With the new generation of information technology changes, the deep integration of artificial intelligence and edge computing has become a trend, and the arithmetic demand for edge data is exploding, and complex application environments cause large fluctuations in the workload of edge heterogeneous computing platforms, resulting in unstable performance and reduced system utilization. Load prediction for heterogeneous platforms can guide the reasonable allocation of tasks, which can improve the task balance of heterogeneous platforms and enhance the performance of heterogeneous platforms. Compared with statistical regression, classical machine learning, deep learning models and other methods, the Informer model enhances the LSTF (Long Sequence Time-series Forecasting, LSTF) prediction capability and reduces the computational overhead, but the Informer model is applicable to the LSTF of the periodic model, and the edge heterogeneous computing platform's loads are non-stationary long time series with strong fluctuations, for which the accuracy of prediction is lacking. In this paper, a DWR-Informer (Discrete Wavelet Resolution, DWR) model based on the Informer model is proposed to improve the ability of long sequence prediction of the Informer model by utilizing the multi-resolution analysis property of the discrete wavelet transform to obtain sequence-enriched features from different frequency bands. It is experimentally verified that the model has high accuracy in load prediction for edge heterogeneous computing platforms, which can guide the balance of resource allocation under complex networks, improve the task throughput of the network, and enhance the performance of the network.