The operating time of ultra-high voltage circuit breakers has significant dispersion. In order to enhance the accuracy of predicting the action time of phase controlled switches, this paper conducted on-site experiments on 550kV hydraulic operated circuit breakers. Subsequently, a random forest (RF) regression prediction model was established based on the relationship between external parameters and the actual operation time dispersion of circuit breakers, with three variables of control voltage, environmental temperature, and oil pressure as inputs. The improved whale optimization algorithm (WOA) was used to optimize the RF algorithm and improve its prediction accuracy. Due to the difficulty of deploying neural network algorithms in embedded devices and the low accuracy of prediction methods used for circuit breaker closing time in practical engineering, this paper proposes a compensation method for circuit breaker action time prediction based on cloud edge collaborative computing architecture. Firstly, an improved WOA-RF model is trained on big data in a cloud computer, and node information is calculated using the built model. Then, based on the node information, Hermite interpolation is used to calculate the closing time of the circuit breaker in the embedded device on site and compensate for it. At the same time, historical closing data records are transmitted to the cloud computer system to continuously improve the database. The research results indicate that using this scheme can more accurately predict the action time of circuit breakers.