CN-116798541-B - Method and system for predicting content of sulfide and nitride in catalytic cracking purified flue gas

CN116798541BCN 116798541 BCN116798541 BCN 116798541BCN-116798541-B

Abstract

The invention relates to the technical field of industrial catalytic cracking, in particular to a method and a system for predicting sulfide and nitride contents in catalytic cracking purified flue gas. The method comprises the steps of obtaining catalytic cracking process data, respectively adopting a plurality of machine learning algorithms to conduct feature selection on the catalytic cracking process data, evaluating importance scores of all features, selecting features with importance scores meeting requirements, generating a plurality of feature subsets, conducting cross-validation on the feature subsets obtained by the plurality of machine learning algorithms, respectively obtaining feature subsets with performance meeting requirements obtained by the plurality of machine learning algorithms, conducting weighted fusion to obtain a common feature set, conducting nonlinear fitting on the common feature set by adopting an improved transducer model, predicting sulfide and nitride contents in catalytic cracking purified flue gas, and outputting the sulfide and nitride contents. The method has the advantages of high prediction precision, high calculation efficiency, wide application range and the like, and has wide application value.

Inventors

DU WENLI
LONG JIAN
ZHONG WEIMIN
QIAN FENG
YANG MINGLEI

Assignees

华东理工大学

Dates

Publication Date: 20260512
Application Date: 20230628

Claims (7)

1. The method for predicting the content of sulfide and nitride in the catalytic cracking purified flue gas is characterized by comprising the following steps of: S1, acquiring catalytic cracking process data; Step S2, respectively adopting a plurality of machine learning algorithms to perform feature selection on the catalytic cracking process data, evaluating the importance score of each feature, selecting the features with the importance scores meeting the requirements, and generating a plurality of feature subsets; Step S3, performing cross verification on a plurality of feature subsets obtained by a plurality of machine learning algorithms to respectively obtain feature subsets with performance meeting requirements obtained by the plurality of machine learning algorithms, and performing weighted fusion to obtain a common feature set; s4, adopting an improved transducer model to perform nonlinear fitting on the common characteristic set, predicting the sulfide and nitride content in the catalytic cracking purified flue gas and outputting the sulfide and nitride content; wherein, the step S3 further includes: Obtaining a final evaluation result of the feature subset according to the cross verification, and selecting the feature subset with the final evaluation result meeting the requirement as the final feature subset; distributing weights to the final feature subsets obtained by each machine learning algorithm according to the final evaluation result; weighting and fusing the final feature subsets according to the weights to obtain a common feature set; the step S4 further includes: Calculating the similarity between the feature vectors in the common feature set by adopting a product quantization method; Generating weights among the feature vectors according to the similarity; Weighting and fusing the feature vectors by using weights to obtain fused feature vectors; Generating a prediction result according to the fused feature vector; The calculating the similarity between the feature vectors in the common feature set by using the product quantization method further comprises: Dividing the feature vector into a plurality of sub-vectors; generating an independent quantization table for each sub-vector; quantizing each sub-vector using the generated quantization table; restoring the original feature vector through the quantized sub-vector; and calculating the similarity between the feature vectors by adopting an optimized distance calculation method.
2. The method for predicting the content of sulfides and nitrides in the catalytic cracking cleaned flue gas according to claim 1, wherein the multiple machine learning algorithms in step S2 further comprise an extreme gradient boosting algorithm, The extreme gradient lifting algorithm corresponds to an evaluation function expression as follows: ; Wherein G X is the sum of the first partial derivatives of leaf node X, G Y is the sum of the first partial derivatives of leaf node Y, H X is the sum of the second partial derivatives of leaf node X, H Y is the sum of the second partial derivatives of leaf node Y, And Respectively represent Penalty term Penalty term.
3. The method for predicting the content of sulfides and nitrides in the catalytic cracking cleaned flue gas according to claim 1, wherein the multiple machine learning algorithms in step S2 further comprise a random forest algorithm, The random forest algorithm corresponds to an evaluation function expression as follows: ; Wherein, the Representing the average change amount of node splitting non-purity of the ith variable in all trees of the random forest, wherein N is the number of random forest classification numbers; The corresponding expression is: ; ; Wherein M is a variable Number of occurrences on the kth tree; representing the Gini index at node m; And Representing Gini indexes of child node l and child node r split at node m, respectively.
4. The method for predicting the content of sulfides and nitrides in catalytic cracking cleaned flue gas according to claim 1, wherein the cross-validation method comprises the steps of: dividing the feature subset into a plurality of subsets with equal size; selecting one of the subsets as a verification set and the remaining subset as a training set; training a model on a training set, carrying out model evaluation on a verification set, and recording evaluation index values; Repeating the steps until all subset evaluations are completed; And summing all the evaluation index values to obtain a final evaluation result of the feature subset.
5. The method for predicting the content of sulfides and nitrides in the catalytic cracking cleaned flue gas according to claim 1, wherein the step S4 further comprises: An improved transducer model is adopted as a nonlinear fitting method and is used for establishing a mapping relation between an input characteristic vector and an output label, wherein the output label is a predicted value of sulfide and nitride content in catalytic cracking purified flue gas.
6. The method for predicting the sulfide and nitride content in a catalytic cracking cleaned flue gas according to claim 1, wherein the modified converter model of step S4 further comprises a coding component and a decoding component: The coding assembly consists of a multi-layer coder, and the decoding assembly consists of decoders with the same layer number; each encoder is composed of a self-attention layer and a feedforward neural network layer.
7. A system for predicting the content of sulfides and nitrides in catalytic cracking purified flue gas, comprising: a memory for storing instructions executable by the processor; a processor for executing the instructions to implement the method of any one of claims 1-6.

Description

Method and system for predicting content of sulfide and nitride in catalytic cracking purified flue gas Technical Field The invention relates to the technical field of industrial catalytic cracking, in particular to a method and a system for predicting sulfide and nitride contents in catalytic cracking purified flue gas. Background Catalytic cracking (Fluid CATALYTIC CRACKING, FCC) is an important process for refining petroleum to convert heavy petroleum into light gasoline and other high value products by spraying hydrocarbon oil into a riser reactor and contacting the hydrocarbon oil with a high temperature catalyst to decompose the hydrocarbon oil into smaller molecules, the cracked hydrocarbon vapors are separated and further processed, and the deactivated high temperature catalyst is regenerated. The catalyst is warmed during regeneration and then recycled back to the riser reactor to provide the required heat for the endothermic reaction and the heat required for the vaporization of the feedstock. The catalytic cracking process has various advantages, including firstly, the catalytic cracking process can convert heavy crude oil into light gasoline and other products with high added value, thereby improving the product yield, being beneficial to meeting the market demand for high-quality gasoline and petrochemical products and improving the profit margin of a refinery, secondly, the catalytic cracking process can adapt to petroleum raw materials with different types and qualities, including heavy and low-quality petroleum raw materials, thereby improving the flexibility and adaptability of the refinery, and furthermore, the catalytic cracking process can also improve the yield and selectivity of petroleum fractions and reduce the energy consumption and environmental pollution of petroleum processing. Therefore, the catalytic cracking technology has important significance for guaranteeing energy safety and environmental protection, and has wide application prospect. However, in the catalytic cracking process, the amount of purified flue gas discharged is large and the discharged components such as nitrogen oxides and sulfides are complex, which constitutes a potential threat to air quality and health. Therefore, the importance of predicting and controlling the sulfide and nitride content in the exhaust cleaned flue gas is becoming increasingly prominent. Disclosure of Invention The invention aims to provide a method and a system for predicting the content of sulfides and nitrides in catalytic cracking purified flue gas, which solve the problem that the content of sulfides and nitrides in the purified flue gas of a catalytic cracking process is difficult to predict accurately in the prior art. In order to achieve the above object, the present invention provides a method for predicting sulfide and nitride contents in catalytic cracking purified flue gas, comprising the steps of: S1, acquiring catalytic cracking process data; Step S2, respectively adopting a plurality of machine learning algorithms to perform feature selection on the catalytic cracking process data, evaluating the importance score of each feature, selecting the features with the importance scores meeting the requirements, and generating a plurality of feature subsets; Step S3, performing cross verification on a plurality of feature subsets obtained by a plurality of machine learning algorithms to respectively obtain feature subsets with performance meeting requirements obtained by the plurality of machine learning algorithms, and performing weighted fusion to obtain a common feature set; and S4, performing nonlinear fitting on the common feature set by adopting an improved transducer model, predicting the content of sulfide and nitride in the catalytic cracking purified flue gas, and outputting. In one embodiment, the plurality of machine learning algorithms in step S2, further comprises an extreme gradient boosting algorithm, The extreme gradient lifting algorithm corresponds to an evaluation function expression as follows: Wherein G X is the sum of the first partial derivatives of leaf node X, G Y is the sum of the first partial derivatives of leaf node Y, H X is the sum of the second partial derivatives of leaf node X, H Y is the sum of the second partial derivatives of leaf node Y, and gamma and lambda represent the L 1 penalty and the L 2 penalty, respectively. In one embodiment, the plurality of machine learning algorithms in step S2 further comprises a random forest algorithm, The random forest algorithm corresponds to an evaluation function expression as follows: Wherein, the Representing the average change amount of node splitting non-purity of the ith variable in all trees of the random forest, wherein N is the number of random forest classification numbers; The corresponding expression is: Wherein GI l and GI r represent Gini indices of two child nodes split at node M, respectively, when variable X i occurs M times in the kth t