CN110987862A - Diesel oil on-line blending method - Google Patents
Diesel oil on-line blending method Download PDFInfo
- Publication number
- CN110987862A CN110987862A CN201911075166.9A CN201911075166A CN110987862A CN 110987862 A CN110987862 A CN 110987862A CN 201911075166 A CN201911075166 A CN 201911075166A CN 110987862 A CN110987862 A CN 110987862A
- Authority
- CN
- China
- Prior art keywords
- property
- value
- blending
- component oil
- model
- 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.)
- Pending
Links
- 238000002156 mixing Methods 0.000 title claims abstract description 120
- 238000000034 method Methods 0.000 title claims abstract description 47
- 239000002283 diesel fuel Substances 0.000 title claims abstract description 22
- 238000012937 correction Methods 0.000 claims abstract description 95
- 238000005259 measurement Methods 0.000 claims abstract description 66
- 238000013528 artificial neural network Methods 0.000 claims abstract description 20
- 230000008569 process Effects 0.000 claims abstract description 18
- 238000010183 spectrum analysis Methods 0.000 claims abstract description 12
- 238000012216 screening Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims abstract description 8
- 230000002159 abnormal effect Effects 0.000 claims description 20
- 210000002569 neuron Anatomy 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 11
- 238000001514 detection method Methods 0.000 claims description 6
- 238000009499 grossing Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000010238 partial least squares regression Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000003595 spectral effect Effects 0.000 claims description 6
- 229910052717 sulfur Inorganic materials 0.000 claims description 6
- 238000001228 spectrum Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 230000004069 differentiation Effects 0.000 claims description 4
- 239000000126 substance Substances 0.000 claims description 3
- 238000002329 infrared spectrum Methods 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims description 2
- 239000003921 oil Substances 0.000 description 95
- 230000006870 function Effects 0.000 description 11
- 238000005457 optimization Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 3
- 238000005265 energy consumption Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000012797 qualification Methods 0.000 description 3
- 238000007430 reference method Methods 0.000 description 3
- CIWBSHSKHKDKBQ-JLAZNSOCSA-N Ascorbic acid Chemical compound OC[C@H](O)[C@H]1OC(=O)C(O)=C1O CIWBSHSKHKDKBQ-JLAZNSOCSA-N 0.000 description 2
- 238000000692 Student's t-test Methods 0.000 description 2
- NINIDFKCEFEMDL-UHFFFAOYSA-N Sulfur Chemical compound [S] NINIDFKCEFEMDL-UHFFFAOYSA-N 0.000 description 2
- 238000009833 condensation Methods 0.000 description 2
- 230000005494 condensation Effects 0.000 description 2
- 238000004821 distillation Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000011593 sulfur Substances 0.000 description 2
- 238000012353 t test Methods 0.000 description 2
- 238000004497 NIR spectroscopy Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
- G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
- G01N21/359—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using near infrared light
Abstract
The invention relates to the technical field of finished oil property prediction and regulation, and discloses an online diesel blending method, which comprises the following steps: performing spectral analysis on component oil samples of different types and different concentrations to obtain a property measurement value of each component oil sample; screening each component oil sample according to the property measurement value to establish a correction set, and fitting the correction set to obtain a property correction model between the concentration and the property measurement value; collecting a historical blending formula, taking the property value of the component oil in the historical blending formula as input, taking the property value of the finished oil obtained according to the historical blending formula as output, and training a neural network to obtain a blending model; and carrying out on-line control on the diesel oil blending process according to the property correction model and the blending model. The invention can carry out on-line control on the diesel oil blending process and has high blending accuracy.
Description
Technical Field
The invention relates to the technical field of prediction and regulation of the oil quality of finished products, in particular to an on-line diesel oil blending method.
Background
The traditional diesel oil property measuring method is long in time consumption and is not suitable for the application of diesel oil on-line blending. In addition, in the process of blending oil products, because the quantity of the treated component oil is extremely large and various influencing factors are mutually related, the mechanism of the processes is not studied at present, but an effective solution is not available in the face of numerous problems in actual production, so that the primary blending qualification rate is low, secondary blending is required, and the labor and energy consumption is caused.
Disclosure of Invention
The invention aims to overcome the technical defects and provide an on-line diesel blending method, which solves the technical problems of long blending time consumption and low blending accuracy in the prior art.
In order to achieve the technical purpose, the technical scheme of the invention provides an on-line diesel blending method, which comprises the following steps:
performing spectral analysis on component oil samples of different types and different concentrations to obtain a property measurement value of each component oil sample;
screening each component oil sample according to the property measurement value to establish a correction set, and fitting the correction set to obtain a property correction model between the concentration and the property measurement value;
collecting a historical blending formula, taking the property value of the component oil in the historical blending formula as input, taking the property value of the finished oil obtained according to the historical blending formula as output, and training a neural network to obtain a blending model;
and carrying out on-line control on the diesel oil blending process according to the property correction model and the blending model.
Compared with the prior art, the invention has the beneficial effects that: the invention firstly utilizes the spectral analysis to obtain the property measured value of the component oil sample, the spectral analysis technology has high analysis speed, more output and no damage to the sample, and provides possibility for the on-line blending of the diesel oil. And then establishing a correction set, fitting according to the correction set to obtain a property correction model, wherein the property correction model can predict the properties of the component oil, so that a foundation is provided for diesel oil online blending. And finally, collecting a historical blending formula as sample data, training a neural network to obtain a blending model between the properties of the component oil and the properties of the finished oil, so that the property requirements of the component oil can be reversely deduced according to the property requirements of the finished oil, the concentration requirements of the component oil can be reversely deduced according to the property correction model, and the blending process can be adjusted on line according to the concentration requirements. The harmonic model obtained by the neural network method realizes the description of complex physical phenomena in the diesel oil harmonic process, does not need to limit the form and parameters of the harmonic model in advance, and has good application value for predicting the diesel oil harmonic property. The blending process of the diesel oil is adjusted on line in real time, so that the primary blending qualification rate can be improved, and the manpower and energy consumption caused by secondary blending can be reduced; the blending time is shortened, and the continuity and the stability of a diesel blending production line are ensured.
Drawings
FIG. 1 is a flow chart of an embodiment of the diesel oil on-line blending method provided by the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Example 1
As shown in fig. 1, embodiment 1 of the present invention provides an online blending method of diesel oil, comprising the steps of:
performing spectral analysis on component oil samples of different types and different concentrations to obtain a property measurement value of each component oil sample;
screening each component oil sample according to the property measurement value to establish a correction set, and fitting the correction set to obtain a property correction model between the concentration and the property measurement value;
collecting a historical blending formula, taking the property value of the component oil in the historical blending formula as input, taking the property value of the finished oil obtained according to the historical blending formula as output, and training a neural network to obtain a blending model;
and carrying out on-line control on the diesel oil blending process according to the property correction model and the blending model.
The embodiment firstly utilizes the spectral analysis to obtain the property measured value of the component oil sample, and the spectral analysis technology has the advantages of high analysis speed, high yield and no damage to the sample, thereby providing possibility for the on-line blending of the diesel oil. And then establishing a correction set, fitting according to the correction set to obtain a property correction model, wherein the property correction model can predict the properties of the component oil, so that a foundation is provided for diesel oil online blending. And finally, collecting a historical blending formula as sample data, training a neural network to obtain a blending model between the properties of the component oil and the properties of the finished oil, so that the property requirements of the component oil can be reversely deduced according to the property requirements of the finished oil, the concentration requirements of the component oil can be reversely deduced according to the property correction model, and the blending process can be adjusted on line according to the concentration requirements. The harmonic model obtained by the neural network method realizes the description of complex physical phenomena in the diesel oil harmonic process, does not need to limit the form and parameters of the harmonic model in advance, and has good application value for predicting the diesel oil harmonic property.
Specifically, the component oil property values include sulfur content, density, distillation range, condensation point, cold filter plugging point, cetane number and the like, and the finished oil property values include sulfur content, density, distillation range, condensation point, cold filter plugging point, cetane number and the like.
In particular, the property measurements of the calibration set of component oil samples need to contain all the properties that need to be monitored and predicted. The concentration of the concentrated corrected component oil sample should be greater than the range of concentration variation for the unknown sample analyzed using the harmonic model. The component concentrations of the same type of component oil sample of the calibration set are preferably evenly distributed throughout the range of variation. The calibration set should have a sufficient number of component oil samples to ensure that the mathematical relationship between the property measurements and concentrations can be analyzed; the property measurements of the calibration pooled component oil samples are preferably determined using current standards or conventional methods.
According to the invention, the diesel blending process is adjusted on line in real time according to the property correction model and the blending model, an optimized formula can be generated quickly, the quality of finished diesel is monitored in real time, the blending rule is corrected at any time, the quality of the finished diesel is ensured, the primary blending qualification rate is improved, and the manpower and energy consumption caused by secondary blending is reduced; the blending time is shortened, and the continuity and the stability of a diesel blending production line are ensured.
Preferably, the spectral analysis is performed on component oil samples of different types and different concentrations to obtain property measurement values of each component oil sample, specifically:
and performing spectral analysis on the component oil samples of different types and different concentrations by using a near-infrared spectrum analyzer to obtain the property measured value of each component oil sample.
Preferably, the screening of the property measurements of each of the component oil samples to establish a calibration set further comprises: pre-processing the data for the property measurements;
preprocessing the data of the property measurement value, specifically:
smoothing the property measurement value by adopting a window moving average algorithm:
wherein the content of the first and second substances,the property measurement value after the k wavelength point smoothing processing is shown, 2w +1 is the window width, xk+iProperty measurement values before smoothing processing are the k + i wavelength points;
derivative processing of the property measurements eliminates spectral baseline drift:
X'(i)=[x(i+g)-x(i)]/g
X”(i)=[x(i+g)-2x(i)+x(i-g)]/g2
wherein, X '(i) is the property measurement value after the first-order differentiation of the ith wavelength point, X' (i) is the property measurement value after the second-order differentiation of the ith wavelength point, X (i + g) is the property measurement value before the derivative processing of the ith wavelength point, and X (i) is the property measurement value before the derivative processing of the ith wavelength point;
data centralizing and normalizing the property prediction values:
wherein x' (i) is the property measurement value after the centering and the standardization of the ith wavelength point data, x (i) is the property measurement value before the centering and the standardization of the ith wavelength point data, u is the property measurement mean value of each component oil sample, and sigma is the standard deviation of the property measurement value of each component oil sample.
Specifically, in the present embodiment, the window moving average algorithm selects a smooth window with a width of 2w +1, and uses the average value of the property measurement values at all wavelength points in the window to replace the property measurement value at the center wavelength point. The derivative processing is a relatively ideal preprocessing method in the aspect of eliminating the spectral baseline drift, and is one of the most common methods in the preprocessing of the diffuse reflection spectrum. The first order differential can remove wavelength independent drift and the second order differential can remove wavelength linearly dependent drift. Data centralization and normalization cancel errors caused by different dimensions, self-variation or large numerical differences in regression analysis.
Preferably, the screening of each component oil sample according to the property measurement value establishes a calibration set, specifically:
selecting two component oil samples with the farthest Euclidean distance between the property measurement values and adding the two component oil samples into an initial calibration set;
respectively calculating the Euclidean distance between the property measured value of each component oil sample which is not added into the correction set and the property measured value of each component oil sample which is added into the correction set, and adding the component oil sample with the farthest Euclidean distance into the correction set;
and judging whether the number of the samples of the correction set is greater than a set threshold value, if so, outputting the correction set, and otherwise, turning to the previous step.
Specifically, the euclidean distance between the property measurements for the two component oil samples is calculated as:
wherein d isx(p, q) denotes a component oil sample xpProperty measurement of (2) with component oil sample xqN is the number of spectral wave points of the component oil sample, xp(j) As component oil sample xpProperty measurement at jth spectral wavelength point, xq(j) As component oil sample xqProperty measurements at the jth spectral wavelength point.
Preferably, the correction set is fitted to obtain a property correction model between the concentration and the property measurement value, specifically:
dividing a spectrum region into a plurality of equal-width wave bands, performing partial least squares regression modeling on the correction set on each wave band to obtain a local regression model of each wave band, calculating the RMSECV value of the local regression model of each wave band, taking the local regression model with the minimum RMSECV value as a first selected model, and taking the wave band corresponding to the local regression model with the minimum RMSECV value as a first selected wave band;
respectively combining other wave bands except the selected wave band with the previous selected wave band to obtain a plurality of combined wave bands, performing partial least squares regression modeling on the correction set on each combined wave band to obtain a local regression model of each combined wave band, taking the local regression model with the minimum RMSECV value as the next selected model, and taking the combined wave band corresponding to the local regression model with the minimum RMSECV value as the next selected wave band;
judging whether all the wave bands are combined, if so, outputting the selected wave band corresponding to the selected model with the minimum RMSECV value as an optimal interval, and otherwise, turning to the previous step to perform next wave band combination;
and performing partial least squares regression modeling on the correction set on the optimal interval to obtain the property correction model.
When a mathematical model is established, the spectrum interval needs to be considered, and the selection of the spectrum interval determines the number of variables participating in fitting. Therefore, the present embodiment first obtains the optimal interval, and then establishes the property correction model on the optimal interval.
Specifically, the formula for calculating the RMSECV value is as follows:
wherein RMSECV is the RMSECV value, the number of n component oil samples, yiIs a property measurement of the ith component oil sample,is a predicted value of the property of the ith component oil sample.Is a predicted value of a property obtained from a local regression model or a property correction model.
Preferably, the screening of each of the component oil samples according to the property measurements to establish a calibration set further comprises: removing abnormal points in the correction set:
calculating the mahalanobis distance for the concentration of each of the component oil samples:
MD=sT(SST)-1s
wherein MD is the Mahalanobis distance, s is the score vector of the correction set, and s isTIs a transposed matrix of S, S is a score matrix of the correction set, STIs a transposed matrix of S;
judging whether MD is more than 3k/n, wherein n is the number of component oil samples in the correction set, k is the number of main components, if so, judging that the corresponding component oil samples are first-class abnormal points, and rejecting the first-class abnormal points;
and judging whether a second type of abnormal points exist in the correction set by adopting a t detection method:
wherein, tiMeasurement of the i component oil sample, yiIs the ithA property measurement of a sample of the component oil,is the predicted value of the property of the ith component oil sample, MD is the mahalanobis distance of the concentration of the ith component oil sample, d is the degree of freedom of the property correction model, d is n-k-1,to correct the predicted values of properties of the pooled component oil samples; y is the actual value of the property of the corrected concentrated component oil sample;
inquiring a boundary value corresponding to the degree of freedom d in a t boundary value table, and judging the detection statistic tiAnd judging whether the degree of freedom d is smaller than a corresponding threshold value of the degree of freedom d, if not, judging that the ith component oil sample is a second type abnormal point, and rejecting the second type abnormal point.
The modeling process of the property correction model has two abnormal points, wherein the first abnormal point is compared with other component oil samples in the correction set, and the component concentration of the component oil sample is extreme; the second type of abnormal points are the significant difference between the property predicted value obtained by the property correction model and the property measured value obtained by the reference method, the second type of abnormal points indicate that the property measured value obtained by the reference method is possibly wrong and needs to be measured again, and if the property measured value is correct, the property correction model is not suitable for the prediction of the component oil sample and needs to eliminate the abnormal points. In this example, the reference method is near infrared spectroscopy. The first type of abnormal point can be detected by means of mahalanobis distance values, and the second type of abnormal point can be detected by means of t-test.
Preferably, the method further comprises:
predicting the properties of the concentrated component oil sample to obtain a property predicted value by using the property correction model;
and detecting the property predicted value and the property measured value by adopting a t detection method:
wherein t is the test statistic of the correction set,average of the differences between the predicted and measured values of the properties of the respective component oil samples, SdIs the standard deviation of the difference between the predicted value of the property and the measured value of the property, and m is the number of component oil samples;
setting a significance level value, inquiring a boundary value corresponding to the significance level in a t boundary value table, and checking whether a statistic t is smaller than the boundary value corresponding to the significance level, wherein if the statistic t is smaller than the boundary value corresponding to the significance level, the property correction model meets the precision requirement, otherwise, the property correction model does not meet the precision requirement.
In order to ensure the availability of the property correction model, a t-test method may be used to check whether there is a significant difference between the predicted property value and the measured property value in the correction set<t(m-1,α),t(m-1,α)And the boundary value corresponding to α -0.05 in the t boundary value table indicates that the predicted value and the measured value of the property have no significant difference and meet the requirement of prediction accuracy.
Preferably, the method further comprises: and updating the correction set, and reestablishing the property correction model according to the updated correction set.
If a new component oil sample needs to be added into the correction set, the application range of the property correction model is expanded, the property correction model needs to be updated, the operation of establishing the property correction model, checking abnormal points, verifying the model and the like is carried out according to the new correction set and the method, so that the new property correction model is obtained, the new property correction model is suitable for the type of the new component oil sample, and the accuracy of the new property correction model is ensured.
Preferably, a historical blending formula is collected, the property values of the component oils in the historical blending formula are used as input, the property values of the finished oil obtained according to the historical blending formula are used as output, a neural network is trained, and a blending model is obtained, and the method specifically comprises the following steps:
and taking the quantity of the properties of the component oil in the historical blending formula as the quantity of neurons of an input layer of the neural network, taking the properties of the component oil in the historical blending formula as the initial value of the neurons of the input layer of the neural network, calculating the value of the neurons of a hidden layer of the neural network by taking a sigmoid function as an activation function according to the value of the neurons of the input layer, and training the neural network by taking the properties of the product oil of the historical blending formula as the value of the neurons of an output layer of the neural network to obtain the blending model.
Calculating the value of the neuron of the hidden layer and the value of the neuron of the output layer by adopting a sigmoid function as an activation function according to the value of the neuron of the input layer:
in the formula:for the linear sum of the input layers (note: custom intermediate variable values, dimensionless, without specific meaning, just for convenience of simplification), xqIs the input value of the input layer after standardization;is the weight coefficient from the input layer to the hidden layer;m is the node number of the input layer;input parameters from the hidden layer to the output layer; f () represents a sigmoid function; z is a radical of(3)Is the output value of the output layer;hidden layer to output layer weight coefficients; b(3)For hidden layer to output layer biasing, n is the number of nodes of the hidden layer.
wherein, delta(3)For the output layer sample fitting error, ypAnd E is the total error.
wherein the content of the first and second substances,f' () is derived for an activation function for self-defining intermediate variables (note: self-defining intermediate variable values, dimensionless, no specific meaning, and convenient simplification);
biasing term b to the output layer(3)Updating:
adjusting the learning rate:
in the formula: μ is the updated value of the learning rate, μiThe initial value of the learning rate, r is the attenuation rate of the learning rate, n is the total running round and s is the updating frequency;
and calculating an error value of the updated harmonic model, judging whether the error value is within a set error range, outputting the harmonic model if the error value is within the set error range, and otherwise, updating the harmonic model next time.
Preferably, the diesel blending process is controlled on line according to the property correction model and the blending model, and the method specifically comprises the following steps:
setting a constraint condition according to the property requirement of the product oil, setting a target function according to the cost requirement, obtaining an optimal blending formula according to the constraint condition, the target function and a blending model, calculating an optimal blending concentration ratio corresponding to the optimal blending formula according to the property correction model, and performing online control on the diesel blending process according to the optimal blending concentration ratio.
In this embodiment, the objective function and the constraint condition are as follows:
g(x)≤0
Ω(Y)=True
0≤x≤U,Yjk∈{True,False}(x∈Rn)
in the formula: z (x) is an objective function on profit set according to cost requirements; piIs the price of the ith component oil sample; xiThe blending amount of the ith product oil is; xjThe amount of the jth component oil used to blend the ith product oil; fjIs the price of the jth component oil sample; b isiThe unit blending cost of the ith finished oil is; the relationship of V is OR; y isjkIs a Boolean variable when YjkIn order to be true, the corresponding constraint h must be taken into accountjk(x) Less than or equal to 0, if YjkIf false, ignore the corresponding constraint hjk(x) Less than or equal to 0; g (x) is less than or equal to 0 and is a material balance constraint condition hjk(x) Less than or equal to 0 is the product property constraint, g (x) is the available inventory constraint of the desired component diesel, YjkWhether the product oil takes into account the logical value of the property constraint, hjk(x) Is a constraint value of the product property index.
The constraint conditions comprise linear constraint and nonlinear constraint, so the optimization problem is a nonlinear programming problem. In this model, the objective function is a linear function. The number of equality constraints in the constraint conditions is small, the form is simple, and most of the constraints are inequality constraints.
The diesel oil on-line blending firstly needs to set the upper and lower limits of the target property value, blending quantity, constraint conditions and the like of the final blended product. After the operation conditions are set, the blending optimization control stage is started to enter, the blending optimization control stage is completed by being divided into a series of small periods, the blending optimization module in each small period optimally designs a formula for blending each component according to the previously set operation conditions and the quality value of each component oil fed back by the near-infrared analyzer on line by using a blending rule, the formula is downloaded to the execution control module, field equipment is controlled to blend according to the formula, the quality of the blended product finally meets the constraint condition, and the quality of the product is guaranteed to be qualified.
In each small period, the analyzer monitors the blended oil product on line to obtain the real quality value of the blended oil product, the blending optimization module predicts the estimated quality value of the oil product after blending according to the respective analysis values of each group and the actual blending proportion on site by using the blending rule, the oil quality value obtained by the analyzer is compared with the oil quality value estimated by the blending optimization module to obtain an estimated deviation through mathematical treatment, and the deviation value is replaced into the blending rule to correct the blending rule, so that the oil product is correctly guided to be blended on line.
The above-described embodiments of the present invention should not be construed as limiting the scope of the present invention. Any other corresponding changes and modifications made according to the technical idea of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. An on-line diesel blending method is characterized by comprising the following steps:
performing spectral analysis on component oil samples of different types and different concentrations to obtain a property measurement value of each component oil sample;
screening each component oil sample according to the property measurement value to establish a correction set, and fitting the correction set to obtain a property correction model between the concentration and the property measurement value;
collecting a historical blending formula, taking the property value of the component oil in the historical blending formula as input, taking the property value of the finished oil obtained according to the historical blending formula as output, and training a neural network to obtain a blending model;
and carrying out on-line control on the diesel oil blending process according to the property correction model and the blending model.
2. The on-line diesel blending method according to claim 1, wherein the spectral analysis is performed on component oil samples of different types and concentrations to obtain property measurement values of each of the component oil samples, specifically:
and performing spectral analysis on the component oil samples of different types and different concentrations by using a near-infrared spectrum analyzer to obtain the property measured value of each component oil sample.
3. The on-line diesel blending method of claim 1, wherein the screening of the property measurements of each of the component oil samples to establish a calibration set further comprises: pre-processing the data for the property measurements;
preprocessing the data of the property measurement value, specifically:
smoothing the property measurement value by adopting a window moving average algorithm:
wherein the content of the first and second substances,the property measurement value after the k wavelength point smoothing processing is shown, 2w +1 is the window width, xk+iProperty measurement values before smoothing processing are the k + i wavelength points;
derivative processing of the property measurements eliminates spectral baseline drift:
X'(i)=[x(i+g)-x(i)]/g
X”(i)=[x(i+g)-2x(i)+x(i-g)]/g2
wherein, X '(i) is the property measurement value after the first-order differentiation of the ith wavelength point, X' (i) is the property measurement value after the second-order differentiation of the ith wavelength point, X (i + g) is the property measurement value before the derivative processing of the ith wavelength point, and X (i) is the property measurement value before the derivative processing of the ith wavelength point;
data centralizing and normalizing the property prediction values:
wherein x' (i) is the property measurement value after the centering and the standardization of the ith wavelength point data, x (i) is the property measurement value before the centering and the standardization of the ith wavelength point data, u is the property measurement mean value of each component oil sample, and sigma is the standard deviation of the property measurement value of each component oil sample.
4. The diesel on-line blending method according to claim 1, wherein a calibration set is established by screening each component oil sample according to the property measurement value, specifically:
selecting two component oil samples with the farthest Euclidean distance between the property measurement values and adding the two component oil samples into an initial calibration set;
respectively calculating the Euclidean distance between the property measured value of each component oil sample which is not added into the correction set and the property measured value of each component oil sample which is added into the correction set, and adding the component oil sample with the farthest Euclidean distance into the correction set;
and judging whether the number of the samples of the correction set is greater than a set threshold value, if so, outputting the correction set, and otherwise, turning to the previous step.
5. The on-line diesel blending method according to claim 1, wherein the correction set is fitted to obtain a property correction model between the concentration and the property measurement values, specifically:
dividing a spectrum region into a plurality of equal-width wave bands, performing partial least squares regression modeling on the correction set on each wave band to obtain a local regression model of each wave band, calculating the RMSECV value of the local regression model of each wave band, taking the local regression model with the minimum RMSECV value as a first selected model, and taking the wave band corresponding to the local regression model with the minimum RMSECV value as a first selected wave band;
respectively combining other wave bands except the selected wave band with the previous selected wave band to obtain a plurality of combined wave bands, performing partial least squares regression modeling on the correction set on each combined wave band to obtain a local regression model of each combined wave band, taking the local regression model with the minimum RMSECV value as the next selected model, and taking the combined wave band corresponding to the local regression model with the minimum RMSECV value as the next selected wave band;
judging whether all the wave bands are combined, if so, outputting the selected wave band corresponding to the selected model with the minimum RMSECV value as an optimal interval, and otherwise, turning to the previous step to perform next wave band combination;
and performing partial least squares regression modeling on the correction set on the optimal interval to obtain the property correction model.
6. The on-line diesel blending method of claim 1, wherein screening each of the component oil samples according to the property measurements to establish a calibration set further comprises: removing abnormal points in the correction set:
calculating the mahalanobis distance for the concentration of each of the component oil samples:
MD=sT(SST)-1s
wherein MD is the Mahalanobis distance, s is the score vector of the correction set, and s isTIs a transposed matrix of S, S is a score matrix of the correction set, STIs a transposed matrix of S;
judging whether MD is more than 3k/n, wherein n is the number of component oil samples in the correction set, k is the number of main components, if so, judging that the corresponding component oil samples are first-class abnormal points, and rejecting the first-class abnormal points;
and judging whether a second type of abnormal points exist in the correction set by adopting a t detection method:
wherein, tiMeasurement of the i component oil sample, yiIs a property measurement of the ith component oil sample,is the predicted value of the property of the ith component oil sample, MD is the mahalanobis distance of the concentration of the ith component oil sample, d is the degree of freedom of the property correction model, d is n-k-1,to correct the predicted values of properties of the pooled component oil samples; y is the actual value of the property of the corrected concentrated component oil sample;
inquiring a boundary value corresponding to the degree of freedom d in a t boundary value table, and judging the detection statistic tiAnd judging whether the degree of freedom d is smaller than a corresponding threshold value of the degree of freedom d, if not, judging that the ith component oil sample is a second type abnormal point, and rejecting the second type abnormal point.
7. The on-line diesel blending method of claim 1, further comprising:
predicting the properties of the concentrated component oil sample to obtain a property predicted value by using the property correction model;
and detecting the property predicted value and the property measured value by adopting a t detection method:
wherein t is the test statistic of the correction set,average of the differences between the predicted and measured values of the properties of the respective component oil samples, SdIs the standard deviation of the difference between the predicted value of the property and the measured value of the property, and m is the number of component oil samples;
setting a significance level value, inquiring a boundary value corresponding to the significance level in a t boundary value table, and checking whether a statistic t is smaller than the boundary value corresponding to the significance level, wherein if the statistic t is smaller than the boundary value corresponding to the significance level, the property correction model meets the precision requirement, otherwise, the property correction model does not meet the precision requirement.
8. The on-line diesel blending method of claim 1, further comprising: and updating the correction set, and reestablishing the property correction model according to the updated correction set.
9. The diesel on-line blending method according to claim 1, wherein a historical blending recipe is collected, a property value of a component oil in the historical blending recipe is used as an input, a property value of a finished oil obtained according to the historical blending recipe is used as an output, a neural network is trained, and a blending model is obtained, specifically:
and taking the quantity of the properties of the component oil in the historical blending formula as the quantity of neurons of an input layer of the neural network, taking the properties of the component oil in the historical blending formula as the initial value of the neurons of the input layer of the neural network, calculating the value of the neurons of a hidden layer of the neural network by taking a sigmoid function as an activation function according to the value of the neurons of the input layer, and training the neural network by taking the properties of the product oil of the historical blending formula as the value of the neurons of an output layer of the neural network to obtain the blending model.
10. The diesel on-line blending method according to claim 1, wherein the diesel blending process is controlled on line according to the property correction model and the blending model, specifically:
setting a constraint condition according to the property requirement of the product oil, setting a target function according to the cost requirement, obtaining an optimal blending formula according to the constraint condition, the target function and a blending model, calculating an optimal blending concentration ratio corresponding to the optimal blending formula according to the property correction model, and performing online control on the diesel blending process according to the optimal blending concentration ratio.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911075166.9A CN110987862A (en) | 2019-11-06 | 2019-11-06 | Diesel oil on-line blending method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911075166.9A CN110987862A (en) | 2019-11-06 | 2019-11-06 | Diesel oil on-line blending method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110987862A true CN110987862A (en) | 2020-04-10 |
Family
ID=70083497
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911075166.9A Pending CN110987862A (en) | 2019-11-06 | 2019-11-06 | Diesel oil on-line blending method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110987862A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112396322A (en) * | 2020-11-18 | 2021-02-23 | 广东赛意信息科技有限公司 | Production plan-based process technology unit productivity assessment method and system |
CN113159435A (en) * | 2021-04-29 | 2021-07-23 | 浙江吉利控股集团有限公司 | Method and system for predicting remaining driving mileage of new energy vehicle |
CN113191547A (en) * | 2021-04-29 | 2021-07-30 | 浙江吉利控股集团有限公司 | New energy vehicle power battery charging optimization method and system |
US11702600B2 (en) | 2021-02-25 | 2023-07-18 | Marathon Petroleum Company Lp | Assemblies and methods for enhancing fluid catalytic cracking (FCC) processes during the FCC process using spectroscopic analyzers |
US11802257B2 (en) | 2022-01-31 | 2023-10-31 | Marathon Petroleum Company Lp | Systems and methods for reducing rendered fats pour point |
US11835450B2 (en) | 2021-02-25 | 2023-12-05 | Marathon Petroleum Company Lp | Methods and assemblies for determining and using standardized spectral responses for calibration of spectroscopic analyzers |
US11891581B2 (en) | 2017-09-29 | 2024-02-06 | Marathon Petroleum Company Lp | Tower bottoms coke catching device |
US11898109B2 (en) | 2021-02-25 | 2024-02-13 | Marathon Petroleum Company Lp | Assemblies and methods for enhancing control of hydrotreating and fluid catalytic cracking (FCC) processes using spectroscopic analyzers |
US11905479B2 (en) | 2020-02-19 | 2024-02-20 | Marathon Petroleum Company Lp | Low sulfur fuel oil blends for stability enhancement and associated methods |
US11905468B2 (en) | 2021-02-25 | 2024-02-20 | Marathon Petroleum Company Lp | Assemblies and methods for enhancing control of fluid catalytic cracking (FCC) processes using spectroscopic analyzers |
US11970664B2 (en) | 2023-05-08 | 2024-04-30 | Marathon Petroleum Company Lp | Methods and systems for enhancing processing of hydrocarbons in a fluid catalytic cracking unit using a renewable additive |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101987979A (en) * | 2009-08-07 | 2011-03-23 | 中国石油天然气股份有限公司 | Method for re-blending refined diesel oil |
CN103065204A (en) * | 2012-12-25 | 2013-04-24 | 浙江大学 | VERNA-genetic algorithm (GA) gasoline concoction optimized dispatching method |
CN104297201A (en) * | 2014-11-03 | 2015-01-21 | 天津工业大学 | Method for quickly, accurately and quantitatively detecting ratio of various oil components in blend oil |
CN104965967A (en) * | 2015-03-26 | 2015-10-07 | 华东理工大学 | Yield real-time prediction method for atmospheric and vacuum distillation unit |
CN106680241A (en) * | 2017-01-13 | 2017-05-17 | 北京化工大学 | Novel spectrum multi-analysis classification and identification method and application thereof |
CN107748146A (en) * | 2017-10-20 | 2018-03-02 | 华东理工大学 | A kind of crude oil attribute method for quick predicting based near infrared spectrum detection |
CN110095429A (en) * | 2019-04-30 | 2019-08-06 | 济南弗莱德科学仪器有限公司 | A kind of product oil method for quickly detecting quality |
-
2019
- 2019-11-06 CN CN201911075166.9A patent/CN110987862A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101987979A (en) * | 2009-08-07 | 2011-03-23 | 中国石油天然气股份有限公司 | Method for re-blending refined diesel oil |
CN103065204A (en) * | 2012-12-25 | 2013-04-24 | 浙江大学 | VERNA-genetic algorithm (GA) gasoline concoction optimized dispatching method |
CN104297201A (en) * | 2014-11-03 | 2015-01-21 | 天津工业大学 | Method for quickly, accurately and quantitatively detecting ratio of various oil components in blend oil |
CN104965967A (en) * | 2015-03-26 | 2015-10-07 | 华东理工大学 | Yield real-time prediction method for atmospheric and vacuum distillation unit |
CN106680241A (en) * | 2017-01-13 | 2017-05-17 | 北京化工大学 | Novel spectrum multi-analysis classification and identification method and application thereof |
CN107748146A (en) * | 2017-10-20 | 2018-03-02 | 华东理工大学 | A kind of crude oil attribute method for quick predicting based near infrared spectrum detection |
CN110095429A (en) * | 2019-04-30 | 2019-08-06 | 济南弗莱德科学仪器有限公司 | A kind of product oil method for quickly detecting quality |
Non-Patent Citations (3)
Title |
---|
张志宏 等: "开发柴油在线优化调合技术的可行性探讨", 《炼油技术与工程》 * |
徐广通等: "CCD近红外光谱仪在柴油生产控制分析中的应用", 《石油炼制与化工》 * |
李为民等: "基于ANN方法预测柴油调和的凝点和冷滤点", 《石油与天然气化工》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11891581B2 (en) | 2017-09-29 | 2024-02-06 | Marathon Petroleum Company Lp | Tower bottoms coke catching device |
US11920096B2 (en) | 2020-02-19 | 2024-03-05 | Marathon Petroleum Company Lp | Low sulfur fuel oil blends for paraffinic resid stability and associated methods |
US11905479B2 (en) | 2020-02-19 | 2024-02-20 | Marathon Petroleum Company Lp | Low sulfur fuel oil blends for stability enhancement and associated methods |
CN112396322A (en) * | 2020-11-18 | 2021-02-23 | 广东赛意信息科技有限公司 | Production plan-based process technology unit productivity assessment method and system |
US11885739B2 (en) | 2021-02-25 | 2024-01-30 | Marathon Petroleum Company Lp | Methods and assemblies for determining and using standardized spectral responses for calibration of spectroscopic analyzers |
US11835450B2 (en) | 2021-02-25 | 2023-12-05 | Marathon Petroleum Company Lp | Methods and assemblies for determining and using standardized spectral responses for calibration of spectroscopic analyzers |
US11860069B2 (en) | 2021-02-25 | 2024-01-02 | Marathon Petroleum Company Lp | Methods and assemblies for determining and using standardized spectral responses for calibration of spectroscopic analyzers |
US11702600B2 (en) | 2021-02-25 | 2023-07-18 | Marathon Petroleum Company Lp | Assemblies and methods for enhancing fluid catalytic cracking (FCC) processes during the FCC process using spectroscopic analyzers |
US11898109B2 (en) | 2021-02-25 | 2024-02-13 | Marathon Petroleum Company Lp | Assemblies and methods for enhancing control of hydrotreating and fluid catalytic cracking (FCC) processes using spectroscopic analyzers |
US11905468B2 (en) | 2021-02-25 | 2024-02-20 | Marathon Petroleum Company Lp | Assemblies and methods for enhancing control of fluid catalytic cracking (FCC) processes using spectroscopic analyzers |
US11906423B2 (en) | 2021-02-25 | 2024-02-20 | Marathon Petroleum Company Lp | Methods, assemblies, and controllers for determining and using standardized spectral responses for calibration of spectroscopic analyzers |
US11921035B2 (en) | 2021-02-25 | 2024-03-05 | Marathon Petroleum Company Lp | Methods and assemblies for determining and using standardized spectral responses for calibration of spectroscopic analyzers |
CN113159435B (en) * | 2021-04-29 | 2022-11-22 | 浙江吉利控股集团有限公司 | Method and system for predicting remaining driving mileage of new energy vehicle |
CN113191547A (en) * | 2021-04-29 | 2021-07-30 | 浙江吉利控股集团有限公司 | New energy vehicle power battery charging optimization method and system |
CN113159435A (en) * | 2021-04-29 | 2021-07-23 | 浙江吉利控股集团有限公司 | Method and system for predicting remaining driving mileage of new energy vehicle |
US11802257B2 (en) | 2022-01-31 | 2023-10-31 | Marathon Petroleum Company Lp | Systems and methods for reducing rendered fats pour point |
US11970664B2 (en) | 2023-05-08 | 2024-04-30 | Marathon Petroleum Company Lp | Methods and systems for enhancing processing of hydrocarbons in a fluid catalytic cracking unit using a renewable additive |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110987862A (en) | Diesel oil on-line blending method | |
Chen et al. | A fuzzy optimization strategy for the implementation of RBF LSSVR model in vis–NIR analysis of pomelo maturity | |
CN105300923B (en) | Without measuring point model of temperature compensation modification method during a kind of near-infrared spectrometers application on site | |
CN103528990B (en) | A kind of multi-model Modeling Method of near infrared spectrum | |
CN104949936A (en) | Sample component determination method based on optimizing partial least squares regression model | |
CN110555230A (en) | rotary machine residual life prediction method based on integrated GMDH framework | |
CN109324013A (en) | A method of it is quickly analyzed using Gaussian process regression model building oil property near-infrared | |
CN105334185A (en) | Spectrum projection discrimination-based near infrared model maintenance method | |
CN115166618B (en) | Current transformer error evaluation method for non-stable output | |
Kohonen et al. | Multi‐block methods in multivariate process control | |
CN113758890A (en) | Gas concentration calculation method, device, equipment and storage medium | |
CN116088453A (en) | Production quality prediction model training method and device and production quality monitoring method | |
CN113420500B (en) | Intelligent atmospheric and vacuum system | |
Chang et al. | Detection of sugar content in food based on the electrochemical method with the assistance of partial least square method and deep learning | |
CN106485049B (en) | A kind of detection method of the NIRS exceptional sample based on Monte Carlo cross validation | |
Mansour et al. | Methodology of on-line optimisation applied to a chemical reactor | |
CN112651173A (en) | Agricultural product quality nondestructive testing method based on cross-domain spectral information and generalizable system | |
Funes et al. | Predictive ANN models for the optimization of extra virgin olive oil clarification by means of vertical centrifugation | |
Liu et al. | The man-hour estimation models & its comparison of interim products assembly for shipbuilding | |
CN116380816A (en) | Method, program and storage medium for judging faults of coal quality online detection system | |
CN113607683A (en) | Automatic modeling method for near infrared spectrum quantitative analysis | |
Taylan | Estimating the quality of process yield by fuzzy sets and systems | |
Abonyi et al. | Data-driven development and maintenance of soft-sensors | |
CN104049624B (en) | Chemical products production model optimization method, device and continuous type chemical system | |
Altieri et al. | Models for the rapid assessment of water and oil content in olive pomace by near‐infrared spectrometry |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200410 |