A distributed control method and system for an alumina production plant

Abstract

The present invention relates to the technical field of industrial production process control, and specifically to a distributed control method and system for an alumina production plant area. The present invention collects and stores the data of alumina production instruments monitored in real time to obtain first alumina production data and second alumina production data; combines the two types of data with reference standard data for alumina production, a dynamic path algorithm, and Euclidean distance to calculate the similarity of the data, and determines whether to perform abnormal feature analysis through an alumina abnormality threshold; when abnormal analysis is required, combines a time series data abnormality recognition algorithm and time series data prediction to obtain first alumina abnormal feature data, and obtains second alumina abnormal feature data through clustering attribution analysis and performs corresponding self-adjustment measures; through the above method, the DCS system can perform more specific and accurate alumina fault monitoring without affecting the production process, and improve the self-diagnosis scope of the alumina DCS system.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial production process control, and specifically to a distributed control method and system for an alumina production plant area. Background Art

[0002] Alumina is a metal compound with high hardness, good corrosion resistance, high temperature resistance and diverse physical and chemical properties. Due to its numerous advantages, it is widely used in industries, medicine, electronics and other fields.

[0003] The alumina production process includes six stages: raw material preparation, digestion, sedimentation, decomposition, roasting and evaporation. Since the production process is relatively complex, problems in a certain link are likely to affect the production quality of alumina. To solve the quality problems, CN113608560B proposes a control system for the alumina caustic liquor preparation process. Aiming at the problem of high caustic soda concentration in the preparation discharge, the real-time preparation data is monitored and adjusted through a feedforward compensator and a feedback controller to improve the qualified rate of the discharge. CN112144043B proposes an alumina deposition device and a gas supply system. Through an improved alumina deposition device, the argon supply in the gas supply process is improved, effectively reducing the blockage probability of the system during the production process and improving the quality and production capacity of the coating films.

[0004] The above methods improve the production quality of alumina by improving the hardware equipment in the production process. However, the above methods still require a relatively high technical level and operation ability and cannot reduce the complex operations of alumina. In order to achieve an efficient and automated production process, the current alumina production process adopts the common decentralized control system (DCS) in the industry to reduce the complex operations of the alumina process.

[0005] The DCS system is a distributed computer control system. By dispersing each control function of the computer into multiple microprocessors, precise control of each step can be achieved. At the same time, the DCS system centrally manages these microprocessors to achieve better control. And it is divided into multiple levels according to the specific production process to achieve more precise management. At the same time, some studies have improved the current DCS system for alumina. For example, CN112684772B proposes a heat balance control system and method for alumina digestion. Aiming at the lag problem of heat balance adjustment in the current DCS system, by obtaining the temperature data between heaters and introducing a fuzzy inference model to analyze the relationship between each data, when an adjustment is needed, the operating system can analyze the real-time analysis results through the model and make adjustments to solve the lag problem and improve the product quality.

[0006] However, due to the high complexity of the DCS system for alumina, although the system has a certain self-diagnosis function, the diagnosis range is limited. Once a fault occurs in the system, it is difficult to find the specific fault problem, resulting in difficulties in maintenance work. Moreover, regardless of the size of the fault problem, once a fault occurs in the DCS system, it will have a huge impact on the entire alumina production process. At present, there is little research on the fault monitoring of the DCS system for alumina.

[0007] In order to achieve real-time fault monitoring of the alumina DCS system, improve the accuracy of fault monitoring of the alumina DSC system, and expand the scope of self-diagnosis and repair, a distributed control method and system for an alumina production plant area are proposed. Summary of the Invention

[0008] The purpose of the present invention is to provide a distributed control method and system for an alumina production plant area. By collecting and storing the data of alumina production instruments monitored in real time, the first alumina production data and the second alumina production data are obtained. The similarity of the two types of data is calculated by combining the alumina production reference standard data, the dynamic path algorithm, and the Euclidean distance. Whether to perform abnormal feature analysis is judged through the alumina abnormality threshold. When abnormal analysis is required, the first alumina abnormal feature data is obtained by combining the time-series data abnormal recognition algorithm and the time-series data prediction, and the second alumina abnormal feature data is obtained through clustering attribution analysis and corresponding self-adjustment measures are taken. Through the above method, the DCS system can perform more specific and accurate alumina fault monitoring without affecting the production process, and improve the self-diagnosis range of the alumina DCS system.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A distributed control method for an alumina production plant area, comprising:

[0011] According to the DCS system, obtain the alumina production instrument data of each stage in the alumina plant area, and perform real-time monitoring on the alumina production instrument data of each stage to generate the first alumina production data of each stage;

[0012] Generate corresponding stage label data and instrument numbers according to the alumina production stage and the alumina production stage instruments, generate a corresponding database according to the instrument numbers, and sort the alumina production instrument data according to the collection time T;

[0013] After the instrument production is completed, store the alumina production instrument data and label the corresponding stage label data to obtain the first alumina production data;

[0014] The alumina production instrument data refers to the power data, instrument revolution data, temperature data, liquid level data, humidity data, gas content data, pressure data and communication flow data of the alumina production stage instruments in each stage of the alumina production process.

[0015] Based on the interval data of the first alumina production data in each stage, obtain the second alumina production data; compare the first alumina production data with the second alumina production data for anomalies.

[0016] Furthermore, the alumina anomaly threshold includes a first alumina reference threshold and a second alumina reference threshold range.

[0017] Obtain the first alumina similarity through the first alumina production data and the alumina production reference standard data.

[0018] If the first alumina similarity is greater than the first alumina reference threshold, there is no anomaly; if the first alumina similarity is less than the first alumina reference threshold, an anomaly has occurred.

[0019] If the second alumina production data is within the second alumina reference threshold range, there is no anomaly; if the second alumina production data is outside the second alumina reference threshold range, an anomaly has occurred.

[0020] If there is data loss in the first alumina production data and / or the second alumina production data, an anomaly has occurred, it is determined that the alumina anomaly feature analysis is not required, and the second alumina anomaly feature data is output.

[0021] Furthermore, the first alumina similarity is obtained through alumina time series similarity comparison calculation. The alumina time series similarity comparison calculation is to call the alumina production reference standard data of the corresponding stage according to the stage label data of the first alumina production data; take n as the vertical axis and m as the horizontal axis to establish an n*m mapping grid.

[0022] The first alumina production data is mapped along the direction of the vertical axis; the alumina production reference standard data is mapped along the method of the horizontal axis, and the first alumina production data and the alumina production reference standard data are mapped to the n*m mapping grid through dynamic path calculation to obtain the first alumina similarity path.

[0023] The alumina production reference standard data is mapped along the vertical and horizontal axes, and is mapped to the n*m mapping grid through dynamic path calculation to obtain the second alumina similarity path.

[0024] The calculation of the first alumina similarity path and the second alumina similarity path is as follows:

[0025] G(i k ,j k )=min[D{i k ,j k}];

[0026]

[0027] Among them, G(i k ,j k ) is the total path distance calculated for the point at position (i k ,j k ), min[] is the minimum value, D{i k ,j k} is the current path distance calculated for the point at position (i k ,j k ), G(i k -1,j k ) is the total path distance calculated for the point at (i k -1,j k ), G(i k -1,j k -1) is the total path distance calculated for the point at (i k -1,j k -1), G(i k ,j k -1) is the total path distance calculated for the point at (i k ,j k -1), R(i k ,j k ) is the distance from the previous point to the position at (i k ,j k );

[0028] Map the first alumina similarity path and the second alumina similarity path to the same n*m mapping grid. After normalizing the data of the first alumina similarity path and the second alumina similarity path, calculate the Euclidean distance to obtain the first alumina similarity; the calculation is as follows:

[0029]

[0030] Among them, κ D is the first alumina similarity, is the coordinate of the d-th point of the first alumina similarity path after normalization calculation, is the coordinate of the d-th point of the second alumina similarity path after normalization calculation, and M is the total number of point coordinates;

[0031] If the abnormal comparison result meets the alumina abnormality threshold, the DCS system continues to monitor and analyze; if the abnormal comparison result does not meet the alumina abnormality threshold, it is determined whether alumina abnormality feature analysis is required; if so, the first alumina production data and / or the second alumina production data are subjected to abnormality feature analysis to obtain first alumina abnormality feature data; if not, second alumina abnormality feature data is directly generated;

[0032] Further, the first alumina abnormality feature data is obtained through an alumina production abnormality model; the alumina production abnormality model includes an alumina production data input unit, an alumina production feature extraction unit, an alumina production abnormality identification unit, an alumina production abnormality prediction and evaluation unit, and an alumina abnormality data output unit;

[0033] The alumina production data input unit preprocesses the first alumina production data into production time series data to generate alumina production input data;

[0034] The alumina production feature extraction unit extracts the features of the alumina production input data to generate alumina production feature data; the alumina production feature data is input into the alumina production abnormality identification unit and compared with alumina reference abnormality data to generate alumina abnormality form data and alumina abnormality data;

[0035] The alumina production abnormality prediction and evaluation unit predicts the second alumina production data through the alumina production input data to generate second alumina prediction data; the second alumina prediction data and the second alumina production data are compared and evaluated to generate alumina prediction evaluation data;

[0036] Both the alumina production feature extraction unit and the alumina production abnormality prediction and evaluation unit are obtained by using the multi-head attention mechanism and LSTM, and are specifically expressed as follows:

[0037]

[0038] Among them, output is the output, input is the input, Mul-Attention() is the multi-head attention mechanism, mul is the data volume of the multi-head attention mechanism, Pea() is feature splicing, and input is the alumina production input data;

[0039] The output of the alumina production feature extraction unit is the extracted time series feature, and the output of the alumina production abnormality prediction and evaluation unit is the value of the second alumina prediction data;

[0040] The alumina production feature extraction unit is obtained by calculating the feature similarity based on the time series features and the alumina reference abnormal data, and the similarity calculation method is the same as that of the first alumina similarity path and the second alumina similarity path;

[0041] The alumina production abnormal prediction and evaluation unit obtains multiple second alumina prediction data through multiple predictions for comparative evaluation. The specific evaluation calculation is as follows:

[0042]

[0043] Among them, Score is the result of the comparative evaluation, θ s is the second alumina production data, α min and α max are weights, λ ps is the second alumina prediction data, other represents other judgment situations, 1 represents similarity, 0 represents dissimilarity, maxc() is to obtain the value with the highest occurrence frequency, is N second alumina prediction data, N is the total number of the second alumina prediction data, maxf() is the number with the highest occurrence frequency, is the c-th second alumina prediction data;

[0044] The alumina abnormal data output unit outputs the corresponding data according to the generation results of the alumina abnormal form data, the alumina abnormal data, and the alumina prediction evaluation data;

[0045] According to the corresponding stage of the first alumina abnormal feature data, the first alumina abnormal feature data is subjected to abnormal fault attribution analysis in the corresponding stage to obtain the second alumina abnormal feature data;

[0046] Furthermore, obtain the stage label data of the alumina abnormal form data and the alumina abnormal data; call the corresponding alumina production instrument problem clustering model according to the stage label data;

[0047] Obtain the first clustering problem label according to the average value of the alumina abnormal form data and the alumina abnormal data; obtain the second clustering problem label according to the duration of the alumina abnormal form data and the alumina abnormal data; obtain the third clustering problem label according to the occurrence time of the alumina abnormal form data and the alumina abnormal data;

[0048] Analyze the subsets of the first clustering problem label, the second clustering problem label, and the third clustering problem label;

[0049] If there is a unique subset, obtain the alumina production clustering problem label based on the unique subset; if there are multiple subsets, obtain the alumina production clustering problem label based on the multiple subsets; if there is no subset, generate the alumina production clustering problem label based on the first clustering problem label, the second clustering problem label, the third clustering problem label, and the result of no subset;

[0050] The alumina production clustering problem label is the second alumina abnormal feature data;

[0051] Perform alumina self-regulation judgment through the second alumina abnormal feature data. If self-regulation can be performed, based on the second alumina abnormal feature data, the DCS system takes corresponding alumina self-regulation measures for the alumina production instrument data; if self-regulation cannot be performed, send the second alumina abnormal feature data to relevant maintenance personnel for adjustment;

[0052] The alumina self-regulation judgment is that if there is a unique subset in the second alumina abnormal feature data, it is judged that self-regulation can be performed, and the corresponding alumina self-regulation scheme is called based on the unique subset; if there is no unique subset in the second alumina abnormal feature data, it is judged that self-regulation cannot be performed;

[0053] The present invention also provides a distributed control system for an alumina production plant area, including an alumina instrument management module, an alumina instrument anomaly comparison module, an alumina anomaly judgment and model analysis judgment module, an alumina attribution analysis module, and an alumina instrument self-regulation module, specifically:

[0054] The alumina instrument management module, the DCS system obtains the alumina production instrument data at each stage of the alumina plant area, monitors the alumina production instrument data at each stage in real time, and generates the first alumina production data at each stage;

[0055] The alumina instrument anomaly comparison module obtains the second alumina production data based on the interval data of the first alumina production data at each stage; compares the first alumina production data with the second alumina production data for anomalies;

[0056] Further, call the alumina production reference standard data of the corresponding stage according to the stage label data of the first alumina production data; establish an n*m mapping grid with n as the vertical axis and m as the horizontal axis;

[0057] The first alumina production data is mapped along the direction of the vertical axis; the alumina production reference standard data is mapped along the horizontal axis, and the first alumina production data and the alumina production reference standard data are mapped onto the mapping grid of n*m through dynamic path calculation to obtain the first alumina similarity path;

[0058] The alumina production reference standard data is mapped along the vertical axis and the horizontal axis, and is mapped onto the mapping grid of n*m through dynamic path calculation to obtain the second alumina similarity path;

[0059] The first alumina similarity path and the second alumina similarity path are mapped onto the same mapping grid of n*m, and after normalizing the data of the first alumina similarity path and the second alumina similarity path, the Euclidean distance is calculated to obtain the first alumina similarity;

[0060] Alumina anomaly judgment and model analysis judgment module. If the anomaly comparison result meets the alumina anomaly threshold, the DCS system continues to monitor and analyze; if the anomaly comparison result does not meet the alumina anomaly threshold, it is judged whether alumina anomaly feature analysis is required; if so, the first alumina production data and / or the second alumina production data are subjected to anomaly feature analysis to obtain the first alumina anomaly feature data; if not, the second alumina anomaly feature data is directly generated;

[0061] Furthermore, the first alumina anomaly feature data is obtained through an alumina production anomaly model; the alumina production anomaly model includes an alumina production data input unit, an alumina production feature extraction unit, an alumina production anomaly identification unit, an alumina production anomaly prediction and evaluation unit, and an alumina anomaly data output unit;

[0062] The alumina production data input unit preprocesses the production time-series data of the first alumina production data to generate alumina production input data;

[0063] The alumina production feature extraction unit extracts the features of the alumina production input data to generate alumina production feature data; the alumina production feature data is input into the alumina production anomaly identification unit and compared with the alumina reference anomaly data to generate alumina anomaly form data and alumina anomaly data;

[0064] The alumina production anomaly prediction and evaluation unit predicts the second alumina production data through the alumina production input data to generate second alumina prediction data; the second alumina prediction data and the second alumina production data are compared and evaluated to generate alumina prediction evaluation data;

[0065] The alumina anomaly data output unit outputs corresponding data according to the generation results of the alumina anomaly form data, the alumina anomaly data, and the alumina prediction and evaluation data;

[0066] The alumina attribution analysis module performs abnormal fault attribution analysis on the first alumina anomaly characteristic data according to the corresponding stage of the first alumina anomaly characteristic data to obtain the second alumina anomaly characteristic data;

[0067] The alumina instrument self-adjustment module performs alumina self-adjustment judgment through the second alumina anomaly characteristic data. If it can be self-adjusted, the DCS system takes corresponding alumina self-adjustment measures on the alumina production instrument data according to the second alumina anomaly characteristic data; if it cannot be self-adjusted, the second alumina anomaly characteristic data is sent to relevant maintenance personnel for adjustment.

[0068] Compared with the prior art, the beneficial effects of the present invention are:

[0069] 1. In order to obtain specific fault problems in the alumina production process, by collecting the production instrument data of alumina at each stage in real time, mapping with the alumina reference standard data to obtain the similarity path of the alumina instrument data, and calculating whether the alumina production instrument data is lower than the threshold through the similarity path as the judgment standard for further anomaly analysis. By performing anomaly analysis on the anomaly data at each stage, it is possible to troubleshoot fault problems without affecting the production process, and at the same time provide data support and basis for the subsequent monitoring and analysis of specific fault problems in the DCS system, improve the accuracy of fault analysis, and increase the fault monitoring range of the DCS system.

[0070] 2. When it is found that the instrument data is abnormal during production, separate fault judgments are made according to the specific abnormal data. The time series characteristics of the instrument data are obtained through the alumina production anomaly model, and its specific abnormal type is obtained based on the anomaly recognition algorithm for further judgment; at the same time, to prevent connection problems between instruments caused by instrument failures, a prediction model is also added to the model to predict the connection interval through the instrument data, so as to improve the accuracy of the alumina production anomaly model in analyzing instrument fault problems. At the same time, combining multi-model analysis can increase the fault monitoring range of the DCS system for more specific attribution analysis.

[0071] 3. After analyzing the problems in the alumina production anomaly model, in order to conduct further detailed analysis through these anomaly data and obtain the causes of specific problems, the corresponding clustering algorithms of each stage are called to obtain the causes of problems. In order to improve the accuracy of problem analysis, three types of data are used for clustering to generate specific problems more precisely. When specific problems are obtained, for the parts that can be self-regulated, the DCS system can perform automatic regulation. When the system cannot perform self-regulation, the answers analyzed by the system will be sent to relevant personnel for data reference. Through this method, the DCS system can conduct more accurate and wider-range fault monitoring, and can analyze problems by itself, improving the fault self-diagnosis accuracy and range of the DCS system. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 is a schematic flow chart of the method of the present invention;

[0073] Figure 2 is a schematic diagram of the first alumina similarity path and the second alumina similarity path of the present invention;

[0074] Figure 3 is a schematic flow chart of the alumina production anomaly model of the present invention;

[0075] Figure 4 is a schematic flow chart of the system of the present invention;

[0076] Figure 5 is a schematic diagram of the structure of the alumina distributed control system of the present invention;

[0077] Figure 6 is a schematic diagram of the software structure of the alumina distributed control system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0078] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0079] The DCS system can achieve precise management and control of various steps of alumina production through multiple levels and a large number of microprocessors. By splitting the huge production system, more detailed regulation and management can be realized to improve the product quality. However, the current DCS system for alumina has limited fault analysis. When a fault occurs, it is impossible to accurately obtain the location of the fault for further diagnosis and repair. Moreover, the occurrence of a fault will have a huge impact on the quality and quantity of the entire alumina product. In order to improve the fault monitoring accuracy of the DCS system for alumina and expand the self-diagnosis repair range, the present invention provides a distributed control method and system for an alumina production plant, and the technical solution is as follows:

[0080] Refer to Figure 1 As shown, according to the DCS system, obtain the alumina production instrument data of each stage in the alumina plant area, and conduct real-time monitoring on the alumina production instrument data of each stage to generate the first alumina production data of each stage;

[0081] According to the interval data of the first alumina production data of each stage, obtain the second alumina production data; compare the first alumina production data and the second alumina production data for anomalies;

[0082] If the anomaly comparison result meets the alumina anomaly threshold, the DCS system continues to monitor and analyze; if the anomaly comparison result does not meet the alumina anomaly threshold, determine whether alumina anomaly feature analysis is required; if so, conduct anomaly feature analysis on the first alumina production data and / or the second alumina production data to obtain the first alumina anomaly feature data; if not, directly generate the second alumina anomaly feature data;

[0083] According to the corresponding stage of the first alumina anomaly feature data, conduct anomaly fault attribution analysis on the first alumina anomaly feature data to obtain the second alumina anomaly feature data;

[0084] Conduct alumina self-regulation judgment through the second alumina anomaly feature data. If self-regulation is possible, according to the second alumina anomaly feature data, the DCS system takes corresponding alumina self-regulation measures for the alumina production instrument data; if self-regulation is not possible, send the second alumina anomaly feature data to relevant maintenance personnel for adjustment.

[0085] The present invention is based on the DCS system to obtain the alumina production instrument data in real time during production, and through the DCS system, reasonable management and storage of each data are carried out to generate the first alumina production data, which is convenient for subsequent production data analysis and improves the accuracy of subsequent analysis;

[0086] Considering that the production of alumina is a detailed and precise process, the connection of production work between various instruments is also the key to fault analysis. Therefore, record the connection time of instruments in each stage where the first alumina production data is located to obtain the second alumina production data. Compare the normal production data and connection data with the reference data without faults to accurately judge whether a fault anomaly has occurred, reduce the calculation burden of the DCS system, improve the fault analysis scope and efficiency of the DCS system. At the same time, further splitting the steps of the fault analysis problem can also improve the accuracy of the analysis;

[0087] In order to accurately analyze the problems that occur, judge whether the DCS system needs to perform further anomaly analysis work based on the alumina anomaly threshold. If there are fault anomaly data with obvious fault characteristics, directly extract the data for attribution analysis. If the fault characteristics are not obvious, perform anomaly feature analysis on it and conduct further attribution analysis based on the analysis results; Through this method, the anomaly feature analysis work can be classified, saving the calculation resources of the DCS system analysis process while enabling more accurate analysis of each data to achieve subsequent fault handling work and improve the anomaly feature accuracy of the entire fault analysis;

[0088] Based on the extracted anomaly feature data for attribution analysis, the system can accurately extract anomaly features while analyzing the causes of these anomaly features leading to fault anomalies, and generate corresponding fault self-adjusting repair solutions for repair based on the causes to ensure more accurate self-repair work of the DCS system, improve the accuracy of the repair plan. At the same time, for situations where the DCS system cannot perform self-adjustment, it can also be sent to relevant personnel for repair based on the accurate analysis results to improve the repair efficiency of relevant personnel and the repair accuracy of fault problems.

[0089] Embodiment 1

[0090] For specific illustration, the present invention is described in combination with the following embodiments. A distributed control method for an alumina production plant area includes:

[0091] Obtain the alumina production instrument data of each stage in the alumina plant area according to the DCS system, and perform real-time monitoring on the alumina production instrument data of each stage to generate the first alumina production data of each stage;

[0092] Generate corresponding stage label data and instrument numbers according to the alumina production stage and the instruments in the alumina production stage, generate a corresponding database according to the instrument numbers, and sort the alumina production instrument data according to the acquisition time T (for example, but not limited to 1 second, 2 seconds, 3 seconds);

[0093] After the instrument production is completed, store the alumina production instrument data and label the corresponding stage label data to obtain the first alumina production data; the alumina production instrument data is the power data, instrument rotation speed data, temperature data, liquid level data, humidity data, gas content data, pressure data, and communication flow data of the instruments in each stage of the alumina production process; alumina production instruments include, for example, alumina ore crushers, alumina ball mills, and alumina rotary kilns used in the roasting stage, etc.

[0094] The specific stages include the instrument data of the raw material preparation stage, digestion stage, sedimentation stage, decomposition stage, roasting stage, and evaporation stage of alumina; the power data, instrument rotation speed data, temperature data, liquid level data, humidity data, gas content data, pressure data, and communication flow data are uniformly input into the data acquisition system based on the measurement system and monitoring system in the DCS system, and at the same time, the corresponding data acquisition of the 6 stages is carried out in combination with the original acquisition data of the data acquisition system.

[0095] Based on the distributed storage logic of the DCS system, the collected data is stored according to different stages and different steps, which is convenient for subsequent management and invocation, and also convenient for adaptive real-time acquisition of the data acquisition methods for different stages; at the same time, by storing the data distributively, it is also possible to prevent data redundancy caused by excessive data during subsequent analysis and reduce the accuracy of subsequent data analysis in the DCS system.

[0096] According to the interval data of the first alumina production data in each stage, obtain the second alumina production data; compare the first alumina production data and the second alumina production data for anomalies.

[0097] The second alumina production data is the interval time for the connection of production work of each instrument. Since the alumina production process is a relatively precise process, and the sudden change in the connection time between relevant production instruments during the production process is also an important part of analyzing whether there is a hardware failure.

[0098] The interval time is mainly obtained by calculating the corresponding acquisition timestamp by combining the data volume of the current stage obtained from the stage label data with the start recording time, and obtaining the interval time based on the acquisition timestamps of different stages to generate the second alumina production data; in addition to the instrument data, time dimension data is added. Considering the fineness of alumina production compared to other industrial productions, by obtaining the interval time data, it is possible to further obtain the normal or abnormal measurement range of the production instruments in each stage of the DCS system and improve the accuracy of subsequent analysis.

[0099] Furthermore, the alumina anomaly threshold includes the first alumina reference threshold and the second alumina reference threshold interval.

[0100] Obtain the first alumina similarity through the first alumina production data and the alumina production reference standard data;

[0101] If the first alumina similarity is greater than the first alumina reference threshold, it is normal; if the first alumina similarity is less than the first alumina reference threshold, an abnormality has occurred;

[0102] If the second alumina production data is within the second alumina reference threshold range, it is normal; if the second alumina production data is outside the second alumina reference threshold range, an abnormality has occurred;

[0103] If there is data missing in the first alumina production data and / or the second alumina production data, an abnormality has occurred, it is determined that alumina abnormality feature analysis is not required, and the second alumina abnormality feature data is output;

[0104] The first alumina reference threshold is the percentage ratio of similarity, and the second alumina reference threshold range is two fixed value ranges. In the specific implementation process, due to the different production processes of each part, in one implementation mode, the first alumina reference threshold and the second alumina reference threshold range of each stage adopt the same numerical standard in combination with the parameters of the actual production instrument. In another implementation mode, the first alumina reference threshold and the second alumina reference threshold range of the raw material preparation stage, digestion stage, sedimentation stage, decomposition stage, roasting stage, and evaporation stage can adopt different numerical standards for the 6 stages in combination with expert experience and the parameters of the actual production instrument;

[0105] Taking the rotary kiln in the roasting stage as an example for the parameters of the actual production instrument, this table only shows partial data, and the parameters are shown in Table 1:

[0106] Table 1 Relevant parameters of different models of rotary kilns

[0107]

[0108] For data missing, the present invention defaults that if there is data missing for more than 5 consecutive acquisition time points, it is determined that an abnormality has occurred, and the judgment standard for data missing can be modified according to the actual situation in the specific implementation process;

[0109] For the convenience of subsequent analysis and to reduce the subsequent computing resource occupation of the DCS system, after obtaining the real-time alumina production instrument data, it is compared based on the alumina reference threshold to distinguish whether an abnormality has occurred. Through this method, the data can be filtered and processed before analysis, improving the accuracy of subsequent model feature extraction and attribution analysis. At the same time, some obvious faults can also be screened in advance at this stage to further improve the screening of different fault problems and enhance the recognition and processing accuracy of the entire fault process of the DCS system;

[0110] Further, the first alumina similarity is obtained by calculating the alumina time series similarity. The alumina time series similarity calculation is to call the alumina production reference standard data of the corresponding stage according to the stage label data of the first alumina production data; taking n as the vertical axis and m as the horizontal axis, an n*m mapping grid is established;

[0111] The first alumina production data is mapped along the vertical axis direction; the alumina production reference standard data is mapped along the horizontal axis direction. The first alumina production data and the alumina production reference standard data are mapped onto the n*m mapping grid through dynamic path calculation to obtain the first alumina similarity path; the alumina production reference standard data will be compared with the alumina production reference standard data of a certain stage among the raw material preparation stage, digestion stage, sedimentation stage, decomposition stage, roasting stage, and evaporation stage by calling the stage label data corresponding to the data.

[0112] The alumina production reference standard data is mapped along the vertical axis and horizontal axis directions, and is mapped onto the n*m mapping grid through dynamic path calculation to obtain the second alumina similarity path;

[0113] The alumina production reference standard data is obtained based on the historical data of a large number of production instruments collected and combined with expert experience;

[0114] The calculation of the first alumina similarity path and the second alumina similarity path is as follows:

[0115] G(i k ,j k )=min[D{i k ,j k}];

[0116]

[0117] Among them, G(i k ,j k ) is the total path distance calculated for the point at position (i k ,j k ), min[] is the minimum value, D{i k ,j k} is the current path distance calculated for the point at position (i k ,j k ), G(i k -1,j k ) is the total path distance calculated for the point at (i k -1,j k ), G(i k -1,j k -1) is for (i k -1,j k-1), the total path distance calculated for the point G(i k ,j k -1) is for (i k ,j k -1), the total path distance calculated for the point R(i k ,j k ) is the distance from the previous point to the position of (i k ,j k );

[0118] Specifically, the path calculations of the first alumina similarity path and the second alumina similarity path refer to Figure 2 shown, Figure 2 In each table, the numbers are path weights. The formula G(i k -1,j k ) + R(i k ,j k ) calculates the longitudinal distance of the path, i.e., path A in the figure; the formula G(i k ,j k -1) + R(i k ,j k ) calculates the lateral distance of the path, i.e., path C in the figure; the formula G(i k -1,j k -1) + 2R(i k ,j k ) represents the diagonal distance of the path, i.e., path B in the figure;

[0119] Map the first alumina similarity path and the second alumina similarity path to the same n*m mapping grid. After normalizing the data of the first alumina similarity path and the second alumina similarity path, calculate the Euclidean distance to obtain the first alumina similarity; the calculation is as follows:

[0120]

[0121] where κ D is the first alumina similarity, is the coordinate of the d-th point of the first alumina similarity path after normalization calculation, is the coordinate of the d-th point of the second alumina similarity path after normalization calculation, and M is the total number of point coordinates;

[0122] In the similarity comparison stage, since the data of alumina production instruments collected is time-series data, in order to improve the similarity calculated for time-series data, during the data comparison stage, the similarity between the anomaly-free alumina production reference standard data and the first alumina production data is calculated. Combining with the dynamic tracking algorithm, the similarity tracking path of the two time-series data is obtained. At the same time, in order to further improve the accuracy of similarity calculation, an anomaly-free similarity tracking path is obtained based on the two anomaly-free alumina production reference standard data. By transforming the time-series data for comparison calculation, the accuracy of the similarity analysis of the DCS system can be significantly improved, facilitating the feature extraction of subsequent models and enhancing the accuracy of fault analysis.

[0123] If the anomaly comparison result meets the alumina anomaly threshold, the DCS system continues to monitor and analyze; if the anomaly comparison result does not meet the alumina anomaly threshold, it is judged whether alumina anomaly feature analysis is required; if so, the first alumina production data and / or the second alumina production data are subjected to anomaly feature analysis to obtain the first alumina anomaly feature data; if not, the second alumina anomaly feature data is directly generated.

[0124] The directly generated second alumina anomaly feature data mainly includes some relatively obvious anomaly features. For example, during the alumina production process, there are a large number of missing data in the liquid level data, severe data jumps in the power data, and large fluctuations (either a large increase or a large decrease) in the temperature data or humidity data for a long time, etc. For these data, since the features are obvious, there is no need for further analysis, and the corresponding anomaly data can be directly screened by the system through traditional statistical methods such as the three-times error method.

[0125] Furthermore, the first alumina anomaly feature data is obtained through the alumina production anomaly model; the alumina production anomaly model includes an alumina production data input unit, an alumina production feature extraction unit, an alumina production anomaly identification unit, an alumina production anomaly prediction and evaluation unit, and an alumina anomaly data output unit; the process is as follows Figure 3 shown;

[0126] The alumina production data input unit preprocesses the first alumina production data into production time-series data to generate alumina production input data; the main steps of the time-series data preprocessing are time-series data denoising and normalization.

[0127] The alumina production feature extraction unit extracts the features of the alumina production input data to generate alumina production feature data; the alumina production feature data is input into the alumina production anomaly identification unit and compared with the alumina reference anomaly data to generate alumina anomaly form data and alumina anomaly data.

[0128] The alumina production anomaly prediction and evaluation unit predicts the second alumina production data from the alumina production input data to generate the second alumina prediction data; compares and evaluates the second alumina prediction data with the second alumina production data to generate alumina prediction evaluation data;

[0129] Both the alumina production feature extraction unit and the alumina production anomaly prediction and evaluation unit are obtained by using the multi-head attention mechanism and LSTM, which are specifically expressed as follows:

[0130]

[0131] Among them, output is the output, input is the input, Mul-Attention() is the multi-head attention mechanism, mul is the data volume of the multi-head attention mechanism, Pea() is the feature splicing, and input is the alumina production input data;

[0132] Among them, the number of heads of the multi-head attention mechanism of the alumina production feature extraction unit is 16, and the number of LSTM units is 6; the number of heads of the multi-head attention mechanism of the alumina production anomaly prediction and evaluation unit is 8, and the number of LSTM units is 12;

[0133] The multi-head attention mechanism is calculated as follows:

[0134]

[0135] Among them, is the feature of node i, σ is the activation function, W is the weight vector, is the feature of node j, α ij is the feature weight coefficient, exp() is the exponential function with base e, e ij and e ik are the node relevance, N i is the number of nodes;

[0136] Both the alumina production feature extraction unit and the alumina production anomaly prediction and evaluation unit use different historical data for training and testing with the multi-head attention mechanism and LSTM model;

[0137] The output of the alumina production feature extraction unit is the extracted time series feature, and the output of the alumina production anomaly prediction and evaluation unit is the value of the second alumina prediction data;

[0138] The alumina production feature extraction unit is obtained by calculating the feature similarity based on the time series feature and the alumina reference anomaly data. The similarity calculation method is the same as that of the first alumina similarity path and the second alumina similarity path; the alumina reference anomaly data is also selected by combining a large amount of historical data and expert experience;

[0139] The abnormal prediction and evaluation unit for alumina production obtains multiple second alumina prediction data through multiple predictions for comparative evaluation. The specific evaluation calculation is as follows:

[0140]

[0141] Among them, Score is the result of comparative evaluation, θ s is the second alumina production data, α min and α max are weights, λ ps is the second alumina prediction data, other represents other judgment situations, 1 represents similar, 0 represents dissimilar, maxc() is to obtain the value with the highest occurrence frequency, is N second alumina prediction data, N is the total number of second alumina prediction data, maxf() is the number with the highest occurrence frequency, is the c-th second alumina prediction data;

[0142] For relatively unobvious fault abnormal data, features in the first alumina abnormal feature data are extracted through the alumina production abnormal model, and the corresponding abnormal types corresponding to different forms of fault data can be accurately obtained by matching the corresponding abnormal data. At the same time, considering that during the process of instrument work connection, it may be caused by instrument failure that the connection is slow, but it cannot be accurately reflected in the data resulting in fault omission. Therefore, by comparing and evaluating the second alumina prediction data predicted by the prediction model with the actual second alumina production data, some hidden problems can be found more accurately, so as to discover and handle them in a timely manner, improving the comprehensiveness and accuracy of DCS system fault analysis;

[0143] The alumina abnormal data output unit outputs corresponding data according to the generation results of alumina abnormal form data, alumina abnormal data, and alumina prediction evaluation data;

[0144] Among them, the alumina abnormal form data outputs corresponding labels according to the identified form features. For example, the step abnormality is marked as 1, the sudden jump is marked as 2, the violent fluctuation is marked as 3, etc. The quantity and type can be set by oneself according to the specific implementation situation; the alumina abnormal data mainly outputs the corresponding original data and outputs the start time and end time of the corresponding abnormality; the alumina prediction evaluation data outputs the corresponding 1 or 0. If it is 0, the corresponding original data and the start time and end time of the corresponding abnormality are output;

[0145] The present invention tests the similarity calculation accuracy rate and prediction accuracy rate through 50 groups of simulation data, and also makes a comparison with other models. The gap with the real result of the simulation data is within 1%, and it is considered that the model calculation is correct. The present invention has carried out 3 tests, and the average value is selected as the final accuracy rate. The specific comparison results are shown in Tables 2 and 3:

[0146] Table 2 Calculation Results of Feature Similarity Comparison Accuracy of Model Alumina

[0147] Model Accuracy of the first test Accuracy of the second test Accuracy of the third test Final accuracy The model of the present invention 90% (45 groups are correct) 90% (45 groups are correct) 94% (47 groups are correct) 91.33％ Transformer + LSTM 84% (42 groups are correct) 96% (48 groups are correct) 86% (43 groups are correct) 88.67％ Attention + GRU 88% (44 groups are correct) 92% (46 groups are correct) 90% (45 groups are correct) 90.00％

[0148] Table 3 Calculation Results of Prediction Accuracy of Model Alumina Features

[0149] Model Accuracy of the first test Accuracy of the second test Accuracy of the third test Final accuracy The model of the present invention 88% (44 groups are correct) 84% (42 groups are correct) 84% (42 groups are correct) 85.33％ Transformer + LSTM 90% (45 groups are correct) 74% (37 groups are correct) 80% (40 groups are correct) 81.33％ Attention + GRU 76% (38 groups are correct) 76% (38 groups are correct) 78% (39 groups are correct) 76.33％

[0150] According to the corresponding stage of the first alumina abnormal feature data, conduct abnormal fault attribution analysis on the first alumina abnormal feature data corresponding to the stage to obtain the second alumina abnormal feature data;

[0151] Furthermore, obtain the abnormal form data of alumina and the stage label data of the abnormal data of alumina; call the corresponding alumina production instrument problem clustering model according to the stage label data; the alumina production instrument problem clustering model is obtained based on a large amount of historical data in each stage; mainly use the k-means model for clustering; the number k of cluster centers is set according to the data statistics of historical data and expert experience;

[0152] Obtain the first clustering problem label according to the abnormal form data of alumina and the mean value of the abnormal data of alumina; obtain the second clustering problem label according to the abnormal form data of alumina and the duration of the abnormal data of alumina; obtain the third clustering problem label according to the abnormal form data of alumina and the occurrence time of the abnormal data of alumina;

[0153] The first clustering problem label is to use the label of the abnormal form data of alumina as the vertical axis coordinate and the mean value of the obtained abnormal data as the horizontal axis coordinate to generate the corresponding coordinate data for clustering calculation;

[0154] The second clustering problem label is to use the label of the abnormal form data of alumina as the vertical axis coordinate and the duration of the abnormal data of alumina calculated from the start time and end time of the abnormal data as the horizontal axis coordinate to generate the corresponding coordinate data for clustering calculation;

[0155] The third clustering problem label is to use the label of the abnormal form data of alumina as the vertical axis coordinate and the start time of the abnormal data as the horizontal axis coordinate to generate the corresponding coordinate data for clustering calculation;

[0156] Analyze subsets of the first clustering problem labels, the second clustering problem labels, and the third clustering problem labels; if there is a unique subset, obtain the alumina production clustering problem labels based on the unique subset; if there are multiple subsets, obtain the alumina production clustering problem labels based on the multiple subsets; if there is no subset, generate the alumina production clustering problem labels based on the first clustering problem labels, the second clustering problem labels, the third clustering problem labels, and the result of the non - existence of subsets;

[0157] The first clustering problem labels, the second clustering problem labels, and the third clustering problem labels include hardware failure problems and software failure problems; examples of hardware failure problems are abnormal temperature sensors for collection, insufficient power supply for the motor of the rotary kiln, insufficient pressure of the alumina crusher, etc.; examples of software problems are that the data storage software fails to store; the software diagram of the alumina DCS system is referred to Figure 6 as shown;

[0158] Since the DCS system itself has a large computational load, the attribution analysis is separated from the model to reduce the computational load of the DCS system analysis. At the same time, by separating the anomaly analysis step and the attribution step, the accuracy of problem analysis in each step can also be improved, ensuring the accuracy of the attribution analysis. Also, in order to ensure accurate problem acquisition, three types of data are used for clustering analysis to improve the accuracy of relevant problem and self - regulation scheme data during the attribution analysis, ensuring that the DCS system can accurately solve the fault problems and improve the accuracy of self - diagnosis and regulation;

[0159] The alumina production clustering problem labels are the second alumina abnormal characteristic data;

[0160] Based on the second alumina abnormal characteristic data, judge the self - regulation of alumina. If it can be self - regulated, according to the second alumina abnormal characteristic data, the DCS system takes corresponding alumina self - regulation measures for the alumina production instrument data; if it cannot be self - regulated, the second alumina abnormal characteristic data is sent to relevant maintenance personnel for adjustment;

[0161] In a specific implementation, if there is only one subset and the DCS system can perform self - regulation of relevant hardware parameters or software parameters; then the DCS system adjusts it. If there are multiple subsets or no subsets, the DCS system will send all the analyzed hardware problems and / or software problems to relevant maintenance personnel for analysis and repair;

[0162] In another implementation, regardless of the number of subsets, the DCS system sends the analysis results to relevant maintenance personnel. If there is only one subset, after the relevant maintenance personnel agree, the DCS system performs the repair; otherwise, the relevant maintenance personnel directly obtain the problems for analysis and repair;

[0163] In the process of solving practical problems, since there may be multiple possibilities in attribution analysis, when these possibilities cannot be judged, the final result of the DCS system analysis is stored and sent to relevant personnel for judgment, so as to avoid the DCS system causing incorrect repairs in special situations, resulting in the occurrence of new faults. At the same time, it can also ensure the accuracy of the DCS system in self-adjusting faults, improve the efficiency and accuracy of solving fault problems, and ensure that the DCS system of alumina can perform self-adjusting repair work in the context of high efficiency and accuracy;

[0164] The self-adjustment judgment of alumina is that if there is a unique subset of the second alumina abnormal characteristic data, it is judged that self-adjustment can be performed, and the corresponding alumina self-adjustment scheme is called based on the unique subset; if there is no unique subset of the second alumina abnormal characteristic data, it is judged that self-adjustment cannot be performed.

[0165] Embodiment 2

[0166] The present invention also provides a distributed control system for an alumina production plant area. This system is optimized based on the overall process of the DCS system. By using the computing resources and storage resources of the existing DCS system, algorithms are added to perform real-time monitoring and real-time analysis of various parts of the faults. At the same time, it also inherits the distributed concept of the DCS system, splitting different stages, different functions, and different faults step by step, performing precise problem analysis while ensuring that the DCS system can perform real-time monitoring, so as to minimize the occupation of computing resources. At the same time, for the problems that occur, it can also be quickly analyzed and calculated, improving the robustness of the overall system and also improving the accuracy of analyzing various problem faults, including an alumina instrument management module, an alumina instrument anomaly comparison module, an alumina anomaly judgment and model analysis judgment module, an alumina attribution analysis module, and an alumina instrument self-adjustment module. Refer to Figure 4 and Figure 5 as shown Figure 5 All the structures in

[0167] The alumina instrument management module. The DCS system obtains the alumina production instrument data of each stage in the alumina plant area through the data acquisition unit in the data acquisition station in Figure 5 ; the data acquisition station in Figure 5 performs real-time monitoring on the alumina production instrument data of each stage and generates the first alumina production data of each stage;

[0168] The alumina instrument anomaly comparison module, Figure 5 the data calculation station in

[0169] Further, Figure 5 The data calculation station in it calls the alumina production reference standard data for the corresponding stage according to the stage label data of the first alumina production data; taking n as the vertical axis and m as the horizontal axis, an n*m mapping grid is established;

[0170] The first alumina production data is mapped along the direction of the vertical axis; the alumina production reference standard data is mapped along the method of the horizontal axis, and the first alumina production data and the alumina production reference standard data are mapped to the n*m mapping grid through dynamic path calculation to obtain the first alumina similarity path;

[0171] The alumina production reference standard data is mapped along the vertical axis and the horizontal axis directions, and is mapped to the n*m mapping grid through dynamic path calculation to obtain the second alumina similarity path;

[0172] The first alumina similarity path and the second alumina similarity path are mapped to the same n*m mapping grid, and after the data of the first alumina similarity path and the second alumina similarity path are standardized and calculated, the Euclidean distance is calculated to obtain the first alumina similarity;

[0173] In the DCS system, by adding the corresponding fault analysis and judgment model for each stage, it is possible to judge the relevant data without affecting the overall alumina production process. At the same time, based on the distributed processing method of the DCS system, the accuracy of the analysis can also be improved, ensuring that the DCS system can accurately detect faults and avoid a large amount of computing resources being occupied by repeated data analysis;

[0174] For the alumina abnormality judgment and model analysis and judgment module, if Figure 5 the abnormality comparison result of the data calculation station in it meets the alumina abnormality threshold, the data acquisition station continues to monitor and analyze; if the abnormality comparison result does not meet the alumina abnormality threshold, the data calculation station judges whether alumina abnormality feature analysis is required; if so, the first alumina production data and / or the second alumina production data are transmitted to Figure 5 the central computer 2 in it for abnormality feature analysis to obtain the first alumina abnormality feature data; if not, the data calculation station generates the second alumina abnormality feature data and transmits it to Figure 5 the central computer 1 in it;

[0175] Further, the first alumina abnormality feature data is obtained through Figure 5 the alumina production abnormality model stored in the central computer 2; the alumina production abnormality model includes an alumina production data input unit, an alumina production feature extraction unit, an alumina production abnormality identification unit, an alumina production abnormality prediction and evaluation unit, and an alumina abnormality data output unit;

[0176] The alumina production data input unit preprocesses the first alumina production data into production time-series data to generate alumina production input data;

[0177] The alumina production feature extraction unit extracts the features of the alumina production input data to generate alumina production feature data; the alumina production feature data is input into the alumina production anomaly recognition unit and compared with the alumina reference anomaly data to generate alumina anomaly form data and alumina anomaly data;

[0178] The alumina production anomaly prediction and evaluation unit predicts the second alumina production data through the alumina production input data to generate second alumina prediction data; the second alumina prediction data and the second alumina production data are compared and evaluated to generate alumina prediction evaluation data;

[0179] The alumina anomaly data output unit outputs the corresponding data according to the generation results of the alumina anomaly form data, the alumina anomaly data, and the alumina prediction evaluation data;

[0180] In a specific implementation, it can be combined with Figure 5 the actual computer resources of the central computer 2 in. If the computer resources are limited, in the first implementation, anomaly analysis can be performed first and then prediction; in the second implementation, prediction can be performed first and then anomaly analysis; if the computer resources are sufficient, in the third implementation, anomaly analysis and prediction can be performed simultaneously;

[0181] By separating the alumina production anomaly model, it can be analyzed when it is ensured that an abnormal fault actually occurs, which can significantly reduce the consumption of computing resources of the DCS system due to the large model. At the same time, it can also leave sufficient time for the model to analyze to avoid errors caused by model analysis failures due to the simultaneous analysis of a large amount of data, improve the accuracy of model analysis, and ensure that the alumina DCS system can accurately analyze fault problems;

[0182] The alumina attribution analysis module, according to the corresponding stage of the first alumina anomaly feature data, Figure 5 the central computer 1 in performs anomaly fault attribution analysis on the first alumina anomaly feature data for the corresponding stage to obtain the second alumina anomaly feature data;

[0183] The alumina instrument self-regulation module, Figure 5 the central computer 1 in makes an alumina self-regulation judgment through the second alumina anomaly feature data. If it can be self-regulated, then according to the second alumina anomaly feature data, Figure 5 the central computer 1 in sends a network message to Figure 5 the operator workstation at the corresponding stage in or Figure 5The engineer workstation in sends instructions and issues instructions to the communication interface of the corresponding instrument control station in Figure 5 and, based on the alumina production instrument data, takes corresponding alumina self-regulation measures through the corresponding control unit in Figure 5 The control unit in Figure 5 is connected to instruments such as crushers and ball mills for the instrument data of the raw material preparation stage, digestion stage, sedimentation stage, decomposition stage, roasting stage, and evaporation stage of alumina, and can adjust the adjustable parameters of each production instrument; if self-regulation cannot be performed, then Figure 5 the central computer 1 in sends the second alumina abnormal characteristic data to Figure 5 the operator workstation in or Figure 5 the relevant maintenance personnel of the engineer workstation in for adjustment.

[0184] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A decentralized control method for an alumina production plant, characterized in that, Including: Obtaining alumina production instrument data at each stage of the alumina plant area according to the DCS system, monitoring the alumina production instrument data at each stage in real time, and generating first alumina production data at each stage; Obtaining second alumina production data according to the interval data of the first alumina production data at each stage; comparing the first alumina production data and the second alumina production data for anomalies; If the anomaly comparison result meets the alumina anomaly threshold, the DCS system continues to monitor and analyze; if the anomaly comparison result does not meet the alumina anomaly threshold, it is judged whether alumina anomaly feature analysis is required; If so, performing anomaly feature analysis on the first alumina production data and / or the second alumina production data to obtain first alumina anomaly feature data; If not, directly generating second alumina anomaly feature data; According to the corresponding stage of the first alumina anomaly feature data, performing anomaly fault attribution analysis on the first alumina anomaly feature data to obtain the second alumina anomaly feature data; Making an alumina self-regulation judgment through the second alumina anomaly feature data. If it can be self-regulated, according to the second alumina anomaly feature data, the DCS system takes corresponding alumina self-regulation measures for the alumina production instrument data; if it cannot be self-regulated, the second alumina anomaly feature data is sent to relevant maintenance personnel for adjustment.

2. The distributed control method for an alumina production plant area according to claim 1, characterized in that Generating corresponding stage label data and instrument numbers according to the alumina production stage and the alumina production stage instruments, generating a corresponding database according to the instrument numbers, sorting the alumina production instrument data according to the collection time T, and storing the alumina production instrument data and marking the corresponding stage label data after the instrument production is completed to obtain the first alumina production data.

3. A decentralized control method for an alumina production plant according to claim 1, characterized in that, The anomaly comparison of the alumina anomaly threshold includes: The alumina anomaly threshold includes a first alumina reference threshold and a second alumina reference threshold range; obtaining a first alumina similarity through the first alumina production data and the alumina production reference standard data; if the first alumina similarity is greater than the first alumina reference threshold, there is no anomaly; if the first alumina similarity is less than the first alumina reference threshold, an anomaly occurs; if the second alumina production data is within the second alumina reference threshold range, there is no anomaly; if the second alumina production data is outside the second alumina reference threshold range, an anomaly occurs; if there is data missing in the first alumina production data and / or the second alumina production data, an anomaly occurs, it is judged that the alumina anomaly feature analysis is not required, and the second alumina anomaly feature data is output.

4. A distributed control method for an alumina production plant area according to claim 2 or 3, characterized in that The first alumina similarity is obtained through alumina time series similarity comparison calculation, and the alumina time series similarity comparison calculation includes: Invoking the alumina production reference standard data of the corresponding stage according to the stage label data of the first alumina production data; taking n as the vertical axis and m as the horizontal axis to establish an n*m mapping grid; The first alumina production data is mapped along the direction of the vertical axis; the alumina production reference standard data is mapped along the horizontal axis, and the first alumina production data and the alumina production reference standard data are mapped onto the mapping grid of n*m through dynamic path calculation to obtain the first alumina similarity path; The alumina production reference standard data is mapped along the vertical and horizontal axes, and is mapped onto the mapping grid of n*m through dynamic path calculation to obtain the second alumina similarity path; the first alumina similarity path and the second alumina similarity path are mapped onto the same mapping grid of n*m, and after the data of the first alumina similarity path and the second alumina similarity path are standardized and calculated, the Euclidean distance is calculated to obtain the first alumina similarity.

5. A distributed control method for an alumina production plant according to claim 1, characterized in that, The acquisition of the first alumina abnormal feature data includes: The first alumina abnormal feature data is obtained through an alumina production abnormal model; the alumina production abnormal model includes an alumina production data input unit, an alumina production feature extraction unit, an alumina production abnormal identification unit, an alumina production abnormal prediction and evaluation unit, and an alumina abnormal data output unit; The alumina production data input unit preprocesses the first alumina production data into production time series data to generate alumina production input data; the alumina production feature extraction unit extracts the features of the alumina production input data to generate alumina production feature data; the alumina production feature data is input into the alumina production abnormal identification unit and compared with alumina reference abnormal data to generate alumina abnormal form data and alumina abnormal data; The alumina production abnormal prediction and evaluation unit predicts the second alumina production data through the alumina production input data to generate second alumina prediction data; the second alumina prediction data and the second alumina production data are compared and evaluated to generate alumina prediction evaluation data; the alumina abnormal data output unit outputs corresponding data according to the generation results of the alumina abnormal form data, the alumina abnormal data, and the alumina prediction evaluation data.

6. A distributed control method for an alumina production plant according to claim 5, characterized in that Performing abnormal fault attribution analysis on the first alumina abnormal feature data in the corresponding stage to obtain the second alumina abnormal feature data includes: Obtaining the stage label data of the alumina abnormal form data and the alumina abnormal data; calling the corresponding alumina production instrument problem clustering model according to the stage label data; Obtaining the first clustering problem label based on the mean value of the alumina abnormal form data and the alumina abnormal data; obtaining the second clustering problem label based on the duration of the alumina abnormal form data and the alumina abnormal data; obtaining the third clustering problem label based on the occurrence time of the alumina abnormal form data and the alumina abnormal data; Analyze subsets of the first clustering problem labels, the second clustering problem labels, and the third clustering problem labels; if there is a unique subset, obtain the alumina production clustering problem labels based on the unique subset; if there are multiple subsets, obtain the alumina production clustering problem labels based on the multiple subsets; if there is no subset, generate the alumina production clustering problem labels based on the first clustering problem labels, the second clustering problem labels, the third clustering problem labels, and the result of no subset; the alumina production clustering problem labels are the second alumina abnormal feature data.

7. A distributed control method for an alumina production plant area according to claim 1, characterized in that The alumina self-regulation is judged as follows: if there is a unique subset of the second alumina abnormal feature data, it is judged that self-regulation can be performed, and the corresponding alumina self-regulation scheme is called based on the unique subset. If there is no unique subset of the second alumina abnormal feature data, it is judged that self-regulation cannot be performed.

8. A distributed control system for an alumina production plant, characterized in that, Used to implement an alumina production plant area distributed control method as described in claim 4, including: An alumina instrument management module, the DCS system obtains the alumina production instrument data at each stage of the alumina plant area, monitors the alumina production instrument data at each stage in real time, and generates the first alumina production data at each stage. An alumina instrument anomaly comparison module, based on the interval data of the first alumina production data at each stage, obtains the second alumina production data; compares the first alumina production data and the second alumina production data for anomalies. An alumina anomaly judgment and model analysis judgment module, if the anomaly comparison result meets the alumina anomaly threshold, the DCS system continues to monitor and analyze; if the anomaly comparison result does not meet the alumina anomaly threshold, it is judged whether alumina anomaly feature analysis is required; if so, perform anomaly feature analysis on the first alumina production data and / or the second alumina production data to obtain the first alumina abnormal feature data; if not, directly generate the second alumina abnormal feature data. An alumina attribution analysis module, based on the corresponding stage of the first alumina abnormal feature data, performs abnormal fault attribution analysis on the first alumina abnormal feature data of the corresponding stage to obtain the second alumina abnormal feature data. An alumina instrument self-regulation module, performs alumina self-regulation judgment through the second alumina abnormal feature data, if self-regulation can be performed, based on the second alumina abnormal feature data, the DCS system takes corresponding alumina self-regulation measures on the alumina production instrument data; if self-regulation cannot be performed, the second alumina abnormal feature data is sent to relevant maintenance personnel for adjustment.

9. The distributed control system for an alumina production plant according to claim 8, wherein The alumina instrument anomaly comparison module judges that based on the dynamic path algorithm, the first alumina similarity path and the second alumina similarity path are mapped onto an n*m mapping network, calculates the Euclidean distance between the first alumina similarity path and the second alumina similarity path, and obtains the first alumina similarity for judgment.

10. The distributed control system for an alumina production plant according to claim 8, characterized in that The model analysis of the alumina anomaly judgment and model analysis judgment module is obtained based on the alumina production anomaly model.

Images