Bitcoin, as the spearhead of cryptocurrencies, is playing a more and more critical role in the world. With the increasing number of users in Bitcoin, the public has risen the demand for transaction throughput, and many hard forks are generating by the continuous throughput improvement. Meanwhile, the vulnerability of Bitcoin's incentive mechanism also makes it vulnerable to be attacked by generating soft forks. Without a robust throughput scheme, Bitcoin suffers from hard forks, soft forks and possible attacks like undercutting attack. In this work, we create a useful model to describe the throughput and the incentive mechanism of Nakamoto consensus protocol. Based on our model, We find that the throughput, which is far higher than user needs, will result in soft forks which reduce the consistency of Bitcoin in the future. We design a robust throughput scheme to reduce the soft forks and hard forks by automatically adjusting block size according to users' needs. In our experiments, we simulate attacks on Bitcoins and our scheme at different mining powers. We find that the soft fork block rate of Nakamoto consensus protocol without block reward is 6.0%, and our scheme can reduce the soft fork block rate to 0.6% under the attacker with the mining power of 0.25.