Abstract: When a high proportion of distributed photovoltaic (PV) generation is integrated into the power grid, dispatch and control become more challenging, the depth of network state perception is insufficient, and the uncertainty in control latency also increases. To address this issue, a convolutional neural network-bidirectional long short-term memory network-attention mechanism (CNN-BiLSTM-Attention) based model for estimation of latency in dispatch and control of distributed photovoltaics (DPV-DCLE) is proposed. Using task completion rate, root mean square error , mean absolute error, mean bias error, R-squared , and curve similarity as evaluation metrics, combined with the dynamic time warping (DTW) algorithm, a simulation experiment is conducted to compare and analyze four cases: method one based on convolutional neural network-long short-term memory (CNN-LSTM), method two based on CNN-LSTM-Attention, the proposed method, and actual latency samples. Experimental results indicate that the proposed method achieves higher accuracy in estimating the control latency of distributed photovoltaic systems, and outperforms the other two methods in prediction accuracy, which helps to reduce the uncertainty in the control latency of such systems. At this time, the integration of the delay estimation module on the distributed photovoltaic cluster control terminal provides technical support for refined management of photovoltaic grid connection and offers certain engineering application value.