Research On Modeling And Optimization Of Anticorrosion Injection Agent In Oil Distillation

With the continuous growth of the national economy, China's demand of crude oil increases dramatically. Domestic crude oil can't meet the need of development of economy. Because of the quality difference of the import crude oil compare greatly, the content of sulfur and acid Value are higher. Enterprises have to face the problem of equipment corrosion.In oil refining enterprises, "Desalting and three injections" is an available way to prevent corrosion at topping system of distillation towers in the crude distillation unit (CDU). Injections agent's adjustment is stayed on expert experience. Adjustment and the relation between adjustment and oil parameters may be further discovered. The paper discussed modeling and optimization on injection agent in both theory and practice, and found out the optimal dose of injections agent under various conditions.At present, the frequent change of oil and process data, make the characteristics of the sample data have too many properties. If the direct use of such data for the establishment of SVM prediction model, which will increase SVM calculation, and reduce the SVM training speed and precision. This problem can be resolved by classified data to identify the actual regularity implicit in the process data. The method is a fuzzy classification theory of process parameters on the basis of fuzzy mathematics and fuzzy reasoning. It mainly provides support for analysising and optimizing process parameters by establishing fuzzy rules model. For different enterprises of different equipment corrosion resistance, this thesis used a variety of ways to classify the data.SVM is a new learning method based on Statistical Learning Theory (SLT). SVM, which based on Structural Risk Minimization (SRM) principle. Based on the analysis of experts on the basis of empirical data, the application of SVM method build and acid value of crude oil, sulphur and drainage pH, the concentration of iron model, has different types of injection and dose of acid value of crude oil, sulphur content And the drainage of pH, iron ion concentration of data input and output; And use of SVM establishment of the injection of ammonia, note inhibitor of prediction model and BP neural networks forecast results were compared. The simulation results show that support vector machines overcome the neural network, and other methods of some inherent shortcomings, improve the ability of the generalization of the model, in a small sample of modeling has a special advantage.Based on the vague classification of the data will be collected at the scene of the data classification process, using SVM note of the establishment of the relationship between the data model, a clear relationship between the data, although the macro level to reflect note of the general process of law, Are not the types of crude oil production is the optimal amount of ammonia and injection of corrosion inhibitor injection. In this paper,"a note of ammonia, note the inhibitor 0-1 integer programming model and application binary PSO solution to the refinery production, injection and dose of acid value of crude oil, sulphur and drainage pH, iron concentrations in different Under the conditions (such as satisfaction, and so on) for optimal results.After PSO for the optimal solution to meet the technological requirements and different satisfaction, under the premise of the moderate dose injection, not only to achieve the purpose of the equipment preservative, preservative and in order to avoid the excessive running, causing massive waste of the phenomenon.Finally, Summary of this thesis is given, and future research is prospected.