Current quantum programs are generally represented by quantum circuits, including various quantum gates. If the program contains gates that are directly represented as unitary matrices, these gates need to be transformed into quantum circuits composed of basic gates. This step is called quantum circuit synthesis. However, current synthesis methods may generate circuits with thousands of gates. The quality of these quantum circuits is low and they are very likely to output incorrect results when deployed to real noisy quantum hardware. When the number of qubits is increased to 8 while ensuring a small number of gates, the quantum circuit synthesis takes weeks or even months. This study proposes a quantum circuit synthesis method, realizing the fast synthesis from unitary matrices to high-quality quantum circuits. Firstly, an iterative method is introduced to approximate the target unitary matrix by inserting circuit modules. During the iteration, a look-ahead strategy with a reward mechanism is proposed to reduce redundant quantum gates. In the acceleration process of quantum circuit synthesis, the study proposes a pruning method to reduce the space of candidate circuit modules. The method first describes the closure of each candidate circuit module to characterize the representation space of the circuit, and then prunes based on the overlap rate of the representation spaces of the modules, thus constructing a small and high-quality candidate set. Furthermore, to reduce the overhead of searching for optimal gate parameters, this study packs the selected candidates with the target unitary into a uniform circuit so that we can quickly obtain the approximation distance by calculating its expectation on the ground state. Experiments show that, compared with the current optimal quantum circuit synthesis methods QuCT and QFAST, this study reduces the number of gates to 37.0%–62.5%, and achieve a 3.7–20.6 times acceleration in the 5 to 8 qubit quantum circuit synthesis.