Partial discharge of electrical equipment is not only the main factor of insulation deterioration, but also an important parameter to effectively characterize insulation defects. The accurate detection of partial discharge can detect the potential faults that endanger the safety of equipment in time. UHF detection has the advantages of good real-time performance and strong anti-interference, and is widely used in discharge detection. However, the existing microstrip antenna sensor is limited by the structure size, and the working bandwidth is difficult to increase. In this paper, a new flexible microstrip antenna sensor based on polyimide is developed. The partial flooring technology combined with the beveled meandering technology is used to improve the structure. The nonlinear relationship between antenna size and working bandwidth is considered to optimize the size. The working bandwidth is expanded while keeping the antenna area unchanged. In order to solve the problem that the antenna performance is unstable due to the adjustment of single size parameters in the process of size optimization, a relationship model between multi-size and working bandwidth is proposed by using radial basis function (RBF) neural network, and an improved beluga whale optimization (IBWO) algorithm is used to optimize antenna size. The simulation results show that the size of the new flexible microstrip antenna is reduced by 59.59%. The operating bandwidth is increased from 0.598~0.6 GHz to 0.3~3 GHz, which fully covers the design requirements of partial discharge detection. By simulating partial discharge test and comparing with Archimedean spiral antenna and three-dimensional spiral antenna, the results show that the new flexible microstrip antenna has more reliable detection performance.