Evolutionary multi-objective optimization platform
Automated metric calculation without GUI is supported. Users can specify the metrics to display or save by setting the value of 'metName' when calling the main function platemo() with parameters.
Modify the way of defining gradient functions, where a method CalGrad is defined instead of CalObjGrad and CalConGrad in PROBLEM class, and a parameter 'gradFcn' is defined instead of 'objGradFcn' and 'conGradFcn' in UserProblem class.
Add a bi-level evolutionary algorithm BL-SAEA, three constrained multi-objective evolutionary algorithms IMTCMO_BS, MFO-SPEA2, and MOEA/D-2WA, a sparse multi-objective evolutionary algorithm SCEA, a surrogate-assisted multi-objective evolutionary algorithm SFA-DE, and two multi-objective feature selection algorithms MFFS and PRDH. There are currently 260 algorithms in the platform.
Add 15 EvoXBench problems CitySegMOP1-15 and 12 constrained multi-objective benchmark problems LSCM1-LSCM12. There are currently 508 problems in the platform.
Enhance the GUI with new features.
Add two sparse multi-objective evolutionary algorithms MGCEA and NUCEA. There are currently 252 algorithms in the platform.
Add a deep reinforcement learning based multi-objective evolutionary algorithm MOEA/D-DQN, two many-objective evolutionary algorithms HEA and SSCEA, two constrained multi-objective evolutionary algorithms MSCEA and TPCMaO, two surrogate-assisted evolutionary algorithms L2SMEA and MO-L2SMEA, and three surrogate-assisted constrained multi-objective evolutionary algorithms MGSAEA, RGA-M1-2, and RGA-M2-2. There are currently 250 algorithms in the platform.
Add a sparse multi-objective evolutionary algorithm S-NSGA-II, a multimodal multi-objective evolutionary algorithm CoMMEA, four surrogate-assisted multi-objective evolutionary algorithms ADSAPSO, EMMOEA, ESBCEO, and KTS, and three constrained multi-objective evolutionary algorithms CMEGL, IMTCMO, and MCCMO. There are currently 240 algorithms in the platform.
Add 15 constrained multi-objective benchmark problems SDC1-SDC15. There are currently 481 problems in the platform.
Add one multi-objective evolutionary algorithm TS-NSGA-II, six constrained multi-objective evolutionary algorithms CMME, CMOCSO, CMOEMT, CMOQLMT, C-TSEA, DP-PPS, two multi-modal multi-objective evolutionary algorithms CMMO and HHC-MMEA, one surrogate-assisted multi-objective evolutionary algorithm PC-SAEA, and one sparse multi-objective evolutionary algorithm SGECF. Refactor the code of CSEA. There are currently 231 algorithms in the platform.
Add 18 multi-objective neural architecture search benchmark problems C10MOP1-C10MOP9 and IN1KMOP1-IN1KMOP9. There are currently 466 problems in the platform.
Automated function creation is supported. Users can input a dataset as an objective function or constraint function when solving user-defined problems, where a function will be automatically fitted according to the dataset.
Add two large-scale multi-objective evolutionary algorithms FLEA and LERD, one expensive multi-objective optimization algorithm SMOA, and one constrained multi-objective evolutionary algorithm C3M. There are currently 220 algorithms in the platform.
Add 16 constrained multi-objective benchmark problems ZXH_CF1-ZXH_CF16. There are currently 448 problems in the platform.
Dynamic optimization, multitasking optimization, bilevel optimization, and robust optimization are now supported in PlatEMO.
Hybrid encoding is now supported in PlatEMO, where a problem can include real variables, integral variables, label variables, binary variables, and permutation variables simultaneously.
Maximum runtime is provided as a new termination criterion, which can be set instead of maximum number of function evaluations.
More algorithms and problems for single-objective optimization, multi-objective optimization, constrained optimization, sparse optimization, expensive optimization, multimodal optimization, dynamic optimization, multitasking optimization, bilevel optimization, and robust optimization. There are currently 216 algorithms and 432 problems in the platform.
More efficient and powerful GUI, where the execution of algorithms in the test module and application module is highly accelerated.
More performance metrics for different types of optimization problems, and the metrics are also tagged with labels. Different metrics will be shown in the dropdown lists when selecting different labels in the GUI.
Gradient based search is now supported in PlatEMO, where users can define gradient functions to accelerate the convergence via mathematical programming algorithms and gradient assisted evolutionary algorithms.
Remake the application module, a more powerful and friendly interface enables users to define problems more easily. The defined problems can also be saved into files and solved in other modules.
Add two multi-objective evolutionary algorithms MOEA/D-DYTS and MOEA/D-UR, two surrogate-assisted multi-objective evolutionary algorithms PB-NSGA-III and PB-RVEA, a constrained multi-objective evolutionary algorithm DSPCMDE, three large-scale multi-objective evolutionary algorithms FDV, IM-MOEA/D, and LMOEA-DS, two sparse multi-objective evolutionary algorithm SLMEA and SparseEA2, and four single-objective mathematical programming methods Adam, Nelder-Mead, RMSProp, and SD. There are currently 176 algorithms in the platform.
Add four multi-objective evolutionary algorithms DEA-GNG, ICMA, PeEA, and RVEA-iGNG. There are currently 162 algorithms in the platform.
Add five constrained multi-objective optimization problems FCP1-FCP5 and a sparse multi-objective optimization problem Sparse_KP. There are currently 345 problems in the platform.
When solving user-defined problems, the objective and constraint functions can be either @(x,d) or @(x).