Backward-chaining evolutionary algorithms

Starting from some simple observations on a popular selection method in Evolutionary Algorithms (EAs)—tournament selection—we highlight a previously-unknown source of inefficiency. This leads us to rethink the order in which operations are performed within EAs, and to suggest an algorithm—the EA with efficient macro-selection—that avoids the inefficiencies associated with tournament selection. This algorithm has the same expected behaviour as the standard EA but yields considerable savings in terms of fitness evaluations. Since fitness evaluation typically dominates the resources needed to solve any non-trivial problem, these savings translate into a reduction in computer time. Noting the connection between the algorithm and rule-based systems, we then further modify the order of operations in the EA, effectively turning the evolutionary search into an inference process operating in backward-chaining mode. The resulting backward-chaining EA creates and evaluates individuals recursively, backward from the last generation to the first, using depth-first search and backtracking. It is even more powerful than the EA with efficient macro-selection in that it shares all its benefits, but it also provably finds fitter solutions sooner, i.e., it is a faster algorithm. These algorithms can be applied to any form of population based search, any representation, fitness function, crossover and mutation, provided they use tournament selection. We analyse their behaviour and benefits both theoretically, using Markov chain theory and space/time complexity analysis, and empirically, by performing a variety of experiments with standard and back-ward chaining versions of genetic algorithms and genetic programming.     

Spin-flip symmetry and synchronization

In the context of optimization by evolutionary algorithms (EAs), epistasis, deception, and scaling are well-known examples of problem difficulty characteristics. The presence of one such characteristic in the representation of a search problem indicates a certain type of difficulty the EA is to encounter during its search for globally optimal configurations. In this paper, we claim that the occurrence of symmetry in the representation is another problem difficulty characteristic and discuss one particular form, spin-flip symmetry, characterized by fitness invariant permutations on the alphabet. Its usual effect on unspecialized EAs, premature convergence due to synchronization problems, is discussed in detail. We discuss five different ways to specialize EAs to cope with the symmetry: adapting the genetic operators, changing the fitness function, using a niching technique, using a distributed EA, and attaching a highly redundant genotype-phenotype mapping.     

Parameter control system of evolutionary algorithm that is aided by the entire search history

In solving problems with evolutionary algorithms (EAs), the performance of the EA will be affected by its properties. As the properties of EA depend on the parameter setting, users need to tune the parameters to optimize the performance on different problems. In the case that the user does not have any prior knowledge of the problem, parameter tuning is very difficult and time consuming. One needs to try different combinations of parameter values to find the best setting. To solve this problem, one way is to control the parameters during the EA run. This paper proposes a new adaptive parameter control system, called Parameter Control system using entire Search History (PCSH). It is a general add-on system which is not restricted to a specific class of EA. Users are only required to know the range of the parameters. It automatically adjusts the parameters of an EA according to the entire search history, in a parameter-less manner. To illustrate the performance of PCSH, it is applied to control the parameters of three common classes of EAs: (1) canonical Genetic Algorithm (GA), (2) Particle Swarm Optimization (PSO) and (3) Differential Evolution (DE). For GA, we show that PCSH can automatically control the crossover operator, crossover values (uniformly sampled from the range) and mutation operator. For DE, we show that PCSH can automatically control the crossover operator, crossover values and the differential amplification factor (uniformly sampled from the ranges). For PSO, we show that PCSH can automatically control the two learning factors and the inertia weight (uniformly sampled from the range). Moreover, no special provision is needed at the initialization. 34 benchmark functions are used to evaluate the performance comprehensively. The test results show that, in most of the benchmark functions, the performance of the test EAs are improved or similar after adopting PCSH. It shows that PCSH keeps or improves the performance of the test EAs while relieving the heavy burden of the algorithm designer on the setting of some parameters.     

Surrogate assisted‐hybrid differential evolution algorithm using diversity control

Evolutionary algorithms (EAs) being a major optimization framework, typically require a considerable number of function evaluations to locate an optimal solution for computationally intensive real‐world optimization problems. In order to solve complex multimodal problems within a limited computational budget, surrogate models are integrated with EA. The overall performance of such algorithms not only depends on the construction and integration procedure of the model, but also on the efficiency of the underlying EA in overcoming premature convergence. This can be achieved through diversity control and parameter adaptation methodology in EAs. In this paper, an improved algorithm, namely Diversity Controlled Parameter adapted Differential Evolution with Local Search (DCPaDE‐LS) is integrated into two dynamic surrogate models and two variants, namely Surrogate Assisted Parameter adapted Differential Evolution with Artificial Neural Networks and Response Surface Methodology (SAPDE‐ANN and SAPDE‐RSM) are proposed. They reduce the exact function evaluations for complex, multimodal problems. The performance of the proposed variants are compared on a set of 26 bound‐constrained benchmark functions scalable with 10 and 30 dimensions, with respect to average number of function evaluations (NFE), success rate (SR) and % reduction in NFE in 30 independent trials. The SAPDE variants are compared with Self‐adaptive Differential Evolution, DCPaDE‐LS, Increasing Population Size Covariance Matrix Adaptation Evolution Strategy and Comprehensive Learning Particle Swarm Optimization. The SAPDE variants are able to reduce the NFE without loss in SR for all the functions. The algorithms are also validated using 12 solvable functions from CEC 2005. Of the two variants, SAPDE‐ANN reports reduced NFE in more functions compared with SAPDE‐RSM. Results reveal that the proposed SAPDE algorithm can be applied to real‐world optimization problems.     

Evolutionary algorithms - how to cope with plateaus of constantfitness and when to reject strings of the same fitness

The most simple evolutionary algorithm (EA), the so-called (1 + 1) EA, accepts an offspring if its fitness is at least as large (in the case of maximization) as the fitness of its parent. The variant (1 + 1)* EA only accepts an offspring if its fitness is strictly larger than the fitness of its parent. Here, two functions related to the class of long-path functions are presented such that the (1 + 1) EA maximizes one in polynomial time and needs exponential time for the other while the (1 + 1)* EA has the opposite behavior. These results demonstrate that small changes of an EA may change its behavior significantly. Since the (1 + 1) EA and the (1 + 1)* EA differ only on plateaus of constant fitness, the results also show how EAs behave on such plateaus. The (1 + 1) EA can pass a path of constant fitness and polynomial length in polynomial time. Finally, for these functions, it is shown that local performance measures like the quality gain and the progress rate do not describe the global behavior of EAs     

Bag of local landscape features for fitness landscape analysis

Evolutionary computation is a research field dealing with black-box and complex optimization problems whose fitness landscapes are usually unknown in advance. It is difficult to select an appropriate evolutionary algorithm and parameters for a given problem due to the black-box setting although many evolutionary algorithms have been developed. In this context, several landscape features have been proposed and their usefulness examined for understanding the problem. In this paper, we propose a novel feature vector by focusing on the local landscape in order to characterize the fitness landscape. The proposed landscape features are a vector form and composed of a histogram of quantized local landscape features. We introduce two implementation methods of this concept, called the bag of local landscape patterns (BoLLP) and the bag of evolvability (BoEvo). The BoLLP uses the fitness pattern of the neighbors of a certain candidate solution, and the BoEvo uses the number of better candidate solutions in the neighbors as the local landscape features. Furthermore, the hierarchical versions of the BoLLP and the BoEvo, concatenated feature vectors with different sample sizes, are considered to capture the landscape characteristic with various resolutions. We extract the proposed landscape feature vectors from well-known continuous optimization benchmark functions and the BBOB benchmark function set to investigate their properties; the visualization of the proposed landscape features, clustering and running time prediction experiments are conducted. Then the effectiveness of the proposed landscape features for the fitness landscape analysis is discussed based on the experimental results.     

一种改进的基于差分进化的多目标进化算法

近年来运用进化算法（EAs）解决多目标优化问题（Multi-objective Optimization Problems MOPs）引起了各国学者们的关注。作为一种基于种群的优化方法,EAs提供了一种在一次运行后得到一组优化的解的方法。差分进化（DE）算法是EA的一个分支,最开始是用来解决连续函数空间的问题。提出了一种改进的基于差分进化的多目标进化算法（CDE）,并且将它与另外两个经典的多目标进化算法（MOEAs）NSGA-II和SPEA2进行了对比实验。     

Multiobjective optimization and multiple constraint handling withevolutionary algorithms. I. A unified formulation

In optimization, multiple objectives and constraints cannot be handled independently of the underlying optimizer. Requirements such as continuity and differentiability of the cost surface add yet another conflicting element to the decision process. While “better” solutions should be rated higher than “worse” ones, the resulting cost landscape must also comply with such requirements. Evolutionary algorithms (EAs), which have found application in many areas not amenable to optimization by other methods, possess many characteristics desirable in a multiobjective optimizer, most notably the concerted handling of multiple candidate solutions. However, EAs are essentially unconstrained search techniques which require the assignment of a scalar measure of quality, or fitness, to such candidate solutions. After reviewing current revolutionary approaches to multiobjective and constrained optimization, the paper proposes that fitness assignment be interpreted as, or at least related to, a multicriterion decision process. A suitable decision making framework based on goals and priorities is subsequently formulated in terms of a relational operator, characterized, and shown to encompass a number of simpler decision strategies. Finally, the ranking of an arbitrary number of candidates is considered. The effect of preference changes on the cost surface seen by an EA is illustrated graphically for a simple problem. The paper concludes with the formulation of a multiobjective genetic algorithm based on the proposed decision strategy. Niche formation techniques are used to promote diversity among preferable candidates, and progressive articulation of preferences is shown to be possible as long as the genetic algorithm can recover from abrupt changes in the cost landscape     

Quality of Service constrained routing optimization using Evolutionary Computation

In this work, a novel optimization framework is proposed that allows the improvement of Quality of Service levels in TCP/IP based networks, by configuring the routing weights of link-state protocols such as OSPF. Since this is a NP-hard problem, some algorithms from Evolutionary Computation were considered, working over a mathematical model that allows the definition of flexible cost functions that can take into account several measures of the network behaviour, such as network congestion and end-to-end delays. A number of experiments were performed, over a large set of network topologies, where Evolutionary Algorithms (EAs), Differential Evolution, local search methods and common heuristics were compared. EAs make the most promising alternative leading to solutions with an effective network performance, even under unfavourable scenarios. A number of state of the art multi-objective optimization algorithms were also tested, but the proposed EAs still hold as the most consistent method for network optimization.     

Finding Mount Everest and handling voids

Evolutionary algorithms (EAs) are randomized search heuristics that solve problems successfully in many cases. Their behavior is often described in terms of strategies to find a high location on Earth's surface. Unfortunately, many digital elevation models describing it contain void elements. These are elements not assigned an elevation. Therefore, we design and analyze simple EAs with different strategies to handle such partially defined functions. They are experimentally investigated on a dataset describing the elevation of Earth's surface. The largest value found by an EA within a certain runtime is measured, and the median over a few runs is computed and compared for the different EAs. For the dataset, the distribution of void elements seems to be neither random nor adversarial. They are so-called semirandomly distributed. To deepen our understanding of the behavior of the different EAs, they are theoretically considered on well-known pseudo-Boolean functions transferred to partially defined ones. These modifications are also performed in a semirandom way. The typical runtime until an optimum is found by an EA is analyzed, namely bounded from above and below, and compared for the different EAs. We figure out that for the random model it is a good strategy to assume that a void element has a worse function value than all previous elements. Whereas for the adversary model it is a good strategy to assume that a void element has the best function value of all previous elements.     

Advanced fitness landscape analysis and the performance of memetic algorithms

Memetic algorithms (MAs) have demonstrated very effective in combinatorial optimization. This paper offers explanations as to why this is so by investigating the performance of MAs in terms of efficiency and effectiveness. A special class of MAs is used to discuss efficiency and effectiveness for local search and evolutionary meta-search. It is shown that the efficiency of MAs can be increased drastically with the use of domain knowledge. However, effectiveness highly depends on the structure of the problem. As is well-known, identifying this structure is made easier with the notion of fitness landscapes: the local properties of the fitness landscape strongly influence the effectiveness of the local search while the global properties strongly influence the effectiveness of the evolutionary meta-search. This paper also introduces new techniques for analyzing the fitness landscapes of combinatorial problems; these techniques focus on the investigation of random walks in the fitness landscape starting at locally optimal solutions as well as on the escape from the basins of attractions of current local optima. It is shown for NK-landscapes and landscapes of the unconstrained binary quadratic programming problem (BQP) that a random walk to another local optimum can be used to explain the efficiency of recombination in comparison to mutation. Moreover, the paper shows that other aspects like the size of the basins of attractions of local optima are important for the efficiency of MAs and a local search escape analysis is proposed. These simple analysis techniques have several advantages over previously proposed statistical measures and provide valuable insight into the behaviour of MAs on different kinds of landscapes.     

演化算法求解TSP问题的最新进展

TSP是一个著名的NP-hard问题.对近期出现的一些新的求解TSP问题的演化算法进行了比较全面的综述.其中有一类算法属于郭涛算法及其相应的改进算法,能够得到比传统演化算法更好的解,还有一类采用了实数编码的染色体表示方式,对求解TSP问题的新的染色体表示方式进行了尝试,还有的属于并行演化算法,通过增加并行进程的方式能够在原有算法的基础上得到更好的解.在综述这些算法的同时,还对比了它们的求解能力.最终的目的是希望通过对上述算法的研究,得到更合理的算法,推动演化算法研究TSP问题的进程.     

Multiobjective evolutionary algorithms: a comparative case studyand the strength Pareto approach

Evolutionary algorithms (EAs) are often well-suited for optimization problems involving several, often conflicting objectives. Since 1985, various evolutionary approaches to multiobjective optimization have been developed that are capable of searching for multiple solutions concurrently in a single run. However, the few comparative studies of different methods presented up to now remain mostly qualitative and are often restricted to a few approaches. In this paper, four multiobjective EAs are compared quantitatively where an extended 0/1 knapsack problem is taken as a basis. Furthermore, we introduce a new evolutionary approach to multicriteria optimization, the strength Pareto EA (SPEA), that combines several features of previous multiobjective EAs in a unique manner. It is characterized by (a) storing nondominated solutions externally in a second, continuously updated population, (b) evaluating an individual's fitness dependent on the number of external nondominated points that dominate it, (c) preserving population diversity using the Pareto dominance relationship, and (d) incorporating a clustering procedure in order to reduce the nondominated set without destroying its characteristics. The proof-of-principle results obtained on two artificial problems as well as a larger problem, the synthesis of a digital hardware-software multiprocessor system, suggest that SPEA can be very effective in sampling from along the entire Pareto-optimal front and distributing the generated solutions over the tradeoff surface. Moreover, SPEA clearly outperforms the other four multiobjective EAs on the 0/1 knapsack problem     

Motif difficulty (MD): a predictive measure of problem difficulty for evolutionary algorithms using network motifs

One of the major challenges in the field of evolutionary algorithms (EAs) is to characterise which kinds of problems are easy and which are not. Researchers have been attracted to predict the behaviour of EAs in different domains. We introduce fitness landscape networks (FLNs) that are formed using operators satisfying specific conditions and define a new predictive measure that we call motif difficulty (MD) for comparison-based EAs. Because it is impractical to exhaustively search the whole network, we propose a sampling technique for calculating an approximate MD measure. Extensive experiments on binary search spaces are conducted to show both the advantages and limitations of MD. Multidimensional knapsack problems (MKPs) are also used to validate the performance of approximate MD on FLNs with different topologies. The effect of two representations, namely binary and permutation, on the difficulty of MKPs is analysed.     

An iterative local search approach based on fitness landscapes analysis for the delay-constrained multicast routing problem

This paper presents the first fitness landscape analysis on the delay-constrained least-cost multicast routing problem (DCLC-MRP). Although the problem has attracted an increasing research attention over the past decade in telecommunications and operational research, little research has been conducted to analyze the features of its underlying landscape. Two of the most commonly used landscape analysis techniques, the fitness distance correlation analysis and the autocorrelation analysis, have been applied to analyze the global and local landscape features of DCLC-MRPs. A large amount of simulation experiments on a set of problem instances generated based on the benchmark Steiner tree problems in the OR-library reveals that the landscape of the DCLC-MRP is highly instance dependent with different landscape features. Different delay bounds also affect the distribution of solutions in the search space. The autocorrelation analysis on the benchmark instances of different sizes and delay bounds shows the impact of different local search heuristics and neighborhood structures on the fitness distance landscapes of the DCLC-MRP. The delay bound constraint in the DCLC-MRP has shown a great influence on the underlying landscape characteristic of the problem. Based on the fitness landscape analysis, an iterative local search (ILS) approach is proposed in this paper for the first time to tackle the DCLC-MRP. Computational results demonstrate the effectiveness of the proposed ILS algorithm for the problem in comparison with other algorithms in the literature.     

On the role of population size and niche radius in fitness sharing

We propose a characterization of the dynamic behavior of an evolutionary algorithm (EA) with fitness sharing as a function of both the niche radius and the population size. Such a characterization, given in terms of the mean and the standard deviation of the number of niches found during the evolution, can be applied to any EA employing a proportional selection mechanism and does not make any assumption on either the fitness landscape or the internal parameters of the EA itself. On the basis of the proposed characterization, a method for estimating the optimal values for the population size and the niche radius without any a priori information on the fitness landscape is presented and tested on a standard set of functions. The proposed method also provides the best solution for the problem at hand, i.e., the solution obtained in correspondence of such optimal values, at no additional cost.     

动态环境中的进化算法

目前关于进化算法（EA）的研究主要局限于静态优化问题,然而很多现实世界中的问题是动态的,对于这类时变的优化问题通常并不是要求EA发现极值点,而是需要EA能够尽可能紧密地跟踪极值点在搜索空间内的运行轨迹.为此,综述了使EA适用于动态优化问题的各种方法,如增加种群多样性、保持种群多样性、引入某种记忆策略和采用多种群策略等.     

An analysis of the interface between evolutionary algorithm operators and problem features for water resources problems. A case study in water distribution network design

Evolutionary Algorithms (EAs) have been widely employed to solve water resources problems for nearly two decades with much success. However, recent research in hyperheuristics has raised the possibility of developing optimisers that adapt to the characteristics of the problem being solved. In order to select appropriate operators for such optimisers it is necessary to first understand the interaction between operator and problem. This paper explores the concept of EA operator behaviour in real world applications through the empirical study of performance using water distribution networks (WDN) as a case study. Artificial networks are created to embody specific WDN features which are then used to evaluate the impact of network features on operator performance. The method extracts key attributes of the problem which are encapsulated in the natural features of a WDN, such as topologies and assets, on which different EA operators can be tested. The method is demonstrated using small exemplar networks designed specifically so that they isolate individual features. A set of operators are tested on these artificial networks and their behaviour characterised. This process provides a systematic and quantitative approach to establishing detailed information about an algorithm's suitability to optimise certain types of problem. The experiment is then repeated on real-world inspired networks and the results are shown to fit with the expected results.     

Unveiling evolutionary algorithm representation with DU maps

Evolutionary algorithms (EAs) have proven to be effective in tackling problems in many different domains. However, users are often required to spend a significant amount of effort in fine-tuning the EA parameters in order to make the algorithm work. In principle, visualization tools may be of great help in this laborious task, but current visualization tools are either EA-specific, and hence hardly available to all users, or too general to convey detailed information. In this work, we study the Diversity and Usage map (DU map), a compact visualization for analyzing a key component of every EA, the representation of solutions. In a single heat map, the DU map visualizes for entire runs how diverse the genotype is across the population and to which degree each gene in the genotype contributes to the solution. We demonstrate the generality of the DU map concept by applying it to six EAs that use different representations (bit and integer strings, trees, ensembles of trees, and neural networks). We present the results of an online user study about the usability of the DU map which confirm the suitability of the proposed tool and provide important insights on our design choices. By providing a visualization tool that can be easily tailored by specifying the diversity (D) and usage (U) functions, the DU map aims at being a powerful analysis tool for EAs practitioners, making EAs more transparent and hence lowering the barrier for their use.     

Probabilistic properties of fitness-based quasi-reflection in evolutionary algorithms

Evolutionary algorithms (EAs) excel in optimizing systems with a large number of variables. Previous mathematical and empirical studies have shown that opposition-based algorithms can improve EA performance. We review existing opposition-based algorithms and introduce a new one. The proposed algorithm is named fitness-based quasi-reflection and employs the relative fitness of solution candidates to generate new individuals. We provide the probabilistic analysis to prove that among all the opposition-based methods that we investigate, fitness-based quasi-reflection has the highest probability of being closer to the solution of an optimization problem. We support our theoretical findings via Monte Carlo simulations and discuss the use of different reflection weights. We also demonstrate the benefits of fitness-based quasi-reflection on three state-of-the-art EAs that have competed at IEEE CEC competitions. The experimental results illustrate that fitness-based quasi-reflection enhances EA performance, particularly on problems with more challenging solution spaces. We found that competitive DE (CDE) which was ranked tenth in CEC 2013 competition benefited the most from opposition. CDE with fitness-based quasi-reflection improved on 21 out of the 28 problems in the CEC 2013 test suite and achieved 100% success rate on seven more problems than CDE.