A method and system for intelligent control of steel pipe galvanizing equipment

Abstract

This invention provides an intelligent control method and system for steel pipe galvanizing process equipment, belonging to the field of intelligent production control technology. The method includes: collecting raw process operation datasets for galvanized steel pipe production and constructing multiple process order groups; generating multiple order state sequences and identifying multiple order segments within each order state sequence; constructing a qualified sample set and a risk sample set; extracting multiple statistical features corresponding to each order segment in the qualified and risk sample sets; classifying the multiple order segments by order type; constructing risk and stable order sets corresponding to the multiple order segments; traversing the qualified and risk sample sets; constructing multiple candidate recovery paths for each order state sequence and fusing them to generate multiple process optimization model libraries; and optimizing the control of the steel pipe galvanizing process equipment based on the process optimization model libraries. This invention improves the level of intelligent control in the production process of galvanized steel pipe products.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology in production, and in particular to an intelligent control method and system for steel pipe galvanizing process equipment. Background Technology

[0002] The galvanizing process for steel pipes typically involves multiple continuous operations, including pretreatment, heating, zinc immersion, cooling, and shaping. These processes involve diverse and highly coupled parameters, with production cycle time and environmental conditions significantly impacting coating quality. With increasing automation, existing equipment is often equipped with various sensors for temperature, speed, liquid level, and air pressure, and these parameters can be automatically adjusted via programmable logic controllers (PLCs) or host computer systems.

[0003] However, traditional control methods, such as those based on single-parameter feedback or fixed control logic, rely on empirical thresholds or model-based formulas for adjustment. While these methods can achieve overall process stability, they may struggle to capture and correct subtle quality deviations when faced with complex operating conditions, raw material variations, or slight fluctuations in equipment status. For instance, when production conditions are generally normal but minor defects such as uneven coating or differences in surface gloss occur, parameter feedback control logic may fail to identify the causes of these subtle changes, making it difficult to achieve targeted adaptive adjustments through existing parameter feedback.

[0004] As industrial process control systems evolve towards intelligence and collaboration, a key research focus in intelligent manufacturing is how to achieve dynamic identification and self-recovery control of equipment operating status through structured representation and correlation learning of multi-source process data without relying on complex physical models. There is an urgent need to establish an intelligent control method capable of autonomously identifying risk states and matching optimal recovery paths based on historical operating patterns to improve the stability and product consistency of galvanized steel pipe production systems. Summary of the Invention

[0005] This invention proposes an intelligent control method and system for steel pipe galvanizing process equipment, aiming to solve at least one technical problem existing in the above-mentioned background art.

[0006] The first aspect of this invention provides an intelligent control method for steel pipe galvanizing process equipment, comprising:

[0007] Collect raw data sets of process operation data for galvanized steel pipe production, including multiple sets of historical production trajectory data, and construct multiple process order groups based on the raw data sets of process operation data;

[0008] Based on multiple process order groups, sliding window order matching is performed on each group of historical production trajectory data to generate an order state sequence for each group of historical production trajectory data, and multiple order segments of each order state sequence are identified.

[0009] Multiple order state sequences are classified to construct qualified sample sets and risk sample sets, and multiple statistical features corresponding to each order segment in the qualified sample set and risk sample set are extracted respectively;

[0010] Based on statistical characteristics, the frequency difference parameter and duration difference parameter of the order segments are calculated, and the order types of multiple order segments are divided to construct risk order sets and stable order sets corresponding to multiple order segments.

[0011] Based on the risk order set and the stable order set, the qualified sample set and the risk sample set are traversed to construct multiple candidate recovery paths for each order state sequence. Multiple candidate recovery paths are merged to construct multiple process optimization model libraries. Based on the process optimization model libraries, the control optimization of steel pipe galvanizing process equipment is realized.

[0012] Preferably, a sliding window order matching is performed on each group of historical production trajectory data according to multiple process order groups to generate an order state sequence for each group of historical production trajectory data, and multiple order segments of each order state sequence are identified, including:

[0013] Determine the order status items for each process order group. The order status items include the order status parameters corresponding to the matching items and the unmatched items, respectively.

[0014] Sliding window order matching is performed on historical production trajectory data to identify the matching status of historical production trajectory data with multiple process order groups in each window, and to construct the order code of historical production trajectory data in each window;

[0015] Generate an order state sequence of historical production trajectory data based on multiple order codes, traverse each order state sequence, extract the duration feature of each order code, and select multiple order segments from multiple order codes based on the duration feature.

[0016] Preferably, the frequency difference parameter and duration difference parameter of the order segments are calculated based on statistical characteristics, and the order types of multiple order segments are classified to construct risk order sets and stable order sets corresponding to multiple order segments, including:

[0017] The statistical characteristics of ordered segments include the segment frequency characteristics and segment duration characteristics in the qualified sample set and the risk sample set, respectively;

[0018] Frequency difference parameters are calculated based on the frequency characteristics of ordered segments in the qualified sample set and the risk sample set, respectively. Duration difference parameters are calculated based on the duration characteristics of ordered segments in the qualified sample set and the risk sample set, respectively.

[0019] The frequency difference parameter and the duration difference parameter are fused to calculate the sample state weight of each order segment. The order type of each order segment is divided by combining multiple preset state ranges, and the order type label of each order segment is determined. The order type label includes stable order, neutral order and risk order. Based on multiple stable order and risk order, a stable order set and a risk order set are constructed.

[0020] Preferably, based on the risk order set and the stable order set, the qualified sample set and the risk sample set are traversed to construct multiple candidate recovery paths for each order state sequence. These multiple candidate recovery paths are then fused to construct multiple process optimization phantom libraries, including:

[0021] Traverse multiple order state sequences in the qualified sample set and the risk sample set, identify the risk order and stable order in each order state sequence based on the risk order set, and extract multiple candidate recovery paths in each order state sequence based on recovery logic, where the starting node of the candidate recovery path is the risk order and the target node is the stable order;

[0022] Based on the steel pipe galvanizing process, multiple candidate recovery paths are classified into process stages, and a candidate recovery sample set corresponding to each process stage is constructed. The multiple candidate recovery paths in each candidate recovery sample set are subjected to state evolution fusion based on state semantic constraints to generate a process optimization model library for each candidate recovery sample set. The process optimization model library includes recovery models for multiple process stages.

[0023] Preferably, the process optimization phantom library for each candidate recovery sample set is generated by performing state evolution fusion based on state semantic constraints on multiple candidate recovery paths in each candidate recovery sample set, including:

[0024] Determine the starting order code and target order code of each candidate recovery path in the candidate recovery sample set. Calculate the starting state distance and target state distance between any two candidate recovery paths based on the starting order code and target order code. Determine the state semantic similarity between two candidate recovery paths based on the starting state distance and target state distance.

[0025] Extract the flipped sequence from each candidate recovery path except for the starting order code and the target order code, perform statistical analysis on each flipped sequence, generate the frequency feature vector of each flipped sequence, and calculate the state evolution similarity between any two candidate recovery paths based on the frequency feature vector.

[0026] A path similarity matrix is ​​constructed based on state semantic similarity and state evolution similarity. Multiple candidate recovery paths in the candidate recovery sample set are clustered based on the path similarity matrix to obtain multiple process stage recovery modalities for each candidate recovery sample set.

[0027] Preferably, for the process stage recovery phantom, it further includes:

[0028] Multiple candidate recovery paths in the candidate recovery sample set are clustered to produce multiple recovery path clusters. Based on the starting order code and the target order code of each candidate recovery path, the risk state and the stable state of the recovery path cluster are constructed. Based on multiple candidate recovery paths, the motif control vector of the recovery path cluster is constructed. By fusing the risk state, the stable state and the motif control vector, the process stage recovery motif of each recovery path cluster is obtained.

[0029] A second aspect of the present invention provides an intelligent control system for steel pipe galvanizing process equipment, for implementing the above-mentioned intelligent control method for steel pipe galvanizing process equipment, comprising:

[0030] The production data acquisition module is used to collect the raw dataset of the process operation of galvanized steel pipe production, including multiple sets of historical production trajectory data, and to construct multiple process order groups based on the raw dataset of the process operation.

[0031] The order fragment recognition module is used to perform sliding window order matching on each group of historical production trajectory data based on multiple process order groups, generate an order state sequence for each group of historical production trajectory data, and identify multiple order fragments in each order state sequence.

[0032] The sample set construction module is used to classify multiple order state sequences to construct qualified sample sets and risk sample sets, and extract multiple statistical features corresponding to each order segment in the qualified sample set and risk sample set respectively;

[0033] The order type classification module is used to calculate the frequency difference parameter and duration difference parameter of order segments based on statistical characteristics, classify multiple order segments into order types, and construct risk order sets and stable order sets corresponding to multiple order segments;

[0034] The galvanizing process optimization module is used to traverse the qualified sample set and the risk sample set based on the risk order set and the stable order set, construct multiple candidate recovery paths for each order state sequence, merge multiple candidate recovery paths to construct multiple process optimization model libraries, and realize the control optimization of steel pipe galvanizing process equipment based on the process optimization model library.

[0035] The present invention has the following beneficial effects:

[0036] This invention constructs a multi-dimensional process order group and generates an order state sequence based on sliding window matching, thereby achieving a structured state characterization of the entire galvanizing production process. By quantifying the differences in frequency and duration of order segments in qualified and risk samples, it accurately identifies and constructs risk order sets and stable order sets, effectively distinguishing potential operating modes that cause subtle fluctuations in coating quality. Furthermore, based on the conversion relationship between risk and stable orders, it mines and integrates historical samples to form a multi-stage process optimization model library, significantly improving the early and sensitive perception of gradual instability risks caused by factors such as raw material fluctuations and equipment status changes. It can accurately match the optimal recovery path according to the real-time status and drive the collaborative adaptive regulation of multiple process parameters through model control vectors, thereby proactively correcting the process state before subtle quality deviations appear, effectively suppressing the generation of hidden defects such as uneven coating and gloss differences, and significantly improving the consistency of galvanized steel pipe products and the level of intelligent control in the production process. Attached Figure Description

[0037] Figure 1 This is a schematic flowchart of an intelligent control method for steel pipe galvanizing equipment according to an embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the structure of an intelligent control system for steel pipe galvanizing equipment in an embodiment of the present invention. Detailed Implementation

[0039] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0040] To facilitate understanding of this invention, the following description, using a practical production scenario of steel pipe galvanizing, further illustrates a method and system for intelligent control of equipment in the steel pipe galvanizing process. This method is primarily applied to the process control of continuous steel pipe galvanizing production lines. Through structured analysis and pattern mining of equipment operating data, it achieves the identification and adaptive control of production status. Please refer to [link to relevant documentation]. Figure 1 As one aspect of the present invention, an intelligent control method for steel pipe galvanizing process equipment is provided, including the following:

[0041] Step S1: Collect the original process operation dataset of galvanized steel pipe production, including multiple sets of historical production trajectory data, and construct multiple process order groups based on the original process operation dataset.

[0042] Specifically, in the galvanizing process of steel pipes, the strip steel sequentially passes through multiple process units, including the preheating zone, reduction zone, zinc immersion bath, cooling zone, and shaping zone, to obtain galvanized steel pipe products. To comprehensively analyze the equipment's operating status, sensors deployed at key locations, such as those for temperature, air pressure, liquid level, belt speed, current, and gas concentration, can collect raw data sets of the production line's process operation over multiple historical periods. These data, combined with product quality indicators corresponding to different production signal data, such as coating thickness, form a data foundation containing multiple sets of historical production trajectory data.

[0043] After obtaining the raw data, several process order groups can be preset according to the correlation and control logic of different process links. Each process order group consists of two or more types of process parameters, used to describe the relative relationship or trend of change between parameters. For example, from the perspective of spatial relationship, for multi-point temperature monitoring data in the zinc pot, the temperature of the upper heating zone can be defined as higher than that of the lower heating zone, and the air pressure in the left zone can be defined as higher than that in the right zone; from the perspective of equipment correlation, the air knife pressure changes proportionally to the belt speed; from the perspective of individual process parameters, the liquid level is lower than the median threshold, etc. Those skilled in the art can combine the actual configuration of the collected production equipment and the number and location of the sensor network to reasonably set multiple process order groups to characterize the relative relationship between different process parameters. This embodiment does not specifically limit them. Through multiple process order groups, the structured state of the process can be characterized without relying on specific numerical values.

[0044] Step S2: Perform sliding window order matching on each group of historical production trajectory data according to multiple process order groups to generate an order state sequence for each group of historical production trajectory data, and identify multiple order segments in each order state sequence.

[0045] Specifically, for the constructed multiple process order groups, a sliding window matching process is performed on the historical production trajectory data of each group. Within each time window, it is determined whether the relationship between the order groups is valid, and an order code for that time window is generated accordingly. The order codes under consecutive time windows are arranged in chronological order to form a complete sequence of order states. Furthermore, by detecting the continuous occurrence of order codes in the sequence, multiple order segments can be identified to represent process segments that maintain similar operating patterns within a certain period of time.

[0046] In one optional implementation, generating an order state sequence for each set of historical production trajectory data, and identifying multiple order segments for each order state sequence includes:

[0047] The order status items for each process order group are determined. The order status items include the order status parameters corresponding to matched items and unmatched items. Specifically, each process order group is marked as a matched item when a match is successful and as an unmatched item when the set relationship is not met. The corresponding order status parameters are recorded as "1" and "0" respectively, which are used to indicate the establishment status of the order group at the current sampling time.

[0048] Then, sliding window order matching is performed on the historical production trajectory data to identify the matching status of the historical production trajectory data with multiple process order groups in each window. For example, if the temperature of the upper heating zone is higher than that of the lower heating zone in the current time window, the order status parameter of the process order group is recorded as 1. The matching results of all order groups under multiple windows are combined in the same order to construct the order code corresponding to different time windows. That is, it is composed of the order status parameter corresponding to each process order group, which is used to describe the comprehensive process status of different time periods.

[0049] The sequence codes obtained from multiple windows in historical production trajectory data are arranged in chronological order to generate a sequence of order states for the historical production trajectory data. Then, each sequence is traversed to extract the duration feature of each sequence code, i.e., the number of consecutive occurrences of each sequence code in the sequence is counted. When the number of consecutive occurrences of a sequence code exceeds a preset duration threshold, it indicates that the process state it represents has maintained a relatively stable period during production. Using this method, multiple order segments are selected from multiple sequence codes based on the duration feature. Each order segment corresponds to a relatively stable operating state in the galvanized steel pipe production process.

[0050] Step S3: Classify multiple order state sequences to construct a qualified sample set and a risk sample set, and extract multiple statistical features corresponding to each order segment in the qualified sample set and the risk sample set respectively.

[0051] Specifically, multiple order state sequences are classified according to product quality indicators, constructing a qualified sample set and a risk sample set. The qualified sample set corresponds to the operational trajectory where the production result quality meets the standards, while the risk sample set corresponds to the operational trajectory with minor flaws or deviations. For each order segment identified above, its frequency of occurrence, average duration, and other characteristic parameters are statistically analyzed in both the qualified and risk sample sets to reflect the performance differences of that order under different quality states.

[0052] It is worth noting that for sample data with significant quality problems, since the corresponding causes of failure are relatively clear, such as obvious uneven coating, temperature runaway, sensor malfunction, and other serious issues, the order sequence of these samples is unstable. Therefore, the analysis mainly focuses on the transitional segment between risk samples and qualified samples—that is, samples where the overall product is qualified but local parameters have deviations, thickness fluctuations, or minor defects—and samples where the product fully meets quality requirements. For risk samples with minor defects, those skilled in the art can reasonably set various indicators such as coating thickness, uniformity, and appearance defects based on actual production standards.

[0053] Step S4: Calculate the frequency difference parameter and duration difference parameter of the order segments based on statistical characteristics, classify the order types of multiple order segments, and construct risk order sets and stable order sets corresponding to multiple order segments.

[0054] Specifically, for the statistical features of each extracted order segment, the statistical features corresponding to each order segment can be the following indicators, such as the segment frequency feature, which characterizes the number of times the order segment appears in different sample types, and the segment duration feature, which reflects the average duration of the order segment at each occurrence. Based on these statistical feature information, the frequency difference parameter and duration difference parameter of each order segment are calculated and fused into the sample state weight.

[0055] Based on the numerical value and direction of the sample state weights, multiple order segments are classified into categories. For example, states with high weights and frequent occurrences in risky samples are classified as risky orders, while those with low weights and frequent occurrences in qualified samples are classified as stable orders, thus constructing risky order sets and stable order sets respectively. These represent operational characteristics that may lead to quality deviations and the ideal operational states that should be maintained, respectively.

[0056] In one optional implementation, frequency difference parameters and duration difference parameters of order segments are calculated based on statistical characteristics. Multiple order segments are then classified into order types, and risk order sets and stable order sets corresponding to these segments are constructed, including:

[0057] Frequency difference parameters are calculated based on the frequency characteristics of ordered segments in the qualified sample set and the risk sample set, respectively. Duration difference parameters are calculated based on the duration characteristics of ordered segments in the qualified sample set and the risk sample set, respectively.

[0058] In this process, for the frequency difference parameter, the difference between the frequency characteristics of ordered segments in the risk sample set and the qualified sample set is calculated, along with the sum of their frequency characteristics. The ratio of this ratio is used as the frequency difference parameter, ranging from -1 to 1. A value closer to 1 indicates that the ordered segment is more common in the risk samples. For the duration difference parameter, the difference between the duration characteristics of ordered segments in the risk sample set and the qualified sample set is calculated, along with the sum of their duration characteristics. The ratio is used as the duration difference parameter. Similarly, the duration difference parameter ranges from -1 to 1, with a value closer to 1 indicating that the ordered segment is more persistent in the risk samples.

[0059] After calculating the frequency difference parameter and the duration difference parameter, the frequency difference parameter and the duration difference parameter are further fused. The fusion weights of the two can be reasonably set to represent the relative importance of the probability difference and the duration difference. The frequency difference parameter and the duration difference parameter are weighted and fused by the set fusion weights to obtain the sample state weight of the ordered segment.

[0060] Based on this, each order segment can be classified into order types. To analyze the differences between qualified and risky samples, multiple order segments can be divided into three order types, with θ as the boundary, dividing into three preset state ranges: >θ, -θ to θ, and <-θ, respectively denoted as risky order, neutral order, and stable order. This yields an order type label for each order segment. Risky order represents order segments that tend to appear in risky samples, neutral order represents regular order segments that exist in both risky and qualified samples without a clear tendency, and stable order represents order segments that tend to appear in qualified samples. Finally, all segments marked as stable order are aggregated to form a stable order set, which characterizes the operating mode that is prevalent in qualified products and reflects the steady-state characteristics of the process; all segments marked as risky order are aggregated to form a risky order set, which characterizes the abnormal patterns that occur frequently in products with defects or deviations.

[0061] Step S5: Based on the risk order set and the stable order set, traverse the qualified sample set and the risk sample set, construct multiple candidate recovery paths for each order state sequence, merge multiple candidate recovery paths to construct multiple process optimization model libraries, and realize the control optimization of steel pipe galvanizing process equipment based on the process optimization model library.

[0062] Specifically, based on the constructed risk and stable order sets, multiple order state sequences are traversed across the qualified and risk sample sets. Order transition segments from risk to stable states that may exist in each order state sequence are extracted, forming multiple candidate recovery path sets. These sets are then aggregated and filtered to generate multiple representative process optimization models, thus constructing a process optimization model library. During the production of galvanized steel pipes, the equipment control system, in real-time operation, can automatically call the corresponding optimization model based on the matching results between the current state and historical models, outputting control adjustment commands to achieve adaptive adjustment and control optimization of the galvanizing process equipment.

[0063] In one optional implementation, based on the risk order set and the stable order set, the qualified sample set and the risk sample set are traversed to construct multiple candidate recovery paths for each order state sequence. These multiple candidate recovery paths are then fused to construct multiple process optimization phantom libraries, including:

[0064] The algorithm iterates through multiple order state sequences in both the qualified and risky sample sets. Based on the risky order set, it identifies the risky and stable orders within each order state sequence. From these identified risky and stable orders, it selects risky and stable order nodes and constructs one or more candidate recovery paths that may exist within the order state sequence. Specifically, based on the recovery logic of transitioning from a risky state to a stable state, the algorithm scans the order state sequence. Upon encountering the first risky order, it marks it as the starting node of a candidate recovery path. The search continues backward from the starting node. For the first stable order that appears, it is marked as the target node of the candidate recovery path. The sequence consisting of multiple order codes involved from the starting node to the target node is recorded as the candidate recovery path.

[0065] The galvanizing process of steel pipes can generally be divided into several continuous stages, including heating, immersion in zinc, cooling, and shaping. The operational characteristics and control logic of each stage differ significantly. To improve the targeting of path analysis, based on the information of the steel pipe galvanizing process flow, multiple candidate recovery paths are classified according to the process flow stages according to the time position of each candidate recovery path, constructing a candidate recovery sample set corresponding to each process stage. The steel pipe galvanizing process flow may differ between the production processes of different products. Those skilled in the art can reasonably determine multiple process stages according to actual needs; this embodiment does not specifically limit them.

[0066] After constructing multiple candidate recovery sample sets, state evolution fusion based on state semantic constraints is performed on multiple candidate recovery paths in each candidate recovery sample set to generate a process optimization model library for each candidate recovery sample set.

[0067] In one optional implementation, the process optimization phantom library for generating each candidate recovery sample set includes:

[0068] From a state semantic dimension, the starting and target order codes of each candidate recovery path in the candidate recovery sample set are determined, i.e., the order codes corresponding to the starting and target nodes of the path, respectively. The starting state distance is obtained by calculating the Hamming distance between the starting and ending order codes of any two candidate recovery paths, and the target state distance is obtained by calculating the Hamming distance between the target order codes of two candidate recovery paths. The mean of the starting and target state distances between two candidate paths is calculated, and the mean of the starting and target state distances is normalized using the total number of order bits, i.e., the total number of order state parameters in the order code, to obtain the state semantic similarity between the two candidate recovery paths.

[0069] From the perspective of state evolution, the flip sequence other than the starting order code and the target order code is extracted from each candidate recovery path. The flip sequence includes the flip state of the process order group in any two adjacent order codes, that is, the difference change of the order state item. If the order state parameter of a certain process order group is inconsistent in two adjacent order codes, it is recorded as a flip state. The corresponding process order group and the flip direction are recorded. For example, in the process order group where the wind pressure in the left area is greater than the wind pressure in the right area, the order state parameter changes from "1" to "0" in two adjacent order codes.

[0070] After extracting the flip sequence from each candidate recovery path, frequency statistical analysis is performed on each flip sequence. This includes calculating the flip frequency of each process order group under different flip directions to construct a frequency feature vector for each flip sequence. Each element in the frequency feature vector represents the flip frequency of one of the process order groups under a specific flip direction. Based on the frequency feature vector, the state evolution similarity between any two candidate recovery paths is calculated. For example, the cosine similarity of two frequency feature vectors is calculated, and the absolute value is taken as the state evolution similarity between the candidate recovery paths.

[0071] Finally, a path similarity matrix for the candidate recovery sample set is constructed based on state semantic similarity and state evolution similarity. This includes correcting the state evolution similarity using state semantic similarity, calculating the product of state semantic similarity and state evolution similarity, and obtaining the path similarity between any two candidate recovery paths in the candidate recovery sample set. State semantic similarity constrains the similarity of the dynamic evolution process at the state semantic level; only when the initial states and target states of two paths are similar is the corresponding dynamic evolution similarity considered valid. Based on the path similarity matrix, distance clustering is performed on multiple candidate recovery paths in the candidate recovery sample set. Using path similarity as the distance index, hierarchical clustering or the DBSCAN algorithm is employed to cluster multiple candidate recovery paths into multiple recovery path clusters.

[0072] For each recovery path cluster, the risk state and stable state of the recovery path cluster are determined based on the starting order codes and target order codes of multiple candidate recovery paths. Specifically, the distribution probability of multiple starting order codes with respect to different order positions in each process order group is statistically analyzed. That is, the aforementioned order state parameters are denoted as "1" and "0" to represent two order positions, respectively. Order positions with a distribution probability greater than a significant threshold are added to the risk state. For process order groups without order positions with a distribution probability greater than a significant threshold, selection can be made through path weighting. For example, the distance between each path and the center of the initial state cluster is calculated and normalized to determine the weight, where the center of the initial state cluster can be selected as the mean of the starting order codes of multiple candidate recovery paths. The closer the candidate recovery path is to the center of the initial state cluster, the greater its corresponding weight. For process order groups without order positions with a distribution probability greater than a significant threshold, the order positions of these process order groups in different candidate recovery paths are determined. The weight of each candidate recovery path is used as the influence parameter of these order positions. The sum of the influence parameters corresponding to the two order positions for each process order group is calculated, and the order position with the larger influence parameter is added to the risk state. Thus, the risk state of the recovery path cluster is constructed. Similarly, the construction of the stable state is also carried out by the above-mentioned probability statistics based on the order position and the path weighting based on the distance between the path and the center of the target state cluster. This will not be elaborated here.

[0073] After determining the risk state and steady state of the recovery path cluster, a modal control vector for each recovery path cluster is constructed based on multiple candidate recovery paths. Specifically, the analysis is conducted from the perspectives of regulation direction, regulation amplitude, response delay, and regulation rate.

[0074] For multiple candidate recovery paths within a recovery path cluster, at the adjustment direction level, the parameter values ​​for each process parameter in the risk and steady states are determined. The deviation between the steady and risk states is calculated; if it is greater than 0, the adjustment direction is positive; otherwise, it is negative. The adjustment directions of multiple candidate recovery paths are statistically analyzed, and the adjustment direction with the highest frequency is selected as the dominant adjustment direction for the process parameter in that recovery path cluster. At the adjustment magnitude level, the mean and variance of the deviations of multiple parameter values ​​are calculated, and the adjustment magnitude is set based on the mean and variance, for example, by selecting the mean ± variance as the adjustment magnitude. At the response delay level, the time axis of each path is normalized to the [0,1] interval to make different sample lengths comparable. The moment when each process parameter first shows a significant deviation from its initial value is determined. This significant deviation can be measured as a percentage of the mean deviation of multiple parameter values; for example, a deviation of one-third of the mean from the initial value represents a significant deviation. This determines the average start-up time for each process parameter, enabling delay analysis of multiple process parameters and obtaining adjustment response delay data—that is, how long each process parameter will be adjusted after the start of regulation compared to the first adjusted parameter. Simultaneously, for different process parameters, the adjustment rate of each parameter in multiple paths is statistically analyzed, and the mean is calculated to obtain the target adjustment rate.

[0075] Finally, based on the adjustment direction, adjustment amplitude, response delay, and target adjustment rate, a phantom control vector is constructed for each recovery path cluster. The risk state, steady state, and phantom control vector of the recovery path cluster are used as the process stage recovery phantoms of the recovery path cluster. Each process stage includes multiple process stage recovery phantoms, forming a process optimization phantom library for optimizing the control of steel pipe galvanizing equipment at different process stages.

[0076] In the process of optimizing the control of steel pipe galvanizing equipment based on the process optimization model library, for the collected real-time process data, the real-time order code and process stage corresponding to the real-time process data are determined. The real-time order code is matched with multiple process stage recovery models in the process optimization model library corresponding to the process stage to determine the risk state closest to the real-time order code. Based on the model control vector, the adjustment direction, adjustment rate, adjustment amplitude, and response delay of each process parameter are determined, and the adjustment is monitored in real time based on the steady state to control the real-time order codes corresponding to multiple process equipment to approach the steady state. This achieves adaptive control optimization based on historical experience, ensuring the high-quality production of galvanized steel pipes. It has the characteristics of self-learning and adaptive adjustment, and can automatically learn from manual detail optimization and production experience data of high-quality products, thereby improving the stability of high-quality galvanized steel pipe production.

[0077] Please see Figure 2As another aspect of the present invention, an intelligent control system for steel pipe galvanizing process equipment is provided, comprising:

[0078] The production data acquisition module is used to collect the raw dataset of the process operation of galvanized steel pipe production, including multiple sets of historical production trajectory data, and to construct multiple process order groups based on the raw dataset of the process operation.

[0079] The order fragment recognition module is used to perform sliding window order matching on each group of historical production trajectory data based on multiple process order groups, generate an order state sequence for each group of historical production trajectory data, and identify multiple order fragments in each order state sequence.

[0080] The sample set construction module is used to classify multiple order state sequences to construct qualified sample sets and risk sample sets, and extract multiple statistical features corresponding to each order segment in the qualified sample set and risk sample set respectively;

[0081] The order type classification module is used to calculate the frequency difference parameter and duration difference parameter of order segments based on statistical characteristics, classify multiple order segments into order types, and construct risk order sets and stable order sets corresponding to multiple order segments;

[0082] The galvanizing process optimization module is used to traverse the qualified sample set and the risk sample set based on the risk order set and the stable order set, construct multiple candidate recovery paths for each order state sequence, merge multiple candidate recovery paths to construct multiple process optimization model libraries, and realize the control optimization of steel pipe galvanizing process equipment based on the process optimization model library.

[0083] The above are merely specific embodiments of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for intelligent control of a steel pipe galvanizing process equipment, characterized by, The method comprises the following steps: Collecting a process operation original data set of galvanized steel pipe production, including multiple sets of historical production trajectory data, and constructing multiple process order groups based on the process operation original data set; According to the multiple process order groups, performing sliding window order matching on each set of historical production trajectory data to generate an order state sequence of each set of historical production trajectory data, and identifying multiple order segments of each order state sequence; Specifically, determining the order state item of each process order group, and the order state item includes order state parameters corresponding to the matching item and the unmatched item respectively; Performing sliding window order matching on the historical production trajectory data, identifying the matching state of the historical production trajectory data with the multiple process order groups in each window, and constructing the order code of the historical production trajectory data in each window; According to the multiple order codes, generating the order state sequence of the historical production trajectory data, traversing each order state sequence, extracting the duration feature of each order code, and selecting multiple order segments from the multiple order codes based on the duration feature; Classifying the multiple order state sequences to construct a qualified sample set and a risk sample set, and extracting multiple statistical features corresponding to each order segment in the qualified sample set and the risk sample set respectively; According to the statistical features, calculating the frequency difference parameter and the duration difference parameter of the order segment, dividing the multiple order segments into order types, and constructing a risk order set and a stable order set corresponding to the multiple order segments; Specifically, the statistical features of the order segment include segment frequency features and segment duration features corresponding to the qualified sample set and the risk sample set respectively; The frequency difference parameter is calculated based on the segment frequency features corresponding to the qualified sample set and the risk sample set respectively, and the duration difference parameter is calculated based on the segment duration features corresponding to the qualified sample set and the risk sample set respectively; Fusing the frequency difference parameter and the duration difference parameter to calculate the sample state weight of each order segment, dividing each order segment into order types according to multiple preset state ranges, determining the order type label of each order segment, wherein the order type label includes stable order, neutral order and risk order, and constructing a stable order set and a risk order set based on multiple stable orders and risk orders; Based on the risk order set and the stable order set, traversing the qualified sample set and the risk sample set to construct multiple candidate recovery paths of each order state sequence, fusing the multiple candidate recovery paths to construct multiple process optimization model libraries, and realizing the control optimization of the galvanized steel pipe production process equipment based on the process optimization model library.

2. A method for intelligent control of galvanizing process equipment for steel pipes as claimed in claim 1, wherein, Based on the risk order set and the stable order set, traversing the qualified sample set and the risk sample set to construct multiple candidate recovery paths of each order state sequence, fusing the multiple candidate recovery paths to construct multiple process optimization model libraries, and realizing the control optimization of the galvanized steel pipe production process equipment based on the process optimization model library. Traversing the multiple order state sequences in the qualified sample set and the risk sample set, identifying the risk order and the stable order in each order state sequence based on the risk order set, and extracting multiple candidate recovery paths in each order state sequence based on the recovery logic, wherein the starting node of the candidate recovery path is the risk order and the target node is the stable order. The steel tube galvanizing process flow classifies a plurality of candidate recovery paths according to process phase, constructs a corresponding candidate recovery sample set for each process phase, performs state evolution fusion on a plurality of candidate recovery paths in each candidate recovery sample set based on state semantic constraints, and generates a process optimization model library for each candidate recovery sample set, wherein the process optimization model library includes a plurality of process phase recovery models. The process of performing state evolution fusion on a plurality of candidate recovery paths in each candidate recovery sample set based on state semantic constraints to generate a process optimization model library for each candidate recovery sample set includes:

3. A method for intelligent control of galvanizing process equipment for steel pipes as claimed in claim 2, wherein, determining a starting order code and a target order code of each candidate recovery path in the candidate recovery sample set, calculating a starting state distance and a target state distance between any two candidate recovery paths based on the starting order code and the target order code, and determining a state semantic similarity between the two candidate recovery paths according to the starting state distance and the target state distance; extracting a flip sequence in each candidate recovery path other than the starting order code and the target order code, performing statistical analysis on each flip sequence, generating a frequency feature vector for each flip sequence, and calculating a state evolution similarity between any two candidate recovery paths based on the frequency feature vector; constructing a path similarity matrix for the candidate recovery sample set based on the state semantic similarity and the state evolution similarity, clustering a plurality of candidate recovery paths in the candidate recovery sample set based on the path similarity matrix, and obtaining a plurality of process phase recovery models for each candidate recovery sample set. For the process phase recovery model, further comprising:

4. A method for intelligent control of galvanizing process equipment for steel pipes as claimed in claim 3, wherein, clustering a plurality of candidate recovery paths in the candidate recovery sample set to produce a plurality of recovery path clusters, constructing a risk state and a stable state of the recovery path cluster according to the starting order code and the target order code of each candidate recovery path, constructing a model control vector of the recovery path cluster based on a plurality of candidate recovery paths, and fusing the risk state, the stable state and the model control vector to obtain a process phase recovery model for each recovery path cluster. The system is used to implement the intelligent control method for the steel tube galvanizing process equipment according to any one of claims 1-4, comprising:

5. An intelligent control system for a steel pipe galvanizing process equipment, characterized by, a production data acquisition module for acquiring a process running original data set of the galvanized steel tube production, including a plurality of historical production trajectory data, and constructing a plurality of process order groups based on the process running original data set; an order segment identification module for performing sliding window order matching on each group of historical production trajectory data according to the plurality of process order groups, generating an order state sequence for each group of historical production trajectory data, and identifying a plurality of order segments for each order state sequence; a sample set construction module for classifying a plurality of order state sequences to construct a qualified sample set and a risk sample set, and extracting a plurality of statistical features corresponding to each order segment in the qualified sample set and the risk sample set, respectively; an order type division module for calculating a frequency difference parameter and a duration difference parameter of the order segment according to the statistical features, performing order type division on a plurality of order segments, and constructing a risk order set and a stable order set corresponding to the plurality of order segments; and an order type division module for calculating a frequency difference parameter and a duration difference parameter of the order segment according to the statistical features, performing order type division on a plurality of order segments, and constructing a risk order set and a stable order set corresponding to the plurality of order segments. The galvanizing process optimization module is used for traversing the qualified sample set and the risk sample set based on the risk order set and the stable order set, constructing a plurality of candidate recovery paths of each order state sequence, fusing the plurality of candidate recovery paths to construct a plurality of process optimization module banks, and realizing the steel pipe galvanizing process equipment control optimization based on the process optimization module bank.

Images