With the development of information technology, blockchain technology has become a key technology to ensure the security of cryptocurrency transactions. However, phishing fraud has become a major criminal offence that undermines the security of transactions in blockchain ecosystem. Phishing fraud detection has become one of the hot issues in blockchain field research. In this paper, we innovatively propose a Phishing Fraud Detection model based on Dynamic Graph Temporal Feature Coding in Ethereum (PFD-TF) for predicting who will be the next target of phishing fraud. First, We constructs a dynamic graph of blockchain transactions with timestamps based on blockchain cryptocurrency transaction data. Then, a timestamp feature encoding strategy is proposed to learn the functional representation of time directly. And the interactions between node features, edge features and the topology of the graph are used to obtain phishing or non-phishing node embeddings. Finally, the acquired node embeddings are trained to be fitted using the cross-entropy loss function to complete the prediction of the next fished node. The experimental results show that the phishing fraud detection model (PFD-TF) proposed in this paper outperforms the results of state-of-the-art methods, indicating that the PFD-TF model is effective in phishing fraud detection in cryptocurrency transactions.