Isopropylbenzene refining process

[0047]The embodiments of the present invention will be further described below in conjunction with the drawings. In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. The indicated orientation or positional relationship is based on the orientation or positional relationship shown in the drawings, which is only a simplified description for the convenience of describing the present invention, and does not indicate or imply that the pointed device or component must have a specific orientation or a specific orientation. The structure and operation cannot therefore be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance. Among them, the terms "first position" and "second position" are two different positions.

[0048]In the description of the present invention, it should be noted that the terms "installation", "connected" and "connected" should be understood in a broad sense, unless otherwise clearly specified and limited. For example, they can be fixed or detachable. Connected or integrally connected; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal communication between two components. For those of ordinary skill in the art, the specific meanings of the above-mentioned terms in the present invention can be understood in specific situations. The various technical solutions in the following embodiments provided in this application, unless they contradict each other, different technical solutions can be combined with each other, and the technical features therein can be replaced with each other.

[0049]Some embodiments of the present invention provide a process flow for refining cumene products, such asfigure 1 withfigure 2 As shown, the following steps can be included:

[0050](1) The product W1 from the alkylation reactor and the product W2 from the alkylation transfer reactor are transported to the raw material mixer 1 for mixing to obtain a mixed feed;

[0051](2) The mixed feed is transported to the high pressure deweighting tower 2, the top extract of the high pressure deweighting tower 2 is transported to the light removal tower 3, and the bottom extract of the high pressure deweighting tower 2 is transported to the low pressure deweighting tower. Heavy tower 4;

[0052](3) The output from the bottom of the light removal tower 3 is a cumene product and is sent to the product tank 5;

[0053](4) The bottom output of the low-pressure weight removal tower 4 is heavy components including polycumene, and the top product of the low-pressure weight removal tower 4 is a cumene product and is sent to the product tank 5.

[0054]Such asfigure 2 As shown, the high-pressure deweighting tower 2 is equipped with a high-pressure de-weighting tower start-up condenser 9 and a high-pressure de-weighting tower start-up reboiler 14, and the low-pressure deweighting tower 4 is equipped with a low-pressure de-weighting tower start-up condenser 11 and a low-pressure de-weighting tower start-up condenser 11 The reboiler 10 and the light removal tower 3 are equipped with a dehydrogenation tower start-up condenser 13 and a dehydrogenation tower start-up reboiler 12. The working principle and start-up timing of the above-mentioned condenser and reboiler can be determined by the existing process requirements. Let me introduce it in detail.

[0055]In the above scheme, the product from the alkylation reactor is mixed with the product from the alkylation transfer reactor and then enters the three-column rectification separation system. The high- and low-pressure de-weighting column is used to achieve the separation of the product cumene and the heavy components. Separation: The product cumene is separated from the light components through the light removal tower, and the product enters the tank area. The deweighting tower is divided into a high pressure tower and a low pressure tower. The separation effect of the product and the heavy component is better through low pressure rectification. The three-tower rectification separation can improve the purity of the cumene product, and by splitting the deweighting tower The two towers are high and low pressure, which can conveniently carry out double-effect rectification heat exchange and save the consumption of public works.

[0056]Further combinefigure 2 withimage 3 As shown, the above process may also include the following steps:

[0057](5) The cumene is delivered to the light removal tower 3 and the low pressure weight removal tower 4 through the cumene tank 6, and the delivered cumene is high-purity fresh cumene. In this scheme, the material volume of the fresh cumene entering the light removal tower 3 and the low pressure weight removal tower 4 and the parameters of the light removal tower 3 and the low pressure weight removal tower 4 are adjusted to ensure the high purity of the final product.

[0058]Further, combinefigure 2 withFigure 4 The above process can also include the following steps: (6) The online adjustment control model 7 is used to adjust and control the process parameters in real time. The online adjustment control model 7 can receive the input parameter S1, and finally output the adjustment value S2 of each process parameter. The line adjustment control model 7 Through real-time monitoring and input of raw material parameters, high-pressure de-weighting tower parameters to the control model system, according to the model algorithm, the corresponding low-pressure de-weighting tower 4, light-removing tower 3 operating parameter adjustment and fresh cumene feed distribution adjustment instructions are obtained. When the material parameters and process operation parameters change in real time, the process parameters can be adjusted according to the real-time changes. Through online adjustment, the quality and purity of the product can be guaranteed.

[0059]Step (6) can be passed asFigure 4 The implementation of the process shown includes:

[0060](6.1) Collect the mass fraction x of benzene in the mixed feed described in step (1)1, The mass fraction of cumene x2, The mass fraction of polycumene x3, Temperature T1, Pressure P1And flow F1As an input parameter of online adjustment control model 7.

[0061](6.2) Acquisition step (2) The top temperature T of the medium and high pressure weight removal tower 22And tower top pressure P2As an input parameter of online adjustment control model 7.

[0062](6.3) Collect the cumene mass fraction Y in the cumene product obtained in the light-removing tower 3 in step (3)1As the target value of the online adjustment control model 7, the cumene mass fraction Y in the cumene product obtained in the low-pressure deweighting tower 4 in step (4)2And the cumene mass fraction Y in the cumene product obtained in the product tank 5 is used as the target value of the online adjustment control model 7.

[0063](6.4) According to the input parameters and target values ​​obtained in the steps (6.1)-(6.3), compare the top pressure P of the light removal tower 3 in step (3)3, Reflux ratio R1, The top pressure P of the low-pressure weight removal tower 4 in step (4)4And reflux ratio R2And the flow rate F of cumene into the light-removing tower in step (5)2And cumene into the low pressure deweighting tower 4 flow F3Adjust so that the mass fraction Y of cumene in step (6.3) reaches the set target value.

[0064]The above scheme, such asFigure 5 As shown, the online adjustment control model in the step (6) includes a prediction model 71 and an optimization adjustment model 72. The former is a mathematical model that takes monitoring variables + control variables as input values ​​and product purity prediction values ​​as output values. The main function is to collect and analyze real-time monitoring values ​​and provide product purity prediction values ​​to the online optimization and adjustment model controller. The latter is a mathematical model in which the product purity prediction value is the input value, and each control variable is the output value. By forming an equipotential line, the product purity and each control variable are displayed in dimensionality reduction, in the reduced dimensionality line model , Find the shortest path according to the deviation between the predicted value and the set value, and then decompose it into specific control variable values ​​to each sub-controller. After completing the parameter change, feedback the actual product purity value as the predicted value for judgment, and then Realize closed-loop optimization.

[0065]In the above solution, the prediction model is obtained in the following way:

[0066](6.01) Extract historical process parameters from the production history data as training samples. The historical process parameters include: the mass fraction of benzene in the mixed feed, the mass fraction of cumene, the mass fraction of polyisopropylbenzene, the pressure, and the flow rate; The top temperature and pressure of the weight removal tower; the top pressure and reflux ratio of the light removal tower; the top pressure and reflux ratio of the low pressure weight removal tower; the flow rate of cumene into the light removal tower and the low pressure of cumene The flow rate of the weight removal tower; and the mass fraction of cumene in the cumene product obtained in the light removal tower under the aforementioned historical process parameters, and the mass fraction of cumene in the cumene product obtained in the low pressure weight removal tower; The quality score of cumene in the cumene product in the product tank; in this step, after mining and cleaning the production history data, the fuzzy clustering algorithm is further applied for data recognition, and finally a machine learning model such as (ANN artificial neural Network model) Construct a mathematical prediction model of the controller to realize the prediction of the target value after online data collection.

[0067]The above data mining and cleaning refers to preprocessing by filling in missing data and eliminating noise data, and then using data standardization to clean up data that does not meet the requirements of the specification. In this example, the data comes from the device's online database, and the data obtained is between 2018-2020 Running data, a total of 10,000 sets of data are standardized, of which 7,500 sets are used as neural network training sets, 2000 sets are used as neural network simulation sets, and the last 500 sets are used for model prediction sets for verification and debugging. Before constructing and debugging the neural network model, use the following formula to give priority to fuzzy clustering processing in order to improve the recognition speed and prediction accuracy of the neural network model. The data has a boundary and a practical difference of ± signs, so the standardized algorithm of this scheme is designed as follows:

[0068]

[0069]In the formula, maxA, MinA, New_maxA, New_minAThey are the maximum and minimum values ​​of the original and transformed attribute values ​​respectively; x is the attribute value. The described fuzzy clustering algorithm, considering that the production data collected in the process flow often corresponds to different production load conditions, it has obvious block clustering characteristics, but at the same time there is relevance, so the fuzzy clustering process can make data recognition more To be accurate and realistic. The design of the fuzzy clustering algorithm in this scheme takes into account the concentration and similarity of the process data, and uses the K-means algorithm as the benchmark to apply fuzzy clustering technology. The algorithm takes n vectors xi, I=1,2,...,n, divide into K fuzzy sets, and find the cluster center of each set to minimize the objective function:

[0070]

[0071]Where:

[0072]

[0073]m is the fuzzy weight index, 1 iAnd membership degree uik The calculation is as follows:

[0074]

[0075]

[0076]Through the above algorithm processing, data preprocessing and mining cleaning can be realized, and the processed data can be used as sample values ​​for machine learning model training, simulation and prediction. The machine learning model can choose the ANN artificial neural network model. The ANN artificial neural network model refers to a mathematical model composed of topological structure, neuron transfer function, learning algorithm, etc., through learning, simulation and prediction of a large amount of effective data after processing, Realize the strong correlation between input variables and output variables, that is, real-time analysis of online monitoring input data and output of predicted values. among them:

[0077]The input and output of each node in the hidden layer are:

[0078] (Number of hidden layer units);

[0079]The input and output of each node in the input layer are:

[0080] (Number of neurons in the output layer);

[0081]After that, the construction and debugging of the neural network model began. The interface graphics of the debugging process are asFigure 6 Shown. After training and learning 7500 sets of data, the neural network model has the initial analysis capability. Then, 2000 sets of data are used for simulation training to debug and improve the model's function parameters, model parameter factors, etc., so that the model's anti-interference degree, Anti-under-fitting/over-fitting ability and prediction ability meet the requirements of use, and then use 500 sets of data for prediction verification, such asFigure 7 As shown,Figure 7 In the results shown, the predicted value and actual value overlap. It can be determined that the prediction model constructed by this program has good accuracy. It is used as the mathematical model of online monitoring analysis and prediction model controller, and new data will be added in the later period to make it more accurate. perfect.

[0082](6.02) The machine learning model is trained using the training samples, and the machine learning model after the training is completed is used as the prediction model 71. Based on the above analysis, the obtained prediction model can be expressed by the following formula:

[0083]

[0084]In the above formula, A, a, b, c, and d are all parameter factors, and Y'is the predicted value of the cumene mass fraction of the cumene product in the product tank.

[0085]Further, the above step (6.4) specifically includes:

[0086](6.41) When the input parameters obtained in the steps (6.1)-(6.3) are used as the input of the prediction model 71, the prediction model 71 outputs the prediction value of the cumene mass fraction of the cumene product in the product tank ; The input parameters obtained in steps (6.1)-(6.3) can be detected by the online meter 16.

[0087](6.42) If the prediction value of the cumene quality score output by the prediction model 71 does not reach the set target value, the prediction value of the cumene quality score is transmitted to the optimization adjustment model 72; if the prediction value of the cumene quality score output by the prediction model 71 is predicted When the value reaches the set target value, it ends.

[0088](6.43) The optimization adjustment model 72 determines the adjustment value of the top pressure of the light removal tower according to the difference between the predicted value of the cumene mass fraction and the set target value, the adjustment value of the reflux ratio of the light removal tower, and the cumene The adjustment value of the flow rate of the inlet and outlet light tower, the adjustment value of the pressure at the top of the low pressure weight removal tower, the adjustment value of the reflux ratio of the low pressure removal tower and the adjustment value of the flow rate of cumene into the low pressure weight removal tower.

[0089]Preferably, the optimization adjustment model 72 adjusts the cumene mass fraction value of the cumene product in the product tank, the top pressure of the light removal tower, the reflux ratio of the light removal tower, the flow rate of the cumene into the light removal tower, and the low pressure weight removal The top pressure of the tower, the reflux ratio of the low-pressure deweighting tower and the flow rate of the cumene into the low-pressure deweighting tower are reduced to obtain a reduced-dimensional equipotential model 721; the deviation of the predicted value of the cumene mass fraction from the set target value is determined The corresponding shortest path in the equipotential line model is decomposed into the adjustment value corresponding to each control parameter after being upgraded.

[0090]The principal component dimensionality reduction algorithm in this scheme is a statistical method that converts the original multiple indicator variables into a few independent comprehensive indicators. Through a comprehensive analysis of the information carried by various indicators, some potential comprehensive indicators (ie principal components) are proposed. Specifically, let X1, X2,..., XPIs the original variable, requires variable Z1,Z2,…,Zm, Satisfy m iWith ZjNot relevant, that is, the correlation coefficient between them is 0, and ZiIt can represent most of the variation information of p original variables xi, which also reduces the number of latitudes of the original variables. To X1, X2,..., XPObserved n times, the observation data matrix is:

[0091]

[0092]Use p vectors of data matrix X (that is, p index vectors) X1, X2,..., XpDo the linear combination as:

[0093]

[0094]Abbreviated as:

[0095]Zi=a1iX1+a2iX2+...+api Xp, I=1,2,...,p;

[0096]Dang Quan XiWhen it is an n-dimensional vector, ZiIt is also an n-dimensional vector. The key here is to require aij (i,j=1,2,...,p; and) Makes Var(Zi) Value is the largest, then solve the constraint equation to obtain the unit vector p, which is the principal component direction, generally 2-3 principal component descriptions can contain the required information, such asFigure 8 Shown. The said equipotential model refers to the mathematical model of independent variable → target value constructed with 2-3 principal component variables as comprehensive variables based on each parameter variable after principal component dimensionality reduction analysis, and then visualized It is an intuitive two-dimensional or three-dimensional image. Use Z1,Z2Two new dimensional variables can represent the relationship between input and output values, which greatly reduces the number of mapping dimensions between input variables and output variables, which facilitates the increase in data processing speed and the construction of visual dimensionality reduction models. In this scheme, the raw material composition, temperature, pressure and each tower operation index are reduced in dimensionality, using the principal component variable Z1,Z2Re-describe the output value, which is the concentration of cumene in the product, to obtain the equipotential model asPicture 9 Shown. The equipotential model is a collection of operating points corresponding to each output variable under different cumene concentrations or the same cumene concentration, which can clearly display the concentration and corresponding position of cumene under the historical and current operating parameters For subsequent model debugging. The predicted value and online monitoring value are transmitted to the optimal adjustment model controller, and the equipotential line model analyzes them, and locates them in the model according to their corresponding input and output values. At the same time, it takes into account that in the chemical production process, each input The adjustment of the parameter variables is not easy to fluctuate excessively and greatly adjust, so the equipotential line model determines the best adjustment strategy by finding the shortest path to the target setting value. The newly obtained target value is positioned and then upgraded and re-decomposed into Recommended adjustment values ​​for each output variable. The above can also include the following steps:

[0097](7) The adjustment value of the top pressure of the light removal tower, the adjustment value of the reflux ratio of the light removal tower, the adjustment value of the flow rate of cumene into the light removal tower, the adjustment value of the top pressure of the low pressure removal tower, and the low pressure removal tower The adjusted value of the reflux ratio and the adjusted value of the flow rate of cumene into the low-pressure deweighting tower are sent to the parameter controller, and the parameter controller adjusts the corresponding regulating valve 15 according to each adjustment value, and then returns to step (6.1).

[0098]Further preferably, refer tofigure 2 In step (2), the double-effect rectification heat exchanger 8 is used to condense the top product of the high-pressure deweighting column; at the same time, in step (4), the double-effect rectification heat exchanger 8 is used as the The bottom reboiler of the low pressure deweighting tower. In this scheme, through the double-effect rectification heat exchanger 8, the top stream of the high-pressure deweighting tower and the bottom stream of the low-pressure deweighting tower are used to exchange heat, thereby achieving high efficiency and energy saving. In specific use, when the system is running, first use public works (such as the low-pressure deweighting tower to start the reboiler 10 to ensure that the system is running, and then switch to the double-effect rectification heat exchanger 8 to replace the public works after the subsequent stability. Consumption, through the above-mentioned double-effect rectification system, the process energy consumption is reduced, and energy saving and high efficiency are reflected.

[0099]The above process will be described below in conjunction with specific examples.

[0100]Taking a cumene plant with an annual output of 530,000 tons in a park as an example, two raw materials from the reaction process are mixed and enter the product refining process. Process referencefigure 2 , The two feeds are first mixed in the raw material mixer 2. The raw material mixer 2 serves as a buffer, and then the material enters the high-pressure weight removal tower 2 from the 22nd plate in the tower, and the pressure of the high-pressure weight removal tower 2 is controlled at 500kPa and the temperature At 160°C, benzene and cumene are extracted from the top of the tower and enter the light removal tower 3, and the stream with control cumene <11wt% and polycumene extracted from the tower bottom enters the low pressure weight removal tower 4. The feed position of the low-pressure weight removal tower 4 is 11 plates, and the tower pressure is controlled at 30-55kPa. The fresh cumene material enters from 10 plates to enhance the separation effect. The low-pressure weight removal tower 4 adjusts the reflux ratio, The tower pressure and fresh material feed amount control the purity of the cumene product at the top of the tower to meet the requirements, and the polycumene obtained from the tower still enters the subsequent heavy component processing system. The feed of the light-removing tower 3 is 33 plates, and the tower pressure is controlled at 125~155kPa. The fresh cumene material enters from 34 plates, which also serves the purpose of enhancing the separation effect. The light-removing tower adjusts the reflux ratio, tower pressure and The fresh material replenishment amount controls the purity of the cumene product at the bottom of the tower to meet the requirements, and the crude benzene obtained from the top of the tower enters the benzene tower to obtain by-product benzene. The two product streams enter the product tank 5. The fresh cumene comes from the cumene tank 6. At the same time, the process is equipped with a double-effect rectification heat exchanger. The main parameters are measured by the online analysis instrument and transmitted to the control system.

[0101]Reference for online control systemFigure 3-Figure 5. Through the online analysis instrument 16, obtain the raw material composition and operating parameters, take a certain working condition for explanation, benzene x1= 0.16, cumene x2=0.76, polycumene x3=0.08, temperature T1=130℃, pressure P1=700kPa, flow F1=89376kg/h, the top temperature T of the weight removal tower2=157, tower top pressure P2=487kPa, the signal is transmitted to the predictive model controller. The controller calculates the product purity prediction value of 99.7% by weight under the current working conditions according to the embedded mathematical model. The value is transmitted to the logic judgment system, which is less than the set value of 99.9% by weight. The detected input variables, control variables, and predicted values ​​are transmitted to the optimized model controller. After data dimensionality reduction is processed, the mathematical model is optimized according to the embedded equipotential line to determine the current operating point, such asimage 3 The arrow starts from the red point, and the model finds the shortest path to the target value.image 3 The red dot at the end of the arrow, after obtaining the variable corresponding to the target value, then upgrade the dimension processing, decompose to obtain the control variable value corresponding to the target operating point, and each variable value is transmitted to the corresponding control valve to complete automatic adjustment through PID parameter adjustment, by increasing the reflux of the tower Ratio, reduce the fresh material replenishment amount, reduce the tower pressure and wait to the target value. At this time, the detected product purity value will be fed back to the logic judgment system as the predicted value, and the adjustment will be stopped if it reaches it. Otherwise, the adjustment and optimization will continue to the target value. Data detection and judgment, the entire system completes an adjustment time of 5 to 12 minutes.

[0102]Double-effect distillation system referencefigure 2 , In stable operation, the top stream of the high-pressure de-weighting tower is used to heat the low-pressure de-weighting tower bottom stream. When driving, each uses its own driving condenser/reboiler to complete the start-up of the tower equipment, and then gradually switches to the double-effect rectification heat exchanger. When shutting down and abnormal working conditions, it is also adjusted by driving the condenser/reboiler. Ensure the safe operation of tower equipment.

[0103]After monitoring for a period of time, the program in the above example works well. It can not only ensure the energy saving and consumption reduction of the process, but more importantly, the quality of the cumene product in the new process design is stable. When the working conditions fluctuate, the online adjustment and optimization model will Real-time optimization and adjustment to offset the impact of fluctuations, and at the same time, remove the weight first and then remove the light, and the two-tower weight reduction model can ensure the minimum product loss.

[0104]The scheme in the above embodiments of the present invention newly constructs a cumene refining process flow that integrates the characteristics of quality optimization, energy optimization, and real-time optimization. The process improves product purity, reduces process energy consumption, and adjusts in real time to deal with the impact of fluctuations in different working conditions and raw material parameters.