The objective of within-document Event Coreference Resolution (ECR) is to group all instances of the same event in a given document into a single cluster. Recent work on ECR has transformed event coreference resolution into a cloze-style task, which is more closely aligned with the pre-training task of a pretrained language model (PLM) - masked language model (MLM) - with the objective of reducing reliance on event encoding capabilities. Nevertheless, this approach continues to employ one-hot labels as the single objective for fine-tuning, which makes it challenging for the model to comprehend the interrelation between events context and coreference words. Furthermore, end-to-end ECR is susceptible to errors originating from Event Detection (ED), which can result in incorrect events. To tackle these two problems, we present a knowledge distillation method for end-to-end ECR with simplified event detection. This approach is designed to assist PLM in learning the relation between event context and coreference words while also reducing error propagation. The experimental results on the KBP corpus demonstrate that our proposed method attains the state-of-the-art (SOTA) performance in both annotated event and predicted event coreference resolution.