Abstract: With a high proportion of renewable energy units connected to the power grid, the equivalent inertia of the power system continues to decline, and the frequency stability deteriorates significantly. To address this issue, a short-term prediction method for system equivalent inertia is proposed based on a temporal convolutional network and a bidirectional long short-term memory network (TCN-BiLSTM). Firstly, variables with a greater impact on the system equivalent inertia are selected for prediction based on the Pearson, Spearman, and Kendall correlation coefficients. Secondly, cluster analysis is performed using the k-means clustering (k-means) algorithm to obtain a data subset with similar fluctuation characteristics. Subsequently, a short-term prediction model for system equivalent inertia based on TCN-BiLSTM is established to mine the mapping relationship between system equivalent inertia and input variables, thereby obtaining the inertia prediction results. Finally, the effectiveness of the proposed method is verified using example data. The results indicate that the proposed method exhibits high prediction accuracy, which can provide reliable information for power grid dispatching and improve the frequency stability of the system.