Multi-objective Optimization Algorithm Based On Decomposition For Faulty Parameter Interval Estimation Of Analog Components

With the working time of the analog circuit increases,the health status of its circuit components is also declining.Timely parameter estimation of components in early fault state can accurately evaluate the health status of equipment,and provide reference for prognostics and health management.At present,most analog circuit fault diagnosis is fault detection and location.There are very few methods can handle the faulty parameter estimation.Based on the transfer function of the circuit under test and the measured faulty response,the possible fault parameter can be deduced reversely.Due to tolerances in the analog components,there are theoretically many parameter combinations that can produce the same fault response.Therefore,the possible fault parameter is an interval rather than a single specific value in fault location.In summary,the faulty parameter interval estimation problem is the research object in this thesis.It is carried out by a multiobjective evolutionary algorithm based on decomposition(MOEA/D).The specific work is as follows.(1)The faulty parameter interval estimation problem is transformed into a multiobjective optimization problem.First,it is transformed into an equality constrained optimization problem by circuit analysis.The optimization objectives are upper and lower bounds.The equality constraint is that the simulated response is equal to the measured fault response.Traditionally,the equality constraint is transformed into an inequality constraint by predefining a positive tolerance value ?.But,it is known that large ?decreases the accuracy while small ? likely result in local optima.Therefore,this thesis transforms equality constraints into an optimization objective and constructs a multiobjective problem.Then,it is solved by corresponding optimization algorithms.(2)Research on the setting method of reference points.Considering the particularity of the research problem,this thesis first analyzes the reference point setting method.Then a logarithmic distribution reference points(LDRP)is proposed.Compared with linear coordinates,logarithmic coordinates have a wider coverage and focus more on the feasible region.At the same time,using the LDRP to deal with equality constraints does not need to predefine the tolerance value ?.It provides an effective new idea for solving equality constraints.(3)A multi-objective evolutionary algorithm based on decomposition with the logarithmic distribution reference points(MOEA/D-LDRP)is proposed.By combining the LDRP with MOEA/D and deriving an adaptive normalization method and an environment selection based on reference point guidance.Then,it accurately and efficiently solves the faulty parameter estimation.Finally,the experimental results show that the proposed algorithm is more accurate and faster than the existing algorithms.(4)The evolutionary operator of the matching problem is used to further optimize the algorithm.In order to improve the accuracy and speed of the algorithm for parameter estimation when dealing with complex circuits.First,the sine and cosine operators are improved by a teacher-supervised learning strategy,and replace the genetic operator.Then,a decomposed multi-objective evolutionary algorithm based on the modified sine cosine operator(MOEA/D-m SCA)is proposed.Finally,the algorithm performance is verified by experiments,and compared with the other algorithms.Meanwhile,the feasibility and correctness of the improved method are proved.At the same time,the cosimulation model of the Simulink and Pspice is built to realize the estimation of the fault parameter interval of the nonlinear analog circuit.It verifies the universality of the proposed algorithm.