CN106281431A - The real-time predicting method of a kind of hydrocracking unit reaction depth and device - Google Patents
The real-time predicting method of a kind of hydrocracking unit reaction depth and device Download PDFInfo
- Publication number
- CN106281431A CN106281431A CN201610738881.6A CN201610738881A CN106281431A CN 106281431 A CN106281431 A CN 106281431A CN 201610738881 A CN201610738881 A CN 201610738881A CN 106281431 A CN106281431 A CN 106281431A
- Authority
- CN
- China
- Prior art keywords
- attribute
- reaction depth
- hydrocracking unit
- operation phase
- oil
- 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.)
- Granted
Links
Classifications
-
- C—CHEMISTRY; METALLURGY
- C10—PETROLEUM, GAS OR COKE INDUSTRIES; TECHNICAL GASES CONTAINING CARBON MONOXIDE; FUELS; LUBRICANTS; PEAT
- C10G—CRACKING HYDROCARBON OILS; PRODUCTION OF LIQUID HYDROCARBON MIXTURES, e.g. BY DESTRUCTIVE HYDROGENATION, OLIGOMERISATION, POLYMERISATION; RECOVERY OF HYDROCARBON OILS FROM OIL-SHALE, OIL-SAND, OR GASES; REFINING MIXTURES MAINLY CONSISTING OF HYDROCARBONS; REFORMING OF NAPHTHA; MINERAL WAXES
- C10G47/00—Cracking of hydrocarbon oils, in the presence of hydrogen or hydrogen- generating compounds, to obtain lower boiling fractions
- C10G47/36—Controlling or regulating
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
Abstract
The present invention provides real-time predicting method and the device of a kind of hydrocracking unit reaction depth.The method includes: divided working status builds steady state data collection, i.e. utilizes the method for similarity measurement to concentrate from historical data and filters out the service data collection corresponding to each steady state condition, and according to plant running stage, the different classification storages of mixed feeding oil attribute；Neutral net optimum training sample set preferred operations, i.e. with the feed oil attribute of current working as foundation, similarity size according to mixing index is concentrated from above-mentioned hydrocracking unit steady state condition service data and is chosen optimum training sample, completes the Fast Training of neutral net；Reaction depth real-time estimate operates.The present invention predicts the current operation phase, currently feeds the reaction depth of hydrocracking unit under attribute, current operational conditions; optimal control for this process quenching hydrogen injection rate provides real time status information, is guard catalyst activity, extension fixture on-stream time, realizes the basis that target product processing capacity is adjusted flexibly.
Description
Technical field
The present invention relates to technical field of data processing, be specifically related to the real-time estimate of a kind of hydrocracking unit reaction depth
Method and device.
Background technology
Petroleum refining industry is the mainstay of the national economy, and its total output value accounts for about 1/3rd of gross national economy, with
Rapid development of economy, demand and the dependence of the energy are grown with each passing day by China, it is ensured that oil supply has become national economy
The core of security strategy.Existing especially into worldwide environmental problem frequency after 21 century, the mankind are to clean energy resource, environment
Protection cry improve constantly, therefore, can be effectively improved oil quality, produce low sulphur fuel oil hydrogen addition technology be increasingly subject to
The attention of countries in the world.Hydrogen addition technology mainly by hydrofinishing, be hydrocracked, be hydrogenated with reconstruct and residual hydrogenation etc. and face hydrogen and face
Catalyst treatment means realize the different of saturated, the normal hydrocarbon of hydrogenation of the removing of organic S, N, O impurity in raw oil, unsaturation hydrocarbon
Structure and the cracking of macromole hydrocarbon, and then reach the requirement of target oil product upgrading modification.
Be hydrocracked flow process mainly include hydrofinishing, be hydrocracked, high-low pressure separate and four main fortune of fractionating system
Row unit (as shown in Figure 1), it designs main operating mode is complete alternation flow process, with straight run light wax oil as primary raw material, mixes refining part and urges
Change diesel oil and circulation tail oil, be primary and secondary product dry gas, low point of air-liquid activating QI and pumice brain to produce heavy naphtha, boat coal and diesel oil
Oil；Take into account once by flow process simultaneously, with straight run light wax oil and wax tailings as primary raw material, mainly produce heavy naphtha, boat
Coal, diesel oil and tail oil, by-product dry gas, low point of air-liquid activating QI and light naphthar.In one-stage serial hydrocracking flow process, can be near
Seemingly thinking that cracking reaction occurs mainly in the 2nd reactor, dominant response can be roughly divided into (1) alkane and be hydrocracked instead
Should；(2) cycloalkane hydrocracking reaction；(3) aromatic hydrogenation cracking reaction；(4) the big class of hygrogenating isomerization reaction four.Cracking reaction
It is by the macromolecule hydrocarbon of high boiling point heaviness in the case of certain hydrogen dividing potential drop, air speed, reaction temperature, at the metal of catalyst
There is hydrogenation/dehydrogenation reaction in center, in the acid centre generation cracking reaction of catalyst, generates and meet target product component requirements
Hydrocarbon micromolecular.These hydrocarbon mixtures generated after cracking according to carbonium ion reaction mechanism are successively divided by high and low pressure
From device and fractionating system, final obtaining light, heavy naphtha fraction, the high-value product such as aerial kerosene fraction and diesel oil distillate (adds
The reaction network of hydrogen cracking process material conversion is as shown in Figure 2).
But, prior art mainly use " conversion ratio " this concept to being hydrocracked response system cracking zone in flow process
The extent of reaction measure, both only carried out dualistic analysis at this process outlet, the fraction of more than 350 DEG C is non-switched tail
Oil, the fraction of less than 350 DEG C is the product converted.The concept of conversion ratio is exactly (feed rate-tail oil flow)/feed stream
Amount * 100%, thus cannot directly reflect generate on earth under current operating condition how many liquefied gas component, light naphthar component,
Heavy naphtha component, boat coal component and diesel component.Prior art cannot directly predict each target product of hydrocracking unit
Yield, it is difficult to become more meticulous tolerance to the reaction depth of response system cracking zone, is unfavorable for the optimization of quenching hydrogen injection rate
Control, be also unfavorable for the further protection of catalyst activity simultaneously.
Summary of the invention
The embodiment of the present invention provides real-time predicting method and the device of a kind of hydrocracking unit reaction depth, is used for solving
Prior art is difficult to quenching hydrogen injection rate because lacking hydrocracking unit cracking zone running state information become more meticulous control
The problem of system, and then provide with reference to information for guard catalyst activity, yield that each sideline product of fractionating system be adjusted flexibly.
Embodiments provide the real-time predicting method of a kind of hydrocracking unit reaction depth, including:
The history data of hydrocracking unit is carried out pretreatment；
Use the method for similarity measurement to carry out stable state differentiation the history data through pretreatment, filter out stable state
Process real time data collection, by described steady-state process real time data collection according to being hydrocracked the operation phase of flow process, mixed feeding oil
The difference of attribute is classified, and forms N number of separate " operation phase-charging attribute-operating condition-outlet yield " data
Block；Wherein, N is the integer more than 0；
According to sideline product flow each in fractionating system, tail oil flow to hydrocracking unit specific run stage, spy
" operation phase-charging attribute-operating condition-outlet yield " data corresponding under the conditions of determining feed oil attribute, specific operation
Block carries out the calculated off line of cracking zone reaction depth and stores；
Calculate the reaction corresponding to each operation phase of hydrocracking unit, various charging attribute, all kinds of operating condition deep
Degree, obtains N number of separate " operation phase-charging attribute-operating condition-reaction depth " data block；
Operation phase of being presently according to hydrocracking unit and currently feed attribute, choose and current charging attribute phase
Multiple " charging attribute-operating condition-reaction depth " data block of predetermined threshold value, the operation of the data block to choose it is more than like degree
Condition is input, and reaction depth is output, Fast Training BP neural network model；
Utilize the BP neural network model set up, the real-time estimate hydrocracking unit current operation phase, currently feed
The lower reaction depth being hydrocracked flow process response system cracking zone that can reach of attribute, current operational conditions.
Alternatively, described history data includes:
It is hydrocracked the Temperature Distribution of the flow process each bed of response system cracking zone, the air speed of all kinds of catalyst, fractionating system
In each sideline product flow, tail oil flow and determine mixed feeding oil attribute various feed oil flows.
Alternatively, the described history data to hydrocracking unit carries out pretreatment, including:
The history data of hydrocracking unit is carried out outlier rejecting and Wavelet Denoising Method.
Alternatively, the differentiation of described steady state data includes with extracting method:
Process operation data are after PCA extracts key message, by contrasting the phase of adjacent sequential load matrix
The process data corresponding to steady state condition is selected like degree.
Alternatively, described according to sideline product flow each in fractionating system, tail oil flow to hydrocracking unit specific
" operation phase-charging attribute-operating condition-go out corresponding under the conditions of operation phase, specific feed oil attribute, specific operation
Mouthful yield " data block carries out the calculated off line of cracking zone reaction depth and stores, including:
Hydrocracking unit is calculated specific run stage, specific feed oil attribute, spy according to formula (1) and formula (2)
Determine the reaction depth of cracking zone under operating condition:
ηk,i,j=w1x1+w2x2+w3x3+w4x4+w5x5+w6x6Formula (1)
Wherein, ηk,i,jRepresenting that hydrocracking unit is i at kth operation phase charging attribute, operating condition is the situation of j
Under the reaction depth of whole response system cracking zone；w1, w2, w3, w4, w5, w6Represent liquefied gas, light naphthar, scheelite brain respectively
Oil, aerial kerosene, diesel oil, the discrete lumped component of tail oil 6 class are for the weight coefficient of reaction depth；x1, x2, x3, x4, x5, x6Point
Do not represent liquefied gas, light naphthar, heavy naphtha, aerial kerosene, diesel oil, the mass percent of the discrete lumped component of tail oil 6 class；Respectively represent be hydrocracked mixed feeding oil in flow process, liquefied gas, light naphthar,
Heavy naphtha, aerial kerosene, diesel oil, the average molecular mass of the discrete lumped component of tail oil six class.
Alternatively, calculate corresponding to each operation phase of hydrocracking unit, various feed oil attribute, all kinds of operating condition
Reaction depth, obtain N number of separate " operation phase-charging attribute-operating condition-reaction depth " data block.
Alternatively, the described operation phase being presently according to hydrocracking unit and current feed oil attribute, choose with
Current feed oil attributes similarity is more than multiple " charging attribute-operating condition-reaction depth " data block of predetermined threshold value, bag
Include:
The operation phase being presently according to hydrocracking unit, from N number of " operation phase-charging attribute-operating condition-
Reaction depth " first data block is screened by " operation phase ", constitute M separate " entering under the current operation phase
Material attribute-operating condition-reaction depth " data block；Wherein, M and N is respectively the integer more than 0, and N > M；
According to current feed oil attribute, respectively in " charging attribute-operating condition-reaction depth " data block individual with above-mentioned M
Charging attribute carry out similarity measurement, choose with current feed oil attributes similarity more than predetermined threshold value multiple " charging belong to
Property-operating condition-reaction depth " and data block, Fast Training neutral net.
Alternatively, described basis current feed oil attribute, " feed attribute-operating condition-reaction deep with above-mentioned M respectively
Degree " charging attribute in data block carries out similarity measurement, including:
According to current feed oil attribute, respectively in " charging attribute-operating condition-reaction depth " data block individual with above-mentioned M
Charging attribute data carry out similarity measurement according to being weighted, by Euclidean distance and co sinus vector included angle value, the mixing index constituted.
Alternatively, utilize described BP neural network model, the real-time estimate hydrocracking unit current operation phase, work as advance
Material attribute, the lower reaction depth being hydrocracked flow process response system cracking zone that can reach of current operational conditions, including:
According to being hydrocracked the on-line checking value of flow process response system cracking zone performance variable, by described BP neutral net
Model, the real-time estimate hydrocracking unit current operation phase, currently feed attribute, current operational conditions lower can reach add
The reaction depth of hydrogen cracking flow process response system cracking zone；
Wherein, the on-line checking value of described performance variable includes: cracker each reaction bed Temperature Distribution current
Value and the currency of all kinds of catalyst space velocities.
Embodiments provide the real-time estimate device of a kind of hydrocracking unit reaction depth, including:
Data pre-processing unit, for carrying out pretreatment to the history data of hydrocracking unit；
First data block forms unit, for the method that the history data through pretreatment is used similarity measurement
Carry out stable state differentiation, filter out steady-state process real time data collection, by described steady-state process real time data collection according to being hydrocracked stream
The operation phase of journey, the difference of mixed feeding oil attribute are classified, and form N number of separate " operation phase-charging genus
Property-operating condition-outlet yield " data block；Wherein, N is the integer more than 0；
Reaction depth calculated off line unit, is used for according to sideline product flow each in fractionating system, tail oil flow hydrogenation
" operation phase-charging genus that cracking unit is corresponding under the conditions of specific run stage, specific feed oil attribute, specific operation
Property-operating condition-outlet yield " data block carries out the calculated off line of cracking zone reaction depth and stores；
Second data block forms unit, is used for calculating each operation phase of hydrocracking unit, various charging attribute, all kinds of
Reaction depth corresponding to operating condition, obtains that N number of separate " operation phase-charging attribute-operating condition-reaction is deep
Degree " data block；
Neural network model training unit, for the operation phase being presently according to hydrocracking unit and current charging
Attribute, chooses and counts more than multiple " the charging attribute-operating condition-reaction depths " of predetermined threshold value with currently charging attributes similarity
According to block, with the operating condition of data block chosen for input, reaction depth is output, Fast Training BP neural network model；
Reaction depth predicting unit, is used for utilizing described BP neural network model, and real-time estimate hydrocracking unit is current
Operation phase, currently feed attribute, current operational conditions lower can reach be hydrocracked the anti-of flow process response system cracking zone
Answer the degree of depth.
The real-time predicting method of the hydrocracking unit reaction depth that the embodiment of the present invention provides and device, by optimal instruction
Practice sample set optimal way and filter out (i.e. current operation phase) and current mixed feeding oil under the conditions of present catalyst activity
The immediate some historical data sample collection of attribute carry out the Fast Training of BP neural network model, wherein choose and are hydrocracked stream
The Temperature Distribution of each reaction bed of journey response system cracking zone, the air speed of all kinds of catalyst are input, and reaction depth is output.Should
Forecast model is first according to operation phase of device and eliminates the interference of a large amount of extraneous data, then according to the mixing of conditional attribute
Index is preferred, and the some historical data sample the most close with current feed oil attribute carry out model construction, and this not only greatly reduces
The training burden of neutral net, it is achieved that Fast Training, and avoid neutral net in the training process because of the most unrelated
The introducing of sample and the not convergence problem that causes.The real-time estimate current operation phase, currently feed under attribute, current operational conditions
The reaction depth of hydrocracking unit, can be that the optimal control of this process quenching hydrogen injection rate provides real time status information, be to protect
Demonstrate,prove catalyst activity, extension fixture on-stream time, realize the basis that target product processing capacity is adjusted flexibly.
Accompanying drawing explanation
In order to be illustrated more clearly that the embodiment of the present invention or scheme of the prior art, below will be to embodiment or existing skill
In art description, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is the one of the present invention
A little embodiments, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to according to these
Accompanying drawing obtains other accompanying drawing.
Fig. 1 is existing hydrocracking unit simplification of flowsheet figure；
Fig. 2 is existing to be hydrocracked lumped reaction schematic network structure；
Fig. 3 is the schematic flow sheet of the real-time predicting method of one embodiment of the invention hydrocracking unit reaction depth；
Fig. 4 is the flow process signal of the real-time predicting method of another embodiment of the present invention hydrocracking unit reaction depth
Figure；
Fig. 5 is the structural representation of the BP neutral net of one embodiment of the invention；
Fig. 6 is the real-time estimate result figure of the hydrocracking unit reaction depth of one embodiment of the invention；
Fig. 7 is the structural representation of the real-time estimate device of the hydrocracking unit reaction depth of one embodiment of the invention
Figure.
Detailed description of the invention
For making the purpose of the embodiment of the present invention, technical scheme and advantage clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is carried out clear, complete description, it is clear that described embodiment is
The a part of embodiment of the present invention rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art
The every other embodiment obtained under not making creative work premise, broadly falls into the scope of protection of the invention.
The reaction depth of the embodiment of the present invention refers in the specific run stage, specific feed oil attribute, specific operation bar
Under part, cracker exit all kinds of target product component the weighted sum calculated according to certain proportion.It is hydrocracked stream
The change of journey target product yield depends primarily on the reaction depth of response system cracking zone, for safety, economic dispatch factor
Considering, it is constant that hydrogen dividing potential drop is usually maintained in about 15MPa, therefore, and the Temperature Distribution of each bed of reactor, the stream of mixed feeding oil
Amount (being characterized as air speed in reactor) just becomes regulation and control reaction depth means the most flexible, maximally effective.Reaction depth is too high, then first
The imflammable gas yield such as alkane, ethane, ethylene are excessive, and hydrogen gas consumption is also too much, is not the optimal operating condition of this flow process
Point；Reaction depth is the lowest, then the target hydrocarbon molecule relative molecular mass ultimately generated is excessive, Petroleum, boat coal, diesel component phase
To less, it it not the most optimal economic optimization running status.Generally, Petrochemical Enterprises is always wished on the premise of ensureing product quality
Improve liquid component yield as far as possible, i.e. ensure the boiling range of each target product component, flash-point, smoke point, octane number, hexadecane
Acquisition Petroleum as much as possible, the boat liquid component such as coal, diesel oil on the premise of the quality index such as value.It is hydrocracked flow process reaction
The reaction depth of system cracking zone determines whole flow process Petroleum, the distribution situation of the boat liquid component such as coal, diesel oil, examines in real time
Survey the status information of cracking zone reaction depth, not only contribute to the Optimal Setting of subsequent fractionation tower each sideline product extracted amount, also
Be conducive to the Precise control of cracking zone reaction depth, for optimal control quenching hydrogen injection rate, guard catalyst activity, extend
Device on-stream time has obvious directive significance.
Fig. 3 is the schematic flow sheet of the real-time predicting method of one embodiment of the invention hydrocracking unit reaction depth.
As it is shown on figure 3, the real-time predicting method of the hydrocracking unit reaction depth of this embodiment includes:
S31: the history data of hydrocracking unit is carried out pretreatment；
S32: use the method for similarity measurement to carry out stable state differentiation the history data through pretreatment, filter out
Steady-state process real time data collection, by described steady-state process real time data collection according to being hydrocracked the operation phase of flow process, being mixed into
The difference of material oil attribute is classified, and forms N number of separate " operation phase-charging attribute-operating condition-outlet yield "
Data block；Wherein, N is the integer more than 0；
S33: according to sideline product flow each in fractionating system, tail oil flow to hydrocracking unit on specific run rank
" operation phase-charging attribute-operating condition-outlet yield " corresponding under the conditions of section, specific feed oil attribute, specific operation
Data block carries out the calculated off line of cracking zone reaction depth and stores；
S34: calculate corresponding to each operation phase of hydrocracking unit, various charging attribute, all kinds of operating condition is anti-
Answer the degree of depth, obtain N number of separate " operation phase-charging attribute-operating condition-reaction depth " data block；
S35: operation phase of being presently according to hydrocracking unit and currently feed attribute, choose and currently feed genus
Property similarity more than multiple " charging attribute-operating condition-reaction depth " data block of predetermined threshold value, with the data block chosen
Operating condition is input, and reaction depth is output, Fast Training BP neural network model；
S36: utilize the BP neural network model set up, the real-time estimate hydrocracking unit current operation phase, current
Charging attribute, the lower reaction depth being hydrocracked flow process response system cracking zone that can reach of current operational conditions.
The real-time predicting method of the hydrocracking unit reaction depth of the embodiment of the present invention, excellent by optimum training sample set
Select mode to filter out under the conditions of present catalyst activity (i.e. current operation phase) if immediate with current feed oil attribute
Dry historical data sample collection carries out the Fast Training of BP neural network model, wherein chooses and is hydrocracked flow process response system cracking
The section Temperature Distribution of each reaction bed, the air speed of all kinds of catalyst are input, and reaction depth is output.First this forecast model is pressed
Operation phase according to device eliminates the interference of a large amount of extraneous data, then according to the mixing index of conditional attribute preferred with work as
Some historical data sample that front feed oil attribute is the most close carry out model construction, and this not only greatly reduces the instruction of neutral net
Practice burden, it is achieved that Fast Training, and avoid neutral net and lead because of the introducing of a large amount of unrelated samples in the training process
The not convergence problem caused.The real-time estimate current operation phase, currently feed hydrocracking unit under attribute, current operational conditions
Reaction depth, can be that the optimal control of this process quenching hydrogen injection rate provides real time status information, be to ensure that catalyst activity, prolong
Growth device on-stream time, realize the basis that target product processing capacity is adjusted flexibly.
Specifically, described history data includes:
It is hydrocracked the Temperature Distribution of the flow process each bed of response system cracking zone, the air speed of all kinds of catalyst, fractionating system
In each sideline product flow, tail oil flow and determine mixed feeding oil attribute various feed oil flows.
Fig. 4 is the flow process signal of the real-time predicting method of another embodiment of the present invention hydrocracking unit reaction depth
Figure.As shown in Figure 4, the embodiment of the present invention carries out pretreatment to the history data of hydrocracking unit, including:
The history data of hydrocracking unit is carried out outlier rejecting and Wavelet Denoising Method.
In actual applications, the embodiment of the present invention, according to the DCS configuration structure of hydrocracking unit, is extracted in this flow process anti-
The history run process datas such as the air speed answering the Temperature Distribution of each bed of system cracking zone, each catalyst, each side in fractionating system
Line product (including liquefied gas, light naphthar, heavy naphtha, aerial kerosene, diesel oil) flow and the history run mistake of tail oil flow
Number of passes evidence, and the upstream device data on flows of real-time characterization mixed feeding oil attribute (each atmospheric and vacuum distillation system can be included
VGO flow, tank field wax oil flow and circulation tail oil flow), reject through outlier, build hydrocracking unit after Wavelet Denoising Method
Essential history running real-time data base.
Through various kinds of sensors from what the process operation data that hydrocracking unit operation scene obtains not only included low frequency it is
System signal, but also comprise the random noise of high frequency and all kinds of outlier introduced in detection or transmitting procedure, for guaranteeing number
According to the true operation conditions that can correctly reflect hydrocracking process, it is necessary to improper data are rejected.Wavelet transformation according to
Signal, noise present the principle of different qualities at different scale, utilize the flexible of wavelet function and translation functions to realize signal and exist
Time domain and the Multiresolution Decomposition of frequency domain, specifically comprise the following steps that
Step1: chooseAs wavelet mother function, noisy process data will be contained
By formulaResolve into the component in different time domain and frequency domain；
Wherein, a is the contraction-expansion factor of wavelet mother function, and τ is the shift factor of wavelet mother function, cj,kFor scale coefficient,
dj,hFor wavelet coefficient, j is Decomposition order, hn-2kAnd gn-2kFor orthogonal mirror image filtering group, n is orthogonal mirror image filtering group median filter
Number；
Step2: to wavelet coefficient dj,hAccording to hard threshold functionIt is filtered processing, wherein
dj,hFor the wavelet coefficient of signal decomposition, d'j,hFor the estimated value of wavelet coefficient, β is threshold value；
Step3: the wavelet coefficient after threshold filtering processes is carried out signal reconstruction, obtains being hydrocracked dress after denoising
Put running data.
Further, the differentiation of described steady state data includes with extracting method:
Process operation data are after PCA extracts key message, by contrasting the phase of adjacent sequential load matrix
The process data corresponding to steady state condition is selected like degree.
It should be noted that for the whole each main device of flow process that is hydrocracked acquired after above-mentioned filtered process
History run process data, for overcoming the shadow to cracking zone reaction depth computational accuracy of the time lag between response system and fractionating system
Ring, need to calculate with the steady state data of whole process the reaction depth of hydrocracking unit, hydrogenation based on similarity measurement
Cracking process steady state detecting method for use specifically comprises the following steps that
Step1: utilize the mobile forms of a length of W to choose L hydrocracking unit running data the most successively
Sample set, obtains X the most respectively(1,W),X(2,W+1),…,X(L,W+L-1)L group sample altogether, and each sample comprises the sampling inspection of W group
Survey data；
Step2: be utilized respectively PCA technology to above-mentioned X(1,W),X(2,W+1),…,X(L,W+L-1)The data matrix of L group sample altogether
Carry out principal component decomposition, extract load matrix P of each group of data sample(1,W),P(2,W+1),…,P(L,W+L-1)；
Step3: calculate the similarity of adjacent two sequential load matrix respectively, if the similarity of L group sampled data is sufficiently large,
Then think that selected data is steady state data, otherwise for unstable state data.
Specifically, calculate the similarity of adjacent sequential load matrix respectively according to equation below, if the phase of L group sample data
It is sufficiently close to 1 (similarity of L sample is both greater than a certain threshold alpha set in advance i.e. continuously), then it is assumed that selected L like degree
Individual history run process data X(1,W),X(2,W+1),…,X(L,W+L-1)For steady state data, it is otherwise unstable state data.
Wherein, γ (Pi,W+i-1,Pi+1,W+i) represent two data samples similarity, A represents the dimension of load matrix, i.e.
The pivot number of PCA.
Further, described according to sideline product flow each in fractionating system, tail oil flow to hydrocracking unit spy
Determine corresponding under the conditions of operation phase, specific feed oil attribute, specific operation " operation phase-charging attribute-operating condition-
Outlet yield " data block carries out the calculated off line of cracking zone reaction depth and stores, including:
Hydrocracking unit is calculated specific run stage, specific feed oil attribute, spy according to formula (1) and formula (2)
Determine the reaction depth of cracking zone under operating condition:
ηk,i,j=w1x1+w2x2+w3x3+w4x4+w5x5+w6x6Formula (1)
Wherein, ηk,i,jRepresenting that hydrocracking unit is i at kth operation phase charging attribute, operating condition is the situation of j
Under the reaction depth of whole response system cracking zone；w1, w2, w3, w4, w5, w6Represent liquefied gas, light naphthar, scheelite brain respectively
Oil, aerial kerosene, diesel oil, the discrete lumped component of tail oil 6 class are for the weight coefficient of reaction depth；x1, x2, x3, x4, x5, x6Point
Do not represent liquefied gas, light naphthar, heavy naphtha, aerial kerosene, diesel oil, the mass percent of the discrete lumped component of tail oil 6 class；Respectively represent be hydrocracked mixed feeding oil in flow process, liquefied gas, light naphthar,
Heavy naphtha, aerial kerosene, diesel oil, the average molecular mass of the discrete lumped component of tail oil six class.
Specifically, hydrocracking unit can be calculated respectively each under i-th kind of mixed feeding oil attribute of kth operation phase
Plant reaction depth value η corresponding to operating conditionk,i, form j separate " operation phase k-charging attribute i-operation bar
Part-reaction depth " data block, its form is as follows:
It is hydrocracked stream from N number of separate " operation phase-charging attribute-operating condition-outlet yield " obtained
Journey data block is chosen identical operation phase difference mixing steady state data corresponding to feed oil attribute, it is thus achieved that kth runs rank
Reaction depth value η corresponding to various operating conditions under section the i-th i kind mixed feeding oil attributek,ii, form jj individual separate
" operation phase k-feeds attribute ii-operating condition-reaction depth " data block, its form is as follows:
Repeat the above steps, until kth operation phase all mixed feedings oil attribute in history run process database
Till reaction depth corresponding to lower all operations condition all calculates, form hydrocracking unit kth operation phase, no
With the reaction depth value of calculation list under the conditions of mixed feeding oil attribute, different operating, its form is as follows:
Choose hydrocracking unit another one operation phase k+1, repeat the above steps, ask for the mixing of this stage difference former
Material oil reaction depth value of calculation corresponding to attribute, different operating condition, constitutes the calculations list of this operation phase reaction depth.
Repeat the above steps, until whole operation phase corresponding in this device history run process database traveled through
Till, finally give N number of separate " operation phase-charging attribute-operating condition-reaction depth " data block.
Further, the described operation phase being presently according to hydrocracking unit and current feed oil attribute, choose
With current feed oil attributes similarity more than multiple " charging attribute-operating condition-reaction depth " data block of predetermined threshold value, wrap
Include:
The operation phase being presently according to hydrocracking unit, from N number of " operation phase-charging attribute-operating condition-
Reaction depth " first data block is screened by " operation phase ", constitute M separate " entering under the current operation phase
Material attribute-operating condition-reaction depth " data block；Wherein, N and M is respectively the integer more than 0, and N > M；
According to current feed oil attribute, respectively in " charging attribute-operating condition-reaction depth " data block individual with above-mentioned M
Charging attribute carry out similarity measurement, choose with current feed oil attributes similarity more than predetermined threshold value multiple " charging belong to
Property-operating condition-reaction depth " and data block, Fast Training neutral net.
Further, described basis current feed oil attribute, respectively with above-mentioned M " charging attribute-operating condition-reaction
The degree of depth " charging attribute in data block carries out similarity measurement, including:
According to current feed oil attribute, respectively in " charging attribute-operating condition-reaction depth " data block individual with above-mentioned M
Charging attribute data carry out similarity measurement according to being weighted, by Euclidean distance and co sinus vector included angle value, the mixing index constituted.
Specifically, the training to BP neural network model includes:
(1) select from N number of " operation phase-charging attribute-operating condition-reaction depth " data block of hydrocracking unit
Select the M corresponding to the current operation phase separate " charging attribute-operating condition-reaction depth " data block；
(2) with the similarity of mixed feeding oil attribute for screening sample foundation, current feed oil attribute is calculated respectively with upper
State corresponding feed oil in M separate " charging attribute-operating condition-reaction depth " data block obtained after primary election
The Euclidean distance of attribute and angle cosine value；
Euclidean distance:
Angle cosine value:
Wherein, c0,jFor the attribute conditions of current reactor inlet mixing raw oil, ci,jFor in historical data sample storehouse not
With the attribute conditions of reactor inlet miscella, i.e. ci,jRepresent that in i-th historical data, jth kind feed component (includes 1# and 2#
Straight run VGO, tank field wax oil, circulation tail oil) flow.
(3) with more than the Euclidean distance of required current feed oil attribute and M feed oil attribute of specific run stage and angle
Based on string value, select certain weight coefficient that the two is weighted summation, constitute novel mixing index Hybrid_index
(i)；
Mixing index: Hybrid_index (i)=λ * cos (θi)+(1-λ)*e-Ed(i)
Wherein, λ is Euclidean distance and the weight of angle cosine value.
(4) it is ranked up according to the size of the hydrocracking unit feed oil conditional attribute mixing index constituted in (3), from
M separate " charging attribute-operating condition-reaction depth " data block filter out closest with current feed oil attribute
Some data samples as the training dataset of BP neutral net.
Further, utilize described BP neural network model, the real-time estimate hydrocracking unit current operation phase, current
Charging attribute, the lower reaction depth being hydrocracked flow process response system cracking zone that can reach of current operational conditions, including:
According to being hydrocracked the on-line checking value of flow process response system cracking zone performance variable, by described BP neutral net
Model, the real-time estimate hydrocracking unit current operation phase, currently feed attribute, current operational conditions lower can reach add
The reaction depth of hydrogen cracking flow process response system cracking zone；
Wherein, the on-line checking value of described performance variable includes: cracker each reaction bed Temperature Distribution current
Value and the currency of all kinds of catalyst space velocities.
In actual applications, substantially operation can be divided into according to the activity of catalyst due to the operation phase of hydrocracking unit
Initial stage, operation mid-term and 3 Main Stage of end-of-run, its initial operating stage and end-of-run rapid catalyst deactivation, time phase
To of short duration, therefore the hydrocracking unit most of the time is in and runs mid-term, and the feature of this operation phase is that catalyst slowly loses
Living, running is relatively steady, therefore the operation interim data of embodiments of the invention selective hydrocracking device is tested.
In order to cause because there is time lag between solution response system and fractionating system, reaction depth result of calculation is inaccurate asks
Topic, utilizes the process operation data corresponding to hydrocracking unit stable state according to formula ηi,j=w1x1+w2x2+w3x3+w4x4+w5x5
+w6x6It is calculated.Wherein w1, w2, w3, w4, w5, w6Be respectively liquefied gas, light naphthar, heavy naphtha, aerial kerosene,
Diesel oil, the discrete lumped component of tail oil 6 class are for the weight coefficient of reaction depth, due to reaction depth and raw oil, liquefied gas, light
Petroleum, heavy naphtha, aerial kerosene, diesel oil, the average molecular mass of tail oil are closely related, and each component is to reaction depth
Contribution its weight coefficient the biggest also should be the biggest, therefore w1, w2, w3, w4, w5, w6Value can according to liquefied gas, light naphthar,
Heavy naphtha, aerial kerosene, diesel oil, tail oil 6 class lumped component are carried out relative to the average molecular mass of mixed feeding oil
Set and (take mixed feeding oil, liquefied gas, light naphthar, heavy naphtha, aerial kerosene, diesel oil, the discrete lumped component of tail oil seven class
Average molecular mass be respectively 350,51,86,128,184,226,300, then w1, w2, w3, w4, w5, w6Corresponding value
Being respectively 6.86,4.07,2.73,1.90,1.55,1.17, after normalization, value is 0.38 respectively, 0.22,0.15,0.10,
0.09,0.06)；x1, x2, x3, x4, x5, x6Represent liquefied gas, light naphthar, heavy naphtha, aerial kerosene, diesel oil, tail oil respectively
The mass percent of the 6 discrete lumped components of class, by the real-time traffic in hydrocracking unit fractionating system each sideline product exit
Data calculate and obtain.Different operating condition institute under the phase specific blend feed oil attribute that is in operation is calculated respectively according to formula (1)
Corresponding response system cracking zone reaction depth, its value of calculation result after normalized is as shown in table 1.
Table 1 is hydrocracked " operating condition-reaction depth " under process flow operation specific blend in mid-term feed oil attribute corresponding pass
System
Utilize and above-mentioned be hydrocracked flow process stable state history data set, calculate the phase different blended that is in operation by formula (1) respectively
The reaction depth being hydrocracked flow process response system cracking zone corresponding under the conditions of closing feed oil attribute, different operating, 5 kinds of allusion quotations
Type mixed feeding oil proportioning situation is as shown in table 2, the reaction that every kind of typical hybrid feed oil is corresponding under the conditions of different operating
Depth calculation value result after normalized is as shown in table 3.
The various feed oil flows that table 2 some typical hybrid feed oil attribute is corresponding
It is right that table 3 is hydrocracked " operating condition-reaction depth " under process flow operation some typical hybrid in mid-term feed oil attribute
Should be related to
In actual applications, preferred and BP neural network model the fast-training procedures of optimum training sample set includes:
Step1: the hydrocracking unit current operation phase built with off-line, " charging attribute-operating condition-reaction was deep
Degree " based on charging attribute in data block, calculate current feed oil attribute respectively and enter with each history in above-mentioned data block
The Euclidean distance of material attribute and angle cosine value；
Based on Step2: the raw oil attribute Euclidean distance calculated in Step1 and angle cosine value, choose conjunction
The two is sued for peace by suitable weight coefficient λ, and (embodiment of the present invention is selected to constitute new mixed type index Hybrid_index (i)
Take λ=0.5)；
Step3: mixing index Hybrid_index (i) constituted in Step2 is ranked up according to its numerical values recited, from
Above-mentioned " charging attribute-operating condition-reaction depth " data block filters out S the sample the most close with current feed oil attribute
Data constitute the optimum training sample set (embodiment of the present invention chooses S=35) of BP neural network model；
Step4: carry out BP god based on conditional attribute mixing index similarity preferred training data sample set with in Step3
Through the Fast Training (as shown in Figure 5) of network, determine input layer, hidden layer, the neuron node number of output layer and currently
Interconnective weight between each layer neuron under feed oil attribute conditions.The number of hidden layer neuron is public according to following experience
Formula is chosen.
Wherein, I, H, O represent the input layer of three layers of BP neutral net, hidden layer, output layer nodes respectively,RepresentValue of calculation round up, the value of the embodiment of the present invention is respectively 11,5,1.
For inspection according to above-mentioned training sample method for optimizing, the precision of prediction of Fast Training BP neural network model, choose
Certain oil plant hydrocracking unit runs 12 groups of different operating conditions under specific feed oil attribute in mid-term (the most currently feeding attribute)
Corresponding history data, as test sample, uses the mean square of absolute error, relative error and 12 test samples
The precision of prediction of above-mentioned model is evaluated by 3 performance indications of root error (RMSE), the specific formula for calculation of each performance indications
As follows, concrete test result is as shown in table 4.
Absolute error:
Relative error:
Root-mean-square error (RMSE):
Table 4 hydrocracking unit runs and currently feeds different operating conditioned response degree of depth real-time estimate result under attribute mid-term
Fig. 6 is the real-time estimate result figure of the hydrocracking unit reaction depth of one embodiment of the invention.Such as Fig. 6 institute
Showing, curve ' True ' represents real reaction depth, and ' common BP ' represents prediction when not being optimized training sample to curve
Reaction depth, ' the preferred BP ' of sample represents the reaction depth of prediction when carrying out preferred to training sample to curve.Can from Fig. 6
Go out, utilize the embodiment of the present invention can reaction depth relatively accurately be predicted, can be this process quenching hydrogen injection rate
Optimal control provides real time status information, is to ensure that catalyst activity, extension fixture on-stream time, realizes target product processing capacity
The basis being adjusted flexibly.
Fig. 7 is the structural representation of the real-time estimate device of one embodiment of the invention hydrocracking unit reaction depth.
As it is shown in fig. 7, the real-time estimate device of the hydrocracking unit reaction depth of the present embodiment includes:
Data pre-processing unit 71, for carrying out pretreatment to the history data of hydrocracking unit；
First data block forms unit 72, for the history data through pretreatment uses the side of similarity measurement
Method carries out stable state differentiation, filters out steady-state process real time data collection, by described steady-state process real time data collection according to being hydrocracked
The operation phase of flow process, the difference of mixed feeding oil attribute are classified, and form N number of separate " operation phase-charging genus
Property-operating condition-outlet yield " data block；Wherein, N is the integer more than 0；
Reaction depth calculated off line unit 73, is used for according to sideline product flow each in fractionating system, tail oil flow adding
" operation phase-charging that hydrogen cracking unit is corresponding under the conditions of specific run stage, specific feed oil attribute, specific operation
Attribute-operating condition-outlet yield " data block carries out the calculated off line of cracking zone reaction depth and stores；
Second data block forms unit 74, is used for calculating each operation phase of hydrocracking unit, various charging attribute, each
Reaction depth corresponding to generic operation condition, obtains N number of separate " operation phase-charging attribute-operating condition-reaction
The degree of depth " data block；Wherein, N is the integer more than 0；
Neural network model training unit 75, for the operation phase being presently according to hydrocracking unit with when advancing
Material attribute, chooses and currently feeds the attributes similarity multiple " charging attribute-operating condition-reaction depths " more than predetermined threshold value
Data block, with the operating condition of data block chosen for input, reaction depth is output, Fast Training BP neural network model；
Reaction depth predicting unit 76, is used for utilizing described BP neural network model, and real-time estimate hydrocracking unit is worked as
The front operation phase, currently feed attribute, current operational conditions lower can reach be hydrocracked flow process response system cracking zone
Reaction depth.
The real-time estimate device of the hydrocracking unit reaction depth of the embodiment of the present invention may be used for performing said method
Embodiment, its principle is similar with technique effect, and here is omitted.
The real-time predicting method of the hydrocracking unit reaction depth that the embodiment of the present invention provides and device, by optimal instruction
Practice sample set optimal way and filter out (i.e. current operation phase) and current feed oil attribute under the conditions of present catalyst activity
Immediate some historical data sample collection carry out the Fast Training of BP neural network model, wherein choose that to be hydrocracked flow process anti-
Answer the Temperature Distribution of each reaction bed of system cracking zone, the air speed of all kinds of catalyst is input, and reaction depth is output.This prediction
Model is first according to operation phase of device and eliminates the interference of a large amount of extraneous data, then according to the mixing index of conditional attribute
Preferably some historical data sample the most close with current feed oil attribute carry out model construction, and this not only greatly reduces god
Training burden through network, it is achieved that Fast Training, and avoid neutral net in the training process because of a large amount of unrelated samples
Introducing and the not convergence problem that causes.The real-time estimate current operation phase, currently feed under attribute, current operational conditions be hydrogenated with
The reaction depth of cracking unit, can be that the optimal control of this process quenching hydrogen injection rate provides real time status information, be to ensure that and urge
Agent activity, extension fixture on-stream time, realize the basis that target product processing capacity is adjusted flexibly.
Those skilled in the art are it should be appreciated that embodiments of the invention can be provided as method, system or computer program
Product.Therefore, the reality in terms of the present invention can use complete hardware embodiment, complete software implementation or combine software and hardware
Execute the form of example.And, the present invention can use at one or more computers wherein including computer usable program code
The upper computer program product implemented of usable storage medium (including but not limited to disk memory, CD-ROM, optical memory etc.)
The form of product.
The present invention is with reference to method, equipment (system) and the flow process of computer program according to embodiments of the present invention
Figure and/or block diagram describe.It should be understood that can the most first-class by computer program instructions flowchart and/or block diagram
Flow process in journey and/or square frame and flow chart and/or block diagram and/or the combination of square frame.These computer programs can be provided
Instruction arrives the processor of general purpose computer, special-purpose computer, Embedded Processor or other programmable data processing device to produce
A raw machine so that the instruction performed by the processor of computer or other programmable data processing device is produced for real
The device of the function specified in one flow process of flow chart or multiple flow process and/or one square frame of block diagram or multiple square frame now.
It should be noted that term " includes ", " comprising " or its any other variant are intended to the bag of nonexcludability
Contain, so that include that the process of a series of key element, method, article or equipment not only include those key elements, but also include
Other key elements being not expressly set out, or also include the key element intrinsic for this process, method, article or equipment.
In the case of there is no more restriction, statement " including ... " key element limited, it is not excluded that including described key element
Process, method, article or equipment in there is also other identical element.
In the description of the present invention, illustrate a large amount of detail.Although it is understood that, embodiments of the invention can
To put into practice in the case of there is no these details.In some instances, it is not shown specifically known method, structure and skill
Art, in order to do not obscure the understanding of this description.Similarly, it will be appreciated that disclose to simplify the present invention and help to understand respectively
One or more in individual inventive aspect, above in the description of the exemplary embodiment of the present invention, each of the present invention is special
Levy and be sometimes grouped together in single embodiment, figure or descriptions thereof.But, should be by the method solution of the disclosure
Release in reflecting an intention that i.e. the present invention for required protection requires than the feature being expressly recited in each claim more
Many features.More precisely, as the following claims reflect, inventive aspect is less than single reality disclosed above
Execute all features of example.Therefore, it then follows claims of detailed description of the invention are thus expressly incorporated in this detailed description of the invention,
The most each claim itself is as the independent embodiment of the present invention.
Above example is merely to illustrate technical scheme, is not intended to limit；Although with reference to previous embodiment
The present invention is described in detail, it will be understood by those within the art that: it still can be to aforementioned each enforcement
Technical scheme described in example is modified, or wherein portion of techniques feature is carried out equivalent；And these are revised or replace
Change, do not make the essence of appropriate technical solution depart from the spirit and scope of various embodiments of the present invention technical scheme.
Claims (9)
1. the real-time predicting method of a hydrocracking unit reaction depth, it is characterised in that including:
The history data of hydrocracking unit is carried out pretreatment；
Use the method for similarity measurement to carry out stable state differentiation the history data through pretreatment, filter out steady-state process
Real time data collection, by described steady-state process real time data collection according to being hydrocracked the operation phase of flow process, mixed feeding oil attribute
Difference classify, form N number of separate " operation phase-charging attribute-operating condition-outlet yield " data block；
Wherein, N is the integer more than 0；
According to sideline product flow each in fractionating system, tail oil flow to hydrocracking unit the specific run stage, specific enter
Under the conditions of material oil attribute, specific operation, corresponding " operation phase-charging attribute-operating condition-outlet yield " data block is entered
The calculated off line of row cracking zone reaction depth also stores；
Calculate the reaction depth corresponding to each operation phase of hydrocracking unit, various feed oil attribute, all kinds of operating condition,
Obtain N number of separate " operation phase-charging attribute-operating condition-reaction depth " data block；
Operation phase of being presently according to hydrocracking unit and currently feed attribute, choose and currently feed attributes similarity
More than multiple " charging attribute-operating condition-reaction depth " data block of predetermined threshold value, the operating condition of the data block to choose
For input, reaction depth is output, Fast Training BP neural network model；
Utilize the BP neural network model set up, the real-time estimate hydrocracking unit current operation phase, currently feed attribute,
The lower reaction depth being hydrocracked flow process response system cracking zone that can reach of current operational conditions.
The real-time predicting method of hydrocracking unit reaction depth the most according to claim 1, it is characterised in that described in go through
History service data includes:
It is hydrocracked in the Temperature Distribution of the flow process each bed of response system cracking zone, the air speed of all kinds of catalyst, fractionating system each
Sideline product flow, tail oil flow and the various feed oil flows of decision mixed feeding oil attribute.
The real-time predicting method of hydrocracking unit reaction depth the most according to claim 1, it is characterised in that described right
The history data of hydrocracking unit carries out pretreatment, including:
The history data of hydrocracking unit is carried out outlier rejecting and Wavelet Denoising Method.
The real-time predicting method of hydrocracking unit reaction depth the most according to claim 1, it is characterised in that described surely
State discriminating data includes with extracting method:
Process operation data are after PCA extracts key message, by contrasting the similarity of adjacent sequential load matrix
Select the process data corresponding to steady state condition.
The real-time predicting method of hydrocracking unit reaction depth the most according to claim 1, it is characterised in that described
According to sideline product flow each in fractionating system, tail oil flow, hydrocracking unit is belonged in specific run stage, specific feed oil
Property, " operation phase-charging attribute-operating condition-outlet yield " data block corresponding under the conditions of specific operation carry out cracking
The section calculated off line of reaction depth also stores, including:
Hydrocracking unit is calculated specific run stage, specific feed oil attribute, specific behaviour according to formula (1) and formula (2)
The reaction depth of the cracking zone under the conditions of work:
ηk,i,j=w1x1+w2x2+w3x3+w4x4+w5x5+w6x6Formula (1)
Wherein, ηk,i,jRepresenting that hydrocracking unit is i at kth operation phase charging attribute, operating condition is whole in the case of j
The reaction depth of individual response system cracking zone；w1, w2, w3, w4, w5, w6Represent liquefied gas, light naphthar, heavy naphtha, boat respectively
Empty kerosene, diesel oil, the discrete lumped component of tail oil 6 class are for the weight coefficient of reaction depth；x1, x2, x3, x4, x5, x6Represent respectively
Liquefied gas, light naphthar, heavy naphtha, aerial kerosene, diesel oil, the mass percent of the discrete lumped component of tail oil 6 class；Respectively represent be hydrocracked mixed feeding oil in flow process, liquefied gas, light naphthar,
Heavy naphtha, aerial kerosene, diesel oil, the average molecular mass of the discrete lumped component of tail oil six class.
The real-time predicting method of hydrocracking unit reaction depth the most according to claim 1, it is characterised in that described
The operation phase being presently according to hydrocracking unit and current feed oil attribute, choose big with current feed oil attributes similarity
In multiple " charging attribute-operating condition-reaction depth " data block of predetermined threshold value, including:
The operation phase being presently according to hydrocracking unit, from N number of " operation phase-charging attribute-operating condition-reaction
The degree of depth " first data block is screened by " operation phase ", constituting under the current operation phase M, separate " charging belongs to
Property-operating condition-reaction depth " data block；Wherein, M is the integer more than 0, and N > M；
According to current feed oil attribute, respectively with entering in above-mentioned M " charging attribute-operating condition-reaction depth " data block
Material attribute carries out similarity measurement, chooses and " feeds attribute-behaviour with current feed oil attributes similarity more than the multiple of predetermined threshold value
Make condition-reaction depth " data block, Fast Training neutral net.
The real-time predicting method of hydrocracking unit reaction depth the most according to claim 6, it is characterised in that described
According to current feed oil attribute, respectively with the charging attribute in above-mentioned M " charging attribute-operating condition-reaction depth " data block
Carry out similarity measurement, including:
According to current feed oil attribute, respectively with entering in above-mentioned M " charging attribute-operating condition-reaction depth " data block
Material attribute data carries out similarity measurement according to the mixing index being made up of Euclidean distance and the weighting of co sinus vector included angle value.
The real-time predicting method of hydrocracking unit reaction depth the most according to claim 1, it is characterised in that utilize institute
State BP neural network model, the real-time estimate hydrocracking unit current operation phase, currently feed under attribute, current operational conditions
The reaction depth being hydrocracked flow process response system cracking zone that can reach, including:
According to being hydrocracked the on-line checking value of flow process response system cracking zone performance variable, pass through set up BP neutral net
Model, the real-time estimate hydrocracking unit current operation phase, currently feed attribute, current operational conditions lower can reach add
The reaction depth of hydrogen cracking flow process response system cracking zone；
Wherein, the on-line checking value of described performance variable includes: the currency of cracker each reaction bed Temperature Distribution with
And the currency of all kinds of catalyst space velocities.
9. the real-time estimate device of a hydrocracking unit reaction depth, it is characterised in that including:
Data pre-processing unit, for carrying out pretreatment to the history data of hydrocracking unit；
First data block forms unit, for using the method for similarity measurement to carry out the history data through pretreatment
Stable state differentiates, filters out steady-state process real time data collection, by described steady-state process real time data collection according to being hydrocracked flow process
Operation phase, the difference of mixed feeding oil attribute are classified, and form N number of separate " operation phase-charging attribute-behaviour
Make condition-outlet yield " data block；
Reaction depth calculated off line unit, is used for according to sideline product flow each in fractionating system, tail oil flow being hydrocracked
" operation phase-charging attribute-behaviour that device is corresponding under the conditions of specific run stage, specific feed oil attribute, specific operation
Make condition-outlet yield " data block carries out the calculated off line of cracking zone reaction depth and stores；
Second data block forms unit, is used for calculating each operation phase of hydrocracking unit, various charging attribute, each generic operation
Reaction depth corresponding to condition, obtains N number of separate " operation phase-charging attribute-operating condition-reaction depth " number
According to block；Wherein, N is the integer more than 0；
Neural network model training unit, for operation phase of being presently according to hydrocracking unit with currently feed genus
Property, choose and currently feed attributes similarity multiple " charging attribute-operating condition-reaction depth " data more than predetermined threshold value
Block, with the operating condition of data block chosen for input, reaction depth is output, Fast Training BP neural network model；
Reaction depth predicting unit, is used for utilizing described BP neural network model, and real-time estimate hydrocracking unit currently runs
Stage, currently to feed attribute, the lower reaction being hydrocracked flow process response system cracking zone that can reach of current operational conditions deep
Degree.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610738881.6A CN106281431B (en) | 2016-08-26 | 2016-08-26 | A kind of real-time predicting method and device of hydrocracking unit reaction depth |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610738881.6A CN106281431B (en) | 2016-08-26 | 2016-08-26 | A kind of real-time predicting method and device of hydrocracking unit reaction depth |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106281431A true CN106281431A (en) | 2017-01-04 |
| CN106281431B CN106281431B (en) | 2018-02-16 |
Family
ID=57677829
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610738881.6A Active CN106281431B (en) | 2016-08-26 | 2016-08-26 | A kind of real-time predicting method and device of hydrocracking unit reaction depth |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106281431B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108229721A (en) * | 2017-12-04 | 2018-06-29 | 西北大学 | The Forecasting Methodology of pyrolysis of coal product based on Speed Controlling Based on Improving BP Neural Network |
| CN110751986A (en) * | 2019-10-15 | 2020-02-04 | 山东省科学院能源研究所 | Calculation model and detection method of polyolefin grafting rate based on artificial neural network |
| CN111849545A (en) * | 2019-04-28 | 2020-10-30 | 中国石油化工股份有限公司 | Hydrocracking product quality prediction method, device and memory |
| CN111849544A (en) * | 2019-04-28 | 2020-10-30 | 中国石油化工股份有限公司 | Hydrocracking product quality automatic control method, device and storage |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW201322149A (en) * | 2011-11-22 | 2013-06-01 | Univ Shu Te | Method of establishing system equivalent model combined with Volterra system and its computer program product |
| CN103605325A (en) * | 2013-09-16 | 2014-02-26 | 华东理工大学 | Complete period dynamic optimization method for industrial ethylene cracking furnace and based on surrogate model |
| CN104804761A (en) * | 2015-03-26 | 2015-07-29 | 华东理工大学 | Real-time yield prediction method for hydrocracking device |
| CN105740960A (en) * | 2014-12-06 | 2016-07-06 | 中国石油化工股份有限公司 | Optimization method of industrial hydrocracking reaction condition |
-
2016
- 2016-08-26 CN CN201610738881.6A patent/CN106281431B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW201322149A (en) * | 2011-11-22 | 2013-06-01 | Univ Shu Te | Method of establishing system equivalent model combined with Volterra system and its computer program product |
| CN103605325A (en) * | 2013-09-16 | 2014-02-26 | 华东理工大学 | Complete period dynamic optimization method for industrial ethylene cracking furnace and based on surrogate model |
| CN105740960A (en) * | 2014-12-06 | 2016-07-06 | 中国石油化工股份有限公司 | Optimization method of industrial hydrocracking reaction condition |
| CN104804761A (en) * | 2015-03-26 | 2015-07-29 | 华东理工大学 | Real-time yield prediction method for hydrocracking device |
Non-Patent Citations (1)
| Title |
|---|
| 王国荣: "神经网络技术在加氢裂化产品收率预测模型中的应用", 《石化技术与应用》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108229721A (en) * | 2017-12-04 | 2018-06-29 | 西北大学 | The Forecasting Methodology of pyrolysis of coal product based on Speed Controlling Based on Improving BP Neural Network |
| CN111849545A (en) * | 2019-04-28 | 2020-10-30 | 中国石油化工股份有限公司 | Hydrocracking product quality prediction method, device and memory |
| CN111849544A (en) * | 2019-04-28 | 2020-10-30 | 中国石油化工股份有限公司 | Hydrocracking product quality automatic control method, device and storage |
| CN111849545B (en) * | 2019-04-28 | 2021-08-06 | 中国石油化工股份有限公司 | Hydrocracking product quality prediction method, device and memory |
| CN111849544B (en) * | 2019-04-28 | 2021-08-06 | 中国石油化工股份有限公司 | Hydrocracking product quality automatic control method, device and storage |
| CN110751986A (en) * | 2019-10-15 | 2020-02-04 | 山东省科学院能源研究所 | Calculation model and detection method of polyolefin grafting rate based on artificial neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106281431B (en) | 2018-02-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101634605B (en) | Intelligent gearbox fault diagnosis method based on mixed inference and neural network | |
| CN106281431B (en) | A kind of real-time predicting method and device of hydrocracking unit reaction depth | |
| CN104789256B (en) | A kind of yield real-time predicting method of catalytic cracking unit | |
| CN106845796B (en) | One kind is hydrocracked flow product quality on-line prediction method | |
| Jiang et al. | Prediction of ecological pressure on resource-based cities based on an RBF neural network optimized by an improved ABC algorithm | |
| Winkowski | Classification of forecasting methods in production engineering | |
| CN110456756A (en) | A method of suitable for continuous production process overall situation operation conditions online evaluation | |
| CN109815616A (en) | A kind of simulation of molecular level hydrogenation plant and optimization method | |
| CN107797931A (en) | A kind of method for evaluating software quality and system based on second evaluation | |
| CN104484714B (en) | A kind of real-time predicting method of catalytic reforming unit yield | |
| CN104392098A (en) | Method for predicting yield of catalytically cracked gasoline | |
| CN104463343A (en) | Method for predicting catalytic cracking light oil yield | |
| CN104463327A (en) | Method for predicting catalytic cracking coke yield | |
| CN106709654A (en) | Global operating condition evaluating and quality tracing method for hydrocracking process | |
| CN102213949A (en) | Method for optimizing value of ethylene device | |
| CN108416439A (en) | Oil refining process Method For Predicting The Products and system based on variable weighting deep learning | |
| Zhang et al. | Research on ethanol coupling to prepare C4 olefins based on BP neural network and cluster analysis | |
| Woo et al. | Understanding the long-term emergence of autonomous vehicles technologies | |
| CN106874644A (en) | The real-time predicting method and its system of hydrogenolysis degree in a kind of hydrofinishing | |
| CN105740960A (en) | Optimization method of industrial hydrocracking reaction condition | |
| Qing | Global practice of AI and big data in oil and gas industry | |
| CN111581882B (en) | Rescheduling trigger judging method under uncertainty condition of production unit | |
| CN105976127A (en) | Risk evaluation method and apparatus for integrated electric propulsion system | |
| Zhu et al. | Multiple input self-organizing-map ResNet model for optimization of petroleum refinery conversion units | |
| Nekrasov et al. | Applying Predictive Machine Learning Algorithms to Petroleum Refining Processes as Part of Intelligent Automation |
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 |