Mobile robot operation efficiency evaluation method and system based on industrial Internet

Abstract

The present invention provides a method and system for evaluating the operation efficiency of a mobile robot based on the industrial Internet, which relates to the technical field of operation efficiency evaluation, and includes: collecting multimodal operation data, selecting a transmission channel and uploading it to an edge computing node, performing preprocessing, completing information and performing filtering operations, and generating a first semantic feature and an operation feature descriptor; selecting an optimal transmission path and a service function chain and uploading the operation feature descriptor to a cloud server, performing real-time analysis on the operation feature descriptor, generating an initial operation efficiency evaluation result, performing analysis and comparison, determining efficiency bottlenecks and generating initial optimization and improvement suggestions; sending the initial operation efficiency evaluation result to an edge computing node, performing semantic enhancement and logical reasoning, generating a detailed efficiency evaluation report, analyzing the correlation and influencing mechanism between different evaluation indicators, determining key efficiency bottleneck factors, generating comprehensive optimization suggestions, collecting operation data in real time and generating a comprehensive efficiency evaluation report.

Description

Technical Field

[0001] The present invention relates to the technical field of operation efficiency evaluation, and in particular to a method and system for evaluating the operation efficiency of a mobile robot based on the industrial Internet. Background Art

[0002] With the rapid development of artificial intelligence and robotics technology and the deepening of industrial transformation, mobile robots have been widely used in industrial production, logistics warehousing, smart medical care and other fields. Mobile robots have realized the automation and intelligence of the production process through autonomous navigation, environmental perception, task planning and other functions, significantly improving production efficiency and flexibility.

[0003] However, in a complex and dynamic industrial environment, the operating efficiency of mobile robots is easily affected by many factors, such as task scheduling, path planning, fault diagnosis, energy consumption management, etc., which makes it difficult to continuously optimize system performance and the operation and maintenance costs remain high.

[0004] At present, there have been some research works on the evaluation and optimization of mobile robot operation efficiency. One type of method is based on mathematical modeling and simulation analysis, and the other type of method is based on data-driven performance evaluation. However, the existing methods still lack systematic evaluation of the operation efficiency of the entire life cycle and the generalization ability of the efficiency evaluation model is insufficient, and it cannot adapt to different working conditions and task requirements.

[0005] Therefore, a solution is urgently needed to solve the problems existing in the prior art. Summary of the invention

[0006] The embodiments of the present invention provide a method and system for evaluating the operating efficiency of a mobile robot based on the industrial Internet, which can at least solve some of the problems existing in the prior art.

[0007] A first aspect of an embodiment of the present invention provides a method for evaluating the operating efficiency of a mobile robot based on the industrial Internet, comprising:

[0008] The multimodal operation data of the mobile robot is collected by a heterogeneous sensor array arranged on the mobile robot, and a transmission channel is selected based on the semantic importance and delay sensitivity of the multimodal operation data through an industrial wireless communication protocol based on cognitive radio, and the data is uploaded to an edge computing node in a hierarchical manner. The edge computing node preprocesses the multimodal operation data, completes the information and performs filtering operations in combination with a multimodal autoencoder network, and generates a first semantic feature and a corresponding operation feature descriptor;

[0009] The edge computing node selects the optimal transmission path and service function chain according to the knowledge graph embedding technology and uploads the operation feature descriptor to the cloud server. The cloud server analyzes the operation feature descriptor in real time through the operation efficiency evaluation model, generates an initial operation efficiency evaluation result, and analyzes and compares the historical operation data to determine the efficiency bottleneck and generate corresponding initial optimization improvement suggestions;

[0010] The cloud server sends the initial operating efficiency evaluation result to the edge computing node, combines the knowledge graph reasoning technology and the pre-set expert rule base for semantic enhancement and logical reasoning, and generates a detailed efficiency evaluation report. The edge computing node analyzes the correlation and influence mechanism between different evaluation indicators in the detailed efficiency evaluation report through a causal inference algorithm, determines the key efficiency bottleneck factors, generates a comprehensive optimization suggestion in combination with a multi-objective optimization algorithm and sends it to the mobile robot, collects the operating data of the mobile robot adjusted according to the comprehensive optimization suggestion in real time, and generates a comprehensive efficiency evaluation report.

[0011] In an optional embodiment,

[0012] The multimodal operation data of the mobile robot is collected by a heterogeneous sensor array arranged on the mobile robot. The transmission channel is selected based on the semantic importance and delay sensitivity of the multimodal operation data and the data is uploaded to the edge computing node in stages through the industrial wireless communication protocol based on cognitive radio. The edge computing node preprocesses the multimodal operation data, completes the information and performs filtering operations in combination with the multimodal autoencoder network, and generates a first semantic feature and a corresponding operation feature descriptor, including:

[0013] Collecting multimodal operation data of the mobile robot through a heterogeneous sensor array arranged on the mobile robot, wherein the heterogeneous sensor array includes a laser radar, a camera, an inertial measurement unit and an encoder, and the multimodal operation data includes environmental data, a color image, a linear acceleration, and wheel speed and mileage information corresponding to the mobile robot;

[0014] A semantic analysis module is added to the data link layer of the sensor array to evaluate the semantic importance score corresponding to each frame of multimodal data in real time, and the data with the semantic importance score higher than a preset importance threshold and delay sensitivity is marked as high priority. The cognitive radio controller dynamically adjusts the wireless channel parameters according to the priority of the multimodal data. For high-priority data, a dedicated channel is used and the transmission power is increased. For non-high-priority data, the transmission channel is reused and a preset power control strategy is configured. The marking and transmission are repeated until all the multimodal data are sent to the edge computing node.

[0015] The multimodal data is aligned in time and space by the edge computing node, converted to the robot body coordinate system by the external parameter calibration result, and the multimodal data is organized into semantic data frames in combination with the semantic relevance of the multimodal data. For the semantic data frame of each modality, a multi-layer sparse autoencoder network is set and the dictionary basis corresponding to each modality is determined and completed by back propagation of data reconstruction errors, and data filtering is performed according to the Kalman filter algorithm to generate standard semantic data;

[0016] The standard semantic data is added to a pre-set multimodal fusion network, and the semantic interaction between different modal data is modeled through the graph attention mechanism. The multi-frame semantic features are temporally modeled in combination with the gated recurrent unit to generate spatiotemporal semantic features and combine them with the semantic descriptor generation module. The semantic information is adaptively aggregated in combination with the attention mechanism to generate the first semantic feature and the corresponding running feature descriptor.

[0017] In an optional embodiment,

[0018] Setting up a multi-layer sparse autoencoder network and back-propagating the data reconstruction error to determine the dictionary basis corresponding to each modality is shown in the following formula:

[0019]

[0020] Among them, H j represents the dictionary basis of the jth mode, X j represents the input semantic data frame of the jth modality, W j (L) represents the weight matrix of the Lth layer, represents the bias vector of the Lth layer, represents the hidden layer output of the L-1th layer, λ represents the regularization coefficient, L represents the total number of layers of the multi-layer sparse autoencoder network, f() represents the activation function, and W j (l) represents the weight matrix of the lth layer, represents the bias vector of the lth layer, represents the hidden layer representation of the l-1th layer.

[0021] In an optional embodiment,

[0022] The edge computing node selects the optimal transmission path and service function chain based on the knowledge graph embedding technology and uploads the operation feature descriptor to the cloud server. The cloud server analyzes the operation feature descriptor in real time through the operation efficiency evaluation model, generates initial operation efficiency evaluation results, and analyzes and compares historical operation data to determine efficiency bottlenecks and generate corresponding initial optimization improvement suggestions, including:

[0023] The operation feature descriptors corresponding to the mobile robot are semantically mapped with the edge network topology and service functions as multi-dimensional heterogeneous information, and a knowledge graph containing entities, attributes and relationships is constructed. The knowledge graph is embedded and represented by a graph neural network to generate semantic similarities between different entities. Combined with the service function chain requirements, the operation feature descriptors with the highest similarity and satisfying the optimal transmission path and service function chain combination are determined in the embedding space through a graph search algorithm and uploaded to the cloud server.

[0024] The cloud server maps the historical operation data of the mobile robot to a common semantic space in combination with multi-view representation learning based on the historical operation data of the mobile robot, including historical operation feature descriptors, operation indicators, and manual scores, determines the association relationship between different view data through multi-view contrast learning, constructs an efficiency evaluation neural network based on the association relationship in combination with an attention mechanism, trains the efficiency evaluation neural network based on the historical operation data and constructs a mapping relationship between the operation feature descriptors and the efficiency scores, updates the hyperparameters in the efficiency evaluation neural network in combination with a contrast loss function, repeats the training until the model converges, and obtains the operation efficiency evaluation model;

[0025] Based on the operation efficiency evaluation model, the operation feature descriptor is inferred in real time, the efficiency score of the current operation feature is calculated through forward propagation, and an initial operation efficiency evaluation result is generated. The operation efficiency evaluation model is reversely parsed through a layer-by-layer correlation propagation algorithm to obtain the efficiency score contribution corresponding to each operation feature dimension. Based on the efficiency score contribution, features with contributions greater than the average contribution are found and a key feature subset is generated. Based on the key feature subset and the historical operation data, bottleneck factors are determined through comparative analysis, and the current efficiency bottleneck is determined based on the bottleneck factors, and initial optimization improvement suggestions are generated.

[0026] In an optional embodiment,

[0027] The operation efficiency evaluation model is reversely analyzed by the layer-by-layer correlation propagation algorithm to obtain the efficiency score contribution corresponding to each operation feature dimension as shown in the following formula:

[0028]

[0029] in, represents the efficiency score contribution of the i-th neuron in the m-th layer, α i represents the activation value of the i-th neuron, w ik represents the connection weight from the ith neuron in the mth layer to the kth neuron in the m+1th layer, Represents the efficiency score contribution of the k-th neuron in the m+1-th layer.

[0030] In an optional embodiment,

[0031] The cloud server sends the initial operation efficiency evaluation result to the edge computing node, combines the knowledge graph reasoning technology and the pre-set expert rule base for semantic enhancement and logical reasoning, and generates a refined efficiency evaluation report. The edge computing node analyzes the correlation and influence mechanism between different evaluation indicators in the refined efficiency evaluation report through a causal inference algorithm, determines the key efficiency bottleneck factors, generates a comprehensive optimization suggestion in combination with a multi-objective optimization algorithm and sends it to the mobile robot, collects the operation data of the mobile robot adjusted according to the comprehensive optimization suggestion in real time, and generates a comprehensive efficiency evaluation report including:

[0032] Perform semantic mapping on the initial operating efficiency evaluation results and the conceptual entities in the knowledge graph, determine the evaluation correlation between the evaluation indicators in combination with the ontology reasoning mechanism, and perform logical reasoning on the evaluation correlation in combination with the domain knowledge in the expert knowledge base acquired in advance to obtain the refined efficiency evaluation report;

[0033] The refined efficiency evaluation report is represented as a directed acyclic graph, in which each node corresponds to an evaluation indicator, and the edges between the nodes represent the causal structure between the evaluation indicators. The dependency relationship between the evaluation indicators is determined by a causal inference algorithm based on a graph model. Multiple counterfactual assumptions are set by a counterfactual analysis method. For each indicator, the change in overall efficiency when the values ​​are different is compared to determine the corresponding causal effect. The evaluation indicators are arranged in descending order according to the size of the causal effect, and the evaluation indicator ranked first is selected as the key bottleneck factor;

[0034] For the bottleneck factors, candidate optimization measures are retrieved from the optimization strategy library, and a multi-objective evaluation is performed on the candidate optimization measures. With the goal of maximizing the improvement in operating efficiency and minimizing resource consumption, the solution is performed in combination with the actual constraints of the mobile robot to generate a comprehensive optimization suggestion, wherein the comprehensive optimization suggestion includes sensor parameter adjustment, motion control strategy modification and task replanning. The comprehensive optimization suggestion is sent to the mobile robot and the operating data of the mobile robot after parameter adjustment according to the comprehensive optimization suggestion is collected. The operating data is uploaded to the edge computing node and the optimization effect is evaluated to generate a comprehensive efficiency evaluation report.

[0035] In an optional embodiment,

[0036] The dependency relationship between the evaluation indicators is determined by a causal inference algorithm based on a graphical model as shown in the following formula:

[0037]

[0038] Among them, Y represents the dependency relationship, θ represents the parameter set in the Bayesian network, and v q represents the qth evaluation index, Pa(v q ) represents the qth evaluation index v q The parent node set, logP(v q |Pa(v q )) indicates that under the given parent node set condition, the evaluation index v q The logarithm of the probability, η represents the regularization coefficient, θ qe Represents node v q and its parent node v e The edge weight between them, and n represents the total number of nodes.

[0039] A second aspect of an embodiment of the present invention provides a mobile robot operation efficiency evaluation system based on the industrial Internet, comprising:

[0040] A first unit is used to collect multimodal operation data of the mobile robot through a heterogeneous sensor array arranged on the mobile robot, select a transmission channel based on the industrial wireless communication protocol based on cognitive radio and in combination with the semantic importance and delay sensitivity of the multimodal operation data, and upload the data to an edge computing node in a hierarchical manner, wherein the edge computing node preprocesses the multimodal operation data, performs information completion and filtering operations in combination with a multimodal autoencoder network, and generates a first semantic feature and a corresponding operation feature descriptor;

[0041] The second unit is used for the edge computing node to select the optimal transmission path and service function chain according to the knowledge graph embedding technology and upload the operation feature descriptor to the cloud server. The cloud server analyzes the operation feature descriptor in real time through the operation efficiency evaluation model, generates an initial operation efficiency evaluation result, and analyzes and compares the historical operation data to determine the efficiency bottleneck and generate corresponding initial optimization improvement suggestions;

[0042] The third unit is used for the cloud server to send the initial operation efficiency evaluation result to the edge computing node, combine the knowledge graph reasoning technology and the pre-set expert rule base to perform semantic enhancement and logical reasoning, and generate a detailed efficiency evaluation report. The edge computing node analyzes the correlation and influence mechanism between different evaluation indicators in the detailed efficiency evaluation report through the causal inference algorithm, determines the key efficiency bottleneck factors, generates comprehensive optimization suggestions in combination with the multi-objective optimization algorithm and sends them to the mobile robot, collects the operation data of the mobile robot adjusted according to the comprehensive optimization suggestions in real time, and generates a comprehensive efficiency evaluation report.

[0043] According to a third aspect of the embodiments of the present invention,

[0044] An electronic device is provided, comprising:

[0045] processor;

[0046] a memory for storing processor-executable instructions;

[0047] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0048] A fourth aspect of the embodiments of the present invention is:

[0049] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0050] In the present invention, an industrial wireless communication protocol based on cognitive radio is introduced. According to the semantic importance and delay sensitivity of the data, the transmission channel and priority are adaptively selected. While ensuring the real-time transmission of key data, the communication efficiency and robustness are improved. The edge computing node is used for data preprocessing, feature extraction and semantic enhancement, which reduces the computing load of the cloud and realizes the local processing of data. The knowledge graph embedding technology is integrated to select the service function chain and extract the operation feature descriptor. The prior knowledge is introduced into the efficiency evaluation process, and the semantic enhancement and logical reasoning of the evaluation results are realized. The operation efficiency evaluation model realizes the end-to-end mapping of operation characteristics and efficiency indicators, overcomes the limitations of traditional models that rely on domain knowledge and artificial features. On the basis of the efficiency evaluation report, the causal inference algorithm mines the causal relationship and influence mechanism between different evaluation indicators. By revealing the causes and transmission paths of the efficiency bottleneck, the causal inference provides an important basis for targeted optimization decisions. In summary, the present invention realizes the intelligent, dynamic and closed-loop management of the operation efficiency of the mobile robot, improves the environmental adaptability, autonomous decision-making level and overall efficiency of the mobile robot, and is of great significance for promoting the application and popularization of mobile robots and industrial development. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of a method for evaluating the operating efficiency of a mobile robot based on the industrial Internet according to an embodiment of the present invention;

[0052] Figure 2 This is a structural diagram of a mobile robot operation efficiency evaluation system based on the industrial Internet according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0054] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0055] Figure 1 Schematic diagram of the process of the method for evaluating the operating efficiency of a mobile robot based on the industrial Internet according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0056] S1. Collect multimodal operation data of the mobile robot through a heterogeneous sensor array arranged on the mobile robot, select a transmission channel based on the industrial wireless communication protocol of cognitive radio in combination with the semantic importance and delay sensitivity of the multimodal operation data, and upload the data to the edge computing node in a hierarchical manner, wherein the edge computing node preprocesses the multimodal operation data, performs information completion and filtering operations in combination with a multimodal autoencoder network, and generates a first semantic feature and a corresponding operation feature descriptor;

[0057] The heterogeneous sensor array refers to an array composed of multiple types of sensors, such as visual sensors, lidars, and inertial measurement units. The multimodal operation data refers to multiple types and multiple sources of data collected by the heterogeneous sensor array during the operation of the mobile robot, such as image data, point cloud data, inertial data, etc. The semantic importance refers to the importance of data for understanding the semantics of the environment and completing task objectives. The delay sensitivity refers to the sensitivity of data to transmission and processing delays. The multimodal autoencoder network is a deep learning model for learning multimodal data representation.

[0058] In an optional embodiment,

[0059] The multimodal operation data of the mobile robot is collected by a heterogeneous sensor array arranged on the mobile robot. The transmission channel is selected based on the semantic importance and delay sensitivity of the multimodal operation data and the data is uploaded to the edge computing node in stages through the industrial wireless communication protocol based on cognitive radio. The edge computing node preprocesses the multimodal operation data, completes the information and performs filtering operations in combination with the multimodal autoencoder network, and generates a first semantic feature and a corresponding operation feature descriptor, including:

[0060] Collecting multimodal operation data of the mobile robot through a heterogeneous sensor array arranged on the mobile robot, wherein the heterogeneous sensor array includes a laser radar, a camera, an inertial measurement unit and an encoder, and the multimodal operation data includes environmental data, a color image, a linear acceleration, and wheel speed and mileage information corresponding to the mobile robot;

[0061] A semantic analysis module is added to the data link layer of the sensor array to evaluate the semantic importance score corresponding to each frame of multimodal data in real time, and the data with the semantic importance score higher than a preset importance threshold and delay sensitivity is marked as high priority. The cognitive radio controller dynamically adjusts the wireless channel parameters according to the priority of the multimodal data. For high-priority data, a dedicated channel is used and the transmission power is increased. For non-high-priority data, the transmission channel is reused and a preset power control strategy is configured. The marking and transmission are repeated until all the multimodal data are sent to the edge computing node.

[0062] The multimodal data is aligned in time and space by the edge computing node, converted to the robot body coordinate system by the external parameter calibration result, and the multimodal data is organized into semantic data frames in combination with the semantic relevance of the multimodal data. For the semantic data frame of each modality, a multi-layer sparse autoencoder network is set and the dictionary basis corresponding to each modality is determined and completed by back propagation of data reconstruction errors, and data filtering is performed according to the Kalman filter algorithm to generate standard semantic data;

[0063] The standard semantic data is added to a pre-set multimodal fusion network, and the semantic interaction between different modal data is modeled through the graph attention mechanism. The multi-frame semantic features are temporally modeled in combination with the gated recurrent unit to generate spatiotemporal semantic features and combine them with the semantic descriptor generation module. The semantic information is adaptively aggregated in combination with the attention mechanism to generate the first semantic feature and the corresponding running feature descriptor.

[0064] The data link layer is the second layer in the communication protocol stack, responsible for the reliable transmission of data frames between two adjacent nodes. The semantic analysis module is a component in the mobile robot system responsible for semantic understanding and interpretation of multimodal operation data. The wireless channel parameters are a set of indicators that describe the quality of wireless communication links, such as signal strength, signal-to-noise ratio, bit error rate, etc. The external parameter calibration refers to estimating the relative posture relationship between different sensor coordinate systems, that is, the rotation and translation matrix. The semantic data frame refers to a data unit encapsulated in the data link layer and containing semantic information. The multi-layer sparse autoencoder network is a deep autoencoder model that learns the sparse representation of data by introducing sparsity constraints. The data error reconstruction refers to reconstructing the original input data using the data representation learned by the autoencoder network and calculating the reconstruction error. The dictionary basis is a concept in sparse representation learning, which refers to a set of basis vectors, and their linear combination can be used to represent data samples. The operation feature descriptor is a compact representation of the operation state of the mobile robot, which integrates the feature information of multimodal operation data.

[0065] A heterogeneous sensor array is set up on the mobile robot, including a laser radar, a camera, an inertial measurement unit, and an encoder. The laser radar is used to collect environmental data around the mobile robot to obtain the geometric features and spatial layout information of the environment. The camera is used to collect color image data within the field of view of the mobile robot to obtain the visual features and semantic information of the environment. The inertial measurement unit is used to collect the linear acceleration data of the mobile robot to obtain the motion state information of the robot. The encoder is used to collect the wheel speed and odometer information of the mobile robot to obtain the position and heading information of the robot.

[0066] A semantic analysis module is added to the data link layer of the heterogeneous sensor array to perform real-time semantic analysis on the collected multimodal data. The semantic analysis module uses a pre-trained deep learning model to extract and understand the semantic content of each frame of multimodal data. According to the task requirements and scene characteristics, the semantic importance of each frame of multimodal data is scored, and the data with semantic importance higher than the preset threshold and sensitive to delay is marked as high priority, and the rest of the data is marked as normal priority.

[0067] The cognitive radio controller dynamically adjusts the wireless channel parameters and transmission strategy according to the priority marking of multimodal data. For high-priority data, the controller allocates dedicated channel resources and increases the transmission power to ensure the real-time and reliability of the data. For ordinary priority data, the controller adopts a channel reuse strategy and sets a predetermined power control scheme to save communication resources. The controller repeatedly marks and transmits multimodal data until it confirms that all data has been successfully sent to the edge computing node.

[0068] After receiving the multimodal operation data, the edge computing node first performs spatiotemporal alignment to eliminate the spatiotemporal deviation between different sensor data. The heterogeneous sensor external parameters obtained by pre-calibration are used to convert the different sensor data into a unified robot body coordinate system. According to the semantic relevance of the multimodal data, the spatiotemporal aligned data is reorganized to generate a semantic data frame. For each modal semantic data frame, a corresponding multi-layer sparse autoencoder network model is designed. The sparse autoencoder network extracts high-order features of the semantic data frame through unsupervised learning and learns a compact dictionary-based representation. The data reconstruction error is used as the loss function, and the parameters of the sparse autoencoder network are optimized through the back propagation algorithm. The input semantic data frame is reconstructed using the learned dictionary basis, and the missing or abnormal data is supplemented according to the reconstruction error. The Kalman filter algorithm is applied to the supplemented semantic data frame to further remove noise and generate standardized semantic data.

[0069] The standardized semantic data is input into a pre-designed multimodal fusion network to extract cross-modal semantic features. The multimodal fusion network adopts a graph attention mechanism to model the semantic interactions and associations between data of different modalities. After the graph attention module, a gated recurrent unit is connected to perform temporal modeling on multi-frame semantic features and extract spatiotemporal semantic information. The spatiotemporal semantic features are input into the semantic descriptor generation module, which dynamically aggregates semantic information at different levels in combination with the attention mechanism. The semantic descriptor generation module outputs the first semantic feature of the current operating state of the mobile robot and the corresponding operating feature descriptor.

[0070] In this embodiment, multi-dimensional and multi-scale status data of the mobile robot during its operation is obtained through the collaborative collection of heterogeneous sensor arrays, providing a comprehensive and three-dimensional perception of the robot's operating status, laying a data foundation for subsequent analysis and decision-making. Through semantic importance evaluation, the robot can adaptively adjust the priority of data processing and focus on key signals. The semantic importance-driven cognitive radio transmission control optimizes the communication efficiency of the mobile robot, ensures the real-time and reliability of key data while saving communication resources. In summary, this embodiment realizes comprehensive perception, intelligent understanding, efficient transmission, robust representation, and knowledge accumulation of the mobile robot's operating status, has good generalization and adaptability, and lays a solid data and knowledge foundation for subsequent operation efficiency evaluation and optimization decision-making, and is expected to significantly improve the autonomy, intelligence, and efficiency of the mobile robot.

[0071] In an optional embodiment,

[0072] Setting up a multi-layer sparse autoencoder network and back-propagating the data reconstruction error to determine the dictionary basis corresponding to each modality is shown in the following formula:

[0073]

[0074] Among them, H j represents the dictionary basis of the jth mode, X j represents the input semantic data frame of the jth modality, W j (L) represents the weight matrix of the Lth layer, represents the bias vector of the Lth layer, represents the hidden layer output of the L-1th layer, λ represents the regularization coefficient, L represents the total number of layers of the multi-layer sparse autoencoder network, f() represents the activation function, and W j (l) represents the weight matrix of the lth layer, represents the bias vector of the lth layer, represents the hidden layer representation of the l-1th layer.

[0075] In this embodiment, a multi-layer autoencoder network is used to map high-dimensional semantic data frames to low-dimensional latent spaces, extract the intrinsic structure and key features of the data, reduce the complexity of data processing and storage, and improve computational efficiency. Sparse representation enhances the distinctiveness and robustness of features, and helps to improve the accuracy of semantic understanding and efficiency evaluation. The multi-layer sparse autoencoder network adaptively learns the dictionary base in a data-driven manner and can characterize the intrinsic patterns of the data. In summary, this embodiment can significantly improve the semantic feature quality of mobile robot operation data, and lay a solid foundation for subsequent semantic fusion, operation efficiency evaluation, and optimization decision-making.

[0076] S2. The edge computing node selects the optimal transmission path and service function chain according to the knowledge graph embedding technology and uploads the operation feature descriptor to the cloud server. The cloud server analyzes the operation feature descriptor in real time through the operation efficiency evaluation model, generates an initial operation efficiency evaluation result, and analyzes and compares the historical operation data to determine the efficiency bottleneck and generate corresponding initial optimization improvement suggestions;

[0077] The knowledge graph embedding technology refers to mapping entities and relationships in the knowledge graph into a continuous low-dimensional vector space, so that it is convenient for calculation and application while retaining semantic information. The operation efficiency evaluation model refers to a computational model used to evaluate the efficiency of mobile robots in performing tasks. The service function chain refers to a chain that connects a series of service functions (such as firewalls, load balancing, intrusion detection, etc.) in a certain order to achieve specific functions. The efficiency bottleneck refers to the highest efficiency value that the system can achieve under the current conditions and technical level, and this value is constrained by certain limiting factors.

[0078] In an optional embodiment,

[0079] The edge computing node selects the optimal transmission path and service function chain based on the knowledge graph embedding technology and uploads the operation feature descriptor to the cloud server. The cloud server analyzes the operation feature descriptor in real time through the operation efficiency evaluation model, generates initial operation efficiency evaluation results, and analyzes and compares historical operation data to determine efficiency bottlenecks and generate corresponding initial optimization improvement suggestions, including:

[0080] The operation feature descriptors corresponding to the mobile robot are semantically mapped with the edge network topology and service functions as multi-dimensional heterogeneous information, and a knowledge graph containing entities, attributes and relationships is constructed. The knowledge graph is embedded and represented by a graph neural network to generate semantic similarities between different entities. Combined with the service function chain requirements, the operation feature descriptors with the highest similarity and satisfying the optimal transmission path and service function chain combination are determined in the embedding space through a graph search algorithm and uploaded to the cloud server.

[0081] The cloud server maps the historical operation data of the mobile robot to a common semantic space in combination with multi-view representation learning based on the historical operation data of the mobile robot, including historical operation feature descriptors, operation indicators, and manual scores, determines the association relationship between different view data through multi-view contrast learning, constructs an efficiency evaluation neural network based on the association relationship in combination with an attention mechanism, trains the efficiency evaluation neural network based on the historical operation data and constructs a mapping relationship between the operation feature descriptors and the efficiency scores, updates the hyperparameters in the efficiency evaluation neural network in combination with a contrast loss function, repeats the training until the model converges, and obtains the operation efficiency evaluation model;

[0082] Based on the operation efficiency evaluation model, the operation feature descriptor is inferred in real time, the efficiency score of the current operation feature is calculated through forward propagation, and an initial operation efficiency evaluation result is generated. The operation efficiency evaluation model is reversely parsed through a layer-by-layer correlation propagation algorithm to obtain the efficiency score contribution corresponding to each operation feature dimension. Based on the efficiency score contribution, features with contributions greater than the average contribution are found and a key feature subset is generated. Based on the key feature subset and the historical operation data, bottleneck factors are determined through comparative analysis, and the current efficiency bottleneck is determined based on the bottleneck factors, and initial optimization improvement suggestions are generated.

[0083] The semantic similarity is a measure of the degree of semantic similarity between two words, phrases or entities. The graph search algorithm is an algorithm for traversing or searching for paths or nodes in a graph structure (a data structure composed of nodes and edges). The common semantic space refers to mapping data from different sources or types into a unified representation space so that the data can be directly compared and calculated in this space. The efficiency evaluation neural network is a model based on a neural network for evaluating the efficiency of a system or process. The layer-by-layer relevance propagation algorithm is a method for interpreting the prediction results of a neural network. Through layer-by-layer back propagation, the relevance of the prediction results is assigned to the input features, thereby determining the contribution of each feature to the prediction results. The efficiency score contribution refers to the influence or contribution of each input feature to the overall efficiency score. The bottleneck factor refers to the key factor or link that limits the performance of a system or process.

[0084] Based on the pre-acquired operation feature descriptors, combined with the topological structure of the edge network (such as the connection relationship between nodes) and service functions (such as data processing capacity and storage capacity), multi-dimensional heterogeneous information is formed. The operation feature descriptors, network topology and service functions are mapped into a knowledge graph containing entities, attributes and relationships. Entities include mobile robots, network nodes and service functions, attributes are their respective characteristics, and relationships represent the association between entities. The knowledge graph is embedded and represented by a graph neural network to generate semantic similarity between different entities.

[0085] Combined with the service function chain requirements, the graph search algorithm is used to find the operation feature descriptor with the highest semantic similarity and the optimal transmission path and service function chain combination in the embedding space, and the optimal operation feature descriptor found is uploaded to the cloud server for further processing;

[0086] Collect historical operation data of mobile robots, including historical operation feature descriptors, operation indicators and manual scores, map the historical operation data to a common semantic space, analyze and determine the correlation between data through multi-view representation learning (for example, data fusion from different perspectives), further clarify the correlation between data from different views through multi-view contrast learning, build an efficiency evaluation neural network in combination with the attention mechanism, train the efficiency evaluation neural network based on historical operation data, learn the mapping relationship between operation feature descriptors and efficiency scores, use the contrast loss function to update the hyperparameters of the neural network, repeat the training until the model converges, and obtain the final operation efficiency evaluation model;

[0087] Using the constructed operation efficiency evaluation model, real-time inference is performed on the current operation feature descriptor, and the efficiency score of the current operation feature is calculated through forward propagation to generate the initial operation efficiency evaluation result. The efficiency evaluation model is reversely parsed through the layer-by-layer correlation propagation algorithm to calculate the efficiency score contribution of each operation feature dimension. Based on the efficiency score contribution, features with a contribution greater than the average contribution are found to generate a key feature subset. Based on the key feature subset and historical operation data, the bottleneck factors that limit the current efficiency are determined through comparative analysis. Based on the identified bottleneck factors, the current efficiency bottleneck is determined, initial optimization improvement suggestions are generated, and specific improvement measures are proposed.

[0088] In this embodiment, the operation characteristics, network topology and service functions of the mobile robot are integrated together through semantic mapping and knowledge graph construction to form a structured information representation with semantic association, so that the information between different data sources can be interconnected, which is convenient for subsequent embedded representation learning and analysis. By performing graph search in the embedded space, the operation feature descriptor with the highest semantic similarity and satisfying the optimal transmission path and service function chain combination can be effectively found, which not only improves the accuracy of data matching, but also ensures the efficiency and reliability of the system in actual operation. Multi-view representation learning and comparative learning can make full use of the diversity and complexity of historical operation data. Complexity, identifying and strengthening the correlation between data from different views, improving the depth and breadth of data analysis, and helping to build a more accurate and robust efficiency evaluation model. The combination of real-time reasoning and layer-by-layer correlation propagation algorithm enables the system to quickly and accurately calculate the efficiency score of the current operating features, and clarify the contribution of each feature, providing technical support for quickly identifying and responding to efficiency issues. In summary, this embodiment can achieve a comprehensive evaluation and optimization of the operating efficiency of the mobile robot, ensuring the efficiency, stability and reliability of the system in actual operation, which not only improves the operating performance of the mobile robot, but also provides advanced technical means for the efficiency evaluation and optimization of complex systems.

[0089] In an optional embodiment,

[0090] The operation efficiency evaluation model is reversely analyzed by the layer-by-layer correlation propagation algorithm to obtain the efficiency score contribution corresponding to each operation feature dimension as shown in the following formula:

[0091]

[0092] in, represents the efficiency score contribution of the i-th neuron in the m-th layer, α i represents the activation value of the i-th neuron, w ik represents the connection weight from the ith neuron in the mth layer to the kth neuron in the m+1th layer, Represents the efficiency score contribution of the k-th neuron in the m+1-th layer.

[0093] In this embodiment, through the layer-by-layer correlation propagation algorithm, the complex operation efficiency evaluation model can be effectively analyzed, revealing the contribution and weight of each neuron on the final effect. This helps engineers and decision makers understand the operation mechanism of the model and promote further optimization and improvement of the model. Through accurate contribution analysis, the redundancy of the model and unnecessary calculations can be reduced, and the overall efficiency and performance of the system can be optimized. In summary, this embodiment provides an in-depth model interpretation and optimization path by calculating the contribution of neurons, which significantly improves the efficiency evaluation accuracy and operational practicality of the system.

[0094] S3. The cloud server sends the initial operating efficiency evaluation result to the edge computing node, combines the knowledge graph reasoning technology and the pre-set expert rule base for semantic enhancement and logical reasoning, and generates a detailed efficiency evaluation report. The edge computing node analyzes the correlation and influence mechanism between different evaluation indicators in the detailed efficiency evaluation report through the causal inference algorithm, determines the key efficiency bottleneck factors, generates comprehensive optimization suggestions in combination with the multi-objective optimization algorithm and sends them to the mobile robot, collects the operating data of the mobile robot adjusted according to the comprehensive optimization suggestions in real time, and generates a comprehensive efficiency evaluation report.

[0095] The knowledge graph reasoning technology is an intelligent reasoning method based on graph structure. It makes inferences and decisions by utilizing the associated information of entities, attributes and relationships in the knowledge graph. The expert rule base is a set of rules built based on expert knowledge and experience, which is used to solve problems in specific fields or make decision inferences. The refined efficiency evaluation report is a report that conducts a detailed analysis and evaluation of the efficiency of a system or process, aiming to identify potential efficiency improvement points and optimization directions. The causal inference algorithm is a statistical method used to identify and quantify the causal relationship between variables, rather than just the correlation relationship. The key efficiency bottleneck factor refers to the main cause or limiting factor that affects the efficiency of a system, process or organization.

[0096] In an optional embodiment,

[0097] The cloud server sends the initial operation efficiency evaluation result to the edge computing node, combines the knowledge graph reasoning technology and the pre-set expert rule base for semantic enhancement and logical reasoning, and generates a refined efficiency evaluation report. The edge computing node analyzes the correlation and influence mechanism between different evaluation indicators in the refined efficiency evaluation report through a causal inference algorithm, determines the key efficiency bottleneck factors, generates a comprehensive optimization suggestion in combination with a multi-objective optimization algorithm and sends it to the mobile robot, collects the operation data of the mobile robot adjusted according to the comprehensive optimization suggestion in real time, and generates a comprehensive efficiency evaluation report including:

[0098] Perform semantic mapping on the initial operating efficiency evaluation results and the conceptual entities in the knowledge graph, determine the evaluation correlation between the evaluation indicators in combination with the ontology reasoning mechanism, and perform logical reasoning on the evaluation correlation in combination with the domain knowledge in the expert knowledge base acquired in advance to obtain the refined efficiency evaluation report;

[0099] The refined efficiency evaluation report is represented as a directed acyclic graph, in which each node corresponds to an evaluation indicator, and the edges between the nodes represent the causal structure between the evaluation indicators. The dependency relationship between the evaluation indicators is determined by a causal inference algorithm based on a graph model. Multiple counterfactual assumptions are set by a counterfactual analysis method. For each indicator, the change in overall efficiency when the values ​​are different is compared to determine the corresponding causal effect. The evaluation indicators are arranged in descending order according to the size of the causal effect, and the evaluation indicator ranked first is selected as the key bottleneck factor;

[0100] For the bottleneck factors, candidate optimization measures are retrieved from the optimization strategy library, and a multi-objective evaluation is performed on the candidate optimization measures. With the goal of maximizing the improvement in operating efficiency and minimizing resource consumption, the solution is performed in combination with the actual constraints of the mobile robot to generate a comprehensive optimization suggestion, wherein the comprehensive optimization suggestion includes sensor parameter adjustment, motion control strategy modification and task replanning. The comprehensive optimization suggestion is sent to the mobile robot and the operating data of the mobile robot after parameter adjustment according to the comprehensive optimization suggestion is collected. The operating data is uploaded to the edge computing node and the optimization effect is evaluated to generate a comprehensive efficiency evaluation report.

[0101] The ontology reasoning mechanism refers to an ontology-based knowledge representation and reasoning system, which is used to infer new knowledge or relationships from known facts. The evaluation correlation describes the degree of association or mutual influence between different evaluation indicators. The directed acyclic graph is a graph structure in which the edges in the graph are directed and there is no path between any nodes that can form a loop. It is often used to represent and analyze causal relationships, dependencies or process control, etc. The causal structure refers to the organization and structuring of causal relationships between variables. The counterfactual analysis method is a statistical analysis technique for inferring causal effects. By comparing observed data under different situations, it attempts to estimate the changes in other variables that will occur if a certain variable changes. The causal effect refers to how a change in one variable (cause) leads to a change in another variable (result).

[0102] Semantically map the initial operating efficiency evaluation results with the conceptual entities in the knowledge graph, use the ontology reasoning mechanism, combine with the domain knowledge base, determine the evaluation correlation between the evaluation indicators, and use the concepts, attributes and relationships defined in the ontology to infer the logical relationship and semantic connection between different evaluation indicators;

[0103] Based on the pre-acquired expert knowledge base, logical reasoning is performed to analyze the correlation between evaluation indicators to obtain a detailed efficiency evaluation report. Each node in the report corresponds to an evaluation indicator. The directed edges between nodes represent the causal structure. The detailed efficiency evaluation report is represented as a directed acyclic graph. Through the causal inference algorithm based on the graph model, the dependency between the evaluation indicators is analyzed to determine the degree of influence of each evaluation indicator on the overall efficiency, and the evaluation indicators are sorted according to the causal effect. Using the counterfactual analysis method, multiple counterfactual assumptions are set to compare the overall efficiency changes under different situations, determine the size of the causal effect of each indicator, and sort the evaluation indicators in descending order according to the size of the causal effect to determine the key bottleneck factors.

[0104] For the identified bottleneck factors, relevant candidate optimization measures are retrieved from the optimization strategy library and multi-objective evaluation is performed. The goal is to maximize the improvement in operating efficiency and minimize resource consumption. Combined with the actual constraints of the mobile robot, comprehensive optimization suggestions are generated, including suggestions on sensor parameter adjustment, motion control strategy modification, and task replanning. The comprehensive optimization suggestions are sent to the mobile robot, and the operation data is collected after the robot implements the comprehensive optimization suggestions. The operation data of the mobile robot after implementing the comprehensive optimization suggestions are collected and uploaded to the edge computing node for evaluation of the optimization effect. Based on the collected data, a comprehensive efficiency evaluation report is generated to evaluate the actual effect and improvement of the optimization measures.

[0105] In this embodiment, semantic mapping is performed between conceptual entities in the knowledge graph and the initial operating efficiency evaluation results, and the ontology reasoning mechanism is used to infer the correlation between evaluation indicators, so as to accurately capture the relationship and influence between different evaluation indicators, thereby providing an in-depth theoretical basis and data support for subsequent efficiency evaluation. Based on the directed acyclic graph and causal inference algorithm, the causal relationship between the evaluation indicators is systematically analyzed. By setting multiple counterfactual assumptions and comparing the overall efficiency changes under different situations, the specific contribution of each evaluation indicator to the overall efficiency is determined. It not only provides a profound causal understanding, but also can accurately identify the key bottleneck factors affecting efficiency, and provide a scientific basis for the formulation of optimization measures. The operating data of the mobile robot after the implementation of the optimization suggestions is collected and analyzed through the edge computing node to generate a comprehensive efficiency evaluation report. The detailed evaluation and display of the actual effects of the optimization measures can not only quantify the optimization effects, but also provide decision support and reference basis for future decisions. In summary, this embodiment can effectively identify, analyze and solve key efficiency problems in the operation of mobile robots, and provide scientific and reliable technical support and method guidance for improving operating efficiency and performance.

[0106] In an optional embodiment,

[0107] The dependency relationship between the evaluation indicators is determined by a causal inference algorithm based on a graphical model as shown in the following formula:

[0108]

[0109] Among them, Y represents the dependency relationship, θ represents the parameter set in the Bayesian network, and v q represents the qth evaluation index, Pa(v q ) represents the qth evaluation index v q The parent node set, logP(v q |Pa(v q )) indicates that under the given parent node set condition, the evaluation index v q The logarithm of the probability, η represents the regularization coefficient, θ qe Represents node v q and its parent node v e The edge weight between them, and n represents the total number of nodes.

[0110] In this embodiment, by maximizing the log-likelihood or Bayesian probability, it is ensured that the model can accurately capture the conditional dependency between the evaluation indicators. By adjusting the regularization coefficient, the fitting ability and generalization ability of the model can be balanced to avoid overfitting, while promoting the learned Bayesian network structure to be more sparse, thereby making the model easier to interpret and generalize to new data. By minimizing the dependency strength, the structure between the nodes in the Bayesian network can be obtained, thereby revealing the causal relationship between the evaluation indicators. In summary, this embodiment can effectively analyze and determine the dependency between the evaluation indicators, can accurately understand and quantify the impact between the evaluation indicators, and provide a scientific basis and data support for subsequent efficiency optimization and decision-making.

[0111] Figure 2 FIG. 1 is a schematic diagram of the structure of a mobile robot operation efficiency evaluation system based on the industrial Internet according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0112] A first unit is used to collect multimodal operation data of the mobile robot through a heterogeneous sensor array arranged on the mobile robot, select a transmission channel based on the industrial wireless communication protocol based on cognitive radio and in combination with the semantic importance and delay sensitivity of the multimodal operation data, and upload the data to an edge computing node in a hierarchical manner, wherein the edge computing node preprocesses the multimodal operation data, performs information completion and filtering operations in combination with a multimodal autoencoder network, and generates a first semantic feature and a corresponding operation feature descriptor;

[0113] The second unit is used for the edge computing node to select the optimal transmission path and service function chain according to the knowledge graph embedding technology and upload the operation feature descriptor to the cloud server. The cloud server analyzes the operation feature descriptor in real time through the operation efficiency evaluation model, generates an initial operation efficiency evaluation result, and analyzes and compares the historical operation data to determine the efficiency bottleneck and generate corresponding initial optimization improvement suggestions;

[0114] The third unit is used for the cloud server to send the initial operation efficiency evaluation result to the edge computing node, combine the knowledge graph reasoning technology and the pre-set expert rule base to perform semantic enhancement and logical reasoning, and generate a detailed efficiency evaluation report. The edge computing node analyzes the correlation and influence mechanism between different evaluation indicators in the detailed efficiency evaluation report through the causal inference algorithm, determines the key efficiency bottleneck factors, generates comprehensive optimization suggestions in combination with the multi-objective optimization algorithm and sends them to the mobile robot, collects the operation data of the mobile robot adjusted according to the comprehensive optimization suggestions in real time, and generates a comprehensive efficiency evaluation report.

[0115] According to a third aspect of the embodiments of the present invention,

[0116] An electronic device is provided, comprising:

[0117] processor;

[0118] a memory for storing processor-executable instructions;

[0119] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0120] A fourth aspect of the embodiments of the present invention is:

[0121] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0122] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A mobile robot operation efficiency evaluation method based on the industrial Internet, characterized in that: include: The multimodal operation data of the mobile robot is collected by a heterogeneous sensor array arranged on the mobile robot, and a transmission channel is selected based on the semantic importance and delay sensitivity of the multimodal operation data through an industrial wireless communication protocol based on cognitive radio, and the data is uploaded to an edge computing node in a hierarchical manner. The edge computing node preprocesses the multimodal operation data, completes the information and performs filtering operations in combination with a multimodal autoencoder network, and generates a first semantic feature and a corresponding operation feature descriptor; The edge computing node selects the optimal transmission path and service function chain according to the knowledge graph embedding technology and uploads the operation feature descriptor to the cloud server. The cloud server analyzes the operation feature descriptor in real time through the operation efficiency evaluation model, generates an initial operation efficiency evaluation result, and analyzes and compares the historical operation data to determine the efficiency bottleneck and generate corresponding initial optimization improvement suggestions; The cloud server sends the initial operating efficiency evaluation result to the edge computing node, combines the knowledge graph reasoning technology and the pre-set expert rule base for semantic enhancement and logical reasoning, and generates a detailed efficiency evaluation report. The edge computing node analyzes the correlation and influence mechanism between different evaluation indicators in the detailed efficiency evaluation report through a causal inference algorithm, determines the key efficiency bottleneck factors, generates a comprehensive optimization suggestion in combination with a multi-objective optimization algorithm and sends it to the mobile robot, collects the operating data of the mobile robot adjusted according to the comprehensive optimization suggestion in real time, and generates a comprehensive efficiency evaluation report.

2. The method according to claim 1, characterized in that The multimodal operation data of the mobile robot is collected by a heterogeneous sensor array arranged on the mobile robot. The transmission channel is selected based on the semantic importance and delay sensitivity of the multimodal operation data and the data is uploaded to the edge computing node in stages through the industrial wireless communication protocol based on cognitive radio. The edge computing node preprocesses the multimodal operation data, completes the information and performs filtering operations in combination with the multimodal autoencoder network, and generates a first semantic feature and a corresponding operation feature descriptor, including: Collecting multimodal operation data of the mobile robot through a heterogeneous sensor array arranged on the mobile robot, wherein the heterogeneous sensor array includes a laser radar, a camera, an inertial measurement unit and an encoder, and the multimodal operation data includes environmental data, a color image, a linear acceleration, and wheel speed and mileage information corresponding to the mobile robot; A semantic analysis module is added to the data link layer of the sensor array to evaluate the semantic importance score corresponding to each frame of multimodal data in real time, and the data with the semantic importance score higher than a preset importance threshold and delay sensitivity is marked as high priority. The cognitive radio controller dynamically adjusts the wireless channel parameters according to the priority of the multimodal data. For high-priority data, a dedicated channel is used and the transmission power is increased. For non-high-priority data, the transmission channel is reused and a preset power control strategy is configured. The marking and transmission are repeated until all the multimodal data are sent to the edge computing node. The multimodal data is aligned in time and space by the edge computing node, converted to the robot body coordinate system by the external parameter calibration result, and the multimodal data is organized into semantic data frames in combination with the semantic relevance of the multimodal data. For the semantic data frame of each modality, a multi-layer sparse autoencoder network is set and the dictionary basis corresponding to each modality is determined and completed by back propagation of data reconstruction errors, and data filtering is performed according to the Kalman filter algorithm to generate standard semantic data; The standard semantic data is added to a pre-set multimodal fusion network, and the semantic interaction between different modal data is modeled through the graph attention mechanism. The multi-frame semantic features are temporally modeled in combination with the gated recurrent unit to generate spatiotemporal semantic features and combine them with the semantic descriptor generation module. The semantic information is adaptively aggregated in combination with the attention mechanism to generate the first semantic feature and the corresponding running feature descriptor.

3. The method according to claim 2, characterized in that Setting up a multi-layer sparse autoencoder network and back-propagating the data reconstruction error to determine the dictionary basis corresponding to each modality is shown in the following formula: Among them, H j represents the dictionary basis of the jth mode, X j represents the input semantic data frame of the jth modality, represents the weight matrix of the Lth layer, represents the bias vector of the Lth layer, represents the hidden layer output of the L-1th layer, λ represents the regularization coefficient, L represents the total number of layers of the multi-layer sparse autoencoder network, f() represents the activation function, represents the weight matrix of the lth layer, represents the bias vector of the lth layer, represents the hidden layer representation of the l-1th layer.

4. The method according to claim 1, characterized in that: The edge computing node selects the optimal transmission path and service function chain based on the knowledge graph embedding technology and uploads the operation feature descriptor to the cloud server. The cloud server analyzes the operation feature descriptor in real time through the operation efficiency evaluation model, generates initial operation efficiency evaluation results, and analyzes and compares historical operation data to determine efficiency bottlenecks and generate corresponding initial optimization improvement suggestions, including: The operation feature descriptors corresponding to the mobile robot are semantically mapped with the edge network topology and service functions as multi-dimensional heterogeneous information, and a knowledge graph containing entities, attributes and relationships is constructed. The knowledge graph is embedded and represented by a graph neural network to generate semantic similarities between different entities. Combined with the service function chain requirements, the operation feature descriptors with the highest similarity and satisfying the optimal transmission path and service function chain combination are determined in the embedding space through a graph search algorithm and uploaded to the cloud server. The cloud server maps the historical operation data of the mobile robot to a common semantic space in combination with multi-view representation learning based on the historical operation data of the mobile robot, including historical operation feature descriptors, operation indicators, and manual scores, determines the association relationship between different view data through multi-view contrast learning, constructs an efficiency evaluation neural network based on the association relationship in combination with an attention mechanism, trains the efficiency evaluation neural network based on the historical operation data and constructs a mapping relationship between the operation feature descriptors and the efficiency scores, updates the hyperparameters in the efficiency evaluation neural network in combination with a contrast loss function, repeats the training until the model converges, and obtains the operation efficiency evaluation model; Based on the operation efficiency evaluation model, the operation feature descriptor is inferred in real time, the efficiency score of the current operation feature is calculated through forward propagation, and an initial operation efficiency evaluation result is generated. The operation efficiency evaluation model is reversely parsed through a layer-by-layer correlation propagation algorithm to obtain the efficiency score contribution corresponding to each operation feature dimension. Based on the efficiency score contribution, features with contributions greater than the average contribution are found and a key feature subset is generated. Based on the key feature subset and the historical operation data, bottleneck factors are determined through comparative analysis, and the current efficiency bottleneck is determined based on the bottleneck factors, and initial optimization improvement suggestions are generated.

5. The method according to claim 4, characterized in that The operation efficiency evaluation model is reversely analyzed by the layer-by-layer correlation propagation algorithm to obtain the efficiency score contribution corresponding to each operation feature dimension as shown in the following formula: in, represents the efficiency score contribution of the i-th neuron in the m-th layer, α i represents the activation value of the i-th neuron, w ik represents the connection weight from the ith neuron in the mth layer to the kth neuron in the m+1th layer, Represents the efficiency score contribution of the k-th neuron in the m+1-th layer.

6. The method according to claim 1, characterized in that The cloud server sends the initial operation efficiency evaluation result to the edge computing node, combines the knowledge graph reasoning technology and the pre-set expert rule base for semantic enhancement and logical reasoning, and generates a refined efficiency evaluation report. The edge computing node analyzes the correlation and influence mechanism between different evaluation indicators in the refined efficiency evaluation report through a causal inference algorithm, determines the key efficiency bottleneck factors, generates a comprehensive optimization suggestion in combination with a multi-objective optimization algorithm and sends it to the mobile robot, collects the operation data of the mobile robot adjusted according to the comprehensive optimization suggestion in real time, and generates a comprehensive efficiency evaluation report including: Perform semantic mapping on the initial operating efficiency evaluation results and the conceptual entities in the knowledge graph, determine the evaluation correlation between the evaluation indicators in combination with the ontology reasoning mechanism, and perform logical reasoning on the evaluation correlation in combination with the domain knowledge in the expert knowledge base acquired in advance to obtain the refined efficiency evaluation report; The refined efficiency evaluation report is represented as a directed acyclic graph, in which each node corresponds to an evaluation indicator, and the edges between the nodes represent the causal structure between the evaluation indicators. The dependency relationship between the evaluation indicators is determined by a causal inference algorithm based on a graph model. Multiple counterfactual assumptions are set by a counterfactual analysis method. For each indicator, the change in overall efficiency when the values ​​are different is compared to determine the corresponding causal effect. The evaluation indicators are arranged in descending order according to the size of the causal effect, and the evaluation indicator ranked first is selected as the key bottleneck factor; For the bottleneck factors, candidate optimization measures are retrieved from the optimization strategy library, and a multi-objective evaluation is performed on the candidate optimization measures. With the goal of maximizing the improvement in operating efficiency and minimizing resource consumption, the solution is performed in combination with the actual constraints of the mobile robot to generate a comprehensive optimization suggestion, wherein the comprehensive optimization suggestion includes sensor parameter adjustment, motion control strategy modification and task replanning. The comprehensive optimization suggestion is sent to the mobile robot and the operating data of the mobile robot after parameter adjustment according to the comprehensive optimization suggestion is collected. The operating data is uploaded to the edge computing node and the optimization effect is evaluated to generate a comprehensive efficiency evaluation report.

7. The method according to claim 6, characterized in that The dependency relationship between the evaluation indicators is determined by a causal inference algorithm based on a graphical model as shown in the following formula: Among them, Y represents the dependency relationship, θ represents the parameter set in the Bayesian network, and v q represents the qth evaluation index, Pa(v q ) represents the qth evaluation index v q The parent node set, logP(v q |Pa(v q )) indicates that under the given parent node set condition, the evaluation index v q The logarithm of the probability, η represents the regularization coefficient, θ qe Represents node v q and its parent node v e The edge weight between them, and n represents the total number of nodes.

8. A mobile robot comprehensive operation efficiency evaluation system based on the industrial Internet, used to implement the method described in any one of claims 1 to 7, characterized in that: include: A first unit is used to collect multimodal operation data of the mobile robot through a heterogeneous sensor array arranged on the mobile robot, select a transmission channel based on the industrial wireless communication protocol based on cognitive radio and in combination with the semantic importance and delay sensitivity of the multimodal operation data, and upload the data to an edge computing node in a hierarchical manner, wherein the edge computing node preprocesses the multimodal operation data, performs information completion and filtering operations in combination with a multimodal autoencoder network, and generates a first semantic feature and a corresponding operation feature descriptor; The second unit is used for the edge computing node to select the optimal transmission path and service function chain according to the knowledge graph embedding technology and upload the operation feature descriptor to the cloud server. The cloud server analyzes the operation feature descriptor in real time through the operation efficiency evaluation model, generates an initial operation efficiency evaluation result, and analyzes and compares the historical operation data to determine the efficiency bottleneck and generate corresponding initial optimization improvement suggestions; The third unit is used for the cloud server to send the initial operation efficiency evaluation result to the edge computing node, combine the knowledge graph reasoning technology and the pre-set expert rule base to perform semantic enhancement and logical reasoning, and generate a detailed efficiency evaluation report. The edge computing node analyzes the correlation and influence mechanism between different evaluation indicators in the detailed efficiency evaluation report through the causal inference algorithm, determines the key efficiency bottleneck factors, generates comprehensive optimization suggestions in combination with the multi-objective optimization algorithm and sends them to the mobile robot, collects the operation data of the mobile robot adjusted according to the comprehensive optimization suggestions in real time, and generates a comprehensive efficiency evaluation report.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images