The flow separation caused by the increase in load on aero-engine compressors often results in flow separation, leading to decreased flow stability and increased flow loss. This paper investigates active flow control strategies based on a highly-loaded compressor cascade with Coanda jet flaps to effectively suppress flow separation and decrease aerodynamic losses of aero-engine compressors. To analyse the influence of the jet on aerodynamic performance, three-dimensional unsteady computational fluid dynamics simulations are conducted. The dynamic characteristic model of the controlled object is established using the time-delay neural network (TDNN) method based on time-domain data obtained from wind tunnel experiments. Three control strategies, including adaptive radial basis function (ADA-RBF) neural network control, PID control, and RBF-PID control strategies, are designed to achieve a fast-response and adaptive control effect. Simulation results demonstrate that RBF-PID control achieves faster convergence speeds than conventional PID control. Given that the adaptive RBF neural network control method, provides both faster convergence speed and more accurate initial jet flow rates, corresponding experimental studies are carried out to validate the proposed approach. The findings demonstrate that ADA-RBF neural network control can effectively inhibit flow separation, even with continuously changing inflow velocity and pressure in the wind tunnel. The total pressure loss coefficient is decreased by 6.13% and 7.87% at Mach numbers of 0.4 and 0.6, respectively.