Accurate prediction of solar irradiance is crucial for advancing renewable energy and optimizing power grid scheduling. However, solar irradiance is highly influenced by weather conditions, particularly cloud movements, resulting in frequent and significant fluctuations. These short-term, high-frequency variations add complexity to the prediction process. To address this challenge, this study employs lidar, widely used in meteorological observations, to obtain cloud information through volume scan mode. We propose a hybrid neural network method for ultra-short-term irradiance prediction based on cloud characteristics inferred from lidar data. Initially, lidar echo signal profiles are analyzed to retrieve cloud parameters from individual scan data. Subsequently, to capture cloud features relevant to irradiance, we extract the region intersecting the solar path during the radar scan cycle and construct a multidimensional feature matrix. Finally, a hybrid neural network model integrating a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network is employed, leveraging the local feature extraction capabilities of CNNs and the temporal dependency capturing strengths of LSTMs. This model predicts irradiance by integrating feature matrices of meteorological data, cloud characteristics, and historical irradiance data. The method was validated using data from lidar and environmental monitoring instruments installed at the same photovoltaic power plant. Results demonstrate that the proposed method, utilizing cloud features inferred from lidar, significantly improves the accuracy of ultra-short-term irradiance prediction for photovoltaic power plants, showing considerable engineering applicability.