dividing a planning time of the large-scale satellite network with resources to be allocated into T time slots;

analyzing a type of each of to-be-served service requests arriving at the large-scale satellite network at a t^th time slot of the T time slots to determine the type of each of the to-be-served service requests; and analyzing, according to the type of each of the to-be-served service requests, attributes of each of the to-be-served service requests to determine the attributes of each of the to-be-served service requests;

calculating, according to the attributes of each of the to-be-served service requests, edge attributes constructed for connecting the to-be-served service requests by using a service priority formula, to generate a service relationship graph corresponding to each type of the to-be-served service requests; and extracting, according to the service relationship graphs, a service demand feature of each of the to-be-served service requests by using a trained service request representation network;

encoding, by using a service request encoder, the service demand feature of each of the to-be-served service requests to generate a network service status of each of the to-be-served service requests; obtaining, according to the network service status of each of the to-be-served service requests, a decoding hidden status of each of the to-be-served service requests by using a service request decoder, calculating, by using a service request served probability formula based on the decoding hidden status of each of the to-be-served service requests, a probability of being served of each of the to-be-served service requests, and determining, according to the probability of being served of each of the to-be-served service requests, a service order of the to-be-served service requests; wherein the service request encoder and the service request decoder are constructed by long short-term memory networks;

calculating, by using a service communication and computing resource demand formula, a demand of each of the to-be-served service requests for communication and computing resources according to the type and the attributes of each of the to-be-served service requests, available resource status information of each satellite node of the large-scale satellite network and the service order of the to-be-served service requests, to generate an available service mode set satisfying each of the to-be-served service requests; and calculating, according to the available service mode set, a probability of being selected of each available service mode in the available service mode set of each of the to-be-served service requests by using an Actor network, and selecting, according to the probability of being selected of each available service mode in the available service mode set of each of the to-be-served service requests, a service strategy of each of the to-be-served service requests;

until the t^th time slot, each of the to-be-served service requests selects a service strategy, or available resources in the large-scale satellite network with resources to be allocated are insufficient; and

obtaining a service strategy of each of to-be-served service requests within the T time slots;

wherein the trained service request representation network, the service request encoder, the service request decoder, and the Actor network are obtained through training in a reinforcement learning framework with a goal of maximizing a completion rate of the to-be-served service requests, and the service request representation network comprises a graph convolutional network; and

wherein the service request served probability formula is expressed as follows:

wherein p(M^t) represents a probability of being served of each of the to-be-served service requests in a to-be-served service request sequence M^t={m|m=1, 2, . . . , |M^t|} in the t^th time slot, softmax(·) represents a normalized exponential function, Uⁱ represents an input value of each of the to-be-served service requests in the normalized exponential function softmax(·) during an i^th service, U_jⁱ represents an input value of a j^th to-be-served service request in the normalized exponential function softmax(·) during the i^th service, tanh(·) represents a hyperbolic tangent function, N_j represents a network service status of the j^th to-be-served service request, H_i represents a decoding hidden status for calculating an i^th to-be-served service request, V, W_e and W_d each represent a weight parameter, SR_jⁱ represents whether the j^th to-be-served service request is served during the i^th service, * represents a matrix multiplication operation, and T represents a matrix transpose operation.