Browsing by Author "Fox, Matthew"

Now showing 1 - 3 of 3
  • Thumbnail Image
    ItemOpen Access
    Evolutionary Computation for Dynamic Optimisation Problems with Different Requirement Satisfaction
    (De Montfort University, 2023-08) Fox, Matthew
    In many real-world optimization problems, the search space changes over a period of time. Unlike the case of static optimization, in these time-varying optimization scenarios known as Dynamic Optimization Problems (DOPs), learning from previous evaluations can be beneficial in tackling the current environment under the assumption that the properties of the problems and the position of their optima cannot change significantly between two consecutive environmental changes. There are many approaches to dealing with DOPs, but most research focuses on tracking moving optima or maintaining population diversity. Studies in Evolutionary Dynamic Optimization (EDO) typically allow for many individual evaluations before the environment changes. When the environment changes very quickly, unsolved challenges arise where typical population-based algorithms may be negatively affected by this limiting number of samples. This thesis aims to solve Dynamic Optimization Problems where the environment is fast changing and has a limited budget of how many fitness evaluations an optimizer can take before changing. The work presented in this thesis builds on the essential area of performance evaluation and benchmarking in EDO research. Here, a benchmark problem is created that allows for predicting the movement of optima. Furthermore, a new optimization algorithm that solves fast-changing environments is proposed as a proof-of-concept to test this benchmark. The new benchmark, called Moving Peaks Benchmark with Attractors (MPBA), incorporates an attractor heuristic that attracts peaks to a specific location in the environment. The proposed benchmark is fully flexible, where the dynamics of the attractors and the rate at which a peak is attracted to such attractors can be modified. When these characteristics are adjusted, certain movement styles can be achieved by a peak. A new performance measure that focuses on comparing algorithms that use prediction is also introduced in response to this benchmark. Furthermore, a new optimization algorithm, named Fast Environmental Changes Particle Swarm Optimizer (FEC-PSO) is designed as a proof-of-concept to test the proposed benchmark. This algorithm uses an optimizer to gather information about the environment so that a surrogate model can be built to estimate the location of the global optimum. It is shown that the surrogate model, given a good enough set of samples and an appropriate interpolation function, can accurately represent the environment.
  • Thumbnail Image
    ItemOpen Access
    An experimental study of prediction methods in robust optimization over time
    (IEEE Press, 2020-07-19) Fox, Matthew; Yang, Shengxiang; Caraffini, Fabio
    Robust Optimization Over Time (ROOT) is a new method of solving Dynamic Optimization Problems in respect to choosing a robust solution, that would last over a number of environment changes, rather than the approach that chooses the optimal solution at every change. ROOT methods currently show that ROOT can be solved by predicting an individual fitness for a number of future environment changes. In this work, a benchmark problem based on the Modified Moving Peaks Benchmark (MMPB) is proposed that includes an attractor heuristic, that guides optima to a determined location in the environment, resulting in a more predictable optimum. We study a number of time series forecasting methods to test different prediction methods of future fitness values in a ROOT method. Four time series regression techniques are considered as the prediction method: Linear and Quadratic Regression, an Autoregressive model, and Support Vector Regression. We find that there is not much difference in choosing a simple Linear Regression to more advanced prediction methods. We also suggest that current benchmark problems that cannot be predicted will deceive the optimizer and ROOT framework as the peaks may move using a random walk. Results show an improvement in comparison with MMPB used in most ROOT studies.
  • Thumbnail Image
    ItemOpen Access
    A New Moving Peaks Benchmark with Attractors for Dynamic Evolutionary Algorithms
    (Elsevier, 2022-07-16) Fox, Matthew; Yang, Shengxiang; Caraffini, Fabio
    Prediction in evolutionary dynamic optimization (EDO), such as predicting the movement of optima, or when and how an environment will change, is a topic that is still under investigation and presents unsolved challenges. A few studies approach prediction based on re-initialising a population or requirement satisfaction problems such as Robust Optimization Over Time. The benchmark problems in these studies inherently use randomly changing parameters and therefore such randomness may make it difficult to compare these algorithms with other EDO approaches. In this paper, we introduce a new benchmark, called Moving Peaks Benchmark with Attractors, which incorporates an attractor heuristic that attracts peaks to a certain location in the environment into the moving peaks problem. The proposed benchmark is fully flexible where the dynamics of the attractors and the rate at which a peak is attracted to such attractors can be modified. By adjusting these characteristics, certain styles of movements can be achieved by a peak. We also introduce a new performance measure that focuses on the comparison of algorithms that use prediction. Seven EDO algorithms based on different working logics are chosen to give a wide representation of the state-of-the-art in this area. We argue that having predictable characteristics in the benchmark problem is more adequate for studying the performances and behaviours of those algorithms that embed prediction mechanisms. Experimental results obtained with the proposed benchmark show it's suitability for the EDO domain as all algorithms featuring prediction capabilities display higher accuracy than their competitors.