Design of a Traveling Wave Tube Multistage Depressed Collector Based on Back Propagation Neural Network–Random Forest Algorithm

The multistage depressed collector (MDC) is a significant component of the traveling wave tube (TWT) that collects the wasted energy of the electron beam after beam-wave interaction. However, more stages of the collectors means the design process of the MDC is more complex and needs more calculations due to more parameter combinations. To simplify the design process and shorten the design time, the number of samples should be reduced during parameter scanning. To solve the above problems, this article combines two powerful algorithms, random forest (RF) and back propagation neural network (BPNN), as the BPNN-RF algorithm. Finally, this novel algorithm successfully forecasts the power collected by a 100-W TWT’s three-stage MDC with lower root mean square error (RMSE) than using these two algorithms. The MDC’s efficiency of this MDC after optimization is up to 85% when the secondary electron yield (SEY) is 1.2 in the CST studio suite.