Solving high-dimensional fractional differential equations has consistently remained one of the most signiffcant challenges within the realm of fractional calculus. Classical numerical methods often grapple with extensive history datasets, resulting in decreased computational efffciency. Existing non-classical methods frequently result in huge dimensional integer-order systems. To address these issues, an improved non-classical method is established in this paper. By introducing an integral interval decomposition strategy, the decay rate of the integrand in the non-classical method is signiffcantly accelerated, achieving a transition from algebraic decay to exponential decay. Consequently, there is a substantial reduction in the dimensionality of the approximate integerorder systems. Simultaneously, the introduction of the interpolation quadrature method further improves the computational accuracy. Theoretical analysis and numerical examples demonstrate that the proposed method offers substantial performance advantages in tackling fractional differential equations. In conclusion, the dynamic behavior of a high-dimensional memristive-coupled Hopffeld neural network was effectively investigated using the proposed method by phase diagrams and bifurcation diagrams. The proposed method demonstrates the capability to accurately identify both chaotic and periodic behavior within the system. Furthermore, it offers a substantial reduction in computation time for high-dimensional systems. This efffciency in capturing complex dynamics while decreasing computational overhead can be highly advantageous in various applications. It holds crucial theoretical implications for the numerical solution of fractional differential equation. Furthermore, it has the potential to foster and stimulate research and practical applications within the realm of fractional order systems.