A dynamic optimization method for industrial automation system based on 5G private network

Abstract

The present invention discloses a dynamic optimization method for an industrial automation system based on a 5G private network. By carefully constructing a system architecture, including modular design to enhance flexibility, data-driven to achieve accurate decision-making, and building a security protection system, the system is ensured to operate stably and reliably. In the data acquisition and preprocessing stages, multi-source data fusion, real-time processing, and intelligent cleaning are performed to provide high-quality data for the system. Control strategy optimization is achieved based on artificial intelligence, such as adaptive adjustment of neural network models, reinforcement learning to seek optimal decisions, and model predictive control to improve performance. At the same time, communication resources are managed and optimized, including dynamic allocation of 5G private network slice resources, intelligent scheduling of wireless resources, and network congestion control. Finally, with the help of real-time monitoring of key performance indicators, performance evaluation through big data analysis, and iterative updates of optimization strategies, the system performance is continuously improved, promoting the development of industrial automation towards intelligence and efficiency.

Description

Technical Field

[0001] The present invention relates to the field related to industrial automation systems based on 5G private networks, and specifically to a dynamic optimization method for industrial automation systems based on 5G private networks. Background Art

[0002] With the rapid development of Industry 4.0 and intelligent manufacturing, industrial automation systems play a vital role in improving production efficiency, reducing costs, and improving product quality. Traditional industrial automation systems have gradually exposed many limitations in the face of increasingly complex production environments and diversified production needs.

[0003] In terms of communication, traditional industrial networks have problems such as limited bandwidth, high transmission delay, and small network coverage, which makes it difficult to meet the real-time, high-speed, and stable data transmission requirements of a large number of devices in industrial automation systems. For some control applications with extremely high real-time requirements, such as industrial robot collaborative operations and high-precision motion control, the delay and jitter of traditional networks may cause control instructions to fail to be executed in a timely and accurate manner, affecting production efficiency and product quality.

[0004] Traditional industrial automation systems are mostly based on preset rules and fixed procedures, lacking the ability to adapt to changes in the production environment and intelligent decision-making capabilities. Faced with uncertain factors in the production process, the system cannot automatically adjust the control strategy in time, resulting in low production efficiency and serious waste of resources.

[0005] In addition, traditional industrial automation systems are relatively weak in data processing and analysis capabilities, making it difficult to extract valuable information from massive production data and unable to provide effective data support for corporate production management and decision-making. At the same time, the security and reliability of the system also face severe challenges, and security incidents such as cyber attacks and data leaks may cause significant losses to corporate production. Summary of the invention

[0006] The purpose of the present invention is to provide a dynamic optimization method for an industrial automation system based on a 5G private network to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a dynamic optimization method for an industrial automation system based on a 5G private network, comprising the following steps: Step 1: System architecture construction and optimization: Modular design enhances system flexibility: The industrial automation system is divided into multiple independent and functionally clear modules, including data acquisition module, control module, execution module and communication module. The modules interact with each other through standardized interfaces, so that the system can easily add, remove or replace modules to adapt to production needs and technology updates, shorten iteration cycles and reduce maintenance costs. Data-driven accurate decision-making: In a 5G private network environment, production data such as equipment operating status, product quality parameters, and environmental data are collected in real time through sensors and actuators. The high-speed, low-latency characteristics of 5G are used to transmit the data to the data processing center in real time. Artificial intelligence algorithms are used for in-depth analysis, and control strategies are dynamically adjusted based on the analysis results to ensure optimization of the production process. Security protection system ensures stable operation: A security protection system is built from the hardware, software and network levels. Encryption chips and secure storage devices are used in hardware to ensure data security; vulnerability scanning and repair are performed on the software, and access rights are set; 5G private network slicing technology is implemented on the network to isolate different business data. At the same time, a fault warning and emergency response mechanism is established. When an abnormality is detected, the plan is quickly launched to ensure system security and stability; Step 2: Intelligent data collection and preprocessing: Multi-source data fusion improves information integrity: With the high bandwidth characteristics of 5G private networks, multi-source data from different sensors and devices, including equipment operation data, production environment data, and product quality inspection data, are integrated, and data fusion algorithms are used to eliminate redundant and conflicting information, improve data accuracy and integrity, and provide a more reliable basis for subsequent control decisions; Real-time data processing ensures timeliness: Taking advantage of the low latency of 5G private networks, real-time pre-processing of data is performed at the data collection end, including data cleaning, feature extraction, and data compression, to reduce the burden of data transmission and processing, improve system response speed, ensure the timeliness of data processing, and meet the real-time requirements of industrial automation systems; Intelligent data cleaning improves data quality: Use machine learning and deep learning algorithms to automatically identify and process noise, outliers, and missing values ​​in the data, use clustering algorithms to detect outliers, use regression algorithms or neural networks to fill in missing values, and perform data standardization and normalization to improve data quality and provide accurate data support for the system; Step 3: Control strategy optimization based on artificial intelligence: Adaptive adjustment of neural network models: Build a neural network model suitable for industrial automation systems, use the large amount of production data transmitted in real time by 5G private networks to train and optimize the model, so that the model can adaptively adjust the control strategy, continuously optimize the weights and thresholds through online learning, and adjust the control parameters in real time according to changes in the production process, thereby improving system adaptability and control accuracy; Reinforcement learning achieves optimal decision-making: The control problem in the industrial automation system is modeled as a reinforcement learning problem, and the state space, action space and reward function are defined. The intelligent agent selects actions based on the system state and obtains reward feedback by interacting with the environment. The 5G private network is used to provide real-time feedback information, and the reinforcement learning algorithm continuously optimizes the strategy, so that the system can make the best decision in a complex environment and improve production efficiency and stability. Model predictive control optimization performance: Combining system mathematical models and real-time data, using the fast data transmission capability of 5G private networks, model predictive control of industrial automation systems is implemented to predict the future behavior of the system, optimize control inputs based on the prediction results, improve control performance through rolling optimization and feedback correction, reduce fluctuations and errors, and improve product quality consistency; Step 4: Communication resource management and optimization: Dynamic allocation of 5G private network slice resources: According to the needs of different services in the industrial automation system, such as real-time control services, data collection services, and video surveillance services, the network slicing technology of 5G private networks is used to allocate independent network resources to each service. By dynamically adjusting slice resources, low latency and high reliability of key services are ensured, thereby improving the overall performance of the system. Intelligent scheduling of wireless resources: Adopt intelligent scheduling algorithms to manage wireless resources in 5G private networks, dynamically allocate resources based on real-time device needs and channel quality, predict device data transmission needs, allocate resources in advance, and use channel status information to optimize resource allocation, improve spectrum efficiency and system capacity, and meet the needs of large-scale device access and data transmission in industrial automation systems; Network congestion control and optimization: Real-time monitoring of the network congestion status of 5G private networks, timely detection of congestion using traffic monitoring technology and congestion detection algorithms, and measures such as traffic control, routing adjustment, and resource allocation optimization to alleviate congestion, ensure the stability and reliability of data transmission, and ensure the normal operation of key services of industrial automation systems; Step 5: System performance evaluation and continuous improvement: Real-time monitoring of key performance indicators: Through the 5G private network, key performance indicators of industrial automation systems, such as production efficiency, product quality, equipment failure rate, and energy consumption, are collected in real time. Data analysis technology is used to monitor and visualize the indicators in real time, so that operators can understand the system operation status in a timely manner and provide a basis for optimizing decisions. Performance evaluation based on big data analysis: collect and store a large amount of historical production data, use big data analysis technology to mine potential patterns and rules in the data, establish a performance evaluation model, evaluate system performance and predict future trends, find system bottlenecks and optimization space based on the evaluation results, and provide direction for continuous improvement; Iterative update of optimization strategies: Based on the system performance evaluation results, the optimization strategies are generated using artificial intelligence algorithms, and the optimization strategies are sent to various devices in the industrial automation system in real time through the 5G private network. The effectiveness of strategy implementation is continuously monitored, and the optimization strategies are iteratively updated based on feedback information to achieve continuous improvement of system performance.

[0008] Furthermore, in the modular design, the functions of each module are as follows: Data acquisition module: responsible for collecting various data in the industrial production process, including equipment operation parameters, environmental parameters, product quality parameters, etc.; this module integrates various types of sensors, such as temperature sensors, pressure sensors, displacement sensors, and image sensors, and transmits the collected data to the data processing center in real time through the 5G private network to ensure the accuracy and timeliness of the data; Control module: Receives control instructions from the data processing center and controls the execution module according to the instructions; this module uses advanced control algorithms, such as PID control algorithm, model predictive control algorithm, and fuzzy control algorithm, to generate accurate control signals based on the needs of the production process and real-time data, to achieve precise control of the equipment; Execution module: According to the instructions of the control module, it performs corresponding operations, such as driving motors, controlling valves, and adjusting the working status of equipment. The execution module has high reliability and fast response capabilities to ensure that control instructions can be accurately executed to achieve automated control of the production process. Communication module: responsible for data communication of the entire system, including data transmission between modules and data interaction with external systems; this module is based on 5G private network technology to achieve high-speed and stable data transmission, and supports multiple communication protocols, such as TCP / IP protocol and OPC UA protocol, to ensure the compatibility and reliability of data communication.

[0009] Furthermore, in the data-driven decision-making, artificial intelligence algorithms include but are not limited to the following: Machine learning algorithms: such as decision tree algorithms, support vector machine algorithms, and neural network algorithms, which build data models by learning from historical data to predict equipment failures, optimize production processes, and improve product quality; Deep learning algorithms: such as convolutional neural network algorithms, recurrent neural network algorithms, and generative adversarial network algorithms, which are used to process complex image, voice, and text data to achieve intelligent monitoring, fault diagnosis, and quality inspection of the production process; Reinforcement learning algorithms: such as Q-learning algorithm, deep Q-network algorithm, and policy gradient algorithm, learn the optimal control strategy through the interaction between the intelligent agent and the environment, realize intelligent control of production equipment, optimal allocation of resources, and maximization of production efficiency.

[0010] Furthermore, in the security protection system, the encryption chip at the hardware level adopts the Advanced Encryption Standard algorithm, the secure storage device supports hardware encryption and access control, vulnerability scanning and repair at the software level are performed regularly and automatically, access permission settings are managed hierarchically according to user roles and permissions, and the 5G private network slicing technology at the network level is implemented using software-defined networking and network function virtualization technology.

[0011] Furthermore, in the multi-source data fusion, the data fusion algorithm adopts weighted averaging method, Kalman filtering method and Bayesian estimation method, performs weighted processing according to the reliability and importance of the data, eliminates redundancy and conflict between data, and improves the accuracy and completeness of the data.

[0012] Furthermore, in the real-time data processing, data cleaning adopts outlier detection and missing value processing technology based on statistical methods and machine learning algorithms, feature extraction adopts principal component analysis algorithm, linear discriminant analysis algorithm, wavelet transform algorithm, data compression adopts lossless compression algorithm and lossy compression algorithm, and the appropriate compression method is selected according to data characteristics and transmission requirements.

[0013] Furthermore, in the adaptive adjustment of the neural network model, the neural network model adopts a feedforward neural network, a feedback neural network, a convolutional neural network, and a recurrent neural network, and is trained and optimized through a back propagation algorithm, a stochastic gradient descent algorithm, and an adaptive learning rate adjustment algorithm to improve the convergence speed and generalization ability of the model.

[0014] Furthermore, in the reinforcement learning to achieve optimal decision-making, the state space includes equipment operating status, production environment status, and task execution progress; the action space includes equipment control operations, resource allocation decisions, and production plan adjustments; the reward function is designed according to production goals and constraints, such as maximizing production efficiency, optimizing product quality, and minimizing costs, and strategy optimization is performed through Q learning algorithm, deep Q network algorithm, and proximal strategy optimization algorithm.

[0015] Furthermore, in the dynamic allocation of 5G private network slice resources, dynamic bandwidth allocation algorithm, priority scheduling algorithm and resource reservation algorithm are used to allocate resources according to the service priority, bandwidth requirement and latency requirement. The creation, configuration, monitoring and management of slices are realized through the network slice management system to ensure the service quality of key services.

[0016] Furthermore, in the performance evaluation and continuous improvement of the system, big data analysis technology adopts data mining algorithms, machine learning algorithms, and statistical analysis methods; the performance evaluation model adopts hierarchical analysis method, fuzzy comprehensive evaluation method, and neural network evaluation method; the iterative update of the optimization strategy adopts genetic algorithm, particle swarm optimization algorithm, and simulated annealing algorithm to automatically adjust the optimization strategy according to changes in system performance to achieve continuous optimization of the system.

[0017] Compared with the prior art, the present invention has the following beneficial effects: Through the high-speed and low-latency characteristics of 5G private networks, fast and stable transmission of data between devices in industrial automation systems can be achieved, ensuring that control instructions can be issued in real time and accurately, significantly improving the system response speed, and effectively reducing production efficiency losses caused by communication delays. For example, in the collaborative operation scenarios of industrial robots, more precise motion synchronization can be achieved, improving operating efficiency and quality.

[0018] The high reliability of 5G private networks ensures the stability of data transmission, reduces the risk of data loss and transmission errors, and combines a variety of measures in the security protection system to effectively resist network attacks and fault interference, ensure the stable operation of industrial automation systems, reduce downtime caused by system failures, and improve production continuity.

[0019] Data-driven decision-making and control strategy optimization based on artificial intelligence algorithms enable the system to adjust control strategies in real time according to changes in the production environment and equipment status, and adaptively respond to various complex working conditions, such as changes in raw material properties, equipment aging, etc., to ensure that the production process is always kept in the best state and improve product quality consistency.

[0020] Dynamic allocation of 5G private network slice resources and intelligent scheduling of wireless resources can reasonably allocate network resources according to different business needs, improve spectrum efficiency and system capacity, ensure low latency and high reliability of key services, avoid waste of resources, and meet the communication needs of large-scale equipment access and diversified services in industrial automation systems.

[0021] The system can monitor the equipment operating status and production environment in real time, and realize intelligent energy-saving operation of the equipment by optimizing the control strategy, such as reasonably adjusting the equipment power according to the production task, optimizing the equipment start and stop time, reducing energy consumption, and improving energy utilization efficiency, which is in line with the green manufacturing concept.

[0022] The adaptive adjustment of the neural network model, reinforcement learning to achieve optimal decision-making and model predictive control optimization performance enable the system to accurately control the production process, optimize the production process, reduce fluctuations and errors in the production process, and improve product quality stability. For example, in the precision machining process, higher machining accuracy and product qualification rate can be achieved.

[0023] The real-time monitoring and intelligent early warning system uses artificial intelligence algorithms to analyze equipment operation data, realize early prediction and intelligent diagnosis of equipment failures, formulate maintenance plans in advance, reduce production interruptions caused by sudden equipment failures, improve equipment maintenance efficiency, reduce maintenance costs, and ensure production continuity and stability.

[0024] Intelligent data collection and preprocessing provide high-quality data support for the system, and combined with artificial intelligence technology, it can achieve deep insights and intelligent decision-making in the production process, promote the development of industrial automation systems in the direction of intelligence, realize advanced manufacturing models such as personalized customized production and intelligent production scheduling, and enhance the market competitiveness of enterprises.

[0025] The dynamic optimization method of industrial automation systems based on 5G private networks and artificial intelligence brings new application scenarios and business models to the industrial field, such as smart factories, smart logistics, and remote equipment control. It promotes industrial upgrading and innovative development, and drives the transformation of manufacturing to high-end, intelligent, and green. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] See also Figure 1 , a specific implementation method of a dynamic optimization method for an industrial automation system based on a 5G private network provided by the present invention: 1. System architecture construction and optimization Modular design Based on the functional requirements and business processes of the industrial automation system, the system is divided into data acquisition module, control module, execution module and communication module. The division principle follows high cohesion and low coupling to ensure that each module is independent and complete, and the interface between modules is clear and concise, which is easy to maintain and expand.

[0029] For example, in an automated production line of an automobile manufacturing plant, the data acquisition module is responsible for collecting the operating parameters of production equipment, environmental parameters, and product quality inspection data. The control module analyzes and processes the collected data according to the preset production process and control strategy, and generates corresponding control instructions. The execution module is responsible for converting control instructions into actual operations, such as driving the motor to run, controlling the welding equipment to work, and adjusting the transmission speed of the production line. The communication module ensures smooth data transmission between modules and with external systems.

[0030] Standardized communication protocols, such as OPCUA protocol, are used to realize data interaction between modules. OPCUA protocol has the advantages of cross-platform, security, reliability, and easy expansion, and can meet the requirements of industrial automation systems for real-time, stability, and interoperability.

[0031] The data transmission between modules adopts a combination of event-driven and timed polling. When the data acquisition module detects a change in the device status or collects new data, it immediately triggers an event and sends the data to the control module. At the same time, the control module sends a polling request to the data acquisition module at a certain time interval to obtain the latest device operation data to ensure the timeliness and accuracy of control decisions.

[0032] In order to improve the reliability and fault tolerance of the system, the communication module adopts a redundant design. For example, multiple base stations and communication links are deployed in the 5G private network. When the main communication link fails, the system can automatically switch to the backup link to ensure uninterrupted data transmission.

[0033] Data-driven decision making According to the characteristics and needs of the industrial production process, select appropriate sensors and data acquisition equipment. In the chemical production process, high-precision thermocouple sensors, pressure sensors and liquid level sensors are used to monitor the temperature, pressure, liquid level and other parameters in the reactor in real time. For the measurement of material flow, electromagnetic flowmeters or mass flowmeters are selected to ensure the accuracy and reliability of data.

[0034] The deployment locations of sensors and data acquisition equipment are carefully designed to obtain the most representative data. For example, torque sensors and angle sensors are installed at the joints of industrial robots to monitor the robot's motion state and force in real time, providing a basis for robot control and fault diagnosis. Environmental sensors are deployed in key areas of the production workshop to monitor environmental parameters such as temperature and humidity, light intensity, and harmful gas concentration to ensure that the production environment meets the requirements.

[0035] The high speed and low latency characteristics of the 5G private network are used to transmit the collected data to the data processing center in real time. The data is encrypted during transmission and the Advanced Encryption Standard algorithm is used to encrypt the data to ensure data security.

[0036] In the data processing center, the received data is first decoded and decrypted, and then cleaned and preprocessed. During the data cleaning process, outlier detection and missing value processing techniques based on statistical methods and machine learning algorithms are used to remove noise and outliers in the data and fill in missing data points. For example, for the temperature data collected by the sensor, if there are outliers that are significantly deviated from the normal range, the mean filtering algorithm is used to replace them with the average value of the surrounding normal data. For missing pressure data, other sensor data related to the sensor are used to predict and fill in by establishing a regression model.

[0037] The cleaned and preprocessed data is analyzed and processed using corresponding artificial intelligence algorithms according to different business needs and application scenarios. For equipment failure prediction, deep learning algorithms are used to model and analyze equipment operation data, explore potential patterns and features in the data, and predict the type and time of possible equipment failures. In terms of production process optimization, machine learning algorithms are used to model and optimize the relationship between key parameters in the production process and product quality indicators, find the optimal combination of production parameters, and improve production efficiency and product quality.

[0038] Safety protection system Choose devices with hardware encryption functions, such as encryption chips and secure storage devices. Encryption chips use national encryption algorithms to encrypt and store key data in the device to prevent data from being illegally stolen and tampered with. Secure storage devices support hardware encryption and access control, and only authorized users or devices can access stored data.

[0039] Trusted computing technology is used in key equipment of industrial automation systems. Trusted measurement and verification of the equipment startup process, system software integrity, and operating status are performed through hardware trusted root chips to ensure that the equipment runs in a safe and trusted environment. For example, when an industrial server is started, the trusted root chip performs integrity checks on the server's BIOS, operating system kernel, driver, etc. If any tampering or abnormality is found, the server is immediately blocked from starting and a security alarm is issued.

[0040] Regularly scan and repair system software vulnerabilities, use professional vulnerability scanning tools to comprehensively scan the operating system, application software, database, etc. in the industrial automation system, and promptly discover and repair known vulnerabilities. At the same time, establish a software vulnerability management mechanism to track the discovery, repair and verification process of vulnerabilities to ensure the security of system software.

[0041] Implement strict access rights management, and divide different levels of permissions according to user roles and responsibilities, such as administrators, operators, and ordinary users. Use a role-based access control model to define corresponding operating permissions for each role to ensure that users can only operate within their authorized scope. For example, administrators have full management rights over the system, including equipment configuration, user management, system maintenance, etc. Operators can only perform daily operations and monitoring of equipment and cannot modify system configurations; ordinary users can only view system operating status and production data and cannot perform any operational control.

[0042] Conduct code security audits on control programs and application software in industrial automation systems, use static code analysis tools and dynamic testing tools to check code security, discover and fix security vulnerabilities in the code, and improve software security and stability.

[0043] Using the network slicing technology of 5G private networks, independent virtual network slices are created for industrial automation systems, and different services are isolated in different slices to ensure data security and isolation between services. Network slicing technology is based on software-defined networking and network function virtualization technology, and provides customized network services for each slice through flexible allocation and management of network resources.

[0044] Deploy network security equipment such as firewalls, intrusion detection systems, and intrusion prevention systems to monitor and filter network traffic of 5G private networks in real time. Firewalls block unauthorized network access and malicious attack traffic based on preset security policies; IDS monitors abnormal behaviors and attack features in network traffic in real time, such as port scanning, DDoS attacks, malware propagation, etc., and immediately issues an alarm once an abnormality is found; IPS, based on IDS, can automatically take measures to block attack traffic, such as disconnecting connections, discarding malicious data packets, etc., to protect industrial automation systems from network attacks.

[0045] Establish a network security situation awareness system, and realize comprehensive perception and visual display of network security situation by real-time collection, analysis and correlation of operation data of network equipment, security equipment, application systems, etc. in 5G private network. The network security situation awareness system can timely discover security threats and risks in the network, predict the development trend of security incidents, and provide support for network security decision-making. For example, through the analysis of network traffic, equipment logs, security incidents and other data, it is found that the network traffic in a certain area suddenly increases abnormally, and there may be a risk of DDoS attack. The system will issue an early warning in time and provide corresponding response strategies, such as starting traffic cleaning services and adjusting network configuration.

[0046] Establish equipment failure prediction models, use machine learning and deep learning algorithms to analyze equipment operation data, and predict possible equipment failures. For example, use the support vector machine algorithm to model the vibration data of the motor, and predict the bearing failure of the motor based on characteristics such as vibration frequency and amplitude. When it is predicted that the equipment may fail, the system will issue early warning information to notify maintenance personnel to carry out maintenance, avoiding production interruptions caused by sudden equipment failures.

[0047] Design a comprehensive emergency response plan, and formulate corresponding emergency handling procedures and measures for different types of security incidents and failure situations. The emergency response plan includes incident reporting process, emergency command system, resource allocation plan, recovery measures, etc. For example, when a network attack occurs, immediately activate the emergency response plan, cut off the connection between the attacked network equipment and the external network, start the backup network equipment, and restore the normal operation of the system; at the same time, organize security experts to investigate and analyze the attack incident, and take measures to strengthen network security protection to prevent similar incidents from happening again.

[0048] Regularly conduct drills and evaluations on emergency response plans to ensure their effectiveness and operability. By simulating different types of security incidents and failure scenarios, we can test the coordination ability, response speed, and processing effect of the emergency response team, and optimize and improve the emergency plan. For example, we organize two network security emergency drills every year to simulate scenarios such as DDoS attacks and malware infections. After the drills, we summarize and evaluate the drills, and revise and improve the emergency plan based on the problems found.

[0049] 2. Intelligent Data Collection and Preprocessing Multi-source data fusion According to the characteristics and fusion requirements of the data in the industrial automation system, select the appropriate data fusion algorithm. In terms of sensor data fusion, the weighted average method is used for the fusion of environmental parameters such as temperature, pressure, and humidity. Different weights are assigned to each sensor based on factors such as sensor accuracy and reliability, and the collected multiple sensor data are processed by weighted average to obtain more accurate environmental parameter estimates. For example, for temperature measurement in a workshop, there are three temperature sensors in different locations, one of which has a high-precision sensor with a weight of 0.5, and the other two ordinary sensors with weights of 0.25 respectively. The comprehensive temperature value of the workshop is calculated by weighted average.

[0050] For the fusion of multimodal data, Kalman filtering or Bayesian estimation is used. In the industrial robot vision guidance system, the image data collected by the robot vision system is fused with the joint position sensor data. The Kalman filtering method is used to fuse the target position information in the image data and the robot joint position information in the sensor data to improve the robot's positioning accuracy and grasping accuracy of the target object. The Bayesian estimation rule takes into account the uncertainty and prior knowledge of the data in the data fusion process. By establishing a Bayesian network model, probabilistic reasoning and fusion of multi-source data are performed, which is suitable for data fusion tasks in complex industrial environments.

[0051] The data acquisition module collects raw data from multiple sensors and data sources, and pre-processes the data at the data fusion node, including data format conversion and time synchronization. This ensures that the data from different data sources are consistent in time and format for subsequent fusion operations.

[0052] For data fusion using the weighted average method, the weight coefficient is dynamically adjusted according to the real-time performance indicators of the sensor. When a sensor fails or the measurement error is large, its weight coefficient is reduced and the weight of other normal sensor data is increased to ensure the accuracy of the fusion result. In the application of Kalman filtering and Bayesian estimation, according to the dynamic characteristics of the system and the characteristics of observation noise, the filtering parameters and Bayesian network structure are optimized to improve the performance and efficiency of data fusion.

[0053] The quality of the fused data is evaluated, and the deviation between the fused data and the true value is measured using indicators such as root mean square error and mean absolute error. If the evaluation result is not ideal, the parameters of the data fusion algorithm are readjusted or other more appropriate algorithms are selected to continuously optimize the data fusion effect.

[0054] Real-time data processing At the data collection end, statistical methods are used to perform preliminary cleaning on the real-time collected data. Calculate statistical indicators such as the mean and standard deviation of the data, set a reasonable threshold range, and treat data that exceeds the threshold range as outliers. For continuously collected temperature data, if the deviation of a data point from the mean exceeds three times the standard deviation, it will be marked as an outlier.

[0055] Machine learning algorithms are used to further detect outliers in the data. The Isolation Forest Algorithm isolates data points into tree nodes by constructing a random forest, and determines the degree of abnormality based on the depth of the data point in the tree; the Local Anomaly Factor Algorithm identifies outliers by calculating the density difference between a data point and its neighboring data points. The detected outliers are corrected or discarded according to the specific situation. If the outlier is caused by a short-term interference of the sensor, the average value of the adjacent data points can be used for correction; if the outlier is caused by a serious problem such as equipment failure, it is discarded and an alarm is issued to notify maintenance personnel to handle it.

[0056] Use principal component analysis algorithms, linear discriminant analysis algorithms, etc. to extract features. In industrial equipment fault diagnosis, multiple vibration signal features are collected during equipment operation, and these high-dimensional features are reduced to a few principal component features through the PCA algorithm. These principal component features can retain the main information of the original data, while reducing the data dimension and the amount of subsequent calculations. The LDA algorithm takes into account the category information of the data while reducing the dimension, and is suitable for feature extraction in classification tasks. For example, the equipment operating status is divided into normal and faulty categories, and the LDA algorithm is used to extract features that can effectively distinguish the two types of status.

[0057] The data is compressed using lossless compression algorithms and lossy compression algorithms. For key parameter data with high data accuracy requirements, a lossless compression algorithm is used to ensure that no information is lost during the compression and decompression process. For data with high real-time requirements and relatively low data accuracy requirements, a lossy compression algorithm can be used to minimize the amount of data and improve data transmission efficiency while ensuring the basic characteristics of the data. In wavelet transform compression, the appropriate wavelet basis function and decomposition layer number are selected according to the frequency characteristics of the data, and the data is decomposed and reconstructed at multiple scales to remove high-frequency noise and redundant information to achieve data compression.

[0058] Use hardware acceleration technologies such as field programmable gate arrays and application-specific integrated circuits to accelerate computationally intensive tasks in the data processing process. FPGAs have reconfigurability and parallel computing capabilities, and can flexibly configure hardware circuits according to specific data processing requirements to achieve efficient data processing. ASICs are customized for specific data processing algorithms and have higher processing efficiency and performance. For example, in industrial image processing, algorithms such as image filtering and edge detection are solidified in FPGA chips to achieve high-speed image processing and improve the real-time performance of the system.

[0059] Optimize the code implementation of data processing algorithms, use efficient programming languages ​​and parallel computing technology, make full use of the computing resources of multi-core processors and GPUs, and improve data processing speed. For data tasks that can be processed in parallel, use multi-threading or multi-process technology for parallel computing to shorten data processing time. For example, in the data cleaning process, use multi-threading technology to clean and process data collected by multiple sensors at the same time to improve the system throughput.

[0060] Intelligent data cleaning For missing values ​​in the data, machine learning methods such as K nearest neighbor algorithm and regression algorithm are used to fill them. In the industrial production process, for missing values ​​in the equipment operation time series data, the time point where the missing value is located is taken as the center, and the data of several time points before and after are selected as features. The KNN algorithm is used to find the K neighbors most similar to the data point, and the missing values ​​are predicted and filled according to the values ​​of the neighbor data. If the data has a certain linear relationship, the linear regression algorithm can also be used to establish a regression model between the data and other related variables, and the missing values ​​can be predicted through the model.

[0061] When using machine learning algorithms to process missing values, select appropriate algorithms and parameters based on the distribution characteristics and missing patterns of the data. In the case of uneven data distribution, preprocess the data to improve the accuracy of the algorithm. At the same time, select the optimal algorithm parameters through methods such as cross-validation, such as the K value in the KNN algorithm, to ensure the accuracy and reliability of missing value filling.

[0062] Construct deep learning models such as convolutional neural networks and recurrent neural networks for anomaly detection. In industrial equipment monitoring, the collected equipment vibration signals or current signals are used as input data, and feature extraction and classification are performed through the convolution layer, pooling layer, and fully connected layer of the CNN model to determine whether the data is abnormal. The RNN model is suitable for processing data with time series characteristics, such as time series data of equipment operating status. By learning from historical data, the trend of future data is predicted. When the actual data deviates greatly from the predicted data, it is determined to be an abnormal situation.

[0063] During the training process of the deep learning model, a large amount of normal data and a small amount of abnormal data are used for training, so that the model can learn the patterns and characteristics of normal data, so as to better identify abnormal data. At the same time, data enhancement technology is used to expand the training data and improve the generalization ability of the model. After the model training is completed, the accuracy and recall rate of anomaly detection are balanced by adjusting the threshold to ensure that the real abnormal situation can be discovered in time while reducing the false alarm rate. For example, in the product quality inspection of industrial production lines, the deep learning model is used to detect the product appearance image. If the product appearance is found to have defects, it is marked as an abnormal product in time for subsequent processing.

[0064] Adaptive adjustment of neural network models Select the appropriate neural network structure according to the control requirements and input-output characteristics of the industrial automation system. For control scenarios with time-series data, use recurrent neural networks or long short-term memory networks. RNN can handle the time-series dependencies in sequence data, while LSTM has better memory capabilities when processing long sequence data and can effectively avoid the gradient vanishing problem. In image processing-related control tasks, use convolutional neural networks to automatically extract image features using their convolutional layers and pooling layers to improve control accuracy.

[0065] Collect a large amount of industrial production process data, including equipment operating status, control input, environmental parameters and corresponding control target data, to train the neural network model. Divide the data into training set, validation set and test set, and use optimization algorithms such as stochastic gradient descent algorithm, Adagrad algorithm, Adadelta algorithm to update the weights and biases of the model. During the training process, set appropriate learning rate, number of iterations, batch size and other parameters, and prevent overfitting or underfitting by monitoring the loss function and accuracy on the validation set. For example, when training a neural network model for an industrial temperature control system, collect temperature set values, ambient temperature, equipment operating power and actual temperature measurement values ​​under different working conditions as training data. After multiple iterations of training, the model can accurately predict the temperature change trend under different input conditions and output appropriate control signals.

[0066] The production data transmitted in real time by the 5G private network is used to learn and update the neural network model online. When new data is input into the model, the model automatically adjusts the weights and biases according to the difference between the new data and the existing knowledge. The use of incremental learning algorithms, such as the online sequential extreme learning machine algorithm, allows the model to be updated quickly when new data arrives without retraining the entire model, improving the adaptability and real-time performance of the model.

[0067] Establish a model performance evaluation index system, such as mean square error, mean absolute error, determination coefficient, etc., to monitor the performance of the model in the actual production process in real time. When the model performance degrades, trigger the model retraining or optimization mechanism. Automatically adjust the structure and parameters of the neural network model according to the changes in working conditions during the production process. For example, when the performance of equipment on an industrial production line degrades after running for a period of time, resulting in unstable temperature control, the system automatically detects that the model prediction error has increased. At this time, the model optimization program is started to increase the number of hidden layers of the neural network or adjust the learning rate so that the model can adapt to changes in equipment performance and re-achieve accurate temperature control.

[0068] Reinforcement learning for optimal decision making Model the control environment of the industrial automation system and determine the state space, action space, and reward function. The state space includes factors that can affect the control decision, such as the current operating status of the equipment, the progress of the production task, and environmental parameters. These state information are quantified and encoded as the input of the reinforcement learning agent. For example, in an automated warehousing system, the state space can be represented as a combination of information such as the robot's position coordinates, the storage location of the goods, the remaining space on the shelf, and the handling task queue.

[0069] The action space is defined as the control actions that the agent can take in each state, such as the movement direction of the industrial robot, the grasping action, the start and stop operation of the equipment, the adjustment of the control parameters, etc. According to the actual control requirements of the industrial automation system, the size and scope of the action space are reasonably determined to ensure that the agent can make effective control decisions within the limited action space.

[0070] Design a reward function to quantitatively evaluate the results of the agent's actions based on the control objectives and constraints. The design of the reward function is to guide the agent to take actions that can maximize the long-term cumulative rewards. In the industrial production process, the reward function can include factors such as improved production efficiency, improved product quality, reduced energy consumption, and penalties for violating constraints. For example, in a chemical production process, if the control actions taken by the agent keep the temperature and pressure in the reactor within the optimal range, improve product quality and reduce energy consumption, a positive reward will be given; conversely, if the temperature is out of control or the pressure is too high, a negative penalty will be given.

[0071] According to the characteristics and control requirements of the industrial automation system, select the appropriate reinforcement learning algorithm, such as Q learning algorithm, deep Q network algorithm, proximal policy optimization algorithm, etc. The Q learning algorithm is suitable for control problems with small and discrete state and action spaces. By building a Q table to store the value of state-action pairs, the agent selects the optimal action based on the Q value. The DQN algorithm combines deep learning with Q learning and uses neural networks to approximate the Q function. It is suitable for processing complex state spaces and continuous action spaces, and can better cope with high-dimensional data and nonlinear relationships in industrial automation systems. The PPO algorithm is a policy gradient-based algorithm with high sample efficiency and stability. It can quickly converge to a better strategy and is suitable for industrial control scenarios with high real-time requirements.

[0072] During the training process of the reinforcement learning algorithm, the agent interacts with the industrial automation system environment, selects actions based on the current state, obtains rewards and the next state from the environment after executing the action, and then updates the strategy based on the reward signal. The experience replay technology is used to store the experience of the agent in the process of interacting with the environment in the experience replay buffer. During training, samples are randomly selected for learning to break the correlation between data and improve learning efficiency and stability. At the same time, a suitable exploration rate is set to encourage the agent to explore more and try different actions in the early stage of training to find a better strategy; as the training progresses, the exploration rate is gradually reduced so that the agent can make more decisions based on the learned experience. For example, in the path planning task of the industrial robot, the agent initially randomly selects different moving paths with a high exploration rate. As the training progresses, it gradually moves according to the optimal path strategy that has been learned, and continuously optimizes the path planning strategy to complete the handling task with the shortest path and the least time.

[0073] Model Predictive Control Optimizing Performance According to the physical characteristics and operating rules of the industrial automation system, a dynamic mathematical model of the system is established. For linear time-invariant systems, state space models, transfer function models, etc. can be used to describe them; for nonlinear systems, nonlinear autoregressive sliding average models, neural network models, etc. are used for modeling. In the process of modeling, the system identification technology is used to estimate the unknown parameters in the model by collecting and analyzing the input and output data of the industrial automation system. For example, in the frequency control of an electric power system, a state space model is established according to the dynamic characteristics of the electric power system. By measuring the output power, frequency and other data of the generator, the least squares method and other parameter identification methods are used to determine the inertia constant, damping coefficient and other parameters in the model.

[0074] In order to improve the accuracy and adaptability of the model, the system dynamic model is updated and verified regularly. As the industrial production process progresses, factors such as equipment aging and process adjustments may cause changes in system characteristics. At this time, the model parameters are re-identified using newly collected data to ensure that the model can accurately reflect the current operating status of the system. At the same time, model verification technology is used to evaluate the effectiveness of the model. If the model performance deteriorates, it is adjusted or re-modeled in a timely manner.

[0075] Based on the established system dynamic model, the model predictive control algorithm is used to control the industrial automation system. In each control cycle, the MPC algorithm predicts the output behavior of the system in the future period based on the current system state and the prediction model, and solves a finite time domain optimization problem through the optimization algorithm to obtain the optimal control sequence in the future period. In the optimization process, the control objectives and constraints are considered to ensure the feasibility and effectiveness of the control strategy. For example, in the temperature control of an industrial heating furnace, the MPC algorithm predicts the temperature change in the furnace in the next few sampling moments based on the current furnace temperature, set temperature, and the dynamic relationship model between heating power and temperature change. The optimization algorithm is used to determine the heating power control sequence that makes the furnace temperature track the set temperature as quickly as possible and minimizes the control energy consumption.

[0076] The prediction time domain, control time domain, weight coefficient and other parameters in the MPC algorithm are optimized and adjusted to improve the control performance. The prediction time domain determines the prediction length of the algorithm for the future system behavior. A longer prediction time domain can provide more comprehensive information, but it will also increase the computational complexity; the control time domain determines the length of the control sequence actually applied in each optimization calculation. Through simulation analysis and actual debugging, the appropriate combination of prediction time domain and control time domain is found to balance the control performance and computing resource consumption of the system. The weight coefficient is used to weigh the relative importance of the control objectives. According to the specific requirements of the industrial production process, the tracking error weight, control energy weight and other coefficients are reasonably adjusted to achieve fine adjustment of the control performance. For example, in a production process with high requirements for product quality but not very sensitive to energy consumption, the tracking error weight can be appropriately increased and the control energy weight can be reduced to ensure the stability of product quality; conversely, in the case of high energy costs, the control energy weight can be appropriately increased to reduce energy consumption while ensuring basic production requirements.

[0077] 4. Communication Resource Management and Optimization Dynamic allocation of 5G private network slice resources Carry out detailed classification and demand analysis of different services in the industrial automation system, and determine the key performance indicator requirements of each type of service. For example, real-time control services have extremely high requirements for latency and reliability, data collection services have certain requirements for bandwidth, and video surveillance services need to ensure a certain bandwidth and image quality. According to the differences in business needs, use the network slicing technology of 5G private networks to create independent network slices for different services. For example, create an ultra-low latency, high-reliability network slice for real-time control of industrial robots; create a medium-bandwidth, high-reliability network slice for production equipment status monitoring data collection; create a large bandwidth and meet certain image quality requirements for workshop video surveillance.

[0078] In the process of network slicing, the physical layout and business distribution characteristics of the industrial automation system are considered, and slice resources are planned reasonably. For real-time control devices centrally deployed in the same area, they can be divided into the same network slice to reduce the complexity of network switching and resource management; for devices distributed in different areas but with similar business needs, they can also be divided into the same slice, and flexible allocation and isolation of resources can be achieved through network virtualization technology. At the same time, different levels of resource guarantees are allocated to different slices according to the importance and priority of the business to ensure the normal operation of key businesses.

[0079] Adopt a dynamic resource allocation strategy based on service priority and real-time demand. Monitor the traffic changes, latency requirements, reliability requirements, etc. of services in each network slice in real time, and dynamically adjust the resource allocation of the slice based on this information. When the traffic of the real-time control service suddenly increases or the latency requirements become more stringent, prioritize more bandwidth and computing resources for the slice to ensure that control instructions can be transmitted in a timely and accurate manner. For data collection services, if the collection frequency decreases or the amount of data decreases, appropriately recycle some resources and allocate them to other slices that need resources.

[0080] Develop and apply resource allocation algorithms, such as resource allocation algorithms based on utility functions and dynamic bandwidth allocation algorithms. The utility function calculates the utility value of each slice based on the KPI indicators and resource usage of the business. The resource allocation algorithm allocates resources based on the size of the utility value so that resource allocation can maximize the satisfaction of business needs. The DBA algorithm dynamically adjusts the bandwidth resources allocated to each slice based on the real-time bandwidth requirements of the business to improve bandwidth utilization. For example, in an industrial automation system, the utility value of the real-time control slice is calculated to be 0.8 through the utility function, the utility value of the data acquisition slice is 0.5, and the utility value of the video surveillance slice is 0.3. Based on the size of the utility value, more wireless resources are allocated to the real-time control slice first to ensure its high-performance operation.

[0081] Establish a slice resource adjustment trigger mechanism. When the system detects the following situations, the slice resource adjustment process is started: first, business demand changes significantly, such as the addition of high-priority services, sudden growth in business traffic, or adjustment of business priorities; second, network resource utilization is too high or too low, affecting overall system performance; third, network failure or interference causes slice resource damage. For example, when a new automated production line is put into use, bringing a large number of new real-time control services, the system automatically detects changes in business demand and triggers the resource adjustment mechanism.

[0082] In the process of adjusting slice resources, a combination of gradual adjustment and batch adjustment is adopted. For some minor changes in business requirements, a gradual adjustment method is adopted to gradually increase or decrease slice resources to avoid excessive impact on the system caused by resource adjustment. For major changes in business requirements or serious imbalance of network resources, a batch adjustment method is adopted to quickly reallocate resources to ensure that the system can adapt to new business requirements as soon as possible. At the same time, after the resources are adjusted, the system performance is monitored and evaluated in real time, and the resource allocation plan is further optimized based on the evaluation results to ensure the rationality and effectiveness of resource allocation.

[0083] Intelligent scheduling of wireless resources Use machine learning and time series analysis techniques to predict and model the communication needs of equipment in industrial automation systems. Collect historical communication data of equipment, including data transmission volume, transmission frequency, transmission time interval and other information, and establish a communication demand prediction model. Use time series prediction models such as ARIMA model and LSTM model to predict the future communication needs of equipment based on historical data. For example, for an industrial sensor, by analyzing its data transmission volume at different time periods every day in the past week, establish an ARIMA model to predict the communication needs of the sensor in various time periods in the next week, and provide a basis for wireless resource scheduling.

[0084] Consider the impact of various factors in the industrial production process on the communication needs of equipment, such as changes in production tasks, adjustments to the operating status of equipment, and changes in environmental factors. Use these factors as input variables of the model to improve the accuracy of communication demand prediction. For example, in a foundry, when the casting process switches from sand casting to metal casting, the operating parameters and data collection frequency of the production equipment will change, resulting in changes in the communication needs of the equipment. By incorporating the casting process type as an input variable into the communication demand prediction model, the communication needs of the equipment under different process conditions can be more accurately predicted.

[0085] Design intelligent wireless resource scheduling algorithms based on prediction results, such as dynamic resource allocation algorithms and proportional fair scheduling algorithms. The DRA algorithm dynamically allocates wireless resources to devices based on the real-time communication needs and channel quality of the devices, giving priority to devices with urgent needs and good channel quality. The PFS algorithm allocates resources based on the historical communication volume and current demand ratio of the device while ensuring a certain degree of fairness, so that resource allocation can meet the requirements of high-demand devices while taking into account the basic communication needs of other devices.

[0086] In the process of implementing the intelligent scheduling algorithm for wireless resources, the characteristics and resource constraints of 5G private networks are taken into consideration. The flexible frame structure and multi-antenna technology of 5G are fully utilized to optimize the resource allocation scheme. For example, in the time division duplex mode of 5G, the uplink and downlink time slot ratios are reasonably configured according to the time distribution characteristics of the device communication requirements to improve the spectrum utilization. For multi-antenna devices, spatial multiplexing technology is used to provide services to multiple devices at the same time and increase system capacity. At the same time, considering the limited wireless resources in 5G private networks, reasonable upper and lower limits of resource allocation are set to prevent over-allocation or under-allocation of resources.

[0087] A channel quality monitoring system is established to monitor the channel quality between each device and the base station in the 5G private network in real time. Using channel state information feedback technology, the device regularly sends CSI reports to the base station, including information such as channel fading, signal-to-noise ratio, and interference level. Based on the received CSI reports, the base station understands the channel quality status in real time and provides a basis for wireless resource scheduling. For example, in an industrial IoT environment, channel quality may change due to factors such as device movement and obstacle obstruction. The channel quality monitoring system can timely grasp these changes in order to adjust resource allocation.

[0088] Design a channel quality feedback mechanism to ensure that the base station can obtain accurate channel quality information in a timely manner. Optimize the cycle and content of CSI feedback to minimize feedback overhead while ensuring the timeliness and accuracy of feedback information. Dynamically adjust the feedback cycle according to changes in channel quality. When the channel quality changes dramatically, shorten the feedback cycle so that the base station can make resource scheduling decisions more quickly; when the channel quality is relatively stable, appropriately extend the feedback cycle to reduce system overhead. At the same time, use efficient coding and compression technology to reduce the transmission volume of CSI reports and improve feedback efficiency.

[0089] Network congestion control and optimization Determine key indicators for monitoring network congestion, such as network delay, packet loss rate, buffer occupancy, link utilization, etc. By deploying monitoring modules at key nodes of the 5G private network, data of these indicators can be collected in real time. For example, the packet transmission delay of each device connection is monitored at the base station. If the delay exceeds a certain threshold, there may be a risk of congestion; at the same time, the buffer occupancy is monitored. When the buffer occupancy is close to or exceeds the set safety threshold, it indicates that the network congestion is high.

[0090] According to the characteristics and requirements of the industrial automation system, select the appropriate congestion control algorithm. For industrial control services with extremely high real-time requirements, feedback-based congestion control algorithms are used, such as the TCP Vegas algorithm and the congestion control mechanism in the RTP / RTSP protocol. These algorithms monitor the network status in real time and adjust the sending rate in time to avoid data loss and increased delays caused by congestion. For data acquisition services with high reliability requirements, window-based congestion control algorithms, such as the TCP Reno algorithm, can be used to control data flow by dynamically adjusting the sending window size to ensure the reliability of data transmission.

[0091] When network congestion is detected, flow control measures are adopted to limit the sending rate of the data source. For real-time control services, priority is given to ensuring the minimum bandwidth requirements to ensure that control instructions can be transmitted in a timely manner, while appropriately limiting the traffic of other non-critical services. For example, in an industrial automation system, if network congestion causes an increase in the transmission delay of real-time control instructions, the system automatically reduces the sending rate of data acquisition services and video surveillance services, and gives priority to ensuring the bandwidth of real-time control services such as industrial robot control instructions.

[0092] Adjust network resource allocation according to congestion conditions. Dynamically adjust the resource allocation of 5G private network slices, allocate more resources to severely congested slices or services, and alleviate congestion. At the same time, optimize wireless resource scheduling, reasonably adjust the parameters such as device transmission power and channel allocation, and improve spectrum efficiency. For example, in congested areas, reduce the transmission power of some devices to reduce interference, and at the same time give priority to allocating channel resources to severely congested device connections to ensure the transmission of critical data.

[0093] Use cache management strategies to optimize the use of buffers. When the buffer occupancy rate is high, some low-priority or outdated data is selectively discarded based on the priority and timeliness of the data to make room for high-priority data. For example, in video surveillance services, if the buffer is full, some older video frames are discarded first to ensure the smoothness of real-time video transmission.

[0094] When network congestion is relieved, the sending rate of restricted services is gradually restored to avoid unnecessary impact on services. A progressive recovery strategy is adopted to slowly increase the sending rate of services according to the improvement of network status until it returns to normal levels. For example, after congestion is relieved, the sending rate of data collection services is increased by 10% at regular intervals until it returns to the rate before congestion, while continuously monitoring the network status to prevent congestion from occurring again.

[0095] Establish a congestion prevention mechanism to take measures in advance to avoid congestion by predicting changes in network traffic and resource requirements. Use machine learning algorithms to analyze historical network traffic data, predict future peak and trough periods of network traffic, and adjust resource allocation and network configuration in advance. For example, based on historical data, it is predicted that a certain production line will have a data transmission peak at 10:00 or 11:00 a.m. every day. The system will start to increase network resources in the corresponding area at around 9:30 a.m., such as increasing the transmission power of the base station and adjusting slice resource allocation, to prevent congestion.

[0096] 5. System performance evaluation and continuous improvement Real-time monitoring of key performance indicators Sensors and monitoring software are deployed at key nodes of the industrial automation system to collect key performance indicator data of the system in real time, such as equipment operation status data, production process data, and communication network data. The collected data is transmitted to the performance evaluation center in real time by utilizing the low latency and high reliability characteristics of the 5G private network.

[0097] In order to ensure the accuracy and integrity of data transmission, a data verification and retransmission mechanism is adopted. The collected data is verified and calculated at the data sending end, and the received data is verified and verified at the receiving end. If data errors or loss are found, a retransmission request is immediately sent to the sending end to ensure the reliability of performance indicator data.

[0098] Construct a performance monitoring system architecture, including data collection layer, data transmission layer, data processing layer and display layer. The data collection layer is responsible for collecting performance indicator data from various data sources; the data transmission layer uses the 5G private network to transmit data to the data processing layer; the data processing layer cleans, preprocesses and analyzes the received data, and calculates the real-time values ​​and statistical values ​​of key performance indicators; the display layer displays the processed performance indicator data to operators and managers in the form of intuitive charts, so as to facilitate real-time understanding of the system operation status.

[0099] The monitoring system has a real-time alarm function. When the key performance indicators exceed the preset threshold range, an alarm message will be issued immediately. Relevant personnel will be notified through SMS, email, sound and light alarms, etc. to ensure that the problem can be handled in time. For example, when the temperature of the equipment exceeds the set safety temperature limit, the system automatically sends a text message to notify the equipment maintenance personnel, and at the same time triggers an sound and light alarm in the monitoring center to remind the operator to pay attention to the equipment operation status.

[0100] Performance evaluation based on big data analysis Establish a big data storage platform to store various types of data such as historical performance indicator data, production data, equipment maintenance data, etc. accumulated by industrial automation systems over a long period of time. Use distributed file systems and distributed databases to store and manage data to ensure high availability, high scalability, and high-performance reading and writing of data.

[0101] Classify and label the stored data to facilitate subsequent data query and analysis. For example, classify the data according to data source, time period, data type and other dimensions to provide a good data foundation for big data analysis.

[0102] Use data mining algorithms and machine learning algorithms to analyze big data and establish a performance evaluation model. Use cluster analysis algorithms to cluster equipment operation status data, identify different equipment operation modes and states, and provide a basis for equipment fault diagnosis and performance evaluation. For example, cluster analysis is performed on the joint motion data of industrial robots, and their operation status is divided into different categories such as normal operation, minor faults, and serious faults. By real-time monitoring of the category to which the robot's operation data belongs, its performance status can be evaluated.

[0103] Regression analysis algorithms are used to establish a relationship model between production efficiency and equipment operating parameters and process parameters, predict production efficiency under different parameter settings, and provide guidance for production process optimization. For example, through regression analysis of historical production data, a mathematical model is established between product output and factors such as equipment processing speed, raw material supply, and ambient temperature. Production parameters are adjusted according to the model prediction results to improve production efficiency.

[0104] Use time series analysis algorithms to analyze and predict the trends of system performance indicator data to identify potential performance problems in advance. For example, perform time series analysis on network delay data to predict the changing trend of network delay in the future. If it is predicted that network delay will continue to increase, take network optimization measures in advance, such as increasing network bandwidth and optimizing routing configuration.

[0105] The performance evaluation results are presented to users in an intuitive and visual way, such as drawing performance indicator trend graphs, comparison graphs, distribution graphs, etc. Through the visual interface, users can clearly understand the changes in system performance at different times, different equipment, and different production links, and quickly find performance bottlenecks and abnormal situations. For example, draw a trend graph of product scrap rate in different production workshops in the past month, compare the production quality of each workshop, and find the workshop with a high scrap rate for key analysis and improvement.

[0106] Generate performance evaluation reports regularly, including statistical analysis results of key performance indicators of the system, performance trend forecasts, existing problems and recommended measures. Performance evaluation reports provide decision-making support for enterprise management, helping them understand the overall operation of industrial automation systems and formulate reasonable production plans and improvement strategies. For example, generate a detailed performance evaluation report every month to report to the company's senior management on changes in key indicators such as system production efficiency, equipment failure rate, and product quality, analyze the main factors affecting performance, and put forward targeted improvement suggestions, such as equipment upgrade and transformation plans, process optimization plans, and personnel training needs.

[0107] Optimize strategy iterative update Based on the performance evaluation results, artificial intelligence algorithms are used to generate optimization strategies. For example, genetic algorithms are used to optimize the control parameters of industrial automation systems, and the control parameters are encoded as chromosomes. By simulating genetic operations such as natural selection, crossover, and mutation, the optimal control parameter combination is found. In terms of production scheduling optimization, simulated annealing algorithms are used to simulate the physical properties of solid annealing processes, randomly search for the optimal production scheduling solution in the solution space, and improve production efficiency and resource utilization.

[0108] Establish a decision support system to provide managers with the selection and decision-making basis of optimization strategies. The decision support system comprehensively considers the performance evaluation results, the expected effects of the optimization strategy, the implementation cost, the risk and other factors, and sorts and evaluates different optimization strategies through multi-criteria decision analysis methods to help managers choose the optimization strategy that best suits the actual situation of the enterprise. For example, when deciding whether to upgrade the equipment of a production line, the decision support system comprehensively analyzes the production efficiency improvement potential after the equipment upgrade, the investment cost, the difficulty of equipment maintenance and other factors to provide decision-making suggestions for managers.

[0109] The selected optimization strategy is sent to each device and control node of the industrial automation system in real time through the 5G private network to ensure that the optimization strategy can take effect in a timely manner. In the process of strategy implementation, automated deployment tools and technologies are used, such as the controller in the software-defined network sending configuration commands and automated script execution, to achieve fast and accurate strategy deployment. For example, when it is necessary to adjust the resource allocation strategy of the network slice, the SDN controller automatically generates the corresponding configuration command according to the optimization strategy and sends it to the 5G base station and core network equipment to achieve dynamic adjustment of network resources.

[0110] After the optimization strategy is implemented, continuously monitor the changes in system performance to verify the effectiveness of the optimization strategy. If it is found that the system performance has not been significantly improved or new problems have arisen, adjust or reformulate the optimization strategy in a timely manner. For example, after implementing the equipment control parameter optimization strategy, continuously monitor the equipment's operating efficiency and product quality. If it is found that the equipment's operating efficiency has not improved but has decreased, the reason for analysis may be that the optimization parameters do not match the actual production conditions. At this time, it is necessary to readjust the optimization parameters and optimize again.

[0111] Record and summarize the implementation process and effects of the optimization strategy to form experience knowledge. Store the relevant information of each optimization strategy in the knowledge base to provide reference for future optimization decisions. For example, record a case where a production line improved production efficiency by optimizing the production process, including specific process improvement measures, production data comparison before and after implementation, etc. When other production lines encounter similar problems, they can draw on the experience of this case for optimization.

[0112] Use machine learning algorithms to learn and mine the empirical data in the knowledge base to discover the potential laws and patterns between optimization strategies and system performance. Through the analysis of a large amount of historical empirical data, the optimization strategy generation algorithm and decision support system are continuously improved to improve the accuracy and efficiency of optimization decisions. For example, through machine learning algorithms, it is found that certain types of equipment can achieve better results by adopting a certain optimization strategy under specific working conditions. This law is applied to future optimization decisions to achieve more accurate system optimization.

Claims

1. A dynamic optimization method for industrial automation system based on 5G private network, characterized in that: The following steps are involved: Step 1: System architecture construction and optimization: Modular design enhances system flexibility: The industrial automation system is divided into multiple independent and functionally clear modules, including data acquisition module, control module, execution module and communication module. The modules interact with each other through standardized interfaces, so that the system can easily add, remove or replace modules to adapt to production needs and technology updates, shorten iteration cycles and reduce maintenance costs. Data-driven accurate decision-making: In a 5G private network environment, production data such as equipment operating status, product quality parameters, and environmental data are collected in real time through sensors and actuators. The high-speed, low-latency characteristics of 5G are used to transmit the data to the data processing center in real time. Artificial intelligence algorithms are used for in-depth analysis, and control strategies are dynamically adjusted based on the analysis results to ensure optimization of the production process. Security protection system ensures stable operation: Build a security protection system from the hardware, software and network levels. Encryption chips and secure storage devices are used in hardware to ensure data security; vulnerability scanning and repair are performed on the software, and access rights are set; Implement 5G private network slicing technology on the network to isolate different business data, and establish a fault warning and emergency response mechanism to quickly launch emergency plans when abnormalities are detected to ensure system security and stability; Step 2: Intelligent data collection and preprocessing: Multi-source data fusion improves information integrity: With the high bandwidth characteristics of 5G private networks, multi-source data from different sensors and devices, including equipment operation data, production environment data, and product quality inspection data, are integrated, and data fusion algorithms are used to eliminate redundant and conflicting information, improve data accuracy and integrity, and provide a more reliable basis for subsequent control decisions; Real-time data processing ensures timeliness: Taking advantage of the low latency of 5G private networks, real-time pre-processing of data is performed at the data collection end, including data cleaning, feature extraction, and data compression, to reduce the burden of data transmission and processing, improve system response speed, ensure the timeliness of data processing, and meet the real-time requirements of industrial automation systems; Intelligent data cleaning improves data quality: Use machine learning and deep learning algorithms to automatically identify and process noise, outliers, and missing values ​​in the data, use clustering algorithms to detect outliers, use regression algorithms or neural networks to fill in missing values, and perform data standardization and normalization to improve data quality and provide accurate data support for the system; Step 3: Control strategy optimization based on artificial intelligence: Adaptive adjustment of neural network models: Build a neural network model suitable for industrial automation systems, use the large amount of production data transmitted in real time by 5G private networks to train and optimize the model, so that the model can adaptively adjust the control strategy, continuously optimize the weights and thresholds through online learning, and adjust the control parameters in real time according to changes in the production process, thereby improving system adaptability and control accuracy; Reinforcement learning achieves optimal decision-making: The control problem in the industrial automation system is modeled as a reinforcement learning problem, and the state space, action space and reward function are defined. The intelligent agent selects actions based on the system state and obtains reward feedback by interacting with the environment. The 5G private network is used to provide real-time feedback information, and the reinforcement learning algorithm continuously optimizes the strategy, so that the system can make the best decision in a complex environment and improve production efficiency and stability. Model predictive control optimization performance: Combining system mathematical models and real-time data, using the fast data transmission capability of 5G private networks, model predictive control of industrial automation systems is implemented to predict the future behavior of the system, optimize control inputs based on the prediction results, improve control performance through rolling optimization and feedback correction, reduce fluctuations and errors, and improve product quality consistency; Step 4: Communication resource management and optimization: Dynamic allocation of 5G private network slice resources: According to the needs of different services in the industrial automation system, such as real-time control services, data collection services, and video surveillance services, the network slicing technology of 5G private networks is used to allocate independent network resources to each service. By dynamically adjusting slice resources, low latency and high reliability of key services are ensured, thereby improving the overall performance of the system. Intelligent scheduling of wireless resources: Adopt intelligent scheduling algorithms to manage wireless resources in 5G private networks, dynamically allocate resources based on real-time device needs and channel quality, predict device data transmission needs, allocate resources in advance, and use channel status information to optimize resource allocation, improve spectrum efficiency and system capacity, and meet the needs of large-scale device access and data transmission in industrial automation systems; Network congestion control and optimization: Real-time monitoring of the network congestion status of 5G private networks, timely detection of congestion using traffic monitoring technology and congestion detection algorithms, and measures such as traffic control, routing adjustment, and resource allocation optimization to alleviate congestion, ensure the stability and reliability of data transmission, and ensure the normal operation of key services of industrial automation systems; Step 5: System performance evaluation and continuous improvement: Real-time monitoring of key performance indicators: Through the 5G private network, key performance indicators of industrial automation systems, such as production efficiency, product quality, equipment failure rate, and energy consumption, are collected in real time. Data analysis technology is used to monitor and visualize the indicators in real time, so that operators can understand the system operation status in a timely manner and provide a basis for optimizing decisions. Performance evaluation based on big data analysis: collect and store a large amount of historical production data, use big data analysis technology to mine potential patterns and rules in the data, establish a performance evaluation model, evaluate system performance and predict future trends, find system bottlenecks and optimization space based on the evaluation results, and provide direction for continuous improvement; Iterative update of optimization strategies: Based on the system performance evaluation results, the optimization strategies are generated using artificial intelligence algorithms, and the optimization strategies are sent to various devices in the industrial automation system in real time through the 5G private network. The effectiveness of strategy implementation is continuously monitored, and the optimization strategies are iteratively updated based on feedback information to achieve continuous improvement of system performance.

2. The method for dynamic optimization of industrial automation system based on 5G private network according to claim 1 is characterized in that: In the modular design, the functions of each module are as follows: Data acquisition module: responsible for collecting various data in the industrial production process, including equipment operation parameters, environmental parameters, product quality parameters, etc.; this module integrates various types of sensors, such as temperature sensors, pressure sensors, displacement sensors, and image sensors, and transmits the collected data to the data processing center in real time through the 5G private network to ensure the accuracy and timeliness of the data; Control module: Receives control instructions from the data processing center and controls the execution module according to the instructions; this module uses advanced control algorithms, such as PID control algorithm, model predictive control algorithm, and fuzzy control algorithm, to generate accurate control signals based on the needs of the production process and real-time data, to achieve precise control of the equipment; Execution module: According to the instructions of the control module, it performs corresponding operations, such as driving motors, controlling valves, and adjusting the working status of equipment. The execution module has high reliability and fast response capabilities to ensure that control instructions can be accurately executed to achieve automated control of the production process. Communication module: responsible for the data communication of the entire system, including data transmission between modules and data interaction with external systems; The module is based on 5G private network technology to achieve high-speed and stable data transmission and supports multiple communication protocols, such as TCP / IP protocol and OPC UA protocol, to ensure the compatibility and reliability of data communication.

3. The method for dynamic optimization of industrial automation system based on 5G private network according to claim 1 is characterized in that: In the data-driven decision-making, artificial intelligence algorithms include but are not limited to the following: Machine learning algorithms: such as decision tree algorithms, support vector machine algorithms, and neural network algorithms, which build data models by learning from historical data to predict equipment failures, optimize production processes, and improve product quality; Deep learning algorithms: such as convolutional neural network algorithms, recurrent neural network algorithms, and generative adversarial network algorithms, which are used to process complex image, voice, and text data to achieve intelligent monitoring, fault diagnosis, and quality inspection of the production process; Reinforcement learning algorithms: such as Q-learning algorithm, deep Q-network algorithm, and policy gradient algorithm, learn the optimal control strategy through the interaction between the intelligent agent and the environment, realize intelligent control of production equipment, optimal allocation of resources, and maximization of production efficiency.

4. The method for dynamic optimization of industrial automation system based on 5G private network according to claim 1 is characterized in that: In the security protection system, the encryption chip at the hardware level adopts the Advanced Encryption Standard algorithm, the secure storage device supports hardware encryption and access control, vulnerability scanning and repair at the software level are performed regularly and automatically, access permission settings are managed hierarchically according to user roles and permissions, and the 5G private network slicing technology at the network level is implemented using software-defined networking and network function virtualization technology.

5. The method for dynamic optimization of industrial automation system based on 5G private network according to claim 1 is characterized in that: In the multi-source data fusion, the data fusion algorithm adopts weighted averaging method, Kalman filtering method and Bayesian estimation method, performs weighted processing according to the reliability and importance of the data, eliminates redundancy and conflict between data, and improves the accuracy and completeness of the data.

6. The method for dynamic optimization of industrial automation system based on 5G private network according to claim 1 is characterized in that: In the real-time data processing, data cleaning adopts outlier detection and missing value processing technology based on statistical methods and machine learning algorithms, feature extraction adopts principal component analysis algorithm, linear discriminant analysis algorithm, wavelet transform algorithm, data compression adopts lossless compression algorithm and lossy compression algorithm, and the appropriate compression method is selected according to data characteristics and transmission requirements.

7. The method for dynamic optimization of industrial automation system based on 5G private network according to claim 1 is characterized in that: In the adaptive adjustment of the neural network model, the neural network model adopts a feedforward neural network, a feedback neural network, a convolutional neural network, and a recurrent neural network, and is trained and optimized through a back propagation algorithm, a stochastic gradient descent algorithm, and an adaptive learning rate adjustment algorithm to improve the convergence speed and generalization ability of the model.

8. The method for dynamic optimization of industrial automation system based on 5G private network according to claim 1 is characterized in that: In the reinforcement learning to achieve optimal decision-making, the state space includes the equipment operation status, the production environment status, and the task execution progress; the action space includes equipment control operations, resource allocation decisions, and production plan adjustments; the reward function is designed according to production goals and constraints, such as maximizing production efficiency, optimizing product quality, and minimizing costs. Strategy optimization is performed through Q learning algorithm, deep Q network algorithm, and proximal strategy optimization algorithm.

9. The method for dynamic optimization of industrial automation system based on 5G private network according to claim 1 is characterized in that: In the dynamic allocation of 5G private network slice resources, dynamic bandwidth allocation algorithm, priority scheduling algorithm and resource reservation algorithm are used to allocate resources according to the service priority, bandwidth requirement and latency requirement. The creation, configuration, monitoring and management of slices are realized through the network slice management system to ensure the service quality of key services.

10. The method for dynamic optimization of industrial automation system based on 5G private network according to claim 1, characterized in that: In the performance evaluation and continuous improvement of the system, big data analysis technology adopts data mining algorithms, machine learning algorithms, and statistical analysis methods; the performance evaluation model adopts hierarchical analysis method, fuzzy comprehensive evaluation method, and neural network evaluation method; the iterative update of the optimization strategy adopts genetic algorithm, particle swarm optimization algorithm, and simulated annealing algorithm to automatically adjust the optimization strategy according to changes in system performance to achieve continuous optimization of the system.

Images