<p>This paper proposes a neural network-based distributed parameter modeling (DPM-NN) method for permanent magnet synchronous machines (PMSMs), incorporating spatial harmonics and iron losses. The proposed method uses magnetic coenergy data from finite element analysis (FEA) to accurately represent both the spatial harmonic characteristics and the magnetic saturation effects of the machine. To account for the effects of iron losses, rotor speed is introduced as an additional input dimension. In contrast to conventional distributed parameter models (DPM), the proposed approach decomposes the magnetic coenergy into spatially fluctuating and mean components, each modeled using a Fourier orthogonal basis neural network. By only retaining the dominant frequency-domain components, the model achieves a substantial reduction in dimensional complexity. Based on the proposed model, an analytical expression for electromagnetic torque is derived. Simulation and test bench validations demonstrate that the proposed model accurately predicts torque outputs under both steady-state and transient conditions, confirming its effectiveness in high-precision motor modeling and torque observation.</p>