Font Size: a A A

LSTM Enhanced Evolutionary Algorithm For Biased Representation Discrete Optimization Problem

Posted on:2022-04-23Degree:MasterType:Thesis
Country:ChinaCandidate:J R SunFull Text:PDF
GTID:2518306533472464Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
In engineering practice and scientific research,there are various types of discrete optimization problems.Due to the strong discrete constraints of the solution,it is difficult to solve these types of problems.Evolutionary algorithm is a kind of robust global search algorithm,which provides a feasible way for solving discrete optimization problems.In recent years,it has achieved abundant research results.However,evolutionary algorithms often need to perform coding representations of solutions,coding-based evolution,and decoding operations,etc.,which will cause the algorithm operation object to deviate from the real feasible discrete solution,mislead the evolutionary search direction,and reduce the algorithm optimization accuracy.Even infeasible solutions are obtained on discrete problems.At present,this problem has not attracted attention in the research of evolutionary algorithm.In response to this problem,this paper combines the advantages of evolutionary optimization methods and deep learning methods to design a solution,and applies it to multiple actual discrete optimization data sets to verify the effectiveness and practicability of the algorithm.The main content of this article includes the following two parts:(1)Evolutionary optimization algorithm based on LSTM estimating deviation:Based on the biased representation of the discrete optimization problem,firstly,the general expression of the discrete optimization problem considering the deviation is given,which is modeled as an optimization problem with dynamic changes in deviation,and the optimization goal is defined as convergence to the true optimal discrete solution.Secondly,the definition of the intelligent optimization fitness function based on the general expression is given.In order to extract the deviation information in depth,a proxy model based on the long and short-term memory network is designed to predict the deviation direction and change amount,and the dynamic deviation estimated by the model is used to guide the population's evolution direction.Finally,redesign the operating mechanism of the genetic operator in the genetic algorithm,strengthen the role of the long and short-term memory network predictive agent model,and realize the efficient solution of complex discrete optimization problems.The algorithm is applied to eight typical test functions in the field of evolutionary optimization.The experimental results show that the proposed method can improve the adverse effects of data deviation on optimization,and can effectively track the true optimal solution.(2)Discrete personalized search evolutionary optimization algorithm based on DBN surrogate model and LSTM estimating bias: In research(1),the discrete optimization model used to test the performance of the algorithm is discretized from the continuous test function.Without considering that the actual complexity discrete optimization problems often do not have a displayed mathematical form,so they cannot provide an accurate fitness function for genetic algorithms.Because of the existence of deviation in the coding and decoding process of genetic algorithm,the quality of the output solution of the algorithm corresponding to the real solution in the solution space is poor.In view of this,based on the research content(1),a discrete personalized search evolutionary optimization algorithm based on DBN surrogate model and LSTM bias estimation is proposed.First,based on personalized search discrete data,a deep belief network used to establish a model of the relationship between product information and user preferences,use it as fitness function and objective function of genetic algorithms.And the method proposed in the research content(1)is empolyed to solve the model.Secondly,the relationship model between data deviation and user preferences is established,and the method of research content(1)is used again to solve the model,thus forming a two-layer intelligent solution framework.Finally,the position of the real solution is obtained by fusing the two optimization results.The algorithm is applied to four real personalized search data sets.The experimental results show that the proposed algorithm effectively improves the accuracy of personalized search,reduces the negative effects of coding deviation,and improves the user's personalized search experience.Aiming at the personalized search problem with user-generated content,this paper based on the intelligent evolutionary algorithm,fused with deep learning methods,and designed an LSTM enhanced evolutionary algorithm for the biased representation discrete optimization problem.The population evolution of this method is more directional,and can reduce the data deviation,so that the output solution of the algorithm is more suitable for the real optimal solution.The application of complex test functions and actual personalized search discrete data sets verifies the effectiveness of the proposed algorithm.The paper has 16 pictures,9 tables,and 107 references.
Keywords/Search Tags:discrete optimization, evolutionary algorithm, biased representation, long short-term memory network, deep belief network
PDF Full Text Request
Related items