To address the issues of weight update stagnation and drift in traditional radial basis function networks(RBFN) caused by the reliance of weight adaptive laws on the persistent excitation (PE) condition, as well as the boundary chattering phenomenon in permanent magnet synchronous motor(PMSM) speed constraint induced by high gains and repeated differentiation in high-order control barrier functions(HOCBF), a constraint control strategy combining a novel weight adaptive law and a dynamic compensation-based HOCBF is proposed for the PMSM position control system. First, a neural network weight composite-adaptation law based on both historical and current errors is designed. By utilizing historical error information to construct a correction term, this law maintains parameter learning dynamics under insufficient excitation, thereby avoiding weight update stagnation and suppressing adaptable weight drift caused by accumulated approximation errors. Second, addressing the speed constraint chattering, a dynamic compensation-based HOCBF is proposed. Adopting a design of "low-gain constraint plus dynamic boundary compensation," this method initially confines speed states within a conservative boundary using lower gains to mitigate chattering. Subsequently, the compensation mechanism smoothly adjusts the constraint boundary to the desired value, thereby effectively resolving boundary chattering caused by high gains and differential amplification while ensuring constraint effectiveness. Furthermore, the system stability is analyzed in detail using lyapunov stability theory, proving that the system is uniformly ultimately bounded(UUB). Finally, the effectiveness of the proposed method is verified through experiments. The results demonstrate that this method effectively solves the stagnation and drift problems in traditional neural network weight learning and achieves smooth, chatter-free chatter-free speed constraint, significantly enhancing the safety and dynamic performance of the system while improving control accuracy.