Soft measurement method for key quality index of fractionation product

Abstract

The invention discloses a soft measurement method for a key quality index of a fractionation product, and the method comprises the steps: selecting an auxiliary variable for prediction model training through calculating a Spearman correlation coefficient between a feature variable collected by a sensor and the key quality index; calculating a delay time relationship between the auxiliary variable and the key quality index by combining a mutual information method and a differential evolution algorithm, reconstructing the auxiliary variable according to the delay time, and finally establishing a regression model by adopting a random forest algorithm to realize real-time prediction of the key quality index of the fractionation product. According to the method, the characteristic variables for soft measurement can be accurately selected, and the prediction precision is further improved by analyzing the delay time of different variables to the key quality parameters, so that an efficient and reliable product key quality index monitoring scheme is provided for the refining industry, and the refinement and intelligence level of the production process is improved.

Description

Technical Field

[0002] This invention relates to the field of predicting the quality indicators of fractionated products in refining and chemical enterprises, specifically a soft measurement method for key quality indicators of fractionated products. Background Technology

[0004] Fractionation is a typical production process in refining and chemical enterprises, and the quality indicators of its products require strict control. Currently, the key quality indicators of fractionated products mainly rely on manual, timed sampling and sending them to the laboratory for analysis. This method is time-consuming, resulting in poor real-time quality monitoring. By the time product defects are discovered, significant economic losses may have already occurred.

[0005] With the development of the Industrial Internet of Things (IIoT) and intelligent manufacturing technologies, refining and chemical enterprises have deployed a large number of sensors for temperature, pressure, flow, and liquid level in their production facilities, enabling real-time sampling of process data. Analyzing the intrinsic correlation between process data and final product quality indicators to construct a soft measurement prediction model for key quality indicators has become an important trend in recent years.

[0006] However, appropriately selecting process data as feature variables to build the model is crucial. Clearly, insufficient variables will fail to accurately capture changes in key quality parameters; conversely, selecting too many variables can easily introduce noise interference, negatively impacting the model's predictive accuracy. Furthermore, even with appropriate feature variable selection, the influence of different variables on key quality parameters has a time dependence, which must be considered during modeling.

[0007] Based on the above, this invention proposes a soft measurement method for key quality indicators of fractionation products. First, sensor data is collected as feature variables. Then, the Spearman correlation coefficient between the feature variables and the key quality indicators is calculated to select auxiliary variables for training the prediction model. Next, the mutual information method and differential evolution algorithm are combined to calculate the time delay relationship between the auxiliary variables and the key quality indicators. Based on the time delay, the auxiliary variables are reconstructed. Finally, a random forest algorithm is used to establish a regression model, enabling real-time prediction of key quality indicators of fractionation products. Summary of the Invention

[0009] This invention addresses the problems existing in the background technology by proposing a soft measurement method for key quality indicators of fractionation products, enabling real-time prediction of key quality indicators of fractionation products. The method specifically includes the following steps:

[0010] Step 1: Read the laboratory test quality index data as the target variable Y, and the historical operating data of the sensors in the synchronous fractionation unit distributed control system as the feature variable X, and detect and remove abnormal samples.

[0011] Step 2: Calculate the Spearman coefficients between each feature variable X and the target variable Y, and select the K variables with the largest absolute values ​​of correlation coefficients as auxiliary variables. The Spearman coefficients are calculated as follows:

[0012]

[0013] Where cov represents covariance. Represents variance. and It is the order of X and Y.

[0014] Step 3: Based on the auxiliary variables selected in Step 2), with the optimization objective of maximizing the average mutual information (AMI) between the auxiliary variables and the target variable, the differential evolution (DE) algorithm is used to globally search for the optimal set of delay time parameters for each auxiliary variable and the target variable. :

[0015] 1) Without considering the delay interval, the method for calculating the average mutual information (AMI) of all auxiliary variables and the target variable is as follows:

[0016]

[0017]

[0018] in, For the mutual information between the i-th auxiliary variable and the target variable Y, and yes and The probability density function, It is a variable and The joint probability density function, where t is the sampling time of the sample. This represents the number of auxiliary variables.

[0019] 2) The optimization problem of the Differential Evolution (DE) algorithm is described as follows:

[0020]

[0021] in, Let i be the delay interval of the i-th auxiliary variable. This represents the maximum delay interval for all auxiliary variables.

[0022] Step 4: Based on the delay time parameter set obtained in Step 3), reconstruct the auxiliary variable input matrix, generate the dataset, and divide the training set. and test set ;

[0023] Step 5: Perform data standardization on each auxiliary variable and the target variable using the Z-Score method. The Z-Score data standardization method is as follows:

[0024]

[0025] in, The original values ​​of the data. The value after standardization. The mean of the sample data. This represents the variance of the sample data.

[0026] Step 6: A prediction model is built using random forest regression on the standardized data. Multiple training subsets are generated through bootstrap sampling, and base decision trees are trained for each subset. The weighted mean square error (MSE) is minimized as the splitting criterion to learn the nonlinear mapping relationship between auxiliary variables and target variables. The weighted mean square error (MSE) of the random forest decision tree during splitting is calculated as follows:

[0027]

[0028] Where S is the population set of the sample, and m is the total number of samples. This indicates that the decision tree splits based on the i-th feature. The set of samples contained in the left child node The number of samples, The set of samples contained in the right child node The number of samples, These represent the true value and the predicted value for a single sample, respectively.

[0029] Step 7, using the test set Evaluate the model's generalization ability by determining whether the mean absolute percentage error (MAPE) on the test set meets the required requirements. If it does, proceed to step 8); otherwise, modify the hyperparameters and proceed to step 6). The formula for calculating the mean absolute percentage error (MAPE) is as follows:

[0030]

[0031] Where n is the total number of samples in the test set. Actual value Corresponding predicted value.

[0032] Beneficial effects:

[0033] This invention discloses a soft measurement method for key quality indicators of fractionation products. It selects auxiliary variables for training the prediction model by calculating the Spearman coefficient, and then combines mutual information and differential evolution algorithms to calculate and reconstruct the time lag relationship between the auxiliary variables and the key quality indicators. This method can accurately select characteristic variables for soft measurement and further improve prediction accuracy by analyzing the time lag of different variables on key quality parameters. Therefore, it provides the refining and chemical industry with an efficient and reliable solution for monitoring key product quality indicators, improving the refinement and intelligence of the production process. Attached Figure Description

[0035] Figure 1 This is a flowchart of a soft measurement method for key quality indicators of fractionation products according to the present invention.

[0036] Figure 2 The curve shows the relationship between the gasoline partial pressure at the top of the fractionation tower and the normalized final distillation point of the crude gasoline product.

[0037] Figure 3 The maximum average mutual information change is used to optimize the differential evolution algorithm.

[0038] Figure 4 This is a comparison curve of the true and predicted values ​​on a portion of the test set. Detailed Implementation

[0040] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0041] This case study uses actual operating data of the fractionation unit collected by the distributed control system of an oil refinery from December 2022 to February 2025, along with manually sampled crude gasoline final distillation point test data, to verify the effectiveness of the soft measurement method for key quality indicators of fractionation products proposed in this invention.

[0042] The overall system flowchart of the present invention is as follows: Figure 1 As shown, the specific steps include:

[0043] 1) Laboratory test quality index data were used as the target variable, and historical operating data from sensors in the synchronous fractionation unit's distributed control system were used as characteristic variables. The criteria were used to detect and remove abnormal samples, resulting in 2065 analytical values ​​for the crude gasoline final distillation point and 95 characteristic variables, including the sensor sampling interval for the characteristic variable. .

[0044] 2) Calculate the Spearman coefficients between each feature variable and the target variable, and select the top K=20 variables with the largest absolute values ​​of correlation coefficients as auxiliary variables. Table 1 shows some of the selected auxiliary variables and their Spearman coefficients. Figure 2 The curve showing the relationship between the gasoline partial pressure at the top of the fractionation tower and the normalized change of the crude gasoline product final distillation point, which is one of the screening characteristics.

[0045] Table 1. Some auxiliary variables and Spearman coefficients

[0046] Auxiliary variable name Spearman coefficient Cold reflux to fractionation column flow rate 0.4304 Differential pressure of fractionation tower 0.5267 Top circulating oil return tower temperature 0.3736 The extraction temperature of intermediate oil from the fractionation tower -0.4919 Absorbent oil return temperature 0.3892 The return temperature of the second fractionated oil to the tower 0.3918 Boundary gas pressure regulation 0.4092 Reaction pressure 0.4832 Oil and gas temperature at the top outlet of the fractionation tower 0.5322 Gasoline partial pressure at the top of the fractionation tower 0.4997 ...... ......

[0047] 3) Using the selected auxiliary variables, with the optimization objective of maximizing the average mutual information (AMI) between the auxiliary variables and the target variable, the differential evolution (DE) algorithm is used to globally search for the optimal set of delay time parameters for each auxiliary variable and the target variable. .

[0048] Without considering the delay interval, the method for calculating the average mutual information (AMI) of all auxiliary variables and the target variable is as follows:

[0049]

[0050]

[0051] in, For the mutual information between the i-th auxiliary variable and the target variable Y, and yes and The probability density function, It is a variable and The joint probability density function, where t is the sampling time of the sample. This represents the number of auxiliary variables.

[0052] The optimization problem of the Differential Evolution (DE) algorithm is described as follows:

[0053]

[0054] in, Let i be the delay interval of the i-th auxiliary variable. This represents the maximum delay interval for all auxiliary variables.

[0055] Based on the actual conditions of this fractionation device, the maximum delay of the differential evolution algorithm is set to 13 (sampling time is 5 minutes, and the maximum delay time does not exceed 1 hour), the population size is 50, the scaling factor is 0.6, the crossover probability is 0.7, and the maximum number of iterations is 300.

[0056] The convergence of the maximum average mutual information in the differential evolution algorithm is as follows: Figure 3 As shown, the optimal average mutual information (AMI) is 0.73, and the optimal delay parameter set is D = [8,5,5,1,3,1,0,0,2,1,8,2,0,1,3,0,5,0,2,0]. Multiplying these values ​​by the sampling time yields the optimal delay time. Table 2 shows the optimal delay times for some auxiliary variables.

[0057] Table 2 Optimal Delay Time for Some Auxiliary Variables

[0058] Auxiliary variable name Optimal delay time / min Cold reflux to fractionation column flow rate 40 Differential pressure of fractionation tower 25 Top circulating oil return tower temperature 25 The extraction temperature of intermediate oil from the fractionation tower 5 Absorbent oil return temperature 15 The return temperature of the second fractionated oil to the tower 5 Boundary gas pressure regulation 0 Reaction pressure 0 Oil and gas temperature at the top outlet of the fractionation tower 10 Gasoline partial pressure at the top of the fractionation tower 5 ...... ......

[0059] By reconstructing the auxiliary variable input matrix using the optimal delay time parameter set, a dataset X is generated, and the training set is divided in an 8:2 ratio. and test set .

[0060] 4) Perform data standardization using the Z-Score method on each auxiliary variable and the target variable. The Z-Score data standardization method is as follows:

[0061]

[0062] in, The original values ​​of the data. The value after standardization. The mean of the sample data. This represents the variance of the sample data.

[0063] 5) A prediction model is built using random forest regression on standardized data. Multiple training subsets are generated through bootstrap sampling, and base decision trees are trained for each subset. The weighted mean squared error (MSE) is minimized as the splitting criterion to learn the nonlinear mapping relationship between auxiliary variables and target variables. The weighted mean squared error (MSE) of the random forest decision tree during splitting is calculated as follows:

[0064]

[0065] Where S is the population set of the sample, and m is the total number of samples. This indicates that the decision tree splits based on the i-th feature. The set of samples contained in the left child node The number of samples, The set of samples contained in the right child node The number of samples, These represent the predicted and actual values ​​for a single sample, respectively.

[0066] Before training, the number of hyperparameter decision trees for the random number model was set to 100. The maximum number of features considered when splitting each tree was the square root of the total number of features. The maximum tree depth was 15. Minimizing the weighted mean square error (MSE) was used as the splitting criterion. Bootstrap sampling was also set.

[0067] The final root mean square error (RMSE) on the test set was 0.52, and the model's coefficient of determination was... =0.91, and the mean absolute percentage error (MAPE) is 0.19%, which fully meets the requirements for prediction accuracy. Figure 4 This is a comparison curve of the true and predicted values ​​on a portion of the test set.

[0068] As can be seen, this method, through the soft measurement model of key quality indicators of fractionation products established by the process data of sensors in the production unit, has successfully achieved accurate prediction of the final distillation point of crude gasoline products, providing the refining and chemical industry with an efficient and reliable solution for monitoring key product quality indicators.

Claims

1. A soft measurement method for key quality indicators of fractionation products, characterized in that, First, the Spearman coefficient between the feature variables and the target variable is calculated, and the top K most strongly correlated variables are selected as auxiliary variables. Then, the lag time between the feature auxiliary variables and the target variable is calculated using the mutual information method and the differential evolution algorithm. The auxiliary variables are then reconstructed based on the lag time. Finally, a regression model is established using the random forest algorithm to achieve real-time prediction of key quality indicators of fractionated products. The specific steps include the following: 1) Read the laboratory test quality index data as the target variable Y, and the historical operation data of the sensors in the synchronous fractionation unit distributed control system as the feature variable X, and detect and remove abnormal samples. 2) Calculate the Spearman coefficient between each feature variable X and the target variable Y, and select the top K feature variables with the largest absolute values ​​of the correlation coefficients as auxiliary variables; 3) Based on the auxiliary variables selected in step 2), with the optimization objective of maximizing the average mutual information (AMI) between the auxiliary variables and the target variable, the differential evolution (DE) algorithm is used to globally search for the optimal set of delay time parameters for each auxiliary variable and the target variable. ; 4) Based on the optimal delay time parameter set obtained in step 3), reconstruct the auxiliary variable input matrix, generate the dataset, and divide it into a training set. and test set ; 5) Standardize the data using the Z-Score method for each auxiliary variable and the target variable; 6) A prediction model is built using random forest regression on standardized data. Multiple training subsets are generated by bootstrap sampling, and base decision trees are trained separately. The minimum weighted mean square error (MSE) is used as the splitting criterion to learn the nonlinear mapping relationship between auxiliary variables and target variables. 7) Through the test set Evaluate the model's generalization ability and determine whether the model's mean absolute percentage error (MAPE) on the test set meets the requirements. If it does, the soft measurement model is complete; otherwise, modify the hyperparameters and proceed to step 6.

2. The soft measurement method for key quality indicators of fractionation products according to claim 1, characterized in that, Spearman coefficient The calculation method is as follows: Where cov represents covariance. Represents variance. and It is the order of X and Y.

3. The soft measurement method for key quality indicators of fractionation products according to claim 1, characterized in that, Without considering the delay interval, the method for calculating the average mutual information (AMI) of all auxiliary variables and the target variable is as follows: in, For the i-th auxiliary variable Mutual information with the target variable Y, and yes and The probability density function, It is a variable and The joint probability density function, where t is the sampling time of the sample. This represents the number of auxiliary variables.

4. The soft measurement method for key quality indicators of fractionation products according to claim 1, characterized in that, The optimization problem of the Differential Evolution (DE) algorithm is described as follows: in, Let i be the delay interval of the i-th auxiliary variable. This represents the maximum delay interval for all auxiliary variables.

5. The soft measurement method for key quality indicators of fractionation products according to claim 1, characterized in that, The Z-Score data standardization method is as follows: in, The original values ​​of the data. The value after standardization. The mean of the sample data. This represents the variance of the sample data.

6. The soft measurement method for key quality indicators of fractionation products according to claim 1, characterized in that, The weighted mean squared error (MSE) of the decision tree in a random forest during splitting is calculated as follows: in, This represents the loss value when the decision tree splits the population set S based on the i-th feature, where S is the population set and m is the total number of samples. This indicates that the decision tree splits based on the i-th feature. The set of samples contained in the left child node The number of samples, The set of samples contained in the right child node The number of samples, , These represent the true value and the predicted value for a single sample, respectively.

7. The soft measurement method for key quality indicators of fractionation products according to claim 1, characterized in that, The formula for calculating the Mean Absolute Percentage Error (MAPE) is as follows: Where n is the total number of samples in the test set. For the true value The corresponding predicted value.

Images