Abstract: A novel dual-channel wind power forecasting model based on the temporal convolutional network (TCN) was proposed. This model adopted a dual-channel structure to achieve the deep fusion of multi-source features from wind farms and model computations. The spatiotemporal perception channel extracted the short-term variation response relationship between meteorological data and wind power by cascading a multi-scale entropy routing attention (MERA) mechanism with the TCN structure. Meanwhile, the temporal feature channel constructed a multi-step dependent temporal convolutional network (MST) to deeply mine the long-term temporal dependency features of wind power. The fused dual-channel feature data was fitted by a Kolmogorov-Arnold network (KAN), utilizing the nonlinear mapping capability of its spline functions to output the final prediction results. Results show that the necessity and effectiveness of each module in the model are verified. Compared with other baseline forecasting models, the proposed model exhibits performance advantages.