Chemical process simulation data management system and method based on mathematical model

Abstract

The invention discloses a mathematical model-based chemical process simulation data management system and method, and relates to the technical field of data management. The system comprises a data feature system construction module, a dynamic mapping rule generation module, a cross-scale collaborative data processing module and a matching relation adaptive optimization module. The method comprises the following steps: constructing a multi-scale data quantization feature system and generating feature vectors; establishing a dynamic mapping mechanism and an association matching rule set, wherein the association matching rule set comprises a cross-scale collaborative weight and a working condition dynamic threshold factor; checking data in multiple dimensions, complementing missing attributes and pushing the data as required; the feature vectors and mapping rules may be iteratively optimized through deviation analysis. The problem of data and model hidden mismatching is solved, working condition fluctuation is adapted, cross-scale simulation accuracy and robustness are improved, and safe and efficient circulation of data is achieved.

Description

Technical Field

[0001] This invention relates to the field of data management technology, specifically to a chemical process simulation data management system and method based on mathematical models. Background Technology

[0002] The design optimization, operating condition prediction, and production control of chemical processes rely heavily on the synergistic application of multi-scale mathematical models, such as molecular property models, unit equipment models, and whole-process coupling models.

[0003] In existing technologies, the management of chemical simulation data mainly focuses on basic aspects such as format standardization, storage expansion, and query efficiency, without in-depth design for the differentiated data requirements of multi-scale models. Mathematical models at different scales have significantly different requirements for data granularity, accuracy, and spatiotemporal dimensions. For example, molecular property models require microscopic, high-precision molecular parameter data, while full-process coupled models require macroscopic, wide-temporal process operation data. However, existing data management methods have not established a mapping and correlation mechanism between model scale and data attributes.

[0004] In cross-scale model collaborative simulations, the aforementioned deficiencies can easily lead to implicit mismatches in data retrieval, resulting in a mismatch between model input data and actual requirements. This not only reduces the reliability of simulation operations but also causes significant deviations between simulation results and actual chemical process operation, making it difficult to support precise optimization and scientific decision-making in chemical production. Therefore, there is an urgent need to propose a data management scheme for chemical process simulations that is compatible with multi-scale mathematical models. Summary of the Invention

[0005] The purpose of this invention is to provide a chemical process simulation data management system and method based on mathematical models to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for managing chemical process simulation data based on mathematical models includes the following steps: S1. For mathematical models of chemical processes at different scales, extract the granularity level, precision threshold, and spatiotemporal dimension features of the corresponding data. After quantifying and defining each feature, generate the data attribute feature vector of the model at each scale. S2. Based on the feature vectors generated in step S1, construct a mapping relationship table, define the matching priority and filtering threshold of models and data attributes at each scale, and form a set of callable association matching rules; S3. During cross-scale model collaborative simulation, identify the currently activated target model scale, call the association matching rule set in step S2, perform multi-dimensional attribute verification on the input chemical simulation raw data in terms of granularity, accuracy, and spatiotemporal dimension, filter unqualified data and fill in missing attributes, and push the verified data to the simulation calculation module.

[0007] Furthermore, the quantization definition of features and the generation process of data attribute feature vectors in S1 specifically include: The scale levels of the multi-scale mathematical model for chemical engineering are divided and differentiated. The granularity level, accuracy threshold, and spatiotemporal dimension of the data corresponding to each scale model are quantified separately. Among them, the smaller the quantization result of the accuracy threshold, the higher the data accuracy. The spatiotemporal dimension quantization is calculated by combining the time sampling interval, the actual spatial range covered by the data, and the total spatial range of the corresponding model, and is obtained through weight allocation. The quantification results of the three types of features are processed by the maximum-minimum normalization method. After normalization, the standardized results of each feature are obtained and combined to generate the data attribute feature vectors of the models at each scale.

[0008] Furthermore, the establishment of the dynamic mapping mechanism and the formation of the association matching rule set in S2 specifically include: A mapping table is constructed based on the feature vectors generated by S1. The elements in the table are the basic fit between the model and the data attributes at each scale. The basic fit directly represents the matching priority. The larger the fit value, the higher the matching priority of the corresponding feature dimension. A constraint is set: if the sum of the components of the feature vector is zero, then each feature dimension is processed with equal priority. The strength of data association between scales is quantified by the ratio of the product of the covariance and the corresponding standard deviation of the feature vectors of data attributes in adjacent scale models to obtain cross-scale collaborative weights, and differentiated data sharing or isolation strategies are set according to the weight values. By combining the deviation of the actual process load from the rated load and the deviation of raw material composition, a weighting coefficient is introduced to construct a dynamic threshold factor for the operating condition. Based on the basic adaptability and the dynamic threshold factor for the operating condition, the screening threshold is determined, and finally a set of association matching rules containing a mapping relationship table, collaborative weights and screening thresholds is formed.

[0009] Furthermore, the multi-dimensional attribute validation, missing attribute completion, and targeted push processes in S3 specifically include: Identify the scale type of the target model, call the association matching rule set, and associate adjacent scale collaborative models based on the application rules of cross-scale collaborative weights; The feature vectors of the original data are extracted, and the granularity, precision, and spatiotemporal dimension features are verified using differential verification conditions. Data that does not meet any of the conditions is judged as unqualified data and filtered out, and missing attribute items are marked. For missing attribute items, they are completed by collaboratively using qualified data from adjacent scales, with boundary conditions set: if adjacent single-scale data is missing, then the data from another adjacent scale is used for weighted completion; if both adjacent dual-scale data are missing, then the average of historical data at the same scale is used for completion; the completed data needs to be normalized again to ensure that the corresponding screening threshold is met. The load rate of the simulation computing module is collected. The load rate is the ratio of the current number of computing tasks to the maximum number of tasks that can be carried. Different push strategies are adopted according to the load rate, including direct push, push to the backup module, and load balancing push after caching.

[0010] Furthermore, it also includes an adaptive iterative optimization step, specifically: S4. Collect the deviation between the simulation output and the actual chemical process operation data. When the deviation exceeds the preset threshold, correct the matching priority and screening threshold in the mapping relationship table based on the principle of minimum deviation, and update the data attribute feature vector synchronously to complete the adaptive iterative optimization of the model and data matching relationship. The deviation value handling and adaptive iterative optimization process in S4 specifically includes: Collect simulation output results and actual process data, calculate the standardized deviation within the same operating condition range, and divide the same operating condition range according to the preset process load range and raw material component deviation range; when the standardized deviation is greater than the preset threshold, trigger the adaptive optimization process; Multiple differentiated data matching schemes are generated based on the association matching rule set. The deviation value corresponding to each scheme is calculated to obtain the maximum, minimum and average deviation values. The attribution coefficient is calculated based on the three values, and the constraints are set: if the average deviation value is zero, the deviation is determined to meet the standard and no optimization is required; when the attribution coefficient meets the preset conditions, the root cause of the deviation is determined to be a data matching rule problem, and the hierarchical optimization stage is entered. The matching relationship is optimized in two layers based on the principle of minimum deviation: the feature layer optimizes and updates the data attribute feature vector generated by S1, and after optimization, the maximum and minimum value normalization process is re-executed to ensure that each component of the feature vector is within the preset value range; the rule layer optimizes and updates the matching priority and filtering threshold of S2. The optimized feature vectors and association matching rule set are applied to a new round of cross-scale model collaborative simulation. The standardized deviation is recalculated. If the deviation meets the standard, the optimization results are solidified. If the deviation does not meet the standard, the attribution judgment and hierarchical optimization are repeated until the deviation meets the standard.

[0011] A chemical process simulation data management system based on mathematical models includes: a data feature system construction module, a dynamic mapping rule generation module, and a cross-scale collaborative data processing module; The data feature system construction module extracts and quantifies the granularity level, precision threshold, and spatiotemporal dimension features of the corresponding data for mathematical models of chemical processes at different scales, generating data attribute feature vectors for models at each scale. The dynamic mapping rule generation module constructs a mapping relationship table based on data attribute feature vectors, defines matching priorities and filtering thresholds, calculates cross-scale collaborative weights and sets dynamic threshold factors for working conditions, and forms a set of associated matching rules. The cross-scale collaborative data processing module identifies the target model scale and calls the association matching rule set to perform multi-dimensional attribute verification and missing attribute completion on the raw chemical simulation data. Combined with the load rate of the simulation calculation module, it performs targeted data push.

[0012] Furthermore, the data feature system construction module includes a scale-level partitioning unit and a feature quantization and vector generation unit; The scale hierarchy of the chemical multi-scale mathematical model is divided and distinguished, and the simulation objectives and corresponding data feature boundaries of each scale model are clearly defined. The feature quantization and vector generation unit quantizes three types of features for the data corresponding to each scale model: granularity level, precision threshold, and spatiotemporal dimension. The quantization results of the three types of features are then normalized by the maximum and minimum values ​​and combined to generate the data attribute feature vectors for each scale model.

[0013] Furthermore, the dynamic mapping rule generation module includes a basic adaptability calculation unit and a collaborative weight and threshold setting unit; The basic fit calculation unit constructs a mapping relationship table based on the data attribute feature vectors, calculates the basic fit between the model at each scale and the data attributes in the table, and sets constraints: if the sum of the components of the feature vector is zero, then each feature dimension is processed with equal priority. The collaborative weight and threshold setting unit calculates cross-scale collaborative weights and defines their application rules to control the data sharing strategy between adjacent scale models. At the same time, it calculates the dynamic threshold factor of the operating condition by combining the process load ratio, raw material composition deviation and weight coefficient. Based on the basic fit and threshold factor, it determines the screening threshold and forms a complete set of association matching rules.

[0014] Furthermore, the cross-scale collaborative data processing module includes a data verification and missing data completion unit and a load-aware data push unit; The data validation and missing data completion unit extracts the feature vector of the original data, validates the granularity, precision and spatiotemporal dimension features respectively, filters out unqualified data and marks missing attribute items; at the same time, it completes missing attributes by weighting qualified data of adjacent scales, and sets boundary conditions: when adjacent single-scale data is missing, it uses data of another adjacent scale to complete; when both adjacent dual-scale data are missing, it uses the mean of historical data of the same scale to complete; the completed data is re-normalized and the threshold is validated. The load-aware data push unit collects the load rate of the simulation computing module, which is the ratio of the current number of computing tasks to the maximum number of tasks it can handle. Based on the load rate, different push strategies are adopted, including direct push, push to the backup module, and load balancing push after caching.

[0015] Furthermore, it also includes a matching relationship adaptive optimization module, which is used to collect the deviation values ​​between the simulation output results and the actual process data, determine the root cause of the deviation through attribution analysis, and optimize the feature vector and mapping rules in layers based on the principle of minimum deviation to achieve adaptive iteration of the matching relationship.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention constructs a quantitative feature system encompassing granularity, accuracy, and spatiotemporal dimensions, and establishes dynamic mapping rules between this system and the model scale. This invention fundamentally changes traditional data management methods by intelligently identifying the differentiated data requirements of models at different scales and performing multi-dimensional attribute verification and targeted push, thereby completely avoiding the "hidden mismatch" problem caused by the mismatch between data attributes and model requirements, significantly improving the accuracy and reliability of cross-scale collaborative simulation results. By introducing cross-scale collaborative weights and corresponding data sharing and isolation strategies, this invention can intelligently control the safe and efficient flow of data between models at different scales while ensuring data quality. By dynamically adjusting the data screening threshold based on key operating parameters such as actual process load and raw material composition, this invention enables the data management strategy to respond in real-time to fluctuations in production conditions. This ensures that even under non-design operating conditions or when raw materials change, the data input to the model still meets the accuracy and range requirements of the current state, thus significantly improving the applicability and robustness of the simulation system in complex actual operating environments, making it more closely aligned with actual production. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a module of a chemical process simulation data management system based on a mathematical model, according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1The present invention provides the following technical solution: A chemical process simulation data management system based on mathematical models includes: a data feature system construction module, a dynamic mapping rule generation module, and a cross-scale collaborative data processing module; The data feature system construction module extracts and quantifies the granularity level, precision threshold, and spatiotemporal dimension features of the corresponding data for mathematical models of chemical processes at different scales, generating data attribute feature vectors for models at each scale. The dynamic mapping rule generation module constructs a mapping relationship table based on data attribute feature vectors, defines matching priorities and filtering thresholds, calculates cross-scale collaborative weights and sets dynamic threshold factors for working conditions, and forms a set of associated matching rules. The cross-scale collaborative data processing module identifies the target model scale and calls the association matching rule set to perform multi-dimensional attribute verification and missing attribute completion on the raw chemical simulation data. Combined with the load rate of the simulation calculation module, it performs targeted data push.

[0020] The data feature system construction module includes a scale-level partitioning unit and a feature quantization and vector generation unit; The scale hierarchy of the chemical multi-scale mathematical model is divided and distinguished, and the simulation objectives and corresponding data feature boundaries of each scale model are clearly defined. The feature quantization and vector generation unit quantizes three types of features for the data corresponding to each scale model: granularity level, precision threshold, and spatiotemporal dimension. The quantization results of the three types of features are then normalized by the maximum and minimum values ​​and combined to generate the data attribute feature vectors for each scale model.

[0021] The dynamic mapping rule generation module includes a basic adaptability calculation unit and a collaborative weight and threshold setting unit; The basic fit calculation unit constructs a mapping relationship table based on the data attribute feature vectors, calculates the basic fit between the model at each scale and the data attributes in the table, and sets constraints: if the sum of the components of the feature vector is zero, then each feature dimension is processed with equal priority. The collaborative weight and threshold setting unit calculates cross-scale collaborative weights and defines their application rules to control the data sharing strategy between adjacent scale models. At the same time, it calculates the dynamic threshold factor of the operating condition by combining the process load ratio, raw material composition deviation and weight coefficient. Based on the basic fit and threshold factor, it determines the screening threshold and forms a complete set of association matching rules.

[0022] The cross-scale collaborative data processing module includes a data verification and missing data completion unit and a load-aware data push unit; The data validation and missing data completion unit extracts the feature vector of the original data, validates the granularity, precision and spatiotemporal dimension features respectively, filters out unqualified data and marks missing attribute items; at the same time, it completes missing attributes by weighting qualified data of adjacent scales, and sets boundary conditions: when adjacent single-scale data is missing, it uses data of another adjacent scale to complete; when both adjacent dual-scale data are missing, it uses the mean of historical data of the same scale to complete; the completed data is re-normalized and the threshold is validated. The load-aware data push unit collects the load rate of the simulation computing module, which is the ratio of the current number of computing tasks to the maximum number of tasks it can handle. Based on the load rate, different push strategies are adopted, including direct push, push to the backup module, and load balancing push after caching.

[0023] It also includes a matching relationship adaptive optimization module, which is used to collect the deviation values ​​between the simulation output results and the actual process data, determine the root cause of the deviation through attribution analysis, and optimize the feature vector and mapping rules in layers based on the principle of minimum deviation to achieve adaptive iteration of the matching relationship.

[0024] A method for managing chemical process simulation data based on mathematical models includes the following steps: S1. For mathematical models of chemical processes at different scales, extract the granularity level, precision threshold, and spatiotemporal dimension features of the corresponding data. After quantifying and defining each feature, generate the data attribute feature vector of the model at each scale. S2. Based on the feature vectors generated in step S1, construct a mapping relationship table, define the matching priority and filtering threshold of models and data attributes at each scale, and form a set of callable association matching rules; S3. During cross-scale model collaborative simulation, identify the currently activated target model scale, call the association matching rule set in step S2, perform multi-dimensional attribute verification on the input chemical simulation raw data in terms of granularity, accuracy, and spatiotemporal dimension, filter unqualified data and fill in missing attributes, and push the verified data to the simulation calculation module.

[0025] The quantization definition of features and the generation process of data attribute feature vectors in S1 specifically include: The scale levels of the multi-scale mathematical model for chemical engineering are divided and differentiated. The granularity level, accuracy threshold, and spatiotemporal dimension of the data corresponding to each scale model are quantified separately. Among them, the smaller the quantization result of the accuracy threshold, the higher the data accuracy. The spatiotemporal dimension quantization is calculated by combining the time sampling interval, the actual spatial range covered by the data, and the total spatial range of the corresponding model, and is obtained through weight allocation. The quantification results of the three types of features are processed by the maximum-minimum normalization method. After normalization, the standardized results of each feature are obtained and combined to generate the data attribute feature vectors of the models at each scale.

[0026] In this embodiment, the process of quantizing and defining features and generating data attribute feature vectors in S1 specifically includes: The scale levels of the multi-scale mathematical model of chemical engineering are divided by the scale identifier C, where C=1 represents the microscale, C=2 represents the mesoscale, and C=3 represents the macroscale. The three types of features corresponding to the data of each scale model are quantified separately, specifically including: The granularity level quantization value G = Nd / Nt, where Nd is the number of basic data units in the current dataset, and Nt is the total number of basic data units at the finest granularity of the same type of data. The constraint is that Nt≥1, G∈(0,1]; The precision threshold quantization value J = |Xm-Xc| / Xc, where Xm is the measured value of the data, Xc is the theoretical calculated value of the data or the standard reference value, J∈[0,+∞), and the smaller the J value, the higher the data precision; The quantized value of the spatiotemporal dimension is S = α·(1 / Ti) + β·(Sa / St), where Ti is the data time sampling interval, Sa is the actual spatial range covered by the data, St is the total spatial range of the corresponding model, α and β are weight coefficients and α + β = 1, and S ∈ (0, +∞). The maximum-minimum normalization method is applied to G, J, and S respectively, and the normalization formula is unified as follows: F' = (F - Fmin) / (Fmax - Fmin), where F is the feature value before normalization, and Fmax and Fmin are the maximum and minimum values ​​of the corresponding feature in the full-scale model dataset, respectively; after normalization, G', J', and S' are obtained, all of which take values ​​in the range (0,1]. These are combined to generate the data attribute feature vector V for each scale model. C And V C =(G',J',S').

[0027] The establishment of the dynamic mapping mechanism and the formation of the association matching rule set in S2 specifically include: A mapping table is constructed based on the feature vectors generated by S1. The elements in the table are the basic fit between the model and the data attributes at each scale. The basic fit directly represents the matching priority. The larger the fit value, the higher the matching priority of the corresponding feature dimension. A constraint is set: if the sum of the components of the feature vector is zero, then each feature dimension is processed with equal priority. The strength of data association between scales is quantified by the ratio of the product of the covariance and the corresponding standard deviation of the feature vectors of data attributes in adjacent scale models to obtain cross-scale collaborative weights, and differentiated data sharing or isolation strategies are set according to the weight values. By combining the deviation of the actual process load from the rated load and the deviation of raw material composition, a weighting coefficient is introduced to construct a dynamic threshold factor for the operating condition. Based on the basic adaptability and the dynamic threshold factor for the operating condition, the screening threshold is determined, and finally a set of association matching rules containing a mapping relationship table, collaborative weights and screening thresholds is formed.

[0028] In this embodiment, the process of establishing the dynamic mapping mechanism and forming the association matching rule set in S2 specifically includes: Based on the generated feature vector V C Construct a mapping table, where each element represents the basic fit m between the model and the data attributes at each scale. Ck =V Ck / ∑ C∈[1,3] V Ck V Ck For V C The k-th component, where k=1,2,3 correspond to granularity, precision, and spatiotemporal dimension respectively, with the constraint condition ∑ C∈[1,3] V Ck ≠0; if ∑ C∈[1,3] V Ck =0, then the default m Ck =1 / 3 (equal priority for three feature dimensions); m Ck Directly representing the matching priority, m Ck The larger the value, the higher the matching priority of the corresponding feature dimension; Calculate cross-scale collaborative weights W C,C+1 =cov(V C V C+1 ) / σ(V C )·σ(V C+1 This is used for data flow matching logic to associate adjacent scale models, where cov is the covariance, σ is the standard deviation, and W... C,C+1 ∈[-1,1]; and the application rule for collaborative weights is: when W C,C+1 When the value is ≥0, the adjacent scale model adopts a positive collaborative strategy, that is, feature data can be shared across scales, and the basic fit m of the corresponding feature dimension is ≥0. Ck Not lower than the average fit at the same scale; when W C,C+1 When W < 0, an isolation strategy is adopted, and the model data at each scale are filtered independently; for example, when W C,C+1 When W ≥ 0.5, a strong positive collaboration strategy is adopted to directly reuse the fitness data of adjacent scales; when 0 ≤ W C,C+1 When W < 0.5, a weak positive collaboration strategy is adopted, and reused data requires secondary verification; when -0.5 < W C,C+1 When WC < 0, a weak isolation strategy is adopted, and the data are independently filtered and then the filtering results are shared; when WC,C+1≤-0.5, a strong isolation strategy is adopted, and the model data at each scale are completely independent; when W C,C+1When W = 1, data from adjacent scales are directly reused; when W C,C+1 When =-1, after independent screening, data consistency needs to be verified, and data can only be included if the consistency deviation is ≤5%. Set the dynamic threshold factor for the operating condition as K = 1 + γ·|L-1| + δ·ΔZ, where L is the ratio of the actual process load to the rated load, ΔZ is the raw material component deviation, i.e., the difference in mass fraction between the measured component and the designed component; γ and δ are weighting coefficients, and their specific values ​​are adjusted according to the process type. Based on m Ck Determine the screening threshold T with K Ck And T Ck =K·T0·m Ck Where T0 is the baseline threshold; forming a set of association matching rules that includes a mapping table, collaborative weights, and filtering thresholds.

[0029] The multi-dimensional attribute validation, missing attribute completion, and targeted push process in S3 specifically includes: Identify the scale type of the target model, call the association matching rule set, and associate adjacent scale collaborative models based on the application rules of cross-scale collaborative weights; The feature vectors of the original data are extracted, and the granularity, precision, and spatiotemporal dimension features are verified using differential verification conditions. Data that does not meet any of the conditions is judged as unqualified data and filtered out, and missing attribute items are marked. For missing attribute items, they are completed by collaboratively using qualified data from adjacent scales, with boundary conditions set: if adjacent single-scale data is missing, then the data from another adjacent scale is used for weighted completion; if both adjacent dual-scale data are missing, then the average of historical data at the same scale is used for completion; the completed data needs to be normalized again to ensure that the corresponding screening threshold is met. The load rate of the simulation computing module is collected. The load rate is the ratio of the current number of computing tasks to the maximum number of tasks that can be carried. Different push strategies are adopted according to the load rate, including direct push, push to the backup module, and load balancing push after caching.

[0030] In this embodiment, the multi-dimensional attribute verification, missing attribute completion, and targeted push process in S3 specifically includes: Identify the scale identifier C of the target model, invoke the association matching rule set, and apply cross-scale collaborative weights W. C,C+1 The application rules are associated with adjacent scale collaborative models; Extract the original data feature vector Vd=(Gd',Jd',Sd'). The differential verification conditions are: granularity dimension Gd'≥TC1, precision dimension Jd'≤TC2, and spatiotemporal dimension Sd'≥TC3. Data that does not meet any of the conditions is judged as unqualified data and filtered out, and missing attribute items are marked at the same time. For missing attribute items, they are completed collaboratively using qualified data from adjacent scales. Taking mesoscale precision features as an example, the completion formula is J2'=W 1,2 ·J1'+W 2,3 ·J3', where the boundary conditions are: if adjacent single-scale data is missing, then the data of another adjacent scale is used for weighted completion, for example, the weight is 1; if both adjacent dual-scale data are missing, then the average of historical data of the same scale is used for completion; the completed data needs to be normalized again to ensure that the corresponding screening threshold is met. The load rate U of the simulation computing module is collected, where U = the current number of computing tasks / the maximum number of tasks that can be carried. The tasks are pushed according to the load rate. For example, if U < 0.6, the task is pushed directly; if 0.6 ≤ U < 0.8, the task is pushed to the backup module; if U ≥ 0.8, the task is cached and then pushed after load balancing.

[0031] It also includes an adaptive iterative optimization step, specifically: S4. Collect the deviation between the simulation output and the actual chemical process operation data. When the deviation exceeds the preset threshold, correct the matching priority and screening threshold in the mapping relationship table based on the principle of minimum deviation, and update the data attribute feature vector synchronously to achieve adaptive iterative optimization of the model and data matching relationship. The deviation value handling and adaptive iterative optimization process in S4 specifically includes: Collect simulation output results and actual process data, calculate the standardized deviation within the same operating condition range, and divide the same operating condition range according to the preset process load range and raw material component deviation range; when the standardized deviation is greater than the preset threshold, trigger the adaptive optimization process; Multiple differentiated data matching schemes are generated based on the association matching rule set. The deviation value corresponding to each scheme is calculated to obtain the maximum, minimum and average deviation values. The attribution coefficient is calculated based on the three values, and the constraints are set: if the average deviation value is zero, the deviation is determined to meet the standard and no optimization is required; when the attribution coefficient meets the preset conditions, the root cause of the deviation is determined to be a data matching rule problem, and the hierarchical optimization stage is entered. The matching relationship is optimized in two layers based on the principle of minimum deviation: the feature layer optimizes and updates the data attribute feature vector generated by S1, and after optimization, the maximum and minimum value normalization process is re-executed to ensure that each component of the feature vector is within the preset value range; the rule layer optimizes and updates the matching priority and filtering threshold of S2. The optimized feature vectors and association matching rule set are applied to a new round of cross-scale model collaborative simulation. The standardized deviation is recalculated. If the deviation meets the standard, the optimization results are solidified. If the deviation does not meet the standard, the attribution judgment and hierarchical optimization are repeated until the deviation meets the standard.

[0032] In this embodiment, the deviation value processing and adaptive iterative optimization process in S4 specifically includes: Collect the simulated output Ys and the actual process data Yr, and calculate the standardized deviation E=|Ys-Yr| / (Yr_max-Yr_min), where Yr_max and Yr_min are the maximum and minimum values ​​of the actual process data within the same operating condition interval, respectively. When E is greater than the preset threshold E0, the adaptive optimization process is triggered. For example, the criteria for dividing the same operating condition interval are: process load L∈[0.8,1.2] and raw material composition deviation ΔZ≤5% constitute one operating condition interval. Based on the association matching rule set, multiple differentiated data matching schemes are generated, and the corresponding deviation values ​​are calculated for each scheme. The maximum deviation value Emax, minimum deviation value Emin, and average deviation value Eavg are obtained. The attribution coefficient A is calculated, and A = (Emax - Emin) / Eavg. If A > 2, the root cause of the deviation is determined to be a problem with the data matching rules, and the process proceeds to the hierarchical optimization stage. For example, based on the association matching rule set, five differentiated data matching schemes are generated, with the following differentiation dimensions: Scheme 1 (baseline scheme, maintaining the original rules), Scheme 2 (adjusting the matching priority weight ±20%), Scheme 3 (adjusting the screening threshold ±15%), Scheme 4 (strengthening cross-scale collaborative weights), and Scheme 5 (weakening cross-scale collaborative weights). For the attribution coefficient A, the constraint is Eavg ≠ 0. If Eavg = 0, the deviation is determined to be within the acceptable range, and no optimization is needed. Based on the principle of minimum deviation, the matching relationship is optimized in two layers: Feature layer optimization: Update the data attribute feature vector generated by S1, with the update formula as: V Cn =V Co ·(1+λ·E), where V Cn For the optimized feature vector, V Co The feature vector before optimization is given by λ, which is a preset optimization coefficient; where V Cn After the calculation is complete, re-execute the maximum and minimum value normalization process of S1 to ensure that V Cn The value range of each component remains (0,1]. Rule layer optimization: Update the matching priority and filtering threshold of S2, with the following update formulas: Matching priority m Ckn =m Cko • (E0 / E), the filtering threshold T Ckn =T Cko ·(E0 / E); where m Ckn and T Ckn These are the updated matching priority and filtering threshold, respectively; m Cko and T Cko These are the matching priority and filtering threshold before the update; The optimized feature vectors and association matching rule set are applied to a new round of cross-scale model collaborative simulation, and the standardized deviation E is recalculated. If E≤E0, the optimization results are solidified. If E>E0, the attribution judgment and hierarchical optimization are repeated until the deviation reaches the target. The upper limit of the number of optimization iterations can be set according to the actual process requirements.

[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0034] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for managing chemical process simulation data based on mathematical models, characterized in that: The method includes the following steps: S1. For mathematical models of chemical processes at different scales, extract the granularity level, precision threshold, and spatiotemporal dimension features of the corresponding data. After quantifying and defining each feature, generate the data attribute feature vector of the model at each scale. S2. Based on the feature vectors generated in step S1, construct a mapping relationship table, define the matching priority and filtering threshold of models and data attributes at each scale, and form a set of callable association matching rules; S3. During cross-scale model collaborative simulation, identify the currently activated target model scale, call the association matching rule set in step S2, perform multi-dimensional attribute verification on the input chemical simulation raw data in terms of granularity, accuracy, and spatiotemporal dimension, filter unqualified data and fill in missing attributes, and push the verified data to the simulation calculation module.

2. The method for managing chemical process simulation data based on a mathematical model according to claim 1, characterized in that: The process of quantizing the features and generating the data attribute feature vectors in S1 specifically includes: The scale levels of the multi-scale mathematical model for chemical engineering are divided and distinguished. The granularity level, accuracy threshold, and spatiotemporal dimension of the data corresponding to each scale model are quantified separately. The spatiotemporal dimension quantification is calculated by weighting the time sampling interval, the actual spatial range covered by the data, and the total spatial range of the corresponding model. The quantification results of the three types of features are processed using the maximum-minimum normalization method. After normalization, the standardized results of each feature are obtained and combined to generate the data attribute feature vectors of the models at each scale.

3. The method for managing chemical process simulation data based on a mathematical model according to claim 1, characterized in that: The process of establishing the dynamic mapping mechanism and forming the association matching rule set in S2 specifically includes: A mapping table is constructed based on the feature vectors generated by S1. The elements in the table are the basic fit between the model and the data attributes at each scale. The basic fit directly represents the matching priority. Constraints are set such that if the sum of the components of the feature vector is zero, then each feature dimension is processed with equal priority. The strength of data association between scales is quantified by the ratio of the product of the covariance and the corresponding standard deviation of the feature vectors of data attributes in adjacent scale models to obtain cross-scale collaborative weights, and differentiated data sharing or isolation strategies are set according to the weight values. By combining the deviation of the actual process load from the rated load and the deviation of raw material composition, a weighting coefficient is introduced to construct a dynamic threshold factor for the operating condition. Based on the basic adaptability and the dynamic threshold factor for the operating condition, the screening threshold is determined, and finally a set of association matching rules containing a mapping relationship table, collaborative weights and screening thresholds is formed.

4. The method for managing chemical process simulation data based on a mathematical model according to claim 1, characterized in that: The multi-dimensional attribute verification, missing attribute completion, and targeted push process in S3 specifically includes: Identify the scale type of the target model, call the association matching rule set, and associate adjacent scale collaborative models based on the application rules of cross-scale collaborative weights; The feature vectors of the original data are extracted, and the granularity, precision, and spatiotemporal dimension features are verified using differential verification conditions. Data that does not meet any of the conditions is judged as unqualified data and filtered out, and missing attribute items are marked. For missing attribute items, they are completed by collaboratively using qualified data from adjacent scales. Boundary conditions are set: if adjacent single-scale data is missing, then the data from another adjacent scale is used for weighted completion; if both adjacent dual-scale data are missing, then the average of historical data at the same scale is used for completion. The completed data needs to be normalized again. The load rate of the simulation computing module is collected. The load rate is the ratio of the current number of computing tasks to the maximum number of tasks that can be carried. Different push strategies are adopted according to the load rate, including direct push, push to the backup module, and load balancing push after caching.

5. The method for managing chemical process simulation data based on a mathematical model according to claim 1, characterized in that: It also includes an adaptive iterative optimization step, specifically: S4. Collect the deviation values ​​between the simulation output results and the actual chemical process operation data. When the deviation value exceeds the preset threshold, correct the matching priority and screening threshold in the mapping relationship table based on the principle of minimum deviation, and update the data attribute feature vector synchronously to complete the adaptive iterative optimization of the model and data matching relationship.

6. A chemical process simulation data management system based on a mathematical model, applied to the chemical process simulation data management method based on a mathematical model as described in any one of claims 1-5, characterized in that: The system includes: a data feature system construction module, a dynamic mapping rule generation module, and a cross-scale collaborative data processing module; The data feature system construction module extracts and quantifies the granularity level, precision threshold, and spatiotemporal dimension features of the corresponding data for mathematical models of chemical processes at different scales, generating data attribute feature vectors for models at each scale. The dynamic mapping rule generation module constructs a mapping relationship table based on data attribute feature vectors, defines matching priority and filtering threshold, calculates cross-scale collaborative weights and sets dynamic threshold factors for working conditions, and forms an associated matching rule set. The cross-scale collaborative data processing module identifies the target model scale and calls the association matching rule set to perform multi-dimensional attribute verification and missing attribute completion on the original chemical simulation data. Combined with the load rate of the simulation calculation module, it performs targeted data push.

7. A chemical process simulation data management system based on a mathematical model according to claim 6, characterized in that: The data feature system construction module includes a scale-level partitioning unit and a feature quantization and vector generation unit; The scale hierarchy division unit divides and distinguishes the scale hierarchy of the chemical multi-scale mathematical model, and clarifies the simulation target and corresponding data feature boundaries of each scale model. The feature quantization and vector generation unit quantifies three types of features for the data corresponding to each scale model: granularity level, precision threshold, and spatiotemporal dimension. The quantization results of the three types of features are then normalized by the maximum and minimum values ​​and combined to generate the data attribute feature vectors for each scale model.

8. A chemical process simulation data management system based on a mathematical model according to claim 6, characterized in that: The dynamic mapping rule generation module includes a basic adaptability calculation unit and a collaborative weight and threshold setting unit; The basic fit calculation unit constructs a mapping relationship table based on the data attribute feature vectors, calculates the basic fit between each scale model and the data attributes in the table, and sets constraints such that if the sum of each component of the feature vector is zero, then each feature dimension is processed with equal priority. The collaborative weight and threshold setting unit calculates cross-scale collaborative weights and defines their application rules to control the data sharing strategy between adjacent scale models. At the same time, it calculates the dynamic threshold factor of the operating condition by combining the process load ratio, raw material component deviation and weight coefficient, and determines the screening threshold based on the basic fit and threshold factor to form a complete set of association matching rules.

9. A chemical process simulation data management system based on a mathematical model according to claim 6, characterized in that: The cross-scale collaborative data processing module includes a data verification and missing data completion unit and a load-aware data push unit. The data verification and missing data completion unit extracts the feature vector of the original data, verifies the granularity, precision and spatiotemporal dimension features respectively, filters unqualified data and marks missing attribute items; at the same time, it completes missing attributes by weighting qualified data of adjacent scales, and sets boundary conditions: when adjacent single-scale data is missing, it uses data of another adjacent scale to complete; when both adjacent dual-scale data are missing, it uses the average of historical data of the same scale to complete; the completed data is re-normalized and the threshold is verified. The load-aware data push unit collects the load rate of the simulation computing module, which is the ratio of the current number of computing tasks to the maximum number of tasks it can handle. Different push strategies are adopted according to the load rate, including direct push, push to the backup module, and load balancing push after caching.

10. A chemical process simulation data management system based on a mathematical model according to claim 6, characterized in that: It also includes a matching relationship adaptive optimization module, which is used to collect the deviation values ​​between the simulation output results and the actual process data, determine the root cause of the deviation through attribution analysis, and optimize the feature vector and mapping rules in layers based on the principle of minimum deviation to achieve adaptive iteration of the matching relationship.

Images