WO2018137442A1 - 一种数据预测方法及装置 - Google Patents

一种数据预测方法及装置 Download PDF

Info

Publication number
WO2018137442A1
WO2018137442A1 PCT/CN2017/117003 CN2017117003W WO2018137442A1 WO 2018137442 A1 WO2018137442 A1 WO 2018137442A1 CN 2017117003 W CN2017117003 W CN 2017117003W WO 2018137442 A1 WO2018137442 A1 WO 2018137442A1
Authority
WO
WIPO (PCT)
Prior art keywords
individual
population
determining
fitness function
individuals
Prior art date
Application number
PCT/CN2017/117003
Other languages
English (en)
French (fr)
Inventor
章政
Original Assignee
中国银联股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 中国银联股份有限公司 filed Critical 中国银联股份有限公司
Publication of WO2018137442A1 publication Critical patent/WO2018137442A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/004Artificial life, i.e. computing arrangements simulating life
    • G06N3/006Artificial 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]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/12Computing arrangements based on biological models using genetic models
    • G06N3/126Evolutionary algorithms, e.g. genetic algorithms or genetic programming

Definitions

  • the present application relates to the field of data processing technologies, and in particular, to a data prediction method and apparatus.
  • Support vector regression is a new learning method based on statistical learning theory. It has many advantages such as complete theory, strong adaptability, global optimization, short training time and good generalization performance. It has become a hot research topic at home and abroad.
  • the support vector regression machine solves the contradiction between the complexity of the model classifier complexity and the nuclear function method through the result risk minimization criterion and the kernel function method, which has attracted great attention from scholars in the field of pattern recognition. Since then, it has developed rapidly and has been successfully applied in many fields (bioinformatics, text categorization, handwriting recognition, face detection, etc.), and it has been comparable or better than traditional methods in the research process.
  • the present application provides a data prediction method and apparatus for solving the problems of large space occupation, premature convergence, and insufficient mountain climbing ability in the prior art.
  • An embodiment of the present application provides a data prediction method, where the method includes:
  • Constructing a first population comprising N individuals, wherein each individual comprises a model parameter value of a support vector regression machine prediction model, wherein N is greater than or equal to 1;
  • the optimal solution is selected from the first population of the Y-order genetic algorithm iteration as the model parameter of the support vector regression machine prediction model.
  • the tabu search algorithm is used in the process of cross-calculation in the genetic algorithm and the process of mutation calculation to generate a new population, the diversity of the population is increased compared with the prior art. To overcome the problem of insufficient mountain climbing ability in genetic algorithm, to avoid premature convergence to local optimal solution.
  • the determining the individuals selected as the intersecting individuals in the first population comprises:
  • the first fitness function value is a model parameter of the predictive model of the support vector regression machine The difference between the obtained predicted value and the true value is determined;
  • whether the two individuals are cross-over individuals are determined according to the random numbers of the two individuals and the crossover probability, instead of determining the crossover probability by using the setting in the prior art to determine whether the individual is a cross-over individual, and increasing the population Individual diversity.
  • p c is the crossover probability between any two of the individuals
  • p cmax is the maximum crossover probability
  • p cmin minimum crossover probability indicates whether or not to be used for the crossover probability.
  • urien number of iterations for the GA current indicates whether or not to be used for the crossover probability.
  • max gen is the maximum number of iterations
  • F avg is An average of first fitness function values for all individuals in the first population, the F' being a smaller first fitness function value of the two individuals.
  • the crossover probability is related to the first fitness function value of the two individuals, and is related to the current number of iterations, instead of using the fixed probability of the prior art, increasing the diversity of the population and avoiding premature convergence. .
  • the determining, by the tabby search algorithm, the cross-child generation according to the tabu search algorithm to generate the second population including:
  • the crossover individual after the crossover individual is determined, the child children after the intersection are determined, and the first population is updated according to the child generation, and the second population after the intersection is obtained.
  • updating the first population to obtain a second population according to the first fitness function value of the individual of the intersecting individual and the second fitness function value of the child individual including:
  • the first fitness function average of the all intersecting individuals Determining the first fitness function average of the all intersecting individuals, wherein the first fitness function average is determined according to a first fitness function value of all intersecting individuals; if the child is determined And the second fitness function value of the generation individual is greater than the average value of the first fitness function, and the child individual is replaced with a cross individual corresponding to the child individual in the first population to obtain a second population; or
  • the second fitness function value of the child individual is not greater than the average value of the first fitness function, and If the descendant individuals are not in the tabu search list, the child individuals are replaced with the cross individuals corresponding to the child individuals in the first population to obtain a second population.
  • the value of the second fitness function is not greater than the average value of the first fitness function.
  • the individuals whose progeny are not in the tabu search list increase the diversity of the population and avoid premature convergence compared with the prior art.
  • the mutating operation is performed on each individual in the second population to obtain a third population, including:
  • the third population is determined by the tabu search algorithm, which increases the diversity of the population and avoids premature convergence compared with the prior art.
  • determining the optimal individual from the third population according to the tabu search algorithm comprises:
  • the individual corresponding to the largest fourth fitness function value is taken as the optimal individual, wherein the taboo expected level value is according to the The third fitness function value of each individual in the two populations is determined;
  • the maximum fourth adaptation is performed.
  • the individual corresponding to the degree function value is the optimal individual.
  • the value of the fourth fitness function value that is not greater than the taboo expectation level value is accepted. Individuals who are not in the taboo form increase the diversity of the population and avoid premature convergence.
  • the method further includes:
  • a second predicted value is determined based on the predicted value and the adjustment factor.
  • the model can have a better prediction effect on periodic data.
  • the application also provides a data prediction device, including:
  • a first population determining unit for constructing a first population comprising N individuals, wherein each individual comprises a model parameter value of a support vector regression machine prediction model, wherein N is greater than or equal to 1;
  • a second population determining unit configured to perform a genetic algorithm iteration on the first population, for each genetic iteration: determining an individual selected as a crossover individual in the first population, and determining an individual of the intersecting individual according to The tabu search algorithm determines the cross children and generates the second population;
  • a second population variation unit configured to perform a variation iteration on the second population, and perform iteration for each mutation generation: performing a mutation operation on each individual in the second population to obtain a third population; according to the tabu search algorithm Determining the optimal individual in the third population; determining the first population of the next genetic algorithm iteration from the M optimal individuals of the M mutation iterations;
  • the model parameter determining unit is configured to select an optimal solution from the first population of the Y-order genetic algorithm iteration as a model parameter of the support vector regression machine prediction model.
  • the tabu search algorithm is used in the process of cross-calculation in the genetic algorithm and the process of mutation calculation to generate a new population, the diversity of the population is increased compared with the prior art. To overcome the problem of insufficient mountain climbing ability in genetic algorithm, to avoid premature convergence to local optimal solution.
  • the second population determining unit is specifically configured to:
  • the first fitness function value is a model parameter of the predictive model of the support vector regression machine The difference between the obtained predicted value and the true value is determined;
  • p c is the crossover probability between any two of the individuals
  • p cmax is the maximum crossover probability
  • p cmin minimum crossover probability indicates whether or not to be used for the crossover probability.
  • urien number of iterations for the GA current indicates whether or not to be used for the crossover probability.
  • max gen is the maximum number of iterations
  • F avg is An average of first fitness function values for all individuals in the first population, the F' being a smaller first fitness function value of the two individuals.
  • the second population determining unit is specifically configured to:
  • the second population determining unit is specifically configured to:
  • the first fitness function average of the all intersecting individuals Determining the first fitness function average of the all intersecting individuals, wherein the first fitness function average is determined according to a first fitness function value of all intersecting individuals; if the child is determined And the second fitness function value of the generation individual is greater than the average value of the first fitness function, and the child individual is replaced with a cross individual corresponding to the child individual in the first population to obtain a second population; or
  • the child fitness function has a second fitness function value that is not greater than the first fitness function average value, and the child generation individual is not in the taboo search list, and the child generation individual is replaced by the first population group
  • the second population is obtained by the intersecting individuals corresponding to the progeny individuals.
  • the second population variation unit is specifically configured to:
  • the second population variation unit is specifically configured to:
  • the individual corresponding to the largest fourth fitness function value is taken as the optimal individual, wherein the taboo expected level value is according to the The third fitness function value of each individual in the two populations is determined;
  • the maximum fourth adaptation is performed.
  • the individual corresponding to the degree function value is the optimal individual.
  • the device further includes:
  • the adjusting unit is configured to obtain a real value in the set time, and divide the real value into a plurality of intervals according to the set step size, and determine an adjustment factor of the different interval, wherein each interval corresponds to a time factor;
  • a second predicted value is determined based on the predicted value and the adjustment factor.
  • An embodiment of the present application further provides an electronic device, including: at least one processor; and a memory communicably coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor The instructions are executed by the at least one processor to enable the at least one processor to perform the data prediction method of the embodiments of the present application.
  • the embodiment of the present application further provides a non-transitory computer storage medium storing computer executable instructions for causing the computer to perform the implementation of the present application.
  • the data prediction method of the example is a non-transitory computer storage medium storing computer executable instructions for causing the computer to perform the implementation of the present application.
  • the embodiment of the present application further provides a computer program product, the computer program product comprising a computing program stored on a non-transitory computer readable storage medium, the computer program comprising the computer executable instructions, when the computer When the executable instructions are executed by the computer, the computer is caused to perform the data prediction method of the embodiment of the present application.
  • the tabu search algorithm is used in the process of cross-calculation in the genetic algorithm and the process of mutation calculation to generate a new population, the diversity of the population is increased compared with the prior art. To overcome the problem of insufficient mountain climbing ability in genetic algorithm, to avoid premature convergence to local optimal solution.
  • FIG. 1 is a schematic flowchart of a data prediction method according to an embodiment of the present application.
  • FIG. 2 is a schematic structural diagram of a data prediction apparatus according to an embodiment of the present application.
  • FIG. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
  • the present application provides a data prediction method, which is performed by an electronic device. As shown in FIG. 1, the method includes:
  • Step 101 constructing a first population comprising N individuals, wherein each individual comprises a model parameter value of a support vector regression machine prediction model, wherein N is greater than or equal to 1;
  • Step 102 Perform a genetic algorithm iteration on the first population, for each genetic iteration: An individual selected as a crossover individual in the first population, and an individual of the intersecting individual is determined to cross the progeny according to a tabu search algorithm to generate a second population;
  • Step 103 performing a variation iteration on the second population, and performing iteration for each mutation generation: performing a mutation operation on each individual in the second population to obtain a third population; and from the third population according to the tabu search algorithm Determining the optimal individual; determining the first population of the next genetic algorithm iteration from the M optimal individuals of the M mutation iterations;
  • Step 104 Select an optimal solution from the first population of the Y-order genetic algorithm iteration as a model parameter of the support vector regression machine prediction model.
  • the establishment of the support vector regression machine prediction model may be described as:
  • the x in the data set is mapped into the high-dimensional feature space F, and a linear function f(x) is found in the feature space F, so that f(x) can well in the feature space F Linear regression is performed on the data set S. Assume:
  • the problem of solving the support vector regression machine prediction model is converted into the problem of solving the optimal solution of the model parameters (C, ⁇ 2 , ⁇ ) in the support vector regression machine prediction model.
  • a first population comprising N individuals is constructed, wherein each individual includes three parameter values in the parameter model described above, for example, optionally, assuming that the values of the three model parameters are all in the range of 0 to An integer between 7 can be represented by a 3-bit unsigned binary integer.
  • an individual in the first population is 101110101, indicating that the three parameter values are 5, 6, and 5, respectively.
  • the method further includes the step of constructing an initial population, wherein the number of individuals of the initial population is greater than or equal to the number of individuals in the first population, and the first population is determined by an initial population selection operation.
  • the initial population includes six individuals, namely, individual A, individual B, individual C, individual D, individual E, and individual F.
  • the termination condition is the maximum number of iterations and the minimum error limit, the maximum number of iterations can be set to 1000, and the minimum error limit can be set to 10-8.
  • the fitness function values of the six individuals in the initial population are first determined.
  • the fitness function value of each individual may be determined by the following formula:
  • the fitness function value of each individual is determined based on the difference between the model value represented by each individual and the actual value after substituting the model value into the SVR model.
  • the selection policy is a proportional selection policy.
  • the proportional selection strategy for each individual i, assuming that the function value of the fitness is F i , the selection probability of the individual i can be expressed as:
  • Table 1 Individual fitness function value table
  • the individual selecting the operation is determined, that is, the individual in the first population needs to be determined, that is, the individual not to be selected is eliminated.
  • a selection operation can be performed by the roulette method to determine which individual is ultimately selected.
  • the selection process is 4 times, and the individual A, the individual B, the individual C, and the individual D are respectively selected, and then the 4 individuals are included in the first population.
  • step 102 a cross-algorithm operation of the first population is required.
  • a first fitness function value of each of the N individuals in the first population needs to be determined, and according to the method for determining the first population, the individual in the first population may be determined.
  • the first fitness function value is determined.
  • determining the probability of crossover between any two individuals in the first population and determining the A random number determines whether the two individuals are intersecting individuals.
  • the crossover probability between the two individuals may be determined according to the following formula:
  • p c is the crossover probability between any two of the individuals
  • p cmax is the maximum crossover probability
  • p cmin minimum crossover probability indicates whether or not to be used for the crossover probability.
  • urien number of iterations for the GA current indicates whether or not to be used for the crossover probability.
  • max gen is the maximum number of iterations
  • F avg is An average of first fitness function values for all individuals in the first population, the F' being a smaller first fitness function value of the two individuals.
  • the probability of crossing between two bodies in the first population is randomly determined, for example, the first in the first population.
  • the fitness function value is as shown in Table 1. It is assumed that the crossover probability between the individual A and the individual B is randomly determined. First, it is determined that the first fitness function value of the individual A is lower according to the first fitness function value of the two individuals. F' is the fitness function value of the individual A, and F avg is the average first fitness function value of the four individuals, alternatively, 23.75, and the crossover probability between the individual A and the individual B is p cmax .
  • step 102 after determining the crossover probability between any two individuals in the first population, it is also necessary to determine the random numbers of the two individuals, and determine the two according to the random numbers of the two individuals and the crossover probability. Whether individuals are cross-over individuals.
  • whether the two individuals can be selected as the parent of the crossover algorithm can be determined by randomly generating a random number RN between 0 and 1 for the two individuals, and determining the two individuals by the RN and the determined crossover probability of the two individuals. That is, cross individuals.
  • the chromosome is selected as a parent chromosome.
  • two individuals are individual A and individual B, and individual A And the crossover probability of individual B is 0.9, and the random number generated by individuals to A and B is 0.6, that is, the random number generated by individuals to A and B is smaller than the crossover probability of individual A and individual B, so individual A and individual B Determined to be a cross individual.
  • step 102 after determining the intersecting individuals in the first population, a crossover algorithm is performed to determine the crossed progeny generated after the crossover of the intersecting individuals, and determining the progeny individuals generated by the intersecting individuals and determining the second fitness of the progeny individuals. a function value; updating the first population to obtain the second population according to the first fitness function value of each of the intersecting individuals and the second fitness function value of the child individual.
  • the first population is updated according to the first fitness function value of the individual of the cross individual and the second fitness function value of the child individual, and the specific process is as follows:
  • a first fitness function average of all intersecting individuals Determining a first fitness function average of all intersecting individuals, wherein the first fitness function average is determined according to a first fitness function value of all intersecting individuals; and if the second fitness function value of the child individual is determined to be greater than The average value of the first fitness function replaces the progeny individual with the cross individual corresponding to the progeny individual in the first population to obtain the second population; or
  • the second fitness function value of the child individual is not greater than the average value of the first fitness function, and the child individual is not in the tabu search list, and the child individual is replaced by the child corresponding to the child group in the first population, and the Two populations.
  • the second population may be determined by the following steps:
  • Step 201 Set the average value of the first fitness function value of the parent cross-over individual as the taboo expectation level value, and determine whether the parental individual's fitness value is taboo. If the fitness function value of the progeny individual is better than the taboo expectation level value, the progeny individual is accepted and enters the next generation; if the progeny individual has a fitness function value that is worse than the taboo expected level value, then the process proceeds to step 202. .
  • Step 202 If the child individual is not in the taboo list, the child individual is accepted and enters the next generation. If the offspring individual is in the taboo list, then the one with the better fitness value in the crossover individual is selected to enter the next generation.
  • the selected crossover individuals in the first population are four individuals A, B, C, and D, and the intersection manner is A, B individuals as cross individuals, and C and D individuals as crosses.
  • the crossover algorithm is respectively performed to obtain the cross children a, b, c, and d, respectively.
  • a single point cross algorithm and a multi-point cross algorithm may be selected to determine a crossover individual.
  • the second fitness function value of each cross child is determined by formula (4), assuming that the determined second fitness function value of a is 25, the fitness function value of b is 20, and the fitness function value of c is 21
  • the fitness function value of d is 26, the average value of the first fitness function value of the intersecting individual is 23.75, and 23.75 is regarded as the taboo expectation level.
  • the individual a and the individual d are accepted as the child individuals, and the second fitness function value of the individual b and the individual c is lower than the taboo expectation level. It is necessary to determine whether the individual b and the individual c are in the contraindication list.
  • the individual b and the individual c are in the taboo list, it indicates that the individual b and the individual c have been regarded as the optimal solution and accepted in the previous genetic algorithm, so if the individual b and the individual c In the taboo list, the two progeny individuals are not accepted, and the individuals with the first fitness value of the individual B and the individual C are accepted.
  • the first fitness value of the parent (in the individual C and the individual D) of the receiving individual c is selected.
  • the individual with the first fitness value is the individual C
  • the second population is the individual a, the individual b, the individual C, and the individual d.
  • the child since the value of the second fitness function of the child individual is not greater than the average value of the first fitness function when the child is accepted, and the child is not in the tabu search list, the child is The generation of individuals replaces the case of the intersecting individuals corresponding to the offspring of the first population, so that the diversity of the individuals in the second population is more abundant than in the prior art.
  • step 103 after obtaining the second population, the second population is subjected to a mutation operation, first determining a third fitness function value of each individual in the second population; and then determining a mutation probability of each individual in the second population. And determining the random number of each individual, determining whether the individual is a variant individual according to the mutation probability and the mutation probability of each individual; performing a mutation operation on the individual determined to be the variant individual, and determining the result of the mutation operation, determining the first according to the result of the mutation operation Three populations.
  • the third fitness function value of each individual in the second population may be determined by determining the second population, for example, the second population is individual a, individual b, individual C, and individual. d, the third fitness function value of each individual has been determined.
  • the method for determining a variant individual is similar to the method for determining a crossover individual.
  • the mutation probability of each of the second population may be determined according to the following formula:
  • p m represents the probability of variation
  • p mmax represents the maximum variation of probability
  • p mmin represents a minimum probability variation
  • F represents individual values of the fitness function.
  • each individual in the second population After determining the mutation probability of each individual in the second population, it is possible to determine whether each individual is determined by RM and the determined crossover probability of each individual by randomly generating a random number RM between 0 and 1 for each individual. Can be selected as a variant individual.
  • the determined mutation probability of the individual a is 0.3, and the random number generated for the individual a is 0.1, and then the individual a is determined to be a mutant individual.
  • individuals in the second population may be mutated, and some individuals may not mutate.
  • the result of the variation of the second population is that the individual a is mutated to If the individual aa and the other second population do not mutate, the newly generated third population is the individual aa, the individual b, the individual C, and the individual d.
  • the iteration termination condition may be set to 1000 times, that is, if it has been mutated 1000 times, Stop the mutation, otherwise determine the optimal solution.
  • the optimal individual is determined from the third population according to the tabu search algorithm, and the specific method is as follows:
  • step 301 it is determined whether the optimal candidate solution reaches the taboo expectation level. If the taboo expectation level is reached, the taboo expectation level value is updated and the current solution is updated to the candidate solution and the next variation is entered. If the taboo expectation level is not reached, go to step 302.
  • the taboo expectation level is an average value of the third fitness function value of the second population, and if the determined optimal solution in the variant progeny reaches the taboo expectation level, the taboo expectation level value is updated, and Update the optimal solution in the variant progeny to the candidate solution, and liberate the optimality in the variant progeny into the taboo list, and then proceed to the next mutation process, which is the largest in the variant progeny.
  • the fourth fitness function value corresponds to the individual.
  • Step 302 in the embodiment of the present application, if the optimal solution of the generated variant child does not reach the tabu expectation level, then the candidate optimal solution is selected as the current solution, if and only if the optimal solution is not in the taboo table, Free the best of the variant offspring into the taboo list and proceed to the next mutation process.
  • the fourth fitness function value of the currently optimal variant individual is not greater than the third fitness function average value when the variant individual is accepted, and the individual is not in the tabu search list, Therefore, compared with the prior art, the diversity of solutions generated by the variation is increased.
  • the generated third population is re-entered as the first population to step 101, and the process of crossing and mutating is continued.
  • the third population is (aaa, bb, c, ddd), then in step 101, the individual in the first population is (aaa, bb). , c, ddd).
  • step 104 after determining that the genetic algorithm satisfies the termination condition, the obtained optimal individual in the first population is the optimal solution of the SVR model parameters to be solved, wherein the optimal solution is the first population.
  • the individual in the substituting into the optimal fitness function value solved in the SVR model is the optimal solution of the SVR model parameters to be solved, wherein the optimal solution is the first population.
  • the concept of the adjustment factor is introduced according to the temporal characteristics of the data.
  • the adjustment factor can be expressed as:
  • S i is an adjustment factor and y i represents a true value.
  • y i represents a true value.
  • i j, j + d, ..., j + (m - 1) d, 0 ⁇ j ⁇ d, d represents the step size of the time series, and m represents the maximum time length of the time series.
  • the values of the plurality of adjustment factors may be determined by the following formula:
  • the adjustment factors corresponding to each time period are obtained by summing the adjustment factors of the plurality of time periods by the root.
  • T represents the last time period point corresponding to the predicted value.
  • the optional set time is the transaction data for 2000-2010 10 years. If d is set to one month, there are 12 transaction data for each year. In contrast, there are 12 forecast data for each year.
  • an adjustment factor can be determined according to formula (8), that is, there are 12 adjustment factors for January, and 12 adjustment factors for February, and so on.
  • the obtained 12 adjustment factors for each month are summed and then divided by 12, and then the root number is opened, and the obtained value is the adjustment factor for the month.
  • the predicted value that needs to be predicted is the transaction data for January 2011.
  • the corresponding adjustment factor for January is determined.
  • the adjustment factor for each January in the 10 years is squared and then divided by 12 and then the root number is obtained.
  • the value is then multiplied by the 1st estimate in January 2011 to obtain the adjusted predicted value.
  • the embodiment of the present application further provides a data prediction apparatus, as shown in FIG. 2, including:
  • a first population determining unit S1 for constructing a first population comprising N individuals, wherein each individual comprises a model parameter value of a support vector regression machine prediction model, wherein N is greater than or equal to 1;
  • a second population determining unit S2 configured to perform a genetic algorithm iteration on the first population, for each genetic iteration: determining an individual selected as a crossover individual in the first population, and the individual of the intersecting individual Determining the cross children according to the tabu search algorithm to generate the second population;
  • a second population variation unit S3 configured to perform a variation iteration on the second population, and perform iteration for each mutation generation: performing a mutation operation on each individual in the second population to obtain a third population; according to the tabu search algorithm Determining an optimal individual in the third population; determining a first population of the next genetic algorithm iteration from the M optimal individuals of the M mutation variants, wherein M is greater than or equal to 1, less than or equal to the number of termination iterations of the mutation algorithm;
  • the model parameter determining unit S4 is configured to select an optimal solution from the first population of the Y-order genetic algorithm iteration as a model parameter of the support vector regression machine prediction model, where Y is the number of termination iterations of the genetic algorithm.
  • the second population determining unit S2 is specifically configured to:
  • the first fitness function value is a model parameter of the predictive model of the support vector regression machine The difference between the obtained predicted value and the true value is determined;
  • p c is the crossover probability between any two of the individuals
  • p cmax is the maximum crossover probability
  • p cmin minimum crossover probability indicates whether or not to be used for the crossover probability.
  • urien number of iterations for the GA current indicates whether or not to be used for the crossover probability.
  • max gen is the maximum number of iterations
  • F avg is An average of first fitness function values for all individuals in the first population, the F' being a smaller first fitness function value of the two individuals.
  • the second population determining unit S2 is specifically configured to:
  • the second population determining unit S2 is specifically configured to:
  • the first fitness function average of the all intersecting individuals Determining the first fitness function average of the all intersecting individuals, wherein the first fitness function average is determined according to a first fitness function value of all intersecting individuals; if the child is determined And the second fitness function value of the generation individual is greater than the average value of the first fitness function, and the child individual is replaced with a cross individual corresponding to the child individual in the first population to obtain a second population; or
  • the child fitness function has a second fitness function value that is not greater than the first fitness function average value, and the child generation individual is not in the taboo search list, and the child generation individual is replaced by the first population group
  • the second population is obtained by the intersecting individuals corresponding to the progeny individuals.
  • the second population variation unit S3 is specifically configured to:
  • the second population variation unit S3 is specifically configured to:
  • the individual corresponding to the largest fourth fitness function value is taken as the optimal individual, wherein the taboo expected level value is according to the The third fitness function value of each individual in the two populations is determined;
  • the maximum fourth adaptation is performed.
  • the individual corresponding to the degree function value is the optimal individual.
  • the device further includes:
  • the adjusting unit S5 is configured to obtain a real value in the first set time and a predicted value obtained according to the support vector regression machine prediction model, and divide the real value and the predicted value into multiple steps according to the set step size. For each interval, determining an adjustment factor corresponding to the interval according to the true value corresponding to the interval and the predicted value; and obtaining a prediction model according to a support vector regression machine for obtaining a second set time a first predicted value, and determining an interval corresponding to the predicted value, acquiring an adjustment factor corresponding to the interval according to the interval; determining a second according to the first predicted value of the second set time and the adjustment factor Predictive value.
  • each module involved in the above embodiments is a logic module.
  • a logical unit may be a physical unit, a part of a physical unit, or multiple physical entities. A combination of units is implemented.
  • a unit that is not closely related to solving the technical problem proposed by the present application is not introduced, but this does not mean that there are no other units in the present embodiment.
  • the embodiment of the present application further provides an electronic device, as shown in FIG. 3, including: at least one processor 210; a transceiver 200; and a memory 220 communicatively coupled to the at least one processor 210;
  • One or more processors 210 and memory 220 are exemplified by one processor 210 in FIG.
  • the electronic device of the interface display method of the electronic device may further include: a transceiver 200, an input device 230, and an output device 240.
  • the transceiver 200, the processor 210, the memory 220, the input device 230, and the output device 240 may be connected by a bus or other means, as exemplified by a bus connection in FIG.
  • the memory 220 is a non-volatile computer readable storage medium and can be used for storing a non-volatile software program, a non-volatile computer executable program, and a module, such as a program corresponding to the data prediction method provided in the embodiment of the present application.
  • An instruction/module (for example, the first population determining unit S1, the second population determining unit S2, the second population variation unit S3, the model parameter determining unit S4, the adjusting unit S5, and the like shown in FIG. 2).
  • the processor 210 executes various functional applications of the server and data processing by executing non-volatile software programs, instructions, and modules stored in the memory 220, that is, implementing the data prediction method in the above embodiments.
  • the memory 220 may include a storage program area and an storage data area, wherein the storage program area may store an operating system, an application required for at least one function; the storage data area may store data created according to usage of the data device predicted by the data, and the like.
  • memory 220 can include high speed random access memory 220, and can also include non-volatile memory 220, such as at least one disk storage 220 piece, flash memory device, or other non-volatile solid state memory 220 piece.
  • memory 220 can optionally include memory 220 remotely located relative to processor 210, which can be coupled to the processing device of the data prediction method over a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
  • Input device 230 can receive input numeric or character information and generate key signal inputs related to user settings and function controls of the electronic device.
  • the output device 240 can include a display device such as a display screen.
  • One or more modules are stored in memory 220, and when executed by one or more processors 210, perform the data prediction methods provided in any of the method embodiments described above.
  • the processor 210 can: construct a first population comprising N individuals, wherein each individual comprises a model parameter value of a support vector regression machine prediction model, wherein N is greater than or equal to 1; genetic algorithm is performed on the first population Iterating, for each genetic iteration: determining an individual selected as a cross-over individual in the first population, and determining an inter-child of the intersecting individual according to a tabu search algorithm to generate a second population; The two populations are subjected to mutation iteration, and each iteration is iterative: a mutation operation is performed on each individual in the second population to obtain a third population; and an optimal individual is determined from the third population according to the tabu search algorithm; The first population of the next genetic algorithm iteration is determined among the M optimal individuals of the iterative generation, wherein M is greater than or equal to 1, less than or equal to the number of termination iterations of the mutation algorithm; and the most selected from the first population of Y genetic iterations
  • the optimal solution is a model parameter of the support vector regression machine
  • the processor 210 is configured to: determine a first fitness function value of each of the N individuals in the first population, wherein the first fitness function value is to use the individual as the Determining a difference between a predicted value obtained by predicting a model parameter of the model and a true value; determining a crossover probability between any two individuals in the first population, and determining a random number of the two individuals, Whether the two individuals are intersecting individuals is determined based on the random numbers of the two individuals and the crossover probability.
  • the processor 210 is capable of determining a crossover probability between the two individuals according to the following formula:
  • p c is the crossover probability between any two of the individuals
  • p cmax is the maximum crossover probability
  • p cmin minimum crossover probability indicates whether or not to be used for the crossover probability.
  • urien number of iterations for the GA current indicates whether or not to be used for the crossover probability.
  • max gen is the maximum number of iterations
  • F avg is An average of first fitness function values for all individuals in the first population, the F' being a smaller first fitness function value of the two individuals.
  • the processor 210 can: determine a child generation generated by the intersecting individual; determine a second fitness function value of the child individual; and according to the first fitness function of each of the intersecting individuals And the value and the second fitness function value of the child individual, updating the first population to obtain a second population.
  • the processor 210 is configured to: determine the first fitness function average value of the all intersecting individuals, wherein the first fitness function average value is based on the first fitness of the all intersecting individuals Determining a function value; if it is determined that the second fitness function value of the child individual is greater than the first fitness function average value, replacing the child generation individual with the child generation individual in the first population Crossing the individual to obtain the second population; or the second fitness function value of the child individual is not greater than the average of the first fitness function, and the child is not in the tabu search list, then the child The individual replaces the intersecting individual corresponding to the progeny individual in the first population to obtain a second population.
  • the processor 210 is capable of: determining a third fitness function value of each individual in the second population; determining a mutation probability of each individual in the second population, and determining a random number of each individual Determining whether the individual is a variant individual according to a mutation probability of each individual and the mutation probability; performing a mutation operation on the individual determined to be the variant individual, and determining a mutation operation result, and determining a third according to the mutation operation result Population.
  • the processor 210 is configured to: determine a fourth fitness function value of each individual in the third population and determine a maximum fourth fitness function value; if the maximum fourth fitness function value is greater than If the expected level value is contraindicated, the individual corresponding to the maximum fourth fitness function value is taken as the optimal individual, wherein the tabu expected level value is based on the third fitness function of each individual in the second population. The value is determined; or, if the maximum fourth fitness function value is not greater than the contraindication expectation level value, and the individual corresponding to the maximum fourth fitness function value is not in the taboo table, then the maximum is The individual corresponding to the fourth fitness function value is the optimal individual.
  • the processor 210 is further configured to: acquire a real value in a first set time and a predicted value obtained according to a support vector regression machine prediction model, and use the real value and the The predicted value is divided into a plurality of sections according to the set step size, and for each section, the adjustment factor corresponding to the section is determined according to the true value corresponding to the section and the predicted value; and the second set time is acquired.
  • the first predicted value obtained by the prediction according to the support vector regression machine prediction model, and the interval corresponding to the predicted value is determined, and the adjustment factor corresponding to the interval is obtained according to the interval; according to the second set time
  • a predicted value and the adjustment factor determine a second predicted value.
  • the electronic device of the embodiment of the present application exists in various forms, including but not limited to:
  • Mobile communication devices These devices are characterized by mobile communication functions and are mainly aimed at providing voice and data communication.
  • Such terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones.
  • Ultra-mobile personal computer equipment This type of equipment belongs to the category of personal computers, has computing and processing functions, and generally has mobile Internet access.
  • Such terminals include: PDAs, MIDs, and UMPC devices, such as the iPad.
  • Portable entertainment devices These devices can display and play multimedia content. Such devices include: audio, video players (such as iPod), handheld game consoles, e-books, and smart toys and portable car navigation devices.
  • the tabu search algorithm is used in the process of cross-calculation in the genetic algorithm and the process of mutation calculation to generate a new population, the diversity of the population is increased compared with the prior art. To overcome the problem of insufficient mountain climbing ability in genetic algorithm, to avoid premature convergence to local optimal solution.
  • the present application provides a non-transitory computer storage medium storing computer-executable instructions for causing the computer to perform the above A data prediction method in any of the embodiments.
  • the present application provides a computer program product comprising a computing program stored on a non-transitory computer readable storage medium, the computer program comprising the computer executable instructions
  • the computer executable instructions When executed by a computer, the computer is caused to perform the data prediction method of any of the above embodiments.
  • the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, ie may be located A place, or it can be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment. Those of ordinary skill in the art can understand and implement without deliberate labor.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Physiology (AREA)
  • Genetics & Genomics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种数据预测方法及装置,方法包括:构建包括N个个体的第一种群,其中每个个体包括支持向量回归机预测模型的模型参数值,其中N大于等于1(101);对第一种群进行遗传算法迭代,针对每一次遗传迭代:确定在第一种群中被选择作为交叉个体的个体,并将交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群(102);对第二种群进行变异迭代,针对每一次变异代迭代:对第二种群中每个个体进行变异操作,得到第三种群;根据禁忌搜索算法从第三种群中确定最优个体;从M次变异代迭代的M个最优个体中确定下一次遗传算法迭代的第一种群(103);从Y次遗传算法迭代的第一种群中选择最优解作为支持向量回归机预测模型的模型参数(104)。上述方法增加了种群的多样性,克服遗传算法中爬山能力不足的问题,避免过早收敛到局部最优解。

Description

一种数据预测方法及装置
本申请要求在2017年1月25日提交中国专利局、申请号为201710061325.4、发明名称为“一种数据预测方法及装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及数据处理技术领域,尤其涉及一种数据预测方法及装置。
背景技术
支持向量回归机是根据统计学习理论提出的一种新的学习方法。其具有理论完备、适应性强、全局优化、训练时间短、泛化性能好等诸多优点,已经成为目前国内外研究的热点。支持向量回归机通过结果风险最小化准则和核函数方法,较好地解决了模式分类器复杂性核推广性之间的矛盾,引起了模式识别领域学者的极大关注。从此迅速的发展起来,现在已经在许多领域(生物信息学,文本分类、手写体识别、人脸检测等)都取得了成功的应用,并且在研究过程中,取得了与传统方法可比或更好的结果,还丰富了自身的内容(如快速训练算法等),从而更加推动了它在其他模式识别领域的应用。由于其出色的学习性能,该技术已成为当前国际机器学习界的研究热点,在模式识别、回归估计、函数逼近等领域有了广泛的应用。
现有技术中,通常可以利用各种优化算法来确定支持向量回归机的模型参数,例如粒子群算法、遗传算法或者禁忌搜索算法等方法来确定模型参数,但是这些现有的方法中总是存在占用空间大,过早收敛,爬山能力不足等问题。
发明内容
本申请提供一种数据预测方法及装置,用于解决现有技术中存在占用空间大,过早收敛,爬山能力不足等问题的问题。
本申请实施例提供一种数据预测方法,所述方法包括:
构建包括N个个体的第一种群,其中每个个体包括支持向量回归机预测模型的模型参数值,其中N大于等于1;
对所述第一种群进行遗传算法迭代,针对每一次遗传迭代:确定在所述第一种群中被选择作为交叉个体的个体,并将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群;
对所述第二种群进行变异迭代,针对每一次变异代迭代:对所述第二种群中每个个体进行变异操作,得到第三种群;根据禁忌搜索算法从所述第三种群中确定最优个体;从M次变异代迭代的M个最优个体中确定下一次遗传算法迭代的第一种群;
从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数。
本申请实施例中,由于在遗传算法中的交叉计算的过程中和变异计算的过程中使用了禁忌搜索算法来产生新的种群,与现有技术相比,增加了种群的多样性,更好地克服遗传算法中爬山能力不足的问题,避免过早收敛到局部最优解。
进一步地,所述确定在所述第一种群中被选择作为交叉个体的个体,包括:
确定所述第一种群中N个个体中的每个个体的第一适应度函数值,其中所述第一适应度函数值是将所述个体作为所述支持向量回归机预测模型的模型参数后得到的预测值与真实值之差确定的;
确定所述第一种群中任意两个个体之间的交叉概率,并确定所述两个个体的随机数,根据所述两个个体的随机数以及所述交叉概率确定所述两个个体是否为交叉个体。
本申请实施例中,根据两个个体的随机数以及交叉概率确定所述两个个体是否为交叉个体,而不是采用现有技术中设置确定交叉概率来确定个体是否为交叉个体,增加了种群中个体的多样性。
进一步地,根据下列公式确定所述任意两个个体之间的交叉概率:
Figure PCTCN2017117003-appb-000001
其中,pc为所述任意两个个体之间的交叉概率,pcmax为最大交叉概率,pcmin为最小交叉概率,curgen为当前遗传算法的迭代次数,max gen为最大迭代次数,Favg为所述第一种群中所有个体的第一适应度函数值的平均值,所述F′为所述两个个体中较小的第一适应度函数值。
本申请实施例中,交叉概率是与两个个体的第一适应度函数值相关,并与当前迭代次数有关,而不是采用现有技术的固定概率,增加了种群的多样性,避免过早收敛。
进一步地,所述将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群,包括:
确定所述交叉个体产生的子代个体;
确定所述子代个体的第二适应度函数值;
根据每个所述交叉个体的所述第一适应度函数值和所述子代个体的第二适应度函数值,更新所述第一种群得到第二种群。
本申请实施例中,在确定交叉个体后,确定交叉后的子代个体,并且根据子代个体来更新第一种群,得到交叉后的第二种群。
进一步地,所述根据所述交叉个体的个体的第一适应度函数值和所述子代个体的第二适应度函数值,更新所述第一种群得到第二种群,包括:
确定所述所有交叉个体的所述的第一适应度函数平均值,其中所述第一适应度函数平均值是根据所述所有交叉个体的第一适应度函数值确定的;若确定所述子代个体的第二适应度函数值大于所述第一适应度函数平均值,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群;或
所述子代个体的第二适应度函数值不大于第一适应度函数平均值,且所 述子代个体不在禁忌搜索列表中,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群。
本申请实施例中,不仅接受了第二适应度函数值大于所述第一适应度函数平均值的子代个体的,也接受了第二适应度函数值不大于第一适应度函数平均值,且所述子代个体不在禁忌搜索列表中的个体,与现有技术相比,增加了种群的多样性,避免了过早收敛。
进一步地,所述对所述第二种群中每个个体进行变异操作,得到第三种群,包括:
确定所述第二种群中每个个体的第三适应度函数值;
确定所述第二种群中每个个体的变异概率,并确定每一个个体的随机数,根据每个个体的变异概率以及所述变异概率确定所述个体是否为变异个体;
对确定为所述变异个体的个体进行变异操作,并确定变异操作结果,根据所述变异操作结果确定第三种群。
本申请实施例中,第三种群是通过禁忌搜索算法来确定的,与现有技术相比,增加了种群的多样性,避免了过早收敛。
进一步地,所述根据禁忌搜索算法从所述第三种群中确定最优个体,包括:
确定所述第三种群中每个个体的第四适应度函数值并确定最大的第四适应度函数值;
若所述最大的第四适应度函数值大于禁忌期望水平值,则将所述最大的第四适应度函数值对应的个体作为最优个体,其中,所述禁忌期望水平值是根据所述第二种群中每个个体的第三适应度函数值确定的;
或,若所述最大的第四适应度函数值不大于所述禁忌期望水平值,且所述最大的第四适应度函数值对应的个体不在禁忌表中,则将所述最大的第四适应度函数值对应的个体作为最优个体。
本申请实施例中,与现有技术相比,除了接受了大于禁忌期望水平值的个体,也接受了不大于所述禁忌期望水平值,但最大的第四适应度函数值对 应的个体不在禁忌表中的个体,增加了种群的多样性,避免了过早收敛。
进一步地,所述从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数后,还包括:
获取设定时间内的真实值,并将所述真实值按照设定步长划分为多个区间,确定不同区间的调整因子,其中每个区间对应一个时间因子;
获取根据支持向量回归机预测模型进行预测得到的第一预测值,并确定所述第一预测值对应的时间因子,根据所述时间因子获取调整因子;
根据所述预测值以及所述调整因子确定第二预测值。
本申请实施例中,通过增加调整因子,能够让模型对周期性数据具有较好的预测效果。
本申请还提供一种数据预测装置,包括:
第一种群确定单元,用于构建包括N个个体的第一种群,其中每个个体包括支持向量回归机预测模型的模型参数值,其中N大于等于1;
第二种群确定单元,用于对所述第一种群进行遗传算法迭代,针对每一次遗传迭代:确定在所述第一种群中被选择作为交叉个体的个体,并将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群;
第二种群变异单元,用于对所述第二种群进行变异迭代,针对每一次变异代迭代:对所述第二种群中每个个体进行变异操作,得到第三种群;根据禁忌搜索算法从所述第三种群中确定最优个体;从M次变异代迭代的M个最优个体中确定下一次遗传算法迭代的第一种群;
模型参数确定单元,用于从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数。
本申请实施例中,由于在遗传算法中的交叉计算的过程中和变异计算的过程中使用了禁忌搜索算法来产生新的种群,与现有技术相比,增加了种群的多样性,更好地克服遗传算法中爬山能力不足的问题,避免过早收敛到局部最优解。
进一步地,所述第二种群确定单元具体用于:
确定所述第一种群中N个个体中的每个个体的第一适应度函数值,其中所述第一适应度函数值是将所述个体作为所述支持向量回归机预测模型的模型参数后得到的预测值与真实值之差确定的;
确定所述第一种群中任意两个个体之间的交叉概率,并确定所述两个个体的随机数,根据所述两个个体的随机数以及所述交叉概率确定所述两个个体是否为交叉个体。
进一步地,根据下列公式确定所述任意两个个体之间的交叉概率:
Figure PCTCN2017117003-appb-000002
其中,pc为所述任意两个个体之间的交叉概率,pcmax为最大交叉概率,pcmin为最小交叉概率,curgen为当前遗传算法的迭代次数,max gen为最大迭代次数,Favg为所述第一种群中所有个体的第一适应度函数值的平均值,所述F′为所述两个个体中较小的第一适应度函数值。
进一步地,所述第二种群确定单元具体用于:
确定所述交叉个体产生的子代个体;
确定所述子代个体的第二适应度函数值;
根据每个所述交叉个体的所述第一适应度函数值和所述子代个体的第二适应度函数值,更新所述第一种群得到第二种群。
进一步地,所述第二种群确定单元具体用于:
确定所述所有交叉个体的所述的第一适应度函数平均值,其中所述第一适应度函数平均值是根据所述所有交叉个体的第一适应度函数值确定的;若确定所述子代个体的第二适应度函数值大于所述第一适应度函数平均值,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群;或
所述子代个体的第二适应度函数值不大于第一适应度函数平均值,且所述子代个体不在禁忌搜索列表中,则将所述子代个体替换所述第一种群中所 述子代个体对应的交叉个体,得到第二种群。
进一步地,所述第二种群变异单元具体用于:
确定所述第二种群中每个个体的第三适应度函数值;
确定所述第二种群中每个个体的变异概率,并确定每一个个体的随机数,根据每个个体的变异概率以及所述变异概率确定所述个体是否为变异个体;
对确定为所述变异个体的个体进行变异操作,并确定变异操作结果,根据所述变异操作结果确定第三种群。
进一步地,所述第二种群变异单元具体用于:
确定所述第三种群中每个个体的第四适应度函数值并确定最大的第四适应度函数值;
若所述最大的第四适应度函数值大于禁忌期望水平值,则将所述最大的第四适应度函数值对应的个体作为最优个体,其中,所述禁忌期望水平值是根据所述第二种群中每个个体的第三适应度函数值确定的;
或,若所述最大的第四适应度函数值不大于所述禁忌期望水平值,且所述最大的第四适应度函数值对应的个体不在禁忌表中,则将所述最大的第四适应度函数值对应的个体作为最优个体。
进一步地,所述装置还包括:
调整单元,用于获取设定时间内的真实值,并将所述真实值按照设定步长划分为多个区间,确定不同区间的调整因子,其中每个区间对应一个时间因子;
获取根据支持向量回归机预测模型进行预测得到的第一预测值,并确定所述第一预测值对应的时间因子,根据所述时间因子获取调整因子;
根据所述预测值以及所述调整因子确定第二预测值。
本申请实施例还提供了一种电子设备,包括:至少一个处理器;以及与所述至少一个处理器通信连接的存储器;其中,所述存储器存储有可被所述至少一个处理器执行的指令,所述指令被所述至少一个处理器执行,以使所述至少一个处理器能够执行本申请实施例的数据预测方法。
本申请实施例还提供了一种非易失性计算机存储介质,所述非暂态计算机可读存储介质存储有计算机可执行指令,所述计算机可执行指令用于使所述计算机执行本申请实施例的数据预测方法。
本申请实施例还提供了一种计算机程序产品,所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算程序,所述计算机程序包括所述计算机可执行指令,当所述计算机可执行指令被计算机执行时,使所述计算机执行本申请实施例的数据预测方法。
本申请实施例中,由于在遗传算法中的交叉计算的过程中和变异计算的过程中使用了禁忌搜索算法来产生新的种群,与现有技术相比,增加了种群的多样性,更好地克服遗传算法中爬山能力不足的问题,避免过早收敛到局部最优解。
附图说明
图1为本申请实施例提供的一种数据预测方法的流程示意图;
图2为本申请实施例提供的一种数据预测装置的结构示意图;
图3为本申请实施例提供的一种电子设备的结构示意图。
具体实施方式
为了使本申请的目的、技术方案和优点更加清楚,下面将结合附图对本申请作进一步地详细描述,显然,所描述的实施例仅仅是本申请一部份实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例,都属于本申请保护的范围。
本申请提供一种数据预测方法,该方法由电子设备执行,如图1所示,所述方法包括:
步骤101,构建包括N个个体的第一种群,其中每个个体包括支持向量回归机预测模型的模型参数值,其中N大于等于1;
步骤102,对所述第一种群进行遗传算法迭代,针对每一次遗传迭代:确 定在所述第一种群中被选择作为交叉个体的个体,并将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群;
步骤103,对所述第二种群进行变异迭代,针对每一次变异代迭代:对所述第二种群中每个个体进行变异操作,得到第三种群;根据禁忌搜索算法从所述第三种群中确定最优个体;从M次变异代迭代的M个最优个体中确定下一次遗传算法迭代的第一种群;
步骤104,从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数。
在步骤101中,支持向量回归机SVR模型把预测函数设置为f(x)=wφ(x)+b,然后求得函数的参数,从而获得预测结果。本申请实施例中,可以设置支持向量回归机预测模型为函数f(x)=wφ(x)+b,具体来说,支持向量回归机预测模型的建立可以描述为:
设有一数据集为S=(xi,yi),i=1,2,…,n,xi∈Rp,yi∈R,xi是输入数据,为p维向量,yi为实际输出数据。通过一个非线性变换φ,将数据集中的x映射到高维的特征空间F中,并且在特征空间F中寻找一线性函数f(x),使得f(x)能够很好地在特征空间F中对数据集S进行线性回归。设:
f(x)=wφ(x)+b
(1)其中,w为回归系数向量,或称为权向量,b为回归常数,或称分类阈值。通过结构风险最小化的原则,在取损失函数为ε不敏感损失函数的前提下,求解回归函数f(x)即为求解如下优化问题:
Figure PCTCN2017117003-appb-000003
求解二次规划问题(2),有
Figure PCTCN2017117003-appb-000004
其中,
Figure PCTCN2017117003-appb-000005
也就是说,在本申请实施例中,求解支持向量回归机预测模型的问题就转换为求解支持向量回归机预测模型中的模型参数(C,σ2,ε)最优解的问题。
在步骤101中,构建包括N个个体的第一种群,其中每一个个体都包括上述参数模型中的三个参数值,例如,可选的,假设三个模型参数的值的范围都是0~7之间的整数,则每个个体都可以用3位无符号二进制整数来表示,例如,在第一种群中的一个个体为101110101,表示的是三个参数值分别为5,6,5。
在步骤101前,还包括构建初始种群的步骤,其中初始种群的个体数量大于等于第一种群中个体数量,第一种群是通过初始种群选择操作后确定的。
在本申请实施例中,例如,初始种群中包括6个个体,分别为个体A、个体B、个体C、个体D、个体E以及个体F,在进行遗传算法的选择操作之前,首先需要确定是否到达遗传算法的终止条件。可选的,终止条件为最大迭代数目和最小误差限,最大迭代数目可以设置为1000,最小误差限可以设置为10-8。
若确定没有到达终止条件,则首先确定初始种群中六个个体的适应度函数值,可选的,在本申请实施例中,可以通过下列公式来确定每个个体的适应度函数值:
Figure PCTCN2017117003-appb-000006
也就是说,每个个体的适应度函数值是根据将每个个体表示的模型值代入SVR模型后,与真实值之间的差来确定的。
在本申请实施例中,差值越小,适应度函数值越大。
可选的,在本申请实施例中,还有其它确定适应度函数值的方法,由于方法众多,得到的结果也不同,可以统一设置越接近真实值的个体的适应度函数值越大。
可选的,在本申请实施例中,选择策略为正比例选择策略。在正比例选择策略中,对于每个个体i,假定适应度的函数值为Fi,则个体i的选择概率可以表示为:
Figure PCTCN2017117003-appb-000007
假设,在本申请实施例中,六个个体的适应度函数值如表1所示:
表1:个体的适应度函数值表
Figure PCTCN2017117003-appb-000008
如表1中所示,确定了初始种群中每个个体的适应度函数值后,确定选择操作的个体,也就是说,需要确定第一种群中的个体,即将不被选择的个体淘汰。
可选的,可以通过轮盘赌方法进行选择操作,确定哪个个体最终被选择。
例如,在本申请实施例中,选择过程为4次,分别选取了个体A、个体B、个体C以及个体D,则第一种群中就为这4个个体。
在步骤102中,需要对第一种群进行交叉算法的操作。在本申请实施例中,首先需要确定第一种群中N个个体中的每个个体的第一适应度函数值,根据上述确定第一种群的方法,可以确定第一种群中的每个个体的第一适应度函数值为已确定的。
然后确定第一种群中任意两个个体之间的交叉概率,并确定两个个体的 随机数,根据两个个体的随机数以及交叉概率确定两个个体是否为交叉个体。
可选的,在本申请实施例中,可以根据下列公式确定所述任意两个个体之间的交叉概率:
Figure PCTCN2017117003-appb-000009
其中,pc为所述任意两个个体之间的交叉概率,pcmax为最大交叉概率,pcmin为最小交叉概率,curgen为当前遗传算法的迭代次数,max gen为最大迭代次数,Favg为所述第一种群中所有个体的第一适应度函数值的平均值,所述F′为所述两个个体中较小的第一适应度函数值。
例如,在本申请实施例中,假设第一种群为[A,B,C,D]4个个体,随机确定第一种群中两两个体之间的交叉概率,例如第一种群中的第一适应度函数值如表1所示,假设随机确定个体A以及个体B之间的交叉概率,首先根据两个个体的第一适应度函数值确定个体A的第一适应度函数值较低,则F′为个体A的适应度函数值,Favg为四个个体的平均第一适应度函数值,可选的,为23.75,个体A与个体B之间的交叉概率为pcmax
可选的,在本申请实施例中,为了使算法具有良好的全局搜索能力,并避免陷入局部最优,我们令pcmax=0.9,pcmin=0.1。
在步骤102中,在确定第一种群中任意两个个体之间的交叉概率后,还需要确定两个个体的随机数,根据所述两个个体的随机数以及所述交叉概率确定所述两个个体是否为交叉个体。
具体的,可以通过对两个个体随机产生一个0到1之间的随机数RN,通过RN以及确定的两个个体的交叉概率来确定两个个体是否能够被选择作为交叉算法的父代个体,也就是交叉个体。
可选的,在本申请实施例中,如果RN≤pc,则该染色体被选为父代染色体。
例如,在本申请实施例中,两个个体分别为个体A以及个体B,个体A 以及个体B的交叉概率为0.9,对个体对A和B产生的随机数为0.6,也就说个体对A和B产生的随机数小于个体A以及个体B的交叉概率,所以个体A以及个体B确定为交叉个体。
在步骤102中,确定了第一种群中的交叉个体后,进行交叉算法,确定交叉个体交叉后产生的交叉子代,并确定交叉个体产生的子代个体以及确定子代个体的第二适应度函数值;根据每个交叉个体的第一适应度函数值和子代个体的第二适应度函数值,更新第一种群得到第二种群。
可选的,在本申请实施例中,根据交叉个体的个体的第一适应度函数值和子代个体的第二适应度函数值,更新第一种群得到第二种群,具体的过程如下:
确定所有交叉个体的的第一适应度函数平均值,其中第一适应度函数平均值是根据所有交叉个体的第一适应度函数值确定的;若确定子代个体的第二适应度函数值大于第一适应度函数平均值,则将子代个体替换第一种群中子代个体对应的交叉个体,得到第二种群;或
子代个体的第二适应度函数值不大于第一适应度函数平均值,且子代个体不在禁忌搜索列表中,则将子代个体替换第一种群中子代个体对应的交叉个体,得到第二种群。
可选的,在本申请实施例中,可以通过下列步骤确定第二种群:
步骤201:设置父代交叉个体的第一适应度函数值的平均值为禁忌期望水平值,以子代个体的适应度值为标准判断是否被禁忌。如果子代个体的适应度函数值比禁忌期望水平值更优,则该子代个体被接受,进入下一代;若子代个体的适应度函数值比禁忌期望水平值要差,则转入步骤202。
步骤202:如果该子代个体不在禁忌列表中,则该子代个体被接受,进入下一代。如果该子代个体在禁忌列表中,则选择交叉个体中适应度值较好的一个进入下一代。
例如,在本申请实施例中,第一种群中被选择的交叉个体为A、B、C、D四个个体,且交叉方式为A、B个体作为交叉个体,C、D个体作为交叉个 体,分别进行交叉算法,分别得到交叉子代a、b以及c、d。
可选的,在本申请实施例中,可以选择单点交叉算法、多点交叉算法来确定交叉个体。
通过公式(4)确定每个交叉子代的第二适应度函数值,假设确定的a的第二适应度函数值为25,b的适应度函数值为20,c的适应度函数值为21,d的适应度函数值为26,交叉个体的第一适应度函数值的平均值为23.75,将23.75作为禁忌期望水平。
个体a以及个体d的第二适应度函数值优于禁忌期望水平,则接受个体a以及个体b作为子代个体,个体b以及个体c的第二适应度函数值低于禁忌期望水平,则还需要确定个体b以及个体c是否在禁忌列表中。
在本申请实施例中,若个体b以及个体c在禁忌列表中,就表明在之前的遗传算法过程中,曾将个体b以及个体c作为最优解并接受了,所以若是个体b以及个体c在禁忌列表中,则不接受两个子代个体,接受个体B以及个体C的第一适应度值较好的个体。
例如,在本申请实施例中,个体b不在禁忌列表中,则接收个体b,个体c在禁忌列表中,则选择接收个体c的父代(个体C和个体D中)第一适应度值较好的个体。例如,第一适应度值较好的个体为个体C,则第二种群为个体a,个体b,个体C以及个体d。
在本申请实施例中,由于在接受子代个体时,考虑了子代个体的第二适应度函数值不大于第一适应度函数平均值,且子代个体不在禁忌搜索列表中,则将子代个体替换第一种群中子代个体对应的交叉个体的情况,所以与现有技术相比,得到第二种群中个体的多样性更加丰富。
在步骤103中,在得到第二种群后,需要对第二种群进行变异操作,首先确定第二种群中每个个体的第三适应度函数值;然后确定第二种群中每个个体的变异概率,并确定每一个个体的随机数,根据每个个体的变异概率以及变异概率确定个体是否为变异个体;对确定为变异个体的个体进行变异操作,并确定变异操作结果,根据变异操作结果确定第三种群。
可选的,在本申请实施例中,通过确定第二种群即可确定第二种群中每个个体的第三适应度函数值,例如,第二种群为个体a,个体b,个体C以及个体d,每个个体的第三适应度函数值为已经确定过的。
在本申请实施例中,确定变异个体的方法与确定交叉个体的方法类似,具体为,可以根据下列公式确定第二种群中每个的变异概率:
Figure PCTCN2017117003-appb-000010
其中,pm表示为变异概率,pmmax表示最大变异概率,pmmin表示最小变异概率,F表示个体的适应度函数值。为了使得算法在初始迭代的过程中染色体能够有足够的多样性,并且在算法进行的后期提升局部搜索能力,可选的,设置pmmax=0.4,pmmin=0.01。
在确定第二种群中每个个体的变异概率后,可以通过对每个个体随机产生一个0到1之间的随机数RM,通过RM以及确定的每个个体的交叉概率来确定每个个体是否能够被选择作为变异个体。
例如,在本申请实施例中,确定的个体a的变异概率为0.3,对个体a产生的随机数为0.1,则确定个体a为变异个体。
可选的,在本申请实施中,第二种群中有的个体可以发生变异,有的个体没有发生变异,例如,在本申请实施例中,第二种群发生变异后的结果为个体a变异为个体aa,其它第二种群中的个体未发生变异,则新生成的第三种群为个体aa,个体b,个体C以及个体d。
在第三种群中确定最优解前,还需要确定第三种群中变异的次数是否大于了设置的迭代终止条件,例如,可以设置迭代终止条件为1000次,即若已经变异了1000次,在停止变异,否则确定最优解。
可选的,在本申请实施例中,根据禁忌搜索算法从第三种群中确定最优个体,具体方法如下:
确定第三种群中每个个体的第四适应度函数值并确定最大的第四适应度 函数值;若最大的第四适应度函数值大于禁忌期望水平值,则将最大的第四适应度函数值对应的个体作为最优个体,其中,禁忌期望水平值是根据第二种群中每个个体的第三适应度函数值确定的;或,若最大的第四适应度函数值不大于禁忌期望水平值,且最大的第四适应度函数值对应的个体不在禁忌表中,则将最大的第四适应度函数值对应的个体作为最优个体。
具体步骤如下:
步骤301,判断最优的候选解是否达到禁忌期望水平。如果达到禁忌期望水平,则更新禁忌期望水平值,并用将当前解更新为该候选解,并进入下一次变异。如果未达到禁忌期望水平,转入步骤302。
在本申请实施例中,禁忌期望水平为第二种群的第三适应度函数值的平均值,若确定的变异子代中的最优解的达到禁忌期望水平,则更新禁忌期望水平值,并将变异子代中的最优解更新为该候选解,且将变异子代中的最优解放入禁忌列表中,然后进入下一次变异过程,最优解指的是在变异子代中,最大的第四适应度函数值对应的个体。
步骤302,在本申请实施例中,若产生的变异子代的最优解未达到禁忌期望水平,则选择候选最优解为当前解,当且仅当该最优解并非在禁忌表中,将变异子代中的最优解放入禁忌列表中,然后进入下一次变异过程。
在本申请实施例中,由于在接受变异个体时,考虑了当前最优的变异个体的第四适应度函数值不大于第三适应度函数平均值,且该个体不在禁忌搜索列表中的情况,所以与现有技术相比,增加了变异产生的解的多样性。
在本申请实施例中,当确定达到变异算法的停止条件后,产生的第三种群作为第一种群重新输入给步骤101,继续完成交叉、变异的过程。
例如,在本申请实施例中,当确定达到变异算法的停止条件时,第三种群为(aaa,bb,c,ddd),则在步骤101中,第一种群中的个体为(aaa,bb,c,ddd)。
在步骤104中,当确定遗传算法满足终止条件后,得到的第一种群中的最优个体,就是需要求解的SVR模型参数的最优解,其中最优解为第一种群 中的个体代入到SVR模型中求解的最优的适应度函数值。
在本申请实施例中,在确定了SVR模型的参数后,为了更准确对数据进行预测,根据数据的时间特性,引入了调整因子的概念。
调整因子可以表示为:
Figure PCTCN2017117003-appb-000011
其中,Si为调整因子,yi表示真实值,
Figure PCTCN2017117003-appb-000012
表示预测值,i=j,j+d,…,j+(m-1)d,0<j≤d,d表示时间序列的步长,m表示时间序列的最大时间长度。可选的,在本申请实施例中,可以根据获取的数据来确定d的长度,例如,d=30,表明计算的调整因子是按照月来更新的等等。
在本申请实施例中,由于真实值以及预测值都是可以获得的,且交易数据可以是多个时间周期构成的,则可以通过下列公式确定多个调整因子的值:
Figure PCTCN2017117003-appb-000013
Figure PCTCN2017117003-appb-000014
也就是说,若存在j个时间周期,每一个时间周期对应的调整因子都是根将多个该时间周期的调整因子求和平均后得到。
在求出每个时间周期对应的调整因子后,预测值会被调整为:
Figure PCTCN2017117003-appb-000015
其中,l=j+(i-1)d,i=1,2,…,0<j≤d,T表示预测值对应的最后一个时间周期点。
在本申请实施例中,以银联数据的交易数据为例,获取设定时间内的交 易数据的真实值,以及设定时间内的预测值。可选的设定时间内为2000年-2010年10年的交易数据,设置d为一个月,则对于每一年有12个交易数据,相对的,对于每一年都有12个预测数据。
按照每一年的每个月,都可以根据公式(8)来确定一个调整因子,也就是说,对于一月份的调整因子有12个,二月份的调整因子也有12个,等等。
将获取的每个月份的12个调整因子求平方和然后除以12,然后再开根号,得到的值就是该月份的调整因子。
例如,需要预测的预测值为2011年1月份的交易数据,首先确定1月份对应的调整因子,即将10年中每个1月份的调整因子求平方和然后除以12然后再开根号得到的值,然后将2011年1月份的预测值乘以1调整因子获得调整后的预测值。
基于同样的构思,本申请实施例还提供一种数据预测装置,如图2所示,包括:
第一种群确定单元S1,用于构建包括N个个体的第一种群,其中每个个体包括支持向量回归机预测模型的模型参数值,其中N大于等于1;
第二种群确定单元S2,用于对所述第一种群进行遗传算法迭代,针对每一次遗传迭代:确定在所述第一种群中被选择作为交叉个体的个体,并将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群;
第二种群变异单元S3,用于对所述第二种群进行变异迭代,针对每一次变异代迭代:对所述第二种群中每个个体进行变异操作,得到第三种群;根据禁忌搜索算法从所述第三种群中确定最优个体;从M次变异代迭代的M个最优个体中确定下一次遗传算法迭代的第一种群,其中M大于等于1,小于等于变异算法的终止迭代次数;
模型参数确定单元S4,用于从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数,其中Y为所述遗传算法的终止迭代次数。
进一步地,所述第二种群确定单元S2具体用于:
确定所述第一种群中N个个体中的每个个体的第一适应度函数值,其中所述第一适应度函数值是将所述个体作为所述支持向量回归机预测模型的模型参数后得到的预测值与真实值之差确定的;
确定所述第一种群中任意两个个体之间的交叉概率,并确定所述两个个体的随机数,根据所述两个个体的随机数以及所述交叉概率确定所述两个个体是否为交叉个体。
进一步地,根据下列公式确定所述任意两个个体之间的交叉概率:
Figure PCTCN2017117003-appb-000016
其中,pc为所述任意两个个体之间的交叉概率,pcmax为最大交叉概率,pcmin为最小交叉概率,curgen为当前遗传算法的迭代次数,max gen为最大迭代次数,Favg为所述第一种群中所有个体的第一适应度函数值的平均值,所述F′为所述两个个体中较小的第一适应度函数值。
进一步地,所述第二种群确定单元S2具体用于:
确定所述交叉个体产生的子代个体;
确定所述子代个体的第二适应度函数值;
根据每个所述交叉个体的所述第一适应度函数值和所述子代个体的第二适应度函数值,更新所述第一种群得到第二种群。
进一步地,所述第二种群确定单元S2具体用于:
确定所述所有交叉个体的所述的第一适应度函数平均值,其中所述第一适应度函数平均值是根据所述所有交叉个体的第一适应度函数值确定的;若确定所述子代个体的第二适应度函数值大于所述第一适应度函数平均值,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群;或
所述子代个体的第二适应度函数值不大于第一适应度函数平均值,且所述子代个体不在禁忌搜索列表中,则将所述子代个体替换所述第一种群中所 述子代个体对应的交叉个体,得到第二种群。
进一步地,所述第二种群变异单元S3具体用于:
确定所述第二种群中每个个体的第三适应度函数值;
确定所述第二种群中每个个体的变异概率,并确定每一个个体的随机数,根据每个个体的变异概率以及所述变异概率确定所述个体是否为变异个体;
对确定为所述变异个体的个体进行变异操作,并确定变异操作结果,根据所述变异操作结果确定第三种群。
进一步地,所述第二种群变异单元S3具体用于:
确定所述第三种群中每个个体的第四适应度函数值并确定最大的第四适应度函数值;
若所述最大的第四适应度函数值大于禁忌期望水平值,则将所述最大的第四适应度函数值对应的个体作为最优个体,其中,所述禁忌期望水平值是根据所述第二种群中每个个体的第三适应度函数值确定的;
或,若所述最大的第四适应度函数值不大于所述禁忌期望水平值,且所述最大的第四适应度函数值对应的个体不在禁忌表中,则将所述最大的第四适应度函数值对应的个体作为最优个体。
进一步地,所述装置还包括:
调整单元S5,用于获取第一设定时间内的真实值以及根据支持向量回归机预测模型进行预测得到的预测值,并将所述真实值以及所述预测值按照设定步长划分为多个区间,针对每个区间,根据所述区间对应的所述真实值以及所述预测值确定所述区间对应的调整因子;获取第二设定时间内的根据支持向量回归机预测模型进行预测得到的第一预测值,并确定所述预测值对应的区间,根据所述区间获取所述区间对应的调整因子;根据所述第二设定时间的第一预测值以及所述调整因子确定第二预测值。
值得一提的是,以上实施方式中所涉及到的各模块均为逻辑模块,在实际应用中,一个逻辑单元可以是一个物理单元,也可以是一个物理单元的一部分,还可以以多个物理单元的组合实现。此外,为了突出本申请的创新部 分,本实施方式中并没有将与解决本申请所提出的技术问题关系不太密切的单元引入,但这并不表明本实施方式中不存在其它的单元。
本申请实施例还提供一种电子设备,如图3所示,包括:至少一个处理器210;收发器200;以及与至少一个处理器210通信连接的存储器220;
一个或多个处理器210以及存储器220,图3中以一个处理器210为例。
电子设备的界面显示方法的电子设备还可以包括:收发器200,输入装置230和输出装置240。
收发器200、处理器210、存储器220、输入装置230和输出装置240可以通过总线或者其他方式连接,图3中以通过总线连接为例。
存储器220作为一种非易失性计算机可读存储介质,可用于存储非易失性软件程序、非易失性计算机可执行程序以及模块,如本申请实施方式中提供的数据预测方法对应的程序指令/模块(例如,附图2所示的第一种群确定单元S1,第二种群确定单元S2,第二种群变异单元S3,模型参数确定单元S4和调整单元S5等)。处理器210通过运行存储在存储器220中的非易失性软件程序、指令以及模块,从而执行服务器的各种功能应用以及数据处理,即实现上述实施方式中的数据预测方法。
存储器220可以包括存储程序区和存储数据区,其中,存储程序区可存储操作系统、至少一个功能所需要的应用程序;存储数据区可存储根据数据预测的处理装置的使用所创建的数据等。此外,存储器220可以包括高速随机存取存储器220,还可以包括非易失性存储器220,例如至少一个磁盘存储器220件、闪存器件、或其他非易失性固态存储器220件。在一些实施例中,存储器220可选包括相对于处理器210远程设置的存储器220,这些远程存储器220可以通过网络连接至数据预测方法的处理装置。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
输入装置230可接收输入的数字或字符信息,以及产生与电子设备的用户设置以及功能控制有关的键信号输入。输出装置240可包括显示屏等显示设备。
一个或者多个模块存储在存储器220中,当被一个或者多个处理器210执行时,执行上述任意方法实施方式中提供的数据预测方法。
具体而言,处理器210能够:构建包括N个个体的第一种群,其中每个个体包括支持向量回归机预测模型的模型参数值,其中N大于等于1;对所述第一种群进行遗传算法迭代,针对每一次遗传迭代:确定在所述第一种群中被选择作为交叉个体的个体,并将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群;对所述第二种群进行变异迭代,针对每一次变异代迭代:对所述第二种群中每个个体进行变异操作,得到第三种群;根据禁忌搜索算法从所述第三种群中确定最优个体;从M次变异代迭代的M个最优个体中确定下一次遗传算法迭代的第一种群,其中M大于等于1,小于等于变异算法的终止迭代次数;从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数,其中Y为所述遗传算法的终止迭代次数。
进一步地,所述处理器210能够:确定所述第一种群中N个个体中的每个个体的第一适应度函数值,其中所述第一适应度函数值是将所述个体作为所述支持向量回归机预测模型的模型参数后得到的预测值与真实值之差确定的;确定所述第一种群中任意两个个体之间的交叉概率,并确定所述两个个体的随机数,根据所述两个个体的随机数以及所述交叉概率确定所述两个个体是否为交叉个体。
进一步地,所述处理器210能够:根据下列公式确定所述任意两个个体之间的交叉概率:
Figure PCTCN2017117003-appb-000017
其中,pc为所述任意两个个体之间的交叉概率,pcmax为最大交叉概率,pcmin为最小交叉概率,curgen为当前遗传算法的迭代次数,max gen为最大迭代次数,Favg为所述第一种群中所有个体的第一适应度函数值的平均值,所述F′为所述 两个个体中较小的第一适应度函数值。
进一步地,所述处理器210能够:确定所述交叉个体产生的子代个体;确定所述子代个体的第二适应度函数值;根据每个所述交叉个体的所述第一适应度函数值和所述子代个体的第二适应度函数值,更新所述第一种群得到第二种群。
进一步地,所述处理器210能够:确定所述所有交叉个体的所述的第一适应度函数平均值,其中所述第一适应度函数平均值是根据所述所有交叉个体的第一适应度函数值确定的;若确定所述子代个体的第二适应度函数值大于所述第一适应度函数平均值,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群;或所述子代个体的第二适应度函数值不大于第一适应度函数平均值,且所述子代个体不在禁忌搜索列表中,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群。
进一步地,所述处理器210能够:确定所述第二种群中每个个体的第三适应度函数值;确定所述第二种群中每个个体的变异概率,并确定每一个个体的随机数,根据每个个体的变异概率以及所述变异概率确定所述个体是否为变异个体;对确定为所述变异个体的个体进行变异操作,并确定变异操作结果,根据所述变异操作结果确定第三种群。
进一步地,所述处理器210能够:确定所述第三种群中每个个体的第四适应度函数值并确定最大的第四适应度函数值;若所述最大的第四适应度函数值大于禁忌期望水平值,则将所述最大的第四适应度函数值对应的个体作为最优个体,其中,所述禁忌期望水平值是根据所述第二种群中每个个体的第三适应度函数值确定的;或,若所述最大的第四适应度函数值不大于所述禁忌期望水平值,且所述最大的第四适应度函数值对应的个体不在禁忌表中,则将所述最大的第四适应度函数值对应的个体作为最优个体。
进一步地,所述处理器210还能够:获取第一设定时间内的真实值以及根据支持向量回归机预测模型进行预测得到的预测值,并将所述真实值以及所 述预测值按照设定步长划分为多个区间,针对每个区间,根据所述区间对应的所述真实值以及所述预测值确定所述区间对应的调整因子;获取第二设定时间内的根据支持向量回归机预测模型进行预测得到的第一预测值,并确定所述预测值对应的区间,根据所述区间获取所述区间对应的调整因子;根据所述第二设定时间的第一预测值以及所述调整因子确定第二预测值。
本申请实施例的电子设备以多种形式存在,包括但不限于:
(1)移动通信设备:这类设备的特点是具备移动通信功能,并且以提供话音、数据通信为主要目标。这类终端包括:智能手机(例如iPhone)、多媒体手机、功能性手机,以及低端手机等。
(2)超移动个人计算机设备:这类设备属于个人计算机的范畴,有计算和处理功能,一般也具备移动上网特性。这类终端包括:PDA、MID和UMPC设备等,例如iPad。
(3)便携式娱乐设备:这类设备可以显示和播放多媒体内容。该类设备包括:音频、视频播放器(例如iPod),掌上游戏机,电子书,以及智能玩具和便携式车载导航设备。
(4)其他具有数据交互功能的电子装置。
本申请实施例中,由于在遗传算法中的交叉计算的过程中和变异计算的过程中使用了禁忌搜索算法来产生新的种群,与现有技术相比,增加了种群的多样性,更好地克服遗传算法中爬山能力不足的问题,避免过早收敛到局部最优解。
基于相同的发明构思,本申请提供一种非易失性计算机存储介质,所述非暂态计算机可读存储介质存储有计算机可执行指令,所述计算机可执行指令用于使所述计算机执行上述任一实施方式中的数据预测方法。
基于相同的发明构思,本申请提供一种计算机程序产品,所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算程序,所述计算机程序包括所述计算机可执行指令,当所述计算机可执行指令被计算机执行时,使所述计算机执行上述任一实施方式中的数据预测方法。
以上所描述的装置实施例仅仅是示意性的,其中所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。本领域普通技术人员在不付出创造性的劳动的情况下,即可以理解并实施。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到各实施方式可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件。基于这样的理解,上述技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品可以存储在计算机可读存储介质中,如ROM/RAM、磁碟、光盘等,包括若干指令用以使得一台计算机装置(可以是个人计算机,服务器,或者网络装置等)执行各个实施例或者实施例的某些部分所述的方法。
最后应说明的是:以上实施例仅用以说明本申请的技术方案,而非对其限制;尽管参照前述实施例对本申请进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围。

Claims (26)

  1. 一种数据预测方法,应用于电子设备,其特征在于,所述方法包括:
    构建包括N个个体的第一种群,其中每个个体包括支持向量回归机预测模型的模型参数值,其中N大于等于1;
    对所述第一种群进行遗传算法迭代,针对每一次遗传迭代:确定在所述第一种群中被选择作为交叉个体的个体,并将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群;
    对所述第二种群进行变异迭代,针对每一次变异代迭代:对所述第二种群中每个个体进行变异操作,得到第三种群;根据禁忌搜索算法从所述第三种群中确定最优个体;从M次变异代迭代的M个最优个体中确定下一次遗传算法迭代的第一种群,其中M大于等于1,小于等于变异算法的终止迭代次数;
    从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数,其中Y为所述遗传算法的终止迭代次数。
  2. 根据权利要求1所述的方法,其特征在于,所述确定在所述第一种群中被选择作为交叉个体的个体,包括:
    确定所述第一种群中N个个体中的每个个体的第一适应度函数值,其中所述第一适应度函数值是将所述个体作为所述支持向量回归机预测模型的模型参数后得到的预测值与真实值之差确定的;
    确定所述第一种群中任意两个个体之间的交叉概率,并确定所述两个个体的随机数,根据所述两个个体的随机数以及所述交叉概率确定所述两个个体是否为交叉个体。
  3. 根据权利要求2所述的方法,其特征在于,根据下列公式确定所述任意两个个体之间的交叉概率:
    Figure PCTCN2017117003-appb-100001
    其中,pc为所述任意两个个体之间的交叉概率,pcmax为最大交叉概率,pcmin 为最小交叉概率,curgen为当前遗传算法的迭代次数,max gen为最大迭代次数,Favg为所述第一种群中所有个体的第一适应度函数值的平均值,所述F′为所述两个个体中较小的第一适应度函数值。
  4. 根据权利要求2所述的方法,其特征在于,所述将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群,包括:
    确定所述交叉个体产生的子代个体;
    确定所述子代个体的第二适应度函数值;
    根据每个所述交叉个体的所述第一适应度函数值和所述子代个体的第二适应度函数值,更新所述第一种群得到第二种群。
  5. 根据权利要求4所述的方法,其特征在于,所述根据所述交叉个体的个体的第一适应度函数值和所述子代个体的第二适应度函数值,更新所述第一种群得到第二种群,包括:
    确定所述所有交叉个体的所述的第一适应度函数平均值,其中所述第一适应度函数平均值是根据所述所有交叉个体的第一适应度函数值确定的;若确定所述子代个体的第二适应度函数值大于所述第一适应度函数平均值,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群;或
    所述子代个体的第二适应度函数值不大于第一适应度函数平均值,且所述子代个体不在禁忌搜索列表中,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群。
  6. 根据权利要求4所述的方法,其特征在于,所述对所述第二种群中每个个体进行变异操作,得到第三种群,包括:
    确定所述第二种群中每个个体的第三适应度函数值;
    确定所述第二种群中每个个体的变异概率,并确定每一个个体的随机数,根据每个个体的变异概率以及所述变异概率确定所述个体是否为变异个体;
    对确定为所述变异个体的个体进行变异操作,并确定变异操作结果,根 据所述变异操作结果确定第三种群。
  7. 根据权利要求6所述的方法,其特征在于,所述根据禁忌搜索算法从所述第三种群中确定最优个体,包括:
    确定所述第三种群中每个个体的第四适应度函数值并确定最大的第四适应度函数值;
    若所述最大的第四适应度函数值大于禁忌期望水平值,则将所述最大的第四适应度函数值对应的个体作为最优个体,其中,所述禁忌期望水平值是根据所述第二种群中每个个体的第三适应度函数值确定的;
    或,若所述最大的第四适应度函数值不大于所述禁忌期望水平值,且所述最大的第四适应度函数值对应的个体不在禁忌表中,则将所述最大的第四适应度函数值对应的个体作为最优个体。
  8. 根据权利要求1~7任一所述的方法,其特征在于,所述从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数后,还包括:
    获取第一设定时间内的真实值以及根据支持向量回归机预测模型进行预测得到的预测值,并将所述真实值以及所述预测值按照设定步长划分为多个区间,针对每个区间,根据所述区间对应的所述真实值以及所述预测值确定所述区间对应的调整因子;
    获取第二设定时间内的根据支持向量回归机预测模型进行预测得到的第一预测值,并确定所述预测值对应的区间,根据所述区间获取所述区间对应的调整因子;
    根据所述第二设定时间的第一预测值以及所述调整因子确定第二预测值。
  9. 一种数据预测装置,其特征在于,包括:
    第一种群确定单元,用于构建包括N个个体的第一种群,其中每个个体包括支持向量回归机预测模型的模型参数值,其中N大于等于1;
    第二种群确定单元,用于对所述第一种群进行遗传算法迭代,针对每一 次遗传迭代:确定在所述第一种群中被选择作为交叉个体的个体,并将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群;
    第二种群变异单元,用于对所述第二种群进行变异迭代,针对每一次变异代迭代:对所述第二种群中每个个体进行变异操作,得到第三种群;根据禁忌搜索算法从所述第三种群中确定最优个体;从M次变异代迭代的M个最优个体中确定下一次遗传算法迭代的第一种群,其中M大于等于1,小于等于变异算法的终止迭代次数;
    模型参数确定单元,用于从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数,其中Y为所述遗传算法的终止迭代次数。
  10. 根据权利要求9所述的装置,其特征在于,所述第二种群确定单元具体用于:
    确定所述第一种群中N个个体中的每个个体的第一适应度函数值,其中所述第一适应度函数值是将所述个体作为所述支持向量回归机预测模型的模型参数后得到的预测值与真实值之差确定的;
    确定所述第一种群中任意两个个体之间的交叉概率,并确定所述两个个体的随机数,根据所述两个个体的随机数以及所述交叉概率确定所述两个个体是否为交叉个体。
  11. 根据权利要求10所述的装置,其特征在于,根据下列公式确定所述任意两个个体之间的交叉概率:
    Figure PCTCN2017117003-appb-100002
    其中,pc为所述任意两个个体之间的交叉概率,pcmax为最大交叉概率,pcmin为最小交叉概率,curgen为当前遗传算法的迭代次数,max gen为最大迭代次数,Favg为所述第一种群中所有个体的第一适应度函数值的平均值,所述F′为所述两个个体中较小的第一适应度函数值。
  12. 根据权利要求10所述的装置,其特征在于,所述第二种群确定单元具体用于:
    确定所述交叉个体产生的子代个体;
    确定所述子代个体的第二适应度函数值;
    根据每个所述交叉个体的所述第一适应度函数值和所述子代个体的第二适应度函数值,更新所述第一种群得到第二种群。
  13. 根据权利要求12所述的装置,其特征在于,所述第二种群确定单元具体用于:
    确定所述所有交叉个体的所述的第一适应度函数平均值,其中所述第一适应度函数平均值是根据所述所有交叉个体的第一适应度函数值确定的;若确定所述子代个体的第二适应度函数值大于所述第一适应度函数平均值,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群;或
    所述子代个体的第二适应度函数值不大于第一适应度函数平均值,且所述子代个体不在禁忌搜索列表中,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群。
  14. 根据权利要求12所述的装置,其特征在于,所述第二种群变异单元具体用于:
    确定所述第二种群中每个个体的第三适应度函数值;
    确定所述第二种群中每个个体的变异概率,并确定每一个个体的随机数,根据每个个体的变异概率以及所述变异概率确定所述个体是否为变异个体;
    对确定为所述变异个体的个体进行变异操作,并确定变异操作结果,根据所述变异操作结果确定第三种群。
  15. 根据权利要求14所述的装置,其特征在于,所述第二种群变异单元具体用于:
    确定所述第三种群中每个个体的第四适应度函数值并确定最大的第四适应度函数值;
    若所述最大的第四适应度函数值大于禁忌期望水平值,则将所述最大的第四适应度函数值对应的个体作为最优个体,其中,所述禁忌期望水平值是根据所述第二种群中每个个体的第三适应度函数值确定的;
    或,若所述最大的第四适应度函数值不大于所述禁忌期望水平值,且所述最大的第四适应度函数值对应的个体不在禁忌表中,则将所述最大的第四适应度函数值对应的个体作为最优个体。
  16. 根据权利要求9~15任一所述的装置,其特征在于,所述装置还包括:
    调整单元,用于获取第一设定时间内的真实值以及根据支持向量回归机预测模型进行预测得到的预测值,并将所述真实值以及所述预测值按照设定步长划分为多个区间,针对每个区间,根据所述区间对应的所述真实值以及所述预测值确定所述区间对应的调整因子;获取第二设定时间内的根据支持向量回归机预测模型进行预测得到的第一预测值,并确定所述预测值对应的区间,根据所述区间获取所述区间对应的调整因子;根据所述第二设定时间的第一预测值以及所述调整因子确定第二预测值。
  17. 一种电子设备,其特征在于,包括:至少一个处理器;以及与所述至少一个处理器通信连接的存储器;
    所述存储器存储有可被所述至少一个处理器执行的指令,所述指令被所述至少一个处理器执行,以使所述至少一个处理器能够:
    构建包括N个个体的第一种群,其中每个个体包括支持向量回归机预测模型的模型参数值,其中N大于等于1;
    对所述第一种群进行遗传算法迭代,针对每一次遗传迭代:确定在所述第一种群中被选择作为交叉个体的个体,并将所述交叉个体的个体根据禁忌搜索算法确定交叉子代,产生第二种群;
    对所述第二种群进行变异迭代,针对每一次变异代迭代:对所述第二种群中每个个体进行变异操作,得到第三种群;根据禁忌搜索算法从所述第三种群中确定最优个体;从M次变异代迭代的M个最优个体中确定下一次遗传算法迭代的第一种群,其中M大于等于1,小于等于变异算法的终止迭代次数;
    从Y次遗传算法迭代的第一种群中选择最优解作为所述支持向量回归机预测模型的模型参数,其中Y为所述遗传算法的终止迭代次数。
  18. 根据权利要求17所述的电子设备,其特征在于,所述处理器能够:
    确定所述第一种群中N个个体中的每个个体的第一适应度函数值,其中所述第一适应度函数值是将所述个体作为所述支持向量回归机预测模型的模型参数后得到的预测值与真实值之差确定的;
    确定所述第一种群中任意两个个体之间的交叉概率,并确定所述两个个体的随机数,根据所述两个个体的随机数以及所述交叉概率确定所述两个个体是否为交叉个体。
  19. 根据权利要求18所述的电子设备,其特征在于,所述处理器能够:
    根据下列公式确定所述任意两个个体之间的交叉概率:
    Figure PCTCN2017117003-appb-100003
    其中,pc为所述任意两个个体之间的交叉概率,pcmax为最大交叉概率,pcmin为最小交叉概率,curgen为当前遗传算法的迭代次数,max gen为最大迭代次数,Favg为所述第一种群中所有个体的第一适应度函数值的平均值,所述F′为所述两个个体中较小的第一适应度函数值。
  20. 根据权利要求18所述的电子设备,其特征在于,所述处理器能够:
    确定所述交叉个体产生的子代个体;
    确定所述子代个体的第二适应度函数值;
    根据每个所述交叉个体的所述第一适应度函数值和所述子代个体的第二适应度函数值,更新所述第一种群得到第二种群。
  21. 根据权利要求20所述的电子设备,其特征在于,所述处理器能够:
    确定所述所有交叉个体的所述的第一适应度函数平均值,其中所述第一适应度函数平均值是根据所述所有交叉个体的第一适应度函数值确定的;若确定所述子代个体的第二适应度函数值大于所述第一适应度函数平均值,则 将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群;或
    所述子代个体的第二适应度函数值不大于第一适应度函数平均值,且所述子代个体不在禁忌搜索列表中,则将所述子代个体替换所述第一种群中所述子代个体对应的交叉个体,得到第二种群。
  22. 根据权利要求20所述的电子设备,其特征在于,所述处理器能够:
    确定所述第二种群中每个个体的第三适应度函数值;
    确定所述第二种群中每个个体的变异概率,并确定每一个个体的随机数,根据每个个体的变异概率以及所述变异概率确定所述个体是否为变异个体;
    对确定为所述变异个体的个体进行变异操作,并确定变异操作结果,根据所述变异操作结果确定第三种群。
  23. 根据权利要求22所述的电子设备,其特征在于,所述处理器能够:
    确定所述第三种群中每个个体的第四适应度函数值并确定最大的第四适应度函数值;
    若所述最大的第四适应度函数值大于禁忌期望水平值,则将所述最大的第四适应度函数值对应的个体作为最优个体,其中,所述禁忌期望水平值是根据所述第二种群中每个个体的第三适应度函数值确定的;
    或,若所述最大的第四适应度函数值不大于所述禁忌期望水平值,且所述最大的第四适应度函数值对应的个体不在禁忌表中,则将所述最大的第四适应度函数值对应的个体作为最优个体。
  24. 根据权利要求17~23任一所述的电子设备,其特征在于,所述处理器还能够:
    获取第一设定时间内的真实值以及根据支持向量回归机预测模型进行预测得到的预测值,并将所述真实值以及所述预测值按照设定步长划分为多个区间,针对每个区间,根据所述区间对应的所述真实值以及所述预测值确定所述区间对应的调整因子;获取第二设定时间内的根据支持向量回归机预测模型进行预测得到的第一预测值,并确定所述预测值对应的区间,根据所述 区间获取所述区间对应的调整因子;根据所述第二设定时间的第一预测值以及所述调整因子确定第二预测值。
  25. 一种非易失性计算机存储介质,其特征在于,所述非暂态计算机可读存储介质存储有计算机可执行指令,所述计算机可执行指令用于使所述计算机执行权利要求1-8任一项所述的数据预测方法。
  26. 一种计算机程序产品,其特征在于,所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算程序,所述计算机程序包括所述计算机可执行指令,当所述计算机可执行指令被计算机执行时,使所述计算机执行权利要求1-8任一项所述的数据预测方法。
PCT/CN2017/117003 2017-01-25 2017-12-18 一种数据预测方法及装置 WO2018137442A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710061325.4 2017-01-25
CN201710061325.4A CN106845627A (zh) 2017-01-25 2017-01-25 一种数据预测方法及装置

Publications (1)

Publication Number Publication Date
WO2018137442A1 true WO2018137442A1 (zh) 2018-08-02

Family

ID=59121466

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/117003 WO2018137442A1 (zh) 2017-01-25 2017-12-18 一种数据预测方法及装置

Country Status (2)

Country Link
CN (1) CN106845627A (zh)
WO (1) WO2018137442A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106845627A (zh) * 2017-01-25 2017-06-13 中国银联股份有限公司 一种数据预测方法及装置
CN110704956B (zh) * 2019-08-09 2022-04-19 太原科技大学 一种冷轧机数据驱动的工艺参数优化方法
CN110866632B (zh) * 2019-10-17 2022-05-06 杭州电子科技大学 一种发电机机组优化的方法
CN111641595B (zh) * 2020-05-11 2021-04-20 湖南大学 一种电力网络安全风险评估方法及系统
CN113806409A (zh) * 2020-05-28 2021-12-17 华为技术有限公司 一种数据配对的方法及其相关设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101018164A (zh) * 2007-02-28 2007-08-15 西南科技大学 一种tcp/ip网络性能评估预测方法
CN101572687A (zh) * 2009-06-05 2009-11-04 北京邮电大学 正交频分复用信号处理方法及系统
CN101807797A (zh) * 2010-04-14 2010-08-18 华中科技大学 一种用于微网的快速故障诊断方法
CN104200263A (zh) * 2014-07-23 2014-12-10 浙江工业大学 一种基于禁忌差分进化和gis的配电网络线路规划方法
CN106845627A (zh) * 2017-01-25 2017-06-13 中国银联股份有限公司 一种数据预测方法及装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101018164A (zh) * 2007-02-28 2007-08-15 西南科技大学 一种tcp/ip网络性能评估预测方法
CN101572687A (zh) * 2009-06-05 2009-11-04 北京邮电大学 正交频分复用信号处理方法及系统
CN101807797A (zh) * 2010-04-14 2010-08-18 华中科技大学 一种用于微网的快速故障诊断方法
CN104200263A (zh) * 2014-07-23 2014-12-10 浙江工业大学 一种基于禁忌差分进化和gis的配电网络线路规划方法
CN106845627A (zh) * 2017-01-25 2017-06-13 中国银联股份有限公司 一种数据预测方法及装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ZHANG, ZHENG: "Research on Data-driven Optimization Learning Method for Short- and Medium-Term Electricity Demand Prediction", WANFANG DATA KNOWLEDGE SERVICE PLATFORM, 5 April 2016 (2016-04-05), pages 46 - 52 *

Also Published As

Publication number Publication date
CN106845627A (zh) 2017-06-13

Similar Documents

Publication Publication Date Title
WO2018137442A1 (zh) 一种数据预测方法及装置
Liu et al. A modified consensus model in group decision making with an allocation of information granularity
WO2020094060A1 (zh) 推荐方法及装置
WO2021155713A1 (zh) 基于权重嫁接的模型融合的人脸识别方法及相关设备
Liu et al. The multiobjective evolutionary algorithm based on determined weight and sub-regional search
CN109840154B (zh) 一种移动云环境下基于任务依赖的计算迁移方法
WO2019120019A1 (zh) 用户性别预测方法、装置、存储介质及电子设备
Li et al. Hybrid optimization algorithm based on chaos, cloud and particle swarm optimization algorithm
Fan et al. Boosted hunting-based fruit fly optimization and advances in real-world problems
CN111814626B (zh) 一种基于自注意力机制的动态手势识别方法和系统
TW201923665A (zh) 風控決策臨界值的確定方法及裝置
CN108829865A (zh) 信息检索方法及装置
WO2019109724A1 (zh) 一种推荐物品的方法和装置
CN106250381A (zh) 表格数据中的用于输入/输出优化的列排序
Yu et al. An iterative locally auto-weighted least squares method for microarray missing value estimation
CN112785005B (zh) 多目标任务的辅助决策方法、装置、计算机设备及介质
CN114492279A (zh) 一种模拟集成电路的参数优化方法及系统
CN110991621A (zh) 一种基于通道数搜索卷积神经网络的方法
WO2023103864A1 (zh) 抵抗联邦学习中歧视传播的节点模型的更新方法
Wu et al. Hyperparameter learning for deep learning-based recommender systems
Zhong et al. Multi-agent simulated annealing algorithm based on differential evolution algorithm
Sun et al. A tent marine predators algorithm with estimation distribution algorithm and Gaussian random walk for continuous optimization problems
Tilahun et al. Vector optimisation using fuzzy preference in evolutionary strategy based firefly algorithm
TWI701629B (zh) 共用產品的推薦方法、裝置以及電子設備
Du et al. Structure tuning method on deep convolutional generative adversarial network with nondominated sorting genetic algorithm II

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17894607

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17894607

Country of ref document: EP

Kind code of ref document: A1