[Haichuan Intelligence] Operation of the “intelligent diagnosis and analysis model” for Guangdong Petrochemical’s catalytic cracking unit – key performance indicators of the stabilization system

Thread Content

As of the end of October, more than a month after its operation began, it has not only enabled accurate prediction of the final boiling point temperature of crude gasoline, with the error remaining at an industry-leading level of ±1 degree Celsius, but it has also increased the compliance rate for this parameter to 100%; meanwhile, the plant’s key overall energy consumption was reduced by 0.5 kilograms of standard oil per ton. 　　As the \"heart\" of oil refining, the final boiling point of the crude oil produced by catalytic cracking units directly affects the quality grade and market competitiveness of gasoline products. For a long time, this indicator has relied on manual sampling and analysis conducted between 7 a.m. and 7 p.m.; operators could obtain data only twice a day, and the 6 to 12 hour analysis cycle led to severe delays in parameter adjustment. “In the past, when receiving instructions for schedule adjustments, one could only rely on experience to carry out the operations; it was not until the test results came in that it became clear whether everything was up to standard, which made the situation very passive. ”Teng Budong, the shift leader of Refining Department 2, recalled that under the traditional approach, indicators fluctuated frequently, and the process of regulation was like trying to cross a river by feeling the stones beneath one’s feet. 　　To address this challenge in the industry, in July of this year, the Second Refining Department of Guangdong Petrochemical, together with the Planning and Technology Information Department, established a task force to develop intelligent prediction methods for the final boiling point of crude gasoline. The former, by leveraging in-depth understanding of the entire production process, systematically analyzed the vast amount of process data accumulated over the past two years, and accurately identified the key influencing variables among 211 complex parameters, ensuring that technology development met the actual needs of production ； Leveraging the integrated innovation platform for intelligent refining and chemical processing established jointly by Guangdong Petrochemical and Kunlun Digital, the latter quickly assembled a team of artificial intelligence algorithm engineers and development engineers, and established a technical framework covering the entire process from data cleaning to feature selection, model training, and iterative optimization. 　　“The core of model development is to let the data ‘speak’. ”Mo Mouyi, a technical specialist from Kunlun Digital Intelligence, explained that the team utilizes an intelligent factory data platform to collect in real time dynamic process data such as device temperature, pressure, and flow rates, and compares this data with laboratory test results on a one-to-one basis, thereby generating tens of thousands of high-quality deep learning samples. Thanks to AI algorithms with self-learning capabilities, the model automatically identifies the 10 most relevant key parameters, providing data support for precise control. 　　During the collaborative efforts, the team also addressed key issues such as data drift and algorithm fitting through close coordination via \"weekly meetings, monthly reviews, and on-site collaboration,\" thus creating a positive cycle in which production experience feeds back into technology development, while technological advancements improve production processes. 　　At around 8 a.m. on October 30, Teng Budong, who was about to hand over his shift, noticed during his inspection that the predicted values over a 12-hour period differed by less than 1 degree Celsius. “After making these adjustments, the results can be seen in real time on the screen; there’s no longer a need to wait for test results, and the device operates much more smoothly. ”Tembudong said. 　　Chen Yunqing, an expert in petrochemical information engineering from Guangdong Petrochemical, said that in the next step, the company will further expand the application scenarios of AI models to develop an \"intelligent quality prediction system for the entire process\" applicable to catalytic cracking units. At the same time, the team will also develop functions such as abnormal condition diagnosis and online operation optimization, in order to gradually achieve intelligent sensing and autonomous decision-making by the equipment, thereby providing a \"Guangdong Petrochemical Solution\" for the intelligent upgrading of the refining and petrochemical industry.