A virtual-real integrated method for energy consumption monitoring and diagnostic analysis throughout the entire chemical production process

Abstract

This invention belongs to the field of process variable detection in industrial production processes and discloses a method for predicting the quality of chemical production processes by integrating multi-condition analysis. The method innovatively proposes using Particle Swarm Optimization (PSOELM) as a sub-model to address the problem of poor model integration accuracy. This not only improves the prediction accuracy and speed of the soft measurement model integrated from sub-models but also reduces the differences between sub-models, enabling better integration. To improve the efficiency of the ensemble model, this invention proposes using a genetic algorithm with fast convergence to select sub-models and selectively integrates models built by PSOELM. This ensures the applicability of the model while significantly reducing storage space and computational complexity.

Description

Technical Field

[0001] This invention belongs to the field of energy management in chemical production processes, specifically relating to a method for monitoring and diagnosing energy consumption throughout the entire chemical production process based on a combination of virtual and real analysis. Background Technology

[0002] Energy management in chemical production is a crucial issue in industrial automation. As a high-energy-consuming production process, chemical enterprises pay close attention to the monitoring and diagnostic analysis of energy status during production to achieve energy conservation, emission reduction, and sustainable development. However, due to the large scale, multiple variables, and coupled operation units inherent in chemical production processes, reliable and effective monitoring and diagnosis of the energy status throughout the entire chemical production process presents significant challenges.

[0003] Currently, energy status monitoring and diagnostic analysis mainly fall into two categories: modeling based on mechanistic knowledge and modeling based on data-driven methods. Considering the complexity of chemical production processes, obtaining accurate mechanistic analysis knowledge is extremely time-consuming and difficult to achieve. Therefore, energy status monitoring and diagnosis for the entire production process often considers historical data analysis. Traditional data-driven methods build black-box models based on input and output variable data, neglecting the correlations between variables within the production process and making fault tracing analysis difficult. Distributed data-driven methods divide the entire production process into several sub-process modules and use multivariate statistical analysis methods for process monitoring and diagnostic analysis, delving deeper into local operational units for status monitoring and fault variable diagnostic analysis. However, current distributed data-driven methods rely heavily on historical data for sub-process module classification, heavily depending on the quality of historical data, and neglecting prior knowledge of the production process and the causal relationships between variables.

[0004] In summary, large-scale chemical production processes involve numerous process units, such as reactors, stripping towers, distillation towers, and stripping towers. These process units form an interconnected production process network, and the internal process variables have a significant impact on energy management during production. Among existing methods for monitoring and diagnosing energy status in chemical production, mechanistic analysis-based methods lack universality and struggle to obtain accurate monitoring and diagnostic models. Data-driven methods, on the other hand, rely on the quality of historical data and lack consideration of prior knowledge. Therefore, how to fully utilize the prior knowledge of the entire chemical production process and combine it with data analysis methods to construct a hybrid virtual-real model to describe the causal relationships between process variables, deeply monitor energy utilization within the production process, and trace the root causes of energy status anomalies is a practical problem that needs to be solved. Summary of the Invention

[0005] To address the aforementioned practical production problems, the purpose of this invention is to provide a virtual-real integrated method for energy consumption monitoring and diagnostic analysis throughout the entire chemical production process. This method delves into the internal analysis of the chemical production process, combining prior knowledge of the process to analyze the interrelationships between various operational units and their process variables. The aim is to achieve scientific and reasonable energy detection and diagnosis based on the operational characteristics of the production process, providing an analytical basis for subsequent optimization of production operation parameters and on-site personnel operation guidance.

[0006] The technical solution of the present invention:

[0007] A method for monitoring and diagnosing energy consumption throughout the entire chemical production process, combining virtual and real methods, comprises the following steps:

[0008] (1) Data preprocessing and analysis

[0009] Historical data on energy and product variables collected in chemical production are validated for data consistency using the Grubbs criterion.

[0010]

[0011] Outliers in the data are removed based on formula (1); where n is the number of samples and α is the significance level. The significance level α and the number of data points n are the test values, and T(n,α) is the test value of Grubbs' criterion; and through...

[0012]

[0013] The data after removing outliers is standardized; where x i,j For the j-th sample data of the i-th input variable, Let x′ be the maximum value of this variable. i,j The data is normalized and standardized;

[0014] (2) Construction of a virtual-real integrated model

[0015] a. Physical Model Construction

[0016] Based on the process control flow chart and knowledge of chemical production processes, the chemical production process is divided into sub-process modules according to the production process unit. The production sequence and cyclical relationship of each sub-process module are considered. In order to ensure the accuracy and comprehensiveness of energy consumption monitoring and diagnosis, the energy flow variables and material flow variables related to energy consumption are analyzed according to the chemical production process to determine the input and output variables of each sub-process module.

[0017] b. Data Model Construction

[0018] Based on prior knowledge of chemical production processes, causal relationship analysis is performed on each sub-process module, and each sub-process module is assigned according to the input X = [x1, x2, ..., x...]. n ] and output Y = [y1, y2, ..., y n The relationship is modeled using probabilistic partial least squares regression analysis, i.e.

[0019] y = B(x - μ) x )+μ y (3)

[0020] Where, μ x μ is the expected value of the sampled values ​​of each input variable x. y Let y be the expected value of each output variable y, and B be the regression coefficient matrix. The common variables among the sub-process modules connect each sub-process module in an orderly manner to form the energy network model of the whole process of chemical production, which is the virtual-real combined model. The virtual-real combined model describes the causal relationship between each sub-process module.

[0021] (3) In-depth monitoring of energy consumption status

[0022] Calculate the energy consumption status monitoring and control limits using historical data on energy and product variables collected during the production process:

[0023]

[0024]

[0025]

[0026]

[0027] in, z represents the set of input variables x and output variables y; T represents the input and output variables respectively. 2 Statistical control limits, D x D y The dimensions of the input and output variables are respectively, N is the number of samples, β is the confidence limit, and F is the value of the input variable. β (D x ,ND x ) is for having D x and ND x F-distribution of degrees of freedom, F β (D y ,ND y ) is for having D y and ND y F-distribution of degrees of freedom, J Q,X J Q,Y These are the control limits for the Q statistic of the input and output variables, respectively. For χ with h degrees of freedom 2 distributed;

[0028] Calculate the monitoring statistics for the collected energy status data respectively:

[0029]

[0030]

[0031]

[0032]

[0033] in, The T values ​​are the input and output variables of the collected energy state data, respectively. 2 Statistic, Q new,X Q new,Y The Q statistics for the input and output variables of the collected energy state data are x and x, respectively. new Input variable data for the collected energy state, y new For the collected energy state output variable data, P and C are the load vector matrices between the input and output variables and the latent variables obtained by the probabilistic partial least squares regression analysis method, respectively, Λ is the eigenvalue matrix, and I is the identity matrix;

[0034] Determine whether energy consumption status is abnormal based on the following criteria:

[0035] ①If Or Q new,X >J Q,X If this happens, an anomaly occurs in the set of input variables;

[0036] ②If Or Q new,Y >J Q,Y If this happens, an anomaly will occur in the set of output variables;

[0037] ③If Or Q new,X >J Q,X and Or Q new,Y >J Q,Y If an anomaly occurs in both the input and output variable sets, then an anomaly will occur simultaneously.

[0038] ④ If none of the above situations occur, then the energy consumption of the production process is normal;

[0039] (5) Diagnosis and source analysis of abnormal energy consumption status

[0040] Based on the in-depth monitoring results of energy consumption status, the cause diagnosis analysis of abnormal energy consumption status is carried out; when an abnormal energy consumption status occurs at the system level of the production process, the energy consumption status level of each sub-process module is judged in depth within the production process, and according to the energy consumption status level of each sub-process module, the sub-process module with abnormal status is constructed in accordance with steps (2) and (3), the input variables of the sub-process module are used as the sub-process cause module, the output variables are used as the sub-process result module, the probabilistic partial least squares method is used for modeling, and the monitoring analysis is carried out in accordance with step (4);

[0041] Calculate the probability values ​​of the anomalies for the variables in the cause and result modules where an anomaly occurs:

[0042]

[0043]

[0044] Among them, z new,sub,i Let F be the set of input variables x and output variables y for the data collected by the i-th module, and let F represent the abnormal state of energy consumption. and p Q (F|z new,sub,i ) represent T values ​​where variables in the module are abnormal. 2 And the probability value of the Q statistic, and p Q (z new,sub,i |F) represent the variables T under abnormal conditions in the module. 2 And the posterior probability value of the Q statistic, and p Q (F) represent the exceptions T in the module. 2 And the prior probability of the Q statistic, and p Q (z new,sub,i ) are respectively T in the module 2 The standardized constants of the Q statistic and the Q statistic.

[0045] Record the number of times each variable encounters an anomaly:

[0046]

[0047] in, γ is the defined importance level, z sub,t,i Let be the set of input variables x and output variables y of the i-th module at the t-th sampling, and T be the total number of samples of the abnormal module;

[0048] Calculate the outlier contribution of each variable within its module:

[0049]

[0050] Where I is the number of all energy consumption variables in the exception module;

[0051] The variable with the highest contribution to the abnormality is the root cause variable of the abnormal energy consumption status; based on the results of the diagnosis and source tracing analysis of the abnormal energy consumption status, corresponding measures are adopted to eliminate the causes of the abnormal energy consumption and improve the energy utilization rate.

[0052] The beneficial effects of this invention are:

[0053] (1) The knowledge and data of chemical production processes are crucial to the accuracy and comprehensiveness of energy management. The virtual and real models constructed by integrating the knowledge of production processes and process data of the entire chemical production process can better analyze the characteristics of internal variables and deeply describe the causal relationships between sub-process modules, providing a causal network analysis basis for monitoring energy consumption status and tracing the source of anomalies.

[0054] (2) Based on the designed distributed monitoring mechanism and the monitoring statistics of each module, the energy consumption status level is analyzed layer by layer from the system layer, process layer and equipment layer, providing an analytical basis for operators to monitor the energy utilization efficiency in the chemical production process.

[0055] (3) Conduct anomaly source tracing and diagnostic analysis on the energy consumption anomaly monitoring results, propose anomaly contribution index to locate abnormal variables, trace the cause of abnormal energy consumption status, and provide a theoretical basis for energy management personnel to make energy management decisions and optimize energy consumption use. Attached Figure Description

[0056] Figure 1 This is the present invention. Figure 1 A flowchart of a method for monitoring and diagnosing energy consumption throughout the entire chemical production process that combines virtual and real data. Detailed Implementation

[0057] The present invention will be further described in detail below with reference to specific embodiments, but the present invention is not limited to the specific embodiments.

[0058] (1) Data preprocessing and analysis

[0059] Historical data collected from chemical production processes are validated for data consistency using the Grubbs criterion.

[0060]

[0061] Remove outliers from the data. The significance level α can be set to 0.01. (This is done by...)

[0062]

[0063] Standardize the data after removing outliers.

[0064] (2) Construction of a virtual-real integrated model

[0065] a. Physical Model Construction

[0066] Based on production process knowledge, sub-process modules are defined, and the input and output variables of each sub-process module are clarified. According to the process control flow chart, the chemical production process can be divided into reaction unit sub-process modules, compression process sub-process modules, separation process sub-process modules, refining process sub-process modules, etc. To ensure the accuracy and comprehensiveness of energy consumption monitoring and diagnosis, energy-related energy flow variables such as the consumption of industrial water, circulating water, demineralized water, high-pressure steam, medium-pressure steam, instrument air, plant air, fuel, nitrogen, and electricity, as well as the output of intermediate and final products, can be used as input and output variables for each sub-process module based on process analysis.

[0067] b. Data Model Construction

[0068] Based on prior knowledge of chemical production processes, a causal relationship analysis is performed on each sub-process module, and the input variables X = [x1, x2, ..., x...] of each sub-process module are defined. n ] and output variable Y = [y1, y2, ..., y n The model is built using probabilistic partial least squares regression analysis, i.e.

[0069] y = B(x - μ) x )+μ y (3)

[0070] The shared variables among the sub-process modules connect them in an orderly manner, forming an energy network model for the entire chemical production process. The constructed virtual-real network model describes the causal relationships between the modules.

[0071] (4) In-depth monitoring of energy consumption status

[0072] Using historical production data, energy consumption status monitoring and control limits are calculated at the system, process, and equipment levels, respectively.

[0073]

[0074]

[0075] For the energy consumption data collected during the production process, monitoring statistics are calculated at the system level, process level, and equipment level, respectively.

[0076]

[0077]

[0078]

[0079]

[0080] The energy consumption status level is analyzed layer by layer according to the energy consumption status monitoring and judgment criteria.

[0081] (5) Diagnosis and source analysis of abnormal energy consumption status

[0082] Based on in-depth monitoring results of energy consumption status, cause diagnosis and analysis are performed on abnormal energy consumption states. When an abnormal energy consumption state occurs at the system level of the production process, the energy consumption status level of each sub-process module is determined by delving into the internal production process. Based on the energy consumption status level of each sub-process module, a virtual-physical hybrid model is constructed for the sub-process modules with abnormal states. The input variables of the sub-process module are used as the cause module, and the output variables are used as the result module. The probabilistic partial least squares method is used for modeling, and monitoring and analysis are conducted.

[0083] Calculate the probability value of the anomaly for the variables in the cause and result modules where an anomaly occurs.

[0084]

[0085]

[0086] Record the number of times each variable encounters an anomaly.

[0087]

[0088] Calculate the outlier contribution of each variable in its respective module.

[0089]

[0090] Where I is the number of all energy consumption variables in the exception module.

[0091] The variable contributing the most to the anomaly is the root cause of the energy consumption anomaly. Based on the results of the energy consumption anomaly diagnosis and source tracing analysis, corresponding measures are adopted to eliminate the causes of the energy consumption anomaly and improve the energy utilization rate.

[0092] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for monitoring and diagnosing energy consumption throughout the entire chemical production process, combining virtual and real methods, characterized in that... The steps are as follows: (1) Data preprocessing and analysis Historical data on energy and product variables collected in chemical production are validated for data consistency using the Grubbs criterion. (1)； Outliers in the data are removed based on formula (1); where, For the number of samples, At the significance level, For this significance level and number of data The test value, The test value is Grubbs' criterion; and it is passed. (2)； The data after removing outliers is then standardized; among them, For the first The first input variable One sample data, The maximum value of this variable. The data is normalized and standardized; (2) Construction of a virtual-real integrated model a. Physical model construction Based on the process control flow chart and knowledge of chemical production processes, the chemical production process is divided into sub-process modules according to the production process unit. The production sequence and cyclical relationship of each sub-process module are considered. In order to ensure the accuracy and comprehensiveness of energy consumption monitoring and diagnosis, the energy flow variables and material flow variables related to energy consumption are analyzed according to the chemical production process to determine the input and output variables of each sub-process module. b. Data Model Construction Based on prior knowledge of chemical production processes, causal relationship analysis is performed on each sub-process module, and each sub-process module is arranged according to its input... and output The relationship is modeled using probabilistic partial least squares regression analysis, i.e. (3)； in, For each input variable Expectation of sampled values For each output variable Expectation of sampled values The regression coefficient matrix; the common variables among the sub-process modules connect the various sub-process modules in an orderly manner to form the energy network model of the entire chemical production process, which is the virtual-real combined model. The virtual-real combined model describes the causal relationship between the various sub-process modules. (3) In-depth monitoring of energy consumption status Using historical data on energy and product variables collected during the production process, calculate the energy consumption status monitoring and control limits respectively: (4)； (5)； (6)； (7)； in, , , , ; Represent each input variable and each output variable A set; , They are respectively input variables and output variables Statistical control limits , These are the dimensions of the input and output variables, respectively. For the number of samples, As confidence limit, For having and degrees of freedom distributed, For having and degrees of freedom distributed, , They are respectively input variables and output variables Statistical control limits For having degrees of freedom distributed; Calculate the monitoring statistics for the collected energy status data respectively: (8)； (9)； (10)； (11)； in, , These are the input and output variables of the collected energy state data, respectively. Statistic, , These are the input and output variables of the collected energy state data, respectively. Statistic, Input variable data for the collected energy state. To collect energy state output variable data, and Let be the loading vector matrix between the input and output variables and the latent variables obtained by the probabilistic partial least squares regression analysis method. For the eigenvalue matrix, It is the identity matrix; Determine whether energy consumption status is abnormal based on the following criteria: ①If or If this happens, an anomaly occurs in the set of input variables; ②If or If this happens, an anomaly will occur in the set of output variables; ③If or and or If an anomaly occurs in both the input and output variable sets, then an anomaly will occur simultaneously. ④ If none of the above situations occur, then the energy consumption of the production process is normal; (4) Diagnosis and source analysis of abnormal energy consumption status Based on the in-depth monitoring results of energy consumption status, conduct cause diagnosis and analysis on the abnormal energy consumption status; when the energy consumption status is abnormal at the system level of the production process, go deep into the production process to judge the energy consumption status level of each sub-process module, and according to the energy consumption status level of each sub-process module, construct the virtual and real combined model of the sub-process module with abnormal status in accordance with steps (2) and (3), take the input variables of the sub-process module as the sub-process cause module, take the output variables as the sub-process result module, use the probability partial least squares method for modeling, and conduct monitoring and analysis in accordance with step (3); Calculate the probability values ​​of the anomalies for the variables in the cause and result modules where an anomaly occurs: (12)； (13)； in, For the first Each module collects data from various input variables. and each output variable The set, This indicates an abnormal state of energy consumption. and These are the variables in the module that are experiencing exceptions. and The probability value of the statistic and These are the variables under abnormal conditions in the module. and The posterior probability value of the statistic. and These are the exceptions in the module. and Prior probability of the statistic and In the modules respectively and Standardized constants of statistics; Record the number of times each variable encounters an anomaly: (14)； in, , For the defined importance level, For the first The module in the first Each input variable during the second sampling and each output variable The set, This represents the total number of samples for this abnormal module; Calculate the outlier contribution of each variable within its module: (15)； in, It is the total number of energy consumption variables in the exception module; The variable with the highest contribution to the abnormality is the root cause variable of the abnormal energy consumption status; based on the results of the diagnosis and source tracing analysis of the abnormal energy consumption status, corresponding measures are adopted to eliminate the causes of the abnormal energy consumption and improve the energy utilization rate.

Images