Oil refining key physical property parameter description method based on nonlinear multi-round median interpolation algorithm

Abstract

The invention discloses an oil refining key physical property parameter description method based on a nonlinear multi-round median interpolation algorithm. The method comprises the following steps: S1, collecting initial experimental data of oil refining key physical property parameters and inputting a system; s2, predicting oil refining key physical property parameters by adopting a median insertion algorithm, specifically comprising the following steps: S2-1, acquiring an actually acquired sampling sequence and an observed value sequence, and setting a multi-round median sampling coefficient M; s2-2, obtaining a point sequence to be interpolated; s2-3, performing multiple rounds of median sampling to obtain a final sampling sequence and a final observation value sequence; and S2-4, all points to be interpolated in the point sequence to be interpolated are judged to belong to sub-intervals of the final sampling sequence and the final observation value sequence, and then a predicted value is obtained through linear interpolation calculation. The method has the following effects: (1) the nonlinear error is obviously reduced; (2) dense sampling is not depended on, and the sampling cost is low; and (3) the algorithm is simple and the calculation cost is extremely low.

Description

Technical Field

[0001] The present invention relates to the field of refinery production optimization, and specifically to a non-linear prediction method for key physical property parameters (such as closed cup flash point, freezing point, viscosity) in the blending process of diesel and gasoline, which solves the problem of linear interpolation error caused by insufficient experimental sampling points and is directly embedded in the process control system to optimize the blending production. Background Art

[0002] In the oil refining industry, due to the complexity of the process, the diversity of the devices and the non-linear characteristics of the process flow, many key physical property parameters (such as freezing point, flash point, smoke point, viscosity, etc.) cannot be directly described by theoretical relational expressions. For example:

[0003] 1) The non-linear nature of physical property correlation: For example, the fractional formula y = 1 / x, the exponential formula y = a x , the logarithmic formula y = log a x or the power function formula y = x a relationships are widespread. Taking the smoke point and the smoke point index as an example, their correlation may be affected by many factors such as raw material composition and reaction conditions, showing a non-linear response.

[0004] 2) The complexity of the blending process: Oil refining production involves a large number of blending processes (such as gasoline blending, diesel blending), and its physical property parameters need to be obtained by interpolating experimental data after multi-component mixing. However, the blending process is affected by the synergistic effect between components, and the linear superposition hypothesis is difficult to hold.

[0005] Due to the monotonicity requirement (such as the freezing point changes monotonically with the component ratio), linear interpolation has become the mainstream method. For example, refineries construct a linear relationship table between the freezing point index and the freezing point through limited experimental data points to quickly estimate the freezing point value of unmeasured points. The principle of linear interpolation is: Linear interpolation assumes that the physical property change between two adjacent data points is a straight line relationship. Given the known points (x1, y1) and (x2, y2), the calculation formula for the target value y in the interval of x1 < x < x2 is:

[0006] Linear interpolation has the following defects:

[0007] (1) Significant non-linear error: The actual physical property relationship is often a curve (such as the exponential change of the smoke point), and linear interpolation will introduce systematic deviation, as Figure 1 shown;

[0008] (2) Dependence on dense sampling: To reduce the error, it is necessary to increase the density of experimental data points, resulting in an increase in cost. For example, changing the original sampling density of the closed cup flash point from every 5°C to every 0.5°C will surely greatly reduce the error, but the cost will increase by 10 times. Table 1 gives an example of the actual sampling points of the closed cup flash point and the closed cup flash point index.

[0009] Table 1 Actual Sampling Points of Closed Cup Flash Point and Closed Cup Flash Point Index

[0010]

[0011] As shown in Table 2, the external sampling closed cup flash points are 7.5, 21.5,..., 47.5, and the actual sampling values of the closed cup flash point index are recorded. Using linear interpolation calculation, as shown in the "Linear Interpolation" column in Table 2, the error is as shown in "Linear Interpolation Error". The average absolute error is 54.66.

[0012] Table 2 Linear Interpolation of Closed Cup Flash Point and Closed Cup Flash Point Index

[0013] Interpolation point Actual sampling Linear interpolation Linear interpolation error 7.5 10512 10335.62 176.50 12.5 6503 6390.01 112.57 17.5 4101 4028.39 72.62 22.5 2634 2586.53 47.41 27.5 1721 1689.65 31.33 32.5 1143 1121.85 20.95 37.5 771 756.37 14.17 42.5 527 517.40 9.70 47.5 366 358.81 6.71

[0014] Combined with Table 2 and Figure 2 , the smaller the closed cup flash point, the steeper the change of the closed cup flash point index, and the greater the linear interpolation error. Summary of the Invention

[0015] In view of the problems existing in the background technology, the present invention proposes a description method for key physical properties parameters of oil refining based on a non-linear multi-round median insertion algorithm.

[0016] Technical Solution:

[0017] A description method for key physical properties parameters of oil refining based on a non-linear multi-round median insertion algorithm, which includes the following steps:

[0018] S1. Initial experimental data collection of key physical properties parameters of oil refining and system input;

[0019] S2. Predict key physical properties parameters of oil refining by using the median insertion algorithm, specifically including:

[0020] S2-1. Obtain the actually collected sampling sequence and observation value sequence, and set the multi-round median sampling coefficient M;

[0021] S2-2. Obtain the sequence of points to be interpolated;

[0022] S2-3. Perform multi-round median sampling to obtain the final sampling sequence and the final observation value sequence;

[0023] S2-4. For all the points to be inserted in the sequence of points to be interpolated, judge the sub-interval to which they belong in the final sampling sequence and the final observation value sequence, and then use linear interpolation calculation to obtain the predicted value.

[0024] Preferably, in step S1, it specifically includes:

[0025] 1) Sample collection. In the blending laboratory, test the key physical properties parameters of oil refining at equal interval steps to generate a sampling sequence;

[0026] 2) Sample storage: Store the sampling sequence in the process database of the Manufacturing Execution System (MES) for algorithm call.

[0027] 3) Data preprocessing: Verify the monotonicity of the sampling data. If there are abnormal points, re - experiment and supplement the measured data.

[0028] Preferably, in step S2 - 3, the specific steps of median sampling include:

[0029] 1) Define the sampling points: Let the equally - spaced sampling point sequence be x1, x2,..., x n , and the corresponding observed value sequence be y1, y2,..., y n ; where n≥4.

[0030] 2) Calculate the adjacent differences Δy of the observed values n :

[0031] Δy i = y i - y i+1 , i = 1, 2,..., n - 1

[0032] Generate the difference sequence Δy = [Δy1, Δy2,..., Δy n-1 .

[0033] 3) Calculate the difference ratio Δp i :

[0034]

[0035] Generate the difference ratio sequence Δp = [Δp1, Δp2,..., Δp n-1 .

[0036] 4) Calculate the square root of the difference ratio sp i :

[0037]

[0038] Generate the square - root - of - difference - ratio sequence sp = [sp1, sp2,..., sp n-2 .

[0039] 5) Expansion of the square root of the difference ratio

[0040] a. Left - hand - side expansion of sp0

[0041] If the number of terms in the difference - ratio sequence sp≥3, select the left - most three terms sp1~sp3. Due to the monotonicity of the observed - value sequence y, sp1~sp3 must also be monotonically increasing or decreasing. The left - hand - side expansion of sp0 is:

[0042] sp0 = sp1 + sp2 - sp3

[0043] If the number of terms in the difference ratio sequence sp = 2, only the leftmost two terms sp1 to sp2 can be selected. Due to the monotonicity of the observed value sequence y, sp1 to sp2 must also be monotonically increasing or decreasing. The left extension sp0 is:

[0044] sp0 = 2sp1 - sp2

[0045] b. Right extension of sp n-1

[0046] If the number of terms in the difference ratio sequence sp ≥ 3, select the rightmost three terms sp n-4 ~sp n-2 , due to the monotonicity of the observed value sequence y, sp n-4 ~sp n-2 must also be monotonically increasing or decreasing. The right extension sp n-1 is:

[0047] sp n-1 = sp n-2 + sp n-3 - sp n-4

[0048] If the number of terms in the difference ratio sequence sp = 2, only the rightmost two terms sp n-3 ~sp n-2 can be selected. Due to the monotonicity of the observed value sequence y, sp n-3 ~sp n-2 must also be monotonically increasing or decreasing. The right extension sp n-1 is:

[0049] sp n-1 = 2sp n-2 - sp n-3

[0050] c. Re - coding

[0051] After the above operations, the number of terms in the difference ratio sequence sp is expanded from the original n - 2 to n; the new difference ratio sequence sp is [sp0, sp1,..., sp n-1 ; keeping the order unchanged, after re - coding it is [sp1, sp2,..., sp n ;

[0052] 6) Construct the median point sequence

[0053]

[0054] Generate the difference sequence x′ = [x1′, x2′,..., x n-1 ′];

[0055] 7) Median point interpolation prediction

[0056] a. Calculate the left interpolation ly′ of the median point i :

[0057]

[0058] b. Calculate the right interpolation ry′ of the median point i :

[0059]

[0060] c. Calculate the interpolation y i ′ of the median point:

[0061]

[0062] 8) Form a new sampling sequence and observation value sequence

[0063] The new sampling sequence is x1, x1′, x2, x2′,..., x n-1 , x n-1 , x n , and the new observation value sequence is y1, y1′, y2, y2′,..., y n-1 , y n-1 , y n , and output the two new sequences.

[0064] Preferably, M rounds of median sampling are performed. In each round of median sampling, the new sampling sequence and observation value sequence newly formed after the previous round of median sampling are used as the input of the current round of median sampling.

[0065] Preferably, it further includes the following steps:

[0066] S3. The APC system adjusts the blending ratio according to the predicted value and issues relevant process parameters; at the same time, it monitors the data of key refinery physical property parameters in real time and transmits them back to the MES system;

[0067] S4. Update the process knowledge base, store the monitored data of the key refinery physical property parameters after actual blending into the process knowledge base, and optimize the initial sampling point selection strategy of the subsequent prediction model.

[0068] Preferably, if the deviation between the measured data of the key refinery physical property parameters and the predicted value exceeds the threshold, trigger the median insertion algorithm to regenerate the sampling sequence.

[0069] Advantages of the present invention

[0070] (1) The non - linear error is significantly reduced;

[0071] (2) It does not rely on dense sampling and has a low sampling cost;

[0072] (3) The algorithm is simple and the calculation cost is extremely low;

[0073] (4) Suitable for predicting physical property correlations with equally spaced sampling of any type and monotonic changes. Description of the Drawings

[0074] Figure 1 It is a schematic diagram of the systematic deviation introduced by linear interpolation in the background technology.

[0075] Figure 2 It is a schematic diagram of the actual sampling points of the closed-cup flash point and the closed-cup flash point index in the background technology.

[0076] Figure 3 It is a system integration flowchart of the present invention. Detailed Implementation Manner

[0077] The present invention will be further described below in conjunction with the specific operation steps of the blending production process, but the protection scope of the present invention is not limited thereto:

[0078] (1) Initial experimental data collection and system input

[0079] 1) Sample collection

[0080] In the blending laboratory, the closed-cup flash point test is carried out at equal interval steps (such as every 5 °C) to generate a sampling sequence.

[0081] 2) Sample storage

[0082] The sampling sequence is stored in the process database of the MES (Manufacturing Execution System) for algorithm call.

[0083] 3) Data preprocessing

[0084] Verify the monotonicity of the sampling data (such as the closed-cup flash point index should decrease monotonically with temperature).

[0085] If there are abnormal points (such as local non-monotonicity), re-experiment and supplement the data.

[0086] (2) Single-round median sampling

[0087] 1) Define the sampling points, and set the equally spaced sampling point sequence as x1, x2,..., x n , and the corresponding observed value sequence as y1, y2,..., y n .

[0088] Where n≥4. Too few sampling points cannot perform interpolation calculation.

[0089] 2) Calculate the adjacent differences of the observed values

[0090] Δy i = yi -y i+1 (i = 1, 2, ..., n - 1)

[0091] Generate a difference sequence Δy = [Δy1, Δy2, ..., Δy n-1 .

[0092] 3) Calculate the difference ratio

[0093]

[0094] Generate a difference ratio sequence Δp = [Δp1, Δp2, ..., Δp n-1 .

[0095] 4) Calculate the square root of the difference ratio

[0096]

[0097] Generate a difference ratio square root sequence sp = [sp1, sp2, ..., sp n-2 .

[0098] Among them, the derivation process of why it is converted to the square root of the difference ratio here is as follows:

[0099] For three consecutive sampling points x i , x i+1 , x i+2 , corresponding to two sampling medians x i ′, x i+1 ′, forming a new sampling sequence x i , x i ′, x i+1 , x i+1 , x i+2 , corresponding to a new observation sequence y i , y i ′, y i+1 , y i+1 ′, y i+1 . It is known that,

[0100]

[0101] Let the ratio between every two adjacent ones in the new observation sequence be β, that is

[0102] y i = β · y i ′ = β 2 · y i+1 = β 3 · y i+1 ′ = β 4 · y i+2

[0103] yi+1 = β·y i+1 ′ = β 2 ·y i+2

[0104] Therefore,

[0105]

[0106] Thus, we obtain

[0107]

[0108] 5) Square root expansion of the difference ratio

[0109] a. Left - hand expansion sp0

[0110] If the number of terms in the difference - ratio sequence sp ≥ 3, select the left - most three terms sp1 to sp3. Due to the monotonicity of the observed - value sequence y, sp1 to sp3 must also be monotonically increasing or decreasing. The left - hand expansion sp0 is

[0111] sp0 = sp1 + sp2 - sp3

[0112] If the number of terms in the difference - ratio sequence sp = 2, only the left - most two terms sp1 to sp2 can be selected. Due to the monotonicity of the observed - value sequence y, sp1 to sp2 must also be monotonically increasing or decreasing. The left - hand expansion sp0 is

[0113] sp0 = 2sp1 - sp2

[0114] b. Right - hand expansion sp n-1

[0115] If the number of terms in the difference - ratio sequence sp ≥ 3, select the right - most three terms sp n-4 ~sp n-2 , due to the monotonicity of the observed - value sequence y, sp n-4 ~sp n-2 must also be monotonically increasing or decreasing. The right - hand expansion sp n-1 is

[0116] sp n-1 = sp n-2 + sp n-3 - sp n-4

[0117] If the number of terms in the difference - ratio sequence sp = 2, only the right - most two terms sp n-3 ~sp n-2 , due to the monotonicity of the observed - value sequence y, sp n-3 ~sp n-2 must also be monotonically increasing or decreasing. The right - hand expansion sp n-1 is

[0118] sp n-1 = 2sp n-2 -sp n-3

[0119] c. Re - coding

[0120] After the above operations, the number of terms of the difference ratio sequence sp is expanded from the original n - 2 to n. The new difference ratio sequence sp is [sp0, sp1,..., sp n-1 . Keeping the order unchanged, after re - coding it becomes [sp1, sp2,..., sp n .

[0121] 6) Construct the median point sequence

[0122]

[0123] Generate the difference sequence x′ = [x1′, x2′,..., x n-1 ′].

[0124] 7) Median point interpolation prediction

[0125] d. Calculate the left interpolation of the median point

[0126]

[0127] e. Calculate the left interpolation of the median point

[0128]

[0129] f. Calculate the interpolation of the median point

[0130]

[0131] 8) Form new sampling sequences and observation value sequences

[0132] The new sampling sequences are x1, x1′, x2, x2′,..., x n-1 , x n-1 ′, x n and the new observation value sequences are y1, y1′, y2, y2′,..., y n-1 , y n-1 ′, y n , and output the two new sequences.

[0133] Expand the original n sampling points to 2n - 1.

[0134] (3) Multi - round median sampling

[0135] A total of M rounds of median sampling are carried out. In each round of median sampling, the sampling sequence and the observation value sequence newly formed after the previous round of median sampling are used as the input for the current round of median sampling. The steps are as follows:

[0136] 1) Median sampling in the j-th round;

[0137] 2) If j = 1, use the actually collected sampling sequence and observation value sequence as the input for median sampling; otherwise, use the sampling sequence and observation value sequence newly formed after the (j - 1)-th round of median sampling as the input for median sampling;

[0138] 3) Perform median sampling and output two new sampling sequences and a new observation value sequence;

[0139] 4) Whether it is the last round, that is, j = M. If not, return to 1); if so, output the final sampling sequence and the final observation value sequence, and end the multi-round median sampling.

[0140] The more the number of M rounds, the denser the final sampling sequence, and the more accurate the subsequent interpolation algorithm. For each additional round, the number of sampling points almost doubles.

[0141] (4) Median interpolation algorithm

[0142] 1) Obtain the actually collected sampling sequence and observation value sequence, and set the multi-round median sampling coefficient M;

[0143] 2) Obtain the sequence of points to be interpolated;

[0144] 3) Perform multi-round median sampling to obtain the final sampling sequence and the final observation value sequence;

[0145] 4) For all the points to be inserted in the sequence of points to be interpolated, determine the sub-intervals of the final sampling sequence and the final observation value sequence to which they belong, and then calculate the predicted values using linear interpolation;

[0146] 5) End

[0147] (5) Generation and execution of harmonic control instructions

[0148] 1) Instruction issuance

[0149] The APC system adjusts the harmonic ratio according to the predicted value and issues relevant process parameters.

[0150] 2) Online monitoring and feedback

[0151] The closed cup flash point after blending is monitored in real time through a near-infrared online analyzer, and the data is transmitted back to the MES system.

[0152] 3) Dynamic correction

[0153] If the deviation between the measured value and the predicted value exceeds the threshold (e.g., ±1°C), trigger the median insertion algorithm to regenerate the sampling sequence.

[0154] (6) Process knowledge base update

[0155] Store the actual flash point data after blending into the process knowledge base to optimize the initial sampling point selection strategy for the subsequent prediction model. The system integration process is as Figure 3 shown.

[0156] (7) Example 1 - Median sampling

[0157] Table 3 Median sampling example

[0158]

[0159] 1) The actual sampling points, observed values are the same as those in Table 1, corresponding to the first two columns of Table 3. For example, x1 = 5, x2 = 10, x3 = 15..., y1 = 13468, y2 = 8247, y3 = 5152...;

[0160] 2) Calculate the difference. For example, Δy1 = y1 - y2 = 13468 - 8247 = 5221;

[0161] 3) Calculate the difference ratio. For example

[0162]

[0163] 4) Calculate the square root of the difference ratio. For example

[0164]

[0165] The difference ratio sequence sp = [1.3, 1.29, 1.27, 1.26, 1.25, 1.24, 1.23, 1.22]

[0166] 5) Expansion of the square root of the difference ratio

[0167] a. Left - hand side expansion sp0

[0168] Select the left - most three items sp1 ~ sp3 and calculate sp0,

[0169] sp0 = sp1 + sp2 - sp3 = 1.3 + 1.28 - 1.27 = 1.31

[0170] b. Right - hand side expansion sp9

[0171] Select the right - most three items sp7 ~ sp9 and calculate sp 10 ,

[0172] sp 10 = sp8 + sp9 - sp7 = 1.21

[0173] The square root of the difference ratio extended sequence is sp = [1.31, 1.3, 1.29, 1.27, 1.26, 1.25, 1.24, 1.23, 1.22, 1.21]

[0174] 6) Construct the median point sequence [7.5, 12.5,..., 47.5]

[0175] 7) Median point interpolation prediction

[0176] a. Calculate the left interpolation of the median point, such as ly1′

[0177]

[0178] b. Calculate the right interpolation of the median point, such as ry1′

[0179]

[0180] Therefore, the median point interpolation y1′ = (10506 + 10518) / 2 = 10512

[0181] 8) Form new sampling sequence and observation value sequence

[0182] The new sampling sequence is [5, 7.5, 10, 12.5,..., 45, 47.5, 50], and the new observation value sequence is [13468, 10512, 8247, 6502,..., 438, 366, 306], and output the two new sequences.

[0183] Expand the original 10 sampling points to 19.

[0184] (8) Example 2 - Median insertion algorithm

[0185] 1) A total of M = 3 rounds of median sampling are performed, as shown in Table 4.

[0186] For example, in the first round of median sampling, a new sampling point 7.5 is generated between the original sampling points 5 and 10; in the second round of median sampling, a new sampling point 6.25 is generated between the sampling points 5 and 7.5; in the third round of median sampling, a new sampling point 5.625 is generated between the sampling points 5 and 6.25.

[0187] There are originally 10 sampling points. After the first round, the number of sampling points is 10×2 - 1 = 19. After the second round, the number of sampling points is 19×2 - 1 = 37. After the third round, the number of sampling points is 37×2 - 1 = 73.

[0188] Table 4 Multi-round median sampling

[0189] Sampling point 5 5.625 6.25 6.875 7.5 8.125 8.75 9.375 10 10.625 Observed value 13468.31 12653.25 11891.33 11178.85 10512.41 9888.87 9305.26 8758.88 8247.15 7767.68 Sampling point 11.25 11.875 12.5 13.125 13.75 14.375 15 15.625 16.25 16.875 Observed value 7318.36 6897.17 6502.25 6131.87 5784.36 5458.20 5151.92 4864.15 4593.77 4339.67 Sampling point 17.5 18.125 18.75 19.375 20 20.625 21.25 21.875 22.5 23.125 Observed value 4100.83 3876.29 3665.10 3466.40 3279.37 3103.22 2937.33 2781.07 2633.85 2495.11 Sampling point 23.75 24.375 25 25.625 26.25 26.875 27.5 28.125 28.75 29.375 Observed value 2364.33 2241.00 2124.65 2014.83 1911.17 1813.32 1720.92 1633.65 1551.21 1473.30 Sampling point 30 30.625 31.25 31.875 32.5 33.125 33.75 34.375 35 35.625 Observed value 1399.65 1329.98 1264.09 1201.75 1142.76 1086.94 1034.10 984.05 936.65 891.72 Sampling point 36.25 36.875 37.5 38.125 38.75 39.375 40 40.625 41.25 41.875 Observed value 849.14 808.78 770.52 734.24 699.82 667.17 636.18 606.76 578.82 552.29 Sampling point 42.5 43.125 43.75 44.375 45 45.625 46.25 46.875 47.5 48.125 Observed value 527.09 503.14 480.39 458.77 438.21 418.66 400.06 382.37 365.53 349.51 Sampling point 48.75 49.375 50 Observed value 334.25 319.72 305.88

[0190] 2) According to the 10 groups of observed values actually sampled, the linear interpolation algorithm and the median insertion algorithm are respectively used for predicting the observed values, and the results are shown in Table 5.

[0191] For example, for the first actually sampled point 11.05, which belongs to the interval [10.625, 11.25] in Table 4 and corresponds to the observed values [7767.68, 7318.36] respectively. Therefore, within this interval, the linear interpolation algorithm is applied to calculate the predicted value of 11.05, that is

[0192]

[0193] At the same time, the absolute errors between the prediction results of the two algorithms and the actual observed values are calculated. The absolute error of the median insertion algorithm is much lower than that of the linear interpolation algorithm. Further calculate the mean absolute errors of the two algorithms. The mean absolute error of the median insertion algorithm is 1.08, and the mean absolute error of the linear interpolation algorithm is 76.19. Therefore, the absolute error of the median insertion algorithm is only 1.08 / 76.19 = 1.42% of that of the linear interpolation algorithm.

[0194] Comparison between the median insertion algorithm and the linear interpolation algorithm in Table 5

[0195] Actual sampling 11.05 43.13 39.37 16.48 27.29 25.23 34.32 40.49 9.22 6.28 Actual observed value 7461.76 502.80 667.44 4499.33 1750.76 2084.01 988.23 612.87 8887.79 11860.80 Median insertion algorithm predicted observed value 7464.74 502.80 667.44 4501.03 1751.29 2084.75 988.32 612.99 8891.16 11862.11 Linear interpolation algorithm predicted observed value 7599.39 512.07 674.05 4598.35 1791.93 2091.72 999.46 616.68 9057.81 12136.25 Median insertion algorithm absolute error 2.98 0.00 0.00 1.70 0.52 0.73 0.10 0.12 3.37 1.31 Linear interpolation algorithm absolute error 137.63 9.27 6.61 99.03 41.17 7.71 11.24 3.81 170.01 275.45

[0196] The specific embodiments described in this article are only illustrative of the spirit of the present invention. Those skilled in the art to which the present invention pertains can make various modifications or supplements to the described specific embodiments or use similar ways to replace them, but will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A description method for key physical properties of oil refining based on a non-linear multi-round median insertion algorithm, characterized in that It includes the following steps: S1. Initial experimental data collection of key refinery physical property parameters and system input; S2. Predict key refinery physical property parameters using the median insertion algorithm, specifically including: S2-1. Obtain the actually collected sampling sequence and observation value sequence, and set the multi-round median sampling coefficient M; S2-2. Obtain the sequence of points to be interpolated; S2-3. Conduct multi-round median sampling to obtain the final sampling sequence and final observation value sequence; S2-4. For all points to be inserted in the sequence of points to be interpolated, determine the sub-intervals of the final sampling sequence and final observation value sequence to which they belong, and then calculate the predicted values using linear interpolation.

2. The method according to claim 1, wherein In step S1, it specifically includes: 1) Sample collection. In the blending laboratory, test the key refinery physical property parameters at equal interval steps to generate a sampling sequence; 2) Sample storage. Store the sampling sequence in the process database of the Manufacturing Execution System (MES) for algorithm call; 3) Data preprocessing. Verify the monotonicity of the sampling data. If there are abnormal points, re-conduct the experiment and supplement the measured data.

3. The method according to claim 1, wherein In step S2-3, the specific steps of median sampling include: 1) Determine the sampling points. Let the equidistant sampling point sequence be x1, x2,..., x n , and the corresponding observed value sequence be y1, y2,..., y n ; where n ≥ 4; 2) Calculate the adjacent difference Δy of the observed values i : Δy i = y i - y i+1 , i = 1, 2, ..., n - 1 Generate a difference sequence Δy = [Δy1, Δy2,..., Δy n-1 ; 3) Calculate the difference ratio Δp i : Generate a difference ratio sequence Δp = [Δp1, Δp2,..., Δp n-1 ; 4) Calculate the square root of the difference ratio sp i : Generate a difference ratio sequence sp = [sp1, sp2,..., sp n-2 ; 5) Square root expansion of the difference ratio a. Left-side expansion sp0 If the number of items in the difference ratio sequence sp ≥ 3, select the leftmost three items sp1 to sp3. Due to the monotonicity of the observation value sequence y, sp1 to sp3 must also be monotonically increasing or decreasing. The left-side expansion sp0 is: sp0 = sp1 + sp2 - sp3 If the number of items in the difference ratio sequence sp = 2, only the leftmost two items sp1 to sp2 can be selected. Due to the monotonicity of the observation value sequence y, sp1 to sp2 must also be monotonically increasing or decreasing. The left-side expansion sp0 is: sp0 = 2sp1 - sp2 b. Right-expanded sp n-1 If the number of terms of the difference ratio sequence sp ≥ 3, select the rightmost three terms of sp n-4 ~sp n-2 , due to the monotonicity of the observed value sequence y, sp n-4 ~sp n-2 must also be monotonically increasing or decreasing, and expand sp on the right n-1 to be: sp n-1 = sp n-2 + sp n-3 - sp n-4 If the number of terms in the difference ratio sequence sp = 2, only the rightmost two terms of sp can be selected n-3 ~sp n-2 , due to the monotonicity of the observed value sequence y, sp n-3 ~sp n-2 must also be monotonically increasing or decreasing. The right side of sp is extended n-1 as follows: sp n-1 = 2sp n-2 -sp n-3 c. Re-encoding After the above operations, the number of terms of the difference ratio sequence sp is expanded from the original n - 2 to n; the new difference ratio sequence sp is [sp0, sp1,..., sp n-1 ; keeping the order unchanged, after re - coding it becomes [sp1, sp2,..., sp n ; 6) Construct the median point sequence Generate the difference sequence x′ = [x1′, x2′,..., x n-1 ′]; 7) Median point interpolation prediction a. Calculate the left interpolation ly of the median point i ′ : b. Calculate the left interpolation ry of the median point i ′ : c. Calculate the median point interpolation y i ′: 8) Form a new sampling sequence and observation value sequence The new sampling sequence is x1, x1′, x2, x2′, ..., x n-1 , x n-1 ′, x n , the new observation value sequence is y1, y1′, y2, y2′, ..., y n-1 , y n-1 ′, y n , and output two new sequences.

4. The method according to claim 3, wherein A total of M rounds of median sampling are performed. In each round of median sampling, the newly formed sampling sequence and observation value sequence after the previous round of median sampling are used as the input for the current round of median sampling.

5. The method according to claim 1, characterized in that It also includes the following steps: S3. The APC system adjusts the blending ratio according to the predicted value and issues relevant process parameters; at the same time, it monitors the data of key refinery physical property parameters in real time and transmits them back to the MES system; S4. Update the process knowledge base. Store the monitored data of the actually blended key refinery physical property parameters in the process knowledge base to optimize the initial sampling point selection strategy of the subsequent prediction model.

6. The method according to claim 5, wherein If the deviation between the measured data of the key refinery physical property parameters and the predicted value exceeds the threshold, trigger the median insertion algorithm to regenerate the sampling sequence.

Images