Traditional satellite observation task scheduling typically involves mission planning systems owned by individual companies responding to individual companies responding to different user observation needs for remote sensing satellite mission planning. This approach often leads to several issues, such as slow response times to satellite tasks due to concurrent multi-user demands, low global efficiency in multi-task scheduling due to separate planning by different systems, and wastage of satellite resources caused by redundant observations of the same target by multiple satellites. The article proposes an intelligent scheduling method for hybrid remote sensing satellite constellation observation tasks based on reinforcement learning. It employs an LSTM network to intelligently predict remote sensing satellite observation tasks and utilizes a reinforcement learning network based on Markov Decision Processes (MDP) to automatically generate observation task plans for multi-agent planning systems. Subsequently, each intelligent planning system updates its task planning according to the scheduling plan, thereby achieving intelligent scheduling of hybrid remote sensing satellite constellation observation tasks. This method was applied to learn from sample data of hybrid remote sensing satellite constellation observation tasks from multiple companies, and a portion of the sample data was used for testing and validation. The experimental results demonstrate that the LSTM model, through deep learning, can predict the timeliness of observation tasks, with the initial prediction results showing high consistency with the validation data. The MDP model, through reinforcement learning, can schedule hybrid satellite constellation observation tasks, making intelligent decisions based on the timeliness of observation targets to automatically generate observation task scheduling plans optimized for the fastest timeliness.