A production device rescheduling method and system based on crude oil property change

Abstract

The application discloses a production device rescheduling method based on crude oil property change, which comprises the following steps: establishing a processing capacity prediction model; establishing an output prediction model; inversely optimizing processing prediction according to output prediction, combining preset material balance information and sales demand information, taking the optimized processing prediction as actual processing capacity of corresponding production device, and completing rescheduling of the production device. According to the production device rescheduling method based on crude oil property change, the method can automatically update device prediction results and realize production rescheduling by modifying crude oil information, and ensures that final device processing capacity and key product output meet production and market requirements.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of chemical engineering and computer science, and in particular to a method and system for rescheduling production units based on changes in crude oil properties. Background Technology

[0002] Crude oil is a key raw material for refining petroleum products, and fluctuations in its properties and composition have a profound impact on the entire refining process. The development of the petrochemical industry makes accurate prediction and optimization of the impact of crude oil fluctuations on all production units particularly important. Parameters such as the API (American Petroleum Institute Heavy Crude Oil Standard), density, sulfur content, and acid value of different crude oil mixtures directly affect the processing performance and output quality of various refining units, from atmospheric distillation units to catalytic cracking units and hydrocracking units. Changes in market demand and fluctuations in equipment status further increase the complexity of crude oil fluctuation management.

[0003] With advancements in petrochemical technology and increased computing power, some enterprises have gradually introduced optimization software based on mathematical models, such as linear programming and integer optimization algorithms. These methods require the prior establishment of a comprehensive mathematical optimization model and the acquisition of optimization results through a solver, which are then used to adjust the load and operating parameters of various units to optimize overall production efficiency. However, these methods still lack sufficient precision and flexibility, especially when facing fluctuations in crude oil supply, changes in market demand, and equipment failures, making it even more difficult to solve the problem of rapid rescheduling for enterprises.

[0004] Traditional production scheduling methods are often limited to single units, making it difficult to comprehensively consider the combined impact of crude oil fluctuations on the entire production process. Traditional linear models struggle to handle these complex factors, often resulting in inaccurate scheduling outcomes or even failure to achieve feasible results, thus affecting overall production efficiency and economic benefits. Furthermore, the overall unit scheduling of oil refineries typically relies on experienced operators and simple linear models. These models are primarily based on static crude oil property data and basic equipment operating parameters, such as temperature, pressure, and flow rate in atmospheric distillation units. However, this approach often struggles to cope with the complex effects of crude oil property fluctuations and changes in market demand, making it difficult to make global decisions in a short period. Therefore, a rapid and comprehensive rescheduling method is urgently needed. Summary of the Invention

[0005] One objective of this invention is to propose a fast and globally controllable method for rescheduling production units based on changes in crude oil properties.

[0006] A method for rescheduling production units based on changes in crude oil properties includes the following steps:

[0007] Based on historical information on crude oil properties and historical information on unit throughput, a throughput prediction model is established.

[0008] Based on historical information on crude oil properties, historical information on unit processing capacity, information on unit location, and historical information on unit output, an output prediction model is established.

[0009] Obtain crude oil property information, input it into the processing volume prediction model, and obtain the processing volume prediction;

[0010] The predicted output and the device orientation information are input into the output prediction model to obtain the predicted output.

[0011] Based on the output forecast, combined with the preset material balance information and sales demand information, the processing forecast is optimized in reverse. The optimized processing forecast is used as the actual processing capacity of the corresponding production unit, and the rescheduling of the production unit is completed.

[0012] The production unit rescheduling method based on changes in crude oil properties proposed in this invention can automatically update unit forecasting results and achieve production rescheduling by modifying crude oil information, ensuring that the final unit throughput and key product output meet production and market requirements. Driven by crude oil properties, this method establishes a production unit and rescheduling model, achieving unit forecasting and rescheduling optimization. It can more effectively meet sales demand and product production plan requirements even when crude oil supply changes, providing effective technical support for production.

[0013] In addition, according to the present invention, it may also have the following additional technical features:

[0014] Furthermore, the historical information on crude oil properties includes blending ratio, API value, density, sulfur content, and acid value.

[0015] Furthermore, the processing capacity prediction model is a neural network model established using historical information on crude oil properties as input and historical information on the device's processing capacity as output, with the objective function being:

[0016]

[0017] Where n is the number of samples, o is the number of devices, and y true,i,j y represents the historical processing volume of device j in sample i. pred,i,j Let m be the predicted processing capacity of device j in sample i. i,j The mask value for device j in sample i is used to filter y. true,i,j Elements greater than 0 in the set.

[0018] Furthermore, the device destination information is converted into a binary form by one-hot encoding based on the destination of the device output and the classification data.

[0019] Furthermore, the output prediction model uses historical information on crude oil properties, historical information on unit processing capacity, and information on unit movement as inputs, and historical information on unit output as output. The established neural network model has the following objective function:

[0020]

[0021] Where n is the number of samples, o is the number of devices, and y true,i,j For the historical device output of device j in sample i, y pred,i,j For the predicted device output of device j in sample i, m i,j The mask value for device j in sample i is used to filter y. true,i,j Elements greater than 0 in the set.

[0022] Furthermore, the reverse optimization process for predicting quantities is performed using the gradient descent method.

[0023] Furthermore, the step of reverse optimization processing the predicted quantity includes:

[0024] (1) Define the objective function

[0025] Constraints include material balance information and sales demand information:

[0026] Ax = b

[0027] Where A is the constraint matrix, s1 represents the output quantity of all devices involved in material balance, r1 represents the quantity of materials involved in material balance, and x represents the output vector. b is the material vector.

[0028] Cy≥d

[0029] Where C is the constraint matrix, s2 represents the quantity produced by all devices involved in sales demand, r2 represents the quantity of products involved in sales demand, and y represents the output vector. d is the product vector.

[0030] Objective function:

[0031]

[0032] Where, p i To predict the processing volume, d i 'o' represents the demand, and 'o' represents the number of devices.

[0033] (2) Calculate the gradient of the objective function.

[0034] Determine the objective function:

[0035]

[0036] For each p i Find the partial derivative:

[0037]

[0038] The gradient vector is:

[0039]

[0040] (3) Iteratively update the processing parameters until the objective function converges and meets the convergence criterion. The iterative process is as follows: Let the initial parameter vector be θ. (0)

[0041]

[0042] Where η is the learning rate;

[0043] The convergence criterion is:

[0044]

[0045] Where, ∈<10 ―3 .

[0046] Furthermore, it also includes:

[0047] The future processing capacity of the unit is obtained by using a processing capacity prediction model, and the future output capacity of the unit is predicted by using an output prediction model. Then, based on the material balance information and sales demand information, reverse optimization is performed using the future processing capacity and future output capacity of the unit.

[0048] Another object of the present invention is to provide a production unit rescheduling system based on changes in crude oil properties using the above-described method, comprising:

[0049] The throughput prediction model building module establishes a throughput prediction model based on historical information on crude oil properties and historical information on unit throughput.

[0050] The output prediction model building module establishes an output prediction model based on historical information on crude oil properties, historical information on unit processing capacity, information on unit location, and historical information on unit output.

[0051] The processing forecast acquisition module obtains crude oil property information, inputs it into the processing forecast model, and obtains the processing forecast.

[0052] The output forecast acquisition module inputs the processed forecast and equipment orientation information into the output forecast model to obtain the output forecast.

[0053] The reverse optimization and rescheduling module optimizes the processing forecast based on the output forecast, combined with preset material balance information and sales demand information. The optimized processing forecast is then used as the actual processing capacity of the corresponding production unit, thus completing the rescheduling of that production unit.

[0054] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0055] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0056] Figure 1 This is a flowchart illustrating the first embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be thorough and complete.

[0058] Please see Figure 1 The first embodiment of the present invention proposes a production unit rescheduling method based on changes in crude oil properties, which includes the following steps.

[0059] S1. Establish a throughput prediction model based on historical information on crude oil properties and historical information on unit throughput.

[0060] In this embodiment, the historical information on crude oil properties includes blending ratio, API value, density, sulfur content, and acid value.

[0061] Specifically, the crude oil properties and unit throughput model uses crude oil blending ratio, blended oil density, blended oil API, blended oil sulfur content, and blended oil residual carbon as inputs, and unit throughput as output. The units involved in the throughput include: I. Atmospheric and vacuum distillation unit, II. Atmospheric and vacuum distillation unit, delayed coking unit, continuous reforming unit, I. Catalytic cracking unit, II. Catalytic cracking unit, I. Gas separation unit, II. Gas separation unit, I. Hydrogenation unit, II. Hydrogenation unit, III. Hydrogenation unit, reforming pre-hydrogenation unit, hydrocracking unit, IV. Hydrogenation unit, jet fuel hydrogenation unit, and aromatics unit.

[0062] Taking crude oil property data from the past two years as an example, the processing volume prediction model is a neural network model established with historical crude oil property information as input and historical processing volume information of the unit as output. The objective function is:

[0063]

[0064] Where n is the number of samples, O is the number of devices, and y true,i,j y represents the historical processing volume of device j in sample i. pred,i,j Let m be the predicted processing capacity of device j in sample i. i,j The mask value for device j in sample i is used to filter y. true,i,j Elements greater than 0 in the set.

[0065] Furthermore, the steps to build a neural network model include:

[0066] (1) Data preprocessing: First, crude oil property data (such as API, density, sulfur content, acid value, etc.), unit throughput data, and product flow data are collected. Then, missing values, outliers, and negative values ​​are processed, missing values ​​are filled using interpolation, and outliers are statistically analyzed and corrected to clean the data.

[0067] (2) Network Structure Selection: Based on the complexity of the crude oil dispatching system and the characteristics of the data, a suitable BP neural network structure was designed. The input layer contains all preprocessed crude oil property data, unit processing volume data, and direction data. The ReLU function was selected as the activation function for the intermediate layers to improve the model's nonlinear fitting ability and training speed. The number of nodes in the output layer is consistent with the target variable to be predicted (such as the processing volume and output of each unit), and the predicted values ​​are output.

[0068] (3) Model Training: Supervised learning was performed using a large amount of collected historical data to train the BP neural network model. The backpropagation algorithm was used to optimize the model parameters and minimize the prediction error. During training, mini-batch gradient descent was used to balance training speed and stability. To prevent overfitting, the Dropout regularization method was used to randomly drop some neurons to improve the model's generalization ability.

[0069] (4) Model Validation: After training, the model performance is evaluated using an independent validation dataset. The mean squared error (MSE) metric is used to measure the model's predictive accuracy. Based on the validation results, hyperparameters are tuned, such as adjusting the learning rate, the number of hidden layer nodes, and the regularization parameter, to further improve model performance.

[0070] S2. Based on historical information on crude oil properties, historical information on unit processing capacity, information on unit location, and historical information on unit output, establish an output prediction model.

[0071] In this embodiment, the device destination information is converted into a binary form by one-hot encoding based on the destination of the device output and the classification data.

[0072] Specifically, the crude oil properties, unit throughput, direction, and unit output model uses crude oil blending ratio, blended oil density, blended oil API, blended oil sulfur content, blended oil residual carbon, unit throughput, and direction as inputs, and unit output as output. The units involved in the unit throughput, direction, and unit output model include: I. Atmospheric and vacuum distillation unit, II. Atmospheric and vacuum distillation unit, delayed coking unit, continuous reforming unit, I. Catalytic cracking unit, II. Catalytic cracking unit, I. Gas separation unit, II. Gas separation unit, I. Hydrogenation unit, II. Hydrogenation unit, III. Hydrogenation unit, reforming pre-hydrogenation unit, hydrocracking unit, IV. Hydrogenation unit, jet fuel hydrogenation unit, and aromatics unit.

[0073] Specifically, the output prediction model uses historical information on crude oil properties, historical information on unit processing capacity, and information on unit movement as inputs, and historical information on unit output as output. The established neural network model has the following objective function:

[0074]

[0075] Where n is the number of samples, o is the number of devices, and y true,i,j For the historical device output of device j in sample i, y pred,i,j For the predicted device output of device j in sample i, m i,j The mask value for device j in sample i is used to filter y. true,i,j Elements greater than 0 in the set.

[0076] S3. Obtain crude oil property information, input it into the processing volume prediction model, and obtain the processing volume prediction.

[0077] S4. Input the processing forecast and device orientation information into the output forecast model to obtain the output forecast.

[0078] This embodiment also includes simulating the impact of crude oil delays on crude oil properties, with the following steps:

[0079] (1) Based on historical data and current forecasts, simulate the impact of delays in specific crude oil varieties on the overall crude oil blending properties;

[0080] (2) Use future crude oil properties data as model input to re-predict unit throughput and output.

[0081] It should be noted that the output of the prediction model includes the predicted values ​​of the processing capacity and output of the device, which are an important part of the production scheduling scheme, and these prediction values ​​provide data support for the adjustment of the production scheduling strategy.

[0082] S5. Based on the output forecast, combined with the preset material balance information and sales demand information, the processing forecast is optimized in reverse, and the optimized processing forecast is used as the actual processing quantity of the corresponding production unit to complete the rescheduling of the production unit.

[0083] In this embodiment, the reverse optimization process for predicting the quantity is performed using the gradient descent method.

[0084] Specifically, the steps for reverse optimization of the predicted values ​​include:

[0085] (1) Define the objective function

[0086] Constraints include material balance information and sales demand information:

[0087] Ax = b

[0088] Where A is the constraint matrix, s1 represents the output quantity of all devices involved in material balance, r1 represents the quantity of materials involved in material balance, and x represents the output vector. b is the material vector.

[0089] Cy≥d

[0090] Where C is the constraint matrix, s2 represents the quantity produced by all devices involved in sales demand, r2 represents the quantity of products involved in sales demand, and y represents the output vector. d is the product vector.

[0091] Objective function:

[0092]

[0093] Where, p i To predict the processing volume, d i 'o' represents the demand, and 'o' represents the number of devices.

[0094] (2) Calculate the gradient of the objective function.

[0095] Determine the objective function:

[0096]

[0097] For each p i Find the partial derivative:

[0098]

[0099] The gradient vector is:

[0100]

[0101] (3) Iteratively update the processing parameters until the objective function converges and meets the convergence criterion. The iterative process is as follows: Let the initial parameter vector be θ. (0)

[0102]

[0103] Where η is the learning rate;

[0104] The convergence criterion is:

[0105]

[0106] Where, ∈<10 ―3 .

[0107] The production unit rescheduling method based on changes in crude oil properties proposed in this invention can automatically update unit forecasting results and achieve production rescheduling by modifying crude oil information, ensuring that the final unit throughput and key product output meet production and market requirements. Driven by crude oil properties, this method establishes a production unit and rescheduling model, achieving unit forecasting and rescheduling optimization. It can more effectively meet sales demand and product production plan requirements even when crude oil supply changes, providing effective technical support for production.

[0108] For example, the crude oil dispatching process includes a crude oil tank unit using crude oil from tankers as raw material, an atmospheric and vacuum distillation unit using mixed crude oil from the crude oil tanks as raw material, a secondary unit using tank supply and direct unit supply as raw material, and a product tank unit using the output of the secondary units as raw material. The specific process is as follows: The crude oil tank unit uses crude oil from different crude oil tankers as raw material, mixing it with residual mixed oil in the tanks to create new mixed oil in the tanks; the atmospheric and vacuum distillation unit uses mixed crude oil from different crude oil tanks as raw material, dispatching one crude oil tank for primary refining and another for blending, converting it into light sludge oil, residue oil, wax oil, etc., which are then sent to various secondary units, intermediate material tanks, or product tanks; the secondary unit uses oil from various intermediate material tanks and direct unit supply materials as raw material, converting it into new outputs which are sent to other secondary units, intermediate material tanks, or product tanks; the product tank unit uses the output of the atmospheric and vacuum distillation unit or the secondary unit as raw material, and delivers the product when the tank is full.

[0109] Modeling of crude oil properties and unit throughput: The input is set as mixed crude oil, where the ratio of the three oils (Shengli Mixed: Busios: Sha Light) is 15.0:46.7:38.3. The density of the mixed crude oil is 0.88 g / ml, the API value is 29.3, the sulfur content is 0.904 wt%, and the carbon residue is 0.47 wt%. Then, the unit throughput is simulated. The comparison results of the calibrated values ​​and simulated values ​​of key parameters are shown in Table 1 (Comparison of Simulation Calculation Results of Unit Throughput).

[0110] Table 1

[0111]

[0112] Modeling of crude oil properties, unit throughput, trend, and unit output: The inputs are set as mixed crude oil, unit throughput, and trend. The ratio of the mixed crude oil is set as 15.0:46.7:38.3 for Shengli Mixed:Busios:Sha Light, with a density of 0.88 g / ml, an API value of 29.3, a sulfur content of 0.904 wt%, and a carbon residue of 0.47 wt%. The unit throughput is the predicted result of the crude oil properties and unit throughput model. The key trend is shown in Table 2 (Trend Display). Then, the unit output is simulated, and the comparison results of the calibrated values ​​and simulated values ​​of key parameters are shown in Table 3 (Comparison of Simulated Calculation Results of Unit Output).

[0113] Table 2

[0114]

[0115] Table 3

[0116]

[0117]

[0118] The tanker delays were simulated, and the throughput was modified to meet constraints: material balance and sales demand. Material balance included hydrogen balance, reformate balance, hydrotreating feed balance, wax oil balance, and residue oil balance; sales demand included gasoline, diesel, naphtha, benzene, etc., as detailed in Table 4 (Sales Demand). Then, gradient descent optimization was performed on the unit's throughput, and the optimized throughput results are shown in Table 5 (Unit Processing Capacity Optimization Results).

[0119] Table 4

[0120]

[0121] Table 5

[0122]

[0123]

[0124] By combining crude oil properties, equipment throughput, and flow data, the optimization method of this invention utilizes a BP neural network model for comprehensive prediction and optimization. In the event of unforeseen circumstances such as crude oil supply delays, dynamic optimization is achieved through simulation and process adjustment, ensuring the rationality and stability of production scheduling.

[0125] This embodiment has the following advantages:

[0126] 1. Performance optimization of crude oil processing units

[0127] By establishing a BP neural network model based on crude oil properties data and unit throughput, the throughput of each unit can be predicted, thereby improving the material allocation efficiency of the units. The BP neural network model, which combines crude oil properties, unit throughput, and flow data, can accurately predict the output of each unit, ensuring coordinated operation of all units during the production process.

[0128] 2. Material balance and meeting sales demand

[0129] By verifying the predicted unit output through material balance and sales demand, and using the gradient descent method to back-optimize the throughput, material balance during production can be ensured to meet sales needs. When crude oil supply is ahead of schedule or delayed, by simulating changes in crude oil properties and using a BP neural network model for prediction and optimization, the unit throughput can be quickly adjusted to ensure the continuity and stability of the production process.

[0130] 3. Energy System Modeling and Optimization

[0131] By establishing an energy system model, key operating parameters in oil refining units can be comprehensively analyzed, helping to determine optimal operating conditions. Optimizing unit throughput and output can reduce energy consumption, achieving energy conservation and emission reduction, and providing guidance for actual production operations.

[0132] 4. Flexibility in responding to supply delays

[0133] This method can solve the problem of production fluctuations caused by the uncertainty of crude oil arrival at the plant. It uses a neural network model to predict future plant throughput and output, and makes optimization adjustments to ensure stable production even if supply is ahead of schedule or delayed.

[0134] 5. Precise data processing

[0135] By performing one-hot encoding on the directional data, the neural network model can better process and utilize directional information, thereby improving the accuracy of predictions and the overall performance of the model.

[0136] The second embodiment of the present invention proposes a production unit rescheduling method based on changes in crude oil properties. The difference between this embodiment and the first embodiment is as follows.

[0137] This embodiment also includes:

[0138] The future processing capacity of the unit is obtained by using a processing capacity prediction model, and the future output capacity of the unit is predicted by using an output prediction model. Then, based on the material balance information and sales demand information, reverse optimization is performed using the future processing capacity and future output capacity of the unit.

[0139] A second embodiment of the present invention proposes a production unit rescheduling system based on changes in crude oil properties, comprising:

[0140] The throughput prediction model building module establishes a throughput prediction model based on historical information on crude oil properties and historical information on unit throughput.

[0141] The output prediction model building module establishes an output prediction model based on historical information on crude oil properties, historical information on unit processing capacity, information on unit location, and historical information on unit output.

[0142] The processing forecast acquisition module obtains crude oil property information, inputs it into the processing forecast model, and obtains the processing forecast.

[0143] The output forecast acquisition module inputs the processed forecast and equipment orientation information into the output forecast model to obtain the output forecast.

[0144] The reverse optimization and rescheduling module optimizes the processing forecast based on the output forecast, combined with preset material balance information and sales demand information. The optimized processing forecast is then used as the actual processing capacity of the corresponding production unit, thus completing the rescheduling of that production unit.

[0145] In this embodiment, the historical information on crude oil properties includes blending ratio, API value, density, sulfur content, and acid value.

[0146] Specifically, the crude oil properties and unit throughput model uses crude oil blending ratio, blended oil density, blended oil API, blended oil sulfur content, and blended oil residual carbon as inputs, and unit throughput as output. The units involved in the throughput include: I. Atmospheric and vacuum distillation unit, II. Atmospheric and vacuum distillation unit, delayed coking unit, continuous reforming unit, I. Catalytic cracking unit, II. Catalytic cracking unit, I. Gas separation unit, II. Gas separation unit, I. Hydrogenation unit, II. Hydrogenation unit, III. Hydrogenation unit, reforming pre-hydrogenation unit, hydrocracking unit, IV. Hydrogenation unit, jet fuel hydrogenation unit, and aromatics unit.

[0147] Taking crude oil property data from the past two years as an example, the processing volume prediction model is a neural network model established with historical crude oil property information as input and historical processing volume information of the unit as output. The objective function is:

[0148]

[0149] Where n is the number of samples, o is the number of devices, and y true,i,j y represents the historical processing volume of device j in sample i. pred,i,j Let m be the predicted processing capacity of device j in sample i. i,j The mask value for device j in sample i is used to filter y. true,i,j Elements greater than 0 in the set.

[0150] Furthermore, the steps to build a neural network model include:

[0151] (1) Data preprocessing: First, crude oil property data (such as API, density, sulfur content, acid value, etc.), unit throughput data, and product flow data are collected. Then, missing values, outliers, and negative values ​​are processed, missing values ​​are filled using interpolation, and outliers are statistically analyzed and corrected to clean the data.

[0152] (2) Network Structure Selection: Based on the complexity of the crude oil dispatching system and the characteristics of the data, a suitable BP neural network structure was designed. The input layer contains all preprocessed crude oil property data, unit processing volume data, and direction data. The ReLU function was selected as the activation function for the intermediate layers to improve the model's nonlinear fitting ability and training speed. The number of nodes in the output layer is consistent with the target variable to be predicted (such as the processing volume and output of each unit), and the predicted values ​​are output.

[0153] (3) Model Training: Supervised learning was performed using a large amount of collected historical data to train the BP neural network model. The backpropagation algorithm was used to optimize the model parameters and minimize the prediction error. During training, mini-batch gradient descent was used to balance training speed and stability. To prevent overfitting, the Dropout regularization method was used to randomly drop some neurons to improve the model's generalization ability.

[0154] (4) Model Validation: After training, the model performance is evaluated using an independent validation dataset. The mean squared error (MSE) metric is used to measure the model's predictive accuracy. Based on the validation results, hyperparameters are tuned, such as adjusting the learning rate, the number of hidden layer nodes, and the regularization parameter, to further improve model performance.

[0155] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0156] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for rescheduling production units based on changes in crude oil properties, characterized in that, Includes the following steps: Based on historical information on crude oil properties and historical information on unit throughput, a throughput prediction model is established. Based on historical information on crude oil properties, historical information on unit processing capacity, information on unit location, and historical information on unit output, an output prediction model is established. Obtain crude oil property information, input it into the processing volume prediction model, and obtain the processing volume prediction; The predicted output and the device orientation information are input into the output prediction model to obtain the predicted output. Based on the output forecast, combined with the preset material balance information and sales demand information, the processing forecast is optimized in reverse. The optimized processing forecast is used as the actual processing capacity of the corresponding production unit, and the rescheduling of the production unit is completed.

2. The production unit rescheduling method based on changes in crude oil properties according to claim 1, characterized in that, The historical information on crude oil properties includes blending ratio, API value, density, sulfur content, and acid value.

3. The production unit rescheduling method based on changes in crude oil properties according to claim 1, characterized in that, The processing capacity prediction model is a neural network model established using historical information on crude oil properties as input and historical information on the processing capacity of the device as output. The objective function is: Where n is the number of samples, o is the number of devices, and y true,i,j y represents the historical processing volume of device j in sample i. pred,i,j Let m be the predicted processing capacity of device j in sample i. i,j The mask value for device j in sample i is used to filter y. true,i,j Elements greater than 0 in the set.

4. The production unit rescheduling method based on changes in crude oil properties according to claim 1, characterized in that, The device trajectory information is converted into a binary form based on the destination of the device output and the classification data through one-hot encoding.

5. The production unit rescheduling method based on changes in crude oil properties according to claim 1, characterized in that, The output prediction model uses historical information on crude oil properties, historical information on unit processing capacity, and information on unit location as inputs, and historical information on unit output as output. The established neural network model has the following objective function: Where n is the number of samples, o is the number of devices, and y true,i,j For the historical device output of device j in sample i, y pred,i,j For the predicted device output of device j in sample i, m i,j The mask value for device j in sample i is used to filter y. true,i,j Elements greater than 0 in the set.

6. The production unit rescheduling method based on changes in crude oil properties according to claim 1, characterized in that, The reverse optimization process for predicting variables is performed using the gradient descent method.

7. The production unit rescheduling method based on changes in crude oil properties according to claim 1 or 6, characterized in that, The steps for reverse optimization of the predicted values ​​include: (1) Define the objective function Constraints include material balance information and sales demand information: Ax = b Where A is the constraint matrix, s1 represents the output quantity of all devices involved in material balance, r1 represents the quantity of materials involved in material balance, and x represents the output vector. b is the material vector. Cy≥d Where C is the constraint matrix, s2 represents the quantity produced by all devices involved in sales demand, r2 represents the quantity of products involved in sales demand, and y represents the output vector. d is the product vector. Objective function: Where, p i To predict the processing volume, d i 'o' represents the demand, and 'o' represents the number of devices. (2) Calculate the gradient of the objective function. Determine the objective function: For each p i Find the partial derivative: The gradient vector is: (3) Iteratively update the processing parameters until the objective function converges and meets the convergence criterion. The iterative process is as follows: Let the initial parameter vector be θ. (0) Where η is the learning rate; The convergence criterion is: Where, ∈<10 ―3 .

8. The production unit rescheduling method based on changes in crude oil properties according to claim 1, characterized in that, Also includes: The future processing capacity of the unit is obtained by using a processing capacity prediction model, and the future output capacity of the unit is predicted by using an output prediction model. Then, based on the material balance information and sales demand information, reverse optimization is performed using the future processing capacity and future output capacity of the unit.

9. A production unit rescheduling system based on changes in crude oil properties, characterized in that, include: The throughput prediction model building module establishes a throughput prediction model based on historical information on crude oil properties and historical information on unit throughput. The output prediction model building module establishes an output prediction model based on historical information on crude oil properties, historical information on unit processing capacity, information on unit location, and historical information on unit output. The processing forecast acquisition module obtains crude oil property information, inputs it into the processing forecast model, and obtains the processing forecast. The output forecast acquisition module inputs the processed forecast and equipment orientation information into the output forecast model to obtain the output forecast. The reverse optimization and rescheduling module optimizes the processing forecast based on the output forecast, combined with preset material balance information and sales demand information. The optimized processing forecast is then used as the actual processing capacity of the corresponding production unit, thus completing the rescheduling of that production unit.

Images