CN104361153B - A kind of method for predicting RFCC settler coking amount - Google Patents
A kind of method for predicting RFCC settler coking amount Download PDFInfo
- Publication number
- CN104361153B CN104361153B CN201410584065.5A CN201410584065A CN104361153B CN 104361153 B CN104361153 B CN 104361153B CN 201410584065 A CN201410584065 A CN 201410584065A CN 104361153 B CN104361153 B CN 104361153B
- Authority
- CN
- China
- Prior art keywords
- settler
- value
- slurry oil
- rate
- coking
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Production Of Liquid Hydrocarbon Mixture For Refining Petroleum (AREA)
Abstract
The present invention relates to a kind of method for predicting RFCC settler coking amount, comprise the following steps：(1) obtain the basic data related to the condensation rate, capture rate, coking yield of settler, and condensation rate, capture rate and coking yield actual value, above-mentioned each group of data is pre-processed, then is normalized；(2) model training is carried out using neutral net, so that basic data is as input value obtained by step (1), using actual value as desired output, obtains the model of Predicted settlement device condensation rate, capture rate and coking yield；(3) basic data related to the condensation rate, capture rate, coking yield of settler to collection in worksite is pre-processed, substitute into respectively in the model that step (2) is obtained, obtain the predicted value of settler condensation rate, capture rate and coking yield, slurry oil mass flow in collection in worksite settler, calculating obtains settler coking amount predicted value.The method that the present invention is provided is easy, accurate, can apply to actual industrial production.
Description
Technical field
The present invention relates to petrochemical industry, and in particular to a kind of side of prediction RFCC settler coking amount
Method.
Background technology
RFCC (RFCC) occupies critical role in petroleum refining industry of China, be refinery increase economic efficiency it is important
Means.But heaviness, in poor quality with crude oil, the blending ratio of China's RFCC raw materials are improved constantly, feedstock property worse and worse,
RFCC products is distributed and product quality variation, and exacerbate RFCC device cokings.Wherein settler coking endangers most
Greatly, while occurring most universal, gently then cause catalyst circulation not smooth and catalyst largely runs agent, it is heavy then cause in catalyst circulation
Disconnected, device is forced unplanned shutdown, and the unplanned shutdown caused by coking has a strong impact on the economic benefit of RFCC devices.
At present, numerous studies are studied slowing down settler coking problem, are mainly included：Control the matter of feedstock oil
Measure, improve atomizing raw materials effect, suitable operating condition is selected according to reaction depth, anti-scorch quantity of steam and temperature, reinforcement is controlled
Settler is incubated, suitably uses high efficiency steam stripping technology, reduction oil gas residence time, does quiet run and public work guarantee well
Deng.Although these researchs play certain effect to delayed fallout coking problem, coking problem is still present, and it is annual all by
This causes refinery facilities unplanned shutdown.Therefore, the coking amount of on-line prediction catalytic cracking subsider, online judgment means coking
Situation is particularly significant to grasping plant running state, can be by adjusting feedstock property, catalyst once coking amount is too high
The modes such as property, operating parameter or scheduled shutdown, slow down the economic loss that unplanned shutdown is brought to refinery.
The content of the invention
In order to solve problems of the prior art, realization judges precipitator device coking situation online, and the present invention is carried
A kind of method for predicting RFCC settler coking amount has been supplied, the described method comprises the following steps：
(1) basic data related to the condensation rate, capture rate, coking yield of settler, and condensation rate, capture rate are obtained
With the actual value of coking yield, above-mentioned each group of data is pre-processed, then is normalized；
(2) model training is carried out using neutral net, using the basic data after normalized obtained by step (1) as defeated
Enter value, using actual value as desired output, obtain the model of Predicted settlement device condensation rate, capture rate and coking yield；
(3) basic data related to the condensation rate, capture rate, coking yield of settler to collection in worksite is located in advance
Reason, substitutes into the model that step (2) is obtained, obtains the predicted value of settler condensation rate, capture rate and coking yield, scene is adopted respectively
Collect slurry oil mass flow in settler, calculating obtains settler coking amount predicted value.
The basic data related to condensation rate and condensation rate actual value described in step (1) of the present invention are given birth to from catalytic cracking
Production scene is gathered by normal experiment to be obtained, wherein, the basic data related to condensation rate includes parameter in detail below：At the beginning of slurry oil
Evaporate point, 575 DEG C of quantity of distillate of slurry oil, the recovered (distilled) temperature of slurry oil 10%, the recovered (distilled) temperature of slurry oil 50%, the recovered (distilled) temperature of slurry oil 90%, sedimentation
Slurry oil partial pressure in device outlet temperature and settler.
The basic data related to capture rate and capture rate actual value described in step (1) of the present invention are given birth to from catalytic cracking
Production scene is simulated by computational fluid dynamics (Computational Fluid Dynamics, CFD) to be obtained, wherein, with catching
Obtaining the related basic data of rate includes parameter in detail below：Oil gas in settler linear speed, settler outlet temperature, in settler
Portion's pressure, slurry oil density and slurry oil carbon residue.
The basic data related to coking yield and coking yield actual value described in step (1) of the present invention are given birth to from catalytic cracking
Production scene is gathered by normal experiment to be obtained, wherein, the basic data related to coking yield includes parameter in detail below：Slurry oil is residual
Charcoal, slurry oil density, slurry oil mean molecule quantity, slurry oil saturated hydrocarbon content, slurry oil aromatic hydrocarbon content, slurry oil hydrogen content, slurry oil carbon contain
Amount, slurry oil asphaltene colloid content and settler outlet temperature.
Step (1) of the present invention or (3) described pretreatment are specially：The data in the common period of parameters are taken, it is right
Parameters carry out the rejecting of exceptional value, the exceptional value include zero, negative value, empty data and and the difference of average value be more than 3 times
The data of standard deviation.
Step (1) the of the present invention normalized is：Sample sequence is set to { X (n) }, according to minimax method logarithm
According to being normalized, the minimax method formula is Xk=(Xi-Xmin)/(Xmax-Xmin), wherein, XkFor normalized value,
XiThe measured data value obtained for step (1), XmaxAnd XminIt is the maximum and minimum value of { X (n) } respectively.
Step (2) the of the present invention neutral net is selected from anti-phase propagation (Back Propagation, BP) neutral net, wide
Adopted recurrent neural networks (Generalized Regression Neural Network, GRNN), RBF (Radial
Basis Function, RBF) neutral net, population (Particle Swarm Optimization, PSO) optimization BP nerve
Network or genetic algorithm (Genetic Algorithm, GA) Optimized BP Neural Network；Preferably GRNN neutral nets.
BP neural network described in step (2) of the present invention, has well as a kind of Nonlinear Modeling and Forecasting Methodology
Non-linear quality, high fitting precision and extensive function, are a kind of multilayer feedforward neural networks of one way propagation, the network
It is mainly characterized by signal propagated forward, error back propagation.X1, X2..., XnIt is the input value of BP neural network, Y1, Y2..., Ym
It is the predicted value of BP neural network, wijWithFor BP neural network weights.Input signal from input layer through hidden layer successively from
Reason, until output layer.One layer of neuron state under the influence of each layer of neuron state.If output layer cannot be expected defeated
Go out, be then transferred to backpropagation, network weight and threshold value are adjusted according to predicated error, so that BP neural network prediction output is continuous
Approach desired output.
The BP neural network is trained：BP neural network includes input layer, output layer and each one layer of hidden layer；Its
In, basic data after normalized obtained by step (1) as input layer, condensation rate, capture rate and coking yield respectively as
Output layer, actual value is used as desired output；Node in hidden layer selects reference formula
In formula, m is input number of nodes, preferably 25, n be output node number, preferably 1, be respectively compared training pattern under different nodes
The mean square error of acquisition, finds out optimal the number of hidden nodes H；Tangent S type function tansig conducts are used between input layer and hidden layer
Transmission function, with selectively acting function Lin as the transmission function between output layer and hidden layer, with function input and output number
According to training BP neural network, the network fit non-linear function trained is set to export.
The BP neural network training also includes BP neural network initialization step, and the parameter of the preferably step is：In step
Suddenly 99% group of data is randomly choosed in the data obtained by (1) as training sample, training sample is carried out using BP neural network
Model training, the iterations of setting is 100 times, and learning rate is 0.2, and desired value is 0.00004；When the error of iteration result is small
In allowable error 0.001~0.00001, system finishing iterative calculation, model construction is completed.
Totally four layers of GRNN neutral nets described in step (2) of the present invention, including input layer, mode layer, summation layer and output
Layer；Wherein, the basic data obtained by step (1) after normalized is as input layer, condensation rate, capture rate and coking yield it is pre-
Measured value is equal to the number of learning sample, i.e. input sample as output layer corresponding with input layer, mode layer neuron number
Dimension, summation layer utilizes formula：
Carry out the summation that counts to the output of all mode layer neurons, the connection weight of its mode layer and each neuron is 1,
Transmission function is：
Neuron number in output layer is equal to the dimension of output vector in learning sample；In GRNN neural metwork trainings
In, number, the number of linear neuron of radial direction base neuron are identical with input vector number in input training sample.
The purpose of the GRNN neural metwork trainings is to generate suitable weight matrix LW1,1、LW2,1And threshold vector
b1, the training is specially：Input layer input vector is delivered to hidden layer, and hidden layer has Q neuron, and node function is Gauss
Function, input weight matrix is LW1,1, threshold vector is b1；Signal output layer is specific linear convergent rate layer, equally containing Q
Neuron, weight function is normalization dot product weight function, and node function is purely linear function, and corresponding weight value matrix is LW2,1, net
The output of network can represent y=a with following expression2=purelin (LW2,1×a1/sum(a1)；The specific linear convergent rate layer
There is following characteristic：Hidden layer output is not the input directly as linear neuron, but first by the output of hidden layer and this layer
Weight matrix LW2.1Make to be re-used as power input after normalization dot-product operation and be re-fed into transmission function, the transmission function is linear
Function.
The GRNN neural network predictions process can try to achieve the most optimized parameter i.e. optimal according to the program feature of itself
Spread values, then GRNN networks are built using optimal method, so cause dependent variable predicted value it is corresponding with sample because
Variable is closely.When optimal spread value=0.7, output valve is tried to achieve, model construction terminates.
RBF neural described in step (2) of the present invention is the three_layer planar waveguide of an only one of which hidden layer
Structure, its difference maximum compared with feedforward network is that hidden layer obtains the Gaussian function that transfer function is local acknowledgement, and simultaneously
The function of non-global response；The prototype function of the RBF neural is：Wherein, radbas
For RBF, generally Gaussian function.The RBF network transfer functions radbas is between weight vector and threshold vector
Apart from ‖ dist ‖ as independent variable, the ‖ dist ‖ are obtained by the product of input vector and the row vector of weighting matrix.
The RBF neural parameter is：Mean square error target, preferably 0.0；Spread is the extension of RBF
Speed, preferably 1.0；MN is the maximum number of neuron, preferably 26；DF is the neuron number that is added between showing twice
According to preferably 25.
Step (2) the PSO Optimized BP Neural Networks of the present invention, including input layer, output layer and each one layer of hidden layer；Its
In, basic data after normalized obtained by step (1) as input layer, condensation rate, capture rate and coking yield respectively as
The corresponding output layer of every group of input layer；Node in hidden layer selects reference formula：
In formula, m is input number of nodes, and n is output node number, is respectively compared under different nodes and trains and verify the mean square error that model is obtained
Difference, finds out optimal the number of hidden nodes；With tangent S type functions tansig as transmission function between input layer and hidden layer, with choosing
Selecting property action function Lin is used as the transmission function between output layer and hidden layer.
The PSO Optimized BP Neural Networks are trained：(a) initiation parameter is set, and the initiation parameter includes
Population Size, accelerator coefficient, learning rate, iterations, the iterations based on experience value with calculating speed and fitness
Result of calculation determine；Specifically, the initiation parameter includes：Initial population n=60, scale is m=30, iterations K
=100, individual and speed maximin popmax=1, popmin=-1, Vmax=6, Vmin=-6, speed undated parameter c2=
1.494；C1=1.494, distributes the weights and threshold value of BP networks；(b) setting end condition is：If iterations exceedes step
(a) value set, algorithm terminates；(c) each particle forward-propagating in a network, calculate each particle its output layer output,
Compare and obtain error；(d) in iterative process each time, each particle updates the speed of itself by individual extreme value and global extremum
Degree and position, find its optimal location PbestAdaptive value；(e) by the optimal location P of each particlebestAdaptive value and colony
History optimal location GbestAdaptive value be compared, if PbestAdaptive value be better than GbestAdaptive value, then using the value as
Colony optimal location Gbest, otherwise GbestKeep constant；(f) colony optimal location G is usedbestConstantly update the weights and threshold of network
Value；If (g) meeting end condition, stop algorithm；If being unsatisfactory for end condition, step (d) is turned to；When the mistake of iteration result
Difference is less than allowable error 0.001~0.00001, and system finishing iterative calculation, model construction is completed.
Step (2) the GA Optimized BP Neural Networks of the present invention, use the initial power of genetic algorithm optimization BP neural network
Value and threshold value, enable the BP neural network after optimization that function output is better anticipated；The genetic algorithm optimization BP nerve net
Network includes initialization of population, fitness function, selection operation, crossover operation and mutation operation.
The GA Optimized BP Neural Networks are trained：(a) initialization of population：Because each individual is contained in population
The threshold of connection weight and output layer between threshold value, hidden layer and the output layer of connection weight, hidden layer between input layer and hidden layer
Four numerical value of value, therefore each individual in population contains the full detail of whole neutral net；Initial population is preferably 40,
Iterations is preferably 100, and using the real coding different from traditional GA, gene is used as using each weights and threshold value of network；
(b) individual adaptation degree is trained:Neutral net is trained according to the weights and threshold value obtained by individual, training data is imported
Neutral net, obtains neural network prediction value, using the inverse of predicted value and the mean square error of desired output as individual adaptation degree,
Calculating formula is shown below：
In formula：FiFor the fitness of i-th of individual, yiFor neural network prediction value, y0For desired output, i.e., actual knot
Jiao Liang, N are training number；(c) selection operation：Fitness ratio is based on roulette method and selects individual, and specific formula is：
In formula：K is coefficient, fiFor based on the fitness value under fitness ratio selection strategy, PiFor based on fitness ratio
The select probability of each individual under selection strategy；(d) crossover operation：Intersection is the Main Means for producing new individual, but at random
Individual intersection operation is selected, effective gene may be caused to lack, so as to extend the time of search optimal solution；Crossover probability value is
0.6~0.9, preferably 0.7；(e) mutation operation：Mutation operator is introduced in genetic algorithm, initial population can be provided and be free of
Gene, or give the gene lost in selection course for change, new content be provided for population；Mutation probability value be 0.001~
0.100, preferably 0.007；When the error of iteration result is less than allowable error 0.001~0.00001, system finishing iteration meter
Calculate, model construction is completed.
The data that step (2) of the present invention is also obtained using step (1) are to prediction condensation rate, capture rate, coking yield
Model verified.
Slurry oil mass flow × condensation rate predicted value × is caught in step (3) settler coking amount=settler of the present invention
Obtain rate predicted value × coking yield predicted value；Slurry oil mass flow is obtained by in-site measurement in the settler.
The method for the on-line prediction RFCC settler coking amount that the present invention is provided, is split by analyzing heavy-oil catalytic
Change settler Coking Mechanism, Binding experiment simulation obtains condensation rate, capture rate, the basic data of coking yield, utilizes neutral net
Model training is carried out to experimental data respectively, condensation rate, capture rate, coking yield model are obtained respectively, and model is tied to influence
Burnt online data is calculated, condensation rate, capture rate and the coking yield of on-line prediction RFCC settler, so as to realize
The coking situation of on-line prediction coking amount, at any time judgment means.The method that the present invention is provided is easy, easy to operate, and prediction is accurate, can
To avoid the unplanned emergency shutdown of device caused by device coking excessively, the loss of the business efficiency thereby resulted in is reduced.
Embodiment
Following examples are used to illustrate the present invention, but are not limited to the scope of the present invention.
Various embodiments of the present invention utilize neutral net, according to each neutral net in the content of the invention it is corresponding explanation, step with
And preferred parameter, prediction condensation rate, coking yield, the model of capture rate are built, the refinery catalytic cracking precipitator devices of A are tied
The prediction of Jiao's amount.The all experimentss data that step (1) is related in various embodiments of the present invention are from A refineries collection in worksite, through invention
After content step (1) pretreatment and normalized, the data related with coking yield to condensation rate, capture rate are respectively obtained
Each 100 groups of (including condensation rate, capture rate, coking yield actual value), by length is limited, only lists partial data herein, referring to
Table 1~3.
Table 1：The partial data related to condensation rate
Table 2：The partial data related to capture rate
Table 3：The partial data related to coking yield
Embodiment 1
What coking amount was predicted comprises the following steps that：
(1) basic data and condensation rate, capture rate, coking yield reality related to condensation rate, capture rate, coking yield is obtained
Actual value；Partial data is referring to table 1~3；
(2) in 100 groups of data that step (1) is obtained, 60 groups of random selection is as training data, and 40 groups are used as checking
Data, described in the content of the inventionBP neural networkObtain the model of Predicted settlement device condensation rate, capture rate and coking yield；
(3) basic data of collection in worksite RFCC settler and pre-processed, by it is preprocessed obtain 1
Group data are substituted into the condensation rate, capture rate, coking yield model of step (2) acquisition respectively, obtain settler condensation rate, capture rate
With the predicted value of coking yield；Above-mentioned data are substituted into formula by slurry oil mass flow in collection in worksite settler：Settler coking amount
Slurry oil mass flow × condensation rate predicted value × capture rate predicted value × coking yield predicted value in=settler, calculating is settled
Device coking amount predicted value.
Predicting the outcome for settler coking amount is shown in Table 4.
Embodiment 2
Differ only in compared with Example 1：In 100 groups of basic datas, 80 groups of random selection is used as training data, 20
Group is as checking data, described in the content of the inventionGRNN neutral netsBuild prediction condensation rate, capture rate, the mould of coking yield
Type；Coking amount to the refinery catalytic cracking precipitator devices of A is predicted；Predict the outcome and be shown in Table 4.
Embodiment 3
Differ only in compared with Example 1：In 100 groups of basic datas, 80 groups of random selection is used as training data, 20
Group is as checking data using described in the content of the inventionRBF neuralBuild condensation rate, capture rate, coking yield model；A is refined
The coking amount of factory's catalytic cracking subsider device is predicted；Predict the outcome and be shown in Table 4.
Embodiment 4
Differ only in compared with Example 1：In 100 groups of basic datas, 80 groups of random selection is used as training data, 20
Group is as checking data, described in the content of the inventionPSO Optimized BP Neural NetworksBuild condensation rate, capture rate, coking yield mould
Type；Coking amount to the refinery catalytic cracking precipitator devices of A is predicted；Predict the outcome and be shown in Table 4.
Embodiment 5
Differ only in compared with Example 1：In 100 groups of basic datas, 80 groups of random selection is used as training data, 20
Group is as checking data, described in the content of the inventionGA Optimized BP Neural NetworksBuild condensation rate, capture rate, coking yield mould
Type；Coking amount to the refinery catalytic cracking precipitator devices of A is predicted；Predict the outcome and be shown in Table 4.
Table 4：The predicted value and actual value of the coking amount of the refinery catalytic cracking precipitator devices of A
Actual coking value is obtained after being overhauled to A refineries precipitator device, passes through contrast, the embodiment of the present invention
1~5 obtained coking amount of prediction is respectively 8.64,9.04,8.58,8.12 and 9.5, each predicted value and actual coking amount 9.13 it
Between error be respectively less than 10%；Wherein, between the gained predicted value of embodiment 2 and actual value absolute error is minimum, is only
0.986%.From the above results, the accuracy rate of RFCC settler coking amount Forecasting Methodology that the present invention is provided compared with
Height, can be applied in actual industrial production.
Although above having made to retouch in detail to the present invention with general explanation, embodiment and experiment
State, but on the basis of the present invention, it can be made some modifications or improvements, this is apparent to those skilled in the art
's.Therefore, these modifications or improvements without departing from theon the basis of the spirit of the present invention, are belonged to claimed
Scope.
Claims (4)
1. a kind of method for predicting RFCC settler coking amount, it is characterised in that the described method comprises the following steps：
(1) basic data related to the condensation rate, capture rate, coking yield of settler, and condensation rate, capture rate and life are obtained
The actual value of burnt rate, is pre-processed, then be normalized to above-mentioned each group of data；
The basic data related to condensation rate includes parameter in detail below：Slurry oil initial boiling point, slurry oil 575 DEG C of quantity of distillate, oil
Starch 10% recovered (distilled) temperature, the recovered (distilled) temperature of slurry oil 50%, the recovered (distilled) temperature of slurry oil 90%, slurry oil in settler outlet temperature and settler
Partial pressure；
The basic data related to capture rate includes parameter in detail below：Oil gas linear speed, settler outlet in settler
Temperature, settler internal pressure, slurry oil density and slurry oil carbon residue；
The basic data related to coking yield includes parameter in detail below：Slurry oil carbon residue, slurry oil density, slurry oil mean molecule
Amount, slurry oil saturated hydrocarbon content, slurry oil aromatic hydrocarbon content, slurry oil H content, slurry oil C content, slurry oil asphaltene colloid content and sedimentation
Device outlet temperature；
(2) model training is carried out using neutral net, input is used as using the basic data after normalized obtained by step (1)
Value, using actual value as desired output, obtains the model of Predicted settlement device condensation rate, capture rate and coking yield respectively；
(3) basic data related to the condensation rate, capture rate, coking yield of settler to collection in worksite is pre-processed, point
The predicted value of settler condensation rate, capture rate and coking yield Dai Ru not be obtained, collection in worksite sinks in the model that obtains of step (2)
Slurry oil mass flow in device drops, and calculating obtains settler coking amount predicted value；The settler coking amount predicted value is by following public affairs
Formula is calculated and obtained：Slurry oil mass flow × condensation rate predicted value × capture rate predicted value × life in settler coking amount=settler
Burnt rate predicted value.
2. according to the method described in claim 1, it is characterised in that step (1) or (3) described pretreatment are specially：Take each
Parameters are carried out the rejecting of exceptional value by the data in the common period of parameter, and the exceptional value includes zero, negative value, sky
Data and and the difference of average value be more than the data of 3 times of standard deviations.
3. according to the method described in claim 1, it is characterised in that step (1) described normalized is：Sample sequence is determined
For { X (n) }, data are normalized according to minimax method, the minimax method formula is Xk=(Xi-Xmin)/
(Xmax-Xmin), wherein, XkFor normalized value, XiThe data value obtained for step (1), XmaxAnd XminIt is the maximum of { X (n) } respectively
Value and minimum value.
4. the method according to claims 1 to 3 any one, it is characterised in that step (2) described neutral net is selected from BP
Neutral net, GRNN neutral nets, RBF neural, PSO Optimized BP Neural Networks or GA Optimized BP Neural Networks.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410584065.5A CN104361153B (en) | 2014-10-27 | 2014-10-27 | A kind of method for predicting RFCC settler coking amount |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410584065.5A CN104361153B (en) | 2014-10-27 | 2014-10-27 | A kind of method for predicting RFCC settler coking amount |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104361153A CN104361153A (en) | 2015-02-18 |
| CN104361153B true CN104361153B (en) | 2017-11-03 |
Family
ID=52528413
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410584065.5A Active CN104361153B (en) | 2014-10-27 | 2014-10-27 | A kind of method for predicting RFCC settler coking amount |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104361153B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108664676A (en) * | 2017-03-31 | 2018-10-16 | 中国石油天然气股份有限公司 | Catalytic cracking process modeling method and catalytic cracking process prediction method |
| CN107301264A (en) * | 2017-05-03 | 2017-10-27 | 中国石油大学(北京) | The evaluation method and system of a kind of catalytic cracking reaction kinetic parameter |
| CN108121857A (en) * | 2017-12-08 | 2018-06-05 | 北京神雾电力科技有限公司 | A kind of method for predicting down-flow fluidized bed using ECT fast pyrogenation furnace radiating pipe rate of depreciation |
| CN108120451A (en) * | 2017-12-21 | 2018-06-05 | 苏州大学 | Based on silicon micro accerometer temperature-compensation method, the system for improving PSO optimization neural networks |
| CN110717235A (en) * | 2018-06-26 | 2020-01-21 | 中国石油化工股份有限公司 | Coking prediction method for settler of catalytic cracking unit |
| CN109934417B (en) * | 2019-03-26 | 2023-04-07 | 国电民权发电有限公司 | Boiler coking early warning method based on convolutional neural network |
| CN112730545B (en) * | 2020-12-22 | 2021-10-15 | 华中科技大学 | Online prediction method and system for electrode coking amount in biomass pyrolysis oil electrolysis process |
| CN113189881A (en) * | 2021-05-11 | 2021-07-30 | 华东理工大学 | Multi-objective optimization control method and system for sewage treatment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1456877A (en) * | 2002-05-09 | 2003-11-19 | 石油大学(北京) | Device and method for determining dynamic scaling and drossing trend of catalystic cracking oi slurry |
| CN103235986A (en) * | 2013-05-03 | 2013-08-07 | 上海发电设备成套设计研究院 | Operation and consumption optimization method based on boiler safety analysis |
| CN103679268A (en) * | 2012-09-14 | 2014-03-26 | 宝钢不锈钢有限公司 | Blast furnace slag viscosity prediction method |
-
2014
- 2014-10-27 CN CN201410584065.5A patent/CN104361153B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1456877A (en) * | 2002-05-09 | 2003-11-19 | 石油大学(北京) | Device and method for determining dynamic scaling and drossing trend of catalystic cracking oi slurry |
| CN103679268A (en) * | 2012-09-14 | 2014-03-26 | 宝钢不锈钢有限公司 | Blast furnace slag viscosity prediction method |
| CN103235986A (en) * | 2013-05-03 | 2013-08-07 | 上海发电设备成套设计研究院 | Operation and consumption optimization method based on boiler safety analysis |
Non-Patent Citations (2)
| Title |
|---|
| Modelling of an Industrial Fluid Catalytic Cracking Unit Using Neural Networks;J.MICHALOPOULOS，等;《Chemical Engineering Research & Design》;20010331;第79卷(第2期);正文第1-5节 * |
| 重油催化裂化***安全分析与关键风险评价研究;张进春;《中国优秀硕士学位论文全文数据库 工程科技I辑》;20071215(第06期);第55-61页第4.3部分 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104361153A (en) | 2015-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104361153B (en) | A kind of method for predicting RFCC settler coking amount | |
| Motlaghi et al. | An expert system design for a crude oil distillation column with the neural networks model and the process optimization using genetic algorithm framework | |
| Ochoa-Estopier et al. | Operational optimization of crude oil distillation systems using artificial neural networks | |
| CN103713604B (en) | A kind of industrial pyrolysis furnace real time operation optimizing based on data-driven and control method | |
| CN108764540B (en) | Water supply network pressure prediction method based on parallel LSTM series DNN | |
| CN109815616B (en) | Molecular-level hydrogenation device simulation and optimization method | |
| Ansari et al. | An improved support vector regression model for estimation of saturation pressure of crude oils | |
| CN102626557B (en) | Molecular distillation process parameter optimizing method based on GA-BP (Genetic Algorithm-Back Propagation) algorithm | |
| CN103105246A (en) | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm | |
| CN104463327A (en) | Method for predicting catalytic cracking coke yield | |
| CN104484714A (en) | Real-time prediction method for catalytic reforming device | |
| CN104463343A (en) | Method for predicting catalytic cracking light oil yield | |
| CN104392098A (en) | Method for predicting yield of catalytically cracked gasoline | |
| Alhajree et al. | Modeling and optimization of an industrial hydrocracker plant | |
| CN105740960B (en) | A kind of optimization method of industry hydrocracking reaction condition | |
| Baklacioglu et al. | Metaheuristic approach for an artificial neural network: exergetic sustainability and environmental effect of a business aircraft | |
| CN111027733A (en) | Petrochemical device product yield optimization method based on big data technology | |
| Gueddar et al. | Disaggregation–aggregation based model reduction for refinery-wide optimization | |
| Li et al. | Prediction model of weekly retail price for eggs based on chaotic neural network | |
| CN117012315A (en) | Concrete strength prediction method for optimizing RBF neural network | |
| Vavilova et al. | Application of genetic algorithm for adjusting the structure of multilayered neural network for prediction of investment processes | |
| CN110045236A (en) | Transformer state parametric data prediction technique and system based on core pivot element analysis optimization | |
| CN108803343A (en) | The control method and device of methanol to olefins reaction device | |
| Zahedi Abghari et al. | Determination of suitable operating conditions of fluid catalytic cracking process by application of artificial neural network and firefly algorithm | |
| Bispo et al. | Modeling, optimization and control of a FCC unit using neural networks and evolutionary methods |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |