Updated June, 2023.

Project Experiences

Simple Covariance Matrix Self-Adaptation Evolution Strategy with Repelling Subpopulations

Main Researcher | SUSTech, May 2021
Advisor: Prof. Xin Yao

Goal: multimodal optimisation is an important research field of evolutionary computation. This work aims to adopt Niching method to solve such an optimisation problem and participate in the GECCO 2022 Competition.
Actions & Achievements: proposed a simple covariance matrix self-adaptation evolution strategy with repelling subpopulations (Simple-RS-CMSA-ES). The algorithm uses CMSA-ES to update each Niche, and adopts taboo regions to ensure exploration and exploitation. For functions with low dimensions or few peaks, the algorithm can find out all peaks. With the increase in the number of peaks or dimensions, the PR value of the algorithm decreases due to its limited local search ability. But in general, the performance of this algorithm ranks among the top 3 in the same course. [report]

Main Researcher | SUSTech, Apr 2021
Advisor: Prof. Xin Yao

Goal: Jing Xie et al. Proposed a Bi-level Grouping Optimization model (BIGO) for Grouping Constrained Storage Location Assignment Problems (SLAP-GC). However, this method has limitations in policy initialization such as randomly grouping items, which limits the convergence speed of the algorithm to find the optimal strategy. As such, this work aims to optimize the initialization of item grouping.
Actions & Achievements: three different heuristic information about the picking frequency are used and tested in the initial item grouping. According to Wilcoxon signed rank test statistics, the method used K-Means clustering performs better on our data sets, effectively obtaining better initial grouping strategies and accelerating the convergence speed of tabu search. [report]

Main Researcher | SUSTech, Mar 2021
Advisor: Prof. Xin Yao

Goal: the single objective numerical optimisation problem with bounded constraints is a classical problem in the field of evolutionary computing. CEP, FEP, IFEP and related algorithms have been applied to this problem. IFEP has the best search performance, but there is still optimisation space. This work aims to improve the convergence speed and local search ability of the IFEP algorithm.
Actions & Achievements: added simple arithmetic recombination to the IFEP algorithm, which extends the space of searching and increases the velocity of convergence; adopted elitist strategy in unconditional global local search and proposed an optimised mutation standard deviation coefficient equation and introduced it into the conditional local search, improving the local search capability near the optimal value. [report]

Head of the Electronic Control Group | HIT, Aug 2017 - Aug 2019

Completed the control of five robots, including writing code, building peripheral circuits and debugging robots.
Added the object-oriented concept to the driver code of the peripheral devices of the robot controller, and designed task scheduling and monitoring for the peripheral devices.
Used a gyroscope, accelerometer and other sensors for data fusion to sense the robot's state, realised three-dimensional positioning of PTZ through cascade PID algorithm, and realised motion calculation of omnidirectional chassis.
Guided teammates in developing the upper computer and completed various debugging tools.
Realize the smoothness of robot operation through filtering algorithm, and complete innovative functions such as “gyro rotation”