The vacuum interrupter is widely used due to the advantages of no arc-extinguishing medium and high insulation strength. However, the arc voltage generated by the commonly used CuCr contacts is low. In hybrid DC circuit breakers (DCCB), hybrid automatic transfer switches (ATS), medium voltage compound switches and other fields, it is difficult to rely on vacuum arc to complete natural commutation, which restricts the development of hybrid switches. In order to understand the current commutation process deeply, the influence of internal and external factors on the current commutation is analyzed by experiments. The coupling mathematical model of arc-commutated branch is established. The criterion for the success of current commutation is summarized. The parameters of the arc model are reconstructed through repeated breaking experiments to explore the influence of internal factors on the arc characteristics. Based on this, the influence law of arc current, contacts gap and transverse magnetic field (TMF) is analyzed. An acceleration method of current commutation is proposed. A prototype for accelerating experiments with an electromagnetic repulsion mechanism and TMF is developed. The commutated branch equivalent to practical applications is built. The experimental results show that the commutation time is effectively shortened and meets the requirements of practical applications through the acceleration method, which provides new thought for the development of hybrid switches.

No abstract

Phase-control technology requires a sufficiently small dispersion of switching operation times, however, even though the switch itself is sufficiently stable, action time dispersion is still a key issue in phase-control technology due to various factors. In this paper, by building a simulation and test model of the hybrid mechanism, the action time variation caused by influencing factors such as fast switching temperature, capacitance voltage, capacitance degradation are investigated, and the test and simulation data are collated and analysed form a database. The Generalized Regression neural network has excellent prediction, self-learning, non-linear mapping and tolerance capabilities, and the prediction error of the GRNN algorithm is verified to be less than 0.2ms for the closing time.