Abstract: With the development of long-span flexible photovoltaic (PV) arrays towards configurations featuring increased rows and spans, conventional wind tunnel testing methods for acquiring wind-induced displacement responses face substantially higher costs, extended durations, and greater demands for monitoring points. To overcome the challenges of deploying extensive measurement points and intermittent data gaps in both wind tunnel experiments and field measurements, this study employs two deep learning models—Bidirectional Long Short-Term Memory (BiLSTM) and Temporal Convolutional Network (TCN)—to investigate multi-point response time-history prediction for a five-row, three-span flexible PV array. By utilizing the displacement response time-history data from the preceding 10 seconds, the models achieve accurate prediction of the subsequent 10-second time-history data, demonstrating the applicability of deep learning in forecasting wind-induced responses of long-span flexible PV arrays. The research reveals that both BiLSTM and TCN models exhibit strong generalization performance; specifically, the TCN model demonstrates higher training efficiency, while the BiLSTM model delivers superior testing-phase performance with better evaluation metrics including Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). This work establishes a framework for predicting wind-induced displacement time-history responses using mainstream time-series models, providing a novel technical pathway for future vibration response prediction in flexible PV arrays with expanded row-span configurations.