根据稠油井蒸汽吞吐的生产特点，基于胜利油田胜科采油管理区草4沙四区块稠油井生产和经营等历史数据，利用神经网络技术，建立稠油注汽转周预测模型，预测稠油注汽转周井的产量和完全成本，与稠油注汽转周井近、远期产量和完全成本的历史数据进行对比，持续对稠油注汽转周模型进行优化，实现稠油注汽转周多维度智能预测，提高稠油井产量预测准确率、稠油注汽转周最佳时机预测准确率和最佳稠油注汽转周措施方案编制效率，提升稠油井智能决策分析管理能力，提高采油管理区的效益开发水平。该项技术自2021年在胜科采油管理区稠油区块推广应用以来，为采油管理区有效注汽转周165口井次，稠油注汽转周井措施增油量为7×104 t，与2020年同期相比，措施增油量增加了1×104 t，措施有效增油率提升约为17%。
According to the production characteristics of steam huff and puff in heavy oil wells，this paper establishes a predictive model for the steam injection cycle of heavy oil with the neural network method and the historical production and management data of heavy oil wells in the 4th Member of Eocene Shahejie Formation（Es4）of Block Cao4 in Shengke oil production management area of Shengli Oilfield. The production and full cost of wells for steam injection cyclic of heavy oil are predicted with the proposed model and compared with their historical data of short-and long-term production and full cost.Further，the model is optimized，which enables the multi-dimensional intelligent prediction of steam injection cycle of heavy oil. Moreover，it improves the production prediction accuracy of heavy oil wells，the prediction accuracy of the best time of steam injection cycle of heavy oil，and the preparation efficiency of the best measure scheme for steam injection cycle of heavy oil and enhances the intelligent decision-making，analysis，and management ability of heavy oil wells and the benefit and development level of oil production management areas. Since the technology was popularized and applied in the heavy oil block of Shengke oil production management area in 2021，effective steam injection cycle of heavy oil has been performed for 165 well times in the oil production management area. The cumulative oil increment is 7×104 t in wells for steam injection cyclic of heavy oil. The oil increment is increased by 1×104 t，and the effective oil increase rate is about 17%，compared with those in the same period in 2020.