The Study Of Analog Circuit Variable Structure Differential Evolution And Its Scalability Problem

Analog circuit design by evolution algorithm is an approach for automatic design analog circuit. Analog circuit evolutionary design can be divided into variable structure design and non-variable structure design, according to whether change the circuit topology. Variable structure evolution is more commonly used and more complex than non-variable structure evolution. This thesis is aimed at studying the problem of analog circuit variable structure evolution. The scalability is an important task for circuit evolutionary design. With the increasing scale and complexity of the circuits, the connection modes and the parameter values of circuit devices increase, resulting in the search space of evolutionary design increases exponentially, and eventually influences the performance of evolutionary design. In order to improve the scalability of analog circuit variable structure evolution, the evolution algorithm should still have high search efficiency when the search space increases exponentially. Whether the evolution algorithm is suitable for the problem itself will influence the search efficiency. In order to make the algorithm suitable for circuit variable structure evolution problem, this thesis will take the evolution algorithm itself for analysis, and make the algorithm self-adaptive. After that, the efficiency of the algorithm and the evolutionary performance are improved, resulting in the improvement of the scalability of evolution design.The main works of this thesis are as follows:1. Propose a variable structure evolution algorithm for analog circuit automatic design based on differential evolution. In this algorithm, the so called group crossover operation is used to make the algorithm search target circuits in a variable component space rather than in a fixed component space. This characteristic of group crossover makes the algorithm can find target circuits with different number of component. In addition, the random length strategy of the proposed algorithm doesn’t reduce the diversity of the component number of mutation circuit. The experiment results show that the proposed algorithm can finish the variable structure evolution task successfully, it is convergent and a new algorithm for analog circuit variable structure evolutionary design.2. Analyze the impact on the circuit topology of group crossover of variable structure differential evolution algorithm. Make an analysis about variable length evolution. The analysis and the experiment results both indicate that the variable length evolution can reduce the cost of the component of evolution circuit. Variable length evolution and group crossover can both slow down the descent speed of evolutionary success rate, and the ease the growth trend of evolutionary cost. The variable structure differential evolution can improve the scalability of analog circuit variable structure evolution design.3. Study the problem of self-adaptive operator of variable structure differential evolution algorithm. A self-adaptive control strategy for mutation operator based on mean value and standard deviation, and a self-adaptive control strategy for crossover operator based on histogram statistic are proposed. The comparative experiments show that the proposed two operator self-adaptive control strategies can improve the search efficiency of evolution algorithm, and improve the scalability of circuit evolution efficiently.In brief, in this thesis, a new variable structure evolution algorithm for analog circuit is proposed, and the algorithm combined with the problem of analog circuit variable structure evolution design is analyzed. The theory and experiment both indicate that the proposed algorithm can obtain high evolutionary efficiency and performance, thereby improve the scalability of the evolution problem. Moreover, the proposed algorithm modified by operator self-adaptive control strategies further improve the scalability of the algorithm on the problem of circuit variable structure evolutionary design.