Industrial equipment remote operation method based on distributed identification and control optimization

Abstract

The invention discloses an industrial equipment remote operation method based on distributed identification and control optimization. The method comprises the following steps: S1, constructing a hierarchical adaptive distributed computing architecture; s2, collecting multi-modal data of industrial equipment at an edge layer, and deploying an adaptive sparse convolutional neural network to carry out state detection and fault diagnosis; s3, deploying a distributed computing node cluster in a local layer, and performing local optimization on the control task through a distributed near-end strategy optimization algorithm; s4, processing data by utilizing cloud computing resources in a global layer, and generating a control strategy by adopting a multi-objective optimization algorithm; s5, carrying out data sharing and feedback among the hierarchies; s6, dynamically adjusting task distribution through a task migration mechanism; s7, introducing a hierarchical collaborative optimization mechanism, and optimizing a control strategy; and S8, performing data analysis by using a data-driven analysis technology to form closed-loop control. According to the invention, a distributed calculation and multi-objective optimization method is adopted, and efficient control of remote operation of industrial equipment is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment remote operation, and in particular to an industrial equipment remote operation method based on distributed identification and control optimization. Background Art

[0002] In the context of modern industrial automation and intelligent manufacturing, remote operation technology for industrial equipment has been widely used. This technology allows operators to monitor and operate industrial equipment distributed in different geographical locations through a remote control center, reducing dependence on on-site operations and improving production efficiency. However, with the increasing complexity of industrial equipment, especially in scenarios where multiple devices need to be monitored and operated simultaneously, traditional remote operation methods have gradually shown some obvious shortcomings.

[0003] Existing remote operating systems for industrial equipment usually rely on a centralized control architecture, where all data and control commands need to be processed through a central server. This centralized architecture limits the real-time and responsiveness of the system to a certain extent, especially when facing large-scale data processing and multi-device collaborative operations, where system latency and load bottleneck issues become particularly prominent. In addition, traditional control architectures are easily affected by network conditions during data transmission, resulting in delays or data loss during operation, which affects the safety of the equipment and the stability of operation.

[0004] Traditional methods for monitoring equipment status and fault diagnosis are usually based on static threshold settings or empirical rules, which have obvious limitations when dealing with complex industrial scenarios. For example, when the working environment of the equipment changes or the status of the equipment itself fluctuates, fixed threshold settings are difficult to cope with these changes, resulting in reduced monitoring accuracy. In addition, traditional fault diagnosis methods often rely on the analysis of a single data source and lack the ability to integrate and analyze multimodal data. The integration of multimodal data can provide richer equipment operation information, thereby improving the accuracy of fault diagnosis.

[0005] On the other hand, existing industrial equipment remote operation systems have limited optimization capabilities in task scheduling and resource allocation. In scenarios where multiple devices operate in coordination, the system is usually unable to dynamically adjust the distribution of tasks based on device status, task priority, and system load, leading to problems such as uneven resource allocation, overload of some devices, and delayed task response time. These deficiencies not only affect the operating efficiency of the system, but also reduce the reliability of operation and the service life of the equipment.

[0006] Current control optimization methods mostly focus on local optimization, lack global considerations, and cannot effectively coordinate the task requirements and resource utilization between multiple devices. Especially in large-scale industrial production, the task requirements of different devices are often dynamically changing. Traditional local optimization methods cannot adapt to the complex system environment, resulting in inefficient control strategies for the system as a whole. In the process of generating and optimizing global strategies, most of the multi-objective optimization algorithms in the existing technology fail to fully consider the balance between task rewards and resource consumption, and have limited use of feedback data, making it impossible to dynamically adjust the control strategy in real time, making it difficult for the system to adapt to environmental changes and load fluctuations.

[0007] In terms of data transmission, existing industrial equipment remote control systems usually use traditional network protocols for data transmission. This method is incapable of low-latency and high-frequency data interaction scenarios, especially when facing edge computing and large-scale data transmission from the local layer to the global layer. Existing protocols are difficult to ensure the real-time and integrity of data, and are prone to problems such as data loss and excessive latency, which in turn affects the safety of equipment and the stability of operation.

[0008] Therefore, how to provide a remote operation method for industrial equipment based on distributed identification and control optimization is an urgent problem that technicians in this field need to solve. Summary of the invention

[0009] One purpose of the present invention is to propose a remote operation method for industrial equipment based on distributed identification and control optimization. The present invention realizes efficient control of remote operation of industrial equipment by introducing distributed computing architecture and multi-objective optimization algorithm. Edge computing and embedded devices are used for real-time data collection and preprocessing, and the local and global control strategies of each level are integrated in combination with the hierarchical collaborative optimization algorithm to ensure dynamic adaptive adjustment of task allocation and resource scheduling. The system also realizes real-time monitoring and fault diagnosis of multiple devices through a low-latency data transmission mechanism and a task migration feedback mechanism, and has the advantages of strong real-time performance, high stability, flexible task scheduling, and optimized resource utilization.

[0010] The industrial equipment remote operation method based on distributed identification and control optimization according to an embodiment of the present invention comprises the following steps:

[0011] S1. Build a hierarchical adaptive distributed computing architecture, wherein the hierarchical adaptive distributed computing architecture specifically includes an edge layer, a local layer, and a global layer;

[0012] S2. Collect multimodal data of industrial equipment at the edge layer, including sensor data, image data, and video data, use embedded edge computing devices to pre-process the data, and deploy adaptive sparse convolutional neural networks for state detection and fault diagnosis;

[0013] S3, deploy distributed computing node clusters at the local layer, integrate the data uploaded from the edge layer, locally optimize the control tasks through the distributed proximal strategy optimization algorithm, and dynamically adjust the task distribution and resource allocation according to the device status, task priority and system load;

[0014] S4, using cloud computing resources at the global level to process data from multiple local levels, combining local optimization results and analysis data at the local level, performing large-scale data analysis, and using a multi-objective optimization algorithm to generate a control strategy;

[0015] S5. Data sharing and feedback are carried out between each layer through a low-latency data transmission protocol. The edge layer transmits the processed data to the local layer, and the local layer further uploads the integrated data to the global layer.

[0016] S6. Through the task migration mechanism, dynamically adjust the distribution of tasks according to the monitored computing load and system resource status, and migrate tasks between the edge layer, local layer and global layer;

[0017] S7. Introduce a hierarchical collaborative optimization mechanism to integrate the control strategies of the edge layer, local layer, and global layer. Use a hierarchical collaborative optimization algorithm to fuse local strategies and global strategies, and further optimize the control strategy based on feedback from task migration.

[0018] S8. Use data-driven analysis technology to analyze the historical and real-time data of industrial equipment operation, and send the generated control strategy to the edge layer and local layer for execution to form a closed-loop control.

[0019] Optionally, the S2 specifically includes:

[0020] S21. Collect multimodal data of industrial equipment at the edge layer, including sensor data, image data and video data, and perform time synchronization and spatial alignment on the collected multimodal data. Use a multimodal data fusion algorithm based on mutual information entropy to improve the temporal consistency of each data source.

[0021] S22, preprocessing the multimodal data, including noise reduction, sparse coding and feature enhancement, wherein the noise reduction adaptively adjusts the wavelet basis function through multi-scale wavelet transform; the sparse coding adopts compressed sensing theory combined with adaptive threshold selection to map the data to a low-dimensional sparse representation space; the feature enhancement adopts a nonlinear multi-scale embedding method to project high-dimensional features onto a low-dimensional manifold;

[0022] S23. Deploy an adaptive sparse convolutional neural network. The adaptive sparse convolutional neural network dynamically adjusts the sparsity and hierarchy of the convolution kernel according to the characteristic complexity of the data stream and the computing resource constraints. The sparsity adjustment mechanism is based on an adaptive partial differential integral equation model:

[0023] ;

[0024] in, represents the sparsity adjustment coefficient, represents the velocity field, represents the sparse diffusion coefficient, represents the feature space, The potential energy function representing sparsity adjustment depends on the data entropy and resource utilization , represents the regularization coefficient, Lagrange operator representing the time evolution of sparsity;

[0025] S24, Adaptive sparse convolutional neural network performs multi-scale feature extraction and sparse activation optimization on the preprocessed multimodal data. The weight distribution of the convolution layer is adaptively adjusted through the sparse activation mechanism based on the Gaussian field. The optimization goal of feature extraction is described as:

[0026] ;

[0027] in, represents the sparse activation energy function, represents the activation state variable, represents the activation potential energy function, represents a high-order constraint, represents the activation gradient, express The square norm of and represents the adjustment coefficient;

[0028] S25. Fault diagnosis of equipment status based on multi-scale features of adaptive sparse convolutional neural network, combining sparse attention mechanism with frequency domain conversion, integrating multi-modal dynamic features of historical data and real-time data, generating probability distribution and confidence of fault diagnosis, and adjusting the diagnostic output through nonlinear time-frequency gating mechanism;

[0029] S26. The fault diagnosis results and adjustment parameters are transmitted to the local layer, and the sparsity adjustment coefficient, activation state and network structure of the adaptive sparse convolutional neural network are dynamically updated based on the diagnostic information feedback information.

[0030] Optionally, the S3 specifically includes:

[0031] S31, deploying a distributed computing node cluster at the local layer, wherein the node cluster is composed of multiple heterogeneous computing nodes and connected through a high-speed data exchange network, using a distributed node discovery algorithm to dynamically identify available computing nodes, and constructing a topology-adaptive computing network structure;

[0032] S32, integrating and processing the multimodal data uploaded by the edge layer, including time alignment, data denoising and feature fusion, wherein the data denoising is implemented by a multi-scale sparse transformation algorithm, and the feature fusion adopts an adaptive kernel density estimation algorithm to map the features of different data sources into a high-dimensional feature space to generate a globally consistent feature representation;

[0033] S33. Use the distributed proximal strategy optimization algorithm to perform local optimization of the control task on the integrated data. The distributed proximal strategy optimization algorithm achieves rapid approximation of the global optimal solution through parallel strategy update and gradient restriction strategy:

[0034] ;

[0035] in, represents the updated policy parameters, Indicates the parameters before the policy update. represents the policy parameter gradient, represents the learning rate, Represents the policy function, which means that in the state Take action The probability of represents the advantage function, represents the cropping factor, represents the empirical data set, represents the clipping function, represents the cropping factor or threshold;

[0036] S34. Dynamically adjust the distribution of tasks and resource allocation according to the device status, task priority and system load. Task distribution optimization adopts a multi-objective optimization model based on the Markov decision process. The state space is composed of the device status, task priority and system load, and the strategy space is the resource allocation plan of the node:

[0037] ;

[0038] in, Indicates expected return, represents the mathematical expectation, represents the discount factor, Indicates in status Take action Rewards, Indicates in status Take action The resource consumption cost, represents the resource balance adjustment coefficient, represents the sparsity adjustment coefficient, represents the state-action value function, represents the state gradient, External force fields representing states;

[0039] S35. Based on the task execution feedback and real-time system load monitoring information, an adaptive task migration algorithm is used to adjust the distribution of tasks among different nodes, a task migration strategy is obtained based on a real-time load prediction model, and a recursive neural network is used for multi-step load prediction;

[0040] S36. Utilize the adaptive learning capability of the distributed proximal strategy optimization algorithm to dynamically adjust the strategy update frequency and learning rate based on the real-time feedback from each computing node.

[0041] Optionally, the S4 specifically includes:

[0042] S41, using cloud computing resources at the global layer to process data from multiple local layers, wherein the cloud computing resources receive and integrate multimodal data from multiple local layers through a distributed storage system, and divide and allocate tasks of different priorities using a scheduling algorithm based on resource weights and load balancing;

[0043] S42, combining the local optimization results of the local layer, performing global feature extraction and fusion on the multimodal data; feature extraction projects the data of each local layer into a sparse representation space through a high-dimensional feature projection algorithm to generate a global feature matrix, and uses a multi-scale feature fusion network to generate a feature embedding graph;

[0044] S43, generating a global control strategy through a multi-objective optimization algorithm, wherein the multi-objective optimization algorithm combines a genetic algorithm with a non-dominated sorting multi-objective optimization algorithm based on the Pareto frontier to generate a control strategy:

[0045] ;

[0046] in, represents the multi-objective optimization function, represents the control strategy variable, Represents the total number of tasks, Indicates The reward value of each task, Indicates The resource consumption of each task, represents the adjusted index of returns, represents the regulation index of the resource, The second-order partial derivative of the task measures the curvature change of the task, represents the adjustment parameter, represents the penalty adjustment coefficient, represents the exponential function;

[0047] S44. In the multi-objective optimization process, a dynamic constraint mechanism is introduced to adjust the resource allocation of the control strategy in real time through nonlinear mutation operations based on Lagrange multipliers:

[0048] ;

[0049] in, Represents a Lagrange multiplier The optimization function of Indicates The consumption limit of a resource, Indicates The constraint threshold of a resource, Indicates the total number of resources. represents the gradient operator;

[0050] S45. Generate a global control strategy based on the results of multi-objective optimization, including global task scheduling, data transmission optimization, resource allocation and load balancing rules to optimize the global effect;

[0051] S46. Dynamically adjust the global control strategy through a real-time feedback mechanism, use feedback data from the local layer, and apply an adaptive learning algorithm to incrementally update the strategy parameters.

[0052] Optionally, the S6 specifically includes:

[0053] S61. Dynamically adjust the distribution of tasks according to the monitored computing load and system resource status through the task migration mechanism. The task migration mechanism is based on the hierarchical load balancing algorithm to monitor the computing resources of the edge layer, local layer and global layer in real time and build a resource utilization matrix. , where the matrix elements Indicates at time node To Node The resource usage of the task is monitored and the task migration path is adjusted according to the resource status;

[0054] S62, adopt the task migration decision algorithm based on the improved Markov decision process, define the state space as the resource load state of the system, the action space as the migration operation of the task between nodes, the state transition probability is determined by the resource load matrix, and the expected reward function of the state transition combines the task priority, resource dissipation rate, delay penalty and time weighting factor:

[0055] ;

[0056] in, Indicates status Next action Expected return, Represents the total number of tasks, represents the exponential function, Indicates the priority weight of the task, represents the adjustment coefficient, represents the threshold of task migration, Indicates resource utilization, represents the delay penalty adjustment coefficient, Indicates the delay of the current task, Indicates the maximum allowed delay, represents the adjustment coefficient of the control acceleration term, Representation Task The second time derivative of the return, represents the time decay factor, represents the time adjustment parameter;

[0057] S63. Task migration is optimized by extending the Bellman equation to calculate the optimal migration strategy. The extended Bellman equation combines global resource distribution, return decay and risk factors to dynamically adjust the strategy.

[0058] S64. According to the optimized migration strategy, the tasks are dynamically distributed among the edge layer, the local layer, and the global layer. The migration decision of the tasks follows the principle of minimizing resource consumption and maximizing task returns.

[0059] S65, evaluating the effect of task migration through the adaptive load monitoring system, providing real-time feedback on the load status of the node after migration, and adjusting the parameters of the migration strategy according to the real-time feedback;

[0060] S66. Combine the feedback information to handle the abnormal load situation that occurs during the task migration process. When the load exceeds the preset threshold, the system automatically triggers the task reallocation mechanism:

[0061] ;

[0062] in, Indicates the load utilization, and represents the load threshold, Indicates the abnormal load adjustment factor, represents the nonlinear adjustment index, Represents an exponential function.

[0063] Optionally, the S7 specifically includes:

[0064] S71, generating local control strategies at the edge layer, local layer and global layer respectively, wherein the local control strategies are generated based on the task requirements, resource status and real-time environment status of each layer, and the strategy generation is completed through a local optimization algorithm;

[0065] S72, fusing the local control strategies of each level through a hierarchical collaborative optimization algorithm, and generating a preliminary global control strategy using a weighted fusion method based on constraint optimization;

[0066] S73. Preliminary optimization of the fused global control strategy is performed at the global level through a nonlinear multi-objective optimization function. The optimization objectives include the balance of task completion time, resource utilization, and task priority:

[0067] ;

[0068] in, represents the optimization objective function, represents the policy parameters, Represents the total number of tasks, Indicates The weight of the task, Indicates The reward function of a task, Indicates The resource consumption of each task, and represents the adjustment coefficient, represents the gradient function of the task, represents the smoothness term;

[0069] S74, combining the initially optimized global strategy with the feedback information of task migration, and adjusting the global strategy through real-time collection of feedback data, wherein the feedback data includes the real-time execution status of the task, resource utilization efficiency, and task response time:

[0070] ;

[0071] in, represents the task priority adjustment amount, represents the task migration feedback value, represents the adjustment coefficient, is the feedback sensitivity, represents the feedback threshold, represents the hyperbolic cosine function, Indicates the task reward value;

[0072] S75. Based on the feedback information of task migration, dynamically adjust the resource allocation and task priority in the global control strategy. The resource allocation strategy is adjusted through the adaptive feedback control model, and the priority is adjusted by adjusting the balance between feedback response speed and task reward.

[0073] The beneficial effects of the present invention are:

[0074] First, the present invention effectively solves the resource bottleneck problem existing in the traditional centralized control architecture by constructing a hierarchical adaptive distributed computing architecture. The task allocation and resource scheduling between the various levels can be flexibly adjusted, ensuring the efficiency and stability of the system when processing large-scale data and multi-device collaborative operations. The edge layer uses embedded computing devices to realize real-time collection and preprocessing of multimodal data, and deploys adaptive sparse convolutional neural networks for state detection and fault diagnosis, which significantly improves the accuracy and response speed of equipment monitoring.

[0075] Secondly, through the distributed proximal policy optimization algorithm, the present invention can dynamically adjust the distribution of tasks and resource allocation according to the actual operating status of the equipment, task priority and system load, ensuring the optimal use of system resources. Compared with the traditional static optimization method, the distributed proximal policy optimization algorithm achieves the adaptability of task scheduling through parallel policy updates and task migration mechanisms, and improves the system's ability to respond to emergencies and environmental changes. Especially in the face of complex industrial scenarios, the task migration mechanism flexibly adjusts the task allocation plan based on real-time feedback, so that the system can still maintain efficient operation under load fluctuations and changes in equipment operating status.

[0076] In addition, by generating a global control strategy through a multi-objective optimization algorithm, we can fully consider the balance of task rewards, resource consumption, and task priority at the global level to achieve the global optimal control strategy. The global control strategy combines the local optimization results of the local layer to ensure the consistency and coordination of strategies between levels. The hierarchical collaborative optimization mechanism further integrates the local control strategies of each level with the global control strategy, optimizing the resource utilization and task execution efficiency of the overall system. The strategy parameters are adjusted based on the feedback of task migration to form a dynamic closed-loop control system, which continuously optimizes the control effect and realizes precise control of equipment and efficient fault diagnosis.

[0077] The present invention also introduces a low-latency data transmission protocol to ensure that data sharing and feedback between various levels can be carried out in a timely and effective manner, greatly improving the real-time performance and data transmission efficiency of the system. At the same time, the system adopts a multimodal data fusion and analysis method, and comprehensively utilizes multiple data sources such as sensors, images and videos to enhance the accuracy and reliability of fault detection and status monitoring. Through data-driven analysis technology, the historical data of equipment operation is combined with real-time data to achieve comprehensive monitoring and predictive maintenance of equipment operation status, effectively reducing the occurrence of equipment failures and reducing maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0079] Figure 1 A flowchart of the industrial equipment remote operation method based on distributed identification and control optimization proposed by the present invention;

[0080] Figure 2 Schematic diagram of the hierarchical adaptive distributed computing architecture of the industrial equipment remote operation method based on distributed identification and control optimization proposed in the present invention. DETAILED DESCRIPTION

[0081] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0082] refer to Figure 1 and Figure 2 , a remote operation method for industrial equipment based on distributed identification and control optimization, comprising the following steps:

[0083] S1. Build a hierarchical adaptive distributed computing architecture, wherein the hierarchical adaptive distributed computing architecture specifically includes an edge layer, a local layer, and a global layer;

[0084] S2. Collect multimodal data of industrial equipment at the edge layer, including sensor data, image data, and video data, use embedded edge computing devices to pre-process the data, and deploy adaptive sparse convolutional neural networks for state detection and fault diagnosis;

[0085] S3, deploy distributed computing node clusters at the local layer, integrate the data uploaded from the edge layer, locally optimize the control tasks through the distributed proximal strategy optimization algorithm, and dynamically adjust the task distribution and resource allocation according to the device status, task priority and system load;

[0086] S4, using cloud computing resources at the global level to process data from multiple local levels, combining local optimization results and analysis data at the local level, performing large-scale data analysis, and using a multi-objective optimization algorithm to generate a control strategy;

[0087] S5. Data sharing and feedback are carried out between each layer through a low-latency data transmission protocol. The edge layer transmits the processed data to the local layer, and the local layer further uploads the integrated data to the global layer.

[0088] S6. Through the task migration mechanism, dynamically adjust the distribution of tasks according to the monitored computing load and system resource status, and migrate tasks between the edge layer, local layer and global layer;

[0089] S7. Introduce a hierarchical collaborative optimization mechanism to integrate the control strategies of the edge layer, local layer, and global layer. Use a hierarchical collaborative optimization algorithm to fuse local strategies and global strategies, and further optimize the control strategy based on feedback from task migration.

[0090] S8. Use data-driven analysis technology to analyze the historical and real-time data of industrial equipment operation, and send the generated control strategy to the edge layer and local layer for execution to form a closed-loop control.

[0091] In this implementation, S2 specifically includes:

[0092] S21. Collect multimodal data of industrial equipment at the edge layer, including sensor data, image data and video data, and perform time synchronization and spatial alignment on the collected multimodal data. Use a multimodal data fusion algorithm based on mutual information entropy to improve the temporal consistency of each data source.

[0093] S22, preprocessing the multimodal data, including noise reduction, sparse coding and feature enhancement, wherein the noise reduction adaptively adjusts the wavelet basis function through multi-scale wavelet transform; the sparse coding adopts compressed sensing theory combined with adaptive threshold selection to map the data to a low-dimensional sparse representation space; the feature enhancement adopts a nonlinear multi-scale embedding method to project high-dimensional features onto a low-dimensional manifold;

[0094] S23. Deploy an adaptive sparse convolutional neural network. The adaptive sparse convolutional neural network dynamically adjusts the sparsity and hierarchy of the convolution kernel according to the characteristic complexity of the data stream and the computing resource constraints. The sparsity adjustment mechanism is based on an adaptive partial differential integral equation model:

[0095] ;

[0096] in, represents the sparsity adjustment coefficient, represents the velocity field, represents the sparse diffusion coefficient, represents the feature space, The potential energy function representing sparsity adjustment depends on the data entropy and resource utilization , represents the regularization coefficient, Lagrange operator representing the time evolution of sparsity;

[0097] S24, Adaptive sparse convolutional neural network performs multi-scale feature extraction and sparse activation optimization on the preprocessed multimodal data. The weight distribution of the convolution layer is adaptively adjusted through the sparse activation mechanism based on the Gaussian field. The optimization goal of feature extraction is described as:

[0098] ;

[0099] in, represents the sparse activation energy function, represents the activation state variable, represents the activation potential energy function, represents a high-order constraint, represents the activation gradient, express The square norm of and represents the adjustment coefficient;

[0100] S25. Fault diagnosis of equipment status based on multi-scale features of adaptive sparse convolutional neural network, combining sparse attention mechanism with frequency domain conversion, integrating multi-modal dynamic features of historical data and real-time data, generating probability distribution and confidence of fault diagnosis, and adjusting the diagnostic output through nonlinear time-frequency gating mechanism;

[0101] S26. The fault diagnosis results and adjustment parameters are transmitted to the local layer, and the sparsity adjustment coefficient, activation state and network structure of the adaptive sparse convolutional neural network are dynamically updated based on the diagnostic information feedback information.

[0102] In this implementation, S3 specifically includes:

[0103] S31, deploying a distributed computing node cluster at the local layer, wherein the node cluster is composed of multiple heterogeneous computing nodes and connected through a high-speed data exchange network, using a distributed node discovery algorithm to dynamically identify available computing nodes, and constructing a topology-adaptive computing network structure;

[0104] S32, integrating and processing the multimodal data uploaded by the edge layer, including time alignment, data denoising and feature fusion, wherein the data denoising is implemented by a multi-scale sparse transformation algorithm, and the feature fusion adopts an adaptive kernel density estimation algorithm to map the features of different data sources into a high-dimensional feature space to generate a globally consistent feature representation;

[0105] S33. Use the distributed proximal strategy optimization algorithm to perform local optimization of the control task on the integrated data. The distributed proximal strategy optimization algorithm achieves rapid approximation of the global optimal solution through parallel strategy update and gradient restriction strategy:

[0106] ;

[0107] in, represents the updated policy parameters, Indicates the parameters before the policy update. represents the policy parameter gradient, represents the learning rate, Represents the policy function, which means that in the state Take action The probability of represents the advantage function, represents the cropping factor, represents the empirical data set, represents the clipping function, represents the cropping factor or threshold;

[0108] S34. Dynamically adjust the distribution of tasks and resource allocation according to the device status, task priority and system load. Task distribution optimization adopts a multi-objective optimization model based on the Markov decision process. The state space is composed of the device status, task priority and system load, and the strategy space is the resource allocation plan of the node:

[0109] ;

[0110] in, Indicates expected return, represents the mathematical expectation, represents the discount factor, Indicates in status Take action Rewards, Indicates in status Take action The resource consumption cost, represents the resource balance adjustment coefficient, represents the sparsity adjustment coefficient, represents the state-action value function, represents the state gradient, External force fields representing states;

[0111] S35. Based on the task execution feedback and real-time system load monitoring information, an adaptive task migration algorithm is used to adjust the distribution of tasks among different nodes, a task migration strategy is obtained based on a real-time load prediction model, and a recursive neural network is used for multi-step load prediction;

[0112] S36. Utilize the adaptive learning capability of the distributed proximal strategy optimization algorithm to dynamically adjust the strategy update frequency and learning rate based on the real-time feedback from each computing node.

[0113] In this implementation, S4 specifically includes:

[0114] S41, using cloud computing resources at the global layer to process data from multiple local layers, wherein the cloud computing resources receive and integrate multimodal data from multiple local layers through a distributed storage system, and divide and allocate tasks of different priorities using a scheduling algorithm based on resource weights and load balancing;

[0115] S42, combining the local optimization results of the local layer, performing global feature extraction and fusion on the multimodal data; feature extraction projects the data of each local layer into a sparse representation space through a high-dimensional feature projection algorithm to generate a global feature matrix, and uses a multi-scale feature fusion network to generate a feature embedding graph;

[0116] S43, generating a global control strategy through a multi-objective optimization algorithm, wherein the multi-objective optimization algorithm combines a genetic algorithm with a non-dominated sorting multi-objective optimization algorithm based on the Pareto frontier to generate a control strategy:

[0117] ;

[0118] in, represents the multi-objective optimization function, represents the control strategy variable, Represents the total number of tasks, Indicates The reward value of each task, Indicates The resource consumption of each task, represents the adjusted index of returns, represents the regulation index of the resource, The second-order partial derivative of the task measures the curvature change of the task, represents the adjustment parameter, represents the penalty adjustment coefficient, represents the exponential function;

[0119] S44. In the multi-objective optimization process, a dynamic constraint mechanism is introduced to adjust the resource allocation of the control strategy in real time through nonlinear mutation operations based on Lagrange multipliers:

[0120] ;

[0121] in, Represents a Lagrange multiplier The optimization function of Indicates The consumption limit of a resource, Indicates The constraint threshold of a resource, Indicates the total number of resources. represents the gradient operator;

[0122] S45. Generate a global control strategy based on the results of multi-objective optimization, including global task scheduling, data transmission optimization, resource allocation and load balancing rules to optimize the global effect;

[0123] S46. Dynamically adjust the global control strategy through a real-time feedback mechanism, use feedback data from the local layer, and apply an adaptive learning algorithm to incrementally update the strategy parameters.

[0124] In this implementation manner, S6 specifically includes:

[0125] S61. Dynamically adjust the distribution of tasks according to the monitored computing load and system resource status through the task migration mechanism. The task migration mechanism is based on the hierarchical load balancing algorithm to monitor the computing resources of the edge layer, local layer and global layer in real time and build a resource utilization matrix. , where the matrix elements Indicates at time node To Node The resource usage of the task is monitored and the task migration path is adjusted according to the resource status;

[0126] S62, adopt the task migration decision algorithm based on the improved Markov decision process, define the state space as the resource load state of the system, the action space as the migration operation of the task between nodes, the state transition probability is determined by the resource load matrix, and the expected reward function of the state transition combines the task priority, resource dissipation rate, delay penalty and time weighting factor:

[0127] ;

[0128] in, Indicates status Next action Expected return, Represents the total number of tasks, represents the exponential function, Indicates the priority weight of the task, represents the adjustment coefficient, represents the threshold of task migration, Indicates resource utilization, represents the delay penalty adjustment coefficient, Indicates the delay of the current task, Indicates the maximum allowed delay, represents the adjustment coefficient of the control acceleration term, Representation Task The second time derivative of the return, represents the time decay factor, represents the time adjustment parameter;

[0129] S63. Task migration is optimized by extending the Bellman equation to calculate the optimal migration strategy. The extended Bellman equation combines global resource distribution, return decay and risk factors to dynamically adjust the strategy.

[0130] S64. According to the optimized migration strategy, the tasks are dynamically distributed among the edge layer, the local layer, and the global layer. The migration decision of the tasks follows the principle of minimizing resource consumption and maximizing task returns.

[0131] S65, evaluating the effect of task migration through the adaptive load monitoring system, providing real-time feedback on the load status of the node after migration, and adjusting the parameters of the migration strategy according to the real-time feedback;

[0132] S66. Combine the feedback information to handle the abnormal load situation that occurs during the task migration process. When the load exceeds the preset threshold, the system automatically triggers the task reallocation mechanism:

[0133] ;

[0134] in, Indicates the load utilization, and represents the load threshold, Indicates the abnormal load adjustment factor, represents the nonlinear regulation index, Represents an exponential function.

[0135] In this implementation manner, the S7 specifically includes:

[0136] S71, generating local control strategies at the edge layer, local layer and global layer respectively, wherein the local control strategies are generated based on the task requirements, resource status and real-time environment status of each layer, and the strategy generation is completed through a local optimization algorithm;

[0137] S72, fusing the local control strategies of each level through a hierarchical collaborative optimization algorithm, and generating a preliminary global control strategy using a weighted fusion method based on constraint optimization;

[0138] S73. Preliminary optimization of the fused global control strategy is performed at the global level through a nonlinear multi-objective optimization function. The optimization objectives include the balance of task completion time, resource utilization, and task priority:

[0139] ;

[0140] in, represents the optimization objective function, represents the policy parameters, Indicates the total number of tasks, Indicates The weight of the task, Indicates The reward function of a task, Indicates The resource consumption of each task, and represents the adjustment coefficient, represents the gradient function of the task, represents the smoothness term;

[0141] S74, combining the initially optimized global strategy with the feedback information of task migration, and adjusting the global strategy through real-time collection of feedback data, wherein the feedback data includes the real-time execution status of the task, resource utilization efficiency, and task response time:

[0142] ;

[0143] in, represents the task priority adjustment amount, represents the task migration feedback value, represents the adjustment coefficient, is the feedback sensitivity, represents the feedback threshold, represents the hyperbolic cosine function, Indicates the task reward value;

[0144] S75. Based on the feedback information of task migration, dynamically adjust the resource allocation and task priority in the global control strategy. The resource allocation strategy is adjusted through the adaptive feedback control model, and the priority is adjusted by adjusting the balance between feedback response speed and task reward.

[0145] Embodiment 1:

[0146] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a large chemical plant. The plant relies on a large number of equipment for complex production operations, including multiple industrial reactors, pumps, compressors, transmission pipelines and other equipment. These devices are distributed in different areas of the plant and are centrally controlled through a remote monitoring system. However, traditional remote control systems face some problems, such as the inability to respond to sudden failures in real time, lack of flexibility in task scheduling, resulting in low resource utilization and frequent equipment overload problems. In addition, due to the wide variety of equipment and the complex form of the collected data (such as sensor data, image data, video streams, etc.), the existing single data source monitoring method is difficult to accurately perform equipment status evaluation and fault diagnosis. This poses a great challenge to the safe operation and production efficiency of the factory.

[0147] In order to overcome the above problems, the factory decided to introduce the remote operation method of industrial equipment based on distributed identification and control optimization proposed in this invention. This method controls and optimizes the equipment in layers through a hierarchical adaptive distributed computing architecture, thereby improving the real-time response capability of the system and the stability of equipment operation. In actual applications, the factory divides the control process into three levels: edge layer, local layer, and global layer. The edge layer collects multimodal data of each device in real time through embedded edge computing devices installed on the equipment, including temperature, vibration, pressure, video surveillance images, etc. The collected data is preprocessed and sparse convolutional neural network is used for state detection and fault diagnosis, which reduces the amount of data transmission while ensuring the data processing speed.

[0148] In specific applications, the embedded devices at the edge layer collect 100MB of device status data per second. Through compression algorithms and multimodal data fusion, the data transmission volume is reduced to 20MB per second, effectively reducing the burden on the local and global layers. During a certain device operation, reactor R-03 experienced abnormal vibration. The edge layer device detected the fault within 30 milliseconds and generated a detailed diagnostic report through the fault diagnosis system. The diagnostic report pointed out that the abnormal vibration was caused by the loose transmission shaft. Subsequently, the fault information was transmitted to the local layer for further analysis and processing.

[0149] At the local level, the distributed computing node cluster begins to analyze the equipment status and task scheduling based on the received fault information. The local level integrates the historical data of the reactor R-03 and the real-time data of other equipment, and determines that the equipment needs to be shut down for maintenance immediately to avoid more serious equipment damage. At the same time, the system reallocates the tasks of other equipment through the distributed proximal strategy optimization algorithm to maintain the continuity of the production line. Data shows that under the traditional system, the average response time for equipment shutdown scheduling tasks is 5 minutes, but in actual tests after the introduction of this invention, the response time is reduced to less than 1 minute, which greatly improves the speed of fault handling.

[0150] The global layer uses cloud computing resources to perform large-scale analysis and optimization of data from multiple local layers. A global control strategy is generated through a multi-objective optimization algorithm to ensure that the equipment in the factory can maximize its efficiency during the failure. During the reactor failure handling process, the global layer replanned the production tasks and reduced production losses caused by downtime. The optimization results of the global layer show that the system performs well in load balancing, resource utilization and task scheduling. The overall resource utilization of the factory has increased by 15%, while the downtime of the production line has been reduced by 30%.

[0151] In order to further verify the effectiveness of the present invention, the factory conducted a one-month test under different equipment operating conditions. During the test, the factory recorded the equipment's core indicators such as fault response time, resource utilization, and task scheduling efficiency. The following is part of the test data:

[0152] Table 1 Test data of industrial equipment remote operation system based on distributed identification and control optimization

[0153] Test Date Faulty equipment Fault type Fault response time (seconds) Improvement in resource utilization (%) Task scheduling response time (seconds) 2024-08-01 R-03 Abnormal vibration 25 15 30 2024-08-05 P-07 Temperature exceeds standard 35 10 28 2024-08-10 C-01 Too much pressure 20 12 25 2024-08-15 T-12 Pipeline Leaks 40 14 32 2024-08-20 V-05 Video surveillance failure 30 18 27

[0154] The data in Table 1 above show that the present invention greatly improves the speed of fault response in practical applications, and realizes dynamic adjustment of equipment load through distributed proximal strategy optimization and task migration mechanism, which significantly improves resource utilization. In the vibration fault processing of equipment R-03, the system can respond quickly and adjust task allocation according to feedback information, so that the factory production can continue to run and avoid long-term downtime. In addition, the experimental results also show that by optimizing the global control strategy, the overall production efficiency of the factory has been significantly improved.

[0155] It can be seen from the above application examples that the present invention effectively solves the problems existing in the existing remote operating system of industrial equipment, such as poor real-time performance, equipment overload, and inflexible task scheduling, ensuring that the system can maintain efficient and safe operation in the face of complex production environments. The hierarchical collaborative optimization mechanism proposed in the present invention can integrate control strategies at different levels to form a closed-loop feedback control, ensuring the accuracy and speed of equipment status monitoring and fault diagnosis. At the same time, the present invention further improves the real-time response capability of the system through a low-latency data transmission mechanism, making the collaborative operation between devices smoother and effectively reducing losses and risks in production.

[0156] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A remote operation method for industrial equipment based on distributed identification and control optimization, characterized in that: The steps include: S1. Build a hierarchical adaptive distributed computing architecture, wherein the hierarchical adaptive distributed computing architecture specifically includes an edge layer, a local layer, and a global layer; S2. Collect multimodal data of industrial equipment at the edge layer, including sensor data, image data, and video data, use embedded edge computing devices to pre-process the data, and deploy adaptive sparse convolutional neural networks for state detection and fault diagnosis; S3, deploy distributed computing node clusters at the local layer, integrate the data uploaded from the edge layer, locally optimize the control tasks through the distributed proximal strategy optimization algorithm, and dynamically adjust the task distribution and resource allocation according to the device status, task priority and system load; S4, using cloud computing resources at the global level to process data from multiple local levels, combining local optimization results and analysis data at the local level, performing large-scale data analysis, and using a multi-objective optimization algorithm to generate a control strategy; S5. Data sharing and feedback are carried out between each layer through a low-latency data transmission protocol. The edge layer transmits the processed data to the local layer, and the local layer further uploads the integrated data to the global layer. S6. Through the task migration mechanism, dynamically adjust the distribution of tasks according to the monitored computing load and system resource status, and migrate tasks between the edge layer, local layer and global layer; S7. Introduce a hierarchical collaborative optimization mechanism to integrate the control strategies of the edge layer, local layer, and global layer. Use a hierarchical collaborative optimization algorithm to fuse local strategies and global strategies, and further optimize the control strategy based on feedback from task migration. S8. Use data-driven analysis technology to analyze the historical and real-time data of industrial equipment operation, and send the generated control strategy to the edge layer and local layer for execution to form a closed-loop control.

2. The method for remote operation of industrial equipment based on distributed identification and control optimization according to claim 1 is characterized in that: The S2 specifically includes: S21. Collect multimodal data of industrial equipment at the edge layer, including sensor data, image data and video data, and perform time synchronization and spatial alignment on the collected multimodal data. Use a multimodal data fusion algorithm based on mutual information entropy to improve the temporal consistency of each data source. S22, preprocessing the multimodal data, including noise reduction, sparse coding and feature enhancement, wherein the noise reduction adaptively adjusts the wavelet basis function through multi-scale wavelet transform; the sparse coding adopts compressed sensing theory combined with adaptive threshold selection to map the data to a low-dimensional sparse representation space; the feature enhancement adopts a nonlinear multi-scale embedding method to project high-dimensional features onto a low-dimensional manifold; S23. Deploy an adaptive sparse convolutional neural network. The adaptive sparse convolutional neural network dynamically adjusts the sparsity and hierarchy of the convolution kernel according to the characteristic complexity of the data stream and the computing resource constraints. The sparsity adjustment mechanism is based on an adaptive partial differential integral equation model: ； in, represents the sparsity adjustment coefficient, represents the velocity field, represents the sparse diffusion coefficient, represents the feature space, The potential energy function representing sparsity adjustment depends on the data entropy and resource utilization , represents the regularization coefficient, Lagrange operator representing the time evolution of sparsity; S24, Adaptive sparse convolutional neural network performs multi-scale feature extraction and sparse activation optimization on the preprocessed multimodal data. The weight distribution of the convolution layer is adaptively adjusted through the sparse activation mechanism based on the Gaussian field. The optimization goal of feature extraction is described as: ； in, represents the sparse activation energy function, represents the activation state variable, represents the activated potential energy function, represents a high-order constraint, represents the activation gradient, express The square norm of and represents the adjustment coefficient; S25. Fault diagnosis of equipment status based on multi-scale features of adaptive sparse convolutional neural network, combining sparse attention mechanism with frequency domain conversion, integrating multi-modal dynamic features of historical data and real-time data, generating probability distribution and confidence of fault diagnosis, and adjusting the diagnostic output through nonlinear time-frequency gating mechanism; S26. The fault diagnosis results and adjustment parameters are transmitted to the local layer, and the sparsity adjustment coefficient, activation state and network structure of the adaptive sparse convolutional neural network are dynamically updated based on the diagnostic information feedback information.

3. The industrial equipment remote operation method based on distributed identification and control optimization according to claim 1 is characterized in that: The S3 specifically includes: S31, deploying a distributed computing node cluster at the local layer, wherein the node cluster is composed of multiple heterogeneous computing nodes and connected through a high-speed data exchange network, using a distributed node discovery algorithm to dynamically identify available computing nodes, and constructing a topology-adaptive computing network structure; S32, integrating and processing the multimodal data uploaded by the edge layer, including time alignment, data denoising and feature fusion, wherein the data denoising is implemented by a multi-scale sparse transformation algorithm, and the feature fusion adopts an adaptive kernel density estimation algorithm to map the features of different data sources into a high-dimensional feature space to generate a globally consistent feature representation; S33. Use the distributed proximal strategy optimization algorithm to perform local optimization of the control task on the integrated data. The distributed proximal strategy optimization algorithm achieves rapid approximation of the global optimal solution through parallel strategy update and gradient restriction strategy: ； in, represents the updated policy parameters, Indicates the parameters before the policy update. represents the policy parameter gradient, represents the learning rate, Represents the policy function, which means that in the state Take action The probability of represents the advantage function, represents the cropping factor, represents the empirical data set, represents the clipping function, represents the cropping factor or threshold; S34. Dynamically adjust the distribution of tasks and resource allocation according to the device status, task priority and system load. Task distribution optimization adopts a multi-objective optimization model based on the Markov decision process. The state space is composed of the device status, task priority and system load, and the strategy space is the resource allocation plan of the node: ； in, Indicates expected return, represents the mathematical expectation, represents the discount factor, Indicates in status Take action Rewards, Indicates in status Take action The resource consumption cost, represents the resource balance adjustment coefficient, represents the sparsity adjustment coefficient, represents the state-action value function, represents the state gradient, External force fields representing states; S35. Based on the task execution feedback and real-time system load monitoring information, an adaptive task migration algorithm is used to adjust the distribution of tasks among different nodes, a task migration strategy is obtained based on a real-time load prediction model, and a recursive neural network is used for multi-step load prediction; S36. Utilize the adaptive learning capability of the distributed proximal strategy optimization algorithm to dynamically adjust the strategy update frequency and learning rate based on the real-time feedback from each computing node.

4. The method for remote operation of industrial equipment based on distributed identification and control optimization according to claim 1 is characterized in that: The S4 specifically includes: S41, using cloud computing resources at the global layer to process data from multiple local layers, wherein the cloud computing resources receive and integrate multimodal data from multiple local layers through a distributed storage system, and divide and allocate tasks of different priorities using a scheduling algorithm based on resource weights and load balancing; S42, combining the local optimization results of the local layer, performing global feature extraction and fusion on the multimodal data; feature extraction projects the data of each local layer into a sparse representation space through a high-dimensional feature projection algorithm to generate a global feature matrix, and uses a multi-scale feature fusion network to generate a feature embedding graph; S43, generating a global control strategy through a multi-objective optimization algorithm, wherein the multi-objective optimization algorithm combines a genetic algorithm with a non-dominated sorting multi-objective optimization algorithm based on the Pareto frontier to generate a control strategy: ； in, represents the multi-objective optimization function, represents the control strategy variable, Indicates the total number of tasks, Indicates The reward value of each task, Indicates The resource consumption of each task, represents the adjusted index of returns, represents the regulation index of the resource, The second-order partial derivative of the task measures the curvature change of the task, represents the adjustment parameter, represents the penalty adjustment coefficient, represents the exponential function; S44. In the multi-objective optimization process, a dynamic constraint mechanism is introduced to adjust the resource allocation of the control strategy in real time through nonlinear mutation operations based on Lagrange multipliers: ； in, Represents a Lagrange multiplier The optimization function of Indicates The consumption limit of a resource, Indicates The constraint threshold of a resource, Indicates the total number of resources. represents the gradient operator; S45. Generate a global control strategy based on the results of multi-objective optimization, including global task scheduling, data transmission optimization, resource allocation and load balancing rules to optimize the global effect; S46. Dynamically adjust the global control strategy through a real-time feedback mechanism, use feedback data from the local layer, and apply an adaptive learning algorithm to incrementally update the strategy parameters.

5. The method for remote operation of industrial equipment based on distributed identification and control optimization according to claim 1, characterized in that: The S6 specifically includes: S61. Dynamically adjust the distribution of tasks according to the monitored computing load and system resource status through the task migration mechanism. The task migration mechanism is based on the hierarchical load balancing algorithm to monitor the computing resources of the edge layer, local layer and global layer in real time and build a resource utilization matrix. , where the matrix elements Indicates at time node To Node The resource usage of the task is monitored and the task migration path is adjusted according to the resource status; S62, adopt the task migration decision algorithm based on the improved Markov decision process, define the state space as the resource load state of the system, the action space as the migration operation of the task between nodes, the state transition probability is determined by the resource load matrix, and the expected reward function of the state transition combines the task priority, resource dissipation rate, delay penalty and time weighting factor: ； in, Indicates status Next action Expected return, Indicates the total number of tasks, represents the exponential function, Indicates the priority weight of the task, represents the adjustment coefficient, represents the threshold of task migration, Indicates resource utilization, represents the delay penalty adjustment coefficient, Indicates the delay of the current task, Indicates the maximum allowed delay, represents the adjustment coefficient of the control acceleration term, Indicates the task The second time derivative of the return, represents the time decay factor, represents the time adjustment parameter; S63. Task migration is optimized by extending the Bellman equation to calculate the optimal migration strategy. The extended Bellman equation combines global resource distribution, return decay and risk factors to dynamically adjust the strategy. S64. According to the optimized migration strategy, the tasks are dynamically distributed among the edge layer, the local layer, and the global layer. The migration decision of the tasks follows the principle of minimizing resource consumption and maximizing task returns. S65, evaluating the effect of task migration through the adaptive load monitoring system, providing real-time feedback on the load status of the node after migration, and adjusting the parameters of the migration strategy according to the real-time feedback; S66. Combine the feedback information to handle the abnormal load situation that occurs during the task migration process. When the load exceeds the preset threshold, the system automatically triggers the task reallocation mechanism: ； in, Indicates the load utilization, and represents the load threshold, Indicates the abnormal load adjustment factor, represents the nonlinear regulation index, Represents an exponential function.

6. The method for remote operation of industrial equipment based on distributed identification and control optimization according to claim 1, characterized in that: The S7 specifically includes: S71, generating local control strategies at the edge layer, local layer and global layer respectively, wherein the local control strategies are generated based on the task requirements, resource status and real-time environment status of each layer, and the strategy generation is completed through a local optimization algorithm; S72, fusing the local control strategies of each level through a hierarchical collaborative optimization algorithm, and generating a preliminary global control strategy using a weighted fusion method based on constraint optimization; S73. Preliminary optimization of the fused global control strategy is performed at the global level through a nonlinear multi-objective optimization function. The optimization objectives include the balance of task completion time, resource utilization, and task priority: ； in, represents the optimization objective function, represents the policy parameters, Indicates the total number of tasks, Indicates The weight of the task, Indicates The reward function of a task, Indicates The resource consumption of each task, and represents the adjustment coefficient, represents the gradient function of the task, represents the smoothness term; S74, combining the initially optimized global strategy with the feedback information of task migration, and adjusting the global strategy through real-time collection of feedback data, wherein the feedback data includes the real-time execution status of the task, resource utilization efficiency, and task response time: ； in, represents the task priority adjustment amount, represents the task migration feedback value, represents the adjustment coefficient, is the feedback sensitivity, represents the feedback threshold, represents the hyperbolic cosine function, Indicates the task reward value; S75. Based on the feedback information of task migration, dynamically adjust the resource allocation and task priority in the global control strategy. The resource allocation strategy is adjusted through the adaptive feedback control model, and the priority is adjusted by adjusting the balance between feedback response speed and task reward.

Images