A method for predicting process parameters of a catalytic cracking reaction-regeneration system based on FVM and ANN

Abstract

The application discloses a catalytic cracking reaction-regeneration system process parameter prediction method based on FVM and ANN. The application generates comprehensive and large simulation data by accurately simulating the catalytic cracking reaction-regeneration system through FVM, and realizes fast, real-time and low-cost prediction of the full-space process, flow and reaction parameters of the catalytic cracking reaction-regeneration system by training an ANN model by using part of the simulation data and device historical data. The prediction method can be used as an expansion of the safety warning function of the DCS system, and can also provide a reference for the running state of the device for process and design personnel.

Description

Technical Field

[0001] This invention belongs to the field of petrochemical production process control and optimization technology, specifically relating to a method for predicting catalytic cracking process parameters based on the finite volume method (FVM) and artificial neural network (ANN). Background Technology

[0002] Catalytic cracking is an important refining process, but its process is complex and some raw materials and products have flammable and explosive properties, and some by-products are toxic and harmful. Therefore, ensuring process safety is a key prerequisite for the long-term stable production of catalytic cracking units.

[0003] Currently, refineries widely use DCS systems to monitor and automatically control process parameters, which can, to a certain extent, enable early warning and emergency response to unit risks, thus preventing safety accidents. Traditional DCS systems use sensors distributed throughout the unit to sense its operating status, including sensors for temperature, pressure, flow rate, speed, and valve opening. However, the relevant characteristics of flow and reaction within the unit cannot be easily sensed by existing sensing technologies, such as the residence time of oil, gas, and catalyst in different parts of the unit, the coking reaction rate in different parts of the unit, and the distribution of products leaving the unit. Flow and reaction are closely related to whether the unit is in a normal and safe production state. Furthermore, the traditional sensor layout can only measure parameters at a few locations and cannot sense process parameters throughout the entire unit. Safety monitoring based on process parameters from only a few locations may result in missed or false alarms.

[0004] Although FVM simulation of catalytic cracking units can provide insights into the internal flow and reaction conditions, the computational time cost is extremely high. This is because FVM simulation depends on the size of the grid set Ω, which is typically in the millions to tens of millions for industrial units. Simulating actual flow for 60 seconds would take approximately two weeks. Consequently, the simulation results cannot reflect the actual operating status of the catalytic cracking unit in real time, and therefore cannot achieve the task of safe monitoring of flow, reaction, and other factors. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting process parameters of a catalytic cracking reaction-regeneration system based on FVM and ANN. By performing accurate FVM simulation of the catalytic cracking reaction-regeneration system, comprehensive and large-scale simulation data is generated. Then, a portion of the FVM simulation data and historical data of the device are used to train the ANN model, thereby achieving rapid, real-time, and low-cost prediction of the process, flow, and reaction parameters of the entire space of the catalytic cracking reaction-regeneration system.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for predicting process parameters of a catalytic cracking reaction-regeneration system, comprising the following steps:

[0008] S1. Establish the device geometric model of the system;

[0009] S2. The geometric model of the device is meshed to form a complete set of computational grids Ω with a total number of grids Q;

[0010] S3. Initialize boundary conditions;

[0011] S4. Based on the actual operating parameters of the device, set multiple sets of boundary conditions and perform FVM simulation calculations to obtain the time interval T of the system in the complete set Ω of the computational grid under different operating conditions. s The complete set of physical quantity simulation parameters V;

[0012] In time interval T s Within this framework, the complete set V of physical quantity simulation parameters is divided into a subset V of measurable process parameter simulation data. s and the subset V of simulation data for unmeasurable process parameters R ;

[0013] S5. Set the distance threshold δ for nearby areas. i Based on actual monitoring needs, construct geometric constraints S. i The monitoring location is divided into a subset φ of the monitoring area corresponding to the full set Ω of the computing grid. i ;

[0014] According to the monitored area subset φ i Update the subset V of the measurable process parameter simulation data s ;

[0015] For the updated subset V of the measurable process parameter simulation data s Perform mean calculation to obtain a subset of the simulated data mean.

[0016] S6. Collect measurable process parameter data d of the system during actual operation over a period of time. i Select the process parameters that need to be monitored to form the complete industrial data set D;

[0017] The data frequencies in the complete industrial data set D are unified and aligned to obtain the complete actual data set.

[0018] For the mean subset of the simulated data and the complete set of actual data Perform data standardization separately to obtain a standardized subset of the simulated data mean. and standardized real data set

[0019] S7. Construct the ANN model; using the standardized complete set of actual data. As input data, the standardized subset of the simulated data mean. As output data, the ANN model is iteratively trained until the expected prediction accuracy is met.

[0020] S8. Take a set of prediction output data from the ANN model that meets the expected prediction accuracy. Perform inverse standardization calculations to obtain the predicted measurable process parameter data y;

[0021] Traverse the subset V of the measurable process parameter simulation data S Similarity calculations are performed on the predicted measurable process parameter data y to obtain a set of similarity coefficients C. S ;

[0022] S9. Select the similarity coefficient C. S The maximum value c s，p Simulated data v of unmeasurable process parameters at time p r，p As predicted process, flow, and reaction-related parameters.

[0023] In step S2, the computational grid set Ω is formed by calculation with Ox, Oy, and Oz as the flow directions;

[0024] Ω={(x1, y1, z1, id1), (x2, y2, z2, id2),..., (x Q y Q , z Q id Q )}.

[0025] In step S3, the initial boundary conditions assume that both the oil and gas phase and the catalyst phase in the calculation region are Eulerian fluids with continuous velocity, volume fraction and temperature in the calculation region, and that the catalyst phase and the oil and gas phase coexist and permeate each other.

[0026] In step S3, the governing equations in the complete computational grid Ω include:

[0027] Oil-gas phase continuity equation:

[0028] Catalyst phase continuity equation:

[0029] Oil and gas phase momentum equation:

[0030]

[0031] Catalyst phase momentum equation:

[0032]

[0033] Catalyst phase approximate temperature equation:

[0034]

[0035] Oil and gas responsive force-strain tensor:

[0036]

[0037] Catalyst-related force-strain tensor:

[0038]

[0039] in:

[0040] ε g ε is the gas phase volume fraction; s The solid volume fraction; u β U is the gas phase velocity, in m / s; s For solid phase velocity, m / s; ρ g The density is in the gas phase, kg / m³ 3 ;ρ s The density is the solid phase density, kg / m³. 3 ;p g p is the gas phase pressure, Pa; s The solid pressure is Pa; g is the acceleration due to gravity, m / s². 2 ;τ g For the corresponding force of air, Pa; τ s For solid-state reaction force, Pa; β is the fluid-structure interaction coefficient; θ s For particle temperature, m 2 / s 2 I is the unit tensor, κ s γ is the particle temperature diffusivity, Pa·s; s Energy consumption during collision, J / m 3 ·s; μ g The viscosity is given in Pa·s and μ. s λ is the shear viscosity coefficient; s is the volume viscosity coefficient.

[0041] In step S4, the multiple sets of boundary conditions are multiple sets of operating conditions formed by floating ±20% based on the process parameters of the actual operating conditions.

[0042] The complete set of simulation parameters for the physical quantities is V = [v1, v2, ..., v].n ].

[0043] In step S4, the subset V of the measurable process parameter simulation data S =[v s，1 v s，2 , ..., v s，n The subset V of simulation data for the unmeasurable process parameters R =[v r，1 v r，2 , ..., v r，n ]; V = V S ∪V R ,and For any time t∈[1, n], there exists a set of v r，t With v s，t One-to-one correspondence.

[0044] In step S5, the geometric constraint g i as follows:

[0045]

[0046] Wherein: g i The mathematical expression representing the geometric shape; m is the number of monitoring points. The geometric shape includes common geometric shapes such as planes, curved surfaces, and spheres.

[0047] Through geometric constraints, the monitored area subset φ i Accurately express V s The specific spatial region measured by the corresponding sensor in the catalytic cracking reaction-regeneration system. The subset φ of the monitored region. i , i∈[1,m]. The subset of the mean of the simulated data It can reflect the measurement value of the corresponding sensor at the corresponding time.

[0048] In step S6, the collected measurable process parameter data d i It can be derived from any one of the following: industrial equipment MES database, process simulation data, and pilot-scale experimental data.

[0049] The measurable process parameter data d i , i∈[1,m].

[0050] The complete industrial data set D consists of m process parameters d i Composition, with dimensions of m×N, D∈V s d i With φ i One-to-one correspondence.

[0051] In step S6, the steps for unifying and aligning the data frequencies are as follows:

[0052] (1) Determine the maximum physical time t corresponding to all locations in the complete industrial data set D. max and minimum value t min Define the time interval δ t , and δ t The interval should be greater than or equal to the minimum sampling interval of the data to construct an empty matrix. The size is n×[(t) max -t min ) / / δ t +1], where / / indicates integer division;

[0053] (2) Traverse all elements d in the complete industrial data set D. i，j For i∈[1,n], j∈[1,T], obtain the corresponding physical time t from the original data source. i，j Let i∈[1, n], j∈[1, T], let elements in

[0054] (3) Traversing the matrix All row vectors set up The number of empty elements in the middle is n e If n s >0.7n, then from Delete The final matrix The size is n×M, where M≤(t) max -t min ) / / δ t -1.

[0055] In step S6, the method for selecting the process parameters to be monitored can be either experience-based or any method based on feature selection algorithms.

[0056] In step S6, the complete set of actual data

[0057]

[0058] In the formula, Let x be the i-th process parameter in the complete industrial data set D. i All data.

[0059] In step S6, the complete set of actual data and the subset of the mean of the simulated data The data standardization process is as follows:

[0060] (1) Using the complete set of actual data For example, for the complete set of actual data any site in Perform the following transformation:

[0061]

[0062] in, and They are respectively The minimum and maximum values, where e is the length and... The same unit vector is used to traverse the entire actual data set. Each site in Obtain the standardized complete set of data

[0063] (2) The subset of the mean of the simulated data The standardization method is the same as step (1), resulting in a standardized subset of the simulated data mean.

[0064] In step S7, the expression of the ANN model is as follows:

[0065]

[0066] Where, x j ω represents the input received by the j-th neuron. j b represents the weight of the j-th neuron. j Let f(x) represent the bias of the j-th neuron, and f(x) represent the activation function of the neuron.

[0067] In the ANN model, both the input and output layers contain m neurons, and the hidden layers contain n neurons.

[0068] In the iterative training, the activation function used can be any activation function that meets the prediction accuracy requirements of the actual problem, such as Sigmoid, tanh, ReLU and its improved forms.

[0069] In step S9, the denormalization calculation process is as follows:

[0070]

[0071] in, and These are the complete industrial data sets. The vector formed by the maximum and minimum values ​​of each variable in the vector has a dimension of m.

[0072] In step S9, the similarity calculation metric can be any mathematical metric that meets the needs of the actual problem, such as mean squared error (MSE) or mean absolute error (MAE).

[0073] The similarity coefficient C S =[c s，1 c s，2 c s，n ].

[0074] The beneficial effects achieved by this invention are as follows:

[0075] 1. Compared with existing technologies, this invention uses FVM simulation and ANN training to predict safety indicators based on flow and reaction factors inside catalytic cracking units. The results are reliable, filling a gap in the field and providing substantial guidance in monitoring related safety hazards.

[0076] 2. During the ANN training process, this invention simultaneously utilizes actual operating data of the catalytic cracking unit (the complete set of actual data). ) and first-principles simulation data (the complete set of FVM simulation parameters V, and the transformation results based on V, a subset of the simulation data mean). Standardized subset of simulated data mean This allows the trained ANN model to have certain physical meaning, ultimately forming a hybrid model that has stronger interpretability compared to other related intelligent models.

[0077] 3. The prediction method described in this invention can adjust the prediction target by changing the process parameters contained in the complete data set D, and can also adjust the prediction target by changing g. i The expression form and the threshold δ of the nearest region i This allows for adjustments to the shape and extent of the predicted area, thus enabling flexible responses to different monitoring tasks at different locations.

[0078] 4. The prediction method provided by this invention overcomes the problem of high cost of FVM simulation calculation for catalytic cracking industrial units. The model application process can achieve real-time prediction (second-level) and has a very fast running speed. In terms of training data selection, data from industrial units over a long period of time can be used. Therefore, it can fully learn the flow and reaction conditions inside the catalytic cracking unit and has strong robustness, which is difficult to achieve with traditional methods.

[0079] 5. The prediction method described in this invention can be used as an extension of the safety early warning function of the DCS system, and can also provide a reference for the operating status of the equipment for process and design personnel. Attached Figure Description

[0080] Figure 1 This is a schematic diagram of the overall process of the prediction method provided by the present invention.

[0081] Figure 2 The diagram shows the geometric modeling and mesh generation of a catalytic cracking unit in the embodiment; the left figure shows the geometric modeling, and the right figure shows the mesh generation.

[0082] Figure 3 This is a schematic diagram showing the actual division of the monitoring area in the embodiment; the left image is a top view; the right image is a side view.

[0083] Figure 4 This is a schematic diagram of the topology of the process parameter prediction model ANN in the embodiment.

[0084] Figure 5 This is to compare the predicted temperature value with the actual value using the method of the present invention.

[0085] Figure 6 This is to compare the predicted pressure value with the actual value using the method of the present invention.

[0086] Figure 7 This is a similarity curve when matching data using the method of the present invention. Detailed Implementation

[0087] The present invention will be further described below with reference to specific embodiments, but the present invention is not limited to the following embodiments.

[0088] Unless otherwise specified, all methods described herein are conventional methods.

[0089] Example

[0090] like Figure 1 As shown, the specific steps are as follows:

[0091] (1) Based on the design drawings of a certain catalytic cracking reaction-regeneration system, construct the geometric model of the system;

[0092] (2) Establish a spatial rectangular coordinate system O-xyz, mesh the model, and perform calculations with Ox, Oy, and Oz as the flow directions, forming a computational grid set Ω with a total number of grids Q; (e.g.) Figure 2 (as shown);

[0093] Ω=((x1,y1,z1,id1),(x z ,y2,z2,id2),...,(x Q y Q , z Q id Q )}

[0094] (3) Initialize boundary conditions. Assume that both the oil / gas phase and the catalyst phase in the computational domain are Eulerian fluids. Specify the flow rate, temperature, and pressure parameters of materials such as feedstock, catalyst, and steam, for example, the oil / gas temperature T. oii =500℃ settling tank top pressure P d =272kPa, catalyst mass flow rate G s=580kg / s, etc.

[0095] (4) Multiple sets of boundary conditions were formed by varying the actual operating parameters by ±20%. FVM simulations were performed with a physical flow duration of 300 s and a time interval of 0.05 s to obtain the reaction-regeneration system of the catalytic cracking unit under different operating conditions within the computational grid set Ω and time interval T. s V is the complete set of physical quantities within the range [0.05, 0.10, ..., 300.00].

[0096] In this embodiment, the complete set of physical quantities V is divided into a dataset of measurable process parameters V. S = [T, P], where T represents temperature variables that can be measured by temperature sensors, P represents pressure variables that can be measured by pressure sensors; and V is a subset of data of process, flow, and reaction-related parameters that cannot be directly measured. R =[G c ,φ,t], where G c φ represents the coking reaction rate at various points in the unit, φ represents the yield of various main products, and t represents the residence time of the product oil and gas at various points in the unit.

[0097] (5) Taking the reactor outlet as an example, set the proximity distance threshold δ. reactor-out =0.1, construct the following geometric constraint g f To make it accurately express V s The area near the reactor outlet, such as Figure 3 As shown, this yields the subset of the reactor outlet in Ω, and also the temperature subset T corresponding to the reactor outlet region. reactor-out and the pressure subset P peactor-out T reactor-out ∈T, P reactor-out ∈P;

[0098] The above geometric constraint g i as follows:

[0099]

[0100] Following the steps above, divide all locations that need to be monitored one by one, and then, according to the requirements, modify the original V... s Divide the data into subsets to obtain the updated V. s =[T1, T2, T3, T4, P1, P2].

[0101] For the updated V s Calculate the mean and get This can reflect the measurement value of the corresponding sensor at the corresponding time.

[0102] (6) Extract one year's worth of real-time data from the MES database of the catalytic cracking unit. The data has a dimension of 6 and is aligned to form a complete set of actual data of size 1000.

[0103]

[0104] For the complete set of actual data and the subset of the mean of the simulated data Standardize them separately to obtain a complete set of standardized actual data. and a standardized subset of the mean of the simulated data

[0105] (7) Construct an ANN model with 6 neurons in both the input and output layers and 50 neurons in the hidden layers, such as... Figure 4 As shown, the activation function is Sigmoid, the optimizer is RMSprop, the initial learning rate is set to 0.001, the number of iterations is set to 500, and the batch size is set to 100.

[0106] With the standardized complete set of actual data As input data, a subset of the mean of the standardized simulated data The ANN model is iteratively trained using the output data until the expected prediction accuracy is achieved, such as... Figure 5 , Figure 6 As shown.

[0107] (8) A set of predicted outputs of the ANN model After inverse standardization, the measurable process parameters y = [483, 462, 495, 501, 232, 225] are obtained;

[0108] Traversing V S The similarity coefficient C is obtained by performing similarity calculation on y. s ,like Figure 7 As shown;

[0109] (9) The above similarity coefficient C s The maximum value is c s，479 =97.79%, corresponding to time 479, the non-ideal secondary reaction rate of the corresponding monitoring site is:

[0110] G c，479 =[0.9973, 0.1168, 0.4035, 0.6144, 0.0013, 5.8623E -5 2.6308E -4 4.8577E -4 8.3669E -4 Product yield φ479 = [3.00%, 15.50%, 5.04%, 45.50%, 25.50%, 3.50%, 1.96%], residence time t of each part of the device. 479 = [5.22, 113.72, 40.92], completing the prediction of process, flow, and reaction-related parameters.

[0111] For the catalytic cracking reaction-regeneration system in the above embodiments, the calculation time required for a set of actual operating conditions using FVM simulation is about one month, which is too costly and cannot provide real-time guidance. However, using the prediction method described in this invention, the data generated by previous FVM simulations can be used during the model ANN training phase, and the online prediction time is about 2 seconds. The model application process can achieve real-time prediction (at the second level).

[0112] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for predicting process parameters of a catalytic cracking reaction-regeneration system, comprising the following steps: S1. Establish the device geometric model of the system; S2. The geometric model of the device is meshed to form a total of [number of meshes]. The complete set of computational grids ; S3. Initialize boundary conditions; S4. Based on the actual operating parameters of the device, set multiple sets of boundary conditions and perform FVM simulation calculations to obtain the system under different operating conditions in the full set of the computational grid. time interval Complete set of physical quantity simulation parameters ; In the time interval Within, the complete set of simulation parameters for the physical quantities is included. Divided into a subset of measurable process parameter simulation data and subset of simulation data for unmeasurable process parameters ; S5. Set the distance threshold for nearby areas. Based on actual monitoring needs, construct geometric constraints. The monitoring locations are divided into the entire set of the computing grid. The corresponding subset of monitoring areas ; According to the subset of the monitored area Update the subset of simulated data for the measurable process parameters ; For the updated subset of the simulated data of the measurable process parameters Perform mean calculation to obtain a subset of the simulated data mean. ; S6. Collect measurable process parameter data of the system during actual operation over a period of time. Select the process parameters that need to be monitored to form a complete industrial data set. ; for the complete set of industrial data The data frequencies in the dataset are unified and aligned to obtain the complete set of actual data. ; For the mean subset of the simulated data and the complete set of actual data Perform data standardization separately to obtain a standardized subset of the simulated data mean. and standardized real data set ; S7. Construct the ANN model; using the standardized complete set of actual data. As input data, the standardized subset of the simulated data mean. As output data, the ANN model is iteratively trained until the expected prediction accuracy is met. S8. Take a set of prediction output data from the ANN model that meets the expected prediction accuracy. Inverse standardization calculations are performed to obtain predicted measurable process parameter data. ; Traverse the subset of simulated data for the measurable process parameters For the predicted measurable process parameter data Similarity calculations are performed to obtain a set of similarity coefficients. ; S9. Let the similarity coefficient be... The maximum value is The corresponding time is ,choose The corresponding unmeasurable process parameter simulation data As predicted process, flow, and reaction-related parameters.

2. The method for predicting process parameters of a catalytic cracking reaction-regeneration system according to claim 1, characterized in that, In step S3, the initialization boundary conditions include the complete set of the computational grid. The governing equations include: Oil-gas phase continuity equation: Catalyst phase continuity equation: Oil and gas phase momentum equation: Catalyst phase momentum equation: Catalyst phase approximate temperature equation: Oil and gas responsive force-strain tensor: Catalyst-related force-strain tensor: in: This refers to the volume fraction of the gas phase. It represents the volume fraction of the solid phase. For gas phase velocity, ; For solid phase velocity, ; The density is the gas phase density. ; For solid density, ; For gas phase pressure, ; For solid-phase pressure, ; It is the acceleration due to gravity. ; For the corresponding force of air, ; To solidify the corresponding force, ; The fluid-structure exchange coefficient; For particle temperature, ; For unit tensors, The particle temperature diffusivity, ; For collision energy consumption, ; The viscosity is the gas phase viscosity. ; It is the shear viscosity coefficient; is the volume viscosity coefficient.

3. The method for predicting process parameters of a catalytic cracking reaction-regeneration system according to claim 1 or 2, characterized in that, In step S4, the multiple sets of boundary conditions are multiple sets of operating conditions formed by floating ±20% based on the process parameters of the actual operating conditions.

4. The method for predicting process parameters of a catalytic cracking reaction-regeneration system according to claim 1 or 2, characterized in that, In step S4, ,and For any given moment There is a set of and One-to-one correspondence.

5. The method for predicting process parameters of a catalytic cracking reaction-regeneration system according to claim 1 or 2, characterized in that, In step S5, the geometric constraints as follows: in: Mathematical expressions representing geometric shapes; The number of monitoring sites; This represents the distance threshold to the nearest neighbor region of a geometric shape.

6. The method for predicting process parameters of a catalytic cracking reaction-regeneration system according to claim 1 or 2, characterized in that, In step S6, the collected measurable process parameter data It can be derived from any one of the following: industrial equipment MES database, process simulation data, and pilot-scale experimental data.

7. The method for predicting process parameters of a catalytic cracking reaction-regeneration system according to claim 1 or 2, characterized in that, In step S6, the steps for unifying and aligning the data frequencies are as follows: (1) Determine the complete set of industrial data Maximum physical time corresponding to all sites and minimum value Define time interval ,and The sampling interval should be greater than or equal to the minimum sampling interval of the data, and the construction size should be [size missing]. An empty matrix, in which Indicates integer division; (2) Traverse the entire set of industrial data. All elements in Obtain the corresponding physical time from the original data source. ,make elements in ; (3) Traverse the entire set of actual data All row vectors of the matrix ,set up The number of empty elements in the middle is ,like Then from Delete The final matrix Size is ,in .

8. The method for predicting process parameters of a catalytic cracking reaction-regeneration system according to claim 1 or 2, characterized in that, In step S6, the complete set of actual data and the subset of the mean of the simulated data The data standardization process is as follows: (1) For the complete set of actual data Any row vector in Perform the following transformation: in, and They are respectively The minimum and maximum values, For length and The same unit vector is used to traverse the entire actual data set. Each row vector in To obtain a standardized complete set of actual data ; (2) The mean subset of the simulated data The standardization method is the same as step (1), resulting in a standardized subset of the simulated data mean. .

9. The method for predicting process parameters of a catalytic cracking reaction-regeneration system according to claim 1 or 2, characterized in that, In step S7, the expression of the ANN model is as follows: in, Indicates the first The input received by each neuron Indicates the first The weights of each neuron, Indicates the first Bias of each neuron This represents the activation function of the neuron.

10. The method for predicting process parameters of a catalytic cracking reaction-regeneration system according to claim 1 or 2, characterized in that, In step S9, the denormalization calculation process is as follows: in, and These are the complete industrial data sets. The vector formed by the maximum and minimum values ​​of each variable in the vector has dimensions of . .

Images