In order to solve the problem of large-value transactions that are prone to failure in payment channel networks, we propose an innovative scheme named Multipath Forwarding based on Transaction Splitting (MFTS), considering insufficient channel balance. First, we propose a path-window based transaction segmentation mechanism to divide transactions that do not satisfy the transaction conditions into different small-value transactions. Then, we design a deep reinforcement learning-driven channel probing algorithm to find the best transaction routes to optimize the transaction throughput. Meanwhile, to ensure transaction security, homomorphic encryption is used to encrypt the segmented microtransactions. Finally, we verify MFTS in the Watts-Strogatz small-world model by simulating payments in conjunction with the Kaggle credit card and Ripple dataset, and the results demonstrate the effectiveness of the designed scheme in improving the transaction success rate and the network throughput.