Chemical process flow analog simulation method and system based on big data

Abstract

The invention discloses a big data-based chemical process flow analog simulation method and system, and relates to the field of chemical processes, and the method comprises the steps: receiving an intervention target set by a user, collecting historical data of a target chemical process flow, and carrying out the preprocessing to form a historical data set; training an initial simulation model taking the process parameters as input and taking the historical reference indexes as output by utilizing the historical data set, analyzing the historical data set by adopting a causal discovery algorithm, and constructing a causal graph of the causal relationship between the process variables; based on the intervention target, identifying candidate causal paths from the controllable process parameters to the intervention target from the causal graph of the causal relationship between the process variables; and reconstructing the initial simulation model according to the causal diagram and the candidate causal paths to obtain a causal diagram simulation model. According to the method, deepening from data driving to causal inference is realized, and the interpretability and optimization accuracy of chemical process simulation are improved.

Description

Technical Field

[0001] This invention relates to the field of chemical processes, and in particular to a method and system for simulating chemical process flows based on big data. Background Technology

[0002] In the field of chemical processes, process simulation technology based on big data and artificial intelligence has become an important development direction for achieving process optimization and operation guidance. Traditional methods mainly use machine learning algorithms to train on historical process data and build a black box prediction model between input parameters and output performance. This model can effectively handle complex nonlinear relationships and achieve rapid prediction within the known operating conditions, providing certain support for process status monitoring and parameter adjustment.

[0003] Existing data-driven simulation methods still have significant limitations in supporting in-depth process optimization. Because the models rely on the statistical correlation between variables for prediction, they lack an explicit expression of the process mechanism, resulting in poor model interpretability. When the prediction results deviate from expectations, operators find it difficult to identify the key parameters affecting system performance and their causal paths, making it impossible to formulate targeted process adjustment strategies. Black-box models cannot effectively answer counterfactual reasoning questions such as how the target index will change if a certain parameter is changed, thus limiting their application in fine-grained process optimization and decision support. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a big data-based chemical process simulation method to solve the problem of difficulty in generating optimization strategies caused by the lack of causal mechanism expression in existing data-driven simulation models.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a chemical process simulation method based on big data, which includes receiving an intervention target set by a user, collecting historical data of the target chemical process, and preprocessing it to form a historical dataset. An initial simulation model with process parameters as input and historical benchmark indicators as output is trained using historical datasets. A causal discovery algorithm is then used to analyze the historical datasets and construct a causal graph of causal relationships among process variables. Based on the intervention target, candidate causal paths from controllable process parameters to the intervention target are identified from the causal graph of causal relationships among process variables; Based on the cause-effect graph and candidate causal paths, the initial simulation model is reconstructed to obtain the cause-effect graph simulation model. The process parameters are input into the cause-effect graph simulation model for simulation to obtain the simulation results of the intervention target. The candidate causal paths are analyzed using a counterfactual explanation algorithm to generate a counterfactual explanation report. Based on the counterfactual explanation report, the optimized combination of process parameters is obtained. The optimized combination of process parameters is then input into the cause-effect graph simulation model for verification simulation to obtain the improvement results of the intervention target.

[0007] As a preferred embodiment of the big data-based chemical process simulation method of this invention, the method includes the following steps: receiving user-defined intervention targets, collecting historical data of the target chemical process, and preprocessing the data to form a historical dataset. By receiving user-defined intervention targets aimed at improving product yield; Collect temperature, pressure, and flow data from the distributed control unit in the target chemical process flow; collect product purity data from the laboratory information management unit; and collect valve opening data from the manufacturing execution unit. The isolated forest algorithm is used to identify and remove outliers in temperature, pressure, flow, product purity, and valve opening data. The K-nearest neighbor algorithm is used to fill in missing values ​​in these data, and then Z-Score standardization is performed. The processed Z-Score standardized temperature data, pressure data, flow data, product purity data, and valve opening data are timestamped and integrated to form a historical dataset.

[0008] As a preferred embodiment of the big data-based chemical process simulation method of this invention, the method includes the following steps: training an initial simulation model using historical datasets with process parameters as input and historical benchmark indicators as output. Reaction temperature, reaction pressure, and feed flow rate were extracted from historical datasets as process parameters, and product yield was extracted as a historical benchmark indicator. The extracted process parameters and historical benchmark indicators are divided into training set and test set. The process parameters in the training set are used as input features and the historical benchmark indicators in the training set are used as output labels. The gradient boosting decision tree algorithm is used to train the initial simulation model. The prediction accuracy of the initial simulation model is evaluated on the test set. Based on the evaluation results on the test set, it is determined that the initial simulation model has reached the predetermined accuracy requirement, and the trained initial simulation model is obtained.

[0009] As a preferred embodiment of the big data-based chemical process simulation method of this invention, the method includes the following steps: Analyzing historical datasets using a causal discovery algorithm to construct a causal graph of causal relationships among process variables. Process variables selected from historical datasets for causal analysis include reaction temperature, reaction pressure, feed flow rate, and product yield. For process variables such as reaction temperature, reaction pressure, feed flow rate, and product yield, the NOTEARS causal discovery algorithm was used to calculate the causal strength score between each pair of process variables. Based on the causal strength score between process variables, a causal strength threshold is set, and causal connections with causal strength scores exceeding the causal strength threshold are retained. By combining the retained causal connections, a causal graph is constructed that represents the causal relationships between process variables, with process variables as nodes and directed edges representing causal relationships.

[0010] As a preferred embodiment of the big data-based chemical process simulation method of this invention, the method includes the following steps: Based on the intervention target, identifying candidate causal paths from controllable process parameters to the intervention target from the causal graph of causal relationships between process variables, including the following steps: Based on the intervention objective, determine the position of the intervention objective at the process variable node of product yield; From the causal diagram of the causal relationships between process variables, locate the process variable nodes corresponding to the controllable process parameters of reaction temperature, reaction pressure, and feed flow rate as the starting point; In the causal graph of the causal relationship between process variables, depth-first search is used to search for the directed path of the process variable node corresponding to the process variable node that achieves the product yield intervention target, starting from the process variable node corresponding to the controllable process parameters such as reaction temperature, reaction pressure, and feed flow rate. The directed path from the controllable process parameter node to the intervention target node is identified as a candidate causal path from the controllable process parameter to the intervention target.

[0011] As a preferred embodiment of the big data-based chemical process simulation method of this invention, the following steps are included: Based on the causal graph and candidate causal paths, the initial simulation model is reconstructed to obtain a causal graph simulation model; process parameters are input into the causal graph simulation model for simulation to obtain the simulation results of the intervention target; and a counterfactual explanation algorithm is used to analyze the candidate causal paths and generate a counterfactual explanation report. Based on the structure of the causal graph of the causal relationships between process variables, a graph attention network skeleton is constructed, and the parameter knowledge of the initial simulation model is transferred to the graph attention network skeleton. Based on the candidate causal paths from controllable process parameters to the intervention target, enhance the weights of the edges corresponding to the candidate causal paths in the graph attention network skeleton; The graph attention network skeleton with parameter transfer and weight enhancement was trained and fine-tuned using historical datasets to obtain a causal graph simulation model. Input a set of reaction temperature values, reaction pressure values, and feed flow rate values ​​as process parameters into the cause-effect graph simulation model; The cause-effect graph simulation model calculates based on the input process parameters and outputs the simulation results of product yield. A counterfactual interpretation algorithm is used to calculate the contribution of each candidate causal path to the simulation results of the intervention target for each candidate causal path from controllable process parameters to the intervention target. Based on the contribution of each candidate causal path, an analysis report describing the path and quantifying its impact is generated as a counterfactual explanation report.

[0012] As a preferred embodiment of the big data-based chemical process simulation method of this invention, the following steps are included: obtaining an optimized combination of process parameters based on a counterfactual interpretation report, inputting the optimized combination of process parameters into a cause-effect graph simulation model for verification simulation, and obtaining the improvement results of the intervention target: Based on the contribution ranking of each candidate causal path in the counterfactual explanation report, the candidate causal path with the highest contribution is selected; Based on the counterfactual suggestions corresponding to the candidate causal path with the highest contribution, determine the specific adjustment direction and magnitude of the reaction temperature setpoint, reaction pressure setpoint, and feed flow rate setpoint; The adjusted reaction temperature setpoint, reaction pressure setpoint, and feed flow rate setpoint are combined into an optimized combination of process parameters. The optimized process parameter combination is input into the cause-effect graph simulation model for verification simulation. The cause-effect graph simulation model outputs the product yield simulation value based on the optimized process parameter combination.

[0013] Secondly, the present invention provides a chemical process simulation system based on big data, including a data aggregation module that receives user-defined intervention targets, collects historical data of the target chemical process, and preprocesses the data to form a historical dataset. The path recognition module uses historical datasets to train an initial simulation model with process parameters as input and historical benchmark indicators as output. It then uses a causal discovery algorithm to analyze the historical datasets and construct a causal graph of causal relationships between process variables. The causal constraint module, based on the intervention target, identifies candidate causal paths from controllable process parameters to the intervention target from the causal graph of causal relationships between process variables; The simulation and deduction module reconstructs the initial simulation model based on the cause-effect graph and candidate causal paths to obtain the cause-effect graph simulation model. The process parameters are input into the cause-effect graph simulation model for simulation to obtain the simulation results of the intervention target. The counterfactual explanation algorithm is used to analyze the candidate causal paths and generate a counterfactual explanation report. The verification module, based on the counterfactual explanation report, obtains the optimized combination of process parameters, inputs the optimized combination of process parameters into the cause-effect graph simulation model for verification simulation, and obtains the improvement results of the intervention target.

[0014] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the big data-based chemical process simulation method of the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the big data-based chemical process simulation method of the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by receiving the intervention target set by the user and collecting preprocessed historical data, after training the initial simulation model, a causal graph between process variables is constructed using a causal discovery algorithm, and candidate causal paths from controllable parameters to the intervention target are identified. Based on the causal graph, a causal graph simulation model is obtained. Predictive results are obtained through simulation, and the contribution of each path is analyzed using a counterfactual interpretation algorithm to generate an interpretation report. Based on the report, the combination of process parameters is optimized, and the improved results are obtained through simulation verification. This realizes the deepening from data-driven to causal inference, and improves the interpretability and optimization accuracy of chemical process simulation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a chemical process simulation method based on big data.

[0019] Figure 2 This is a schematic diagram of a chemical process simulation system based on big data.

[0020] Figure 3 This is a flowchart of the initial simulation model training process.

[0021] Figure 4 This is a schematic diagram for identifying candidate causal paths. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a chemical process simulation method based on big data, including the following steps: S1. Receive the intervention target set by the user, collect historical data of the target chemical process, and preprocess it to form a historical dataset.

[0026] S1.1, Interventions aimed at improving product yield by receiving user-defined parameters.

[0027] Furthermore, users can select product yield as the optimization target from a list of preset process performance indicators through the human-computer interaction interface, or directly enter the instruction to improve product yield in the target input box, thereby setting the intervention target to improve product yield.

[0028] S1.2 Collect temperature, pressure, and flow data from the distributed control unit in the target chemical process flow; collect product purity data from the laboratory information management unit; and collect valve opening data from the manufacturing execution unit.

[0029] Furthermore, it collects reaction temperature data, reaction pressure data, and feed flow data recorded by the distributed control unit in the target chemical process flow, product purity data recorded by the laboratory information management unit, and valve opening data recorded by the manufacturing execution unit.

[0030] S1.3. The isolated forest algorithm is used to identify and remove outliers in temperature data, pressure data, flow data, product purity data, and valve opening data. The K-nearest neighbor algorithm is used to fill in the missing values ​​in temperature data, pressure data, flow data, product purity data, and valve opening data, and then Z-Score standardization is performed.

[0031] Furthermore, the isolated forest algorithm is used to identify and remove outliers in the reaction temperature, reaction pressure, feed flow rate, product purity, and valve opening data; the K-nearest neighbor algorithm is used to fill in missing values ​​in the reaction temperature, reaction pressure, feed flow rate, product purity, and valve opening data; and the reaction temperature, reaction pressure, feed flow rate, product purity, and valve opening data that have undergone outlier removal and missing value filling are subjected to Z-Score standardization to normalize all data to the same dimension.

[0032] S1.4. The processed Z-Score standardized temperature data, pressure data, flow data, product purity data, and valve opening data are timestamped and integrated to form a historical dataset.

[0033] Furthermore, for the reaction temperature data, reaction pressure data, feed flow rate data, product purity data, and valve opening data that have undergone Z-Score standardization, a linear interpolation method is used to align the timestamps of all data to a uniform frequency, such as one data point per minute. The timestamp-aligned reaction temperature data, reaction pressure data, feed flow rate data, product purity data, and valve opening data are then integrated to form a historical dataset.

[0034] S2. Use historical datasets to train an initial simulation model with process parameters as input and historical benchmark indicators as output.

[0035] S2.1 Extract reaction temperature, reaction pressure, and feed flow rate from historical datasets as process parameters, and extract product yield as a historical benchmark indicator.

[0036] Furthermore, reaction temperature, reaction pressure, and feed flow rate data columns are extracted from historical datasets as process parameter data, and product yield data columns are extracted as historical benchmark index data.

[0037] S2.2. The extracted process parameters and historical benchmark indicators are divided into training set and test set. The process parameters in the training set are used as input features and the historical benchmark indicators in the training set are used as output labels. The gradient boosting decision tree algorithm is used to train the initial simulation model.

[0038] Furthermore, the extracted reaction temperature, reaction pressure, and feed flow rate data columns are used as process parameter data, and the product yield data column is used as historical benchmark index data. These are divided into training and testing sets in chronological order. The reaction temperature, reaction pressure, and feed flow rate data in the training set are used as input features, and the product yield data in the training set is used as output labels. A gradient boosting decision tree algorithm is employed to train an initial simulation model by minimizing the mean square error between the predicted product yield and the actual product yield.

[0039] It should be noted that the gradient boosting decision tree algorithm trains the initial simulation model by iteratively constructing a series of decision trees. In each iteration, the algorithm fits the prediction residual of the current initial simulation model on the product yield data in the training set, and adds the prediction results of the new tree to the existing initial simulation model with a certain learning rate, gradually reducing the prediction error of the product yield until the loss function of the initial simulation model on the training set converges.

[0040] S2.3. Evaluate the prediction accuracy of the initial simulation model on the test set. Based on the evaluation results on the test set, determine whether the initial simulation model has reached the predetermined accuracy requirement and obtain the trained initial simulation model.

[0041] Furthermore, on the test set, the trained initial simulation model is used to predict the product yield based on the reaction temperature data, reaction pressure data, and feed flow rate data in the test set, and the root mean square error between the predicted product yield and the actual product yield in the test set is calculated as an evaluation index; when the root mean square error is lower than the preset threshold according to the prediction error range allowed by the process, it is confirmed that the initial simulation model has reached the predetermined accuracy requirement, and the trained initial simulation model is obtained.

[0042] S3. Use a causal discovery algorithm to analyze the historical dataset and construct a causal graph of the causal relationships between process variables.

[0043] S3.1 Select process variables such as reaction temperature, reaction pressure, feed flow rate, and product yield from historical datasets for causal analysis.

[0044] Furthermore, reaction temperature, reaction pressure, feed flow rate, and product yield data were selected from historical datasets as process variables for causal analysis.

[0045] S3.2 For the process variables of reaction temperature, reaction pressure, feed flow rate, and product yield, the NOTEARS causal discovery algorithm is used to calculate the causal strength score between each pair of process variables.

[0046] Furthermore, the selected reaction temperature, reaction pressure, feed flow rate, and product yield data were analyzed using the NOTEARS causal discovery algorithm. The NOTEARS algorithm calculates the causal strength score between each pair of process variables by solving a constrained optimization problem. The goal of the optimization problem is to minimize the loss function, which includes a data fitting term and an L1 regularization term. The constraint condition is the acyclicity constraint of the directed acyclic graph. After optimization, a weighted adjacency matrix is ​​output, and each element in the weighted adjacency matrix is ​​the causal strength score between the corresponding pair of process variables.

[0047] The loss function expression for the NOTEARS algorithm is: ; in, To minimize the loss function, For the standardized historical observation data matrix, For regularization terms, The regularization coefficient is . The prediction error matrix, For the Frobenius norm, The total number of valid observation samples in the historical dataset The expression for the acyclicity constraint of a directed acyclic graph is: ; in, For a weighted adjacency matrix, This is the weight matrix. For prior knowledge of the weighted adjacency matrix, This represents the total number of nodes in the cause-effect graph. The standardized historical observation data matrix expression is as follows: ; in, This refers to the specific value of the reaction temperature recorded during the first observation. This refers to the specific value of the reaction pressure recorded during the first observation. This refers to the specific product yield value corresponding to the first observation. This refers to the specific value of the feed flow rate recorded during the first observation. For the first The specific value of the reaction temperature recorded during the first observation. For the first The specific value of the reaction pressure recorded during the second observation. For the first The specific value of the feed flow rate recorded during each observation. For the first The specific value of the product yield corresponding to the next observation; The weighted adjacency matrix expression is:

[0048]

[0049]

[0050] ; in, The reaction temperature To product yield The causal strength score, For reaction pressure To product yield The causal strength score, Feed flow rate To product yield The causal strength score, The causal strength of the reaction temperature on itself. To reflect the causal strength of the pressure on itself, The reaction temperature to reaction pressure The causal strength score, The causal strength of the product yield on itself, For reaction pressure to the reaction temperature The causal strength score, Feed flow rate to the reaction temperature The causal strength score, For product yield to the reaction temperature The causal strength score, Feed flow rate to reaction pressure The causal strength score, For product yield The causal strength fraction of the reaction pressure The reaction temperature To feed flow rate The causal strength score, For reaction pressure To product yield The causal strength score, The causal strength of the feed flow rate on itself, Feed flow rate To product yield The causal strength score.

[0051] S3.3. Based on the causal intensity scores between process variables, set a causal intensity threshold and retain causal connections whose causal intensity scores exceed the causal intensity threshold.

[0052] Furthermore, based on all the causal intensity scores in the weighted adjacency matrix, a causal intensity threshold is set, for example, a causal intensity threshold of 0.05; the weighted adjacency matrix is ​​traversed, causal connections with an absolute value of causal intensity score greater than 0.05 are retained, and causal connections with an absolute value of causal intensity score less than or equal to 0.05 are set to zero.

[0053] S3.4 Combine the retained causal connections to construct a causal graph of causal relationships between process variables, with process variables as nodes and directed edges representing causal relationships.

[0054] Furthermore, all retained causal connections are combined; using reaction temperature, reaction pressure, feed flow rate, and product yield as nodes, and the retained causal connections as directed edges, a directed graph is constructed. This directed graph is the causal graph of the causal relationships between process variables.

[0055] S4. Based on the intervention target, identify candidate causal paths from controllable process parameters to the intervention target from the causal graph of causal relationships between process variables.

[0056] S4.1. Based on the intervention objective, determine the position of the intervention objective at the product yield process variable node.

[0057] Furthermore, for the intervention target set by the receiving user to improve product yield, the product yield process variable node in the causal graph corresponding to the intervention target is determined.

[0058] S4.2 From the causal diagram of the causal relationship between process variables, locate the process variable nodes corresponding to the controllable process parameters of reaction temperature, reaction pressure, and feed flow rate as the starting point.

[0059] Furthermore, from the causal graph of the causal relationships between process variables, the reaction temperature process variable node, reaction pressure process variable node, and feed flow rate process variable node are located as the starting point of the search. The pressure process variable node and the feed flow rate process variable node correspond to controllable process parameters that can be directly adjusted by the operator.

[0060] S4.3 In the causal graph of the causal relationship between process variables, use depth-first search to search for the directed path of the process variable node corresponding to the controllable process parameters such as reaction temperature, reaction pressure, and feed flow rate to achieve the product yield intervention target.

[0061] Furthermore, in the causal graph of the causal relationships between process variables, a depth-first search algorithm is executed starting from the reaction temperature process variable node, the reaction pressure process variable node, and the feed flow rate process variable node. The depth-first search algorithm recursively traverses along the direction of the directed edges in the causal graph, recording all directed paths from the starting node to the product yield process variable node.

[0062] S4.4 Identify the directed paths from the controllable process parameter nodes to the intervention target nodes as candidate causal paths from the controllable process parameters to the intervention target.

[0063] Furthermore, all directed paths discovered by the depth-first search algorithm—from the reaction temperature process variable node to the product yield process variable node, from the reaction pressure process variable node to the product yield process variable node, and from the feed flow rate process variable node to the product yield process variable node—are identified as candidate causal paths from controllable process parameters to the intervention target.

[0064] S5. Based on the cause-effect graph and candidate causal paths, the initial simulation model is reconstructed to obtain the cause-effect graph simulation model. The process parameters are input into the cause-effect graph simulation model for simulation to obtain the simulation results of the intervention target. The candidate causal paths are analyzed using a counterfactual explanation algorithm to generate a counterfactual explanation report.

[0065] S5.1 Based on the structure of the causal graph of the causal relationship between process variables, construct the graph attention network skeleton and transfer the parameter knowledge of the initial simulation model to the graph attention network skeleton.

[0066] Furthermore, based on the directed edge connections of the causal graph of the causal relationships between process variables, a graph attention network skeleton is constructed, with the nodes of the graph attention network skeleton corresponding to the process variables in the causal graph; the fully connected layer parameter knowledge of the initial simulation model after training is transferred to the corresponding node embedding vectors in the graph attention network skeleton.

[0067] S5.2. Based on the candidate causal paths from controllable process parameters to the intervention target, enhance the weights of the edges corresponding to the candidate causal paths in the graph attention network skeleton.

[0068] Furthermore, based on the candidate causal path from controllable process parameters to the intervention target, the attention weights of the edges in the graph attention network skeleton corresponding to the candidate causal path are enhanced. For example, the initial attention weight of the edge that coincides with the candidate causal path is set to 1.0, and the initial attention weight of the edge that does not coincide with the candidate causal path is set to 0.5.

[0069] S5.3. Use historical datasets to train and fine-tune the graph attention network skeleton that has undergone parameter transfer and weight enhancement to obtain a causal graph simulation model.

[0070] Furthermore, the graph attention network skeleton with completed parameter transfer and weight enhancement is trained and fine-tuned using historical datasets. During training, the backpropagation algorithm is used to adjust the weight parameters of the graph attention network skeleton, resulting in a trained causal graph simulation model.

[0071] It should be noted that the training process of the graph attention network adopts a message passing mechanism. Each node aggregates the feature information of its neighboring nodes after attention weighting, and the information is propagated through multiple graph attention layers. Finally, the predicted value is output at the intervention target node (product yield node). The training objective is to minimize the mean square error between the predicted product yield and the actual product yield in the historical dataset.

[0072] S5.4 Input a set of reaction temperature values, reaction pressure values, and feed flow rate values ​​as process parameters into the cause-effect graph simulation model.

[0073] Furthermore, a set of specific reaction temperature values, reaction pressure values, and feed flow rate values ​​are combined into a process parameter vector, which is then input into the cause-effect graph simulation model.

[0074] S5.5 The cause-effect graph simulation model calculates based on the input process parameters and outputs the simulation results of product yield.

[0075] Furthermore, the causal graph simulation model performs forward computation based on the input process parameter vector within the graph attention network skeleton. The reaction temperature, reaction pressure, and feed flow rate values ​​from the process parameter vector are assigned to the corresponding reaction temperature, reaction pressure, and feed flow rate nodes in the graph attention network skeleton, respectively. Then, through the message passing mechanism of the graph attention network, information propagates and aggregates along the directed edges of the causal graph of the causal relationships between process variables. The simulation result of product yield is output at the product yield node in the graph attention network skeleton.

[0076] The simulation result expression for product yield is as follows: ; in, The simulation results are for product yield. for, For process parameter vectors, It is a cause-effect graph structure. These are the weight parameters for the cause-effect graph simulation model.

[0077] S5.6. Using a counterfactual interpretation algorithm, calculate the contribution of each candidate causal path to the simulation results of the intervention target for each candidate causal path from controllable process parameters to the intervention target. Furthermore, a counterfactual interpretation algorithm is adopted to calculate the contribution of each candidate causal path to the simulation result of product yield for each candidate causal path. When calculating the contribution, a minimum intervention vector is applied to the current process parameter vector. The intervention vector only has a non-zero change in the process parameter corresponding to the starting point of the candidate causal path. The difference between the product yield simulation result after intervention and the original product yield simulation result is calculated by the causal graph simulation model as the contribution.

[0078] The contribution expression is: ; in, For the first Candidate causal paths For path Contribution This is the current process parameter vector. To minimize the intervention vector, To discretize each independent candidate causal path.

[0079] It should be noted that, The goal is to quantify only through candidate causal paths (e.g., reaction temperature) →Product Yield The impact of this, in order to achieve this, The intervention cut off all other causal pathways that could potentially affect the reaction temperature (e.g., feed flow rate). Material → Reaction Temperature When executing (reaction temperature) When the reactor heating power is adjusted independently while the feed flow rate is strictly controlled to remain constant, this simulates a real-world scenario where engineers operate the reactor on-site: they change the reaction temperature by adjusting an independent controller (such as a heating valve). At the same time, we try our best to keep other parameters stable. The calculation yields the pure effect of this isolated operation.

[0080] S5.7 Based on the contribution of each candidate causal path, generate an analysis report on the path description and quantitative impact as a counterfactual explanation report.

[0081] Furthermore, based on the contribution value of each candidate causal path, an analysis report containing path description and quantitative impact is generated as a counterfactual explanation report. The counterfactual explanation report describes the contribution value of the expected increase in product yield by adjusting the starting process parameters of the candidate causal path.

[0082] S6. Based on the counterfactual explanation report, obtain the optimized process parameter combination, input the optimized process parameter combination into the cause-effect graph simulation model for verification simulation, and obtain the improvement results of the intervention target.

[0083] S6.1 Based on the contribution ranking of each candidate causal path in the counterfactual explanation report, select the candidate causal path with the highest contribution.

[0084] Furthermore, based on the contribution value of each candidate causal path in the counterfactual explanation report, all candidate causal paths are sorted from highest to lowest contribution; the candidate causal path with the highest contribution value is selected as the target path for subsequent optimization.

[0085] S6.2. Based on the counterfactual suggestions corresponding to the candidate causal path with the highest contribution, determine the specific adjustment direction and range of the reaction temperature setpoint, reaction pressure setpoint, and feed flow rate setpoint.

[0086] Furthermore, based on the counterfactual recommendations corresponding to the candidate causal path with the highest contribution in the counterfactual explanation report, the specific adjustment direction and magnitude of the reaction temperature setpoint, reaction pressure setpoint, and feed flow rate setpoint are determined. If the candidate causal path with the highest contribution is the path from reaction temperature to product yield, and the counterfactual recommendation is that increasing the reaction temperature by five degrees Celsius can significantly improve the product yield, then the reaction temperature setpoint is increased by five degrees Celsius on the original basis, while the reaction pressure setpoint and feed flow rate setpoint remain unchanged for the time being.

[0087] S6.3 Combine the adjusted reaction temperature setting, reaction pressure setting, and feed flow rate setting into an optimized combination of process parameters.

[0088] Furthermore, the adjusted reaction temperature setpoints, reaction pressure setpoints, and feed flow rate setpoints are combined to form an optimized combination of process parameters. S6.4 Input the optimized process parameter combination into the cause-effect graph simulation model for verification simulation. The cause-effect graph simulation model outputs the product yield simulation value based on the optimized process parameter combination.

[0089] Furthermore, the optimized process parameter combination is input into the cause-effect graph simulation model for verification simulation; the cause-effect graph simulation model performs calculations based on the optimized process parameter combination and outputs the product yield simulation value based on the optimized process parameter combination, which is the improvement result of the intervention target.

[0090] This embodiment also provides a chemical process simulation system based on big data, including: a data aggregation module, which receives the intervention target set by the user, collects historical data of the target chemical process, and performs preprocessing to form a historical dataset; The path recognition module uses historical datasets to train an initial simulation model with process parameters as input and historical benchmark indicators as output. It then uses a causal discovery algorithm to analyze the historical datasets and construct a causal graph of causal relationships between process variables. The causal constraint module, based on the intervention target, identifies candidate causal paths from controllable process parameters to the intervention target from the causal graph of causal relationships between process variables; The simulation and deduction module reconstructs the initial simulation model based on the cause-effect graph and candidate causal paths to obtain the cause-effect graph simulation model. The process parameters are input into the cause-effect graph simulation model for simulation to obtain the simulation results of the intervention target. The counterfactual explanation algorithm is used to analyze the candidate causal paths and generate a counterfactual explanation report. The verification module, based on the counterfactual explanation report, obtains the optimized combination of process parameters, inputs the optimized combination of process parameters into the cause-effect graph simulation model for verification simulation, and obtains the improvement results of the intervention target.

[0091] This embodiment also provides a computer device applicable to the simulation method of chemical process flow based on big data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the simulation method of chemical process flow based on big data proposed in the above embodiment.

[0092] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0093] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the big data-based chemical process simulation method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0094] In summary, this invention receives user-defined intervention targets and collects preprocessed historical data. After training an initial simulation model, it uses a causal discovery algorithm to construct a causal graph between process variables, identifying candidate causal paths from controllable parameters to the intervention target. Based on the causal graph, a causal graph simulation model is reconstructed. Predictive results are obtained through simulation, and the contribution of each path is analyzed using a counterfactual interpretation algorithm to generate an interpretation report. Based on the report, the combination of process parameters is optimized, and the improved results are verified through simulation. This invention achieves a deepening from data-driven to causal inference, improving the interpretability and optimization accuracy of chemical process simulation.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A big data based chemical process flow simulation method, characterized in that: The method comprises the following steps: Receiving an intervention target set by a user, collecting historical data of a target chemical process, and preprocessing the historical data to form a historical data set, including the following steps: Receiving a user-set intervention target of improving product yield; Collecting temperature data, pressure data, and flow data of the distributed control unit in the target chemical process, collecting product purity data of the laboratory information management unit, and collecting valve opening data of the manufacturing execution unit; Anomaly values in the temperature data, pressure data, flow data, product purity data, and valve opening data are identified and removed using the Isolation Forest algorithm, and missing values in the temperature data, pressure data, flow data, product purity data, and valve opening data are filled using the K-Nearest Neighbor algorithm, and then Z-Score standardization processing is performed; 2. The big data based chemical process flow simulation method as claimed in claim 1, wherein: The processed Z-Score standardized temperature data, pressure data, flow data, product purity data, and valve opening data are time-stamped and integrated to form a historical data set. An initial simulation model is trained using the historical data set, including the following steps: Extracting reaction temperature, reaction pressure, and feed flow as process parameters and product yield as historical benchmark indicators from the historical data set; The extracted process parameters and historical benchmark indicators are divided into a training set and a test set, the process parameters in the training set are used as input features, the historical benchmark indicators in the training set are used as output labels, and the gradient boosting decision tree algorithm is used to train the initial simulation model; The prediction accuracy of the initial simulation model is evaluated on the test set, and the initial simulation model is determined to meet the predetermined accuracy requirement according to the evaluation result on the test set, and the trained initial simulation model is obtained.

3. The big data based chemical process flow simulation method as claimed in claim 2, wherein: A causal graph of causal relationships between process variables is constructed by analyzing the historical data set using a causal discovery algorithm, including the following steps: Selecting reaction temperature, reaction pressure, feed flow, and product yield as process variables for causal analysis from the historical data set; ​ ​ 4. The big data based chemical process flow simulation method as claimed in claim 3, wherein: ​ ​ The NOTEARS causal discovery algorithm is used to calculate the causal strength scores between each pair of process variables for reaction temperature, reaction pressure, feed flow rate, and product yield process variables; According to the causal strength scores between the process variables, a causal strength threshold is set, and causal connections with causal strength scores exceeding the causal strength threshold are retained; The retained causal connections are combined to construct a causal graph of the causal relationships between process variables, with process variables as nodes and directed edges representing causal relationships.

5. The big data based chemical process flow simulation method as claimed in claim 4, wherein: Based on the intervention target, candidate causal paths from controllable process parameters to the intervention target are identified from the causal graph of the causal relationships between process variables, including the following steps: Based on the intervention target, determine the location of the intervention target on the product yield process variable node; From the causal graph of the causal relationships between process variables, locate the process variable nodes corresponding to the controllable process parameters of reaction temperature, reaction pressure, and feed flow rate as starting points; In the causal graph of the causal relationships between process variables, use depth-first search to search for directed paths from the process variable nodes corresponding to the controllable process parameters of reaction temperature, reaction pressure, and feed flow rate to the process variable node corresponding to the product yield intervention target; The directed paths from the controllable process parameter nodes to the intervention target nodes are identified as candidate causal paths from the controllable process parameters to the intervention target.

6. The big data based chemical process flow simulation method as claimed in claim 5, wherein: According to the causal graph and the candidate causal paths, the initial simulation model is reconstructed to obtain a causal graph simulation model. The process parameters are input into the causal graph simulation model for simulation to obtain the simulation results of the intervention target. The candidate causal paths are analyzed using the counterfactual explanation algorithm to generate a counterfactual explanation report, including the following steps: According to the structure of the causal graph of the causal relationships between process variables, construct a graph attention network skeleton, and transfer the parameter knowledge of the initial simulation model to the graph attention network skeleton; According to the candidate causal paths from the controllable process parameters to the intervention target, enhance the weights of the edges in the graph attention network skeleton corresponding to the candidate causal paths; Use the historical data set to train and fine-tune the graph attention network skeleton with completed parameter transfer and weight enhancement to obtain the causal graph simulation model; Input a set of reaction temperature values, reaction pressure values, and feed flow rate values as process parameters into the causal graph simulation model; The causal graph simulation model calculates based on the input process parameters to output the simulation results of the product yield; Using the counterfactual explanation algorithm, calculate the contribution of each candidate causal path to the simulation results of the intervention target for the candidate causal paths from the controllable process parameters to the intervention target; Based on the contribution of each candidate causal path, generate an analysis report of path description and quantitative impact as a counterfactual explanation report.

7. The big data based chemical process flow simulation method as claimed in claim 6, wherein: Based on the counterfactual explanation report, obtain the optimized process parameter combination, input the optimized process parameter combination into the causal graph simulation model for verification simulation, and obtain the improvement results of the intervention target, including the following steps: Based on the contribution degree ranking of each candidate causal path in the counterfactual explanation report, select the candidate causal path with the highest contribution degree; According to the counterfactual explanation corresponding to the candidate causal path with the highest contribution degree, the specific adjustment direction and amplitude of the reaction temperature set value, the reaction pressure set value and the feed flow rate set value are determined; The adjusted reaction temperature set value, reaction pressure set value and feed flow rate set value are combined into an optimized process parameter combination; The optimized process parameter combination is input into the causal diagram simulation model for verification simulation, and the causal diagram simulation model outputs a product yield simulation value based on the optimized process parameter combination.

8. A big data based chemical process flow simulation system based on any one of the big data based chemical process flow simulation method of claims 1-7, characterized in that: The data aggregation module receives the user-set intervention target, collects historical data of the target chemical process, and pre-processes the historical data to form a historical data set; The path identification module trains an initial simulation model with process parameters as input and historical benchmark indicators as output using the historical data set, analyzes the historical data set using a causal discovery algorithm, and constructs a causal diagram of the causal relationships between process variables; The causal constraint module identifies candidate causal paths from controllable process parameters to the intervention target based on the intervention target and the causal diagram of the causal relationships between process variables; The simulation deduction module reconstructs the initial simulation model based on the causal diagram and the candidate causal paths, obtains a causal diagram simulation model, inputs process parameters into the causal diagram simulation model for simulation, obtains a simulation result of the intervention target, analyzes the candidate causal paths using a counterfactual explanation algorithm, and generates a counterfactual explanation report; The verification module obtains an optimized process parameter combination based on the counterfactual explanation report, inputs the optimized process parameter combination into the causal diagram simulation model for verification simulation, and obtains an improved result of the intervention target. 9.A computer device, comprising a memory and a processor, the memory storing a computer program, and the computer device is characterized in that: The processor executes the computer program to implement the steps of the chemical process simulation method based on big data according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the chemical process simulation method based on big data according to any one of claims 1-7.

Images