Aiming at critical issues in the function-oriented composition of semantic Web services, such as the incapability of classic AI planning methods to handle the dynamically created individuals during Web services' executions and the inadequacy of service matchmaking based methods to fully exploit the semantic connections between types of services' I/O parameters. Following a comparative study of description logics and dynamic logics, Web services' IOPR's (inputs, outputs, preconditions and results) are encoded in axioms of description logics, and AI (artificial intelligence) planning method based on dynamic logic is extended to accommodate the transformation of a service composition problem into reasoning task of description logics. It finally overcomes difficulties in the classic AI planning method and defects in those methods based on service matchmaking.