Many-objective Evolutionary Algorithm Driven by Indicator Under Adaptive Reference Point Adjustment
-
摘要: 在具有不同Pareto前沿形状的优化问题上, 基于参考点的高维多目标进化算法表现出较差的通用性. 为了解决这个问题, 提出参考点自适应调整下评价指标驱动的高维多目标进化算法(Many-objective evolutionary algorithm driven by evaluation indicator under adaptive reference point adjustment, MaOEA-IAR). MaOEA-IAR提出Pareto前沿形状监测基础上的参考点自适应策略, 利用该策略选择一组候选解作为初始参考点; 然后通过曲线参数对参考点位置进行调整; 将最终得到的能够适应不同Pareto前沿的参考点用于计算增强的反世代距离指标, 基于指标值设计适应度函数作为选择标准. 实验证明提出的算法在处理各种Pareto前沿形状的优化问题时能获得较好的性能, 算法通用性高.
-
关键词:
- 参考点自适应 /
- 评价指标 /
- 高维多目标 /
- Pareto前沿形状
Abstract: Many-objective evolutionary algorithms based on reference points show poor versatility on optimization problems with different shapes of Pareto fronts. To address this issue, this paper proposes a many-objective evolutionary algorithm driven by evaluation indicator under adaptive reference point adjustment (MaOEA-IAR). MaOEA-IAR proposes a reference point adaptation strategy based on the Pareto front shape monitoring, and uses this strategy to select a group of candidate solutions as the initial reference points, then adjusts their positions with the curve parameter. The final obtained reference points that can adapt to different Pareto fronts are used to calculate the enhanced inverted generational distance indicator, the fitness function is designed as selection criterion based on the indicator value. The experiment shows that the algorithm proposed in this paper can get good performance and high versatility when dealing with the optimization problems with various shapes of Pareto fronts. -
表 1 测试问题
Table 1 Test questions
问题 目标数目M 决策变量D PF DTLZ1 5, 10, 15, 25 M−1+5 线性 DTLZ2~DTLZ4 5, 10, 15, 25 M−1+10 凹型 DTLZ5~DTLZ6 5, 10, 15, 25 M−1+10 退化 DTLZ7 5, 10, 15, 25 M−1+20 断开 IDTLZ1 5, 10, 15, 25 M−1+5 倒置 IDTLZ2 5, 10, 15, 25 M−1+10 倒置 WFG1 5, 10, 15, 25 M−1+10 混合 WFG2 5, 10, 15, 25 M−1+10 断开 WFG3 5, 10, 15, 25 M−1+10 退化 WFG4~DTLZ9 5, 10, 15, 25 M−1+10 凹型 表 2 不同目标对应的种群规模
Table 2 Population sizes corresponding to different objectives
目标数目M (H1, H2) 种群规模N 5 (6, 0) 126 10 (3, 2) 275 15 (2, 1) 135 25 (2, 1) 350 表 3 不同目标对应的终止条件
Table 3 Termination conditions corresponding to different objectives
目标数目M 种群规模N 进化代数 评价次数 5 126 800 100800 10 275 1000 275000 15 135 1300 175500 25 350 1500 525000 表 4 MOEA-DD、NSGA-III、RVEA、MOMBI-II、AR-MOEA以及MaOEA-IAR在DTLZ1~DTLZ4, WFG4~WFG9上获得IGD值的统一结果(均值和标准差). 最好的结果已被标记
Table 4 The statistical results (mean and standard deviation) of the IGD values obtained by MOEA-DD、NSGA-III、RVEA、MOMBI-II、AR-MOEA and MaOEA-IAR on DTLZ1~DTLZ4, WFG4~WFG9. The best results are highlighted
问题 M MOEA-DD NSGA-III RVEA MOMBI-II AR-MOEA MaOEA-IAR DTLZ1 5 6.3354×10−2
(8.24×10−5)+6.3588×10−2
(2.93×10−4)+6.3365×10−2
(1.44×10−4)+6.7530×10−2
(8.96×10−3)+1.3555×10−3
(7.87×10−6)+2.6870×10−1
(7.95×10−2)10 1.0784×10−1
(3.02×10−4)+1.2073×10−1
(2.91×10−2)+1.0804×10−1
(1.45×10−3)+2.1744×10−1
(2.32×10−2)+1.3680×10−3
(1.14×10−4)+4.7099×10+1
(1.33×10+1)15 1.4088×10−1
(5.94×10−3)+1.9693×10−1
(2.97×10−2)+1.6130×10−1
(7.54×10−3)+2.8789×10−1
(3.92×10−2)+3.1383×10−3
(1.58×10−3)+5.6383×10+1
(2.03×10+1)25 1.0354×10−1
(5.86×10−2)+2.7045×10−1
(1.96×10−1)=1.3372×10−1
(7.61×10−2)+2.0490×10−1
(1.19×10−1)+1.2392×10−1
(7.02×10−2)+2.4067×10−1
(1.44×10−1)DTLZ2 5 1.9506×10−1
(8.00×10−5)+1.9538×10−1
(1.18×10−4)+1.9508×10−1
(7.34×10−5)+2.0140×10−1
(1.93×10−3)=4.3609×10−3
(7.50×10−6)+2.0137×10−1
(1.61×10−3)10 4.2179×10−1
(8.12×10−4)−4.4544×10−1
(3.60×10−2)−4.2291×10−1
(1.61×10−3)−4.4619×10−1
(6.50×10−3)−4.9235×10−2
(1.43×10−4)−4.7588×10−2
(4.60×10−4)15 6.1801×10−1
(1.88×10−2)−6.4108×10−1
(2.01×10−2)−6.3958×10−1
(3.67×10−2)−8.5287×10−1
(6.41×10−2)−1.0058×10−2
(1.56×10−3)=1.1135×10−2
(1.23×10−3)25 4.3015×10−1
(2.45×10−1)−5.3577×10−1
(3.10×10−1)−4.3033×10−1
(2.45×10−1)−7.0528×10−1
(4.12×10−1)−6.1891×10−1
(3.45×10−1)−4.2923×10−1
(2.44×10−1)DTLZ3 5 1.9907×10−1
(3.45×10−3)+7.1945×10−1
(1.21×10+0)+2.0829×10−1
(2.11×10−2)+2.0571×10−1
(4.19×10−3)+4.4165×10−3
(5.43×10−5)+1.0190×10+0
(6.60×10−1)10 4.2392×10−1
(3.09×10−3)+1.1841×10+0
(1.08×10+0)+4.3390×10−1
(1.54×10−2)+6.9344×10−1
(1.66×10−1)+3.1091×10−2
(9.02×10−2)+4.8634×10+2
(9.91×10+1)15 6.6842×10−1
(2.26×10−1)+5.5656×10+0
(2.83×10+0)+8.3685×10−1
(3.72×10−1)+1.1033×10+0
(2.16×10−2)+3.8594×10−1
(5.94×10−1)+6.4283×10+2
(1.08×10+2)25 4.3041×10−1
(2.46×10−1)+8.3189×10+0
(6.16×10+0)−4.3076×10−1
(2.46×10−1)+8.7119×10−1
(5.08×10−1)+4.6567×10−1
(2.38×10−1)+9.8829×10−1
(5.81×10−1)DTLZ4 5 1.9508×10−1
(9.98×10−5)+2.4346×10−1
(9.84×10−2)−1.9514×10−1
(1.62×10−4)+2.5703×10−1
(9.71×10−2)−4.4014×10−3
(1.06×10−4)+2.0129×10−1
(1.72×10−3)10 4.2271×10−1
(1.73×10−3)−4.5035×10−1
(3.18×10−2)−4.3317×10−1
(2.13×10−3)−4.5814×10−1
(1.71×10−2)−5.7202×10−3
(5.61×10−4)−5.6898×10−3
(1.01×10−4)15 6.3495×10−1
(1.18×10−2)−6.4865×10−1
(2.13×10−2)−6.3283×10−1
(7.39×10−3)−6.5937×10−1
(2.31×10−2)−1.1470×10−2
(1.34×10−3)−9.9175×10−3
(1.41×10−3)25 4.3041×10−1
(2.45×10−1)−4.6255×10−1
(2.72×10−1)−4.3053×10−1
(2.46×10−1)−4.3098×10−1
(2.46×10−1)−5.8948×10−1
(2.86×10−1)−4.3021×10−1
(2.44×10−1)WFG4 5 1.2337×10+0
(4.54×10−3)−1.1663×10+0
(3.61×10−3)−1.1649×10+0
(2.28×10−3)−1.8356×10+0
(2.02×10−1)−1.1244×10+0
(1.47×10−2)−1.1192×10+0
(2.58×10−1)10 6.0999×10+0
(1.66×10−1)−4.5156×10+0
(4.39×10−2)−4.3816×10+0
(5.05×10−2)−5.6693×10+0
(4.27×10−1)−4.5463×10+0
(9.53×10−3)−4.0153×10+0
(2.71×10−2)15 9.4478×10+0
(7.47×10−1)−9.1431×10+0
(9.63×10−2)−9.4338×10+0
(3.83×10−1)−2.0211×10+1
(1.35×10+0)−9.3716×10+0
(3.68×10−2)−8.4000×10+0
(1.11×10−1)25 2.4032×10+1
(2.95×10−1)−1.3110×10+1
(7.97×10−1)−1.2394×10+1
(2.92×10−1)−3.7271×10+1
(3.18×10+0)−1.5196×10+1
(2.61×10−1)−1.1405×10+1
(1.89×10−2)WFG5 5 1.2116×10+0
(6.46×10−3)−1.1459×10+0
(4.71×10−3)−1.1554×10+0
(2.17×10−3)−1.5867×10+0
(1.16×10−1)−1.1648×10+0
(2.17×10−4)−1.1148×10+0
(1.31×10−2)10 6.2798×10+0
(1.39×10−1)−4.4654×10+0
(1.82×10−2)−4.3730×10+0
(4.27×10−2)−5.3610×10+0
(2.19×10−2)−4.5296×10+0
(9.52×10−3)−3.9834×10+0
(1.90×10−2)15 1.1246×10+1
(1.85×10−1)−8.9556×10+0
(7.70×10−2)−9.8645×10+0
(1.66×10−1)−2.4601×10+1
(1.34×10+0)−9.3564×10+0
(5.48×10−2)−8.2316×10+0
(8.70×10−2)25 2.2399×10+1
(1.96×10−1)−1.1842×10+1
(1.14×10+0)=1.1407×10+1
(5.45×10−2)=4.5754×10+1
(3.19×10+0)−1.4772×10+1
(2.79×10−1)−1.1370×10+1
(3.12×10−3)WFG6 5 1.2276×10+0
(1.14×10−2)−1.1617×10+0
(2.62×10−3)−1.1638×10+0
(2.18×10−3)−2.1021×10+0
(3.69×10−1)−1.1624×10+0
(1.24×10−3)−1.1443×10+0
(2.89×10−2)10 6.0366×10+0
(1.65×10−1)−4.5795×10+0
(1.88×10−2)−4.4020×10+0
(7.33×10−2)−5.2873×10+0
(2.86×10−2)−4.5333×10+0
(1.12×10−2)−4.0605×10+0
(4.78×10−2)15 1.0818×10+1
(1.01×10+0)−9.3642×10+0
(3.84×10−1)−1.0520×10+1
(4.62×10−1)−1.8732×10+1
(1.89×10+0)−9.3953×10+0
(5.07×10−2)−8.5397×10+0
(1.76×10−1)25 2.1950×10+1
(6.43×10−1)−1.5582×10+1
(9.19×10−1)−1.5281×10+1
(9.27×10−1)−3.5035×10+1
(2.81×10+0)−1.5206×10+1
(3.56×10−1)−1.1375×10+1
(1.29×10−2)WFG7 5 1.2454×10+0
(8.15×10−3)−1.1680×10+0
(3.67×10−3)−1.1693×10+0
(3.05×10−3)−1.8487×10+0
(2.46×10−1)−1.1787×10+0
(8.19×10−4)−1.1499×10+0
(3.17×10−2)10 5.1499×10+0
(2.73×10−1)−4.5094×10+0
(3.76×10−2)−4.3149×10+0
(5.44×10−2)−5.4063×10+0
(5.76×10−2)−4.5123×10+0
(1.07×10−2)−3.9423×10+0
(4.82×10−2)15 8.9281×10+0
(8.74×10−2)−9.0694×10+0
(9.98×10−2)−9.1250×10+0
(2.71×10−1)−1.7892×10+1
(1.65×10+0)−9.4000×10+0
(3.81×10−2)−8.7103×10+0
(1.27×10−1)25 1.6767×10+1
(1.55×10+0)−1.3522×10+1
(7.95×10−1)−1.1672×10+1
(3.01×10−1)=3.2655×10+1
(2.66×10+0)−1.5873×10+1
(1.94×10−1)−1.1572×10+1
(8.12×10−2)WFG8 5 1.2301×10+0
(8.57×10−3)−1.1867×10+0
(9.32×10−2)=1.1770×10+0
(4.82×10−3)−2.9462×10+0
(3.23×10−2)−1.1575×10+0
(2.95×10−3)−1.1675×10+0
(1.66×10−2)10 5.2132×10+0
(3.48×10−1)−4.5532×10+0
(3.00×10−1)−4.3661×10+0
(8.88×10−2)−6.0943×10+0
(8.02×10−1)−4.6273×10+0
(4.43×10−2)−4.1413×10+0
(7.20×10−2)15 9.2706×10+0
(3.66×10−1)−9.1735×10+0
(4.45×10−1)−9.9548×10+0
(7.95×10−1)−2.0437×10+1
(1.84×10+0)−9.3736×10+0
(7.14×10−2)−8.7680×10+0
(2.43×10−1)25 2.2579×10+1
(3.17×10+0)−1.6519×10+1
(6.99×10−1)−1.3760×10+1
(1.73×10+0)−3.9378×10+1
(2.20×10+0)−1.5213×10+1
(2.84×10−1)−1.1409×10+1
(2.03×10−2)WFG9 5 1.2087×10+0
(1.17×10−2)−1.1112×10+0
(1.08×10−2)−1.1315×10+0
(4.39×10−3)−2.6613×10+0
(1.67×10−1)−1.1476×10+0
(2.59×10−3)−1.0867×10+0
(1.42×10−2)10 5.2551×10+0
(3.81×10−1)−4.3117×10+0
(7.01×10−2)−4.3306×10+0
(6.41×10−2)−5.2773×10+0
(6.59×10−2)−4.4943×10+0
(1.01×10−2)−3.8995×10+0
(3.21×10−2)15 8.9598×10+0
(5.37×10−1)−8.6320×10+0
(1.41×10−1)−8.8404×10+0
(2.73×10−1)−2.5164×10+1
(1.97×10+0)−9.1233×10+0
(3.81×10−2)−8.1162×10+0
(1.37×10−1)25 1.5766×10+1
(3.46×10+0)−1.2211×10+1
(7.44×10−1)−1.2343×10+1
(1.28×10+0)−4.8512×10+1
(2.40×10+0)−1.5086×10+1
(2.89×10−1)−1.1166×10+1
(8.75×10−2)+/−/= 10/30/0 7/30/3 10/28/2 8/31/1 10/29/1 “+”表明该算法优于MaOEA-IAR, “−”劣于MaOEA-IAR, “=”则表示与MaOEA-IAR性能相似 表 5 MOEA-DD、NSGA-III、RVEA、MOMBI-II、AR-MOEA以及MaOEA-IAR在DTLZ1~DTLZ4, WFG4~WFG9上获得PD值的统一结果(均值和标准差). 最好的结果已被标记
Table 5 The statistical results (mean and standard deviation) of the PD values obtained by MOEA-DD、NSGA-III、RVEA、MOMBI-II、AR-MOEA and MaOEA-IAR on DTLZ1~DTLZ4, WFG4~WFG9. The best results are highlighted
问题 M MOEA-DD NSGA-III RVEA MOMBI-II AR-MOEA MaOEA-IAR DTLZ1 5 7.0347×10+5
(1.50×10+5)−6.6822×10+5
(2.89×10+5)−1.0144×10+6
(2.23×10+5)−6.5737×10+5
(5.56×10+5)−3.1124×10+5
(1.14×10+5)−1.1335×10+7
(2.48×10+7)10 3.9493×10+9
(4.51×10+8)−4.3920×10+9
(1.31×10+9)−4.4832×10+9
(5.54×10+8)−1.3231×10+8
(1.60×10+8)−3.9031×10+9
(3.47×10+8)−2.1838×10+12
(4.55×10+11)15 2.0771×10+11
(2.15×10+10)−7.2781×10+10
(4.17×10+10)−1.8665×10+11
(5.48×10+10)−2.8086×10+8
(4.77×10+8)−5.4706×10+10
(1.65×10+10)−2.4099×10+13
(8.65×10+12)25 4.2240×10+13
(1.16×10+12)+2.4959×10+13
(1.82×10+13)+1.5947×10+13
(3.82×10+12)+2.6381×10+9
(4.68×10+9)−1.2365×10+13
(6.44×10+12)+1.6674×10+11
(4.44×10+11)DTLZ2 5 4.5169×10+6
(3.69×10+5)−4.1624×10+6
(5.12×10+5)−4.4519×10+6
(4.97×10+5)−3.1301×10+6
(4.36×10+5)−2.2535×10+6
(1.23×10+5)−1.3815×10+7
(1.83×10+6)10 1.0700×10+10
(4.94×10+8)−1.5036×10+10
(4.96×10+9)−1.4787×10+10
(8.21×10+8)−6.3297×10+9
(9.76×10+8)−1.4371×10+10
(9.34×10+8)−4.8565×10+10
(8.50×10+9)15 2.3742×10+11
(5.12×10+10)−5.4986×10+10
(3.53×10+10)−1.5788×10+11
(3.98×10+10)−1.4169×10+9
(1.60×10+9)−2.2431×10+10
(1.21×10+10)−4.6071×10+11
(2.42×10+11)25 1.5327×10+13
(1.82×10+12)−1.9112×10+13
(9.88×10+12)−1.8880×10+13
(3.09×10+12)−1.4360×10+10
(2.20×10+10)−2.1209×10+12
(1.31×10+12)−1.0367×10+15
(3.68×10+13)DTLZ3 5 1.0522×10+7
(3.37×10+6 )+1.8368×10+7
(1.98×10+7 )+9.1974×10+6
(3.08×10+6 )+2.8101×10+6
(4.82×10+5 )+3.0005×10+6
(3.36×10+5 )+2.0182×10+6
(4.79×10+6)10 1.3601×10+10
(2.86×10+9)−2.2578×10+10
(1.68×10+10)−1.6599×10+10
(2.67×10+9)−1.1037×10+9
(1.66×10+9)−4.7006×10+9
(1.96×10+9)−2.9582×10+13
(4.66×10+12)15 3.6156×10+11
(8.59×10+11)−1.1462×10+12
(8.77×10+11)−1.2788×10+11
(2.39×10+11)−3.1557×10+8
(4.47×10+8)−4.9753×10+10
(4.57×10+10)−6.5391×10+14
(1.63×10+14)25 1.6252×10+13
(1.40×10+12)+3.2406×10+15
(2.26×10+15)+1.0735×10+13
(5.81×10+12)+7.3339×10+8
(1.67×10+9)−2.2266×10+13
(2.19×10+13)+9.6558×10+10
(2.77×10+11)DTLZ4 5 3.7620×10+6
(6.71×10+5)−3.3390×10+6
(8.43×10+5)−4.4024×10+6
(6.07×10+5)−2.1806×10+6
(1.14×10+6)−2.4008×10+6
(1.56×10+5)−1.3225×10+7
(1.43×10+6)10 1.3637×10+10
(1.08×10+9)+5.6960×10+9
(2.42×10+9)−1.0848×10+10
(2.29×10+9)+5.9777×10+7
(2.68×10+7)−4.0114×10+9
(1.10×10+9)−9.1937×10+9
(2.85×10+9)15 2.7750×10+9
(1.79×10+9)−5.2847×10+8
(5.23×10+8)−2.0648×10+9
(1.07×10+9)−4.0923×10+7
(1.33×10+8)−5.8144×10+9
(3.64×10+9)−3.9392×10+10
(1.98×10+10)25 3.7432×10+12
(9.47×10+11)−3.1839×10+11
(3.51×10+11)−3.8552×10+11
(1.51×10+11)−2.8053×10+8
(5.19×10+8)−1.2420×10+11
(5.99×10+10)−8.4925×10+14
(3.45×10+13)WFG4 5 1.2326×10+8
(6.16×10+6)−1.3502×10+8
(4.43×10+6)−1.2748×10+8
(7.49×10+6)−1.8889×10+7
(3.43×10+6)−6.4422×10+7
(4.60×10+6)−2.0686×10+8
(1.12×10+7)10 1.4338×10+11
(9.29×10+9)−2.9001×10+11
(1.68×10+10)−1.9392×10+11
(1.22×10+10)−7.4282×10+10
(2.70×10+10)−1.4488×10+11
(8.57×10+9)−6.6989×10+11
(2.52×10+10)15 3.4224×10+12
(5.49×10+11)−5.5478×10+12
(7.84×10+11)−4.0484×10+12
(1.04×10+12)−2.0839×10+10
(1.96×10+10)−2.0698×10+12
(6.60×10+11)−1.9906×10+13
(1.68×10+12)25 9.4061×10+13
(1.83×10+13)−2.2238×10+15
(6.40×10+14)−1.6309×10+14
(2.64×10+13)−2.5978×10+11
(6.99×10+11)−3.9968×10+14
(1.07×10+14)−1.0349×10+16
(5.36×10+14)WFG5 5 1.2123×10+8
(6.36×10+6)−1.4121×10+8
(8.96×10+6)−1.1886×10+8
(4.50×10+6)−2.3887×10+7
(4.70×10+6)−4.1612×10+7
(3.93×10+6)−1.9277×10+8
(1.04×10+7)10 1.9457×10+11
(1.24×10+10)−3.7692×10+11
(1.60×10+10)−2.2741×10+11
(1.20×10+10)−1.1103×10+11
(8.12×10+9)−1.2033×10+11
(6.96×10+9)−6.8603×10+11
(3.29×10+10)15 3.9863×10+12
(7.29×10+11)−8.8868×10+12
(7.95×10+11)−4.9690×10+12
(3.43×10+11)−2.0497×10+10
(3.24×10+10)−1.4482×10+12
(4.81×10+11)−2.0970×10+13
(1.56×10+12)25 2.8409×10+14
(2.86×10+13)−1.6088×10+15
(7.01×10+14)−2.7765×10+14
(3.00×10+13)−1.1499×10+12
(2.16×10+12)−5.7645×10+14
(6.09×10+13)−1.2220×10+16
(5.00×10+14)WFG6 5 1.1994×10+8
(8.37×10+6)−1.2908×10+8
(1.30×10+7)−1.1997×10+8
(7.99×10+6)−1.7038×10+7
(6.02×10+6)−5.4499×10+7
(6.80×10+6)−1.8108×10+8
(1.37×10+7)10 1.6619×10+11
(1.38×10+10)−2.7497×10+11
(1.56×10+10)−1.9869×10+11
(1.18×10+10)−9.2008×10+10
(6.73×10+9)−1.1142×10+11
(6.52×10+9)−6.2142×10+11
(2.98×10+10)15 3.5945×10+12
(7.09×10+11)−4.3937×10+12
(8.35×10+11)−2.6933×10+12
(5.32×10+11)−4.3260×10+10
(8.15×10+10)−5.8023×10+11
(5.03×10+11)−1.7544×10+13
(1.37×10+12)25 5.8142×10+13
(1.39×10+13)−2.2239×10+15
(4.43×10+14)−1.3831×10+14
(3.12×10+13)−1.0474×10+12
(4.06×10+12)−1.3621×10+14
(5.09×10+13)−1.0081×10+16
(4.97×10+14)WFG7 5 1.4307×10+8
(8.30×10+6)−1.5328×10+8
(1.08×10+7)−1.3594×10+8
(7.42×10+6)−2.0429×10+7
(7.25×10+6)−8.4724×10+7
(5.77×10+6)−2.0010×10+8
(1.05×10+7)10 2.5289×10+11
(1.42×10+10)−3.8153×10+11
(3.44×10+10)−2.6215×10+11
(1.24×10+10)−1.2614×10+11
(9.82×10+9)−1.8152×10+11
(7.90×10+9)−6.9190×10+11
(3.57×10+10)15 5.8545×10+12
(7.41×10+11)−9.3356×10+12
(1.86×10+12)−7.4133×10+12
(1.39×10+12)−6.2349×10+10
(7.43×10+10)−2.2599×10+12
(9.62×10+11)−1.4999×10+13
(1.54×10+12)25 6.8227×10+14
(1.72×10+14)−1.8449×10+15
(4.42×10+14)−7.9199×10+14
(1.16×10+14)−2.3362×10+12
(5.57×10+12)−8.9765×10+14
(1.59×10+14)−7.1165×10+15
(4.06×10+14)WFG8 5 1.5013×10+8
(8.11×10+6)−1.9783×10+8
(1.05×10+7)−1.7829×10+8
(7.32×10+6)−3.6308×10+7
(4.44×10+6)−1.4408×10+8
(7.57×10+6)−2.3335×10+8
(1.10×10+7)10 1.8219×10+11
(1.68×10+10)−3.8567×10+11
(5.74×10+10)−2.0046×10+11
(2.71×10+10)−8.2783×10+10
(3.03×10+10)−1.7126×10+11
(1.80×10+10)−6.0173×10+11
(7.82×10+10)15 4.0400×10+12
(9.81×10+11)−9.9234×10+12
(1.83×10+12)=3.1853×10+12
(9.93×10+11)−1.9227×10+11
(1.86×10+11)−1.6871×10+12
(1.26×10+12)−1.2895×10+13
(5.22×10+12)25 1.3971×10+14
(9.91×10+13)−4.8077×10+15
(6.01×10+14)−2.4789×10+14
(8.47×10+13)−4.7555×10+12
(8.90×10+12)−3.1451×10+14
(9.11×10+13)−1.0680×10+16
(8.26×10+14)WFG9 5 1.9141×10+8
(8.37×10+6)−2.3341×10+8
(1.10×10+7)−1.9389×10+8
(7.79×10+6)−3.6203×10+7
(5.92×10+6)−1.5846×10+8
(6.96×10+6)−2.6391×10+8
(1.35×10+7)10 3.7972×10+11
(3.13×10+10)−6.0898×10+11
(4.27×10+10)−4.2790×10+11
(1.62×10+10)−2.0148×10+11
(1.32×10+10)−3.5255×10+11
(1.94×10+10)−9.8787×10+11
(3.03×10+10)15 1.4912×10+13
(1.61×10+12)−2.0973×10+13
(2.10×10+12)−1.2833×10+13
(1.43×10+12)−5.5018×10+10
(8.57×10+10)−1.0318×10+13
(1.36×10+12)−3.2318×10+13
(1.52×10+12)25 4.2407×10+15
(8.53×10+14)−9.2528×10+15
(1.06×10+15)−3.7876×10+15
(6.04×10+14)−4.6745×10+12
(7.48×10+12)−5.3764×10+15
(3.32×10+14)−1.6297×10+16
(4.08×10+14)+/−/= 4/36/0 3/36/1 4/36/0 1/39/0 3/37/0 “+”表明该算法优于MaOEA-IAR, “−”劣于MaOEA-IAR, “=”则表示与MaOEA-IAR性能相似 表 6 MOEA-DD、NSGA-III、RVEA、MOMBI-II、AR-MOEA以及MaOEA-IAR在DTLZ5~DTLZ7, IDTLZ1~IDTLZ2, WFG1~WFG3上获得IGD值的统一结果(均值和标准差). 最好的结果已被标记
Table 6 The statistical results (mean and standard deviation) of the IGD values obtained by MOEA-DD、NSGA-III、RVEA、MOMBI-II、AR-MOEA and MaOEA-IAR on DTLZ5~DTLZ7, IDTLZ1~IDTLZ2, WFG1~WFG3. The best results are highlighted
问题 M MOEA-DD NSGA-III RVEA MOMBI-II AR-MOEA MaOEA-IAR DTLZ5 5 1.0286×10−1
(4.56×10−3)−1.5564×10−1
(6.36×10−2)−1.7052×10−1
(2.08×10−2)−2.6836×10−1
(1.37×10−2)−1.2293×10−1
(2.95×10−3)−9.2396×10−2
(1.58×10−2)10 1.3275×10−1
(1.77×10−2)+2.9646×10−1
(9.03×10−2)+3.7614×10−1
(6.44×10−2)+5.8454×10−1
(1.64×10−1)−9.4984×10−2
(3.11×10−3)+4.8043×10−1
(1.04×10−1)15 1.5268×10−1
(4.95×10−3)−2.5900×10−1
(4.20×10−2)−5.8957×10−1
(1.74×10−1)−6.4004×10−1
(6.56×10−2)−6.0343×10−1
(1.01×10−2)−1.3313×10−1
(1.25×10−2)25 1.4321×10−1
(4.38×10−2)+8.5001×10−1
(5.98×10−1)−3.6054×10−1
(2.03×10−1)−5.7672×10−1
(2.97×10−1)−9.6898×10−2
(1.91×10−2)−3.2482×10−2
(1.69×10−1)DTLZ6 5 1.0630×10−1
(9.13×10−3)+2.8728×10−1
(2.28×10−1)+1.5455×10−1
(3.80×10−2)+3.8061×10−1
(2.25×10−1)=3.3734×10−1
(3.50×10−2)−3.1149×10−1
(5.18×10−2)10 2.0616×10−1
(1.64×10−1)+1.5928×10+0
(5.87×10−1)+2.8394×10−1
(8.30×10−2)+5.4111×10−1
(2.36×10−1)+1.6497×10−1
(8.66×10−3)+2.4129×10+0
(3.39×10−1)15 2.0227×10−1
(1.34×10−1)+2.5854×10+0
(1.02×10+0)=3.2020×10−1
(1.73×10−1)+7.5239×10−1
(2.82×10−1)+1.2810×10−1
(1.01×10−1)+2.9171×10+0
(4.24×10−1)25 1.5709×10−1
(2.14×10−2)−6.1820×10+0
(3.10×10+0)−2.2612×10−1
(6.64×10−2)−5.4816×10−1
(2.83×10−1)−5.0812×10−1
(3.00×10−1)−1.0126×10−1
(2.94×10−2)DTLZ7 5 3.0002×10+0
(9.72×10−4)−3.3903×10−1
(1.25×10−2)+6.0021×10−1
(3.12×10−2)−4.8750×10−1
(8.74×10−2)−8.7917×10−3
(4.95×10−4)+4.0035×10−1
(8.59×10−2)10 2.0394×10+0
(3.09×10−1)−1.6480×10+0
(2.43×10−1)−1.8321×10+0
(4.74×10−1)−4.8283×10+0
(7.64×10−1)−2.8392×10−2
(3.28×10−3)+8.4705×10−1
(3.02×10−2)15 3.4630×10+0
(8.21×10−2)−8.8601×10+0
(7.49×10−1)−2.3863×10+0
(2.46×10−1)−1.1222×10+1
(1.43×10−1)−5.4456×10−1
(3.13×10−1)+1.6482×10+0
(3.88×10−2)25 4.9987×10+0
(4.01×10−2)−1.4671×10+1
(1.41×10+0)−3.5010×10+0
(4.87×10−1)+1.9012×10+1
(2.83×10−1)−3.7797×10+0
(1.17×10+0)+4.0533×10+0
(1.41×10−5)IDTLZ1 5 1.5305×10−1
(4.10×10−2)=1.3547×10−1
(1.54×10−2)=1.5234×10−1
(1.70×10−2)=1.1355×10−1
(3.26×10−4)=6.5841×10−2
(8.22×10−4)+2.5128×10−1
(1.65×10−1)10 2.2790×10−1
(1.05×10−2)−1.4175×10−1
(3.27×10−3)−2.5816×10−1
(2.39×10−2)−1.8633×10−1
(6.51×10−3)−1.2386×10−1
(3.42×10−2)=1.1027×10−1
(9.23×10−4)15 3.2984×10−1
(3.09×10−2)−2.1719×10−1
(4.73×10−3)−3.5590×10−1
(2.89×10−2)−2.1098×10−1
(1.14×10−2)−1.9277×10−1
(1.29×10−2)−1.7215×10−1
(7.39×10−2)25 3.2145×10−1
(1.81×10−2)−1.9936×10−1
(3.93×10−3)−3.8507×10−1
(3.13×10−2)−2.6360×10−1
(6.52×10−3)−4.3902×10−2
(1.27×10−1)−2.1252×10−2
(1.63×10−2)IDTLZ2 5 2.7867×10−1
(5.70×10−3)−2.4322×10−1
(6.88×10−3)−2.9833×10−1
(5.13×10−3)−3.1740×10−1
(1.15×10−3)−2.1362×10−1
(5.25×10−3)−2.0888×10−1
(7.30×10−3)10 7.3889×10−1
(6.96×10−3)−6.0395×10−1
(1.69×10−2)−6.7982×10−1
(7.75×10−3)−6.6946×10−1
(4.82×10−3)−4.4639×10−1
(6.90×10−3)−4.4152×10−1
(6.68×10−3)15 9.4524×10−1
(2.03×10−2)−7.6174×10−1
(1.42×10−2)−8.7051×10−1
(1.47×10−2)−8.5834×10−1
(7.35×10−3)−6.8015×10−1
(1.79×10−2)−5.7753×10−1
(9.19×10−3)25 1.1245×10+0
(1.07×10−2)−8.3706×10−1
(1.63×10−2)−1.0678×10+0
(1.60×10−2)−1.0151×10+0
(4.12×10−3)−8.9235×10−2
(2.87×10−2)-2.8098×10−2
(6.87×10−4)WFG1 5 8.4034×10−1
(1.15×10−1)−5.6876×10−1
(4.95×10−2)−5.3556×10−1
(4.24×10−2)−5.3968×10−1
(6.19×10−2)−5.1692×10−1
(1.26×10−2)−4.4951×10−1
(5.18×10−3)10 1.2623×10+0
(6.70×10−2)−1.2030×10+0
(7.30×10−2)−1.0736×10+0
(6.23×10−2)=1.2539×10+0
(5.56×10−2)−1.1573×10+0
(3.43×10−2)−1.0186×10+0
(3.21×10−2)15 1.9032×10+0
(6.34×10−2)−1.9456×10+0
(8.90×10−2)−1.8780×10+0
(7.27×10−2)−2.5378×10+0
(2.18×10−1)−1.7899×10+0
(4.09×10−2)=1.7788×10+0
(4.73×10−2)25 3.8022×10+0
(5.48×10−2)−3.0121×10+0
(4.19×10−1)−3.1303×10+0
(1.12×10−1)−3.6938×10+0
(9.71×10−2)−3.1942×10+0
(6.67×10−2)−2.6916×10+0
(1.82×10−1)WFG2 5 5.7666×10−1
(1.64×10−2)−4.6948×10−1
(3.19×10−3)+4.4930×10−1
(1.01×10−2)+5.1672×10−1
(6.95×10−2)+4.7596×10−1
(2.55×10−3)+5.4316×10−1
(2.97×10−2)10 1.4487×10+0
(2.06×10−2)−1.2010×10+0
(1.43×10−1)−1.1062×10+0
(3.99×10−2)−1.6478×10+0
(5.00×10−1)−1.4937×10+0
(4.68×10−2)−1.0054×10+0
(1.43×10−2)15 2.1614×10+0
(3.30×10−2)−1.7688×10+0
(7.55×10−2)=1.7815×10+0
(1.09×10−1)=7.5478×10+0
(2.36×10+0)−1.7084×10+0
(3.82×10−2)=1.7263×10+0
(1.03×10−1)25 4.0448×10+0
(1.16×10−2)−3.5543×10+0
(1.21×10−1)−2.8225×10+0
(1.52×10−1)−1.0109×10+1
(5.56×10+0)−3.1864×10+0
(1.60×10−1)−2.7976×10+0
(1.71×10−1)WFG3 5 7.4442×10−1
(3.28×10−2)−5.9020×10−1
(5.58×10−2)−6.8641×10−1
(1.11×10−1)−1.6953×10+0
(1.28×10−1)−6.8013×10−1
(1.54×10−1)−5.4840×10−1
(3.21×10−2)10 2.7654×10+0
(1.17×10−1)+1.4175×10+0
(4.13×10−1)+3.4602×10+0
(5.56×10−1)−8.5627×10+0
(1.89×10−1)−2.4166×10+0
(9.05×10−2)+3.0447×10+0
(5.68×10−1)15 6.4307×10+0
(6.67×10−1)+2.6293×10+0
(4.06×10−1)+6.6442×10+0
(1.19×10+0)+1.3900×10+1
(2.03×10−1)−5.5924×10+0
(2.08×10−1)+7.5080×10+0
(9.65×10−1)25 1.8160×10+1
(3.63×10−2)−1.2519×10+1
(1.40×10+0)−1.1530×10+1
(1.65×10+0)−2.7424×10+1
(3.33×10−2)−9.8933×10+0
(1.41×10−1)−6.8891×10+0
(5.11×10+0)+/−/= 7/24/1 7/22/3 7/22/3 3/27/2 11/18/3 “+”表明该算法优于MaOEA-IAR, “−”劣于MaOEA-IAR, “=”则表示与MaOEA-IAR性能相似 表 7 MOEA-DD、NSGA-III、RVEA、MOMBI-II、AR-MOEA以及MaOEA-IAR在DTLZ5~DTLZ7, IDTLZ1~IDTLZ2, WFG1~WFG3上获得PD值的统一结果(均值和标准差).最好的结果已被标记
Table 7 The statistical results (mean and standard deviation) of the PD values obtained by MOEA-DD、NSGA-III、RVEA、MOMBI-II、AR-MOEA and MaOEA-IAR on DTLZ5~DTLZ7, IDTLZ1~IDTLZ2, WFG1~WFG3. The best results are highlighted
问题 M MOEA-DD NSGA-III RVEA MOMBI-II AR-MOEA MaOEA-IAR DTLZ5 5 1.7925×10+7
(2.11×10+6)−3.5995×10+7
(4.18×10+6)+1.8932×10+7
(4.20×10+6)−1.1044×10+7
(1.59×10+6)−4.8118×10+7
(2.34×10+6)+2.5713×10+7
(2.69×10+6)10 1.9742×10+10
(2.54×10+9)−7.1635×10+10
(5.54×10+9)−2.9432×10+10
(3.78×10+9)−2.1484×10+9
(2.81×10+9)−8.3986×10+10
(3.99×10+9)−1.2087×10+11
(8.00×10+9)15 2.8397×10+11
(7.71×10+10)−1.1841×10+12
(1.38×10+11)−4.1423×10+10
(2.81×10+10)−6.5840×10+9
(4.48×10+9)−1.0895×10+12
(1.09×10+11)−2.9251×10+12
(2.88×10+11)25 4.2895×10+13
(9.57×10+12)+1.4562×10+11
(1.05×10+11)−1.7203×10+13
(1.37×10+13)+4.0293×10+10
(4.56×10+10)−1.0108×10+14
(1.23×10+13)−5.6358×10+14
(2.21×10+14)DTLZ6 5 2.8362×10+7
(3.50×10+6)−5.4971×10+7
(7.11×10+6)+2.6105×10+7
(7.26×10+6)−1.0490×10+7
(2.67×10+6)−5.2864×10+7
(3.81×10+6)+3.1205×10+7
(4.21×10+6)10 6.5293×10+10
(1.58×10+10)−1.8747×10+11
(4.95×10+10)−5.9078×10+10
(1.80×10+10)−2.1985×10+9
(2.83×10+9)−8.4598×10+10
(5.75×10+9)−3.9921×10+11
(2.70×10+10)15 9.1234×10+11
(3.10×10+11)−3.4359×10+12
(5.79×10+11)−7.4856×10+11
(3.95×10+11)−1.0372×10+10
(1.40×10+10)−4.2785×10+11
(3.48×10+11)−7.9758×10+12
(8.70×10+11)25 2.2080×10+14
(1.01×10+14)−2.2455×10+14
(1.56×10+13)−8.7904×10+13
(2.72×10+13)−1.5169×10+11
(1.15×10+11)−3.2540×10+13
(1.77×10+13)−2.3216×10+15
(1.08×10+15)DTLZ7 5 1.6766×10+3
(4.84×10+3)−2.1762×10+7
(4.10×10+6)=1.8563×10+7
(3.07×10+6)=5.1253×10+6
(7.63×10+5)−3.3243×10+7
(4.17×10+6)+2.0833×10+7
(7.02×10+6)10 2.1457×10+9
(1.38×10+9)−3.1794×10+10
(2.69×10+9)−1.6034×10+10
(4.07×10+9)−8.0445×10+9
(1.73×10+9)−2.5524×10+10
(3.65×10+9)−6.8683×10+10
(7.18×10+9)15 1.8990×10+10
(5.95×10+9)−4.7550×10+11
(9.43×10+10)−2.7698×10+11
(6.40×10+10)−4.9554×10+10
(2.76×10+10)−4.0328×10+11
(1.19×10+11)−2.5059×10+12
(3.10×10+11)25 4.2556×10+12
(1.02×10+12)−7.3834×10+13
(1.20×10+13)−4.7217×10+13
(1.00×10+13)−4.4333×10+13
(1.34×10+13)−1.3427×10+9
(6.62×10+8)−1.1453×10+14
(1.61×10+13)IDTLZ1 5 1.4745×10+6
(9.43×10+5)−6.3366×10+6
(3.40×10+6)−2.1610×10+6
(9.83×10+5)−8.4010×10+5
(2.39×10+5)−1.0739×10+7
(2.67×10+6)−1.9799×10+8
(2.32×10+8)10 1.0063×10+8
(1.13×10+8)−9.8146×10+9
(1.30×10+9)−1.0241×10+9
(9.00×10+8)−2.0726×10+9
(3.55×10+8)−7.1952×10+9
(1.15×10+9)−4.5218×10+10
(1.18×10+11)15 4.8749×10+10
(2.39×10+11)−2.7098×10+11
(4.55×10+10)−6.2342×10+10
(5.79×10+10)−3.9405×10+10
(8.72×10+9)−1.2277×10+11
(3.42×10+11)−3.2538×10+12
(1.06×10+13)25 3.8837×10+9
(8.47×10+9)−5.6426×10+13
(1.28×10+13)−7.6673×10+12
(5.54×10+12)−1.6889×10+12
(1.19×10+12)−1.6465×10+14
(4.61×10+14)−5.8266×10+14
(1.42×10+13)IDTLZ2 5 1.2432×10+7
(1.92×10+6)−3.5278×10+7
(7.22×10+6)−1.8123×10+7
(1.50×10+6)−4.8079×10+6
(6.93×10+5)−2.6241×10+7
(1.72×10+6)−5.5348×10+7
(2.41×10+6)10 3.7649×10+9
(3.38×10+8)−4.9622×10+10
(7.67×10+9)−1.1533×10+10
(1.30×10+9)−1.5701×10+10
(1.85×10+9)−9.1979×10+10
(3.71×10+9)−1.6089×10+11
(4.17×10+9)15 7.1923×10+10
(3.07×10+10)−1.0779×10+12
(1.50×10+11)−2.0643×10+11
(6.01×10+10)−2.9274×10+11
(4.07×10+10)−2.6072×10+12
(1.61×10+11)−5.4279×10+12
(1.82×10+11)25 4.4848×10+12
(1.99×10+12)−5.6171×10+14
(8.35×10+13)−1.4306×10+13
(8.54×10+12)−7.0067×10+13
(1.05×10+13)−8.5860×10+14
(4.94×10+13)−2.0550×10+15
(5.47×10+13)WFG1 5 7.7570×10+7
(4.76×10+6)+7.5733×10+7
(7.14×10+6)+6.6169×10+7
(5.02×10+6)+7.6250×10+6
(2.77×10+6)−3.0367×10+7
(2.17×10+6)+2.3345×10+7
(3.37×10+6)10 3.8144×10+10
(3.75×10+9)+8.4159×10+10
(8.64×10+9)+4.6067×10+10
(6.62×10+9)+6.4761×10+9
(1.55×10+9)−4.3604×10+10
(2.26×10+9)+1.8476×10+10
(2.80×10+9)15 1.3448×10+12
(2.24×10+11)+1.2676×10+12
(5.92×10+11)+6.2449×10+11
(1.04×10+11)=1.9369×10+10
(2.14×10+10)−6.0993×10+11
(4.26×10+10)=6.1772×10+11
(1.26×10+11)25 3.3402×10+13
(2.45×10+12)−1.6094×10+14
(4.13×10+13)+1.0889×10+14
(5.98×10+12)+4.8564×10+12
(1.42×10+12)−1.0664×10+14
(3.18×10+12)+8.4524×10+13
(1.67×10+13)WFG2 5 5.1237×10+7
(2.65×10+6)−5.4016×10+7
(3.16×10+6)−6.1875×10+7
(2.87×10+6)−1.1742×10+7
(3.17×10+6)−3.8312×10+7
(2.55×10+6)−7.6371×10+7
(4.16×10+6)10 3.9168×10+10
(1.43×10+9)−6.6429×10+10
(1.01×10+10)−5.3106×10+10
(2.41×10+9)−7.4067×10+9
(5.52×10+9)−3.9309×10+10
(2.97×10+9)−9.1141×10+10
(4.59×10+9)15 4.4512×10+11
(4.65×10+10)−1.5952×10+12
(2.02×10+11)−9.2321×10+11
(1.58×10+11)−1.2053×10+10
(1.52×10+10)−5.7677×10+11
(8.49×10+10)−2.2461×10+12
(1.65×10+11)25 4.1805×10+13
(3.61×10+12)−3.8842×10+14
(1.07×10+14)−1.7542×10+14
(2.40×10+13)−3.8888×10+11
(9.15×10+11)−1.6560×10+14
(1.30×10+13)−4.7295×10+14
(2.67×10+13)WFG3 5 1.2116×10+8
(9.07×10+6)−1.4756×10+8
(7.56×10+6)−1.3033×10+8
(1.40×10+7)−5.1396×10+7
(4.21×10+6)−1.5433×10+8
(7.28×10+6)=1.5501×10+8
(6.95×10+6)10 1.1096×10+11
(9.29×10+9)−2.4982×10+11
(2.39×10+10)−1.6581×10+11
(1.63×10+10)−9.4558×10+8
(1.51×10+8)−1.9220×10+11
(1.05×10+10)−3.1718×10+11
(2.00×10+10)15 3.8028×10+12
(7.77×10+11)−7.4227×10+12
(1.37×10+12)+4.1520×10+12
(7.83×10+11)−1.6200×10+10
(5.00×10+9)−2.7152×10+12
(9.23×10+11)−6.1382×10+12
(5.23×10+11)25 1.0941×10+14
(6.89×10+12)−1.9873×10+15
(6.87×10+14)=5.6632×10+14
(1.69×10+14)−5.1563×10+10
(1.66×10+10)−3.2575×10+14
(2.26×10+13)−1.5572×10+15
(1.06×10+14)+/−/= 4/28/0 7/23/2 4/26/2 0/32/0 6/24/2 “+”表明该算法优于MaOEA-IAR, “−”劣于MaOEA-IAR, “=”则表示与MaOEA-IAR性能相似 表 8 AR-MOEA和MaOEA-IAR在DTLZ1~DTLZ7, IDTLZ1~IDTLZ2, WFG1~WFG9上运行时间的统一结果(均值). 最好的结果已被标记
Table 8 The statistical results (mean) of the time obtained by AR-MOEA and MaOEA-IAR on DTLZ1~DTLZ7, IDTLZ1~IDTLZ2, WFG1~WFG9. The best results are highlighted
问题 M AR-MOEA MaOEA-IAR 问题 M AR-MOEA MaOEA-IAR DTLZ1 5 8.7284×10+1− 5.2877×10+1 WFG1 5 2.1497×10+2− 1.2453×10+2 10 3.2368×10+3− 1.4929×10+3 10 3.0802×10+3− 1.8838×10+3 15 1.5646×10+2+ 4.5986×10+2 15 3.3103×10+2− 2.9192×10+2 25 2.7908×10+3− 1.8621×10+3 25 2.4172×10+3+ 2.8807×10+3 DTLZ2 5 1.4045×10+2+ 3.1151×10+2 WFG2 5 2.5691×10+2− 1.9505×10+2 10 3.4692×10+3− 2.0823×10+3 10 3.6336×10+3− 3.2122×10+3 15 1.9603×10+2+ 4.9719×10+2 15 4.3556×10+2− 3.8969×10+2 25 3.1545×10+3+ 3.4599×10+3 25 3.4466×10+3= 3.3881×10+3 DTLZ3 5 7.6826×10+1− 3.3370×10+1 WFG3 5 3.4018×10+2− 3.2435×10+2 10 3.2959×10+3− 1.5038×10+3 10 4.3267×10+3− 4.1139×10+3 15 4.9418×10+2− 1.5822×10+2 15 5.1973×10+2− 4.3731×10+2 25 2.9858×10+3− 1.7848×10+3 25 3.8047×10+3= 3.7357×10+3 DTLZ4 5 1.4659×10+2+ 3.2766×10+2 WFG4 5 3.3620×10+2− 2.8286×10+2 10 3.8768×10+3− 2.1238×10+3 10 4.2819×10+3− 3.6326×10+3 15 1.9815×10+2+ 5.1050×10+2 15 3.8595×10+2+ 5.2267×10+2 25 3.1269×10+3+ 3.2684×10+3 25 3.7546×10+3− 3.3598×10+3 DTLZ5 5 2.2500×10+2− 1.3912×10+2 WFG5 5 3.3378×10+2− 2.9287×10+2 10 3.0904×10+3− 1.9358×10+3 10 4.2697×10+3− 3.7511×10+3 15 1.8227×10+2+ 4.5443×10+2 15 5.2633×10+2− 4.0438×10+2 25 3.0757×10+3− 1.9646×10+3 25 3.8177×10+3− 3.4843×10+3 DTLZ6 5 1.9591×10+2− 1.2208×10+2 WFG6 5 2.7412×10+2− 2.5095×10+2 10 3.5852×10+3− 1.8312×10+3 10 4.0843×10+3− 3.2581×10+3 15 1.5310×10+2+ 4.9933×10+2 15 3.1397×10+2+ 5.1200×10+2 25 2.6990×10+3− 2.0808×10+3 25 3.7598×10+3− 2.8335×10+3 DTLZ7 5 2.0799×10+2− 1.4207×10+2 WFG7 5 3.7091×10+2− 3.1761×10+2 10 3.8714×10+3− 2.3062×10+3 10 4.3950×10+3− 3.7082×10+3 15 4.8328×10+2− 2.0122×10+2 15 3.7417×10+2+ 5.2910×10+2 25 2.4555×10+3− 1.5033×10+3 25 3.8242×10+3− 3.2364×10+3 IDTLZ1 5 2.1931×10+2− 9.4000×10+1 WFG8 5 2.2219×10+2= 2.1770×10+2 10 3.1281×10+3− 1.9947×10+3 10 3.7129×10+3− 2.6444×10+3 15 3.4287×10+2− 2.5752×10+2 15 2.4552×10+2+ 4.5493×10+2 25 2.0253×10+3− 1.2549×10+3 25 3.7527×10+3− 2.5151×10+3 IDTLZ2 5 3.1594×10+2+ 3.3723×10+2 WFG9 5 3.8469×10+2− 3.2047×10+2 10 4.0353×10+3− 3.4738×10+3 10 4.3729×10+3− 3.9 448×10+3 15 4.4610×10+2− 4.2203×10+2 15 5.3114×10+2− 4.3118×10+2 25 2.5265×10+3− 1.7758×10+3 25 3.8142×10+3− 3.5823×10+3 +/−/= 10/26/0 +/−/= 5/28/3 “+”表明该算法优于MaOEA-IAR, “−”劣于MaOEA-IAR, “=”则表示与MaOEA-IAR性能相似 表 9 AR-MOEA和MaOEA-IAR在MaF1~MaF7上获得IGD值的统一结果(均值和标准差). 最好的结果已被标记
Table 9 The statistical results (mean and standard deviation) of the IGD values obtained by AR-MOEA and MaOEA-IAR on MaF1~MaF7. The best results are highlighted
问题 M AR-MOEA MaOEA-IAR MaF1 5 6.1773×10−2 (6.75×10−2)− 3.2287×10−3 (7.05×10−5) 10 1.0257×10−1 (1.11×10−1)− 6.5011×10−3 (1.44×10−4) 15 1.6987×10−1 (1.85×10−1)− 1.1253×10−2 (2.95×10−4) 25 1.7434×10−1 (1.89×10−1)− 9.2626×10−3 (1.97×10−4) MaF2 5 5.6263×10−2 (5.26×10−2)− 8.5071×10−3 (3.31×10−4) 10 8.3280×10−2 (8.77×10−2)− 6.4968×10−3 (9.61×10−5) 15 1.2184×10−1 (1.24×10−1)− 8.9506×10−3 (2.26×10−4) 25 1.0515×10−1 (1.21×10−1)− 6.2938×10−3 (4.48×10−4) MaF3 5 3.4660×10−2 (4.22×10−2)− 1.0505×10−3 (8.53×10−4) 10 3.4304×10−2 (4.06×10−2)− 4.8646×10−3 (8.72×10−3) 15 1.6992×10+0 (6.87×10+0)− 3.9708×10−3 (3.15×10−3) 25 8.5024×10−1 (2.47×10+0)− 2.0315×10−3 (2.00×10−3) MaF4 5 1.0234×10+0 (1.26×10+0)− 4.5541×10−2 (9.87×10−3) 10 3.6718×10+1 (4.35×10+1)− 1.7669×10+0 (1.34×10−2) 15 1.7572×10+3 (2.01×10+3)− 2.7158×10+2 (5.78×10−1) 25 1.8804×10+6 (2.17×10+6)− 1.7062×10+5 (3.15×10+2) MaF5 5 1.0214×10+0 (1.29×10+0)− 6.3123×10−2 (1.23×10−2) 10 3.7332×10+1 (4.78×10+1)− 6.3090×10−1 (9.64×10−2) 15 1.4852×10+3 (1.90×10+3)− 3.6171×10+0 (1.29×10+0) 25 1.7966×10+6 (2.32×10+6)− 7.1131×10+0 (8.87×10+0) MaF6 5 1.3381×10−3 (1.72×10−3)− 4.2532×10−6 (1.43×10−6) 10 3.1054×10−1 (1.30×10+0)+ 4.8255×10+0 (3.33×10+0) 15 2.1990×10−1 (7.89×10−1)− 4.6076×10−6 (1.37×10−6) 25 3.5425×10−1 (5.51×10−1)− 3.7808×10−6 (4.08×10−6) MaF7 5 1.3383×10−1 (1.61×10−1)= 1.8586×10−1 (2.35×10−1) 10 5.7112×10−1 (6.97×10−1)= 3.5443×10−1 (4.13×10−1) 15 2.2371×10+0 (2.25×10+0)− 7.1977×10−1 (7.75×10−1) 25 3.6581×10+0 (3.86×10+0)− 1.1867×10−1 (8.44×10−1) +/−/= 1/25/2 “+”表明该算法优于MaOEA-IAR, “−”劣于MaOEA-IAR, “=”则表示与MaOEA-IAR性能相似 表 10 不同参数下算法MaOEA-IAR的性能
Table 10 Performance of MaOEA-IAR under different parameters
问题 目标数 r初始值 5 7 9 11 13 DTLZ1 5 2.6870×10−1 2.6135×10−1 2.6051×10−1 2.6382×10−1 2.6351×10−1 10 4.8177×10+1 4.7009×10+1 4.6085×10+1 4.7519×10+1 4.7286×10+1 15 5.7026×10+1 5.6981×10+1 5.5765×10+1 5.6146×10+1 5.6083×10+1 25 4.5156×10−1 4.5013×10−1 3.6557×10−1 2.4067×10−1 2.4052×10−1 DTLZ5 5 9.2396×10−2 8.3615×10−2 8.0045×10−2 8.8463×10−2 8.8732×10−2 10 6.2597×10−1 4.7625×10−1 4.7044×10−1 4.7765×10−1 4.7829×10−1 15 1.3549×10−1 1.3158×10−1 1.2994×10−1 1.3036×10−1 1.3023×10−1 25 3.3464×10−2 3.3391×10−2 3.3301×10−2 3.3362×10−2 3.3485×10−2 IDTLZ1 5 2.5128×10−1 2.3743×10−1 2.1694×10−1 2.2748×10−1 2.2926×10−1 10 1.1236×10−1 1.1149×10−1 1.0017×10−1 1.0594×10−1 1.0542×10−1 15 1.8035×10−1 1.7891×10−1 1.7077×10−1 1.7901×10−1 1.7924×10−1 25 2.2693×10−2 2.2451×10−2 2.2279×10−2 2.2538×10−2 2.2523×10−2 WFG1 5 4.4951×10−1 4.3826×10−1 4.3007×10−1 4.3596×10−1 4.4001×10−1 10 1.1967×10+0 1.1051×10+0 1.0186×10+0 1.0598×10+0 1.0482×10+0 15 1.6956×10+0 1.5941×10+0 1.5076×10+0 1.6582×10+0 1.6677×10+0 25 2.8015×10+0 2.7693×10+0 2.6724×10+0 2.7795×10+0 2.7619×10+0 WFG2 5 5.4316×10−1 5.4029×10−1 5.4002×10−1 5.4174×10−1 5.4169×10−1 10 1.1086×10+0 1.0997×10+0 1.0051×10+0 1.1036×10+0 1.1097×10+0 15 1.8675×10+0 1.7649×10+0 1.7069×10+0 1.8007×10+0 1.7952×10+0 25 2.9598×10+0 2.9413×10+0 2.8039×10+0 2.8796×10+0 2.8976×10+0 WFG4 5 1.1192×10+0 1.1128×10+0 1.1065×10+0 1.1097×10+0 1.1106×10+0 10 4.1062×10+0 4.1057×10+0 4.0388×10+0 4.0263×10+0 4.0589×10+0 15 8.9645×10+0 8.8089×10+0 8.3986×10+0 8.4108×10+0 8.4004×10+0 25 1.9058×10+1 1.8769×10+1 1.6506×10+1 1.5375×10+1 1.6405×10+1 -
[1] Luo J P, Liu Q Q, Yang Y, Li X, Chen M R, Cao W M. An artificial bee colony algorithm for multi-objective optimisation. Applied Soft Computing, 2017, 50: 235-251 doi: 10.1016/j.asoc.2016.11.014 [2] Zhou Y L, Wang J H, Chen J, Gao S C, Teng L Y. Ensemble of many-objective evolutionary algorithms for many-objective problems. Soft Computing, 2017, 21(9): 2407-2419 doi: 10.1007/s00500-015-1955-3 [3] Jain H, Deb K. An evolutionary many-objective optimization algorithm using reference-point based nondominated sorting approach, part II: Handling constraints and extending to an adaptive approach. IEEE Transactions on Evolutionary Computation, 2014, 18(4): 602-622 doi: 10.1109/TEVC.2013.2281534 [4] Cinalli D, Martí L, Sanchez-Pi N, Garcia A C B. Collective intelligence approaches in interactive evolutionary multi-objective optimization. Logic Journal of the IGPL, 2020, 28(1): 95-108 doi: 10.1093/jigpal/jzz074 [5] Kudikala R, Mills A R, Fleming P J, Tanner G F, Holt J E. Aero engine health management system architecture design using multi-criteria optimization. In: Proceedings of the 15th Annual Conference Companion on Genetic and Evolutionary Computation. Amsterdam, The Netherlands: ACM, 2013. 185−186 [6] Saeidi A M, Hage J, Khadka R, Jansen S. A search-based approach to multi-view clustering of software systems. In: Proceedings of the 22nd IEEE International Conference on Software Analysis, Evolution, and Reengineering. Montreal, Canada: IEEE, 2015. 429−438 [7] 陈振兴, 严宣辉, 吴坤安, 白猛. 融合张角拥挤控制策略的高维多目标优化. 自动化学报, 2015, 41(6): 1145-1158Chen Zhen-Xing, Yan Xuan-Hui, Wu Kun-An, Bai Meng. Many-objective optimization integrating open angle based congestion control strategy. Acta Automatica Sinica, 2015, 41(6): 1145-1158 [8] Liu Y, Zhu N B, Li K L, Li M Q, Zheng J H, Li K Q. An angle dominance criterion for evolutionary many-objective optimization. Information Sciences, 2020, 509: 376-399 doi: 10.1016/j.ins.2018.12.078 [9] Yang S X, Li M Q, Liu X H, Zheng J H. A grid-based evolutionary algorithm for many-objective optimization. IEEE Transactions on Evolutionary Computation, 2013, 17(5): 721-736 doi: 10.1109/TEVC.2012.2227145 [10] 余伟伟, 谢承旺, 闭应洲, 夏学文, 李雄, 任柯燕, 等. 一种基于自适应模糊支配的高维多目标粒子群算法. 自动化学报, 2018, 44(12): 2278-2289Yu Wei-Wei, Xie Cheng-Wang, Bi Ying-Zhou, Xia Xue-Wen, Li Xiong, Ren Ke-Yan, et al. Many-objective particle swarm optimization based on adaptive fuzzy dominance. Acta Automatica Sinica, 2018, 44(12): 2278-2289 [11] Elarbi M, Bechikh S, Gupta A, Said L B, Ong Y S. A new decomposition-based NSGA-II for many-objective optimization. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2018, 48(7): 1191-1210 doi: 10.1109/TSMC.2017.2654301 [12] Zhang Q F, Li H. MOEA/D: A multiobjective evolutionary algorithm based on decomposition. IEEE Transactions on Evolutionary Computation, 2007, 11(6): 712-731 doi: 10.1109/TEVC.2007.892759 [13] Wang L P, Xu M N, Yu W, Qiu Q C, Wu F. Decomposition multi-objective evolutionary algorithm based on adaptive neighborhood adjustment strategy. IEEE Access, 2020, 8: 78639-78651. doi: 10.1109/ACCESS.2020.2990193 [14] Cheng R, Jin Y C, Olhofer M, Sendhoff B. A reference vector guided evolutionary algorithm for many-objective optimization. IEEE Transactions on Evolutionary Computation, 2016, 20(5): 773-791 doi: 10.1109/TEVC.2016.2519378 [15] Lu X, Tan Y Y, Zheng W, Meng L L. A decomposition method based on random objective division for MOEA/D in many-objective optimization. IEEE Access, 2020, 8: 103550-103564 doi: 10.1109/ACCESS.2020.2999417 [16] Deb K, Jain H. An evolutionary many-objective optimization algorithm using reference-point-based nondominated sorting approach, part I: Solving problems with box constraints. IEEE Transactions on Evolutionary Computation, 2014, 18(4): 577-601 doi: 10.1109/TEVC.2013.2281535 [17] 封文清, 巩敦卫. 基于在线感知Pareto前沿划分目标空间的多目标进化优化. 自动化学报, 2020, 46(8): 1628-1643Feng Wen-Qing, Gong Dun-Wei. Multi-objective evolutionary optimization with objective space partition based on online perception of Pareto front. Acta Automatica Sinica, 2020, 46(8): 1628-1643 [18] Beume N, Naujoks B, Emmerich M. SMS-EMOA: Multiobjective selection based on dominated hypervolume. European Journal of Operational Research, 2007, 181(3): 1653-1669 doi: 10.1016/j.ejor.2006.08.008 [19] Gómez R H, Coello C A C. Improved metaheuristic based on the R2 indicator for many-objective optimization. In: Proceedings of the 2015 Annual Conference on Genetic and Evolutionary Computation. Madrid, Spain: ACM, 2015. 679−686 [20] Liang Z P, Luo T T, Hu K F, Ma X L, Zhu Z X. An indicator-based many-objective evolutionary algorithm with boundary protection. IEEE Transactions on Cybernetics, 2021, 51(9): 4553-4566 doi: 10.1109/TCYB.2019.2960302 [21] Li F, Li T J, Zhang S N. R2 indicator and objective space partition based evolutionary algorithm for many-objective optimization. In: Proceedings of the 2019 IEEE Symposium Series on Computational Intelligence. Xiamen, China: IEEE, 2019. 1271−1278 [22] Sun Y A, Yen G G, Yi Z. IGD indicator-based evolutionary algorithm for many-objective optimization problems. IEEE Transactions on Evolutionary Computation, 2019, 23(2): 173-187 doi: 10.1109/TEVC.2018.2791283 [23] Hua Y C, Liu Q Q, Hao K R, Jin Y C. A survey of evolutionary algorithms for multi-objective optimization problems with irregular Pareto fronts. IEEE/CAA Journal of Automatica Sinica, 2021, 8(2): 303-318 doi: 10.1109/JAS.2021.1003817 [24] Ma X L, Yu Y A, Li X D, Qi Y T. A survey of weight vector adjustment methods for decomposition-based multiobjective evolutionary algorithms. IEEE Transactions on Evolutionary Computation, 2020, 24(4): 634-649 doi: 10.1109/TEVC.2020.2978158 [25] Yuan J W, Liu H L, Gu F Q, Zhang Q F, He Z S. Investigating the properties of indicators and an evolutionary many-objective algorithm using promising regions. IEEE Transactions on Evolutionary Computation, 2021, 25(1): 75-86 doi: 10.1109/TEVC.2020.2999100 [26] Liu S B, Lin Q Z, Tan K C, Gong M G, Coello C A C. A fuzzy decomposition-based multi/many-objective evolutionary algorithm. IEEE Transactions on Cybernetics, to be published [27] Van D A, Gary V, Lamont B. Multiobjective evolutionary algorithm research: A history and analysis. EvolutionaryComputation, 1998, 8(2): 125−147 [28] Zhou A M, Jin Y C, Zhang Q F, Sendhoff B, Tsang E. Combining model-based and genetics-based offspring generation for multi-objective optimization using a convergence criterion. In: Proceedings of the 2006 IEEE International Conference on Evolutionary Computation. Vancouver, Canada: IEEE, 2006. 892−899 [29] While L, Hingston P, Barone L, Huband S. A faster algorithm for calculating hypervolume. IEEE Transactions on Evolutionary Computation, 2006, 10(1): 29-38 doi: 10.1109/TEVC.2005.851275 [30] Liu C, Zhao Q, Yan B, Gao Y. A new hypervolume-based differential evolution algorithm for many-objective optimization. RAIRO-Operations Research, 2017, 51(4): 1301-1315 doi: 10.1051/ro/2017014 [31] Ishibuchi H, Imada R, Masuyama N, Nojima Y. Dynamic specification of a reference point for hypervolume calculation in SMS-EMOA. In: Proceedings of the 2018 IEEE Congress on Evolutionary Computation. Rio de Janeiro, Brazil: IEEE, 2018. 1−8 [32] Brockhoff D, Wagner T, Trautmann H. On the properties of the R2 indicator. In: Proceedings of the 14th Annual Conference on Genetic and Evolutionary Computation. Philadelphia, USA: ACM, 2012. 465−472 [33] Li F, Liu J C, Song Y X, Shang L L. An adaptive evolutionary multi-objective algorithm based on R2 indicator. In: Proceedings of the 2019 Chinese Control and Decision Conference. Nanchang, China: IEEE, 2019. 692−697 [34] Wang H D, Jin Y C, Yao X. Diversity assessment in many-objective optimization. IEEE Transactions on Cybernetics, 2017, 47(6): 1510-1522 doi: 10.1109/TCYB.2016.2550502 [35] Tian Y, Zhang X Y, Cheng R, Jin Y C. A multi-objective evolutionary algorithm based on an enhanced inverted generational distance metric. In: Proceedings of the 2016 IEEE Congress on Evolutionary Computation. Vancouver, Canada: IEEE, 2016. 5222−5229 [36] Tian Y, Cheng R, Zhang X Y, Cheng F, Jin Y C. An indicator-based multiobjective evolutionary algorithm with reference point adaptation for better versatility. IEEE Transactions on Evolutionary Computation, 2018, 22(4): 609-622 doi: 10.1109/TEVC.2017.2749619 [37] Deb K, Agrawal R B. Simulated binary crossover for continuous search space. Complex Systems, 1995, 9(3): 115-148 [38] Deb K, Goyal M. A combined genetic adaptive search (GeneAS) for engineering design. Computer Science and Informatics, 1996, 26(4): 30-45 [39] Das I, Dennis J E. Normal-boundary intersection: A new method for generating the Pareto surface in nonlinear multicriteria optimization problems. Siam Journal on Optimization, 1998, 8(3): 631-657 doi: 10.1137/S1052623496307510 [40] Li M Q, Zhen L L, Yao X. How to read many-objective solution sets in parallel coordinates. IEEE Computational Intelligence Magazine, 2017, 12(4): 88-100 doi: 10.1109/MCI.2017.2742869