An Asset Data Collection Method Based on the Internet of Things

Abstract

The present application provides an asset data collection method based on the Internet of Things, including: for each of the node groups, constructing a directed acyclic graph model based on task decomposition and dependency relationships, using the critical path analysis algorithm to determine the task execution order and data transfer method among the nodes, and obtaining a task-oriented node collaboration process; for the data fusion result, using the incremental learning algorithm to dynamically update the data model and decision rules of the nodes, using the anomaly detection algorithm to timely detect the abnormal deviation of the node data, and triggering the dynamic adjustment of the node group; constructing a node collaboration incentive mechanism based on the blockchain, dynamically adjusting the credit value and resource allocation of the nodes according to the contribution degree and task completion quality of the nodes, and using the smart contract technology to constrain and arbitrate the collaboration behavior among the nodes.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an asset data collection method based on the Internet of Things. Background Art

[0002] In complex asset monitoring scenarios, sensor nodes need to work together to complete data collection, fusion and reporting tasks. How to dynamically group nodes and assign tasks according to the characteristics of the monitored objects and business needs to achieve collaboration and optimization between nodes is a key issue. Collaborative grouping needs to consider factors such as node capabilities, locations, and energy consumption, and use intelligent algorithms to achieve autonomous organization and scheduling of nodes to improve the efficiency and accuracy of data collection. At the same time, it is also necessary to reduce energy consumption and communication overhead to ensure the stable operation of the data collection system. Summary of the invention

[0003] The present invention provides an asset data collection method based on the Internet of Things, which mainly includes:

[0004] Acquire the capability, location and energy consumption attribute information of the sensor nodes, perform cluster analysis on the attribute information using a clustering algorithm, obtain a subset of nodes with similar attributes, determine the role and task of each subset of nodes in the collaborative grouping, and obtain a node grouping solution driven by business needs;

[0005] A distributed optimization algorithm is used to comprehensively consider the communication cost, energy consumption balance and data redundancy factors between the nodes, and the node grouping scheme is dynamically adjusted to obtain a globally optimal node cooperation scheme. If the node capability or network topology changes, the re-optimization of the grouping scheme is triggered;

[0006] For each node grouping, a directed acyclic graph model based on task decomposition and dependency relationships is constructed, and a critical path analysis algorithm is used to determine the task execution order and data transmission method between the nodes to obtain a task-oriented node collaboration process;

[0007] In the process of node collaboration, a distributed data fusion algorithm is used to pre-process, extract features and semantically map the heterogeneous data collected by the nodes to obtain a unified data representation form, and a multi-sensor data association algorithm is used to mine the spatiotemporal correlation between data from different nodes to achieve cross-node data fusion;

[0008] Based on the data fusion results, an incremental learning algorithm is used to dynamically update the data model and decision rules of the node, and an abnormality detection algorithm is used to timely discover abnormal deviations of the node data, triggering dynamic adjustment of the node grouping;

[0009] In the data reporting stage, an adaptive data compression algorithm is adopted to dynamically adjust the data compression ratio according to the correlation and redundancy of the data. A multi-path routing algorithm is used to optimize the data reporting path by comprehensively considering factors such as link quality, energy consumption balance, and load balance;

[0010] A node cooperation incentive mechanism based on blockchain is constructed. According to the contribution degree and task completion quality of the nodes, the credit value and resource allocation of the nodes are dynamically adjusted. Smart contract technology is used to constrain and arbitrate the cooperation behavior between the nodes.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0012] The present invention discloses an asset data collection method based on the Internet of Things. The method first performs attribute clustering and role assignment on sensor nodes to form a node grouping driven by services. Then, a distributed optimization algorithm is used to dynamically adjust the grouping scheme by comprehensively considering factors such as communication overhead and energy consumption balance. For each grouping, a task dependency graph model is constructed and the cooperation process is determined. In the cooperation process, distributed data fusion and multi-sensor association algorithms are used to process heterogeneous data. At the same time, incremental learning and anomaly detection algorithms are used to dynamically update the data model and trigger grouping adjustment. In the data reporting stage, adaptive compression and multi-path routing algorithms are used to optimize the transmission. Finally, an incentive mechanism based on blockchain is introduced to dynamically adjust resource allocation according to node contributions. The present invention realizes the intelligent cooperation and resource optimization configuration of the sensor network, and improves the overall performance and reliability of the system. Brief Description of the Drawings

[0013] Figure 1 It is a flowchart of an asset data collection method based on the Internet of Things of the present invention.

[0014] Figure 2 It is a schematic diagram of an asset data collection method based on the Internet of Things of the present invention.

[0015] Figure 3 It is another schematic diagram of an asset data collection method based on the Internet of Things of the present invention. Detailed Embodiments

[0016] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] As Figures 1-3 , a specific asset data collection method based on the Internet of Things in this embodiment may include:

[0018] S101. Acquire the attribute information of sensor nodes, such as capability, location and energy consumption, and use intelligent algorithms to perform cluster analysis on the attribute information to obtain a subset of nodes with similar attributes; for each subset of nodes, determine its role and task in the collaborative grouping to obtain a node grouping solution driven by business needs.

[0019] The sensor node's attribute information such as capability, location and energy consumption is obtained. The K-means clustering algorithm is used to perform cluster analysis on the obtained attribute information, and nodes with similar attributes are divided into multiple subsets. For each node subset, the role of each subset in the collaborative grouping is determined by analyzing its attributes such as capability and location, and corresponding tasks are assigned to each role according to business needs, so as to obtain a node grouping scheme driven by business needs. According to the node grouping scheme, the Ql earning reinforcement learning algorithm is used to train the nodes, with the optimization goal of maximizing the task completion quality and node collaboration efficiency, so that each node can effectively collaborate with other nodes according to its own role and task. In the process of node execution, the energy consumption of each node is monitored in real time, and the energy consumption threshold is dynamically set using the control chart method. When the energy consumption of a node exceeds the threshold, the node is marked as a regrouping node, triggering the node regrouping process. According to the current node attribute information and business needs, the genetic algorithm is used to optimize the node grouping scheme. With the minimum node energy consumption and the maximum task completion quality as the optimization goals, the regrouped nodes are redistributed and the roles are adjusted to achieve dynamic optimization of node grouping. For the regrouped node subset, the instance-based transfer learning algorithm is used to select the source node with the highest task similarity with the target node as the knowledge source, and some training parameters and model structures of the source node are migrated to the target node to accelerate the learning and convergence speed of the target node. In the process of node collaboration, the node collaboration relationship graph is constructed by analyzing the communication data between nodes. The graph convolutional neural network algorithm is used to train the relationship graph with the optimization goal of maximizing the node collaboration efficiency. The attribute information and neighbor information of the nodes are aggregated through graph convolution operations, and the implicit collaboration relationship between nodes is mined to guide the nodes to dynamically adjust the collaboration strategy. Continuously monitor changes in business needs, set business demand change detection rules, compare the gap between current business indicators and expected goals, and when the gap exceeds the preset threshold, trigger a business demand change event and re-trigger the node grouping process. Based on the new business needs and node attributes, use the above method to regenerate the node grouping plan to achieve business-driven dynamic optimization of node grouping.

[0020] Exemplarily, after obtaining the attribute information of the sensor node such as capability, location and energy consumption, the K-means clustering algorithm is used to divide the nodes into several subsets. Among them, the K value can be set according to the number of nodes and business requirements, for example, 100 nodes are divided into 5 subsets. In the clustering process, the capability, location and energy consumption of the node are selected as feature vectors, and the similarity between nodes is measured by Euclidean distance. The nodes are divided into the nearest cluster center by iterative optimization until the mean square error of the nodes in the cluster is minimized. For each node subset, its capabilities, location and other attributes are analyzed to determine its role in the collaborative grouping. For example, nodes with strong computing power and close to the target area are divided into "execution nodes" responsible for executing specific tasks; nodes with low energy consumption and wide distribution of locations are divided into "relay nodes" responsible for forwarding and aggregation of data. Corresponding tasks are assigned to each role according to business needs, for example, target tracking tasks are assigned to "execution nodes" and data aggregation tasks are assigned to "relay nodes". The Ql earning reinforcement learning algorithm is used to train the nodes. The task completion quality and node collaboration efficiency are set as the reward function. Through continuous trial and error and learning, each node can effectively collaborate with other nodes according to its own role and task. In the process of node execution, the energy consumption of each node is monitored in real time, and the energy consumption threshold is dynamically set using the control chart method. For example, the upper and lower control limits of node energy consumption are determined based on historical data. When the energy consumption of a node exceeds the upper control limit, it is marked as a regrouping node. The genetic algorithm is used to optimize the node grouping scheme. The fitness function is the minimum node energy consumption and the maximum task completion quality. The regrouping nodes are redistributed and the roles are adjusted through selection, crossover, mutation and other operations to achieve dynamic optimization of node grouping. For the node subset after regrouping, the instance-based transfer learning algorithm is used to select the source node with the highest similarity to the target node task as the knowledge source. For example, the node that has performed similar target tracking tasks is selected as the knowledge source, and some of its training parameters and model structure are transferred to the target node to accelerate the learning speed and convergence speed of the target node. In the process of node collaboration, the node collaboration relationship graph is constructed by analyzing the communication data between nodes. The graph convolutional neural network algorithm is used to maximize the node collaboration efficiency as the loss function, and the attribute information and neighbor information of the node are aggregated through the graph convolution operation. For example, the node's capabilities, location and other attributes and the task execution status of the neighboring nodes are used as the node's features. The hidden layer representation of the node is obtained through graph convolution, and the implicit collaboration relationship between nodes is mined to guide the nodes to dynamically adjust the collaboration strategy. Continuously monitor the changes in business needs, and compare the gap between current business indicators and expected goals by setting business demand change detection rules.For example, when the task completion time exceeds 20% of the expected target, a business requirement change event is triggered, and the node grouping process is restarted. According to the new business requirements and node attributes, the node grouping scheme is regenerated using the above method to achieve dynamic optimization of node grouping driven by business.

[0021] S102. Using a distributed optimization algorithm, comprehensively considering factors such as the communication cost, energy consumption balance, and data redundancy among the nodes, dynamically adjust the node grouping scheme to obtain a globally optimal node cooperation scheme; if the node capabilities or network topology change, trigger the re-optimization of the grouping scheme.

[0022] Obtain the resource information of each node in the network, such as computing power, storage capacity, communication bandwidth, etc., as well as the network topology structure information such as the connection relationship and link quality between nodes. Preprocess the obtained node resource information and network topology structure information, extract key feature parameters, and construct a node resource attribute matrix and a network topology adjacency matrix. Based on the node resource attribute matrix and the network topology adjacency matrix, establish a mathematical model for node grouping optimization. In this model, set optimization objectives such as minimizing the communication cost between nodes, balancing node energy consumption, and minimizing data redundancy, as well as conditions such as node resource capacity limitations and network connectivity constraints. Use the particle swarm optimization algorithm to search for the optimal node grouping scheme on the premise of meeting the optimization objectives and constraint conditions. Through iterative optimization, continuously update the particle positions and velocities, and finally converge to obtain the globally optimal node grouping and cooperation scheme. According to the optimized node grouping scheme, generate configuration instructions for node grouping and cooperation, and send the configuration instructions to the corresponding nodes. After each node receives the configuration instructions, adjust its own working mode and cooperation relationship according to the instructions to form a dynamically optimized node grouping. Deploy monitoring agent programs for node resources and link quality to collect the resource usage and link status information of each node in real time. When the node resource utilization rate exceeds the preset threshold or the link quality is lower than the preset threshold, it is determined that the node capabilities or network topology have changed. When the monitoring program determines that the node capabilities or network topology have changed, trigger the parameter update and re-solving of the node grouping optimization model. According to the solution result of the updated model, generate a new node grouping scheme, send the updated configuration instructions, and achieve dynamic adjustment of node grouping to continuously maintain the globally optimal node cooperation state.

[0023] Exemplarily, after obtaining the resource information and network topology structure information of each node in the network, the data is preprocessed. Key resource parameters such as node CPU frequency, memory size, and storage capacity are extracted to construct a node resource attribute matrix. For example, for a network containing 100 nodes, a 100×5 matrix can be constructed, where each row represents the 5 resource attribute values of a node. At the same time, the connection relationship between nodes and parameters such as link bandwidth and delay are extracted to construct a network topology adjacency matrix, which is used to represent the connectivity and communication cost between nodes. Based on the attribute matrix and the adjacency matrix, a node grouping optimization model is established. Optimization objective functions such as minimizing communication cost, balancing node energy consumption, and reducing data redundancy are set. At the same time, considering the node resource capacity limit and network connectivity constraints, a multi-objective optimization problem is formed. The particle swarm algorithm is used to solve the problem. A group of particles is initialized, and each particle represents a possible node grouping scheme. During the iteration process, the fitness function is used to evaluate the quality of each particle. The particle velocity and position are updated according to the particle's historical optimal position and the global optimal position until the algorithm converges or reaches the maximum number of iterations. The optimized node grouping scheme is converted into configuration instructions and sent to each node. The node adjusts its working mode according to the instructions, establishes a cooperation relationship with the specified node, and forms a dynamically optimized grouping structure. At the same time, a monitoring agent program is deployed to collect information such as node CPU utilization rate, memory usage rate, and link bandwidth occupancy rate in real time. When the CPU utilization rate of a certain node exceeds 80% for 5 consecutive minutes or the link bandwidth occupancy rate exceeds 70% for 5 consecutive minutes, it is determined that the node resources or network topology have changed, triggering the parameter update and re-solving of the grouping optimization model. The model calculates a new optimal node grouping scheme based on the latest resource attribute matrix and adjacency matrix, generates updated configuration instructions, and sends them to each node, so that the node grouping scheme can adapt to network dynamic changes and continuously maintain the optimal state.

[0024] S103. For each of the node groupings, construct a directed acyclic graph model based on task decomposition and dependency relationships; use the critical path analysis algorithm to determine the task execution order and data transfer method between the nodes, and obtain a task-oriented node cooperation process.

[0025] Obtain the node grouping information to understand the existing node groupings in the system and their internal structures. For each node grouping, adopt the task decomposition method to perform fine-grained decomposition of the tasks within the node, obtain each subtask, and clarify the input, output, and processing logic of the tasks. Analyze the dependency relationships between subtasks to identify the sequence and data dependencies between tasks. Based on the dependency relationships, construct a directed acyclic graph model, with subtasks as the nodes of the graph and dependency relationships as the directed edges, forming a task dependency graph. On the basis of the task dependency graph, adopt a critical path analysis algorithm, including either the Critical Path Method (CPM) or the Program Evaluation and Review Technique (PERT), to calculate the earliest start time, latest start time, earliest end time, and latest end time of each subtask, and determine the critical path; the tasks on the critical path have a decisive impact on the completion time of the entire process and need to be focused on and optimized. According to the results of the critical path analysis, determine the execution order of each subtask. For the tasks on the critical path, strictly sort them according to their earliest start time and latest end time to form a critical task execution sequence. For the tasks not on the critical path, their execution order can be appropriately adjusted on the premise of not affecting the critical path to achieve reasonable utilization of resources. For each task in the task execution sequence, analyze its input data and output data to identify the data dependency relationships between tasks. For each task, clarify the upstream tasks from which its required input data is sourced and the downstream tasks that will use its output data. Based on the data dependency relationships, construct a data dependency graph to represent the data flow direction between tasks. Integrate the task execution sequence and the data dependency graph to obtain a task-oriented node collaboration process. This process clarifies the execution order of each task within the node and the data transfer method between tasks. The generation and consumption relationships of data are clear, which can guide the actual task execution and data flow. Optimize the node collaboration process with the goal of minimizing the execution time of the entire process while satisfying the task dependency relationships. Adopt a workflow optimization algorithm, such as the DAG (Directed Acyclic Graph) scheduling algorithm or the critical path optimization algorithm, to adjust the task execution order and data transfer method, reduce the waiting time of tasks and data transmission overhead, and improve the parallelism and efficiency of the process. On the basis of the optimized node collaboration process, further consider the collaboration and communication between node groupings. Analyze the task dependency relationships and data dependency relationships between different node groupings, and construct a global task dependency graph and a data dependency graph. Adopt a distributed workflow management system, such as Apache Airflow or Kubernetes, to coordinate the task scheduling and data transmission between different node groupings, and achieve global task orchestration and optimization. Through the above steps, an optimized task-oriented node collaboration process can be obtained, which clarifies the task execution order and data transfer method within and between nodes.This process can guide the actual task deployment and execution, improve the performance and efficiency of the system, and achieve efficient collaboration among node groups.

[0026] Exemplarily, after obtaining the node group information, the tasks within each node group are decomposed at a fine-grained level. For example, a data processing task is decomposed into sub-tasks such as data collection, preprocessing, feature extraction, model training, and result evaluation, and the input, output, and processing logic of each sub-task are clarified. By analyzing the dependency relationships between sub-tasks, it is identified that data collection is a pre-task for preprocessing, and preprocessing is a pre-task for feature extraction, forming a directed acyclic graph model. Using the critical path analysis algorithm, the critical path is calculated as data collection → preprocessing → feature extraction → model training, and these tasks have a decisive impact on the completion time of the entire process. According to the critical path analysis results, the execution order of each sub-task is determined, and a data dependency graph is constructed to clarify that the output data of the data collection task will be used by the preprocessing task, and the output data of the preprocessing task will be used by the feature extraction task, forming a clear data flow direction. On this basis, the DAG scheduling algorithm is used to optimize the task execution order. By executing independent sub-tasks in parallel, the waiting time of the model training task is reduced from 2 hours to 1 hour, and the execution time of the entire process is optimized from the original 10 hours to 8 hours. Finally, Apache Airflow is used to coordinate the task scheduling and data transmission between different node groups to achieve global task orchestration and optimization. Through the above steps, an optimized task-oriented node collaboration process is obtained, which provides guidance for the actual task deployment and execution and improves the performance and efficiency of the system.

[0027] S104. During the node collaboration process, a distributed data fusion algorithm is adopted to preprocess, extract features, and perform semantic mapping on the heterogeneous data collected by the nodes to obtain a unified data representation form; a multi-sensor data association algorithm is adopted to mine the spatio-temporal correlation between the data of different nodes to achieve cross-node data fusion.

[0028] Obtain heterogeneous raw data collected by distributed nodes, adopt corresponding preprocessing methods for different types and formats of data, clean, denoise, and standardize the data to obtain a normalized data set. According to the data characteristics after preprocessing, use the principal component analysis algorithm to extract the key features of the data, reduce the data dimension, and obtain a concise feature vector representation. By constructing an ontology knowledge base and semantic mapping rules, map the extracted data features to a unified semantic space, eliminate semantic differences between different data sources, and obtain a semantically consistent feature representation form. For the semantic feature data of different nodes, use the mutual information method to calculate the correlation between the data of different nodes, quantify the degree of correlation by calculating the mutual information value between features, and mine the spatio-temporal association patterns of the data. According to the spatio-temporal association intensity of the data, construct a network topology structure for cross-node data fusion, determine the path and order of data fusion, and form a hierarchical data fusion framework. On the basis of the data fusion framework, adopt a distributed data fusion algorithm based on Kalman filtering, run Kalman filtering on each node, and through data exchange and update between nodes, comprehensively utilize the associated data of multiple nodes to generate a consistent result after fusion. Use Tableau to visually display the result after fusion, and intuitively present the effect of data fusion through charts and dashboards. At the same time, use statistical methods such as mean square error and correlation coefficient to evaluate the accuracy of data fusion, and dynamically adjust the parameters of the Kalman filtering algorithm according to the evaluation results, such as the covariance matrices of process noise and measurement noise, to continuously optimize the performance of data fusion. Finally, obtain high-quality fused data to provide support for subsequent data analysis and applications.

[0029] Exemplarily, after obtaining heterogeneous raw data collected by distributed nodes, corresponding preprocessing methods are adopted for different types of data such as text, images, and audio. For text data, regular expressions are used to filter out HTML tags and special characters, and operations such as word segmentation and stop word removal are performed; for image data, median filtering is used to remove salt-and-pepper noise, and histogram equalization is used to enhance contrast; for audio data, wavelet transform is used to denoise the signal and convert it into frequency domain features. After preprocessing, a normalized data set is obtained. According to the data characteristics after preprocessing, the principal component analysis algorithm is adopted. Through eigenvalue decomposition and the selection of the first 5 principal components, the original data is reduced from 1000 dimensions to 50 dimensions, obtaining a concise feature vector representation. By constructing an ontology knowledge base containing 500 concepts and 1000 relationships, and defining semantic mapping rules, such as "temperature" corresponding to "air temperature" and "pressure" corresponding to "atmospheric pressure", the extracted data features are mapped to a unified semantic space. For the semantic feature data of different nodes, the mutual information method is used to calculate the correlation between features. For example, the mutual information value between "temperature" of node A and "humidity" of node B is 8, indicating a strong correlation between them. According to the calculated mutual information matrix, a data fusion network topology containing 10 nodes is constructed, and a hierarchical fusion path of first fusing temperature and humidity data and then fusing air pressure data is determined. On this basis, a distributed data fusion algorithm based on Kalman filtering is adopted, the initial state estimate value and covariance matrix are set, and through 8 iterations, the estimation error is reduced from the initial 10% to 5%, generating a consistent result after fusion. The line chart and scatter chart are drawn for the fusion result using Tableau, intuitively showing the consistency trend of data of different nodes. At the same time, the mean square error of the data before and after fusion is calculated to decrease from 8 to 5, and the correlation coefficient is increased from 6 to 95, proving the effectiveness of data fusion. According to the evaluation results, the process noise covariance matrix parameter of the Kalman filtering algorithm is dynamically adjusted. After 3 adjustments, the fusion accuracy is further improved by 5%. Finally, the high-quality fusion data lays a foundation for subsequent data analysis and applications.

[0030] S105. For the data fusion result, an incremental learning algorithm is adopted to dynamically update the data model and decision rules of the node; an anomaly detection algorithm is adopted to timely detect the abnormal deviation of the node data and trigger the dynamic adjustment of the node grouping.

[0031] Obtain the data fusion result and input it into an incremental learning algorithm such as random forest or support vector machine, etc., to dynamically update the data model and decision rules of the nodes, and obtain the updated node data model and decision rules. Apply the updated node data model and decision rules to the node data, and use anomaly detection algorithms such as isolation forest or local outlier factor, etc., to determine whether there are abnormal deviations in the node data. If there are abnormal deviations, trigger the dynamic adjustment of the node grouping. According to the statistical characteristics of the node data abnormal deviation, such as deviation mean, variance, etc., determine the priority of the node grouping adjustment, and preferentially adjust the node grouping with a larger degree of abnormal deviation. For the node grouping that needs to be adjusted, obtain its historical data within a certain time range from the historical database, and divide the nodes into different groupings through clustering algorithms such as K-means, so that the data models and decision rules of the nodes within the grouping are highly similar. For each newly divided node grouping, based on the data models and decision rules of its internal nodes, generate the overall data model and decision rules of the grouping through ensemble learning algorithms such as Bagging or Boosting. Update the newly generated node grouping data model and decision rules to the local storage or memory of the corresponding node through the configuration file or API interface, and complete the dynamic adjustment of the node grouping. Set a fixed time interval such as 5 minutes to continuously monitor the data fusion result and node data. When abnormal deviations are detected or the preset adjustment threshold is reached, repeat the above steps to achieve the continuous optimization and dynamic evolution of the node data model, decision rules, and grouping.

[0032] Exemplarily, the data fusion result is input into the random forest algorithm for incremental learning. The number of trees is set to 100, and the maximum depth is set to 10. Through 5 iterations of training, the accuracy of the data model of node A is improved from 85% to 92%, and the coverage rate of the decision rule of node B is improved from 90% to 96%. The updated data model and decision rule are applied to detect anomalies in the node data. The isolation forest algorithm is used, the number of trees is set to 50, the subsampling ratio is set to 8, and the anomaly threshold is set to 6. The data anomaly deviation of node C is detected to be 7, which is higher than the threshold, triggering the adjustment of the node grouping. According to the average anomaly deviation of 7 and variance of 2 of node C, its adjustment priority is determined to be high. The historical data of node C in the recent 1 hour is obtained, and the K-means algorithm is used for clustering. The number of clusters k is set to 3, and the number of iterations is set to 10 times. Node C is divided into 2 new groupings. Grouping 1 contains 3 nodes, and grouping 2 contains 2 nodes. The similarity of the data models of the nodes within the grouping reaches more than 95%. The Bagging algorithm is used for ensemble learning on grouping 1, the base learner is set to a decision tree, and the number of ensembles is set to 50 to generate the overall data model of grouping 1. The AdaBoost algorithm is used for ensemble learning on grouping 2, the base learner is set to SVM, and the number of ensembles is set to 30 to generate the overall decision rule of grouping 2. The new grouping models and rules are updated to the corresponding nodes through the API interface to complete the dynamic adjustment. The fusion result and node data are monitored every 5 minutes. When the anomaly deviation of node D reaches 8, the above adjustment process is repeated to continuously optimize the overall performance of the node grouping. The accuracy of anomaly detection is improved from 90% to 95%, and the generalization ability of the node data model is improved from 92% to 98%, realizing the dynamic evolution of the node grouping.

[0033] S106. In the data reporting stage, an adaptive data compression algorithm is adopted to dynamically adjust the data compression rate according to the correlation and redundancy of the data. A multi-path routing algorithm is adopted to optimize the data reporting path by comprehensively considering factors such as link quality, energy consumption balance, and load balance.

[0034] Obtain the original data to be reported, preprocess the data, and use the median filtering method to remove obvious noise and outliers to improve data quality. Analyze the preprocessed data, calculate the correlation between different data segments using the Pearson correlation coefficient, and determine the degree of correlation of the data according to the magnitude of the correlation coefficient. According to the degree of correlation of the data, adopt an adaptive data compression algorithm, set different compression rate thresholds for data segments with different degrees of correlation, and the thresholds can be set according to historical data and empirical values. If the correlation coefficient of a data segment is greater than the corresponding threshold, increase the compression rate of that data segment; otherwise, maintain the original compression rate. The compressed data packets dynamically select appropriate data transmission paths using the multi-attribute decision-making method according to their size and priority. For data packets with a large amount of data and a high priority, select a path with a shorter delay and a larger bandwidth. During the data transmission process, monitor the CPU occupancy rate and memory usage rate of each node in real time. When the CPU occupancy rate of a node exceeds 80% or the memory usage rate exceeds 90%, start the load balancing mechanism between nodes, and use the consistent hashing algorithm to dynamically migrate some data packets to other nodes with lower load for processing. After the data packets reach the aggregation node, use the corresponding decompression algorithm to restore the data, and use the cyclic redundancy check (CRC) method for integrity verification, and report the data that passes the verification to the data center to complete the entire data reporting process.

[0035] Exemplarily, after obtaining the original data to be reported, the data is first preprocessed. The median filtering method is used to remove obvious noise and outliers. The sliding window size is set to 5, and the original data is replaced by taking the median of the data within the window, effectively improving the data quality. Then, the preprocessed data is analyzed. The Pearson correlation coefficient is used to calculate the correlation between different data segments. By setting the threshold to 8, the correlation degree of the data is judged. If the correlation coefficient is greater than 8, the data is considered highly correlated. According to the correlation degree of the data, an adaptive data compression algorithm is adopted. The compression ratio is set to 60% for the data segments with a correlation coefficient greater than 9, 40% for the data segments with a correlation coefficient between 8 and 9, and the original compression ratio of 20% is maintained for the data segments with a correlation coefficient less than 8. The compressed data packets dynamically select appropriate data transmission paths using the multi-attribute decision-making method according to their size and priority. Considering factors such as the data packet size, priority, path delay, and bandwidth, the comprehensive scores of each path are calculated by weighted summation, and the path with the highest score is selected for transmission. During the data transmission process, the CPU occupancy rate and memory usage rate of each node are monitored in real time. When the CPU occupancy rate of node A reaches 85%, the load balancing mechanism between nodes is started, and 30% of the data packets are dynamically migrated to the less-loaded nodes B and C for processing using the consistent hashing algorithm, ensuring the efficiency of data transmission. After the data packets reach the aggregation node, the corresponding decompression algorithm is used to restore the data, and the cyclic redundancy check (CRC) method is adopted for integrity verification. The generating polynomial is set to CRC-16. By comparing the CRC check value of the data packet with the recalculated CRC value, it is ensured that no data is lost or incorrect during the data transmission process. Finally, the data passing the verification is reported to the data center, completing the entire data reporting process and realizing the full-link optimization and guarantee of data from collection to reporting.

[0036] S107. Construct a node cooperation incentive mechanism based on the blockchain, and dynamically adjust the credit value and resource allocation of the node according to the contribution degree and task completion quality of the node; adopt smart contract technology to constrain and arbitrate the cooperation behavior between the nodes.

[0037] According to the pre-established node credit assessment model, combined with the historical behavior data of nodes in the blockchain network, the support vector machine algorithm is used to dynamically calculate the comprehensive credit score of nodes. The credit score is mapped to the corresponding credit level according to the preset interval, and the credit level is written into the blockchain ledger for other nodes to query and reference. The behavior data of nodes participating in task collaboration is obtained through smart contract rules, including task completion time, task quality score, etc. According to the task characteristics and the historical performance of nodes, the weight coefficients of each index are preset, and the weighted average algorithm is used to calculate the contribution degree of nodes in the current task, and the resource allocation weight of nodes is dynamically adjusted according to the contribution degree. Set the expected standard of task completion quality as the threshold for evaluating node performance. If the task completion quality of a node is lower than the preset threshold, the smart contract is triggered to deduct the credit value of the node and reduce its resource allocation weight at the same time; if the task completion quality is higher than the preset threshold, the credit value and resource allocation weight of the node are increased to encourage the contribution of high-quality nodes. When a collaboration dispute occurs between nodes, the blockchain arbitration mechanism is used to judge the dispute behavior. The arbitration mechanism votes on the dispute event through multiple trusted nodes and makes a final decision according to the voting results. The credit values of relevant nodes are rewarded and punished according to the judgment results. The credit value of the violating node is deducted until it is downgraded, and its resource allocation weight is reduced in subsequent task assignments. The asymmetric encryption algorithm (such as RSA) is used to encrypt the collaboration behavior data between nodes to ensure the confidentiality and integrity of the data. The encrypted behavior data is written into the blockchain, and the anti-tampering and traceable characteristics of the blockchain are used to achieve a trusted record and audit of the node collaboration behavior. The deep reinforcement learning algorithm, such as Deep Q-Network (DQN), is introduced. By continuously interacting with the environment and obtaining feedback, it learns to optimize the parameter weights in the node credit assessment model and the decision rules in the resource allocation strategy. According to the operating conditions of the blockchain network, such as indicators such as the credit level distribution of nodes and task completion quality, the incentive and punishment rules in the smart contract are dynamically adjusted to achieve the adaptive evolution of the incentive mechanism. The blockchain consensus mechanism (such as proof of work or proof of stake) is used to achieve the consistency verification of node collaboration behavior, and the authenticity and reliability of the recorded collaboration behavior data are ensured through multi-party verification. The blockchain technology is used to build a trusted collaboration environment to encourage nodes to actively participate in collaboration and improve task completion quality, and maintain the healthy operation of the blockchain network.

[0038] Exemplarily, in the node credit assessment model, the support vector machine algorithm is used to train the historical behavior data of the nodes. By setting the RBF kernel function parameter γ = 1 and the penalty coefficient C = 10, an optimized credit assessment model is constructed. According to the comprehensive credit score of the nodes calculated by the model, it is mapped into 5 credit levels: above 90 points is level A, 80 - 90 points is level B, 70 - 80 points is level C, 60 - 70 points is level D, and below 60 points is level E. The behavior data of the nodes participating in task collaboration is obtained using smart contract rules, including task completion time, quality score, etc. Among them, the weight coefficient of the task completion time is set to 6, and the weight coefficient of the quality score is set to 4. The weighted average algorithm is used to obtain the contribution degree of the nodes. If the task completion quality of the node is lower than the preset threshold of 85 points, the smart contract is triggered to deduct 20 points from the credit value of the node, and reduce its resource weight by 20% in subsequent task assignments; if it is higher than 95 points, the credit value is increased by 10 points, and the resource weight is increased by 10%. When a collaboration dispute occurs between nodes, 5 trusted nodes are used for arbitration voting. If more than 3 nodes vote to determine that node A has violated the regulations, 50 points will be deducted from the credit value of A, and its resource weight will be reduced by 50% in subsequent tasks. The collaboration behavior data between nodes is encrypted using the 2048-bit RSA algorithm to ensure data security. The DQN algorithm is introduced. By setting the learning rate α = 0.1 and the discount factor γ = 0.9, continuous "state-action-reward" learning iterations are carried out to optimize the node credit assessment model and resource allocation strategy. According to the operating conditions of the blockchain network, the incentive and penalty rules in the smart contract are dynamically adjusted. For example, when the average credit level of the nodes is lower than level C, the task completion quality threshold is increased to 90 points to promote the nodes to improve their collaboration performance. Finally, using the proof-of-work consensus mechanism, the nodes are required to provide a proof of work with a computational hash value difficulty of 10^20 to verify the authenticity of the recorded collaboration behavior data and maintain a trusted collaboration environment for the blockchain network.

[0039] Inspired by the above embodiments based on the present invention, through the above description, relevant staff can completely make various changes and modifications without departing from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. An asset data collection method based on the Internet of Things, characterized in that, The method includes: Obtaining the capability, location, and energy consumption attribute information of sensor nodes, using a clustering algorithm to perform clustering analysis on the attribute information to obtain node subsets with similar attributes, and for each of the node subsets, determining its role and tasks in the collaborative grouping to obtain a node grouping scheme driven by business requirements; Using a distributed optimization algorithm, comprehensively considering factors such as communication cost, energy consumption balance, and data redundancy among the nodes, dynamically adjusting the node grouping scheme to obtain a globally optimal node collaboration scheme. If the node capabilities or network topology change, trigger the re-optimization of the grouping scheme; For each of the node groupings, constructing a directed acyclic graph model based on task decomposition and dependency relationships, using a critical path analysis algorithm to determine the task execution order and data transfer method among the nodes to obtain a task-oriented node collaboration process; During the node collaboration process, using a distributed data fusion algorithm to preprocess, extract features, and perform semantic mapping on the heterogeneous data collected by the nodes to obtain a unified data representation form, and using a multi-sensor data association algorithm to mine the spatio-temporal correlation between data of different nodes to achieve cross-node data fusion; For the data fusion result, using an incremental learning algorithm to dynamically update the data model and decision rules of the nodes, and using an anomaly detection algorithm to timely detect abnormal deviations in the node data and trigger the dynamic adjustment of the node grouping; In the data reporting stage, using an adaptive data compression algorithm to dynamically adjust the data compression rate according to the correlation and redundancy of the data, and using a multi-path routing algorithm to comprehensively consider factors such as link quality, energy consumption balance, and load balance to optimize the data reporting path; Constructing a node collaboration incentive mechanism based on blockchain, dynamically adjusting the credit value and resource allocation of the nodes according to the contribution degree and task completion quality of the nodes, and using smart contract technology to constrain and arbitrate the collaboration behavior among the nodes.

2. The method according to claim 1, wherein The obtaining the capability, location, and energy consumption attribute information of sensor nodes, using a clustering algorithm to perform clustering analysis on the attribute information to obtain node subsets with similar attributes; and for each of the node subsets, determining its role and tasks in the collaborative grouping to obtain a node grouping scheme driven by business requirements, includes: Obtaining the capability, location, and energy consumption attribute information of sensor nodes, and for the obtained attribute information, using the K-means clustering algorithm to perform clustering analysis to divide the nodes with similar attributes into multiple subsets; For each node subset, by analyzing its capability and location attributes, determining the role of each subset in the collaborative grouping, and assigning corresponding tasks to each role according to business requirements, thereby obtaining a node grouping scheme driven by business requirements; According to the node grouping scheme, using the Q-learning reinforcement learning algorithm to train the nodes, with maximizing the task completion quality and node collaboration efficiency as the optimization goal, enabling each node to effectively collaborate with other nodes according to its own role and tasks. During the process of node execution, the energy consumption of each node is monitored in real time, and the energy consumption threshold is dynamically set using the control chart method. When the energy consumption of a node exceeds the threshold, the node is marked as a regrouping node, triggering the node regrouping process; According to the current node attribute information and business needs, the genetic algorithm is used to optimize the node grouping scheme, with the minimum node energy consumption and the maximum task completion quality as the optimization goals, and the regrouped nodes are reallocated and the roles are adjusted to achieve dynamic optimization of node grouping; For the reorganized node subset, an instance-based transfer learning algorithm is used to select the source node with the highest task similarity with the target node as the knowledge source, and some training parameters and model structures of the source node are transferred to the target node to accelerate the learning speed and convergence speed of the target node. In the process of node collaboration, the communication data between nodes is analyzed to construct a node collaboration relationship graph. The graph convolutional neural network algorithm is used to train the relationship graph with the optimization goal of maximizing the node collaboration efficiency. The attribute information and neighbor information of the nodes are aggregated through graph convolution operations to mine the implicit collaboration relationship between nodes and guide the nodes to dynamically adjust the collaboration strategy. Continuously monitor changes in business needs, set business demand change detection rules, compare the gap between current business indicators and expected goals, and when the gap exceeds the preset threshold, trigger a business demand change event and re-trigger the node grouping process. Based on the new business needs and node attributes, use the above method to regenerate the node grouping plan to achieve business-driven dynamic optimization of node grouping.

3. The method according to claim 1, wherein The distributed optimization algorithm is used to comprehensively consider the communication cost, energy consumption balance and data redundancy factors between the nodes, dynamically adjust the node grouping scheme, and obtain the global optimal node cooperation scheme; If the node capability or network topology changes, re-optimization of the grouping scheme is triggered, including: Obtain node resource information of each node in the network, including the computing power, storage capacity and communication bandwidth of each node, as well as network topology information, including the connection relationship and link quality between each node; Preprocess the acquired node resource information and network topology information, extract key feature parameters, and construct the node resource attribute matrix and network topology adjacency matrix; According to the node resource attribute matrix and the network topology adjacency matrix, a mathematical model for node grouping optimization is established; In this model, optimization objectives are set, including minimum communication cost between nodes, balanced node energy consumption, and minimum data redundancy, as well as constraints, including node resource capacity limitations and network connectivity; The particle swarm optimization algorithm is used to search for the optimal node grouping scheme under the premise of meeting the optimization objectives and constraints; Through iterative optimization, the particle positions and velocities are continuously updated, and finally convergence is achieved to obtain the global optimal node grouping and collaboration solution; According to the optimized node grouping scheme, generate node grouping and collaboration configuration instructions, and send the configuration instructions to the corresponding nodes; After receiving the configuration instructions, each node adjusts its own working mode and cooperation relationship according to the instructions to form a dynamically optimized node grouping; Deploy monitoring agents for node resources and link quality, and collect the resource usage and link status information of each node in real time; When the node resource utilization rate exceeds the preset threshold or the link quality is lower than the preset threshold, it is determined that the node capabilities or network topology have changed; When the monitoring program determines that the node capabilities or network topology have changed, trigger the parameter update and re - solution of the node grouping optimization model; According to the solution results of the updated model, generate a new node grouping scheme, issue the updated configuration instructions, and realize the dynamic adjustment of node grouping to continuously maintain the globally optimal node cooperation state.

4. The method according to claim 1, wherein For each of the node groupings, construct a directed acyclic graph model based on task decomposition and dependency relationships; use the critical path analysis algorithm to determine the task execution order and data transfer method among the nodes, and obtain the task - oriented node cooperation process, including: Obtain node grouping information to understand the existing node groupings in the system and their internal structures; For each node grouping, use the task decomposition method to perform fine - grained decomposition of the tasks within the node to obtain each subtask, and clarify the input, output, and processing logic of the tasks; Analyze the dependency relationships between subtasks to identify the sequence and data dependencies between tasks; Construct a directed acyclic graph model according to the dependency relationships, use the subtasks as the nodes of the graph, and the dependency relationships as the directed edges to form a task dependency graph; Based on the task dependency graph, use a critical path analysis algorithm, including either the Critical Path Method (CPM) or the Program Evaluation and Review Technique (PERT), to calculate the earliest start time, latest start time, earliest end time, and latest end time of each subtask, and determine the critical path; According to the results of the critical path analysis, determine the execution order of each subtask; For the tasks on the critical path, strictly sort them according to their earliest start time and latest end time to form a critical task execution sequence; For the tasks not on the critical path, adjust their execution order on the premise of not affecting the critical path to achieve reasonable resource utilization; For each task in the task execution sequence, analyze its input data and output data to identify the data dependency relationships between tasks; For each task, clarify which upstream tasks its required input data comes from and which downstream tasks will use its output data; According to the data dependency relationships, construct a data dependency graph to represent the data flow direction between tasks; Fuse the task execution sequence and the data dependency graph to obtain the task - oriented node cooperation process; Optimize the node cooperation process with the goal of minimizing the execution time of the entire process under the premise of satisfying the task dependency relationships; Use a workflow optimization algorithm, including either the Directed Acyclic Graph (DAG) scheduling algorithm or the critical path optimization algorithm, to adjust the task execution order and data transfer method, reduce the waiting time of tasks and data transmission overhead, and improve the parallelism and efficiency of the process; Based on the optimized node cooperation process, further consider the cooperation and communication between node groupings; Analyze the task dependencies and data dependencies between different node groups, and construct a global task dependency graph and a data dependency graph; Adopt a distributed workflow management system, including either Apache Airflow or Kubernetes, to coordinate task scheduling and data transfer between different node groups, and achieve task orchestration and optimization on a global scale.

5. The method according to claim 1, characterized in that, During the process of node collaboration, adopt a distributed data fusion algorithm to preprocess, extract features, and perform semantic mapping on the heterogeneous data collected by the nodes to obtain a unified data representation form; adopt a multi-sensor data association algorithm to mine the spatio-temporal correlation between the data of different nodes and achieve cross-node data fusion, including: Obtain the heterogeneous raw data collected by distributed nodes, and adopt corresponding preprocessing methods for different types and formats of data to clean, denoise, and standardize the data to obtain a normalized data set; According to the data characteristics after preprocessing, adopt the principal component analysis algorithm to extract the key features of the data, reduce the data dimension, and obtain a concise feature vector representation; By constructing an ontology knowledge base and semantic mapping rules, map the extracted data features to a unified semantic space, eliminate the semantic differences between different data sources, and obtain a semantically consistent feature representation form; For the semantic feature data of different nodes, adopt the mutual information method to calculate the correlation between the data of different nodes, quantify the degree of correlation by calculating the mutual information value between features, and mine the spatio-temporal association pattern of the data; According to the spatio-temporal association intensity of the data, construct a network topology structure for cross-node data fusion, determine the path and order of data fusion, and form a hierarchical data fusion framework; Based on the data fusion framework, adopt a distributed data fusion algorithm based on Kalman filtering, run Kalman filtering on each node, and through data exchange and update between nodes, comprehensively utilize the associated data of multiple nodes to generate a fused consistent result; Use Tableau to visually display the fused result, and intuitively present the effect of data fusion through charts and dashboards; At the same time, use statistical methods, including either the mean square error or the correlation coefficient method, to evaluate the accuracy of data fusion, and dynamically adjust the parameters of the Kalman filtering algorithm according to the evaluation results, including the covariance matrices of process noise and measurement noise, and continuously optimize the performance of data fusion; Finally, obtain high-quality fused data to provide support for subsequent data analysis and applications.

6. The method according to claim 1, wherein For the data fusion result, adopt an incremental learning algorithm to dynamically update the data model and decision rules of the nodes; adopt an anomaly detection algorithm to timely detect the abnormal deviation of the node data and trigger the dynamic adjustment of the node group, including: Obtain the data fusion result and input it into an incremental learning algorithm, including at least one of random forest or support vector machine, to dynamically update the data model and decision rules of the nodes and obtain the updated node data model and decision rules; Apply the updated node data model and decision rules to the node data, and determine whether there are abnormal deviations in the node data through an anomaly detection algorithm, which includes at least one of isolation forest or local outlier factor. If there are abnormal deviations, trigger the dynamic adjustment of node grouping. Determine the priority of node grouping adjustment according to the statistical characteristics of node data abnormal deviations, including deviation mean and variance, and preferentially adjust the node grouping with a larger degree of abnormal deviation. For the node grouping that needs to be adjusted, obtain its historical data within a certain time range from the historical database, and divide the nodes into different groupings through the K-means clustering algorithm, so that the data models and decision rules of the nodes within the grouping are more similar. For each newly divided node grouping, based on the data models and decision rules of its internal nodes, generate the overall data model and decision rules of the grouping through an ensemble learning algorithm, including Bagging or Boosting. Update the newly generated node grouping data model and decision rules to the local storage or memory of the corresponding node in the form of a configuration file or API interface to complete the dynamic adjustment of node grouping. Set a fixed time interval, such as 5 minutes, to continuously monitor the data fusion result and node data. When abnormal deviations are detected or the preset adjustment threshold is reached, repeat the above steps to achieve continuous optimization and dynamic evolution of the node data model, decision rules, and grouping.

7. The method according to claim 1, characterized in that In the data reporting stage, an adaptive data compression algorithm is adopted to dynamically adjust the data compression rate according to the correlation and redundancy of the data; a multi-path routing algorithm is adopted to optimize the data reporting path by comprehensively considering factors such as link quality, energy consumption balance, and load balance, including: Obtain the original data to be reported, preprocess the data, and use the median filtering method to remove obvious noise and outliers to improve data quality. Analyze the preprocessed data, calculate the correlation between different data segments using the Pearson correlation coefficient, and determine the degree of correlation of the data according to the magnitude of the correlation coefficient. According to the degree of correlation of the data, adopt an adaptive data compression algorithm to set different compression rate thresholds for data segments with different degrees of correlation, and the thresholds can be set according to historical data. If the correlation coefficient of the data segment is greater than the corresponding threshold, increase the compression rate of the data segment, otherwise maintain the original compression rate. The compressed data packets dynamically select a suitable data transmission path using the multi-attribute decision method according to their size and priority. For data packets with a large amount of data and high priority, select a path with a shorter delay and a larger bandwidth. During the data transmission process, continuously monitor the CPU occupancy rate and memory usage rate of each node. When the CPU occupancy rate of a node exceeds 80% or the memory usage rate exceeds 90%, start the load balancing mechanism between nodes, and use the consistent hashing algorithm to dynamically migrate some data packets to other nodes with lower load for processing. After the data packet arrives at the aggregation node, use the corresponding decompression algorithm to restore the data, and use the cyclic redundancy check CRC method for integrity verification, and report the data that passes the verification to the data center to complete the entire data reporting process.

8. The method according to claim 1, wherein Construct the blockchain-based node collaboration incentive mechanism, dynamically adjust the credit value and resource allocation of the nodes according to the contribution degree and task completion quality of the nodes; adopt smart contract technology to constrain and arbitrate the collaboration behaviors among the nodes, including: According to the pre-established node credit assessment model, combined with the historical behavior data of the nodes in the blockchain network, dynamically calculate the comprehensive credit score of the nodes by using the support vector machine algorithm; Map the credit score to the corresponding credit level according to the preset interval, and write the credit level into the blockchain ledger for other nodes to query and refer to; Obtain the behavior data of the nodes participating in task collaboration through smart contract rules, including task completion time and task quality score; According to the task characteristics and the historical performance of the nodes, preset the weight coefficients of each index, calculate the contribution degree of the nodes in the current task by using the weighted average algorithm, and dynamically adjust the resource allocation weight of the nodes according to the contribution degree; Set the expected standard of task completion quality as the threshold for evaluating the performance of the nodes; If the task completion quality of the node is lower than the preset threshold, trigger the smart contract to deduct the credit value of the node and reduce its resource allocation weight at the same time; If the task completion quality is higher than the preset threshold, increase the credit value and resource allocation weight of the node to incentivize the contributions of high-quality nodes; When a collaboration dispute occurs among the nodes, judge the disputed behavior through the blockchain arbitration mechanism; The arbitration mechanism votes on the disputed event through multiple trusted nodes and makes a final ruling according to the voting results; Reward and punish the credit values of the relevant nodes according to the judgment results, deduct the credit value of the violating nodes until they are downgraded, and reduce their resource allocation weights in subsequent task assignments; Use the asymmetric encryption algorithm RSA to encrypt the collaboration behavior data among the nodes to ensure the confidentiality and integrity of the data; Write the encrypted behavior data into the blockchain, and utilize the anti-tampering and traceable characteristics of the blockchain to achieve the trustworthy recording and auditing of the node collaboration behaviors; Introduce the deep reinforcement learning algorithm, including DeepQ-Network, and continuously interact with the environment and obtain feedback to learn and optimize the parameter weights in the node credit assessment model and the decision rules in the resource allocation strategy; Dynamically adjust the incentive and punishment rules in the smart contract according to the operating conditions of the blockchain network, including the credit level distribution of the nodes and the task completion quality indicators, to achieve the adaptive evolution of the incentive mechanism; Adopt the blockchain consensus mechanism to achieve the consistency verification of the node collaboration behaviors, and ensure the authenticity and reliability of the recorded collaboration behavior data through multi-party verification.

Images