Real-time data processing analysis method and system of industrial PLC controller

Abstract

The invention relates to the technical field of data processing, and discloses a real-time data processing analysis method and system of an industrial PLC. The method comprises the following steps: transmitting temperature, pressure, current, vibration and acoustic parameters acquired by multiple sensors to an industrial PLC (Programmable Logic Controller) in real time to obtain multi-source heterogeneous original data; preprocessing the multi-source heterogeneous original data to obtain standardized fusion data; correlation calculation and anomaly recognition are carried out through the multivariate analysis model, and an abnormal state classification result is obtained; dynamically adjusting data interaction frequency and sampling rate between the edge nodes and the central PLC, and generating a real-time control decision instruction; and matching the real-time control decision instruction with the current motor load fluctuation state, and outputting the optimal frequency conversion control parameter. According to the invention, the response delay of the system is reduced, the control precision and reliability are improved, the dynamic balance between the safety and the energy efficiency is realized, and the system can intelligently adjust the control strategy according to the real-time safety situation.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a real-time data processing and analysis method and system for an industrial PLC controller. Background Art

[0002] Traditional industrial PLC controllers face problems of insufficient real-time performance and limited analysis capabilities when dealing with a large amount of multi-source heterogeneous data. Especially in the context of the widespread application of variable frequency speed regulation technology, industrial control systems need to achieve precise control of energy consumption while ensuring the safe operation and response speed of equipment, which poses a severe challenge to traditional PLC control architectures. In the prior art, the processing of industrial data mostly uses single statistical methods or simple threshold judgments, and cannot effectively identify abnormal states under complex working conditions, resulting in insufficient control accuracy and serious energy waste.

[0003] The communication link of industrial control systems is another issue worthy of attention. In traditional systems, data acquisition and control decisions often adopt a communication mechanism with a fixed frequency, and cannot dynamically adjust data interaction strategies according to changes in the system operation state, which is likely to cause information lag or network congestion in abnormal states. At the same time, there is a lack of a collaborative optimization mechanism between different execution units and the central controller, and edge computing resources are not fully utilized, resulting in system response delays and low control decision-making efficiency. Summary of the Invention

[0004] The present invention provides a real-time data processing and analysis method and system for an industrial PLC controller, which reduces system response delays, improves control accuracy and reliability, achieves a dynamic balance between security and energy efficiency, and enables the system to intelligently adjust control strategies according to the real-time security situation.

[0005] In a first aspect, the present invention provides a real-time data processing and analysis method for an industrial PLC controller, and the real-time data processing and analysis method for the industrial PLC controller includes: Using the OPC UA communication protocol to transmit temperature, pressure, current, vibration, and acoustic parameters collected by multiple sensors to the industrial PLC controller in real time, obtaining multi-source heterogeneous raw data; Performing data cleaning, normalization processing, and weighted clustering fusion on the multi-source heterogeneous raw data to obtain normalized fusion data; Inputting the normalized fusion data into a multivariate analysis model for correlation calculation and anomaly identification to obtain an anomaly state classification result; Dynamically adjusting the data interaction frequency and sampling rate between the edge node and the central PLC controller according to the anomaly state classification result to generate a real-time control decision instruction; Matching the real-time control decision instruction with the current motor load fluctuation state and outputting optimal variable frequency control parameters.

[0006] In a second aspect, the present invention provides a real-time data processing and analysis system for an industrial PLC controller. The real-time data processing and analysis system for the industrial PLC controller includes: An acquisition module, configured to use the OPC UA communication protocol to transmit temperature, pressure, current, vibration, and acoustic parameters collected by multiple sensors to the industrial PLC controller in real time, so as to obtain multi-source heterogeneous raw data; A fusion module, configured to perform data cleaning, standardization processing, and weighted clustering fusion on the multi-source heterogeneous raw data to obtain standardized fusion data; A calculation module, configured to input the standardized fusion data into a multivariate analysis model for correlation calculation and anomaly identification to obtain an anomaly state classification result; A dynamic adjustment module, configured to dynamically adjust the data interaction frequency and sampling rate between the edge node and the central PLC controller according to the anomaly state classification result, and generate a real-time control decision instruction; An output module, configured to match the real-time control decision instruction with the current motor load fluctuation state and output optimal variable frequency control parameters.

[0007] In the technical solution provided by the present invention, by constructing a three-layer industrial control system architecture and a five-layer parallel structure of the PLC controller, and combining the secure transmission mechanism of the OPC UA communication protocol, the data processing ability and communication efficiency of the industrial control system are significantly improved, enabling the system to process multi-source heterogeneous data safely and efficiently. The data cleaning, standardization processing, and weighted clustering fusion methods adopted effectively solve the problem of multi-source heterogeneous data fusion in industry, improve the data quality, and provide a high-quality data basis for subsequent analysis. By integrating parameter correlation analysis, change propagation analysis, and three complementary anomaly detection methods through a multivariate analysis model, various anomaly states in the system can be comprehensively identified, improving the accuracy and reliability of anomaly detection. The "edge-center" cooperation framework dynamically adjusted based on the anomaly state classification result realizes a control mode of "processing nearby and making hierarchical decisions", significantly reducing the system response delay, and improving the control accuracy and reliability. By adopting an adaptive variable frequency control algorithm, it can adopt a differentiated control strategy according to the load fluctuation characteristics, minimize energy consumption while ensuring control accuracy, and effectively balance the relationship between system security and energy efficiency. Through a hierarchical security situation assessment and collaborative optimization control method, through the comprehensive assessment of the execution unit layer, controller layer, and network communication layer, combined with the parameter deviation risk degree assignment, the dynamic balance between security and energy efficiency is achieved, enabling the system to intelligently adjust the control strategy according to the real-time security situation. Description of the Drawings

[0008] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0009] Figure 1 It is a schematic flowchart of the real-time data processing and analysis method for the industrial PLC controller provided by the embodiments of this application; Figure 2 It is a schematic block diagram of the structure of the real-time data processing and analysis system for the industrial PLC controller provided by the embodiments of this application. Detailed Embodiments

[0010] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0011] The flowchart shown in the drawings is only an example illustration, and does not necessarily include all contents and operations / steps, nor does it necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged. Therefore, the actual execution order may change based on the actual situation.

[0012] It should also be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0013] It should be further understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0014] The following will describe in detail some embodiments of this application in conjunction with the drawings. Without conflict, the features in the following embodiments and the embodiments can be combined with each other.

[0015] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of the real-time data processing and analysis method for the industrial PLC controller provided by the embodiments of this application. As Figure 1As shown in the figure, the real-time data processing and analysis method of the industrial PLC controller provided by the embodiment of the present application includes steps S100 to S600.

[0016] Step S100: Use the OPC UA communication protocol to transmit the temperature, pressure, current, vibration, and acoustic parameters collected by multiple sensors to the industrial PLC controller in real time, and obtain multi-source heterogeneous raw data; It can be understood that the execution subject of the present invention can be the real-time data processing and analysis system of the industrial PLC controller, or a terminal or a server. Specifically, it is not limited here. In the embodiment of the present invention, the server is used as the execution subject for illustration.

[0017] Specifically, the industrial control system is structurally divided, and a three-layer industrial control system architecture is constructed according to the logic of the execution layer, control layer, and data acquisition layer. Among them, the execution layer is composed of various actuators, which are used to receive control instructions and complete actual operation actions; the core of the control layer is the industrial PLC controller, which receives data, performs logic control, and outputs control instructions; while the data acquisition layer undertakes the perception tasks of various process parameters, and heterogeneous acquisition devices such as temperature sensors, pressure sensors, current sensors, vibration sensors, and acoustic sensors need to be deployed to build a comprehensive parameter monitoring network. Inside the PLC controller in the control layer, a five-layer parallel structure is implemented. This structure includes a data interface layer, a data preprocessing layer, a logic control layer, a high-level analysis layer, and a decision output layer. And each layer is interconnected by a high-speed bus, and at the same time, a double-buffer mechanism is introduced, so that the reception, processing, and response of data can run in parallel without blocking each other, improving data processing efficiency and response real-time performance. Under the multi-level industrial control architecture, various sensors configured in the data acquisition layer collect the real-time operating status of industrial equipment at a high frequency according to preset sampling frequencies - for example, temperature 10Hz, pressure 50Hz, current 100Hz, vibration 200Hz, and acoustic 500Hz - and encapsulate the heterogeneous raw data into data frames in a standardized data format. The multi-source heterogeneous raw data is encrypted with 256 bits, and an encrypted session is established through a digital certificate authentication mechanism to form an encrypted data packet with a high security level. The encrypted data packet is transmitted through the unified OPC UA communication protocol. This protocol has platform independence, structured information modeling ability, and built-in security mechanisms to ensure that multi-source data is not eavesdropped, tampered with, or forged during the transmission process. When the encrypted data stream arrives at the PLC controller, it is received by the data interface layer, and a decryption program is started to perform data decryption operations on it to verify the integrity and source legality of the data. After successful decryption, the data packet will be parsed and restored to various parameter values to form a standardized set of raw parameter data. The raw parameter data is stored in the double-buffer memory area of the PLC controller, and the data writing and reading are respectively performed, and the execution is switched alternately, so as to ensure that there is no read-write conflict during the high-speed data processing process, and multi-source heterogeneous raw data is obtained.

[0018] Step S200: Perform data cleaning, standardization processing, and weighted clustering fusion on the multi-source heterogeneous raw data to obtain standardized fusion data; Specifically, within the data preprocessing layer of the PLC controller, a data cleaning mechanism is initiated to identify and remove outliers from the multi-source heterogeneous raw data collected from temperature, pressure, current, vibration, and acoustic sensors. The cleaning method adopts a judgment criterion based on threshold setting. By presetting reasonable numerical ranges for various parameters, extreme values and noise points beyond the normal operating conditions are identified and removed, resulting in a preliminarily cleaned multi-parameter data set. Low-pass filtering is performed on the temperature data in the preliminarily cleaned data. A fifth-order Butterworth low-pass filter is selected, and its cut-off frequency is set to 5 Hz to effectively filter out high-frequency interference. The vibration signal exhibits strong non-stationarity and transient characteristics. The wavelet packet transform method is used for multi-scale decomposition and reconstruction. The db4 wavelet basis is selected, and the decomposition level is set to 3 layers. The signal is reconstructed by removing high-frequency wavelet coefficients to obtain more representative and noise-reduced vibration parameter data. A standardization transformation operation is performed on the parameter data set after noise reduction processing. Parameters with different dimensions and value ranges are uniformly mapped to the standard interval of [-1,1]. The standardization method used is range normalization, enabling parameters such as temperature, pressure, current, vibration, and acoustics to be compared and calculated in the same numerical space. After standardization, weight coefficients are assigned to each parameter according to its importance and credibility in industrial control, constructing a weighted parameter set. The setting of the weights is optimized through engineering experience and data analysis based on equipment characteristics and data stability indicators. The weighted Euclidean distance matrix between data points is calculated based on the weighted parameter set. On the basis of the weighted Euclidean distance matrix, a weighted clustering algorithm, namely the improved WECA algorithm, is executed. This algorithm optimizes the class division through an iterative method, minimizing the sum of the weighted distances within the class and maximizing the weighted distances between classes, obtaining a clustering result with optimal internal consistency and maximum external distinguishability, and getting the standardized fusion data.

[0019] Step S300: Input the standardized fusion data into a multi-variable analysis model for correlation calculation and anomaly identification to obtain an anomaly status classification result; Specifically, the standardized fusion data is input into the parameter correlation analysis module in the multivariate analysis model. In this module, an improved Pearson correlation coefficient algorithm is used to calculate the bivariate correlation between each parameter. By setting a correlation threshold, significant parameter pairs are screened out, and a preliminary parameter influence network is constructed based on the high correlation relationship. To eliminate the spurious dependence caused by indirect correlation, a partial correlation analysis method is introduced to determine the direct action path between variables, and a direct parameter influence relationship network is constructed. This network diagram reflects the direct dependence paths between various process parameters, either physically or logically. The direct parameter influence relationship network is input into the parameter change propagation model for calculating the propagation coefficient. This model, based on the construction logic of network edge weights and historical operation data, solves how the changes of different parameters are conducted through this network in the system, quantifies the influence degree and lag relationship of each parameter on other parameters and the overall system state, and obtains a set of conduction influence coefficient matrices describing the mutual influence strength between parameters. Combining the standardized fusion data with the above conduction influence coefficients, they are input into three complementary anomaly detection models. In the statistical anomaly detection model, the conduction influence coefficient is used to weight the importance of parameters, and then combined with the deviation between the current value of the standardized parameter and its historical mean, the weighted mean deviation of each parameter is calculated and compared with a preset differential threshold to output a statistical anomaly score reflecting the overall parameter deviation degree; in the pattern anomaly detection model, the time series feature vector of the standardized fusion data is extracted and dynamically time-warped and matched with the typical operation pattern under the historical normal working condition, and then the similarity between the current operation state and the known normal pattern is judged to obtain a pattern anomaly degree index describing the pattern deviation degree; in the prediction anomaly detection model, the conduction influence coefficient is used to guide the structure and input selection of the LSTM prediction model, and a time series model for short-term parameter trend prediction is obtained through training, and the predicted value is compared with the current observed value. When the deviation exceeds the set tolerance, it is regarded as a higher anomaly degree, and the prediction anomaly degree is output accordingly. The above three anomaly results in different dimensions, namely the statistical anomaly score, the pattern anomaly degree, and the prediction anomaly degree, are weighted and summed according to the preset weighting coefficients to obtain the final comprehensive anomaly score. According to the interval value where the comprehensive anomaly score is located, the current system operation state is divided into four levels: "normal", "slight anomaly", "moderate anomaly", and "severe anomaly", and an anomaly state classification result is generated.

[0020] In this embodiment, the standardized fusion data is structurally processed. The data set is logically grouped according to different parameter types and device sources and divided into multiple parameter matrices. Each matrix represents a parameter set with similar sampling characteristics or physical associations. For each parameter matrix, the mean vector and standard deviation vector are calculated respectively in the time dimension to extract the statistical characteristics of the group of parameters in the time series. These statistical characteristics effectively represent the distribution form and fluctuation range of the parameters. Based on the parameter statistical characteristics, the Pearson algorithm is executed on the standardized fusion data. By normalizing the covariance between any two parameter vectors, their correlation coefficients are calculated, and the numerical range is limited to the interval [-1, 1]. The closer the value is to ±1, the stronger the linear correlation. Through the calculated correlation coefficient matrix between parameters, a preliminary parameter correlation network is constructed. This network reflects the statistical coupling relationship between all parameters in the observed data, that is, the initial correlation relationship. To avoid redundant calculations and misjudgments, the initial correlation relationship is screened. Weak correlation edges below the set threshold are removed, and only the edges with the absolute value of the correlation coefficient higher than the specified threshold are retained. And the strongly correlated parameter combinations are analyzed to extract the strongly correlated network that constitutes the core system structure. The correlation coefficients in the strongly correlated network are converted into Z values through the Fisher Z transformation. Through the transformation formula Z = 0.5×ln((1 + r) / (1 - r)), the original correlation coefficients approximately follow a normal distribution after the transformation, which is convenient for hypothesis testing and conditional independence analysis. Based on the values after the Z transformation, the independence between two parameters under the condition of controlling other variables is estimated, and their conditional independent probabilities are calculated to determine whether there is a direct influence relationship between the variables. If two parameters still have significant statistical dependence after controlling other variables, it means they have a direct correlation. Calculate the correlation strength and directionality for the correlation coefficients in the direct correlation network. The strength is measured by the absolute value of the Z value, and the directionality is determined by combining time lag analysis, Granger causality test or information flow-based indicators to determine the direction of the causal relationship. For example, when the change of one parameter always precedes another parameter and can significantly improve the prediction ability of the latter, it is considered that the former has a directed influence on the latter. Based on the above analysis results, all the edges with associated direction and influence strength are incorporated into the construction process to form a weighted directed graph structure including parameter nodes, directed edges and edge weights. This structure is the parameter direct influence relationship network.

[0021] In this embodiment, the structure of the parameter direct influence relationship network is transformed and mapped into a formal directed weighted graph structure. The nodes in this structure represent the key process parameters in the system, while the edges represent the directed influence relationships between the parameters, and the edge weights correspond to the influence intensity or correlation values. Thus, a system parameter topology graph for propagation modeling is formed. The out-degree and in-degree of each parameter node are statistically calculated. The out-degree represents the total quantity and intensity of the influence of this parameter on other parameters, and the in-degree represents the total amount of influence it receives from other parameters. By combining the calculations of these two indicators, the information conduction ability and structural position of each parameter in the entire system topology are evaluated, and the parameter importance ranking is obtained to identify the dominant variables and key nodes during the system operation. Based on the parameter importance ranking results, the system parameter topology graph is modeled as a state transition probability graph, and a first-order conduction matrix between the parameters is constructed using the Markov chain modeling method. In this conduction matrix, each row represents the direct influence transition probability of one parameter state on other parameter states, and its element values are calculated based on the directed edge weights between the parameters and normalization processing, ensuring that the matrix has the properties of a transition probability matrix, that is, the sum of the elements in each row is 1. This conduction matrix depicts the possibility of the transfer of each parameter state to adjacent parameters at any time step of the system. To simulate the information conduction process of multiple rounds and multiple paths, the first-order conduction matrix is expanded by powers, and its square, cube, and up to the nth power matrix are calculated in sequence to form a set of high-order conduction matrices, which are added in order to construct a multi-order comprehensive conduction matrix, thereby comprehensively reflecting the superimposed influence effects generated by different parameters through various paths in the entire system. The row vectors of the multi-order comprehensive conduction matrix are normalized so that they represent the distribution probability of the influence propagated by this parameter on all paths, and a parameter influence distribution vector group is obtained. Each vector represents the distribution weight of the indirect or direct influence intensity that a parameter change may cause among other parameters in the system, reflecting its potential system interference ability. The parameter influence distribution vector group is subjected to sensitivity analysis, and the response degree of the system state to the input parameter perturbation is quantified as a sensitivity factor. By calculating the partial derivative or response amplitude of the change in the overall state function caused by each parameter change and combining the weights of each dimension in the above distribution vector, the final conduction influence of the parameter change on the system state is comprehensively evaluated, and a set of stable and clearly structured conduction influence coefficients is output.

[0022] Step S400: Dynamically adjust the data interaction frequency and sampling rate between the edge node and the central PLC controller according to the abnormal state classification result, and generate a real-time control decision instruction; Specifically, the classification results of abnormal states are subject to level recognition, and the system operation state is divided into four levels: "normal", "slight abnormality", "moderate abnormality", and "severe abnormality", obtaining the abnormal level judgment result of the current system. This judgment result will serve as the trigger basis for the subsequent dynamic scheduling mechanism, guiding the adjustment of data collection and communication strategies to respond to different system risk levels. According to this abnormal level judgment result, the data interaction frequency between the edge node and the central PLC controller is dynamically reconfigured. In the "normal" state, low-frequency summary communication is adopted; in the "slight abnormality" state, it is upgraded to medium frequency; when the abnormal level rises to "moderate", the frequency is further increased; and in the "severe abnormality" state, it enters the emergency communication mode and implements a real-time push mechanism. Based on these strategies, a data interaction frequency configuration table is generated, which is dynamically loaded into the edge node communication scheduling module according to the level to control the data upload rhythm and bandwidth allocation strategy. At the same time, according to the abnormal level judgment result, the sampling frequencies of various sensors deployed in the data collection layer are adjusted in real time, and a sensor sampling frequency configuration table is constructed. Under normal conditions, the sensors maintain the original configured frequency; when the system is in a slight or moderate abnormal state, the sampling frequencies of key parameters (such as vibration and current) are appropriately increased to enhance the accuracy and sensitivity of abnormal detection; and in the severe abnormal state, the sampling frequencies of all key parameters are increased by more than double to ensure that the dynamic state changes of the system are accurately captured. The above two configuration tables are synchronously transmitted to the hierarchical control unit in the edge-center cooperation framework. This unit includes three hierarchical modules: the basic control layer, the optimization control layer, and the policy control layer. Through this control unit, a multi-dimensional analysis and cross-diagnosis of the current system operation state are carried out to construct a control strategy priority table reflecting the urgency and priority of each control strategy. This table is used to clarify the scheduling order and resource allocation priority of different control behaviors (such as parameter adjustment, response delay suppression, and target redefinition) in the current state. According to the control strategy priority table and the current system operation parameter data, the basic control layer adjusts the coordination strategy between edge nodes to ensure the time synchronization and logical consistency of the actions of different control units; the optimization control layer performs parameter calculation tasks based on the overall system state and preliminarily adjusts the controller output parameters such as frequency, voltage, and flow rate; while the policy control layer corrects and directionally plans the overall control strategy based on the production plan, energy consumption target, and safety boundary conditions, generating a preliminary control parameter adjustment plan. Before entering the execution link, this preliminary plan will perform control margin calculation and safety verification, calculating key indicators such as the occupancy degree of the adjustment action on the system stability boundary, the suppression ability of high-frequency disturbances, and the remaining response space of the controller, so as to ensure that the execution of the plan will not trigger secondary risks. After passing the verification, a real-time control decision instruction is finally generated.

[0023] Step S500: Match the real-time control decision instruction with the current motor load fluctuation state and output the optimal variable-frequency control parameters.

[0024] Specifically, the operating status of target motors at industrial sites under different frequency and load factor combinations is systematically collected, recording multiple key parameters including input power, output power, speed, current, and temperature. Based on these experimental data, a mathematical modeling process is constructed. A fourth-order polynomial fitting method is used, with motor efficiency η as the objective function and frequency f and load factor L as independent variables. The efficiency model η = f(f, L) is constructed to ensure high fitting accuracy at all combinations, such as an R² value exceeding 0.95. Furthermore, an optimization method is used to extract the frequency-load relationship with optimal efficiency and construct an optimal frequency-load comparison table. The real-time motor load data acquired from the acquisition system undergoes fluctuation analysis. An exponentially weighted moving average method is used to eliminate short-term noise interference. Combined with standard deviation calculation within a sliding window, the load fluctuation intensity within a specific time range is identified. Based on a set fluctuation threshold, the current load state is classified into four categories: stable operating condition, small fluctuation condition, moderate fluctuation condition, and large fluctuation condition. The load fluctuation classification results are then output. Based on this classification, a condition-strategy matching table is constructed, with a pre-set control strategy for each condition. For example, a lookup-based constant-value control strategy is used for stable conditions, a PI smoothing frequency modulation strategy is used for conditions with small fluctuations, a short-term prediction mechanism is introduced for medium-fluctuation conditions to achieve advance adjustments, and a segmented optimization control logic is applied for conditions with large fluctuations, using dynamic segmentation and frequency fine-tuning to cope with drastic changes. Real-time control decision instructions are matched with this condition-strategy matching table to determine the most suitable control scheme for the current system state. This ensures that control decisions achieve optimal energy efficiency while also taking into account system response characteristics and operational safety. Based on the matched control solution, it is fused with the optimal frequency-load comparison table to form a control mechanism driven by an empirical model and real-time strategy. For stable operating conditions, the optimal frequency from the comparison table is directly called. For small fluctuations, the frequency from the comparison table is used as the target value, and a PI controller is applied to achieve small, smooth adjustments of the frequency around the target value, with an adjustment step size of no more than 1.0 Hz / s. For medium fluctuations, future load trends are predicted by constructing time series prediction models such as LSTM, and the predicted values are used to guide frequency changes, making the adjustment process more forward-looking. For large fluctuations, the entire load curve is divided into multiple short-period stable segments. The optimal frequency is re-evaluated in each segment to form a set of intra-segment control instructions. Through fast switching and real-time feedback mechanisms, more efficient energy consumption control and stable operating conditions are achieved. This multi-step fusion calculation process outputs the optimal variable frequency control parameters, which include the current target frequency value, frequency adjustment step size, control period, and dynamic correction factor.

[0025] In the embodiments of the present invention, by constructing a three-layer industrial control system architecture and a five-layer parallel structure of the PLC controller, and combining the secure transmission mechanism of the OPC UA communication protocol, the data processing ability and communication efficiency of the industrial control system are significantly improved, enabling the system to securely and efficiently process multi-source heterogeneous data. The data cleaning, standardization processing, and weighted clustering fusion methods adopted effectively solve the problem of multi-source heterogeneous data fusion in industry, improve the data quality, and provide a high-quality data basis for subsequent analysis. Through the multi-variable analysis model, the parameter correlation analysis, change propagation analysis, and three complementary anomaly detection methods are integrated, enabling comprehensive identification of various abnormal states in the system and improving the accuracy and reliability of anomaly detection. Based on the "edge-center" collaboration framework dynamically adjusted according to the anomaly state classification results, the control mode of "proximity processing and hierarchical decision-making" is realized, significantly reducing the system response delay and improving the control accuracy and reliability. The adaptive variable-frequency control algorithm can adopt different control strategies according to the load fluctuation characteristics, minimizing the energy consumption while ensuring the control accuracy, and effectively balancing the relationship between system security and energy efficiency. Through the hierarchical security situation assessment and cooperative optimization control method, through the comprehensive assessment of the execution unit layer, controller layer, and network communication layer, combined with the parameter deviation risk degree assignment, the dynamic balance between security and energy efficiency is achieved, enabling the system to intelligently adjust the control strategy according to the real-time security situation.

[0026] In a specific embodiment, the process of executing step S100 may specifically include the following steps: Divide the industrial control system into three layers, namely the execution layer, control layer, and data acquisition layer. Implement a five-layer parallel structure for the PLC controller in the control layer to obtain a multi-level industrial control system architecture; Deploy temperature sensors, pressure sensors, current sensors, vibration sensors, and acoustic sensors in the data acquisition layer, and collect the corresponding parameters respectively according to the preset sampling frequency to obtain multi-source heterogeneous raw data; Perform 256-bit encryption and certificate authentication processing on the multi-source heterogeneous raw data to obtain a securely encrypted transmission data packet, and transmit the securely encrypted transmission data packet to the data interface layer of the PLC controller through the OPC UA communication protocol to obtain the encrypted data stream to be processed; Perform decryption verification and data unpacking processing on the encrypted data stream to be processed to obtain the original parameter data, and store the original parameter data in the dual-buffer memory area of the PLC controller to obtain multi-source heterogeneous raw data.

[0027] Specifically, based on the function division and data flow characteristics of the industrial automation control system, the entire system is divided into three mutually supportive and collaborative architecture levels, namely the execution layer, the control layer, and the data acquisition layer. Among them, the execution layer is mainly responsible for the final implementation of all specific actions in the industrial field, including equipment such as frequency converters, motors, robotic arms, actuating solenoid valves, and cylinders. It responds immediately according to the instructions issued by the controller and completes production tasks. The control layer, as the core decision-making and computing center of the entire system, is composed of PLC controllers with strong computing and communication capabilities. It receives data acquisition signals from the lower layer, executes logic control, process modeling, anomaly analysis, and strategy output, and accurately issues instructions to the equipment in the execution layer to form a closed-loop response process. At the same time, the data acquisition layer, as the original source of information, is equipped with a variety of high-precision and low-latency sensor devices for real-time collection of various key process parameters, including temperature, pressure, current, vibration, and acoustic information. By collecting and integrating these physical quantities, a dynamic perception portrait of the system state is formed. In the PLC controller within the control layer, in order to break through the technical bottlenecks of limited processing capacity, low module integration, and lagging response speed of the traditional PLC structure, an internal system division of the "five-layer parallel structure" is implemented. This structure consists of five independent but highly collaborative functional modules: the data interface layer, the data preprocessing layer, the logic control layer, the advanced analysis layer, and the decision output layer. Among them, the data interface layer receives external data streams from the acquisition layer and completes preliminary access protocol management and caching operations; the data preprocessing layer filters, denoises, normalizes, and time-aligns the original signals; the logic control layer executes the core logic control instructions of the PLC to achieve real-time response to the industrial process; the advanced analysis layer is used for parameter correlation modeling, trend prediction, and anomaly identification; and the decision output layer is responsible for converting the comprehensive analysis results into specific executable control instructions and feeding them back to the execution layer. At the same time, to ensure efficient data transmission and real-time response between modules, a high-speed data bus is introduced inside the controller, supplemented by a double-buffer structure. One piece of memory is set for data writing and the other for reading and processing in the control loop. The two buffers alternate, effectively avoiding data access conflicts and improving the overall parallel processing ability, forming a high-order control system with the capabilities of real-time data access, parallel analysis, and fast response. In the data acquisition layer, a variety of heterogeneous sensing devices such as temperature sensors (sampling frequency 10Hz), pressure sensors (50Hz), current sensors (100Hz), vibration sensors (200Hz), and acoustic sensors (500Hz) are deployed. The sampling frequency is configured differently according to different physical parameters to ensure optimal perception of the on-site state without wasting system resources. The data collected by all sensors are encapsulated into multi-source heterogeneous raw data packets. Since these data come from different devices, have different physical dimensions and precisions, they are uniformly and securely encapsulated before transmission.The raw data is encrypted using the 256-bit AES symmetric encryption algorithm. Combined with an authentication mechanism based on the X.509 certificate system, the sensor identity and data source are verified to create a secure encrypted data packet. The encrypted data packet is transmitted via the service-based OPC UA communication protocol, which offers excellent cross-platform compatibility, information modeling capabilities, and built-in security mechanisms within the Industrial Internet of Things (IIoT) architecture, supporting stable connections between devices and expressing data context. After the encrypted data packet enters the PLC controller's data interface layer via the OPC UA channel, the data access module initiates decryption, verification, and unpacking. This process includes key identification, symmetric decryption, certificate verification, and data structure parsing, ensuring that the received data has not been tampered with and that its source is trustworthy. The decrypted raw parameter data is written to the write port of the controller's internal double buffer. When the data processing thread completes its reading task from the previous cycle, the controller swaps the buffers by switching memory pointers, automatically making the newly received data in the current cycle the input for the processing thread, resulting in multi-source heterogeneous raw data.

[0028] In a specific embodiment, the process of executing step S200 may specifically include the following steps: Perform data cleaning on multi-source heterogeneous raw data, remove outliers in temperature, pressure, current, vibration, and acoustic parameters, and obtain preliminary cleaned data; Low-pass filtering is performed on the temperature data in the data after preliminary cleaning, and wavelet packet transform is performed on the vibration data in the data after preliminary cleaning to reduce noise, so as to obtain the parameter data after noise reduction; The parameter data after noise reduction is normalized to obtain normalized data with unified dimension, and weight coefficients are assigned to the temperature, pressure, current, vibration and acoustic parameters in the normalized data with unified dimension to obtain a weighted parameter set; The weighted Euclidean distance matrix between data points is calculated based on the weighted parameter set, and weighted clustering is performed on the weighted Euclidean distance matrix to obtain standardized fused data by minimizing the intra-class weighted distance and maximizing the inter-class weighted distance.

[0029] Specifically, for multi-source heterogeneous raw data, data cleaning is performed. Corresponding cleaning thresholds are preset according to the physical ranges of different parameters. In each sampling period, it is determined one by one whether the data points exceed the reasonable range. If they exceed, they are marked as abnormal points and removed. If they are missing data, interpolation or linear filling methods are used for processing to construct a preliminary cleaned data set. After the cleaning is completed, in the face of the time-domain characteristics and frequency-domain structures reflected by different parameters, targeted noise reduction strategies are executed for specific data types to extract more representative information. Among them, for temperature data, since it shows a slow-changing process in the industrial control process and is easily affected by low-frequency disturbances such as process start and stop, a low-pass filter is used for processing. A fifth-order Butterworth low-pass filter is selected as the noise reduction tool for the temperature signal, and its cut-off frequency is set to 5 Hz to effectively filter out the high-frequency interference components in the input signal while retaining the main information of the temperature change trend. For vibration data, its typical characteristics are strong non-stationarity and local mutability. Wavelet packet transform is used to perform multi-scale decomposition and reconstruction processing on it. Db4 is selected as the wavelet basis function and the decomposition depth is set to 3 layers. By performing coefficient compression and denoising operations on the wavelet coefficients in different frequency bands, the denoised vibration signal is reconstructed. The denoised parameter data is subjected to standardization transformation. The range normalization method is used to map each parameter to the interval [-1, 1], unifying the numerical scales of five types of parameters, namely temperature, pressure, current, vibration, and acoustics, so that all parameters are comparable in the same dimension space. To express the importance of each parameter in system control decision-making and state evaluation, a weight coefficient is assigned to each standardized parameter, which is determined by combining multi-dimensional information such as expert experience, parameter stability, and influence on control effects, forming a weighted parameter set. A sample space is constructed based on the five-dimensional parameter vector of each data point, and weights are introduced according to the Euclidean distance formula to calculate the weighted Euclidean distance between data points. The difference in each dimension is multiplied by the square of the corresponding weight and then participates in the calculation of the total distance. This method effectively overcomes the clustering bias caused by unequal weights of each dimension feature and enhances the sensitivity of the clustering boundary to the parameter meaning. The improved weighted clustering algorithm is used to process the weighted distance matrix, and a clustering optimization mechanism aiming to minimize the within-class weighted distance and maximize the between-class weighted distance is selected. By iteratively adjusting the initial clustering center, reassigning the belonging categories of data points, and updating the within-class centroid, it gradually converges to a stable clustering result. The minimization of the within-class weighted distance ensures that the data points within the same cluster have high similarity, and the maximization of the between-class weighted distance enhances the discrimination ability between clusters, obtaining a standardized fusion data set with clear structure, distinct boundaries, and suitable for multi-parameter working condition analysis.

[0030] In a specific embodiment, the process of executing step S300 may specifically include the following steps: Input the standardized fusion data into the parameter correlation analysis model in the multivariate analysis model to calculate the correlation coefficient and construct the relationship network, and obtain the parameter direct influence relationship network; Input the parameter direct influence relationship network into the parameter change propagation model to calculate the propagation coefficient, and obtain the conduction influence coefficient of different parameter changes on the system state; Input the standardized fusion data and the conduction influence coefficient into the statistical anomaly detection model to calculate the mean deviation and threshold judgment, and obtain the statistical anomaly score; Input the standardized fusion data and the conduction influence coefficient into the pattern anomaly detection model to extract time series features and perform pattern matching, and obtain the pattern anomaly degree; Input the standardized fusion data and the conduction influence coefficient into the prediction anomaly detection model to perform time series prediction, and obtain the prediction anomaly degree; Perform weighted summation on the statistical anomaly score, the pattern anomaly degree and the prediction anomaly degree to obtain the comprehensive anomaly score, and generate the anomaly state classification result according to the score interval.

[0031] Specifically, the standardized fusion data is input into the parameter correlation analysis model in the multivariate analysis model. The pairwise correlations of all parameter pairs are calculated by the improved Pearson correlation coefficient method, and the indirect action relationships are eliminated through partial correlation to establish an initial correlation coefficient matrix between parameters. Based on this matrix, for parameter pairs with an absolute correlation value greater than the set threshold, their directed edges are retained, and the correlation values are standardized using the Fisher Z transformation. At the same time, their conditional independence is evaluated through hypothesis testing, and the directly related edges with statistical significance are screened out to construct a real and directional parameter direct influence relationship network. In this network, the nodes represent the key process parameters in the system, the edges represent the direct action relationships between parameters, and the edge weights reflect the action intensity. The formed network has clear structure and also has propagation directionality. As a structural model of the system state evolution, this network is input into the parameter change propagation model to analyze the influence mechanism of the change of any parameter propagating in the network and affecting other parameters and even the overall system state. The parameter direct influence relationship network is converted into an adjacency matrix representation of a weighted directed graph, and a first-order conduction matrix is constructed, where each row vector represents the direct influence probability or weight of the current parameter change on other parameters. On this basis, the conduction process is extended to multi-order paths through matrix power operation, that is, the dynamic process of parameter change propagating to other parts of the system through one-hop, two-hop or even multi-hop relationships is simulated, and a multi-order comprehensive conduction matrix is calculated. After this matrix is normalized, each row represents the influence distribution of a certain parameter on the multi-order propagation path, so as to obtain the conduction influence coefficient describing the internal parameter coupling conduction relationship of the system. The standardized fusion data and the conduction influence coefficient are input into the statistical anomaly detection model for mean deviation calculation and threshold judgment. Based on the statistical characteristics of each parameter in the time series, by calculating the deviation between the current observation value and the historical window mean and combining with the standard deviation for normalization, a mean deviation index is obtained. Since the system already has the conduction influence coefficient, the model introduces weight correction for different parameters when calculating the deviation, so that the parameter deviation values that have a greater impact on the system state obtain higher anomaly sensitivity. Individualized thresholds are set for each parameter, and whether there is an anomaly in the statistical sense at this moment is judged according to whether the deviation value exceeds the threshold. By weighted summarizing the deviation amounts of all parameters, a statistical anomaly score reflecting the current global deviation degree of the system is output. The pattern anomaly detection module uses the time series feature extraction and pattern similarity calculation method to construct the standardized fusion data into a multi-dimensional time window vector, and uses the dynamic time warping algorithm to match it with multiple normal operating condition patterns in history. The dynamic time warping algorithm realizes the optimal matching between sequences of different lengths and rhythms through non-linear time axis alignment, so as to measure the similarity to the greatest extent without destroying the time structure.The matching result is calculated as a similarity score, and 1 minus this score is the degree of pattern deviation. Combining the conduction influence coefficient, the contribution weight of key variables in the pattern to the degree of abnormality is strengthened, and an index of pattern abnormality reflecting the degree of deviation of the operating trajectory from the historical law is obtained. The standardized fusion data and the conduction influence coefficient are input into the prediction anomaly detection model for time series prediction, and a time series prediction model is established to compare the error between the actual value and the predicted value, so as to identify potential abnormal trends. A prediction framework based on the long short-term memory neural network (LSTM) is constructed. The input includes the feature vector after weighting the standardized fusion data and the conduction influence coefficient. The LSTM structure mines the long-range dependencies in time and captures the long-term trends of the system operating state. Through the training process, the network fits the evolution law of each parameter and predicts the values at several future time points. In the detection stage, the model outputs the predicted value in real time and compares it with the actual observed value, calculates the error ratio. If the error exceeds the preset tolerance, it is regarded as a prediction anomaly. The degree of this error forms a prediction anomaly degree measure after scaling and normalization. The statistical anomaly score, the pattern anomaly degree, and the prediction anomaly degree are input into the anomaly comprehensive scoring module. This module performs weighted summation processing on the three sub-indexes according to the set weighting rules to obtain the anomaly comprehensive score. The numerical range of this score is divided into four levels, corresponding to the normal state of the system (score < 0.3), the slight anomaly state (0.3 ≤ score < 0.6), the moderate anomaly state (0.6 ≤ score < 0.8), and the severe anomaly state (score ≥ 0.8). According to this score range, the working condition at the current moment is generated as the final anomaly state classification result.

[0032] In a specific embodiment, the process of inputting the standardized fusion data into the parameter correlation analysis model in the multivariate analysis model to calculate the correlation coefficient and construct the relationship network can specifically include the following steps: The standardized fusion data is grouped to obtain multiple parameter matrices, and the mean vector and standard deviation vector in the time dimension are calculated for each parameter matrix to obtain the parameter statistical characteristics; Based on the parameter statistical characteristics, the Pearson algorithm is executed on the standardized fusion data to calculate the correlation coefficient between each parameter, and the initial correlation relationship between the parameters is generated according to the correlation coefficient; The initial correlation relationship is screened and the strongly correlated parameter groups are analyzed to obtain the strongly correlated network between the parameters; The correlation coefficients in the strongly correlated network are converted into Z values through Fisher Z transformation, and the conditional independence probability between the parameters is calculated to obtain the direct correlation network between the parameters; The correlation strength and directionality are calculated for the correlation coefficients in the direct correlation network to obtain the directed influence network, and the directed influence network is converted into a weighted directed graph structure to obtain the parameter direct influence relationship network.

[0033] Specifically, the standardized fusion data is logically grouped according to parameter attributes or functions, and the data is divided into several parameter sub - matrices. Each matrix contains several statistically relevant parameter variables and their evolution trajectories in the time dimension. In each parameter sub - matrix, the mean vector and standard deviation vector of each parameter are calculated along the time dimension to reflect its central tendency and fluctuation characteristics during the operation cycle. Based on the above - mentioned parameter statistical characteristics, a quantitative analysis of the correlation relationship between parameters is carried out. All standardized parameter vectors are input into the Pearson correlation algorithm, and the linear correlation coefficient between parameter pairs is calculated by dividing the covariance by the product of the standard deviations. The Pearson algorithm is a classic method for measuring the degree of linear coupling between variables, and its output result is a continuous value between - 1 and 1, where 1 represents a perfect positive correlation, - 1 represents a perfect negative correlation, and 0 represents no linear correlation. After calculating the correlation coefficient matrix by pairwise combination of all parameters, a parameter correlation map is constructed. Based on this matrix, an initial correlation relationship network between parameters is generated. Each edge in this network corresponds to a parameter pair, and its edge weight is the correlation coefficient value between the two parameters. Since the correlation coefficient is symmetric, this initial network is an undirected weighted graph, reflecting the overall coupling structure of the industrial system in the multi - dimensional parameter space. Threshold screening is performed on the initial correlation relationship network by setting the absolute value threshold of the correlation coefficient, and only the significantly strongly correlated edges are retained. The graph partitioning algorithm is applied to classify the strongly correlated parameter clusters into several strongly correlated sub - networks. The parameter nodes in these sub - networks form core parameter clusters with strong coupling, fast information propagation, and easy interference conduction. After completing the construction of the strongly correlated network, in order to identify the direct action paths between parameters, a normal transformation and statistical significance analysis of the correlation coefficient are carried out. The Fisher Z - transformation method is introduced to map the correlation coefficient r value to the Z - value space. The formula is Z = 0.5×ln((1 + r) / (1−r)). This transformation process can transform the distribution of the correlation coefficient into an approximately normal distribution, making the correlation analysis applicable to subsequent hypothesis testing. Based on the Fisher Z - value, by setting the confidence level and performing a conditional independence test, the system can identify parameter pairs that still retain significant correlation under the condition of controlling other parameters. If two parameters still have a significant Z - value after controlling all other variables, it is considered that there is a direct correlation between this parameter pair, thus constructing a direct correlation network between parameters. In order to improve the analysis depth and control guidance ability of the parameter relationship network, directionality and intensity information are assigned to the edges in the direct correlation network, that is, a directed influence network is constructed. The directionality judgment is based on time - series causal analysis methods such as Granger causality test, information flow direction estimation, etc. The direction of the influence path is determined by analyzing the lag response relationship between two variables; and the intensity of the edge is measured according to the absolute value of the Fisher Z - value or an index based on response sensitivity, that is, considering the amplitude response of a unit change in one parameter to the change in the downstream parameter.Through this step, each edge not only has a direction but also has a clear weight, forming a directed and weighted influence network structure. Convert this directed influence network into a standard weighted directed graph data structure, that is, construct a direct causal relationship network of parameters. In this graph, all nodes represent parameters themselves, all edges represent their direct causal influence paths, the edge direction represents the acting direction, and the edge weight reflects the influence intensity.

[0034] In a specific embodiment, the process of performing the step of inputting the direct causal relationship network of parameters into the parameter change propagation model to calculate the propagation coefficient and obtaining the conduction influence coefficient of different parameter changes on the system state may specifically include the following steps: Perform a directed weighted graph conversion on the direct causal relationship network of parameters to obtain a system parameter topology graph; Calculate the out-degree and in-degree values for each parameter node in the system parameter topology graph, and determine the parameter importance ranking of the parameters in the system according to the out-degree and in-degree values; Based on the parameter importance ranking, perform a Markov chain modeling on the system parameter topology graph to obtain a first-order conduction matrix between parameters; Perform matrix power operation on the first-order conduction matrix to obtain a multi-order comprehensive conduction matrix, and perform normalization processing on the row vectors of the multi-order comprehensive conduction matrix to obtain a parameter influence distribution vector group; Perform sensitivity analysis on the parameter influence distribution vector group to obtain the conduction influence coefficient of different parameter changes on the system state.

[0035] Specifically, the direct influence relationship network of parameters is converted into a directed weighted graph. The nodes of this graph represent each monitoring or control parameter in the industrial system, while the edges represent the direct action paths of this parameter on other parameters. The directionality of the edges indicates the causal direction of the influence, and the weights of the edges reflect the strength of the influence. This structure is the system parameter topology graph, whose mathematical essence is a weighted directed graph. Its adjacency matrix is asymmetric and contains real weight values. Calculate the out-degree and in-degree values for each parameter node in the system parameter topology graph. The out-degree refers to the number and intensity of the channels through which this parameter sends out influences, while the in-degree represents the degree of convergence of this parameter being influenced by other parameters. The higher the out-degree, the more dominant this parameter is at the source position in the information or perturbation propagation path; the higher the in-degree, the more responsive or subordinate variable it is in the system, undertaking a receiving role. After calculating and combining the out-degree and in-degree values according to the weights, an importance index for each parameter node is formed. By sorting the importance of all nodes, the key variables that have the greatest impact on the evolution of the system state are identified. Based on the parameter importance ranking, a Markov chain model is built for the system parameter topology graph. The adjacency matrix of the system parameter topology graph is probabilistically processed, that is, the weights of all edges in each row are normalized so that the sum of the weights of all outgoing connection edges of each node is 1, and it is transformed into a state transition probability matrix to form a first-order conduction matrix between parameters. Each item of this matrix represents the probability of transitioning from parameter i to parameter j, mapping the possibility or intensity of the direct influence of the change in parameter i on parameter j. Through this step, the parameter influence path is modeled as a first-order Markov process, that is, the propagation behavior between parameters only depends on the current state and does not depend on the historical state, satisfying the memoryless assumption and having high applicability in the deduction of complex system states. To simulate the multi-step propagation process of parameter changes in the system, that is, the combined influence of parameter perturbations on remote parameters after passing through multiple intermediate parameters, the above first-order conduction matrix is extended to a higher order. This extension process is achieved by performing power operations on the conduction matrix, that is, successively calculating the square, cube,... up to the nth power of the conduction matrix to obtain the second-order, third-order... up to the nth-order conduction matrices. These matrices reflect the indirect propagation effects of parameter changes on two-hop, three-hop or even multi-hop paths. By weighted superposition of all power matrices, a multi-order comprehensive conduction matrix is constructed, which covers the total conduction ability between various parameters in the system through different paths. In actual operation, a propagation depth upper limit n is set to control the computational complexity and ensure the convergence of the conduction weights. The comprehensive conduction matrix is normalized so that the total conduction output of each parameter node is unified to 1, facilitating horizontal comparison of the propagation capabilities between different parameters. This processing is achieved by performing L1 norm normalization on each row vector of the comprehensive conduction matrix, and it is transformed into a parameter influence distribution vector, that is, each vector represents the influence distribution of the current parameter on all other parameters under the action of multi-order propagation.Perform sensitivity analysis on the parameter influence distribution vector group to clarify the ultimate conduction influence degree of each parameter change on the system state. Take each parameter influence distribution vector as input, and establish a partial derivative model or response function model between it and the system target state or state function. By simulating the perturbation degree caused by a small change in this parameter to the overall system state, obtain the conduction influence coefficient.

[0036] In a specific embodiment, the process of executing step S400 may specifically include the following steps: Perform level recognition on the abnormal state classification result to obtain an abnormal level judgment result; Dynamically adjust the data interaction frequency between the edge node and the central PLC controller according to the abnormal level judgment result to obtain a data interaction frequency configuration table; Dynamically adjust the sensor sampling frequency based on the abnormal level judgment result to obtain a sensor sampling frequency configuration table; Input the data interaction frequency configuration table and the sensor sampling frequency configuration table into the hierarchical control unit of the edge-center cooperation framework, and perform hierarchical analysis on the system state through the hierarchical control unit to obtain a control strategy priority table; Execute basic control layer coordination, optimized control layer parameter calculation, and policy control layer target adjustment according to the control strategy priority table and the current system state data to obtain a preliminary control parameter adjustment plan; Perform control margin calculation and safety verification on the preliminary control parameter adjustment plan to obtain a real-time control decision instruction.

[0037] Specifically, the classification results of abnormal states are subject to level recognition. Based on the weighted outputs of the three sub-models of statistical anomaly, pattern anomaly, and prediction anomaly, an integrated anomaly score with clear numerical boundaries is calculated, which serves as the direct basis for discriminating abnormal states. According to the pre-set score intervals, the abnormal states are divided into four levels: "normal", "slight anomaly", "moderate anomaly", and "severe anomaly". Among them, a score less than 0.3 is regarded as the normal state, a score between 0.3 and 0.6 is a slight anomaly, between 0.6 and 0.8 is a moderate anomaly, and a score greater than or equal to 0.8 is classified as the severe anomaly state, forming a clear abnormal level judgment result. According to the abnormal level judgment result, the control system initiates the dynamic scheduling strategy adjustment process for the communication mechanism between the edge nodes and the central PLC controller. When the system operates in the normal state, the edge nodes will adopt a low-frequency communication strategy to reduce communication resource consumption and system energy consumption; when entering the slight anomaly state, the communication frequency of the system will be increased to enhance the response speed to potential disturbances; if the system enters the moderate anomaly state, the communication frequency will be further increased, and the data content will be expanded from summary information to raw data segments; in the severe anomaly state, the system starts a real-time push mechanism to continuously send the complete raw data to the central controller and trigger the warning module. This hierarchical communication strategy will form a data interaction frequency configuration table in the system, the content of which clearly defines the reporting time intervals and data types of various edge nodes to the center under different abnormal levels. At the same time, the sampling frequency of the sensors is dynamically adjusted to match the change in the communication frequency to ensure the timeliness and accuracy of the collected data. Under normal working conditions, the sensors maintain the preset standard sampling frequency; when the system enters a slight anomaly, the sampling frequency of key parameters is increased by about 50%, and when a moderate anomaly is detected, the system doubles the sampling of all relevant parameters to enhance the data granularity; if the severe anomaly level is reached, the key sensors will continuously sample at the limit frequency to capture the characteristics of sudden changes in the system state to the greatest extent. All frequency adjustment results will be written into the sensor sampling frequency configuration table, which, together with the data interaction frequency configuration table, constitutes the basis for the response mechanism adjustment of the system in the abnormal state. The data interaction frequency configuration table and the sensor sampling frequency configuration table are input into the hierarchical control unit of the edge-center cooperation framework. This unit consists of a basic control layer, an optimization control layer, and a policy control layer, and has the ability to hierarchically perceive and regulate the system operation state. After receiving the configuration table data, the hierarchical control unit combines the current system operation data, parameter change trends, and historical working condition information to evaluate the current state of the system, and generates a control strategy priority table through rule reasoning or learning algorithms. This table sorts the priority levels of various control actions according to factors such as execution urgency, the influence weight on system stability, and control resource occupancy, clearly indicating which control actions should be immediately executed and which should participate in control feedback when resources are abundant.Based on the control strategy priority table, the control system combines the currently real-time collected system state data to initiate the collaborative working mechanism of the basic control layer, the optimization control layer, and the strategy control layer. Among them, the basic control layer is mainly responsible for coordinating the synchronous execution among edge nodes. For example, it coordinates the start-stop logic and unifies the action time among multiple devices. The optimization control layer performs refined parameter calculations based on real-time data and control models, such as adjusting and calculating key variables such as motor frequency, voltage, current, and pressure to form local control quantities. The strategy control layer focuses on the overall system goals, such as production rhythm, energy consumption efficiency, and safety boundaries, and indirectly affects control behaviors by adjusting control strategy parameters (such as scheduling algorithms and objective function weights). After the three-layer control collaboration is completed, a preliminary control parameter adjustment plan is obtained. Perform control margin calculation and safety verification on the preliminary control parameter adjustment plan. The control margin refers to the response ability and safety buffer space reserved by the current system when performing a certain control adjustment, and is measured by indicators such as the ratio of the control variable adjustment range to the maximum allowable adjustment range, the difference between the system dynamic response change rate and the critical value, and the gap between the equipment operating state and the rated boundary. Perform margin calculation on each control variable in the adjustment plan. If all variables are within the acceptable range, proceed to the next step; otherwise, the system will trigger the adjustment rollback mechanism or enter the exception handling process. After completing the margin verification, enter the safety verification. In this stage, predict the system state after the execution of the control plan by running a redundant model, a simulation engine, or a safety rule engine to determine whether there are risk factors such as out-of-bounds, conflicts, or equipment overload. If the safety verification passes, the control system generates real-time control decision instructions, which are distributed to each execution unit through the control network in the form of an instruction set or a command structure, and trigger the execution of actual control behaviors, such as adjusting the output frequency of the frequency converter, adjusting the valve opening, and modifying the process temperature set point, to achieve the global dynamic control response of the industrial system.

[0038] In a specific embodiment, the process of executing step S500 may specifically include the following steps: Collect and perform polynomial fitting modeling on the operating data of the motor at different frequencies and different loads to obtain the motor efficiency model and the optimal frequency-load comparison table; Perform fluctuation detection on the current motor load data to obtain the classification result of the load fluctuation state; Divide the current working conditions into multiple working conditions according to the classification result of the load fluctuation state, and configure the corresponding control strategy type for each working condition to obtain the working condition-strategy matching table; Match the real-time control decision instruction with the working condition-strategy matching table to obtain a control plan adapted to the working condition; Fuse the control scheme adapted to the working conditions with the optimal frequency-load comparison table, calculate the specific frequency adjustment step and adjustment strategy for different working conditions, and output the optimal variable frequency control parameters.

[0039] Specifically, the operating data of the motor under different frequencies and different loads are collected and modeled by polynomial fitting. During the experimental or production operation stage, the real-time operating data of the motor at multiple frequency points and multiple load rate levels are collected, including key parameters such as input power, output power, current, speed, temperature rise, etc., and all measuring point information is recorded in real-time synchronization through an electric energy metering instrument and a PLC data acquisition module. Under each combination condition of frequency and load rate, the motor efficiency η = output power / input power is calculated, and a polynomial fitting model with frequency f and load rate L as independent variables and efficiency η as the dependent variable is constructed. To ensure the fitting accuracy and generalization ability of the model, a fourth-order polynomial structure is selected to construct a bivariate regression model of η = f(f, L), and parameter estimation is performed for each term. The model coefficients are fitted by the least squares method to ensure that the goodness of fit R² of the model within the entire operating condition range is not less than 0.95. After the modeling is completed, the efficiency values at all combination points are calculated using the model, and the optimal frequency at each load rate is screened out to form a three-dimensional mapping table of "frequency - load - efficiency". Then, the point with the maximum efficiency is extracted to form an "optimal frequency - load comparison table", which indicates the most energy-efficient operating frequency to be adopted under each type of load rate. After the efficiency model is constructed, it enters the real-time evaluation stage of the current motor load state. By continuously sampling the current or power value output by the load sensor in real-time, and using the exponentially weighted moving average algorithm to smooth the data, combined with the calculation of the standard deviation within the sliding window, the fluctuation amplitude and trend of the load are dynamically monitored. The specific fluctuation threshold is set according to the system characteristics. The discrimination result is assigned a working condition level label, which is bound to the time stamp to form the classification result of the load fluctuation state at the current moment, providing the working condition characteristics of the current motor operating state. According to the above classification results, the operating state is divided into four working condition types, namely stable working condition, small-amplitude fluctuation working condition, medium-amplitude fluctuation working condition, and large-amplitude fluctuation working condition, and an optimal control strategy logic is designed for each working condition. For example, in the stable working condition, look-up table control is preferred, and the target frequency value is directly read from the frequency - load comparison table to achieve minimum energy consumption operation; in the small-amplitude fluctuation working condition, a PI control strategy is adopted, and the frequency is slowly and smoothly adjusted dynamically through the proportional term and the integral term to prevent the impact caused by frequent switching; in the medium-amplitude fluctuation working condition, a predictive control method is introduced, and a short-term load prediction model such as an LSTM structure is used to estimate the load trend in the next 10 seconds and make frequency adjustment preparations in advance; for the large-amplitude fluctuation working condition, a segmented adjustment strategy is adopted, decomposing the entire fluctuation process into multiple short-term stable segments, and adjusting the frequency according to the sub-working conditions within each segment to maintain the real-time performance of the motor response and control stability. These strategies are organized into a working condition - strategy matching table through system design to select the most appropriate control response mechanism under different load dynamic characteristics.After the system receives the real-time control decision instruction output by the upstream module, it automatically consults the working condition-strategy matching table according to the current load fluctuation status label, filters out the control strategy matching the current working condition, and encapsulates it into a control scheme adapted to the working condition. This scheme includes multiple parameters such as the type of frequency control model, the maximum adjustment amplitude, the adjustment period, and the frequency adjustment direction. The control scheme adapted to the working condition is fused with the optimal frequency-load comparison table, that is, the theoretical optimal frequency is determined by looking up the table according to the current actual load rate, and then the control strategy logic of the working condition adaptation scheme is applied to the dynamic adjustment of the target frequency. Taking PI control as an example, the system uses the optimal frequency as the set value and the current frequency as the feedback value, calculates the correction value of the frequency error according to the set proportional coefficient Kp and integral coefficient Ki, and the control output is the frequency adjustment step size for the current cycle; for the predictive control method, the system calls the historical load sequence to predict the future short-term load trend, and actively adjusts the frequency according to the predicted change trend to make the motor adapt to the load change in advance; if it is a segmented control strategy, the system divides the fluctuation curve into multiple time periods, searches for the optimal frequency corresponding to the load in each segment and adjusts it slowly to prevent frequency fluctuations caused by frequent changes. Considering the frequency adjustment strategy, the current operating frequency of the motor, the change direction of the load trend, the adjustment step size limit, and the control cycle duration, an optimal variable frequency control parameter is output, which includes elements such as the current target frequency value, the frequency adjustment amplitude, the upper limit of the allowable variable frequency step size, the adjustment period, and the prediction correction term. All parameters are transmitted to the frequency converter through the control bus to achieve real-time variable frequency control.

[0040] In this embodiment, the process of controlling an industrial motor according to the optimal variable-frequency control parameters includes: performing command filtering on the optimal variable-frequency control parameters, generating a smooth frequency reference trajectory and its derivative through a second-order low-pass filter, and obtaining a filtered frequency command signal and a rate signal; constructing a prescribed performance function based on the filtered frequency command signal, where the prescribed performance function does not depend on the initial state of the motor system, defining the convergence boundary of the motor tracking error through an exponential decay structure, and obtaining a performance constraint function; converting the performance constraint function into an unconstrained error variable, applying a backstepping design method to the unconstrained error variable to construct a virtual controller, and obtaining a stepwise smooth virtual control signal; designing an adaptive law for the unknown parameters in the control system, online estimating the system parameters through an error feedback and gradient descent method, and obtaining parameter estimation values; designing a two-parameter switching control mechanism based on the parameter estimation values, automatically switching the control gain parameters according to the system state and the estimation error, generating an adaptive control signal through dynamic gain adjustment, and obtaining a switching control law; combining the switching control law with the virtual control signal, using Lyapunov stability analysis to ensure the fixed-time stability of the system, and obtaining a final control input; applying the final control input to the inverter power distribution module, dynamically distributing the output power according to the actual requirements of the motor drive, and obtaining an optimized motor drive waveform; performing closed-loop monitoring on the optimized motor drive waveform, calculating the compliance between the system tracking error and the prescribed performance in real time, and adjusting the control parameters to ensure that the motor tracking accuracy meets the requirements of industrial production, and obtaining a high-precision variable-frequency control effect.

[0041] Please refer to Figure 2 , Figure 2 , which is a schematic block diagram of the real-time data processing and analysis system 200 of the industrial PLC controller provided by the embodiment of the present application. As Figure 2 shown, the real-time data processing and analysis system 200 of the industrial PLC controller includes: An acquisition module 210, configured to use the OPC UA communication protocol to transmit the temperature, pressure, current, vibration, and acoustic parameters collected by multiple sensors to the industrial PLC controller in real time, and obtain multi-source heterogeneous raw data; A fusion module 220, configured to perform data cleaning, normalization processing, and weighted clustering fusion on the multi-source heterogeneous raw data, and obtain normalized fusion data; A calculation module 230, configured to input the normalized fusion data into a multi-variable analysis model for correlation calculation and anomaly identification, and obtain an anomaly state classification result; A dynamic adjustment module 240, configured to dynamically adjust the data interaction frequency and sampling rate between the edge node and the central PLC controller according to the anomaly state classification result, and generate a real-time control decision instruction; An output module 250, configured to match the real-time control decision instruction with the current motor load fluctuation state, and output the optimal variable-frequency control parameters.

[0042] Through the collaborative cooperation of the above-mentioned various components, by constructing a three-layer industrial control system architecture and a five-layer parallel structure of the PLC controller, combined with the secure transmission mechanism of the OPC UA communication protocol, the data processing capacity and communication efficiency of the industrial control system have been significantly improved, enabling the system to process multi-source heterogeneous data safely and efficiently. The data cleaning, standardization processing, and weighted clustering fusion methods adopted effectively solve the problem of industrial multi-source heterogeneous data fusion, improve the data quality, and provide a high-quality data basis for subsequent analysis. By integrating parameter correlation analysis, change propagation analysis, and three complementary anomaly detection methods through a multi-variable analysis model, various abnormal states in the system can be comprehensively identified, improving the accuracy and reliability of anomaly detection. Based on the "edge-center" collaboration framework dynamically adjusted according to the abnormal state classification results, a control mode of "processing nearby and making hierarchical decisions" is realized, significantly reducing the system response delay and improving the control accuracy and reliability. By adopting an adaptive variable frequency control algorithm, different control strategies can be adopted according to the load fluctuation characteristics, achieving minimum energy consumption while ensuring control accuracy, and effectively balancing the relationship between system security and energy efficiency. Through the hierarchical security situation assessment and collaborative optimization control method, through the comprehensive assessment of the execution unit layer, controller layer, and network communication layer, combined with the parameter deviation risk degree assignment, the dynamic balance between security and energy efficiency is achieved, enabling the system to intelligently adjust the control strategy according to the real-time security situation.

[0043] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0044] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0045] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A real-time data processing and analysis method for an industrial PLC controller, characterized in that, Including: Using the OPC UA communication protocol to transmit the temperature, pressure, current, vibration, and acoustic parameters collected by multiple sensors to the industrial PLC controller in real time, obtaining multi-source heterogeneous raw data; Performing data cleaning, standardization processing, and weighted clustering fusion on the multi-source heterogeneous raw data to obtain standardized fusion data; Inputting the standardized fusion data into a multivariate analysis model for correlation calculation and anomaly identification to obtain an anomaly status classification result; Dynamically adjusting the data interaction frequency and sampling rate between the edge node and the central PLC controller according to the anomaly status classification result to generate real-time control decision instructions; Matching the real-time control decision instructions with the current motor load fluctuation state and outputting optimal variable frequency control parameters.

2. The real-time data processing and analysis method of the industrial PLC controller according to claim 1, wherein The step of using the OPC UA communication protocol to transmit the temperature, pressure, current, vibration, and acoustic parameters collected by multiple sensors to the industrial PLC controller in real time, obtaining multi-source heterogeneous raw data, includes: Dividing the industrial control system into a three-layer architecture, dividing the industrial control system into an execution layer, a control layer, and a data acquisition layer, and implementing a five-layer parallel structure for the PLC controller in the control layer to obtain a multi-level industrial control system architecture; Deploying temperature sensors, pressure sensors, current sensors, vibration sensors, and acoustic sensors in the data acquisition layer, and respectively collecting corresponding parameters according to a preset sampling frequency to obtain multi-source heterogeneous raw data; Performing 256-bit encryption and certificate authentication processing on the multi-source heterogeneous raw data to obtain a securely encrypted transmission data packet, and transmitting the securely encrypted transmission data packet to the data interface layer of the PLC controller through the OPC UA communication protocol to obtain an encrypted data stream to be processed; Performing decryption verification and data unpacking processing on the encrypted data stream to be processed to obtain raw parameter data, and storing the raw parameter data in the double buffer memory area of the PLC controller to obtain multi-source heterogeneous raw data.

3. The real-time data processing and analysis method of the industrial PLC controller according to claim 1, wherein The step of performing data cleaning, standardization processing, and weighted clustering fusion on the multi-source heterogeneous raw data to obtain standardized fusion data, includes: Performing data cleaning on the multi-source heterogeneous raw data, removing outliers in the temperature, pressure, current, vibration, and acoustic parameters to obtain preliminarily cleaned data; Performing low-pass filtering on the temperature data in the preliminarily cleaned data, and simultaneously performing wavelet packet transform denoising on the vibration data in the preliminarily cleaned data to obtain denoised parameter data; Performing standardization transformation on the denoised parameter data to obtain standardized data with a unified dimension, and assigning weight coefficients to the temperature, pressure, current, vibration, and acoustic parameters in the standardized data with a unified dimension to obtain a weighted parameter set; Calculating the weighted Euclidean distance matrix between data points based on the weighted parameter set, and performing weighted clustering on the weighted Euclidean distance matrix. By minimizing the within-class weighted distance and maximizing the between-class weighted distance, standardized fusion data is obtained.

4. The real-time data processing and analysis method of the industrial PLC controller according to claim 1, characterized in that, The step of inputting the standardized fusion data into a multivariate analysis model for correlation calculation and anomaly identification to obtain an anomaly status classification result, includes: Input the standardized fusion data into the parameter correlation analysis model in the multivariate analysis model to calculate the correlation coefficient and construct the relationship network, and obtain the parameter direct influence relationship network; Input the parameter direct influence relationship network into the parameter change propagation model to calculate the propagation coefficient, and obtain the conduction influence coefficient of different parameter changes on the system state; Input the standardized fusion data and the conduction influence coefficient into the statistical anomaly detection model to calculate the mean deviation and threshold judgment, and obtain the statistical anomaly score; Input the standardized fusion data and the conduction influence coefficient into the pattern anomaly detection model to extract time series features and perform pattern matching, and obtain the pattern anomaly degree; Input the standardized fusion data and the conduction influence coefficient into the prediction anomaly detection model to perform time series prediction, and obtain the prediction anomaly degree; Perform weighted summation on the statistical anomaly score, the pattern anomaly degree and the prediction anomaly degree to obtain the comprehensive anomaly score, and generate the anomaly state classification result according to the score interval.

5. The real-time data processing and analysis method of the industrial PLC controller according to claim 4, characterized in that The step of inputting the standardized fusion data into the parameter correlation analysis model in the multivariate analysis model to calculate the correlation coefficient and construct the relationship network, and obtain the parameter direct influence relationship network includes: Perform grouping processing on the standardized fusion data to obtain multiple parameter matrices, and calculate the mean vector and standard deviation vector of the time dimension for each parameter matrix to obtain the parameter statistical features; Based on the parameter statistical features, perform the Pearson algorithm on the standardized fusion data to calculate the correlation coefficient between each parameter, and generate the initial correlation relationship between the parameters according to the correlation coefficient; Perform screening and strong correlation parameter group analysis on the initial correlation relationship to obtain the strong correlation network between the parameters; Convert the correlation coefficient in the strong correlation network into a Z value through the Fisher Z transformation, and calculate the conditional independent probability between the parameters to obtain the direct correlation network between the parameters; Calculate the correlation strength and directionality of the correlation coefficient in the direct correlation network to obtain the directed influence network, and convert the directed influence network into a weighted directed graph structure to obtain the parameter direct influence relationship network.

6. The real-time data processing and analysis method of the industrial PLC controller according to claim 4, characterized in that The step of inputting the parameter direct influence relationship network into the parameter change propagation model to calculate the propagation coefficient, and obtain the conduction influence coefficient of different parameter changes on the system state includes: Perform a directed weighted graph conversion on the parameter direct influence relationship network to obtain the system parameter topology graph; Calculate the out-degree and in-degree values for each parameter node in the system parameter topology graph, and determine the parameter importance ranking of the parameters in the system according to the out-degree and in-degree values; Based on the parameter importance ranking, perform Markov chain modeling on the system parameter topology graph to obtain the first-order conduction matrix between the parameters; Perform matrix power operation on the first-order conduction matrix to obtain the multi-order comprehensive conduction matrix, and perform normalization processing on the row vectors of the multi-order comprehensive conduction matrix to obtain the parameter influence distribution vector group; Perform sensitivity analysis on the parameter influence distribution vector group to obtain the conduction influence coefficient of different parameter changes on the system state.

7. The real-time data processing and analysis method of the industrial PLC controller according to claim 1, characterized in that Dynamically adjusting the data interaction frequency and sampling rate between the edge node and the central PLC controller according to the abnormal state classification result, and generating a real-time control decision instruction, including: Performing level recognition on the abnormal state classification result to obtain an abnormal level judgment result; Dynamically adjusting the data interaction frequency between the edge node and the central PLC controller according to the abnormal level judgment result to obtain a data interaction frequency configuration table; Dynamically adjusting the sensor sampling frequency based on the abnormal level judgment result to obtain a sensor sampling frequency configuration table; Inputting the data interaction frequency configuration table and the sensor sampling frequency configuration table into the hierarchical control unit of the edge-center cooperation framework, and performing hierarchical analysis on the system state through the hierarchical control unit to obtain a control strategy priority table; Executing basic control layer coordination, optimizing control layer parameter calculation, and strategy control layer target adjustment according to the control strategy priority table and the current system state data to obtain a preliminary control parameter adjustment plan; Performing control margin calculation and safety verification on the preliminary control parameter adjustment plan to obtain a real-time control decision instruction.

8. The real-time data processing and analysis method of the industrial PLC controller according to claim 1, characterized in that Matching the real-time control decision instruction with the current motor load fluctuation state, and outputting optimal variable frequency control parameters, including: Collecting and performing polynomial fitting modeling on the operation data of the motor at different frequencies and different loads to obtain a motor efficiency model and an optimal frequency-load comparison table; Performing fluctuation detection on the current motor load data to obtain a load fluctuation state classification result; Dividing the current working condition into multiple working conditions according to the load fluctuation state classification result, and configuring corresponding control strategy types for each working condition to obtain a working condition-strategy matching table; Matching the real-time control decision instruction with the working condition-strategy matching table to obtain a working condition-adapted control plan; Performing data fusion on the working condition-adapted control plan and the optimal frequency-load comparison table, calculating specific frequency adjustment steps and adjustment strategies for different working conditions, and outputting optimal variable frequency control parameters.

9. A real-time data processing and analysis system for an industrial PLC controller, characterized in that, For implementing the real-time data processing and analysis method of the industrial PLC controller according to any one of claims 1-8, the real-time data processing and analysis system of the industrial PLC controller includes: An acquisition module, configured to use the OPC UA communication protocol to transmit the temperature, pressure, current, vibration, and acoustic parameters collected by multiple sensors to the industrial PLC controller in real time to obtain multi-source heterogeneous raw data; A fusion module, configured to perform data cleaning, standardization processing, and weighted clustering fusion on the multi-source heterogeneous raw data to obtain standardized fusion data; A calculation module, configured to input the standardized fusion data into a multi-variable analysis model for correlation calculation and anomaly identification to obtain an abnormal state classification result; A dynamic adjustment module, configured to dynamically adjust the data interaction frequency and sampling rate between the edge node and the central PLC controller according to the abnormal state classification result, and generate a real-time control decision instruction; An output module, configured to match the real-time control decision instruction with the current motor load fluctuation state, and output optimal variable frequency control parameters.

Images