Intelligent control system for alkalization reaction of cellulose ether

Abstract

The invention relates to the technical field of chemical automation control, in particular to a cellulose ether alkalization reaction intelligent control system. Comprising a state acquisition module, a reaction control module, a state simulation module and a feedback processing module, the method comprises the following steps: performing continuous change mapping on state information codes and collected process parameters by acquiring the state information codes of each device in the cellulose ether alkalization reaction to form a state mapping table; traversing each element in the state mapping table, constructing a process parameter iterative relation network according to the condition factor of each element under the ether alkalization reaction, and judging the expected state value corresponding to each element. Aiming at the expected state value of each element, determining an abnormal state tree and a control reaction item corresponding to each process parameter; and obtaining a control training path according to the control reaction item, adjusting the current control strategy according to the number of advance steps on the control training path, and determining a control index after the control strategy is adjusted. The efficiency and the accuracy of cellulose treatment are realized.

Description

Technical Field

[0001] The invention relates to the technical field of chemical automation control, in particular to an intelligent control system for cellulose ether alkalization reaction. Background Art

[0002] As an important chemical raw material, the alkalization reaction process of cellulose ether presents the following technical difficulties: Strong multivariable coupling: The reaction process involves more than 10 key process parameters, such as pH, temperature, flow rate, and pressure. These parameters exhibit complex nonlinear correlations, making it difficult to achieve coordinated optimization using traditional PID control methods. Anomaly detection lag: Existing systems rely on fixed threshold alarms, which are unable to capture abnormal patterns in parameter coordination, resulting in delayed responses and easily causing batch rejection.

[0003] For example, Chinese patent publication number CN116804885A discloses a control system for an octafluorocyclobutane heating reaction and a control method thereof, which includes a heater, an AC power regulation module, a communication interface module, a liquid crystal display (LCD), a switching input and output module, a monitoring and alarm module, a host computer, and a controller; the heater realizes signal acquisition of heating temperatures at different locations on site through multiple temperature acquisition modules; the AC power regulation module is connected to the controller to realize control of the heating power of the heater; the communication interface module uses serial communication to realize two-way data exchange with the controller; the LCD realizes input of control parameters of the heating reaction control system and real-time display of feedback signals and temperature curves; the switching input and output module is connected to the controller to realize input and output control of switching signals in the system; the host computer and the controller realize remote control of the heating reaction through wireless communication.

[0004] For example, Chinese patent publication number CN117234266A discloses a method and system for reverse selective control of a ternary precursor reactor reaction. The method involves setting a numerical value of a reaction control item, combining the reaction condition characteristic parameters and the current pH value based on the set numerical value of the reaction control item, and obtaining a first expected value of each reaction condition characteristic parameter through a relationship model between the reaction condition characteristic parameters and the reaction control item. Furthermore, combining the reaction condition characteristic parameters, the current pH value, and the relationship model between the reaction control items, a reverse deduction method is used to obtain multiple second expected values ​​of the reaction control item corresponding to each first expected value. An ideal value of the reaction control item is selected from the multiple second expected values, and the reaction condition characteristic parameters are adjusted accordingly in the reactor.

[0005] The prior art explains that the thermal coupling of the heating reaction can be adjusted through temperature stages and other methods, and that control adjustment can be completed through the expected ideal values ​​of parameters such as flow rate, pH value and temperature under the reactor; however, when the prior art only uses the expected ideal values ​​for control, it is easy to ignore the relationship between parameters, resulting in problems such as control overshoot and oscillation. Moreover, when the expected values ​​are used for identification, only a single parameter out-of-bounds can be identified, and it cannot be described collaboratively based on multiple parameters, resulting in a long time to identify reaction anomalies and an inability to locate the root cause of the multi-parameter impact. Summary of the Invention

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: an intelligent control system for cellulose ether alkalization reaction, including: a state acquisition module, used to obtain the state information code of each device in the cellulose ether alkalization reaction, and continuously map the state information code of each device with the collected process parameters to form a state mapping table.

[0007] The reaction control module is used to traverse each element in the state mapping table, build a process parameter iterative relationship network based on the condition factors of each element under the ether alkalization reaction, and determine the expected state value corresponding to each element.

[0008] The state simulation module is used to identify the abnormal diffusion state of each process parameter based on the expected state value of each element, determine the abnormal state tree corresponding to each process parameter, and describe the control reaction item of the ether alkalization reaction with the output results of the abnormal state tree.

[0009] The feedback processing module is used to obtain at least one control training path according to the control response item, and adjust the current control strategy according to the number of advancement steps on the control training path to determine various control indicators after the control strategy is adjusted.

[0010] The beneficial effects of the present invention are as follows: First, the present invention constructs a state mapping table based on the continuous mapping of equipment state codes and process parameters. By identifying critical points in the operation scenario, the process boundary is dynamically quantified using the intersection of the neighborhood buffer and the operation boundary. The relevant state information codes in the operation scenario are then combined with the process parameters to facilitate the subsequent identification and processing of process parameters at different stages, achieving continuous spatiotemporal representation of the process state.

[0011] 2. The present invention is based on the topological relationship of conditional factors. After combining the conditional factors of each element in the state mapping table in each scenario with the relationship between the described process parameters, a relationship network of conditional factors is formed. The relationship network is then time-weighted and conducted with the state deviation of the process parameters. By calculating the probability of state deviation conduction, the parameter adjustment items are extracted and the expected state values ​​are set; the correlation between process parameters is explicitly modeled to solve the problem of traditional PID control ignoring parameter coupling.

[0012] 3. The present invention is based on the abnormal type of each process parameter under the abnormal diffusion state, and constructs an abnormal state tree after weighting with the frequency corresponding to the abnormal type, and locates its control reaction item with the shortest path described on the abnormal state tree to locate the link relationship where the abnormal propagation occurs, so as to realize the traceability processing of alarms for different parameter abnormalities; then, according to the process parameters extracted on its shortest path, a control training path is formed, and based on the alarm information chain of the control training path, adjustment data is retrieved from the strategy library and control indicators are generated; segmented control strategy processing is realized by advancing the step traversal, and finally the collaborative processing and optimization of multiple parameters are completed. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The present invention will be further described below with reference to the accompanying drawings and examples.

[0014] Figure 1 It is a system framework diagram of an intelligent control system for cellulose ether alkalization reaction.

[0015] Figure 2 The present invention is a flow chart of a state acquisition module of an intelligent control system for cellulose ether alkalization reaction.

[0016] Figure 3 The present invention is a flow chart of a reaction control module of an intelligent control system for cellulose ether alkalization reaction.

[0017] Figure 4 The present invention is a flow chart of a state simulation module of an intelligent control system for cellulose ether alkalization reaction.

[0018] Figure 5 The present invention is a flow chart of a feedback processing module of an intelligent control system for cellulose ether alkalization reaction. DETAILED DESCRIPTION

[0019] The following embodiments of the present invention are described in detail. The embodiments described below are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention. Where specific techniques or conditions are not specified in the embodiments, the techniques or conditions described in the literature in the art or in the product specifications shall be followed.

[0020] See Figure 1 A cellulose ether alkalization reaction intelligent control system includes: a state acquisition module, a reaction control module, a state simulation module and a feedback processing module; wherein the output end of the state acquisition module is connected to the reaction control module, the output end of the reaction control module is connected to the state simulation module, and the output end of the state simulation module is connected to the feedback processing module.

[0021] The state acquisition module is used to obtain the state information code of each device in the cellulose ether alkalization reaction, and continuously map the state information code of each device with the collected process parameters to form a state mapping table.

[0022] The reaction control module is used to traverse each element in the state mapping table, build a process parameter iterative relationship network based on the condition factors of each element under the ether alkalization reaction, and determine the expected state value corresponding to each element.

[0023] The state simulation module is used to identify the abnormal diffusion state of each process parameter based on the expected state value of each element, determine the abnormal state tree corresponding to each process parameter, and describe the control reaction item of the ether alkalization reaction with the output results of the abnormal state tree.

[0024] The feedback processing module is used to obtain at least one control training path according to the control response item, and adjust the current control strategy according to the number of advancement steps on the control training path to determine various control indicators after the control strategy is adjusted.

[0025] The above-mentioned status information coding is used to indicate the operating status of each device, such as running, stopped, faulty and other expressions; process parameters are key factors affecting the quality and efficiency of cellulose ether alkalization reaction, such as alkali concentration, reaction temperature, reaction time, stirring speed and material ratio, which represent the main parameters of the cellulose ether alkalization reaction process. At this time, the data pairs are set with the process parameters and the current operation steps to describe the process of each process parameter after the corresponding time is removed; for example, the temperature pair is set with the temperature and the operation step number to describe the time and pressure under each temperature pair. In the alkalization stage, its material ratio and low temperature duration are used to describe its brief process; in the etherification stage, the temperature in the initial, middle and late stages are checked, and the reaction pressure is checked; then, by adjusting the temperature, pressure, stirring speed and other parameters, the differences in various parameter combinations under each temperature pair and the progress of batch processing are described; to describe the parameter conditions of the equipment in different states.

[0026] The alkali concentration is typically 18%-50%, affecting cellulose swelling and degradation. The reaction temperature represents the temperature used in the alkalization and etherification stages, e.g., 20-35°C for the alkalization stage and 60-100°C for the etherification stage. The reaction time represents the duration of each stage, e.g., 25-120 minutes for alkalization and 3-6 hours for etherification. This can be adjusted based on the type of etherifying agent and the degree of substitution required. The stirring speed indicates the stirring mode and type. For example, high-speed stirring (e.g., 50-100 rpm) is used in the alkalization stage to promote mixing of the alkali and cellulose; medium-to-low speed stirring (e.g., 30-60 rpm) is used in the etherification stage to avoid excessive shearing and product degradation. The material ratio indicates the ratio of cellulose, alkali, and etherifying agent, used to describe the values ​​currently used in production.

[0027] like Figure 2 As shown, the implementation method of the state acquisition module also includes: starting from the operation scenario corresponding to the state information encoding, setting the operation point and operation factor corresponding to each operation scenario according to the semantic identification of the process parameters in different operation scenarios; at this time, the operation scenario represents the equipment status in different time periods in the etherification stage and the alkalization stage, and the semantic identification will explain information such as the reaction progress at different positions. This information is divided into small specific operations and operation points of state changes, as well as the data contained in each operation point, and the data contained in the operation point is regarded as the operation factor.

[0028] Based on the operation factors of each operation point, the critical point of each operation scenario is identified, and each operation point is coupled with the critical point of each operation scenario to obtain a state mapping map; the state mapping map is output as a state mapping table.

[0029] At this time, the operating scenarios divided by the state mapping, such as the cellulose swelling degree in the alkalization stage, the temperature and pressure gradient in the etherification stage, etc., are identified and processed for the corresponding multiple groups of parameters in these scenarios to find out whether there are critical points. These parts with critical points are coupled to illustrate the current risk situations in the etherification scenario. After these contents are organized into a relative knowledge graph, they are output as the completed mapping data. At this time, they are converted into a table in order to intuitively display the problem points of each device in the corresponding operating scenario. The state mapping graph is obtained first in order to improve the traceable correlation of multi-data queries, so as to improve the efficiency and complete display of combined queries of various forms of data.

[0030] At this time, a system consisting of relevant equipment in the cellulose ether alkalization scenario, such as reactors, metering tanks, transfer pulp, washing and filter pressing equipment, granulation equipment, crushing equipment, drying equipment, nitrogen cylinders, exhaust gas treatment equipment, etc., is used to identify and control the status of each device to prevent the problem of low cellulose ether alkalization production efficiency under the current intelligent control.

[0031] At this point, the operation scenario is broken down into specific operation points, each representing a specific moment of state change. For example, the start and end of the alkalization phase, the start and end of the etherification phase, the temperature reaching the set value, and the pressure starting to rise are considered operation points in the current process. Then, operation factors are defined. For example, the operation factors for the start of the alkalization phase include alkali concentration, initial temperature, and stirring speed; the operation factors for the temperature reaching the set value include actual temperature, set temperature, and time to reach the set value. The information required for each operation scenario is used as the primary parameter to be viewed, and this information is recorded in multiple operation factors to complete the multi-parameter association mapping.

[0032] When identifying critical points later, the boundary values ​​of normal processing in each stage are used for selection. For example, the critical point of the alkalization stage will select the data covered in the scenarios where the alkali concentration is lower than 18% or higher than 50%, the temperature exceeds 35°C, and the stirring speed is lower than a certain value. These data will represent the problem of reduced overall efficiency of cellulose ether alkalization. These relevant values ​​can be directly obtained by querying the data stored in the database to determine the multiple sets of parameters that need to be paid attention to when completing the cellulose ether alkalization treatment. The coupled operation points are based on the critical points of each operation scenario, and the operation points are coupled to form a state mapping diagram. For example, in the operating scenario at the end of the alkalization stage, when the temperature reaches 35°C, the trigger temperature reaches the set value operating point, and checks whether the conditions for entering the etherification stage are met; if other etherification stage conditions are met, such as sufficient alkalization time and appropriate alkali concentration, the point is coupled from the end of the alkalization stage to the operating point at the beginning of the etherification stage, and the stage switching is completed at this time. In other words, when the data corresponding to the identified critical point is abnormal, the executed operation cannot smoothly transition to the next stage, causing problems in the overall process and requiring retrospective rollback and other methods to adjust its control method. The final map should clearly show the logical relationship and conversion conditions between each operating point.

[0033] Preferably, when selecting the above-mentioned critical points, they are set based on the value range under normal processing in each stage, and the confidence interval of their values ​​is used as the basis for identifying the critical points. If the critical points exceed the confidence interval, the corresponding critical points will be extracted. The confidence interval level used here is 95%, to view the data points that may obviously exceed the normal part in each operation scenario.

[0034] Preferably, when identifying the critical points under each operation scenario, it also includes: judging the operation factor probability of each operation point, mapping the historical batch state with the operation factor probability of each operation point, determining the operation boundary of each operation point, and extracting multiple critical points under the historical batch mapping.

[0035] At this time, the probability of occurrence of each operating factor at different operating points is counted, and then the probability distribution is fitted to obtain the operating factor probability. For the obtained operating factor probability, find the similar part between the operating factor probability and the historical batch, and then extract the data close to the critical point in this similar part as the critical point output at this time.

[0036] The above-mentioned operation boundary represents the data range when the data in each operation point can reach the critical point. After querying the operation boundary of the current operation point by comparing the currently obtained operation factor with the relevant data in the historical data, the critical point is selected based on the similarity between the operation points of the current batch and the historical batch.

[0037] When mapping historical batches, the similarity between the current operating factor probability and the corresponding operating factor in the historical batch is calculated using the Pearson correlation coefficient. The operating points with the greatest similarity are used as the mapping data at this time. The portions of these mapped operating points close to the critical point are then output as the critical points for the current identification of each operating scenario. The selected operating points are sorted from large to small according to their similarity values, and the top K points are selected. The K value determines the required number of operating points based on the current process characteristics. The historical batches to be mapped are selected in order of batches with similar timing to the current operation to obtain their operating factor probabilities and related values.

[0038] Preferably, when selecting the number of operating points, you can use the previous 10% of the total number currently identified as the number of operating points to be viewed at this time. If this number is less than the minimum number of operating points extracted from any batch in the historical batch processing, the number of operating points extracted from the previous batch will be used as the number of operating points used at this time.

[0039] As for the occurrence probability of each operation factor calculated above at different operation points, it can be set according to the ratio of the occurrence frequency of its operation factor value at the corresponding operation point to the frequency of the corresponding operation point in the historical data. The probability distribution is fitted by using Gaussian distribution or other methods to fit the probability value of the operation factor identified at this time so that it conforms to the form of normal distribution.

[0040] That is, the critical point extraction method also includes: mapping historical batch status data with the operating factor probability of each operating point, constructing a neighborhood buffer corresponding to each operating point. This neighborhood buffer represents the number of operating points that need to be selected under the current process characteristics. The size of this buffer is set according to the safety threshold corresponding to the process characteristics. This neighborhood buffer represents the data set corresponding to the top K operating points whose similarity is calculated after the similarity is calculated. At this time, a single operating point is calculated with data from different historical batches to obtain the data corresponding to the top K operating points after the similarity calculation for each operating point. The safety threshold is set according to the average value of the safety threshold used in the historical data.

[0041] The intersection area of ​​the neighborhood buffer and the operation boundary of each operating point is identified and defined as the critical influence area; the boundary contour of each critical influence area is extracted, and each operating point on the boundary contour is used as the output critical point.

[0042] When extracting the boundary contour, a convex hull algorithm or other forms of algorithms are used to obtain the operating points of the corresponding data in the current intersection area within the corresponding value range, and the extracted operating points; that is, the data in the intersection area is mapped to the data space, and it is distributed as content in the form of a plane graph, such as using the main elements of the operating point in the corresponding scene as its coordinates. For example, the actual temperature and reaction time are the main parameters to be viewed in the alkalization reaction. At this time, the actual temperature and reaction time are used as its coordinate system. After mapping the data of the corresponding operating point, the convex hull algorithm is used to extract the boundary points that can surround these data. These boundary points will reflect the relative position of the boundary points in the current time series ether alkalization reaction, and assist in identifying the value range of the current ether alkalization reaction at each stage, to prevent the occurrence of under-reaction or over-reaction parts, and to improve the process of adjusting the cellulose ether alkalization.

[0043] In one embodiment of the present invention, the iterative relationship of the process parameters under the ether alkalization reaction is described at this time, and the key parameters under the iterative method, such as temperature, pressure, heating rate, solution concentration, etc., are obtained, and it is described whether each element can reach the expected state value; this expected state value will indicate the progress of the current ether alkalization reaction; after the expected value is obtained, it is promoted to the PID-related dynamic instructions under each operation deployment.

[0044] In the reaction control module, the parameters contained in the state mapping table are the main processing content, and the initial reaction values ​​of each element in the corresponding stage are used as its conditional factors. Then, the parameters are combined with their initial values ​​and their numerical changes to form an iterative relationship network to illustrate the change scenario of the current reaction control module.

[0045] like Figure 3As shown, the implementation method of the reaction control module includes: dividing each element in the state mapping table according to the mapped data, and setting at least one condition factor for each element during the division; the condition factor represents the condition value represented when the operation factor is established at each operation point in the state mapping table, and this value is regarded as the condition factor. For example, the main reaction period of the etherification stage is regarded as the currently processed data, and the gradient temperature increase and pH dynamic control therein are regarded as the operation point. Then the operation factor will include the heating rate, target temperature, holding time, real-time pH value, alkali solution concentration, alkali supplement flow rate and control cycle; then the condition factor needs to make these data associated with the normal operation of the current equipment, and can change with time and appear different situations according to the process; then the condition factor can select its process parameters such as temperature, pH value, alkali solution concentration, and use the content of these values ​​as the condition factor processed at this time to identify whether the process of cellulose ether alkalization reaction is normal.

[0046] The topological relationship between each condition factor is obtained, and according to the correlation coefficient between each condition factor in the topological relationship, each condition factor is used as a network node to generate a relationship network; at this time, the topological relationship between each condition factor indicates whether there is a correlation between each condition factor. After describing whether there is a logical relationship between each condition factor in the cellulose ether alkalization process, the correlation coefficient of the corresponding data in each condition factor is calculated. At this time, the corresponding data of two condition factors are used as input, and after introducing the corresponding data in the historical batches, the correlation coefficient between the condition factors is calculated in the form of the Pearson correlation coefficient, and then a relationship network with correlation is formed.

[0047] At this point, all conditional factors are connected according to the preset topological relationships. For example, alkali flow rate can be associated with temperature, and pH value can be connected with alkali consumption during the reaction. This generates fully connected factor pairs for all conditional factors. Correlation coefficients are calculated for each factor pair, and edges with correlation coefficients greater than 0.6 are retained. All edges are then tested using a t-test. If the p-value is greater than 0.05, all edges are sorted in descending order by the absolute value of the correlation coefficient. Each node is treated as an independent tree, and the maximum-weighted edge is selected in turn. If the connection is to a different tree, it is merged until all nodes are connected. This completes the generation of the relationship network. If the p-value is less than 0.05, the edges in the relationship network are not merged, indicating that there is a significant correlation between the conditional factors and no merging is required.

[0048] Perform time-series weighting on each network node in the relationship network, and use the time-series weighted relationship network as the process parameter to iterate the relationship network.

[0049] At this time, each network node in the relationship network represents a specific condition factor, and the weights between the network nodes are updated each time by the sliding window calculation method to complete the time series weighting; ;in, represents the weight between two network nodes at time t, represents the weight between two network nodes at time t-1, Represents the adjustment coefficient, with a value range of 0.2-0.3. Here, 0.2 is selected to emphasize the weight of the corresponding grid node in the previous time series as much as possible. The value represents the importance of the iterative weighting of the previous and current time series. This is suitable for scenarios where data changes drastically and the latest trends need to be reflected quickly. A value of 0.3 can achieve a balance between the current and historical data, retaining more information from the previous time series. This is suitable for scenarios where data has certain fluctuations but the influence of noise needs to be suppressed. Represents the correlation coefficient between two network nodes. The two network nodes described here refer to any two network nodes with a connection relationship in the current relationship network, to illustrate the weight adjustment form of its network nodes in time series changes. At this time, the adjustment method can be completed by adjusting the weights of different edges in the relationship network through the correlation coefficient of the data in the current batch and the weights in the previous batch.

[0050] As for judging the expected state value of each element, it can be expressed as calculating the state deviation of each network node in the process parameter iteration relationship network, determining the transmission probability of the state deviation between each network node, and using the transmission probability and weight of the network node to extract the parameter adjustment item of the state deviation.

[0051] According to the value of the parameter adjustment item of the state deviation after iteration, the expected state value of each element is set.

[0052] The state deviation of each network node mentioned above represents the difference in the corresponding data on the network node within a continuous time period, and the conduction probability represents the conditional probability corresponding to the state deviation of the two network nodes; then, the parameter adjustment item for the state deviation is selected by using the weight between the corresponding network nodes in the process parameter iteration relationship network. The weight is the value of the relationship network when iterating according to the time series. Then, the conduction probability and weight are used to set the relevant path with the current network node, and then the main parameters related to it are found to illustrate the current value that needs to be viewed. The subsequent constraint iteration represents the value range of the condition factor during iteration, and then represents the expected state value obtained by each element corresponding to the state deviation; that is, after the process parameter iteration relationship network of the relationship network combination is segmented, the values ​​of the parameter adjustment items under multiple time series are checked, and the average value of the same parameter adjustment item obtained after weighting the relationship network under multiple time series is used as the output expected state value.

[0053] For example, the implementation method of extracting parameter adjustment items for state deviation includes: extracting a sub-network of the process parameter iteration relationship network based on the weight of each network node in the process parameter iteration relationship network, and each network node in the sub-network is not an isolated node; at this time, the extracted sub-network is selected based on the content identified in the process parameter iteration relationship network, focusing on an element in a certain state mapping table as its judgment standard, and after screening out multiple network nodes related to the corresponding elements in the state mapping table, the corresponding network nodes are identified with the conduction probability to indicate the expected state value that needs to be judged in the end.

[0054] At this time, the sub-network is extracted by screening out the network nodes related to the sub-network according to the weights between the network nodes. It is required that the weight value of the network node corresponding to the element in the current state mapping table is greater than 0.6, so as to complete the screening of the corresponding sub-network.

[0055] The conduction probability of each network node in the process parameter iteration relationship network is used to determine the node orientation of each network node in the subnetwork, and the terminal node with the maximum conduction probability orientation is used as the parameter adjustment item of the state deviation; this parameter adjustment item will represent the corresponding data item for judging the expected state value of each element, and the value of this data item will be used to describe the value of the state mapping table after the relative relationship is met.

[0056] At this time, the conduction probability will be screened using a probability threshold to obtain the node orientation of each network node in its sub-network. For each node, all outgoing edges pointing to other nodes are found, and then its conduction probability is identified to determine whether the conduction probability of all its outgoing edges is greater than the probability threshold. The probability threshold can be set based on the average value of the conduction probability used in historical data. Then, the outgoing edges greater than the probability threshold are used as their node orientation at this time; the part less than the probability threshold is marked as a pending node, and then the incoming edges of the pending node are checked, that is, the incoming edges from other nodes to the current pending node. If the number of incoming edges is greater than 2 and the conduction probability of the incoming edge is greater than the probability threshold, the node orientation of the incoming edge is inherited; otherwise, it is marked as a candidate termination node, which means that the node at this time does not have a conduction situation with obvious state deviation from other nodes.

[0057] Then, the number of outgoing edges of all candidate terminal nodes is determined. If the number of outgoing edges is 0 or the conduction probability of the outgoing edge is less than the probability threshold, and the same situation occurs in three consecutive sampling periods, the candidate terminal node that meets the conditions is selected as the terminal node. When the maximum conduction probability orientation is selected, the part of each node with the maximum conduction probability when outgoing edges is selected for judgment to find the planting node that meets the conditions at this time. This terminal node will represent the stable part under the current state deviation, which will serve as the control closed-loop point to prevent excessive adjustment of parameters.

[0058] As for the implementation method of setting the expected state value of each element, the method includes: using the average value of the parameter adjustment item of the state deviation under multiple iterations as the expected state value corresponding to each element.

[0059] As shown in Table 1, after completing the extraction and analysis of condition factors and relative data, its state mapping table should be expressed as shown below.

[0060] Table 1. State mapping diagram

[0061]

[0062] Table 1 illustrates the data format represented by the current state mapping table after identifying the conditional factors and corresponding expected state values. These data represent the extraction of relevant conditional factors from the partial data set represented by the operation points after identifying multiple operation points coupled to critical points in the state mapping table. These conditional factors are then associated with the data in the corresponding operation point couplings to determine the multiple sets of expected state values ​​required for the current reaction. The elements shown in Table 1 represent partial data present in the cellulose ether alkalization reaction. These data will indicate whether the cellulose ether alkalization reaction can proceed normally, thereby detecting whether there are corresponding problems in the entire process flow. Ultimately, the corresponding procedures of the cellulose ether alkalization reaction can be adjusted based on the existing expected state values, completing rapid control measures such as data rollback and equipment deactivation.

[0063] The first row in Table 1 illustrates the swelling control of cellulose pulp. The two main conditional factors are alkali concentration and immersion temperature. The alkali concentration can destroy hydrogen bonds and promote penetration, and the immersion temperature will cause degradation when it is greater than 45 degrees Celsius. At this time, these two conditional factors will mainly affect the swelling degree-related content. When the swelling degree is greater than 80%, the alkalization uniformity can be ensured. That is, when the conditional factors are within the relevant value range, the swelling degree can meet the requirements, which means that the elements traversed at this time are in line with expectations. As for the second row, it explains the aging control of alkali cellulose. The two main conditional factors are aging temperature and CO2 concentration. When the CO2 concentration is less than 200ppm, it can prevent carbonation. As for the aging temperature, it will affect the degree of polymerization. At this time, it is mainly necessary to know whether the value of α-cellulose can meet the requirements under the corresponding alkali content. If it cannot meet the requirements, the nitrogen protection system may be triggered. As for the purpose of the last two rows, the purpose is the same as the expression of the above two rows. The conditional factors need to meet certain values. The factors that need to be checked can meet the corresponding values. That is, multiple data-related contents are combined to express whether there is any abnormality in some part of the current cellulose ether alkalization process.

[0064] Preferably, DS as described in Table 1 represents the degree of substitution, which refers to the proportion of hydroxyl groups (-OH) on each glucose unit in a polysaccharide such as cellulose or starch that are substituted with other groups. MS represents the molar substitution, which, for side-chain polymers such as hydroxyalkyl cellulose, describes the total number of substituted groups attached to each glucose unit, including further substitution on the side chains.

[0065] It should be noted that the conditional factors, expected state values, and current state values ​​appearing in Table 1 are all multiple nodes with connected relationships in the process parameter iterative relationship network. These nodes represent the associated factors and relative influencing factors of each element. The expected state value extracted at this time is based on the state value connected to the current element as a condition, and the value range that a certain node needs to reach is ultimately required to ensure that the overall reaction process can maintain the preset standard.

[0066] In one embodiment of the present invention, in the state simulation module, it is necessary to use the expected state value to identify each process parameter to check whether there is an abnormality, and based on the condition factor corresponding to the expected state value, check whether multiple process parameters corresponding to the condition factor are synchronously abnormal. Then, the process parameters with abnormalities are combined according to the abnormality type corresponding to the abnormal diffusion state to generate an abnormal state tree related to abnormal diffusion. According to the sub-nodes and other contents divided by the abnormal state tree, the control reaction items that need to be adjusted and controlled are obtained.

[0067] like Figure 4 As shown, the implementation method of the state simulation module includes: obtaining the abnormal type of each process parameter under the abnormal diffusion state, combining the mapping relationship between each process parameter and the expected state value to form each mapping subgroup. At this time, the mapping relationship between each process parameter and the expected state value represents the process parameter combination contained in each element in the state mapping table when obtaining the expected state value. After extracting the mapping relationship between these process parameters and the corresponding expected state values, a data set of multiple process parameter combinations is formed. This data set will represent a data set of multiple abnormal type combinations under the premise of correlation between multiple conditional factors.

[0068] The weight of each mapping subgroup is set based on the frequency of occurrence of the abnormal type in each mapping subgroup. In this case, the abnormal type represents the abnormal pattern corresponding to the abnormal occurrence of multiple parameters. The weight of each mapping subgroup is set based on the ratio of the frequency of occurrence of the current batch to the historical frequency. The data corresponding to each mapping subgroup is used as the node of the abnormal status tree to construct the abnormal status tree.

[0069] As shown in Table 2, the process parameters included in the mapping subgroup and the values ​​of the relative expected state values.

[0070] Table 2. Mapping subgroups

[0071]

[0072] Table 2 shows the combination of multiple process parameters. These process parameters are combined and extracted according to the association between multiple process parameters in the sub-network extracted from the process parameter iterative relationship network. Then the expected state value range mentioned later will directly describe the expected values ​​of these process parameters. These values ​​are partially different from the contents expressed in Table 1. Table 1 shows that the expected state value is the value that describes the content that the current element needs to view. The expected state value in Table 2 directly indicates the value range that the process parameter normally needs under the current abnormality identification to illustrate the relative mapping relationship; as for the currently identified mapping subgroups, there are more than what is shown in the current Table 2.

[0073] In addition, each row in Table 2 represents a mapping subgroup. When the mapping subgroups are combined into an abnormal state tree, each process parameter in the mapping subgroup can be used as a node of the abnormal state tree. That is, the implementation method of the abnormal state tree includes: taking all the data corresponding to the mapping subgroup as its root node, and then using Gini impurity to divide it into multiple child nodes. Each child node can represent the data under each abnormal type. Gini impurity mainly divides different types of data when dividing data to make all data categories more unified; that is, when the Gini impurity value of each node under the abnormal state tree is the largest, the child nodes and leaf nodes of the abnormal state tree are divided, and then the abnormal state tree is continuously divided. Finally, the data contained in its leaf nodes can be the data set corresponding to a certain parameter in its mapping subgroup. At this time, the setting of the abnormal state tree is completed.

[0074] The shortest path corresponding to each leaf node in the abnormal state tree is obtained, and the process parameters included in the shortest path are used as the output control response items. The shortest path is identified by summing the weights of each node in the abnormal state tree to identify the path with the smallest sum of weights on this shortest path, which is used as the shortest path. The weight of each node in the abnormal state tree is calculated in the same way as the weight of the mapping subgroup, and is set by the ratio of the occurrence frequency of the data in the node to the historical frequency. If the data is multi-type, the frequency of occurrence of multiple types is used for calculation. This is to obtain the shortest path for each leaf node in the abnormal state tree. By using the minimum weight and path, the root node of the abnormal propagation, such as the sudden drop in pH value, and the key propagation path, such as pH→temperature→etherifying agent flow rate, can be quickly located, avoiding computing resources on non-critical branches. The processing content here is different from the analysis of the relationship between conditional factors in the process parameter iterative relationship network. It mainly focuses on decomposing the main path of each process parameter when the main abnormality occurs. Then, differentiated responses can be made for the process parameters included in each abnormality type. At the same time, the cellulose ether alkalinization reaction has strong temporal coupling. For example, the temperature must respond within 30 seconds after the pH changes. The shortest path can capture this lag effect and avoid control delays caused by a too long path.

[0075] In one embodiment of the present invention, the control training path represents multiple groups of paths included in the control reaction item, and then the number of path segments is used as its advancement step number, and the advancement step number can represent the number of process parameters existing on the current path; then, the process parameters with problems on the path are identified, indicating the parts that currently need to adjust the control strategy, and these parts are adjusted to obtain control indicators to complete the rapid response processing of the current cellulose ether alkalization reaction.

[0076] like Figure 5 As shown, the implementation of the feedback processing module involves mapping the process parameters in the control training path to the control strategy, determining the single-point alarm information for the process parameters in the control training path within the control strategy, and traversing the control strategy by advancing the number of steps to determine the alarm information chain corresponding to the process parameters. The alarm information chain is chained by combining the process parameters in each segment by advancing the number of steps, resulting in a section containing multiple groups of process parameters with abnormalities in the control strategy.

[0077] Based on the alarm information chain, data is retrieved to obtain the adjustment data corresponding to the alarm information chain in the control strategy library, and the adjustment data is mapped to the current control strategy to form multiple control indicators. These obtained control indicators are used to complete the adjustment of the current control strategy.

[0078] As shown in Table 3 and Table 4, the current control training path and alarm information chain table are displayed.

[0079] Table 3. Schematic diagram of control training path

[0080]

[0081] Table 3 shows the process parameters extracted from the control reaction items, and then the partial path conditions of these process parameters extracted from the abnormal state data. After describing the number of advancement steps of each path, it is identified whether there are problems with the process parameters, and then these problem parameters are marked as single-point alarm information at this time. The single-point alarm information is then combined to obtain the alarm information chain.

[0082] Table 4. Alarm information list

[0083]

[0084] Table 4 shows that these data form a chain data table, and then the control strategy query is performed on these data to find the control strategy corresponding to the process parameters with current problems, and then generate the relevant alarm information chain. For example, the first three rows are represented as S1→S2→S3, and the fourth row represents itself alone. At this time, the data in this alarm information chain will be corrected in sequence, and then the current values ​​that need to be adjusted are found from the preset control strategy library according to the concentration parameters under these control strategies, and then the relevant parts in the control strategy are modified. For the part of the associated control strategy in Table 4, it only represents the number of the control part corresponding to the process parameter. In addition to the current table representation, other forms of serial number marking can also be used. The modified value retrieved from the control strategy library is then output as the control indicator at this time to complete the adjustment of the current control strategy; ultimately, rapid closed-loop control of complex process abnormalities is achieved, and multivariable strong coupling of cellulose ether alkalization is completed.

[0085] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention, which are still covered by the scope of protection of the present invention.

Claims

1. An intelligent control system for cellulose ether alkalization reaction, characterized in that: include: The state acquisition module is used to obtain the state information code of each device in the cellulose ether alkalization reaction, and continuously map the state information code of each device with the collected process parameters to form a state mapping table; The reaction control module is used to traverse each element in the state mapping table, build a process parameter iterative relationship network based on the condition factors of each element under the ether alkalization reaction, and determine the expected state value corresponding to each element; The state simulation module is used to identify the abnormal diffusion state of each process parameter based on the expected state value of each element, determine the abnormal state tree corresponding to each process parameter, and use the output results of the abnormal state tree to describe the control reaction item of the ether alkalization reaction; The feedback processing module is used to obtain at least one control training path according to the control response item, and adjust the current control strategy according to the number of advancement steps on the control training path to determine various control indicators after the control strategy is adjusted.

2. The intelligent control system for cellulose ether alkalization reaction according to claim 1, characterized in that: The implementation of the status acquisition module also includes: Starting from the operation scenario corresponding to the state information encoding, the operation points and operation factors corresponding to each operation scenario are set according to the semantic identification of the process parameters in different operation scenarios; Based on the operation factors of each operation point, the critical point of each operation scenario is identified, and each operation point is coupled with the critical point of each operation scenario to obtain a state mapping map; the state mapping map is output as a state mapping table.

3. The intelligent control system for cellulose ether alkalization reaction according to claim 2, characterized in that: When identifying critical points in each operating scenario, it also includes: Determine the operating factor probability of each operating point, map the historical batch status based on the operating factor probability of each operating point, and determine the operating boundary of each operating point; The historical batch state mapping data is performed based on the operation factor probability of each operation point, and the neighborhood buffer corresponding to each operation point is constructed; The intersection area of ​​the neighborhood buffer and the operation boundary of each operating point is identified and defined as the critical influence area; the boundary contour of each critical influence area is extracted, and each operating point on the boundary contour is used as the output critical point.

4. The intelligent control system for cellulose ether alkalization reaction according to claim 1, characterized in that: The implementation of the reaction control module includes: Divide each element in the state mapping table according to the mapped data, and set at least one condition factor for each element when dividing; Obtain the topological relationship between each condition factor, and based on the correlation coefficient between each condition factor in the topological relationship, use each condition factor as a network node to generate a relationship network; Perform time-series weighting on each network node in the relationship network, and use the time-series weighted relationship network as the process parameter to iterate the relationship network.

5. The intelligent control system for cellulose ether alkalization reaction according to claim 4, characterized in that: The expected state value of each element is expressed as: Calculate the state deviation of each network node in the process parameter iterative relationship network, determine the transmission probability of the state deviation between each network node, and use the transmission probability and weight of the network node to extract the parameter adjustment item of the state deviation; According to the value of the parameter adjustment item of the state deviation after iteration, the expected state value of each element is set.

6. The intelligent control system for cellulose ether alkalization reaction according to claim 5, characterized in that: The implementation methods of parameter adjustment items for extracting state deviation include: According to the weight of each network node in the process parameter iteration relationship network, a sub-network of the process parameter iteration relationship network is extracted, and each network node in the sub-network is not an isolated node; The conduction probability of each network node in the process parameter iteration relationship network is used to determine the node orientation of each network node in the subnetwork, and the terminal node with the maximum conduction probability orientation is used as the parameter adjustment item of the state deviation.

7. The intelligent control system for cellulose ether alkalization reaction according to claim 5, characterized in that: The implementation methods for setting the expected state value of each element include: The average value of the parameter adjustment item of the state deviation under multiple iterations is used as the expected state value corresponding to each element.

8. The intelligent control system for cellulose ether alkalization reaction according to claim 1, characterized in that: The implementation of the state simulation module includes: Obtain the abnormal type of each process parameter under the abnormal diffusion state, and combine the mapping relationship between each process parameter and the expected state value to form each mapping subgroup; The weight of each mapping subgroup is set based on the frequency of occurrence of the abnormal type in each mapping subgroup, and the data corresponding to each mapping subgroup is used as the node of the abnormal status tree to construct the abnormal status tree; The shortest path corresponding to each leaf node in the abnormal state tree is obtained, and the process parameters contained in the shortest path are used as the output control response items.

9. The intelligent control system for cellulose ether alkalization reaction according to claim 8, characterized in that: The implementation method of the abnormal state tree includes: taking all the data corresponding to the mapping subgroup as its root node, and dividing the abnormal state tree into child nodes and leaf nodes when the Gini impurity value of each node under the abnormal state tree is maximized to complete the setting of the abnormal state tree.

10. The intelligent control system for cellulose ether alkalization reaction according to claim 1, characterized in that: The implementation of the feedback processing module includes: Mapping the process parameters in the control training path to the control strategy, determining the single-point alarm information of the process parameters in the control training path in the control strategy, and traversing the control strategy with the advancement step number to determine the alarm information chain corresponding to the process parameters; Data retrieval is performed based on the alarm information chain to obtain the adjustment data corresponding to the alarm information chain in the control strategy library, and the adjustment data is mapped to the current control strategy to form multiple control indicators.

Images