Auto-Machine Learning-Based <i>W</i>-Band High-Efficiency Oscillator Design

In this article, automatic machine learning (ML) technology and global optimization algorithms have been combined to achieve a fast and high-performance extended interaction oscillator (EIO) design. An EIO model is manually designed to give a comparison. The manually designed EIO model can predict an output power of 7.9 kW with an efficiency of 15.8%. The data to train ML models are generated with CST-MATLAB co-simulation. During dataset creation, some EIO models are derived from high-output-power models by changing voltage and current parameters. By adding these derived models into dataset, power gradients over current and voltage are decreased. This is the key to the success of the whole project. Particle swarm optimization (PSO) and genetic algorithm (GA) are tested and compared to achieve the EIO fast design. PSO algorithm performs better than GA for relatively fixed searching directions and less randomness. An EIO optimized by PSO can predict an output power of 11.9 kW. The EIO electronic efficiency is 19.3%, which is 3.5% higher than that of the manually designed oscillator.