A method and system for intelligent setting of PID parameters based on a PLC control system

Abstract

The application relates to the technical field of PLC control parameter setting, and discloses a PID parameter intelligent setting method and system based on a PLC control system. The method collects process variable signals, set value signals and actuator output signals in real time through a PLC communication protocol, extracts characteristic quantities after smoothing filtering and noise elimination processing, and forms a standardized control data set; the set is used to query matched setting strategy entries, calculate a complexity index and evaluate parameter configuration compliance, and generate a setting state judgment of a single control loop; a plurality of standardized control data sets are combined to detect shared entities between loops and calculate an interaction intensity index, and a multi-loop interaction relationship graph is generated; based on the graph, the setting state judgment is fused, the overall quality score is calculated, and a setting recommendation scheme is formed; after field verification and closed-loop control simulation testing, a final PID parameter setting instruction is output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of PLC control parameter setting, in particular to a PID parameter intelligent setting method and system based on a PLC control system. BACKGROUND

[0002] In the field of industrial automation control, PLC is widely used in the closed-loop control of various production processes as the core control device, and the PID (proportional-integral-derivative) control algorithm becomes the main means to realize accurate regulation in the PLC control system due to its simple structure and strong robustness. The setting quality of PID parameters directly affects the dynamic response performance and steady-state control accuracy of the control loop, and is an important link to ensure the stable operation of the production process and improve product quality.

[0003] Traditional PID parameter setting methods rely on manual experience, and technicians need to manually adjust parameters such as proportional coefficient, integral time and derivative time according to on-site debugging experience, which is not only time-consuming and laborious, but also greatly affected by the skill level of personnel, making it difficult to adapt to the control requirements of complex industrial scenes. With the complication of industrial processes and the increase of control loops, the single-loop independent setting mode gradually exposes obvious limitations. Since there is a coupling relationship between different control loops in terms of matter, energy or information, optimizing the parameters of a single loop may lead to the deterioration of the performance of other related loops, and even cause instability of the entire control system.

[0004] Although existing automatic setting techniques reduce manual intervention to some extent, they still have many shortcomings. Some methods only calculate parameters based on the dynamic response data of a single loop, ignoring the interactive influence between multiple loops, which leads to limited applicability of the setting results in actual complex systems; other methods consider the correlation between loops, but fail to effectively eliminate noise interference in the field signals during data processing, resulting in insufficient accuracy of the extracted feature quantities, which further affects the matching accuracy of the setting strategy; at the same time, most setting systems lack a systematic verification mechanism for the setting scheme, making it difficult to ensure the reliability of parameter settings in actual working conditions. These problems make it difficult for existing technologies to achieve efficient and accurate setting of PID parameters in the face of multivariable and strongly coupled industrial control scenes, restricting the further improvement of the level of industrial automation control. SUMMARY

[0005] The present application provides a PID parameter intelligent setting method and system based on a PLC control system to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides a PID parameter intelligent setting method based on a PLC control system, which comprises:

[0007] The process variable signal, the set value signal and the actuator output signal of the control system are collected in real time through the PLC communication protocol, the collected signals are subjected to smoothing filtering and noise elimination processing, the rising time, the overshoot and the steady-state error characteristic quantities in the signals are extracted, the characteristic quantities are corresponded to the pre-set characteristic parameter field and the time sequence correlation and the causal correlation between the characteristic quantities are set, and a standardized control data set is formed;

[0008] The standardized control data set is used to query the setting strategy item matching the control loop characteristics in the setting knowledge base, to acquire the setting task complexity index, to compare the control data with the performance threshold, the response speed limit and the stable domain range required by the strategy point by point in dependence of the complexity index, to evaluate the parameter configuration compliance, and to generate the setting state judgment of a single control loop;

[0009] The process variables, the associated actuating mechanisms and the detection instruments shared by different control loops are detected in combination with a plurality of the standardized control data sets, the interaction strength index between the loops is calculated in dependence of the frequency and the type of the shared entities, the network topology graph with the control entities as the vertices and the shared connections as the edges is drawn in accordance with the interaction strength index, and the multi-loop interaction relationship graph is generated;

[0010] Based on the multi-loop interaction relationship graph, the setting state judgment of the associated loops is fused, the overall setting quality score is calculated, the setting recommendation scheme containing the quality level is formed, the setting recommendation scheme is verified in the field, the closed-loop control simulation test is performed, and the final PID parameter setting instruction is output.

[0011] Preferably, the acquisition step of the standardized control data set comprises: parallel analysis of the real-time collected multi-channel process signals, separation of the dynamic response component and the steady-state working point component in the signals; the multi-scale waveform decomposition technology is used to extract the characteristic profile of the dynamic response component, and the working point drift tracking algorithm is used to record the change trajectory of the steady-state working point component; the characteristic profile and the change trajectory are spatio-temporally aligned to generate a signal characteristic sequence with a time stamp; the signal characteristic sequence is subjected to pattern matching with a pre-defined characteristic template library, the matched characteristics are subjected to dimension reduction and normalized coding to form a standardized characteristic vector; a characteristic correlation matrix is established based on the time sequence dependence relationship between the standardized characteristic vectors, and the construction of the standardized control data set is completed.

[0012] Preferably, the obtaining of the setting task complexity index comprises: parsing a dynamic characteristic mode and a steady-state working interval of the control loop from the standardized control data set; performing a deep traversal search in a setting rule knowledge graph according to the dynamic characteristic mode to obtain an associated setting rule cluster; analyzing a topological structure feature of the setting rule cluster, including a rule node degree, a rule chain length, and a rule cross-reference density; combining a boundary condition of the steady-state working interval to calculate an activation probability distribution and a conflict detection coefficient of the rule cluster; and performing weighted fusion based on the topological structure feature and the activation probability distribution to generate the setting task complexity index.

[0013] Preferably, the obtaining of the setting state judgment of the single control loop comprises: establishing a similarity measurement space of a control loop dynamic response and an ideal response template; performing multi-dimensional comparison of response trajectories in the measurement space, including phase trajectory coincidence degree, amplitude envelope matching degree, and convergence speed deviation degree; dynamically updating a comparison reference by using an adaptive threshold adjustment mechanism, and performing weighted correction on a comparison result in combination with the setting task complexity index; generating a compliance quantitative score based on the corrected multi-dimensional comparison result, and generating the setting state judgment according to an interval mapping of the score.

[0014] Preferably, the obtaining of the inter-loop interaction intensity index comprises: constructing a control loop correlation hypergraph model, wherein a hyperedge represents a coupling relationship between multiple loops; analyzing node centrality and hyperedge weight distribution of each loop in the hypergraph; calculating energy transfer efficiency and interference propagation path between loops by using a coupling strength detection algorithm based on random walk; generating a multi-dimensional interaction feature vector in combination with control frequency band overlap degree and control timing synchronization; performing principal component analysis and weight allocation on the multi-dimensional interaction feature vector to synthesize the inter-loop interaction intensity index.

[0015] Preferably, the obtaining of the multi-loop interaction relationship graph comprises: establishing a control loop entity knowledge graph, wherein a node contains a loop static attribute and a dynamic running state; learning an implicit relationship between nodes by using a graph neural network technology to mine a potential interaction mode; dynamically calibrating an edge weight based on the interaction intensity index to construct a weighted interaction network; introducing a timing diagram attention mechanism to capture an evolution law of the interaction relationship to generate a multi-loop interaction relationship graph with a time dimension.

[0016] Preferably, the obtaining of the overall setting quality score comprises: performing multi-objective optimization path search on the multi-loop interaction relationship graph to find an optimal path of setting effect propagation; analyzing a setting state transition probability and an effect decay coefficient on the path; adopting a setting effect evaluation method based on Monte Carlo simulation to statistically obtain a robustness index of a setting scheme under uncertain working conditions; combining a topological feature of the interaction relationship graph and the robustness index to construct a setting quality comprehensive evaluation function to generate the overall setting quality score.

[0017] Preferably, the generating step of the setting recommendation scheme comprises: establishing a multimodal distribution model of the setting parameter solution space, locating the high-quality parameter region by using a heuristic region search algorithm; prioritizing the parameter region based on the overall setting quality score; generating parameter boundary cases using the generative adversarial network technology to test the stability boundary of the setting scheme; combining the multi-objective optimization algorithm to find the Pareto optimal solution set in the high-quality parameter region to form the setting recommendation scheme.

[0018] Preferably, the output step of the final PID parameter setting instruction comprises: designing a multi-scenario verification test case library covering typical working conditions and boundary conditions; using digital twin technology to establish a high-fidelity simulation model of the controlled object; performing batch verification testing of the setting recommendation scheme in the simulation environment, collecting dynamic response data; based on the verification results, using a conservative adjustment strategy to calibrate the safety margin of the parameters to generate the final PID parameter setting instruction.

[0019] Preferably, the application further comprises a PLC control system-based PID parameter intelligent setting system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the above-mentioned PLC control system-based PID parameter intelligent setting method.

[0020] Compared with the prior art, the application has the following advantages:

[0021] The PLC control system-based PID parameter intelligent setting method provided by the application collects and processes key signals in real time through the PLC communication protocol to form a standardized control data set. During this process, the collected signals are subjected to smoothing filtering and noise elimination, which can reduce the influence of field interference on data quality. The extracted characteristic quantities such as rise time and overshoot are set through time sequence correlation and causal correlation to make the data more consistent with the internal law of the control process, thereby providing a standardized and reliable data basis for subsequent setting work.

[0022] The standardized control data set is used to query and match the setting strategy items, and the complexity index is calculated to evaluate the parameter configuration compliance. This process determines the task complexity by combining the number of items and the analysis depth, making the parameter evaluation more consistent with the complexity of actual control requirements. The performance threshold, response speed limit and other contents are compared point by point to comprehensively investigate the adaptability of parameter configuration, and the generated single control loop setting state judgment is more targeted and accurate.

[0023] The multi-loop interaction relationship graph is generated in combination with multiple standardized control data sets, the interaction strength index is calculated and the network topology graph is drawn by detecting common process variables, associated actuators and other entities, the internal relationship between different control loops can be clearly presented, the limitation of single-loop setting is broken, a global perspective is provided for parameter optimization of the overall system, and the system imbalance problem caused by isolated setting is avoided.

[0024] Based on the multi-loop interaction relationship graph fusion setting state judgment, the overall quality score is measured and the recommended scheme is formed, and then the final parameter instruction is output through field verification and closed-loop simulation test, the process generates an overall scheme by fusing related loop states, ensures that the parameter setting meets the overall performance requirements of the system, and further verifies the applicability of the scheme in the actual working condition through the field verification and simulation test link, so that the finally output PID parameter instruction can better adapt to the actual needs of the complex industrial control environment, and help to improve the dynamic response performance and steady-state control precision of the control system, and adapt to the control requirements of the multivariable and strong coupling industrial scene. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The working principle diagram of the PID parameter intelligent setting method based on the PLC control system is described.

[0026] Figure 2 The acquisition step flowchart of the standardized control data set is described.

[0027] Figure 3 The acquisition step flowchart of the setting state judgment of a single control loop is described. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0029] Please refer to Figure 1The application provides a PID parameter intelligent setting method and system based on a PLC control system. The method includes a PLC communication protocol for real-time acquisition of process variable signals, set value signals and actuator output signals of the control system. The acquired signals are subjected to smoothing filtering and noise elimination processing to improve data quality. In the signal processing stage, the rise time, overshoot and steady-state error characteristic quantities are extracted. These characteristic quantities are mapped to the preset characteristic parameter field, and the time sequence correlation and causal correlation between the characteristic quantities are established, thereby forming a standardized control data set. The standardized control data set serves as a basic data unit, supporting the input of subsequent analysis modules. The setting knowledge base stores a large number of setting strategy entries matched with the control loop characteristics. The standardized control data set is used for query operation, and the setting task complexity index is calculated according to the number of entries and the analysis depth. The setting task complexity index drives the evaluation process, and the performance threshold, response speed limit and stable domain range of the control data and strategy requirements are compared point by point to evaluate the parameter configuration compliance and generate the setting state judgment of a single control loop. Multiple standardized control data sets are integrated and analyzed to detect process variables, associated actuators and detection instruments shared between different control loops. According to the frequency and type of shared entities, the inter-loop interaction strength index is calculated. The inter-loop interaction strength index is used to construct a network topology graph, in which the control entities are vertices and the shared connections are edges, and a multi-loop interaction relationship map is generated. The multi-loop interaction relationship map integrates the setting state judgment of the associated loops to measure the overall setting quality score and form a setting recommendation scheme containing the quality level. The setting recommendation scheme is subjected to a field verification link and effect confirmation through closed-loop control simulation test, and finally outputs the PID parameter setting instruction.

[0030] Example 1: see Figure 2The construction of the standardized control data set starts with the parallel analysis of the real-time collected multi-channel process signals from various sensors and actuators connected to the PLC control system. The parallel analysis process uses a multi-threaded computing architecture to simultaneously read and preliminarily classify multiple signal channels, separating the dynamic response components and the steady-state working point components in the signals. The dynamic response components reflect the transient behavior of the control system under set value changes or external disturbances, while the steady-state working point components reflect the long-term running level of the system in the equilibrium state. The multi-scale waveform decomposition technique based on wavelet transform theory is used to extract the feature profiles of the dynamic response components, which captures the local features of the signals at coarse and fine granularities. The feature profiles are composed of a series of parameters describing the waveform shape, extreme point position, and oscillation mode. The working point drift tracking algorithm is used to record the change trajectory of the steady-state working point components, which calculates the statistics (such as mean, variance) of the signal using a sliding time window and fits the trend of its change over time to identify slow drift phenomena.

[0031] The feature profiles and change trajectories are spatiotemporally aligned to generate a time-stamped signal feature sequence. The spatiotemporal alignment operation relies on a high-precision time synchronization mechanism to ensure that the features extracted from different components accurately correspond on the time axis, and each feature point is labeled with a unique timestamp. The signal feature sequence is matched with a pre-defined feature template library that stores the ideal feature patterns of various typical control loops under standard working conditions. The pattern matching process calculates the similarity measure (such as Euclidean distance, dynamic time warping distance) between the signal feature sequence and each template, and performs dimension reduction and normalized encoding on the matched features to form a standardized feature vector. Dimension reduction reduces the number of features while preserving the main information through principal component analysis or autoencoder networks, and normalized encoding scales the feature values of different dimensions and ranges to a unified numerical interval (such as the [0, 1] interval). The standardized feature vector facilitates the uniform processing of subsequent algorithms. Based on the temporal dependence relationship between standardized feature vectors, a feature correlation matrix is established to complete the construction of the standardized control data set. The temporal dependence relationship reveals how features at different time points interact with each other, and the feature correlation matrix is a square matrix that quantifies the correlation strength between feature pairs. The correlation strength can be calculated by mutual information, correlation coefficient, etc.

[0032] The feature correlation matrix simultaneously contains quantitative representations of the time sequence correlation and the causal correlation, wherein the time sequence correlation describes the sequential relationship of the feature values on the time axis (such as the time sequence of the overshoot at t1 and the steady-state error at t2), and the causal correlation represents the logical dependence and action relationship between the feature values (that is, the change of one feature value is the reason or result of the change of another feature value, such as the excessive overshoot caused by the excessively large proportional coefficient, and the cumulative steady-state error causes the increase of the actuator output increment). The specific implementation steps of the causal correlation are as follows: first, based on the control theory, a feature causal relationship library is preset (such as the inherent causal logic of “proportional coefficient Kp-overshoot σ”, “integral time Ti-steady-state error ess”, and “derivative time Td-response speed tr” in PID control); second, the causal inference is performed on the standardized feature vector, and the Granger causality test algorithm is used to verify the causal effectiveness between the feature pairs (such as verifying whether “actuator output change” is the Granger reason of “process variable fluctuation”); and finally, the verified causal relationship is quantified as a correlation strength value (the value range is 0-1, and 1 indicates strong causality and 0 indicates no causality), which is weighted and fused with the time sequence correlation strength value (the causal correlation weight is 0.6, and the time sequence correlation weight is 0.4, which can be adaptively adjusted according to the control scene) to serve as the final element value of the feature correlation matrix. Through the above manner, the feature correlation matrix simultaneously encodes the time sequence and the logical causality of the feature values, so that the standardized control data set can accurately reflect the internal law of the control process, and provide a data basis with both time sequence continuity and causal logic for subsequent tuning strategy matching and multi-loop interaction analysis. The standardized control data set serves as a structured data whole, which provides an input for subsequent tuning task complexity analysis. Among them, the causal correlation information in the feature correlation matrix is the key, which can help to quickly locate the strategy directly related to the current control problem (such as directly matching the tuning rule of “reducing the proportional coefficient” through the causal relationship of “overshoot exceeding the standard → Kp being too large”) in the subsequent tuning strategy item matching; and the causal correlation can identify the causal transmission path across loops (such as the “process variable fluctuation” of loop A becoming the reason for the “steady-state error increase” of loop B through the shared actuator) in the multi-loop interaction analysis, thereby providing a logical basis for accurately calculating the interaction strength index.

[0033] The obtaining step of the tuning task complexity index parses the dynamic characteristic mode and the steady-state working interval of the control loop from the standardized control data set. The dynamic characteristic mode describes the response style of the loop to external excitation, such as over-damping, under-damping, oscillation, etc., and the steady-state working interval indicates the value range of the process variable when the loop is normally running. According to the dynamic characteristic mode, a depth-first search is performed in the tuning rule knowledge graph to obtain the associated tuning rule cluster. The tuning rule knowledge graph is a knowledge base organized in a graph structure, where the nodes represent specific tuning rules or conditions, and the edges represent the logical relationship between rules. The depth-first search algorithm explores along the edge structure of the graph to collect all rule nodes related to the current dynamic characteristic mode, forming a tuning rule cluster. The topological structure characteristics of the tuning rule cluster are analyzed, including the rule node degree, rule chain length, and rule cross-reference density. The rule node degree measures how many other nodes a certain rule node is connected to, and the high-degree node is usually the core rule. The rule chain length refers to the number of rule nodes passed from the initial condition to the final decision. The rule cross-reference density reflects the frequency of mutual reference between different rules in the rule cluster, and high density indicates complex rule logic. Combined with the boundary conditions of the steady-state working interval, the activation probability distribution and the conflict detection coefficient of the rule cluster are calculated. The boundary conditions of the steady-state working interval limit the working condition range of the rules, and the activation probability distribution estimates the possibility of each rule in the tuning rule cluster being triggered under a given steady-state working interval. The conflict detection coefficient is used to evaluate whether the output suggestions of different rules activated at the same time are contradictory or irreconcilable.

[0034] The weighted fusion is based on the topological structure characteristics and activation probability distribution to generate the setting task complexity index. The weights of different topological characteristics and probability values are assigned in the weighted fusion process, and the weight values can be obtained through expert experience or historical data learning. The fusion result is a scalar value, and the size of the value directly represents the estimated complexity of the parameter setting task of the control loop. Higher setting task complexity index means that the loop dynamic behavior is complex, the applicable setting rules are numerous and may conflict, and more careful and in-depth analysis is required. The setting task complexity index as a key metadata will guide the subsequent setting state judgment process, such as adjusting the strictness of the performance threshold or selecting a more detailed comparison algorithm. The specific implementation of the multi-scale waveform decomposition technique usually selects a wavelet basis function, and the decomposition layer is adaptively determined according to the main frequency component of the signal. The feature quantity that can represent the key behavior of the signal is extracted from the wavelet coefficients of each layer after decomposition, such as the energy of a specific frequency band, the zero-crossing rate, and the envelope shape parameter. The working point drift tracking algorithm not only tracks the mean shift, but also monitors the variance change to discover the change of system noise characteristics or external disturbance pattern. The algorithm needs to set appropriate window size and update frequency to balance the sensitivity and stability of tracking. The construction of the feature template library is an accumulation and iteration process, which needs to cover the common control loop types and working conditions in the target industrial field. The template matching algorithm needs to have certain noise resistance and tolerance to slight waveform deformation.

[0035] Dimension reduction and normalization encoding are key steps in data preprocessing. Principal component analysis finds the direction with the maximum variance of data through linear transformation for projection, and autoencoder learns the nonlinear low-dimensional representation of data through neural network. Normalization encoding commonly uses min-max scaling or Z-score standardization method. The choice of method depends on the distribution characteristics of data and the requirements of subsequent algorithms. The construction of feature correlation matrix can consider the correlation under different time delays, so as to capture the causality information. The sparsity processing of the matrix helps to improve the calculation efficiency and focus on key correlations. The quality of the setting rule knowledge graph directly affects the effect of deep traversal search. The construction of the graph depends on the control theory knowledge and the experience of field experts, and can also be supplemented and optimized by graph learning technology using historical setting data. Deep traversal search needs to set termination conditions to prevent infinite loops in large graphs, such as limiting search depth or setting relevance threshold. The calculation of topological structure characteristics belongs to the category of graph theory. The rule node degree can be divided into in-degree and out-degree. The rule chain needs to identify all possible paths, and the cross-reference density reflects the coupling degree within the rule cluster.

[0036] Example 2: see Figure 3The acquisition of the setting state judgment of the single control loop starts from establishing a similarity measurement space of the dynamic response of the control loop and an ideal response template. The dynamic response of the control loop is time series data collected from an actual system, and the ideal response template is a desired performance curve defined in advance based on control theory. The similarity measurement space is a mathematical abstraction for quantifying the difference between the actual response and the ideal template. Multi-dimensional comparison of the response trajectory is performed in the similarity measurement space. The response trajectory is a path in a multi-dimensional space, and the multi-dimensional comparison includes phase trajectory coincidence degree, amplitude envelope matching degree and convergence speed deviation degree. The phase trajectory coincidence degree evaluates the alignment degree of the actual response and the ideal template in phase, the amplitude envelope matching degree measures the consistency of the amplitude change profile, and the convergence speed deviation quantifies the difference in the time required for the system to reach a stable state.

[0037] The multi-dimensional comparison needs to combine the causal relationship information in the standardized control data set: the consistency of the actual characteristic quantity causal relationship (such as “Kp=2.5→ overshoot=15%”) and the standard causal relationship in the ideal response template (such as “Kp=2.0→ overshoot=5%”) is verified, and the causal matching degree (i.e. the degree of coincidence between the actual causal chain and the standard causal chain) is calculated. The causal matching degree is an independent comparison dimension (weight 0.3), and is weighted and fused with the phase trajectory coincidence degree (weight 0.3), the amplitude envelope matching degree (weight 0.2) and the convergence speed deviation degree (weight 0.2) to generate a more comprehensive compliance quantitative score, ensuring that the setting state judgment not only focuses on the response result, but also focuses on the rationality of the causal logic behind the result.

[0038] An adaptive threshold adjustment mechanism is used to dynamically update the comparison benchmark. The adaptive threshold adjustment mechanism automatically adjusts the severity of the judgment standard according to historical comparison data and system running state, and combines the complexity of the setting task index to weightedly correct the comparison result. The complexity of the setting task index reflects the estimated difficulty of the loop setting task.

[0039] The compliance quantitative score is generated based on the corrected multi-dimensional comparison result, and the compliance quantitative score is a comprehensive numerical value. The setting state judgment is generated according to the score interval mapping, and the setting state judgment is a clear and conclusive output, such as 'parameter configuration compliance' or 'parameter needs to be reset'. The steps of obtaining the inter-loop interaction intensity index include constructing a control loop correlation hypergraph model, and the control loop correlation hypergraph model is a general graph structure, in which a hyperedge can connect multiple nodes, and is used to represent the complex coupling relationship between multiple control loops. The node centrality and hyperedge weight distribution of the hypergraph are analyzed, the node centrality measures the importance and influence of a specific loop in the entire hypergraph network, and the hyperedge weight distribution reflects the relative strength of different coupling relationships. The coupling strength detection algorithm based on random walk is used to calculate the energy transfer efficiency and interference propagation path in the loop, and the coupling strength detection algorithm based on random walk simulates the propagation process of a virtual signal in the hypergraph network, the energy transfer efficiency calculates the attenuation degree of energy transfer from one loop to another, and the interference propagation path identifies the main way of interference diffusion in the loop network.

[0040] The multi-dimensional interaction feature vector is generated in combination with the loop working frequency band overlap degree and the control timing synchronization, the loop working frequency band overlap degree analyzes the overlapping part of the operation frequency range of different loops, and the control timing synchronization evaluates the coordination relationship of the control action of the loop in time. The multi-dimensional interaction feature vector is subjected to principal component analysis and weight allocation to synthesize the inter-loop interaction intensity index, the principal component analysis is used for dimension reduction and extraction of the most representative components in the interaction feature, and the weight allocation determines the contribution proportion of each interaction feature to the final index value. The data of the dynamic response of the control loop is usually derived from the process variable change curve after the set value change or load disturbance recorded by the PLC system, and the ideal response template is set according to the control target, such as the typical second-order system underdamped response or critical damping response curve. Different distance functions can be selected for the construction of the similarity measurement space, such as the dynamic time warping algorithm for aligning time series with different lengths, and then calculating the Euclidean distance or cosine similarity. The calculation of the phase trajectory coincidence degree needs to convert the time series to the phase space first, and then compare the shape and direction of the phase trajectory. The amplitude envelope matching degree involves extracting the upper envelope line and the lower envelope line of the signal, and comparing them with the envelope of the template. The convergence speed deviation degree focuses on the time point when the response curve enters and remains in the steady-state error band.

[0041] The core of the adaptive threshold adjustment mechanism is a feedback loop, where the initial threshold is set based on theoretical or empirical values, and then dynamically adjusted based on the distribution of the continuous comparison results (such as mean and variance), so that the judgment criteria can adapt to the slow changes of system characteristics. The complexity of the tuning task index is used as a correction factor, which may relax the comparison requirements of some dimensions for high complexity loops, because their ideal performance is more difficult to achieve. The generation of compliance quantitative scores usually uses a linear weighting model, which multiplies the comparison scores of multiple dimensions by their respective weights and sums them up. The weights can be learned based on expert experience or historical data. The score interval mapping is to discretize the continuous score value into a limited number of tuning state levels, and the mapping relationship needs to be defined in advance. The construction of the control loop association hypergraph model needs to identify all potential coupling paths between loops, which may be due to shared physical devices, common energy supply, or process flow associations. The calculation of node centrality can use the feature vector centrality algorithm of hypergraph, and the hyperedge weight can be assigned according to the strength of the coupling physical mechanism or historical interference data. The coupling strength detection algorithm based on random walk performs random walk on the hypergraph and calculates the stationary distribution probability of reaching other nodes from each node. This probability matrix contains rich coupling information. Energy transfer efficiency can be analyzed from the frequency response perspective, and interference propagation path can be identified by analyzing the highest frequency edge in the random walk path.

[0042] Loop operating frequency band overlap analysis needs to obtain the main operating frequency range of each loop and calculate the intersection size of these frequency ranges. Control timing synchronization analysis needs to check the phase relationship between the execution cycles of different loop controllers and the sequence of control actions. The generated multi-dimensional interaction feature vector contains interaction information from different angles. Principal component analysis converts these possibly correlated features into a set of linearly uncorrelated principal components, retaining most of the original variance. The weight allocation process can use objective weighting methods such as entropy weighting method, or combine with subjective weighting methods such as analytic hierarchy process. The final inter-loop interaction strength index is a scalar value, whose size directly reflects the strength of the interaction between any two or more loops. The inter-loop interaction strength index provides a key quantitative input for subsequent construction of multi-loop interaction relationship graph.

[0043] The obtaining steps of the multi-loop interaction relationship graph in embodiment 3 include establishing a control loop entity knowledge graph, which is a method for representing knowledge in a graph structure, wherein a node represents a specific control loop entity, and the node contains loop static attributes and dynamic running states. The loop static attributes include the type of the loop, the process unit to which the loop belongs, the connected actuator and sensor model, and other fixed information. The dynamic running state includes real-time process variable values, controller outputs, set values, and calculated performance indicators such as instantaneous error and integral error. The construction of the control loop entity knowledge graph needs to extract, clean and associate information from scattered configuration data, historical databases and real-time data streams to form a unified, machine-readable knowledge network. The graph neural network technology is used to learn the implicit relationship between nodes and mine potential interaction patterns. The graph neural network technology is a deep learning model specially used for processing graph structure data. It aggregates the information of neighbor nodes through a message passing mechanism to update the representation of the node itself. This process can capture the complex, non-direct and explicitly defined associations between nodes. Learning the implicit relationship between nodes means that the graph neural network model can automatically discover potential causal relationships or influence paths that are not predefined in the control loop entity knowledge graph, such as identifying two loops that are far apart in geographical location but are coupled by energy through process logistics. Mining potential interaction patterns is to discover typical interaction forms existing in the loop group by analyzing the similarity of node representation vectors or detecting graph neural network activation patterns, such as the pattern of multiple loop synchronous oscillation or the pattern of a certain loop as the main disturbance source affecting a piece of loop. The cross-loop causal association strength indicator needs to be included in the multi-dimensional interaction feature vector. Based on the causal association matrix in the standardized control data set, the causal transmission relationship between different loops corresponding to the shared entities (such as the change of the actuator output of loop 1 → the action of the shared actuator → the change of the process variable of loop 2) is extracted. The cross-loop causal influence degree is calculated using the causal transmission efficiency formula (causal transmission efficiency = target loop feature change amount / source loop feature change amount), and this degree is taken as the core component of the multi-dimensional interaction feature vector. The working frequency band overlap degree and control timing synchronization are fused to ensure that the interaction strength index can accurately represent the strength of causal coupling between loops, rather than relying only on the statistical level of association.

[0044] The interaction intensity index dynamically calibrates the edge weight to construct the weighted interaction network. The interaction intensity index is a quantitative indicator calculated from the previous step, representing the strength of interaction between any two or more loops. In the control loop entity knowledge graph, the interaction relationship between loops may initially exist or not exist in the form of a simple, unweighted edge. The interaction intensity index is used as a weight value to give these edges a specific strength meaning, achieving dynamic calibration. Dynamic calibration is reflected in the fact that the interaction intensity index may be updated as the system operating conditions change, resulting in the redistribution of the interaction network edge weight. The result of constructing the weighted interaction network is a graph with each edge attached to a weight value, which directly reflects the strength of the interaction between the corresponding loops. The introduction of the time series graph attention mechanism captures the evolution law of the interaction relationship and generates a multi-loop interaction relationship graph with a time dimension. The time series graph attention mechanism is an extension of the graph attention mechanism in time series. It not only considers the topological structure of the current time graph, but also focuses on the sequence of node and edge attributes over time. Capturing the evolution law of the interaction relationship means that the time series graph attention mechanism can identify how the coupling strength between loops changes over time, for example, some interaction relationships become significant only in a specific production phase. The time series graph attention mechanism will assign different attention weights to the graph states at different time steps, focusing on key evolution moments. Generating a multi-loop interaction relationship graph with a time dimension means that the final graph is not a static snapshot, but a dynamic model that can show the historical changes, current state, and possibly predict future trends of the interaction relationship. This dynamic model can be represented as a sequence of weighted graphs corresponding to time stamps, or a tensor containing a time dimension.

[0045] The obtaining step of the overall tuning quality score includes performing multi-objective optimization path search on the multi-loop interaction relationship graph to find an optimal path of tuning effect propagation. The multi-loop interaction relationship graph depicts the connection and interaction strength between loops, and the multi-objective optimization path search aims to find one or more paths in the graph, so that implementing parameter tuning along these paths can bring the best overall benefit, while balancing multiple possibly conflicting objectives, such as maximizing the overall control performance improvement, minimizing the number of loops involved in tuning actions, and avoiding introducing unstable factors in strongly interactive loops. Finding the optimal path of tuning effect propagation requires defining what is "effect propagation", i.e. the influence model of parameter change of one loop on the performance of its neighbor loops, and the optimal path is the propagation chain with the maximum positive effect and the minimum negative effect. The tuning state transition probability and effect decay coefficient on the path are analyzed. The tuning state transition probability on the path describes the possibility that after a loop is tuned, its adjacent loops change from one tuning state to another due to coupling, and this probability value depends on the interaction strength and the dynamic characteristics of the loop itself. The effect decay coefficient quantifies the degree of weakening of the tuning effect as the propagation distance in the interaction network increases, similar to the attenuation of signal propagation in a medium.

[0046] A tuning effect evaluation method based on Monte Carlo simulation is used to statistically obtain the robustness index of the tuning scheme under uncertain working conditions. The tuning effect evaluation method based on Monte Carlo simulation simulates the uncertainty of the system, such as set value variation, external disturbance, process parameter drift, etc. In each simulation, the tuning scheme to be evaluated is applied, and the performance of the control system under the environment described by the multi-loop interaction relationship graph is observed. Through a large number of independent simulations, the performance distribution of the tuning scheme can be statistically obtained, and the robustness index, such as the mean, variance, and worst-case performance value of the performance index, can be extracted. The robustness index reflects the ability of the tuning scheme to maintain good performance in the face of various uncertain factors. Combined with the topological features of the interaction relationship graph and the robustness index, a tuning quality comprehensive evaluation function is constructed to generate the overall tuning quality score. The topological features of the interaction relationship graph include network connectivity, centrality index, community structure, etc., which affect the breadth and efficiency of tuning effect propagation. The tuning quality comprehensive evaluation function is a mathematical expression that takes the topological features and robustness index as input variables, and outputs a single scalar value, i.e. the overall tuning quality score, through a weighting or combination rule. The design of this function needs to reflect the influence of the topological structure on the tuning robustness, for example, a highly connected network may be more sensitive to tuning actions, requiring a higher robustness of the tuning scheme. The overall tuning quality score is finally used to compare the pros and cons of different tuning recommendation schemes.

[0047] In constructing the control loop entity knowledge graph, static attributes are usually from the configuration files of the control system, and dynamic running states need to be interfaced with real-time databases. The learning process of graph neural network technology is usually unsupervised or semi-supervised, using the structure of the graph itself as a supervision signal. The time series attention mechanism needs to process time series data, which may involve recurrent neural network structures or variants of the Transformer architecture. Multi-objective optimization path search can use multi-objective evolutionary algorithms such as NSGA-II, or special-purpose heuristic algorithms for graph path problems. The estimation of the setting state transition probability may require historical data or physical model-based derivation. The determination of the effect decay coefficient can be done by analyzing the transfer function of the interactive loop or by empirical assignment. The tuning effect evaluation based on Monte Carlo simulation requires the construction of a simulation model of the control system, and the sampling range should cover the expected operating condition changes. The construction of the comprehensive evaluation function of the tuning quality is key, which can be a weighted linear combination or a more complex nonlinear function, and its parameters may need to be determined through expert knowledge or offline learning.

[0048] Overall tuning quality score One example of a comprehensive evaluation function is formalized as follows:

[0049]

[0050] Where: represents the final overall tuning quality score, which is a scalar value. represents the system stability score calculated based on Monte Carlo simulation results, with a value between 0 and 1, where 1 indicates stability in all simulation trials. represents the system average response speed score, which is normalized based on the average of all loop regulation times, with a higher score indicating faster overall response. represents the interactive network impact score, which is calculated based on the topological properties of the multi-loop interaction relationship graph (such as network diameter and average path length), reflecting the efficiency of tuning effect propagation. is the weight coefficient, corresponding to the weights of stability, response speed, and interaction impact, satisfying These weights can be adjusted according to the importance of specific control objectives. This simplified formula integrates system stability, response performance, and network interaction efficiency into a comprehensive score through weighted summation. The setting of the weight coefficient reflects the preference for different performance dimensions. Ultimately, the overall tuning quality score is higher, indicating a better tuning scheme.

[0051] The generating step of the tuning recommendation scheme includes establishing a multi-peak distribution model of the tuning parameter solution space, the tuning parameter solution space covers the value range of all possible PID parameter combinations considered, the multi-peak distribution model uses a probability distribution function to describe the case that there may be multiple excellent performance areas in the tuning parameter solution space, the multi-peak distribution model is constructed by analyzing historical tuning data or prior knowledge based on a physical model, and regions with high probability density are identified, which correspond to potential high-quality parameter settings. A heuristic region search algorithm is used to locate the high-quality parameter region, which can efficiently explore the vast tuning parameter solution space and avoid the computational burden of exhaustive search. The algorithm guides the search direction by evaluating the performance indicators of the sampling points, gradually shrinks to the region with better performance, and finally locates a batch of parameter regions with high tuning quality scores. The overall tuning quality score prioritizes the parameter regions, and the overall tuning quality score is a quantitative value of the comprehensive performance of the parameter region evaluated from the system level. The priority sorting process sorts the multiple high-quality parameter regions according to the overall tuning quality score, and the parameter region with a higher overall tuning quality score is given priority in subsequent processing. The generative adversarial network technology generates parameter boundary cases, the generative adversarial network technology includes a generative network and a discriminative network, the generative network learns the distribution characteristics of the high-quality parameter region and generates new parameter combination cases near the boundary of the network learned high-quality parameter region. These parameter boundary cases are used to test the stability boundary of the tuning scheme, that is, to investigate whether the system performance will deteriorate significantly when the parameters are slightly changed near the boundary of the high-quality region. The multi-objective optimization algorithm searches for the Pareto optimal solution set in the high-quality parameter region, the multi-objective optimization algorithm handles multiple possible conflicting optimization objectives, such as the trade-off between fast response and reduced overshoot, and searches for parameter solutions in a single high-quality parameter region that cannot improve the performance of other objectives without degrading the performance of at least one objective, i.e. the Pareto optimal solution set, forming the tuning recommendation scheme.

[0052] Establishing a multi-modal distribution model of the tuning parameter solution space is the basis of generating tuning recommendation schemes. The multi-modal distribution model structures the continuous parameter space into multiple probability peaks, each of which represents a potential optimal parameter setting cluster. Heuristic regional search algorithms such as particle swarm optimization or differential evolution algorithm conduct concentrated search around these probability peaks and quickly locate high-performance regions. The located high-quality parameter regions need to be accurately characterized, usually recording their center coordinates and boundary ranges. The overall tuning quality score gives each parameter region a global evaluation score, making the ranking have a system perspective. Adversarial generative network technology generates parameter combinations that appear reasonable but are actually at the performance critical point through the generator, and the discriminator tries to distinguish the generated boundary cases from the real high-quality parameters. Through this adversarial training, the generator can generate effective boundary cases for robustness testing. Multi-objective optimization algorithms such as NSGA-II conduct fine search within the determined high-quality parameter regions, and finally output a set of Pareto optimal solutions to provide operators with a series of optional tuning schemes under different performance trade-offs. The tuning parameter solution space is usually defined by the value range of the proportional gain Kp, integral time Ti, and derivative time Td, and the multi-modal distribution model can be fitted using a Gaussian mixture model. The performance of the heuristic regional search algorithm depends on its initialization and parameter settings. The training of the adversarial generative network technology requires sufficient data support. The effectiveness of the multi-objective optimization algorithm depends on the definition of the objective function and the efficiency of the search algorithm. The tuning recommendation scheme is finally presented in the form of a set of Pareto optimal parameter solutions and their corresponding performance prediction indicators, see Table 1.

[0053] Table 1: High-quality parameter region attribute table

[0054]

[0055] Table 1 shows a brief description of the four located high-quality parameter regions. Region number is used to uniquely identify each region. The center coordinates give the core location of the high-quality parameter region in the tuning parameter solution space. The region boundary radius roughly represents the extension of the high-quality parameter region in the parameter space. The overall tuning quality score is the comprehensive score of the high-quality parameter region evaluated at the system level. The performance characteristic tendency is a qualitative description of the general control performance that the parameter set within the high-quality parameter region can bring. The establishment of the multi-peak distribution model of the tuning parameter solution space relies on the in-depth understanding of the system behavior or the accumulation of a large amount of historical data, and the model can reveal the complex nonlinear relationship between parameters and performance. The heuristic region search algorithm conducts directional exploration in such a vast space, and its convergence speed and global search ability are crucial. The overall tuning quality score serves as a unified ruler, allowing schemes from different regions with different performance tendencies to be compared and sorted. The boundary cases generated by the generative adversarial network technology are extreme tests of the robustness of the tuning scheme, which helps to find potential performance vulnerabilities.

[0056] The output step of the final PID parameter setting instruction in embodiment 5 includes designing a multi-scenario verification test case library, which is a structured test case set covering typical operating conditions and boundary conditions. Typical operating conditions refer to operating states that the control system often encounters during normal production processes, such as step changes in set values and steady-state operation under different loads. Boundary conditions represent the limit states or extreme situations of safe system operation, such as maximum allowable pressure, temperature limit, and flow upper limit. The design of the test case library needs to systematically consider various input signal combinations, disturbance types, and system initial states. Each test case clearly defines the initial conditions, input sequence, and expected performance evaluation criteria. A high-fidelity simulation model of the controlled object is established using digital twin technology. Digital twin technology integrates physical models, real-time data, and historical data to build a virtual model highly consistent with the physical object. The high-fidelity simulation model requires accurate reflection of the dynamic characteristics, nonlinear factors, and time-varying behavior of the controlled object, such as the thermal inertia of heat exchangers, pressure transmission delay of fluid pipelines, and friction models of mechanical systems. The establishment of the model relies on in-depth mechanism analysis, system identification experiments, and verification of a large amount of operating data to ensure that the simulation results have high consistency with the behavior of the physical system under a wide range of operating conditions.

[0057] A batch verification test of the tuning recommendation scheme is performed in a simulation environment, which provides a running platform for the digital twin model. The tuning recommendation scheme includes one or more sets of PID parameter configurations optimized and screened through the foregoing steps. The batch verification test means that each set of candidate PID parameter settings is loaded into the high-fidelity simulation model in turn, and all or part of the test cases in the multi-scenario verification test case library are automatically run. This process simulates the application effect of the parameters in the actual control system, collects dynamic response data, and the dynamic response data includes the process variable curve over time, the actuator output signal, the error integral index, the overshoot, the regulation time and other key performance indicators. The batch test realizes comprehensive and efficient evaluation of the tuning scheme under multiple preset scenarios. The verification result is the comprehensive analysis conclusion of all performance data collected in the batch verification test, and the conservative adjustment strategy is a parameter fine-tuning method that tends to enhance the stability and robustness of the system. The core is not to pursue the theoretical optimal performance, but to actively increase the safety margin of the parameters on the premise of maintaining acceptable performance. Safety margin calibration makes subtle adjustments to the PID parameters, such as appropriately reducing the proportional gain to reduce the sensitivity of the system to high-frequency noise, or appropriately increasing the integral time to prevent integral saturation and reduce overshoot. The goal of calibration is to keep the parameter settings stable when facing unmodeled dynamics, equipment aging or unexpected disturbances. A final PID parameter setting instruction is generated, which is a set of explicit parameter values that can be directly recognized and executed by the PLC control system. The format of the instruction matches the requirements of the target PLC programming software or configuration tool, ensuring that the parameters can be accurately downloaded to the controller.

[0058] Consider a specific example of temperature control of a reaction kettle in a chemical process. The multi-scenario verification test case library will include the following test cases: one case simulates the startup process of the reaction kettle from room temperature to the set operating temperature, another case simulates the temperature disturbance recovery due to fluctuations in feed temperature during production, and a boundary condition case simulates the stability of temperature control when the heating medium pressure reaches the upper limit. The high-fidelity simulation model of the reaction kettle established by digital twin technology will include the heat capacity of the reaction kettle, the heat transfer coefficient of the jacket, the measurement delay and noise characteristics of the temperature sensor, the flow characteristics of the heating valve and its nonlinearity. The tuning recommendation scheme may include three different sets of PID parameters, one set for fast heating optimization, one set for disturbance rejection optimization, and one set for a compromise solution. In the simulation environment, the three sets of parameters will be tested separately. For fast heating parameters, they may exhibit excellent heating speed in the startup process test, but may exhibit significant overshoot oscillation in the disturbance rejection test. For disturbance rejection parameters, they may perform smoothly in the disturbance recovery test, but the startup process may be too slow. The batch verification test objectively records the performance data of each set of parameters under all test cases.

[0059] Analysis of the verification results will reveal the pros and cons of each set of parameters. A conservative tuning strategy can be applied to the set of compromised parameter scheme. The analysis results can show that the parameter scheme performs well under most operating conditions, but the stability margin is slightly insufficient under the boundary condition of the upper limit of heating medium pressure. The safety margin calibration criteria will fine-tune the PID parameters of the compromised scheme based on this finding, for example, slightly reducing the proportional gain Kp from the recommended 1.85 to 1.78, and increasing the integral time Ti from 125 seconds to 135 seconds. This calibration comes at the cost of slightly sacrificing the response speed under normal conditions in exchange for more reliable stability under boundary conditions. The calibrated parameter combination, together with the other two sets of unadjusted parameters with detailed performance descriptions, constitutes the final PID parameter setting instruction set. These instructions can be exported in standard file formats or directly sent to the PLC of the reactor temperature control through a communication interface, completing the last step of the parameter tuning process. The output step of the final PID parameter setting instruction significantly improves the probability of a one-time success of the parameters in actual application and reduces the risk of on-site debugging through strict virtual verification and conservative engineering adjustment.

[0060] While embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, combinations, and variations can be undertaken by those skilled in the art without departing from the spirit and scope of the present application, which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent tuning of PID parameters based on a PLC control system, characterized in that, The method includes: The process variable signals, set value signals and actuator output signals of the control system are acquired in real time through the PLC communication protocol. The acquired signals are smoothed, filtered and noise-reduced. The rise time, overshoot and steady-state error characteristics are extracted from the signals. The characteristics are mapped to the preset characteristic parameter fields and the time-series and causal relationships between the characteristics are set to form a standardized control data set. Using the standardized control data set, query the tuning strategy entries in the tuning knowledge base that match the characteristics of the control loop, obtain the tuning task complexity index, and rely on the complexity index to compare the performance threshold, response speed limit and stability range of the control data with the strategy requirements point by point, evaluate the compliance of parameter configuration, and generate a tuning status judgment for a single control loop. The steps for obtaining the tuning task complexity index include: parsing the dynamic characteristic patterns and steady-state operating ranges of the control loop from the standardized control data set; performing a deep traversal search in the tuning rule knowledge graph based on the dynamic characteristic patterns to obtain associated tuning rule clusters; analyzing the topological characteristics of the tuning rule clusters, including rule node degree, rule chain length, and rule cross-reference density; calculating the activation probability distribution and conflict detection coefficient of the rule clusters based on the boundary conditions of the steady-state operating range; and performing weighted fusion based on the topological characteristics and activation probability distribution to generate the tuning task complexity index. By combining multiple standardized control data sets, the shared process variables, associated actuators and detection instrument entities between different control loops are detected. The interaction intensity index between loops is calculated based on the frequency and type of shared entities. Based on the interaction intensity index, a network topology graph with control entities as vertices and shared connections as edges is drawn to generate a multi-loop interaction relationship graph. Based on the multi-loop interaction relationship graph, the tuning status judgment of the associated loops is integrated to calculate the overall tuning quality score and form a tuning recommendation scheme including quality level; the tuning recommendation scheme is verified in the field and the final PID parameter setting command is output through closed-loop control simulation test. The steps for obtaining the tuning state judgment of a single control loop include: establishing a similarity measurement space between the dynamic response of the control loop and the ideal response template; performing multi-dimensional comparison of the response trajectory within the measurement space, including phase trajectory overlap, amplitude envelope matching degree, and convergence speed deviation; dynamically updating the comparison benchmark using an adaptive threshold adjustment mechanism, and weighting the results by combining the comparison with the tuning task complexity index; generating a compliance metric score based on the corrected multi-dimensional comparison results, and generating a tuning state judgment based on the score interval mapping; The steps for obtaining the overall tuning quality score include: performing a multi-objective optimization path search on the multi-loop interaction graph to find the optimal path for the propagation of tuning effects; analyzing the tuning state transition probability and effect attenuation coefficient on the path; using a tuning effect evaluation method based on Monte Carlo simulation to statistically analyze the robustness index of the tuning scheme under uncertain operating conditions; and constructing a comprehensive tuning quality evaluation function by combining the topological characteristics and robustness index of the interaction graph to generate the overall tuning quality score.

2. The intelligent PID parameter tuning method based on a PLC control system according to claim 1, characterized in that, The steps for acquiring the standardized control data set include: parallel analysis of real-time acquired multi-channel process signals to separate the dynamic response component and the steady-state operating point component; extracting the feature contour of the dynamic response component using multi-scale waveform decomposition technology, and recording the change trajectory of the steady-state operating point component using an operating point drift tracking algorithm; spatiotemporally aligning the feature contour and the change trajectory to generate a time-stamped signal feature sequence; performing pattern matching between the signal feature sequence and a predefined feature template library, and performing dimensional reduction and normalization encoding on the successfully matched features to form standardized feature vectors; and establishing a feature correlation matrix based on the temporal dependency relationship between the standardized feature vectors to complete the construction of the standardized control data set.

3. The intelligent PID parameter tuning method based on a PLC control system according to claim 1, characterized in that, The steps for obtaining the inter-loop interaction strength index include: constructing a hypergraph model of control loop associations, where hyperedges represent the coupling relationships between multiple loops; analyzing the node centrality and hyperedge weight distribution of each loop in the hypergraph; using a coupling strength detection algorithm based on random walks to calculate the energy transfer efficiency and interference propagation path between loops; generating a multi-dimensional interaction feature vector by combining the overlap of loop operating frequency bands and the synchronization of control timing; and performing principal component analysis and weight allocation on the multi-dimensional interaction feature vector to synthesize the inter-loop interaction strength index.

4. The intelligent PID parameter tuning method based on a PLC control system according to claim 1, characterized in that, The steps for obtaining the multi-loop interaction relationship graph include: establishing a control loop entity knowledge graph, where nodes contain loop static attributes and dynamic operating states; using graph neural network technology to learn the implicit relationships between nodes and mine potential interaction patterns; dynamically calibrating edge weights based on the interaction intensity index to construct a weighted interaction network; and introducing a temporal graph attention mechanism to capture the evolutionary rules of interaction relationships and generate a multi-loop interaction relationship graph with a time dimension.

5. The intelligent PID parameter tuning method based on a PLC control system according to claim 1, characterized in that, The steps for generating the tuning recommendation scheme include: establishing a multi-peak distribution model of the tuning parameter solution space and using a heuristic region search algorithm to locate high-quality parameter regions; prioritizing the parameter regions based on the overall tuning quality score; generating parameter boundary cases using generative adversarial network technology to test the stability boundary of the tuning scheme; and combining a multi-objective optimization algorithm to find the Pareto optimal solution set in the high-quality parameter regions to form the tuning recommendation scheme.

6. The intelligent PID parameter tuning method based on a PLC control system according to claim 1, characterized in that, The steps for outputting the final PID parameter setting instruction include: designing a multi-scenario verification test case library covering typical operating conditions and boundary conditions; establishing a high-fidelity simulation model of the controlled object using digital twin technology; performing batch verification tests of the recommended tuning scheme in the simulation environment and collecting dynamic response data; and calibrating the parameters with a safety margin using a conservative adjustment strategy based on the verification results to generate the final PID parameter setting instruction.

7. A PID parameter intelligent tuning system based on a PLC control system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent PID parameter tuning method based on the PLC control system as described in any one of claims 1 to 6.

Images