Hybrid algorithm for aero-engine model solving based on Levenberg-Marquardt algorithm
-
摘要:
为了降低航空发动机非线性模型求解的收敛性要求,将模型非线性方程组的求解问题转化为最小二乘问题,提出了基于Levenberg-Marquardt(L-M)算法的混合算法。为了使L-M算法跳出局部解,混合算法使用动力学方法修正局部解;为了提高计算效率,利用Broyden拟牛顿法加速L-M算法。以涡扇发动机为研究对象,应用混合算法、L-M算法、牛顿法和Broyden拟牛顿法进行稳态和瞬态仿真。结果表明:在稳态工况下,L-M算法和混合算法收敛范围更大,在随机初值条件下能达到90%以上的收敛率,远高于牛顿法和Broyden拟牛顿法不到20%的收敛率,且混合算法计算速度与Broyden拟牛顿法相当。在瞬态工况下, L-M算法和混合算法能够在牛顿法和Broyden拟牛顿法都不收敛的强瞬变工况收敛,且混合算法瞬态计算时间仅为Broyden拟牛顿法的1.13倍。仿真结果表明该算法在航空发动机模型求解上具有良好的适用性。
-
关键词:
- 非线性模型 /
- 牛顿法 /
- Broyden拟牛顿法 /
- Levenberg-Marquardt算法 /
- 混合算法
Abstract:In order to reduce the convergence requirements for solving the nonlinear model of aeroengine, the problem of solving the model nonlinear equations was transformed into the least square problem. A hybrid algorithm based on Levenberg-Marquardt (L-M) algorithm was proposed. The hybrid algorithm avoided the local solution by modifying the local solution with dynamic method. Meanwhile, Broyden quasi-Newton method was used to accelerate the L-M algorithm. Targeting turbofan engine, the hybrid algorithm, L-M algorithm, Newton method and Broyden quasi-Newton method were used for steady-state and transient simulation. Results showed that: under the steady-state condition, if L-M algorithm and hybrid algorithm had larger convergence range, the convergence rate can reach more than 90% under the random initial value condition, much higher than the convergence rate of Newton method and Broyden quasi-Newton method which was less than 20%, and the calculation speed of hybrid algorithm was similar to that of Broyden quasi-Newton method. Under transient condition, L-M algorithm and hybrid algorithm can converge at strong transient condition where neither Newton method nor Broyden quasi-Newton method converged, and the transient computation time of hybrid algorithm was only 1.13 times that of Broyden quasi-Newton method. Simulation results show that the algorithm has good applicability in solving aero-engine model solving.
-
表 1 迭代变量范围
Table 1. Range of iteration variables
迭代变量 最小值 最大值 ${\,\beta _{ {\text{fan} } } }$ 0 1 ${\,\beta _{ {\text{lpc} } } }$ 0 1 ${\,\beta _{ {\text{hpc} } } }$ 0 1 ${\,\beta _{ {\text{hpt} } } }$ 0 1 ${\,\beta _{ {\text{lpt} } } }$ 0 1 ${n_1}$ 0.001 1.2 ${n_2}$ 0.001 1.2 表 2 最大推力工况(H=0 m,Ma=0)
Table 2. Maximum thrust operation point (H=0 m,Ma=0)
${R_{{\text{el}}}}$值 牛顿法 Broyden拟牛顿法 L-M算法 混合算法 是否收敛 调动次数 是否收敛 调动次数 是否收敛 调动次数 是否收敛 调动次数 −0.1(0.1) 是(是) 61(76) 是(是) 24(23) 是(是) 76(91) 是(是) 24(24) −0.2(0.2) 是(是) 91(91) 是(是) 31(24) 是(是) 91(106) 是(是) 43(37) −0.3(0.3) 是(是) −(106) 否(是) −(27) 是(是) 121(121) 是(是) 40(53) −0.4(0.4) 否(否) 否(是) −(32) 是(是) 121(136) 是(是) 45(68) −0.5(0.5) 否(否) 否(否) 是(是) 136(151) 是(是) 57(84) −0.6(0.6) 否(否) 否(否) 是(是) 196(151) 是(是) 86(85) −0.7(0.7) 否(否) 否(否) 是(是) 256(181) 是(是) 116(100) −0.8(0.8) 否(否) 否(否) 是(是) 256(181) 是(是) 148(116) 表 3 巡航工况(H=11000 m,Ma=0.8)
Table 3. Cruising operating point (H=11000 m,Ma=0.8)
${R_{{\text{el}}}}$值 牛顿法 Broyden拟牛顿法 L-M算法 混合算法 是否收敛 调动次数 是否收敛 调动次数 是否收敛 调动次数 是否收敛 调动次数 −0.1(0.1) 是(是) 61(61) 是(是) 22(22) 是(是) 91(61) 是(是) 23(22) −0.2(0.2) 是(是) 61(61) 是(是) 27(23) 是(是) 106(91) 是(是) 27(23) −0.3(0.3) 是(是) 91(91) 是(是) 36(26) 是(是) 121(106) 是(是) 35(39) −0.4(0.4) 是(是) 91(121) 是(是) 56(34) 是(是) 136(121) 是(是) 46(54) −0.5(0.5) 否(否) 否(否) 是(是) 151(151) 是(是) 64(69) −0.6(0.6) 否(否) 否(否) 是(是) 151(166) 是(是) 74(85) −0.7(0.7) 否(否) 否(否) 是(是) 166(181) 是(是) 92(100) −0.8(0.8) 否(否) 否(否) 是(是) 151(196) 是(是) 117(116) 表 4 随机输入计算结果对比
Table 4. Comparison of random input calculation results
样本数 方法 最大部件
调动次数最小部件
调动次数平均部件
调动次数中断数 不收敛解
个数收敛解
个数100 牛顿法 136 76 100 81 0 19 Broyden拟牛顿法 50 27 35 86 7 7 L-M算法 391 106 200 5 4 91 混合算法 375 38 102 5 0 95 1000 L-M算法 496 91 213 59 38 903 混合算法 499 23 112 60 4 936 表 5 总计算时间和相对计算时间(工况 2)
Table 5. Total computation time and relative computation time (case 2)
计算方法 总时间/s 相对计算时间 牛顿法 13.80 1.43 Broyden拟牛顿法 9.66 1 L-M算法 16.92 1.75 混合算法 10.92 1.13 -
[1] 施洋,屠秋野,严红明,等. 一种航空发动机全状态性能模型[J]. 航空动力学报,2017,32(2): 373-381.SHI Yang,TU Qiuye,YAN Hongming,et al. A full states performance model for area engine[J]. Journal of Aerospace Power,2017,32(2): 373-381. (in Chinese) [2] 曹建国. 航空发动机仿真技术研究现状、挑战和展望[J]. 推进技术,2018,39(5): 961-970.CAO Jianguo. Status, challenges and perspectives of aero-engine simulation technology[J]. Journal of Propulsion Technology,2018,39(5): 961-970. (in Chinese) [3] 周文祥,黄金泉,窦建平. 涡扇发动机部件级起动模型[J]. 航空动力学报,2006,21(2): 248-253. doi: 10.3969/j.issn.1000-8055.2006.02.004ZHOU Wenxiang,HUANG Jinquan,DOU Jianping. Development of component-level startup model for a turbofan engine[J]. Journal of Aerospace Power,2006,21(2): 248-253. (in Chinese) doi: 10.3969/j.issn.1000-8055.2006.02.004 [4] 苟学中,周文祥,黄金泉. 变循环发动机部件级建模技术[J]. 航空动力学报,2013,28(1): 104-111.GOU Xuezhong,ZHOU Wenxiang,HUANG Jinquan. Compinent-level modeling technology for variable cycle engine[J]. Journal of Aerospace Power,2013,28(1): 104-111. (in Chinese) [5] 陈玉春,徐思远,杨云铠. 改善航空发动机特性计算收敛性的方法[J]. 航空动力学报,2008,23(12): 2242-2248. doi: 10.13224/j.cnki.jasp.2008.12.027CHEN Yuchun,XU Siyuan,YANG Yunkai. Research on the method to solve convergence problem in aero turbo engine performance computation[J]. Journal of Aerospace Power,2008,23(12): 2242-2248. (in Chinese) doi: 10.13224/j.cnki.jasp.2008.12.027 [6] 陈玉春,徐思远,屠秋野,等. 求解航空发动机非线性方程组的变步长牛顿法[J]. 航空计算技术,2009,39(1): 39-41. doi: 10.3969/j.issn.1671-654X.2009.01.010CHEN Yuchun,XU Siyuan,TU Qiuye,et al. Solving aeroengine nonlinear equations based on variable step size Newton method[J]. Aeronautical Computing Technique,2009,39(1): 39-41. (in Chinese) doi: 10.3969/j.issn.1671-654X.2009.01.010 [7] 王元,李秋红,黄向华. 基于自校正Broyden拟牛顿法的航空发动机模型数值计算[J]. 航空动力学报,2016,31(1): 249-256.WANG Yuan,LI Qiuhong,HUANG Xianghua. Numerical calculation of area-engine model based on self-tuning Broyden quasi-Newton method[J]. Journal of Aerospace Power,2016,31(1): 249-256. (in Chinese) [8] 朱正琛,李秋红,王元,等. 基于微分进化算法的航空发动机模型修正[J]. 航空发动机,2016,42(1): 53-58.ZHU Zhengchen,LI Qiuhong,WANG Yuan,et al. Correction of aeroengine model based on differential evolution algorithm[J]. Aeroengine,2016,42(1): 53-58. (in Chinese) [9] 苏三买. 遗传算法及其在航空发动机非线性数学模型中的应用研究[D]. 西安: 西北工业大学, 2002.SU Sanmai. Research on genetic algorithm and its application in nonlinear mathematical model of aeroengine[D]. Xi’an: Northwestern Polytechnical University, 2002. (in Chinese) [10] 苏三买,廉小纯. 遗传算法在航空发动机非线性数学模型中的应用[J]. 推进技术,2004(3): 237-240. doi: 10.3321/j.issn:1001-4055.2004.03.012SU Sanmai,LIAN Xiaochun. Application of genetic algorithm in aeroengine nonlinear mathematical models[J]. Journal of Propulsion Technology,2004(3): 237-240. (in Chinese) doi: 10.3321/j.issn:1001-4055.2004.03.012 [11] 钱海鹰,杨培源,徐松林. 基于粒子群算法的发动机部件模型求解[J]. 推进技术,2012,33(6): 974-980. doi: 10.13675/j.cnki.tjjs.2012.06.019QIAN Haiying,YANG Peiyuan,XU Songlin. Application of particle swarm optimization in obtaining solution of aeroengine component-level model[J]. Journal of Propulsion Technology,2012,33(6): 974-980. (in Chinese) doi: 10.13675/j.cnki.tjjs.2012.06.019 [12] 黄飞燕, 黄洲荣, 陈洪骏. 基于粒子群算法的变循环发动机部件法建模及优化[J]. 数学的实践与认识, 2014, 44(14): 107-116.HUANG Feiyan, HUANG Zhourong, CHEN Hongjun. Application of particle swarm optimization on modeling and optimization of variable cycle engine component-level model[J]. Mathematical in Practice and Theory, 2014, 44(14): 107-116. (in Chinese) [13] 郝旺,王占学,张晓博,等. 基于自适应差分进化算法的变循环发动机模型求解方法研究[J]. 推进技术,2021,42(9): 1-11.HAO Wang,WANG Zhanxue,ZHANG Xiaobo,et al. Solving variable cycle engine model based on adaptive differential evolution algorithm[J]. Journal of Propulsion Technology,2021,42(9): 1-11. (in Chinese) [14] 苏三买,陈永琴. 基于混合遗传算法的航空发动机数学模型解法[J]. 推进技术,2007(6): 661-664. doi: 10.3321/j.issn:1001-4055.2007.06.015SU Sanmai,CHEN Yongqin. Hybrid genetic algorithm in solving aeroengine nonlinear mathematical model[J]. Journal of Propulsion Technology,2007(6): 661-664. (in Chinese) doi: 10.3321/j.issn:1001-4055.2007.06.015 [15] 王军,隋岩峰. 求解航空发动机数学模型的迭代算法及其改进算法的收敛性研究[J]. 系统仿真学报,2014,26(2): 310-314.WANG Jun,SUI Yanfeng. Study on convergence and improvement of iteration methods in aero-engine mathematical models[J]. Journal of System Simulation,2014,26(2): 310-314. (in Chinese) [16] 白洋,段黎明,柳林,等. 基于改进的混合粒子群算法的变循环发动机模型求解[J]. 推进技术,2014,35(12): 1694-1700.BAI Yang,DUAN Liming,LIU Lin,et al. Solving variable cycle engine model based on improved hybrid particle swarm optimization[J]. Journal of Propulsion Technology,2014,35(12): 1694-1700. (in Chinese) [17] 肖红亮,李华聪,李嘉,等. 基于QPSO混合算法的变循环发动机建模方法[J]. 北京航空航天大学学报,2018,44(2): 305-315.XIAO Hongliang,LI Huachong,LI Jia,et al. Modeling method of variable cycle engine based on QPSO hybrid algorithm[J]. Journal of Beijing University of Aeronautics and Astronautics,2018,44(2): 305-315. (in Chinese) [18] 唐世建,童万军. 基于容积法的某涡扇发动机动态建模方法[J]. 航空动力学报,2013,28(3): 709-713. doi: 10.13224/j.cnki.jasp.2013.03.032TANG Shijian,TONG Wanjun. Turbofan engine transient modeling based on inter-component volume method[J]. Journal of Aerospace Power,2013,28(3): 709-713. (in Chinese) doi: 10.13224/j.cnki.jasp.2013.03.032 [19] 黄开明,黄金泉,刘杰,等. 基于无迭代解法的航空发动机实时模型[J]. 航空发动机,2004(2): 35-38. doi: 10.3969/j.issn.1672-3147.2004.02.009HUANG Kaiming,HUANG Jinquan,LIU Jie,et al. An aeroengine real rime model based on non-iterative method[J]. Aeroengine,2004(2): 35-38. (in Chinese) doi: 10.3969/j.issn.1672-3147.2004.02.009 [20] FAWKE A J,SARAVANAMUTTOO H. Digital computer simulation of the dynamic response of a twinspool turbofan with mixed exhaust[J]. Aeronautical Journal,1973,77(753): 471-478. [21] 杨刚,孙健国,黄向华,等. 一种不需要迭代的发动机辅助变量建模方法[J]. 航空动力学报,2003,18(2): 289-294. doi: 10.3969/j.issn.1000-8055.2003.02.026YANG Gang,SUN Jianguo,HUANG Xianghua,et al. A non-iterative method of aero-engine modeling using complementary variables[J]. Journal of Aerospace Power,2003,18(2): 289-294. (in Chinese) doi: 10.3969/j.issn.1000-8055.2003.02.026 [22] RAHMAN N U,WHIDBORNE J F. Real-time transient three spool turbofan engine simulation: a hybrid approach[J]. Journal of Engineering for Gas Turbines and Power,2009,131(5): 051602.1-051602.8. [23] ROBERTS R A,EASTBOURN S M. Modeling techniques for a computational efficient dynamic turbofan engine model[J]. International Journal of Aerospace Engineering,2014,2014: 283479.1-283479.11. [24] SANGHI V,LAKSHMANAN B K. Timescaling control volumes during real time digital simulation of a military turbofan[J]. Journal of Propulsion and Power,2012,18(4): 964-966. [25] FAN J Y. The modified Levenberg-Marquardt method for nonlinear equations with cubic convergence[J]. Mathematics of Computation,2012,81(277): 447-466. doi: 10.1090/S0025-5718-2011-02496-8 [26] FAN J Y. Accelerating the modified Levenberg-Marquardt method for nonlinear equations[J]. Mathematics of Computation,2014,83(287): 1173-1187. [27] CHEN L. A high-order modified Levenberg-Marquardt method for systems of nonlinear equations with fourth-order convergence[J]. Applied Mathematics and Computation,2016,285: 79-93. doi: 10.1016/j.amc.2016.03.031 [28] YANG X. A higher-order Levenberg-Marquardt method for nonlinear equations[J]. Applied Mathematics and Computation,2013,219(22): 10682-10694. doi: 10.1016/j.amc.2013.04.033 [29] 尹大伟. 航空发动机模型求解算法及性能寻优控制中的参数估计研究[D]. 长沙: 国防科学技术大学, 2011.YI Dawei. Algorithms for solving aeroengine nonlinear mathematical model and parameter estimation in performance-seeking control[D]. Changsha: Graduate School of National University of Defense Technology, 2011. (in Chinese) [30] LU J,LU F,HUANG J. Performance estimation and fault diagnosis based on Levenberg-Marquardt algorithm for a turbofan engine[J]. Energies,2018,11(1): 181-198. doi: 10.3390/en11010181 [31] 范金燕, 袁亚湘. 非线性方程组数值解法[M]. 北京: 科学出版社, 2018. [32] FAN J Y,YUAN Y X. On the quadratic convergence of the Levenberg-Marquardt method without non-singularity assumption[J]. Computing,2005,74(1): 23-39. doi: 10.1007/s00607-004-0083-1