1. A method for optimizing a through disc of an atmospheric and vacuum device is characterized by comprising the following steps:

determining the range covered by the constant pressure reduction device and the related variable;

establishing an optimization model of the atmospheric and vacuum device, wherein the optimization model comprises an optimization objective equation and constraint conditions;

obtaining data of the selected variables under the required working condition by a physical logic reconstruction method to obtain initial values, crude oil properties and device performance parameters required by general disc optimization;

and writing an optimization program on the optimization platform according to a programming rule, substituting each variable into the solved initial value, solving the optimization model by using a nonlinear solver and a formulated optimization strategy to enable the optimized value to jump out of local optimization, and finally obtaining a global optimal solution of the target equation and the adjustment direction and amplitude of the corresponding optimized variable.

2. The general disc optimization method according to claim 1, wherein the establishment of the optimization objective equation comprises the following processes:

determining an optimization objective, wherein the optimization objective comprises economic efficiency maximization, product yield maximization/minimization, utility consumption minimization, or a balanced combination thereof;

and establishing an optimization objective equation.

3. The general disc optimization method according to claim 2, wherein the establishment of the constraint condition comprises the following processes:

selecting input and output variables of the model by a dimension reduction method;

training a relation model between a certain output variable and a corresponding input variable by adopting a machine learning method;

establishing an AI model of the atmospheric and vacuum device with a plurality of neuron network groups by adopting a machine learning method and an experience mechanism hybrid algorithm, wherein the AI model is used for equality constraint conditions in an optimization model of the atmospheric and vacuum device;

and (3) counting the historical data by adopting a statistical method to obtain a variable allowable variation range, thereby obtaining an inequality constraint condition in the optimization model of the atmospheric and vacuum device, and combining the equality constraint condition formed in the last step to form a constraint condition in the optimization model of the atmospheric and vacuum device.

4. The general disk optimization method according to claim 1, wherein in each step of optimization direction search, the influence of the selected variables on the optimization target and the potential bottleneck of the whole device is measured in a unified manner.

5. The general disc optimization method according to claim 4, wherein in each step of optimization direction search, whether the optimization model meets production requirements is calculated based on set site constraints, wherein the site constraints comprise five site constraint conditions of material balance, energy balance, phase balance, heat transfer and equipment performance.

6. The general disc optimization method according to claim 5, wherein the constraint conditions are as follows:

y_n＝f_n(F，T，P，y_p) n，p∈N，n≠p

Prop_m＝f_m(F，T，P，y_n)

y_n，L≤y_n≤y_n，U

ΔF_min≤ΔF≤ΔF_max

ΔT_min≤ΔT≤ΔT_max

ΔP_min≤ΔP≤ΔP_max

Prop_m，minspec+Prop_m，accuracy≤Prop_m≤prop_m，maxspec-Prop_m，accuracy

wherein F, T, P respectively indicate the flow, temperature and pressure adjustable independent variables in the optimization, y_nVariables other than the physical properties and adjustable independent variables required by the process index in the side product, y_pShows the physical property, adjustable independent variable and y required by the process index in the products except the side line_nVariables other than; f represents a plurality of physical models of material balance, energy balance, phase balance, pressure balance and heat transfer; n represents the set of 1 to N, i.e., 1, 2, 3 … … N; min represents the minimum value and max represents the maximum value; prop_mShows the physical property, Prop, required by the process index in a side product_m，minspec，Prop_m，maxspecRespectively representing the upper and lower limits of the index requirement, Prop_m,accuracyThe model prediction accuracy indicating this physical property; Δ F, Δ T, Δ P are the variable amplitudes and the associated inequality is the constraint on the optimization amplitude.

7. The general disk optimization method according to any one of claims 3 to 6, wherein the optimization of the variables is classified according to adjustment frequency, wherein the frequencies of the variables are classified into three types, namely high frequency, medium frequency and low frequency, the adjustment level is classified into three levels, the 3-level adjustment comprises all the variables, the 2-level adjustment comprises medium and low frequency variables, the 1-level adjustment comprises low frequency variables, the optimization variables are determined according to the variable categories related to the adjustment level, and the variables not in the adjustment level are fixed to be constant.

8. The general disc optimization method according to claim 5 or 6, wherein the generalized gradient algorithm is adopted for optimization solution, and whether the equipment performance is met is measured and calculated simultaneously based on the high-precision ANN model in each step of optimization direction search, the method comprises the following steps: the tower plate hydraulics performance of the fractionating tower, the heat transfer capacity of a heat exchanger, the operable range of conveying equipment, the load bottleneck of a heating furnace and the high-temperature coking tendency of the equipment.

9. The method for optimizing the communication disc according to claim 1, wherein before the optimization model is used, whether the current working condition exceeds the range of the model or not is further judged, if the current working condition exceeds the range, working condition enhancement is performed in the range near the current working condition, a part of samples are newly added, the newly added part of sample information is gathered into the previous samples to perform correction training on the model, and model self-learning is achieved.

10. The use of the general disc optimization method according to any one of claims 1 to 9 in a normal pressure reduction device of an oil refinery.

Background

The production decision-making mode of the refining industry is measured and calculated by establishing a mechanism model by the current process experts from the past development depending on experience, and changes in two stages are carried out.

An empirical decision stage: lacking a measuring and calculating tool and a measuring and calculating model, planning personnel distribute logistics trends for production plans according to statistical data and experience; the operator gives the operation parameter adjusting direction according to the operation experience and the operation manual of the operator and the production condition of the site, and adjusts the operation parameter in small steps according to the test analysis result.

A mechanism model measuring, calculating and guiding stage for a technical professional: the method comprises the steps of establishing and correcting a mechanism model through working condition calibration, verifying by a process expert according to an optimization thought provided by professional knowledge and by utilizing mechanism model simulation measurement or setting a simple optimization strategy for measurement and calculation, and finally obtaining a production plan and an operation scheme for guiding production. The process simulation software was introduced in the 50 th of the 20 th century, and after decades of development, the process simulation software is widely applied in the industry and cultivates a group of excellent modeling and process optimization experts. The process experts provide guidance for production by utilizing a mechanism model and a professional knowledge measuring and calculating device processing scheme and a production plan, and certain economic benefit is obtained.

However, the production conditions of the atmospheric and vacuum distillation device are often in the dynamic adjustment process, and currently, the process optimization is carried out by depending on experts and simulation software, which is far from keeping up with the changing requirements. Enterprises need an overall solution for optimizing the atmospheric and vacuum distillation unit so as to further improve the decision management level of the crude oil processing industrial chain of the enterprises and convert the business targets of the enterprises into the operation targets in production operation.

The conventional optimization idea usually considers the influence of a single operation variable at a time, and hardly considers the influence of the simultaneous adjustment of a plurality of operation variables on various aspects of the whole device, so how to put one eye on the operation cycle of the whole device, from the perspective of production or economic benefit, quickly obtain a set of feasible and accurate optimization scheme which is globally considered for the current state of the device, and is a technical problem to be solved urgently at present.

Disclosure of Invention

In view of the above problems in the prior art, an object of the present invention is to provide a general disk optimization method for an atmospheric and vacuum distillation unit, which considers the influence of the adjustment of a plurality of operating variables on a system or a device, considers the current state of the unit and the long-period stable operation requirement of the unit, and provides an optimization scheme through global consideration, so that the unit achieves the purposes of energy saving, efficiency improvement and long-period operation.

In order to solve the technical problems, the technical scheme adopted by the invention is as follows:

a method for optimizing a through disc of an atmospheric and vacuum device comprises the following steps:

determining the range covered by the constant pressure reduction device and the related variable;

establishing an optimization model of the atmospheric and vacuum device, wherein the optimization model comprises an optimization objective equation and constraint conditions;

obtaining data of the selected variables under the required working condition by a physical logic reconstruction method to obtain initial values, crude oil properties and device characteristic parameters required by general disc optimization;

and writing an optimization program on the optimization platform according to a programming rule, substituting each variable into the initial value, solving the optimization model by using a nonlinear solver and a formulated optimization strategy to enable the optimization value to jump out of local optimization, and finally obtaining a global optimal solution of the target equation and the adjustment direction and amplitude of the corresponding optimization variable.

Further, the establishment of the optimization objective equation comprises the following processes:

determining an optimization objective, wherein the optimization objective comprises economic efficiency maximization, product yield maximization/minimization, utility consumption minimization, or a balanced combination thereof;

and establishing an optimization objective equation.

Further, the establishment of the constraint condition includes the following processes:

selecting input and output variables of the model by a dimension reduction method;

training a relation model between a certain output variable and a corresponding input variable by adopting a machine learning method; establishing an AI model of the atmospheric and vacuum device with a plurality of neuron network groups by adopting a machine learning method and an experience mechanism hybrid algorithm, wherein the AI model is used for equality constraint conditions in an optimization model of the atmospheric and vacuum device;

and (3) counting the historical data by adopting a statistical method to obtain a variable allowable variation range, thereby obtaining an inequality constraint condition in the optimization model of the atmospheric and vacuum device, and combining the equality constraint condition formed in the last step to form a constraint condition in the optimization model of the atmospheric and vacuum device.

Further, in each step of optimization direction search, the influence of the selected variables on the optimization target and the potential bottleneck of the whole device is measured and calculated in a unified manner, wherein the variables comprise the treatment capacity of each tower, the outlet temperature of the atmospheric furnace and the vacuum furnace, the flow rate of stripping steam at each position, the flow rate of reflux at each position and the product yield.

And further, in each step of optimization direction search, whether the optimization model meets production requirements is measured and calculated simultaneously based on set site constraints, wherein the site constraints comprise five site constraint conditions of material balance, energy balance, phase balance, heat transfer and equipment performance.

Further, the constraint conditions are as follows:

y_n＝f_n(F，T，P，y_p) n，p∈N，n≠p

Prop_m＝f_m(F，T，P，y_n)

y_n，L≤y_n≤y_n，U

ΔF_min≤ΔF≤ΔF_m，ax

ΔT_min≤ΔT≤ΔT_max

ΔP_min≤ΔP≤ΔP_max

Prop_m，minspec+Prop_m，accuracy≤Prop_m≤Prop_m，maxspec-Prop_m，accuracy

wherein F, T, P respectively indicate the flow, temperature and pressure adjustable independent variables in the optimization, y_nVariables other than the physical properties and adjustable independent variables required by the process index in the side product, y_pShows the physical property, adjustable independent variable and y required by the process index in the products except the side line_nVariables other than; f represents a plurality of physical models of material balance, energy balance, phase balance, pressure balance and heat transfer; n represents the set of 1 to N, i.e., 1, 2, 3 … … N; min represents the minimum value and max represents the maximum value; prop_mShows the physical property, Prop, required by the process index in a side product_m，minspec，Prop_m，maxspecRespectively representing the upper and lower limits of the index requirement, Prop_m，accuracyThe model prediction accuracy indicating this physical property; Δ F, Δ T, Δ P are the variable amplitudes and the associated inequality is the constraint on the optimization amplitude.

Further, adjusting frequency classification is carried out on the optimized variables according to attributes, wherein the frequencies of the variables are divided into three types of high frequency, medium frequency and low frequency, the adjusting level is divided into three levels, 3 levels of adjustment comprise all variables, 2 levels of adjustment comprise medium and low frequency variables, 1 level of adjustment comprises low frequency variables, the optimized variables are determined according to the variable types related to the adjusting level, and the variables not in the adjusting level are fixed as constants.

Further, the generalized gradient algorithm is adopted for optimization solution, whether the equipment performance is met is measured and calculated simultaneously in each step of optimization direction search based on the high-precision ANN model, and the method comprises the following steps: the tower plate hydraulics performance of the fractionating tower, the heat transfer capacity of a heat exchanger, the operable range of conveying equipment, the load bottleneck of a heating furnace and the high-temperature coking tendency of the equipment.

And further, before the optimization model is used, whether the current working condition exceeds the range of the model or not is judged, if the current working condition exceeds the range, the working condition is enhanced in the range near the current working condition, a part of samples are newly added, and the newly added part of sample information is gathered into the previous samples to perform correction training on the model.

Compared with the prior art, the general disc optimization method of the atmospheric and vacuum distillation unit has the following technical effects:

1. the invention adopts a general disk optimization method, takes the result of physical logic reconstruction as the basis, comprehensively considers the actual feasibility constraint of the site, takes the improvement of the system economic benefit as the main target under a controllable optimization strategy, utilizes a mathematical programming algorithm, and optimizes and solves the direction and the size of the change of the operation condition, thereby providing a target for the generation of the adjustment step of the next process and further achieving the purposes of energy saving and efficiency improvement.

2. The optimization variables are comprehensive, and the influence of all the variables on the device on the site is comprehensively considered, so that the energy-saving and efficiency-increasing space is more fully excavated.

3. From the reality, an optimization strategy is made according to a field operation method, so that the optimization result is easier to implement, and the unhooking with the field is avoided.

4. The data quantity of the established model is large and is from field data or strict mechanism simulation data, the model can learn by self and continuously learn the characteristics of the device, and therefore the accuracy of the model characterization device is improved.

5. Crude oil properties are obtained through physical logic reconstruction, thereby ensuring the accuracy of optimization.

6. And obtaining an optimized initial value through physical logic reconstruction, and reducing the optimization time to be within 5 minutes.

7. The current running situation of the equipment is evaluated through the model, and the serious operation risk and economic loss of long-period running caused by excessive excavation of short-term optimization benefits are avoided, so that the long-term stable running requirement of the device is met.

Drawings

Fig. 1 is a schematic flow chart of a general disc optimization method of an atmospheric and vacuum distillation device according to an embodiment of the present invention.

Detailed Description

The present invention will be described in further detail below with reference to the accompanying drawings, but the present invention is not limited thereto.

Referring to fig. 1, a general disc optimization method for an atmospheric and vacuum distillation unit disclosed in an embodiment of the present invention includes the following steps:

step S1: the coverage of the device and the variables involved are determined. The atmospheric and vacuum distillation device mainly comprises three rectifying towers (a primary distillation tower, an atmospheric tower and a vacuum tower), two heating furnaces (an atmospheric furnace and a vacuum furnace), an electric desalting system and a heat exchange network, wherein variables comprise the treatment capacity of each tower, the outlet temperature of the atmospheric furnace and the vacuum furnace, the flow rate of stripping steam at each position, the reflux amount of each middle section, the product yield, the bypass flow rate of a heat exchanger and the like.

Step S2: and establishing an atmospheric and vacuum device optimization model which comprises an optimization objective equation and constraint conditions.

The establishment of the optimization objective equation comprises the following steps:

step S21: determining optimization objectives, including economic maximization, product yield maximization/minimization, utility consumption minimization, etc.;

step S22: and establishing an optimization objective equation.

The goal of the general-purpose optimization is to achieve the required output according to the project requirements, such as economic efficiency maximization, capacity maximization, public engineering minimization, etc., or a balanced combination thereof, or any other feasible set goal scheme, taking the maximization of economic efficiency in the operable area as an example, the objective function is substantially a maximum profit function, and the economic efficiency generally includes three parts: the revenue of each sidedraw product, the cost of the raw materials (which is generally fixed given the raw material construction and throughput), and the utility cost, the maximum revenue function is essentially the solution of the revenue of each sidedraw product minus the cost of the raw materials and the utility cost. Thus variables that affect economic efficiency include raw material processing, product production, and utility consumption, then the equation for this goal can be expressed as:

wherein p is the price，F_iFor the production of product i, F_jAs raw material j, processing amount, F_kIs the utility k usage.

The optimization constraint conditions of the atmospheric and vacuum devices comprise equality constraints and inequality constraints. The equality constraint is obtained by combining Artificial Neuron Network (ANN) training and empirical mechanism simulation. The ANN model simulates a network structure of neurons in the human brain and a processing mode of information, realizes the calculation output of input information by adopting a multilayer equation structure formed by connecting a large number of neurons, is a model with a complex structure and high calculation speed, and can simulate various processes with complex structure and numerous parameters. In addition, the ANN model has a self-learning function, so that the method is very suitable for the atmospheric and vacuum devices with variable working conditions. When the working condition is changed greatly, the ANN model has a self-correction function along with the input of new data on site, the model parameters can be adjusted gradually to adapt to the new working condition, and the simulation precision can be improved gradually.

The specific steps for establishing the optimization constraint conditions of the atmospheric and vacuum devices are as follows:

step S23: selecting input and output variables of the model by a dimension reduction method;

step S24: a machine learning method is adopted to train a relation model between a certain output and an input variable thereof, and parameters in machine learning are changed, so that the model obtained by training can better fit training data and can better predict test data;

step S25: if the accuracy of the model does not meet the requirement, changing the input variable corresponding to the output by a trial and error method, and continuing to train the model by adopting the step S24 until the accuracy meets the requirement;

step S26: and establishing an AI model of the atmospheric and vacuum device with a plurality of neuron network groups by adopting a machine learning method and an experience mechanism hybrid algorithm. Specifically, a product property soft instrument prediction model, a crude oil/intermediate stream property soft instrument prediction model, an operation condition soft instrument prediction model, a stream unit enthalpy value prediction model, a column plate gas-liquid phase load soft instrument prediction model, a pressure reducing furnace and transfer line related temperature and pressure prediction model and a heat exchange network model are obtained, and an atmospheric and vacuum plant AI model is formed by combining a quality model, an energy balance model and a pressure reduction tower whole-tower pressure drop related empirical model and is used for equality constraint conditions in an atmospheric and vacuum plant optimization model.

Step S27: and (3) counting the historical data by adopting a statistical method to obtain a variable allowable variation range, thereby obtaining an inequality constraint condition in the optimization model of the atmospheric and vacuum device, and combining the equality constraint condition formed in the last step to form a constraint condition in the optimization model of the atmospheric and vacuum device.

For example, 286 product properties, 93 crude oil/middle stream properties, 42 operating conditions, 124 tray gas-liquid phase loads, 46 stream units, the relevant temperature and pressure of the vacuum furnace and the transfer line and 98 heat exchange temperatures are selected as target variables according to the characteristics of a certain atmospheric and vacuum plant, and input variables affecting the target variables are selected by a dimension reduction method on the basis of hundreds of thousands of sets of operation data. For example, product properties may be targeted variables for the model. And analyzing the influence of other variables including raw material composition, equipment characteristics, operating conditions and the like on the target variable by using a dimension reduction method, thereby determining some variables which have larger influence on the target variable. And training the target variable and variable sample data influencing the target variable by adopting a machine learning method to obtain an AI model of the target variable.

Finally, 286 product property soft instrument prediction models, 93 crude oil/middle stream property soft instrument prediction models, 42 operation condition soft instrument prediction models, 124 tray gas-liquid phase load soft instrument prediction models, 98 heat exchange network models with detailed heat exchanger structures, 46 stream unit enthalpy value prediction models, 2 decompression furnaces and oil transfer line related temperature and pressure prediction models are trained according to requirements. And the constraint conditions of the optimization model of the whole atmospheric and vacuum distillation device are formed by combining the model with 2 pressure drop related models of the whole vacuum tower, 6 material balance models of the whole tower and 3 energy balance models of the whole tower.

The optimization constraint equation is illustrated as follows:

y_n＝f_n(F，T，P，y_p) n，p∈N，n≠p

Prop_m＝f_m(F，T，P，y_n)

y_n，L≤y_n≤y_n，U

ΔF_min≤ΔF≤ΔF_max

ΔT_min≤ΔT≤ΔT_max

ΔP_min≤ΔP≤ΔP_max

Prop_m，minspec+Prop_m，accuracy≤Prop_m≤Prop_m，maxspec-Prop_m，accuracy

wherein F, T, P respectively indicate the flow, temperature and pressure adjustable independent variables in the optimization, y_nVariables other than the physical properties and adjustable independent variables required by the process index in the side product, y_pShows the physical property, adjustable independent variable and y required by the process index in the products except the side line_nVariables other than; f represents a plurality of physical models of material balance, energy balance, phase balance, pressure balance and heat transfer; n represents the set of 1 to N, i.e. 1, 2, 3, …, N; min represents the minimum value and max represents the maximum value; prop_mThe physical properties required by the technical indexes in the side line product are shown; prop_m，minspec；Prop_m，maxspecRespectively representing the upper and lower limits of the index requirement, Prop_m，accuracyThe model prediction accuracy indicating this physical property; Δ F, Δ T, Δ P are the variable amplitudes and the associated inequality is the constraint on the optimization amplitude.

Step S3: and (3) optimizing and solving the physical logic reconstruction model by adopting a nonlinear solver to obtain all parameters representing required working conditions, on one hand, obtaining initial values required by the general disc optimization, and on the other hand, obtaining crude oil properties and device performance parameters adopted during optimization.

Specifically, an AI model of the atmospheric and vacuum device is taken as constraint, processed steady-state data is taken as input, a mathematical programming method is adopted to solve the optimized values of all variables when the difference values of the predicted values and the measured values of all the variables are minimum, the variable data obtained by the method realizes the correction of historical working condition data and real-time working condition data, and soft measurement of the variables is obtained, including crude oil properties and device characteristic parameters, so that the field working conditions are restored, the variable data obtained by the method meet the constraint conditions in the optimized model, and the optimization time can be reduced to be within 5 minutes by taking the set of data as the initial values of general disk optimization.

The physical logic reconstruction model is schematically as follows:

y_j＝f_j(x_cal，i，y_j′) j，j′∈J，j≠j′

x_cal，i＝f_i(x_cal，i′，y_j) j，j′∈I，i≠i′

y_j，L≤y_j≤y_j，U

x_i，L≤x_i，cal≤x_i，U

wherein x is_msd，iThe data of the field measuring instrument after data processing; x is the number of_syserr，iThe system error of the instrument i; x is the number of_cal，iData output for the final calculation; x is the number of_cal，i’Is in addition to x_cal，iThe data output by the last calculation; w is a_iIs the weight of meter i; m_axiIs the maximum value of the instrument i; min_iIs the minimum value of instrument i, y_jAre other variables; including output variables and intermediate variables other than field measurements; y is_j，For other variables, including in addition to field measurements and y_iOutput variables other than and intermediate variables; the function f represents the relation between all x and y and mainly reflects physical models of material balance, energy balance, phase balance, pressure balance, heat transfer and the like; j represents the set of 1 to J, i.e. 1, 2, 3, …, J; i represents the set of 1 to I, i.e. 1, 2, 3, …, I; y is_j，LAnd y_j，UAre each y_jLower and upper limits of (2), x_i，LAnd x_i，UAre respectively x_iLower and upper limits of.

Step S4: the optimization variables are classified into 3 classes of low, medium and high frequencies according to the adjustment frequency, as shown in table 1, the adjustment levels are classified into 1, 2 and 3 levels, and the variable classes to be adjusted are selected according to the adjustment levels, as shown in table 2. According to the variable attribute, the invention sets different adjusting frequencies: for parameters which are not easy to be adjusted repeatedly and have large influence on the system, the adjustment is reduced as much as possible, for example, the change of the outlet temperature of the heating furnace has large influence on the change of the whole system, so that low-frequency adjustment is set; for some operating parameters which need to be frequently adjusted or are assisted by an automatic means, the high-frequency adjustment can be carried out, such as the side line extraction temperature, the steam blowing at the bottom of the atmospheric tower and the like; there are also some operations which are medium frequency regulation, such as the operation of blowing steam at the bottom of a vacuum tower and the flow rate of middle-section reflux. For the level 1 adjustment, only the high frequency variable can be adjusted, the medium frequency variable and the low frequency variable are not adjusted as much as possible, for the level 2 adjustment, the high frequency variable and the medium frequency variable can be adjusted, the low frequency variable is not adjusted as much as possible, and for the level 3 adjustment, the high frequency, the medium frequency and the low frequency can be adjusted, wherein for the parameters needing to be adjusted, the adjustment range is preferably controlled within 20% each time. Aiming at different variables, different frequency adjustments and different adjustment strategies are carried out, and therefore long-period stable operation of the system is achieved.

TABLE 1 variable optimization Categories

TABLE 2 optimization variable adjustment strategy

Step S5: and sequentially defining the variables and the optimization models involved in the steps according to an optimization platform programming rule, assigning the calculated initial values to the variables, and selecting a nonlinear solver and a formulated optimization strategy to carry out optimization solution on the models. The optimization platform can be a GAMS modeling platform, the solver is based on a Generalized Reduced Gradient (GRG) algorithm, the influence of the change of all variables on all potential bottlenecks of the system is comprehensively evaluated when the operating variables are adjusted at each step, and the feasibility of the optimization direction is ensured. In each step of optimization direction search, the influence of variables of a plurality of atmospheric and vacuum devices, including the treatment capacity of each tower, the outlet temperature of the atmospheric furnace, the outlet temperature of the vacuum furnace, the stripping steam flow at each position, the reflux quantity at each position and the like on the optimization target and the potential bottleneck of the whole system is measured and calculated uniformly, namely the variables are adjusted simultaneously, and the adjustment also needs to meet the constraint relation among the variables, so that the optimal optimization direction and amplitude meeting the feasible operation condition of the device are estimated and generated. Meanwhile, in order to ensure that the optimization scheme is feasible, five types of field constraints are simultaneously measured and calculated in each step of optimization direction search based on a high-precision ANN model to determine whether the performance requirements of the system are met, wherein the performance requirements comprise the tray hydraulics performance of the fractionating tower, the heat transfer capacity of a heat exchanger, the operable range of conveying equipment, the load bottleneck of a heating furnace and the high-temperature coking tendency of the equipment.

The method comprises the steps of selecting key variables which have large influence on the performance of the device through sensitivity analysis, dividing a solving space into a plurality of parts through segmenting the key variables, carrying out optimization solving in each space, comparing results of all the spaces, and selecting an optimal result, so that the optimization result is prevented from falling into local optimization.

Step S6: self-learning and self-correcting of the model. When the field equipment performance and the processed raw materials are greatly changed, the working conditions may exceed the range covered by the AI model, and the prediction effect of the model outside the range is poor. Therefore, before the model is used, the method also judges whether the current working condition exceeds the range of the model, if so, the working condition is enhanced in the range near the current working condition, part of samples are newly added, and the newly added part of sample information is gathered into the previous samples to carry out correction training on the model, so that the self-learning and self-correction of the model are realized. In addition, the current characteristics of the device, such as the plate effect of a tower plate, the pressure drop of a transfer line and the like, are obtained through the physical logic reconstruction calculation result, and the characteristic parameters of the device in the model are updated, so that the optimization model is adaptive to the characteristic change of the device.

The optimization constraint in the embodiment of the invention mainly considers five factors of material balance, energy balance, phase balance, heat transfer and equipment performance; the limitation of some adjustment ranges of the variables, such as the product index constraint range, needs to consider the redundancy caused by the model precision, the reflux variation in the key tower plate, the variation range of the three tower treatment capacity, the variation range of the side stream flow to the downstream device, the variation range of the branch flow, and the spray density range of each packing section of the decompression tower.

In this embodiment, the result after the general disk optimization mainly includes: the total benefit change before and after optimization, the product benefit change before and after optimization, the energy consumption change before and after optimization, the operation parameter change before and after optimization, the property change of each side product before and after optimization, the hydrodynamics change of the tower plates before and after optimization, and the like.

In this embodiment, not only can an optimization scheme that is considered globally be proposed for the current state of the device, but also the operation cycle of the whole device can be looked at, and the long-cycle stable operation of the device can be optimized. For example: in short, the reduction of the washing oil amount of the vacuum tower is beneficial to improving the extraction rate of the vacuum tower. However, an excessively small amount of washing oil tends to cause coking of the lower portion of the vacuum tower in the long term, and the performance of the vacuum tower equipment is lowered, resulting in a greater economic loss. The general disc optimization technology evaluates the coking tendency of the high-temperature position of the equipment through an ANN model, and avoids the major operation risk that the short-term optimization benefit is excessively excavated to influence the long-term operation.

The invention utilizes the mechanism model to generate a large amount of data so as to make up the defect that the production data range is narrow and the production device cannot be comprehensively simulated, and the model established by the method can express the characteristics of the production device as the mechanism model and has higher calculation speed than the mechanism model in the aspect of calculation; and the autonomous learning device has real-time characteristics, and the model is corrected according to the real-time characteristics of the device, so that the accuracy of the model is improved.

The general optimization of the invention is based on the result of physical logic reconstruction, the result of physical logic reconstruction is all parameter data representing the current working condition, the result is used as the initial value of general optimization, and crude oil property and device performance parameter data are provided for general optimization, thereby ensuring that the optimization result is obtained quickly and accurately.

The invention determines the optimization direction and amplitude according to the restored field working condition, avoids unhooking from the field working condition, considers the influence of adjusting a plurality of variables on the benefit, accords with the real operation logic of atmospheric and vacuum pressure, can search out the optimal parameter optimization direction in time, maximizes the economic benefit, not only can provide an optimization scheme which is considered globally for the current state of the device, but also can put eyes on the whole device operation period, optimizes the long-period stable operation of the device, and has higher practicability for production.

The foregoing description shows and describes several preferred embodiments of the invention, but as aforementioned, it is to be understood that the invention is not limited to the forms disclosed herein, but is not to be construed as excluding other embodiments and is capable of use in various other combinations, modifications, and environments and is capable of changes within the scope of the inventive concept as expressed herein, commensurate with the above teachings, or the skill or knowledge of the relevant art. And that modifications and variations may be effected by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.