor's value corresponding with the vanquished is unequal, is then reversed this position of victor, is 1 and becomes 0, otherwise become 1, and then the new chromosome after this bit reversal and original chromosome carry out the comparison of fitness value。
3.2. the size of the new chromosomal fitness value after reversion and the fitness value of original chromosome is judged, if the new chromosomal fitness value after reversion is more than the fitness value of original chromosome, then substitute original chromosome with new chromosome, and judge the value after reversion, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N。
If 3.3. the new chromosomal fitness value after reversion is not more than the fitness value of original chromosome, then retain original chromosome, and judge this value of original chromosome, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N。
Step 4, it is judged that algorithm performs whether result reaches the condition of convergence。
Judging whether the updated probability vector of step 3 reaches the condition of convergence, namely the value of each of probability vector all converges to 0 or 1, if probability vector reaches the condition of convergence, algorithm terminates, and otherwise carries out step 5。
Step 5, carries out mutation operation to chromosome。
5.1. judge that each value of probability vector decides whether that the corresponding positions to candidates is individual performs mutation operation。
5.2. if it is determined that a certain place value of probability vector is more than 0.5, then continue to judge to be the value of the chromosomal corresponding position of victor 1 and remain unchanged, be 0 this position of reversing。
5.3. if it is determined that a certain place value of probability vector is less than 0.5, then continue to judge to be the value of the chromosomal corresponding position of victor 0 and remain unchanged, be 1 this position of reversing。
5.4., after traveling through each of probability vector, relatively more consequent new chromosomal fitness value and the chromosomal fitness value of former victor, using chromosome big for fitness value as victor。
Step 6, it is judged that the algebraically that elite retains, and produce new chromosome。
Introduce non-persistent elite retention strategy and parameter alpha, parameter alpha represents the maximum algebraically that elite retains, within α generation, elite individuality can be genetic to the next generation, and is produced a new chromosome by probability vector, and evolutionary generation i adds 1, but more than α generation, elite individuality will be abandoned, and is regenerated two chromosomes, evolutionary generation i=1 by probability vector。Then turning to step 2 to re-execute this algorithm until reaching the condition of convergence after execution of step 6, algorithm terminates。
Below in conjunction with experiment, the application effect of part is described in detail。
One, with ne-TCGA and CGA, TCGA algorithm respectively to function maximizing
1. find a function maximum
Use tri-kinds of algorithms of CGA, TCGA, ne-TCGA respectively to following two function maximizings, and each population number operation of every kind of algorithm is averaged for 10 times。Result is such as shown in Fig. 3, Fig. 4, Fig. 5, Fig. 6, Fig. 3,4 is function curve diagram, Fig. 5,6 is the performance and required evaluation number of times comparing result figure that solve。
2. simulation result and analysis
It can be seen that along with population number purpose increases, the evaluation number of times of three kinds of algorithms all increases, but at identical population number now, the evaluation least number of times of ne-TCGA algorithm, and the performance solved is best, wherein to retain solving speed during value is population algebraically the 50% of algebraically α the fastest for elite。When solved function (2) maximum, although CGA searches the maximum in definition territory in evolutionary process, but final result converges to Local modulus maxima, does not try to achieve maximum, the poor performance solved, and the final result that obtains of TCGA algorithm also not always globally optimal solution。Therefore ne-TCGA algorithm has more superiority than CGA, TCGA algorithm in solved function max problem。
Two, ne-TCGA application in Evolvable Hardware
The fpga chip EP2S30F484I4 development board selecting altera corp StratixII series designs from Evolution System, thereon ne-TCGA algorithm is carried out validation verification。It is currently based on that the implementation of the Evolvable Hardware of FPGA can be divided between plate evolving, plate level chip chamber is evolved, chip-scale is evolved, and chip-scale evolution is to realize evolutionary process on a single chip, save data communication time, also it is can the development trend of evolutionary system, so selecting which to be designed, the realization of its hardware platform such as Fig. 7。
1. design is from the hardware circuit of Evolution System
Design one and wherein comprise the adaptive IP core based on virtual reconfigurable circuits from Evolution System, it be one by the CellArray array of 40 cellularities, such as Fig. 8。This CellArray array has 8 externally inputs and 8 outputs。
Basic logic dispensing unit Cell in the CellArray array of design is as it is shown in figure 9, it is made up of look-up table and MUX。Owing to the evolved circuit model of 3 inputs look-up table (LUT) is most effective, so select 3 input LUT herein。Although the input of CellArray array external is 8, but the input of first 8 Cell be provided by outside 8 input and these 8 input reversely constitute, the Cell input of secondary series is to be made up of 8 outputs of 8 outside inputs and first row, the input of the 3rd row and columns afterwards is all be made up of the output of first two columns, so each Cell unit has 3 16 the selector selecting 1, the input of LUT below is determined again in the output of selector as control bit。Because each 16 select the selector configuration of 1 to need 4 chromosomes, LUT look-up table needs 8, so single Cell configuration needs 20 altogether。Owing to CellArray array comprises 40 Cell unit, so a CellArray array needs 800 chromosomes to configure。
Adaptive IP core has 8 externally inputs, 8 outputs, and one-bit full addres only has 3 inputs and 2 outputs, so the 8 kinds of combinations only taking latter three of outside 8 inputs input as full adder, this operation will realize by C language in NiosIIIDE。When after input test vector, the output of CellArray array carries out inclusive OR operation with the truth table of 8 kinds of inputs, obtain matched signal, and it is high 6 for shielding matched signal to design a MASK depositor, because the output of full adder only has 2, hexadecimal number 0x03 put into by MASK depositor, it and matched signal are carried out AND-operation, output matchdata, in matchdata, the number of 1 is fitness value, owing to full adder input has 8 kinds of combinations, so fitness value is 16 to the maximum, representing develops correct one-bit full addres。
2. simulation result and analysis
Respectively the ne-TCGA algorithm adopted, TCGA algorithm and CGA algorithm being programmed by C language in the soft core of NiosII, then passing through JTAG download cable will carry out the evolution of one-bit full addres on download program to development board, evolution result is as shown in Figure 10。As can be seen from the figure CGA algorithm is relative to other two kinds of algorithms, its evolution efficiency is worst, and although TCGA algorithm evolutionary generation is minimum, but can not develop required circuit, ne-TCGA algorithm successfully develops in limited evolutionary generation and one-bit full addres, therefore ne-TCGA algorithm has effectiveness in hardware develops application, decrease the probability being absorbed in locally optimal solution, convergence and search capability remarkable。Ne-TCGA algorithm will have very big potentiality with the advantage that the simple of its implementation is few with shared memory space within hardware in hardware develops。
The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all any amendment, equivalent replacement and improvement etc. made within the spirit and principles in the present invention, should be included within protection scope of the present invention。
Claims (10)
1. the compact GA method of the trend type based on non-persistent elite retention strategy, it is characterised in that the described compact GA method of the trend type based on non-persistent elite retention strategy includes:
Construct a probability vector P, then randomly generated two chromosomes by probability vector；Calculate two chromosome fitness values of generation respectively, and compare the two fitness value to have decided victor and the vanquished；Relatively two chromosomal each, if two corresponding units values are equal, then continue to compare next bit, otherwise reversed in this position of victor, and carry out update probability vector according to its fitness value of victor；
Evaluation algorithm performs whether result reaches the condition of convergence；If not up to the condition of convergence, then chromosome is carried out mutation operation；Introduce non-persistent elite retention strategy and parameter alpha, it is judged that the algebraically that elite retains；If reaching the condition of convergence, then algorithm terminates。
2. the compact GA method of trend type based on non-persistent elite retention strategy as claimed in claim 1, it is characterised in that the described compact GA method of the trend type based on non-persistent elite retention strategy includes:
Step one, construct a probability vector P the same with chromosome coding length L, and to make each place value of probability vector be 0.5, then being randomly generated two chromosomes by probability vector, each value of probability vector represents that the corresponding place value of chromosome of generation is the probability of 1；
Step 2, calculate two chromosome fitness values of generation respectively, and compare the two fitness value, the big chromosome of fitness value is as victor, and on the contrary, that fitness value is little is then the vanquished, then each of victor and the vanquished is compared, if two corresponding units values are equal, then continue to compare next bit, otherwise carry out step 3；
Step 3, is reversed in this position of victor, then the individuality after reversion and the fitness value between former individuality are compared, if fitness value becomes big after reversion, judge the value of reversion position, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding place value of update probability by reducing the step-length of 1/N；
Step 4, after each has compared, it is judged that whether probability vector reaches the condition of convergence, reaches, and terminates, it does not have reach, and continues next step, and the condition of convergence is that each in probability vector all converges on 1 or 0；
Step 5, carries out mutation operation to chromosome, judges whether more than 0.5 to each of probability vector, if greater than 0.5, then continues to judge the chromosomal corresponding position of this corresponding victor, is 1 and remains unchanged, be 0 this position of reversing；Otherwise the corresponding position of victor is 1 this position of reversing, and is 0 and remains unchanged；Relatively more consequent new chromosomal fitness value and the chromosomal fitness value of former victor, using chromosome big for fitness value as victor；
Step 6, introduces non-persistent elite retention strategy and parameter alpha, it is judged that the algebraically that elite retains, without more than α, then being produced a new chromosome by probability vector, and turn to step 2, then produced two new chromosomes by probability vector more than α, and turn to step 2。
3. the compact GA method of trend type based on non-persistent elite retention strategy as claimed in claim 2, it is characterized in that, step one uses probability vector to describe population, the length of initialization probability vector is L, it is equal with chromosome length, in evolutionary process, every generation randomly generates two separate chromosomes according to probability vector value；
Each place value of initialization probability vector is 0.5, and each value of probability vector represents that the corresponding place value of chromosome of generation is the probability of 1。
4. the compact GA method of trend type based on non-persistent elite retention strategy as claimed in claim 2, it is characterized in that, step 2 calculates two chromosomal fitness values produced by every generation, and these two chromosomes are divided into victor and the vanquished by the size according to fitness value, that fitness value is big is victor, that fitness value is little is the vanquished, and each of victor and the vanquished is compared；
Step 3 carries out according to the following procedure:
The first step, when each of victor and the vanquished is compared, if the value of victor position corresponding with the vanquished is equal, then continues to compare next bit；
Second step, if victor's value corresponding with the vanquished is unequal, then reverses this position of victor, is 1 and becomes 0, otherwise become 1, and then the new chromosome after this bit reversal and original chromosome carry out the comparison of fitness value；
3rd step, if the new chromosomal fitness value after reversion is more than the fitness value of original chromosome, then substitute original chromosome with new chromosome, and judge the value after reversion, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N；
4th step, if the new chromosomal fitness value after reversion is not more than the fitness value of original chromosome, then retain original chromosome, and judge this value of original chromosome, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N；
5th step, repeat this step until by victor with the vanquished each compared。
5. the compact GA method of trend type based on non-persistent elite retention strategy as claimed in claim 2, it is characterised in that step 5 specifically includes:
The first step, if probability vector is not up to the condition of convergence, then carries out mutation operation to chromosome, introduces new mutation operation；
To each of probability vector, second step, judges that whether it is more than 0.5 respectively, if greater than 0.5, then continue to judge to be the value of the chromosomal corresponding position of victor 1 and remain unchanged, be 0 this position of reversing；
3rd step, if it is determined that the corresponding place value of probability vector is less than 0.5, then continues to judge to be the value of the chromosomal corresponding position of victor 0 and remain unchanged, is 1 this position of reversing；
4th step, relatively more consequent new chromosomal fitness value and the chromosomal fitness value of former victor, using chromosome big for fitness value as victor。
6. the compact GA method of trend type based on non-persistent elite retention strategy as claimed in claim 2, it is characterized in that, step 6 introduces non-persistent elite retention strategy and parameter alpha, parameter alpha represents the maximum algebraically that elite retains, and within α generation, elite individuality is genetic to the next generation, and produced a new chromosome by probability vector, but more than α generation, elite individuality will be abandoned, and regenerated two chromosomes by probability vector。
7. the compact genetic system of trend type of the trend type compact GA method based on non-persistent elite retention strategy as claimed in claim 2, it is characterised in that the described compact genetic system of trend type includes:
Constructing module, for constructing a probability vector P the same with chromosome coding length L, and to make each place value of probability vector be 0.5, then being randomly generated two chromosomes by probability vector, each value of probability vector represents that the corresponding place value of chromosome of generation is the probability of 1；
Computing module, for calculating two chromosome fitness values of generation respectively, and compare the two fitness value, the big chromosome of fitness value is as victor, on the contrary, that fitness value is little is then the vanquished, then each of victor and the vanquished is compared, if two corresponding units values are equal, then continue to compare next bit；
Reversal block, for being reversed in this position of victor, then the individuality after reversion and the fitness value between former individuality are compared, if fitness value becomes big after reversion, judge the value of reversion position, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding place value of update probability by reducing the step-length of 1/N；
Comparison module, after having compared for each, it is judged that whether probability vector reaches the condition of convergence, reaches, and terminates, it does not have reach, and continues next step, and the condition of convergence is that each in probability vector all converges on 1 or 0；
Mutation operation module, for chromosome is carried out mutation operation, judges whether more than 0.5 to each of probability vector, if greater than 0.5, then continues to judge the chromosomal corresponding position of this corresponding victor, is 1 and remains unchanged, be 0 this position of reversing；Otherwise the corresponding position of victor is 1 this position of reversing, and is 0 and remains unchanged；Relatively more consequent new chromosomal fitness value and the chromosomal fitness value of former victor, using chromosome big for fitness value as victor；
Introduce module, for introducing non-persistent elite retention strategy and parameter alpha, it is judged that the algebraically that elite retains, without more than α, then being produced a new chromosome by probability vector, then produced two new chromosomes by probability vector more than α。
8. the compact genetic system of trend type as claimed in claim 7, it is characterised in that described reversal block farther includes:
Comparing unit, during for each of victor and the vanquished is compared, if the value of victor position corresponding with the vanquished is equal, then continues to compare next bit；
Fitness value comparing unit, if victor's value corresponding with the vanquished is unequal, then reverses this position of victor, is 1 and becomes 0, otherwise become 1, and then the new chromosome after this bit reversal and original chromosome carry out the comparison of fitness value；
Judging unit, if the new chromosomal fitness value after reversion is more than the fitness value of original chromosome, then substitute original chromosome with new chromosome, and judge the value after reversion, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N；
If the new chromosomal fitness value after reversion is not more than the fitness value of original chromosome, then retain original chromosome, and judge this value of original chromosome, it is 1 just carry out the update probability corresponding place value of vector by increasing the step-length of 1/N, is 0 just carry out the vectorial corresponding value of update probability by reducing the step-length of 1/N；
Repeat comparing unit, for repeat to compare until by victor with the vanquished each compared；
Described mutation operation module specifically includes further:
Introduce unit, if probability vector is not up to the condition of convergence, then chromosome is carried out mutation operation, introduce new mutation operation；
To each of probability vector, judging unit, judges that whether it is more than 0.5 respectively respectively, if greater than 0.5, then continues to judge to be the value of the chromosomal corresponding position of victor 1 and remain unchanged, is 0 this position of reversing；
If it is determined that a certain place value of probability vector is less than 0.5, then continues to judge to be the value of the chromosomal corresponding position of victor 0 and remain unchanged, be 1 this position of reversing；
Comparing unit, relatively more consequent new chromosomal fitness value and the chromosomal fitness value of former victor, using chromosome big for fitness value as victor。
9. the computer system comprised described in claim 1-6 any one based on the compact GA method of the trend type of non-persistent elite retention strategy。
10. one kind comprise described in claim 1-6 any one based on the compact GA method of the trend type of non-persistent elite retention strategy electric design automation control system。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610004867.3A CN105701539A (en) | 2016-01-04 | 2016-01-04 | Tend type compact genetic algorithm (GA) based on non-durable elitism strategy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610004867.3A CN105701539A (en) | 2016-01-04 | 2016-01-04 | Tend type compact genetic algorithm (GA) based on non-durable elitism strategy |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105701539A true CN105701539A (en) | 2016-06-22 |
Family
ID=56226931
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610004867.3A Pending CN105701539A (en) | 2016-01-04 | 2016-01-04 | Tend type compact genetic algorithm (GA) based on non-durable elitism strategy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105701539A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101789044A (en) * | 2010-01-27 | 2010-07-28 | 武汉大学 | Method of implementing cooperative work of software and hardware of genetic algorithm |
| CN102254225A (en) * | 2011-06-03 | 2011-11-23 | 中国科学院西安光学精密机械研究所 | Evolution hardware implementation method based on trend type compact genetic algorithm |
-
2016
- 2016-01-04 CN CN201610004867.3A patent/CN105701539A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101789044A (en) * | 2010-01-27 | 2010-07-28 | 武汉大学 | Method of implementing cooperative work of software and hardware of genetic algorithm |
| CN102254225A (en) * | 2011-06-03 | 2011-11-23 | 中国科学院西安光学精密机械研究所 | Evolution hardware implementation method based on trend type compact genetic algorithm |
Non-Patent Citations (4)
| Title |
|---|
| STOMEO.E ETAL.: "On Evolution of Relatively Large Combinational Logic Circuits", 《IN: PROC.OF THE 2005 NASA/DOD CONFERENCE OF EVOLUTION HARDWARE》 * |
| Y.JEWAJINDA ETAL.: "A Cooperative Approach to Compact Genetic Alogrithm for Evolvable Hardware", 《IN: PROC. OF THE 2006 IEEE CONGRESS ON EVOLUTIONARY COMPUTATION》 * |
| 姜庆辉: "演化硬件技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 朱继祥 等: "基于演化硬件的在线自适应系统", 《计算机科学》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lu et al. | An efficient hardware accelerator for sparse convolutional neural networks on FPGAs | |
| Shim et al. | Adaptive memetic computing for evolutionary multiobjective optimization | |
| CN105049061A (en) | Advanced calculation-based high-dimensional polarization code decoder and polarization code decoding method | |
| CN111461335B (en) | MPI multi-process-based noise-containing single quantum logic gate implementation method and device | |
| CN113222133B (en) | FPGA-based compressed LSTM accelerator and acceleration method | |
| Rahimi et al. | Grapefruit: An open-source, full-stack, and customizable automata processing on FPGAs | |
| Boguslawski et al. | Twin neurons for efficient real-world data distribution in networks of neural cliques: Applications in power management in electronic circuits | |
| CN104881549B (en) | A kind of power consumption optimization method of the Reed Muller logic circuits comprising outlier | |
| CN106919504B (en) | Test data evolution generation method based on GA algorithm | |
| Meng et al. | QTAccel: A generic FPGA based design for Q-table based reinforcement learning accelerators | |
| CN107483178A (en) | A kind of device and smart card for realizing Secure Hash Algorithm SHA3 | |
| Gordon | Exploring models of development for evolutionary circuit design | |
| CN115146782A (en) | Quantum line compiling method, device, compiling framework and quantum operating system | |
| CN105701539A (en) | Tend type compact genetic algorithm (GA) based on non-durable elitism strategy | |
| CN111832144B (en) | Full-amplitude quantum computing simulation method | |
| CN115879562B (en) | Quantum program initial mapping determination method and device and quantum computer | |
| CN115829047B (en) | Quantum program final mapping determination method and device and quantum computer | |
| Huang et al. | CAMA: Energy and memory efficient automata processing in content-addressable memories | |
| CN114064235A (en) | Multitask teaching and learning optimization method, system and equipment | |
| CN111104767B (en) | Variable-precision random gradient descending structure and design method for FPGA | |
| Gangwar et al. | Hardware/software co-design of a high-speed Othello solver | |
| CN111401555A (en) | Model training method, device, server and storage medium | |
| CN117632770B (en) | Multipath coverage test case generation method and system | |
| CN113868242B (en) | Global index distribution strategy adjusting method, computer equipment and storage medium | |
| CN115730670B (en) | Method and device for generating mode file, medium and electronic device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160622 |