本文围绕“基于混频流数据的中国财政收入实时监测”这一中心问题， 采用税收、地方非税和中央非税三大分项财政收入分别进行最优预测模型开 发，再加总得到财政收入实时预测值的方式开展研究。具体地，首先基于总 税收、地方非税和中央非税三大分项财政收入分别进行月度内每日入库比例 的演进趋势和基本特征的归纳分析；然后结合趋势特征，分别构建不同的月 度预测模型并进行相应的预测效果比较；再后，基于三大分项收入的最优模 型加总构建中国财政收入组合预测模型，并与基于财政收入的独立预测模型 进行效果比较，得到的最优预测结果误差率约为 10%左右；最后，总结上述研 究结果并结合未来可能的走势，给出未来财政收入实时监测的思路、方法和 模型建议。

Focusing on the central issue of "real-time monitoring of China's fiscal revenue based on mixed stream data", this essay adopts three sub items of fiscal revenue: tax revenue, local non tax revenue and central non tax revenue to develop the optimal prediction model, and then sums up the real-time prediction value of fiscal revenue. Specifically, firstly, based on the total tax revenue, local non tax revenue and central non tax revenue, this paper analyzes the evolution trend and basic characteristics of the monthly daily warehousing proportion. Then, according to the trend characteristics, it constructs different monthly forecasting models and compares the corresponding forecasting results. Then, it constructs the optimal model of China's daily warehousing proportion based on the three sub revenue. Finally, this paper summarizes the above research results and gives the ideas, methods and model recommendations of real-time monitoring of future fiscal revenue.