A cluster aggregation communication system applied to AI data stream processing

Abstract

The application discloses a kind of cluster collection communication systems applied to AI data stream processing, it is related to distributed system and network engineering technical field, including topological network deployment module, data stream space-time data acquisition module, optimization verification module, state monitoring early warning module and visual traceability module;Topological network deployment module synchronously processes AI data stream, data stream space-time data acquisition module extracts entropy value and covariance matrix, optimization verification module outputs optimization parameter and reliable verification result, state monitoring early warning module sets adaptive threshold and realizes hierarchical early warning, visual traceability module generates three-dimensional thermal diagram and traceability atlas.The application improves data stream processing efficiency by star-ring hybrid topology and low-delay protocol, enhances decision reliability by combining quantum optimization and blockchain verification, realizes accurate early warning by using dynamic weighted monitoring and adaptive threshold, three-dimensional visual traceability accelerates fault location, and provides efficient, intelligent full-stack solution for AI cluster communication.

Description

Technical Field

[0001] This invention relates to the field of distributed systems and network engineering technology, specifically to a cluster aggregation communication system applied to AI data stream processing. Background Technology

[0002] The trunking aggregation communication system is a mobile communication system specifically designed for group dispatch and command. Its core lies in sharing channel resources. The system dynamically allocates all available channels to all users, supports group calls and emergency calls, operates in simplex and half-duplex modes, and improves efficiency through dynamic channel allocation. Its advantages include high spectrum utilization, low network construction cost, high reliability, and flexibility. It is widely used in public safety, transportation, and industrial fields. Through gateways, it can realize multi-standard trunking interconnection, support voice, data, and image transmission, and meet complex dispatching needs.

[0003] To address the challenges of traditional cluster aggregation communication systems in processing high-dimensional dynamic data streams, existing technologies employ a single topology combined with static threshold monitoring. However, these approaches suffer from several drawbacks: topology rigidity leading to a tradeoff between communication efficiency and fault tolerance; insufficient extraction of spatiotemporal features of the data stream causing biased optimization decisions; and centralized verification mechanisms introducing single-point trust risks. These issues ultimately result in poor system scalability, delayed anomaly response, and cross-module collaboration failures. To resolve these problems, a cluster aggregation communication system for AI data stream processing is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a cluster aggregation communication system for AI data stream processing to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a cluster aggregation communication system for AI data stream processing, including a topology network deployment module, a data stream spatiotemporal data acquisition module, an optimization verification module, a status monitoring and early warning module, and a visualization and tracing module;

[0006] The topology network deployment module deploys a star-ring hybrid topology network in the cluster aggregation communication system, and combines a low-latency protocol to synchronously process AI data streams between modules.

[0007] The data stream spatiotemporal data acquisition module, based on a star-ring hybrid topology network, collects and preprocesses operation log data and spatial coordinate data, and extracts the sliding window entropy and covariance matrix of each dimension from them.

[0008] The optimization verification module constructs a quantum optimization model and a blockchain verification model based on the sliding window entropy and covariance matrix, and outputs optimized weight parameters and reliable verification results.

[0009] The status monitoring and early warning module dynamically weights and optimizes the weight parameters and the trusted verification results, outputs a comprehensive judgment value, sets an adaptive threshold for the comprehensive judgment value, judges the status of the AI ​​data stream, and provides hierarchical early warnings.

[0010] The visualization and tracing module generates a three-dimensional spatial heat map and an abnormal event tracing map.

[0011] A further improvement to the technical solution of this invention lies in the fact that the process of synchronously processing AI data streams between modules in the topology network deployment module includes:

[0012] In the cluster aggregation communication system, a star-shaped backbone network is constructed through high-performance routers, which serve as core nodes to connect AI data stream processing terminals and execution terminals. A ring topology subnet driven by a low-latency protocol is deployed between the quantum optimization model and the blockchain verification model to form a closed-loop redundant link. Heterogeneous data caching gateways are deployed at the star-ring intersection nodes to dynamically distribute the peak and average traffic of AI data streams, and the load threshold of the intersection nodes is optimized through a dynamic load balancing algorithm.

[0013] The star backbone employs a priority scheduling mechanism to dynamically load balance AI data streams to target terminals. Transmission latency is constrained by the size of a single data packet and the number of routing hops. Within the ring subnet, each node uses an alternating verification mechanism to process data. After the quantum-optimized node processes the data, the blockchain verification node verifies the integrity of the result along the ring path. The verification cycle is dynamically adjusted by a real-time load feedback factor. When the star backbone exceeds its load limit, the cache gateway triggers a ring-based traffic splitting path, directly connecting the data to the target node via the ring link. The optimal path is selected based on a weighted average of the square root of the load rate and the ring link load rate. When a node within the ring subnet fails, a low-latency protocol initiates a reverse link retransmission.

[0014] A further improvement to the technical solution of this invention lies in that: in the data stream spatiotemporal data acquisition module, the acquisition and preprocessing process of operation log data and spatial coordinate data includes:

[0015] Distributed fiber optic sensors are deployed at the edge nodes of the star backbone network. The distributed fiber optic sensors capture system events at millisecond frequency, collect timestamped operation log data from heterogeneous data streams in real time, and transmit them to the cache gateway through the star backbone network.

[0016] Intelligent edge nodes are deployed on the execution terminal side of the ring topology subnet. The intelligent edge nodes collect spatial coordinate data of device locations and environmental monitoring points in the communication system based on spatial sensors, synchronously generate environmental status labels, and push them to the cache gateway through the low-latency protocol of the ring subnet.

[0017] The noise threshold is dynamically set by the signal energy, and the wavelet threshold noise removal algorithm is used to filter out high-frequency noise in the operation log data. The operation log data and spatial coordinate data are matched by timestamp to construct a spatiotemporal correlation matrix.

[0018] A further improvement to the technical solution of this invention lies in the fact that the process of extracting the sliding window entropy from the preprocessed operation log data in the data stream spatiotemporal data acquisition module includes:

[0019] A dual-channel data interface is configured on the heterogeneous data caching gateway. Channel 1 receives preprocessed star-topology backbone operation log data and uses a sliding time window segmentation technique to divide the continuous log stream into discrete batches, based on the peak data stream rate V. peak Dynamically adjust the time window length W t Within a single window, statistically analyze the distribution of event types and calculate the frequency p of the i-th type of event. i Based on event frequency, the sliding window entropy S is obtained through a parallel computing framework. The calculation process is as follows:

[0020]

[0021] Among them, D d Where n is the size of a single data packet, N is the number of parallel processing windows, and n i Let m be the number of events of type i, and m be the total number of event types within the window.

[0022] A further improvement to the technical solution of this invention lies in the fact that the process of extracting the covariance matrix of each dimension from the preprocessed spatial coordinate data in the data stream spatiotemporal data acquisition module includes:

[0023] Channel 2 receives the preprocessed spatial coordinate data of the ring subnet, inputs it into the covariance matrix algorithm, calculates the mean and covariance of dimension X and dimension Y, and constructs the covariance matrix of each dimension.

[0024] If the coordinate values ​​of a spatial coordinate data point deviate from the mean by more than 3 times the standard deviation of the dimension, it is identified as an outlier and removed.

[0025] A further improvement to the technical solution of this invention lies in the following: the process of constructing a quantum optimization model and outputting optimization weight parameters in the optimization verification module includes:

[0026] The quantum annealing algorithm is used to map the sliding window entropy to a quantum Hamiltonian H, and a coupling coefficient J is constructed. ij With bias term h i The calculation process is as follows:

[0027] H = ∑ i<j J ij σ i σ j+∑ i h i σ i ;

[0028] Where, σ i ∈{-1,+1} represents the spin state of a qubit;

[0029] Initialize the spin state of the qubit and set the initial temperature T. c and final temperature T z Based on the real-time load feedback factor α, the annealing rate T(t) is dynamically adjusted according to an exponential function to generate a quantum optimization model. Through quantum tunneling, the energy barrier is traversed, allowing the cluster communication system to approach the ground state. The final quantum bit state is measured, and the final quantum bit state σ is... i Mean converted to optimized weight parameter y q The calculation process is as follows:

[0030] T(t) = T c ·e -αt ;

[0031]

[0032] Where t is the annealing time and N is the total number of qubits.

[0033] A further improvement to the technical solution of this invention lies in the following: the process of constructing a blockchain verification model and outputting a trusted verification result in the optimized verification module includes:

[0034] Based on smart contract technology, the covariance matrix M is broadcast to each blockchain verification node through a ring topology subnet. Each blockchain verification node uses a dual SHA-256 chain structure to calculate the hash value H(M) of the covariance matrix = SHA256(SHA256(M)). If more than 2 / 3 of the blockchain verification nodes return the same H(M), the data is considered complete.

[0035] A distributed consensus mechanism is adopted to execute smart contracts in three phases: proposal, verification, and submission, thereby constructing a blockchain verification model.

[0036] During the proposal phase, the main blockchain validator encapsulates the covariance matrix and hash value into a block proposal and broadcasts it along a circular path;

[0037] During the verification phase, each node verifies the distribution consistency of the covariance matrix based on the 3σ anomaly removal rule and generates a verification result. If the absolute value of the coordinate value of the k-th data point in dimension i minus the mean of dimension i does not exceed three times the standard deviation of dimension i, the verification result is 1. If the absolute value of the coordinate value of the k-th data point in dimension i minus the mean of dimension i exceeds three times the standard deviation of dimension i, the verification result is 0.

[0038] During the submission phase, if more than 2 / 3 of the blockchain verification nodes return a verification result of 1, the smart contract generates a digital signature N = ECDSA(H(M), private key) and writes the signed result (M, N) into the blockchain.

[0039] The consensus round R is dynamically adjusted based on the real-time load rate L and load feedback factor β of the ring subnet. If consensus is reached, the blockchain verification model outputs a trusted verification result y. v =1, if consensus fails, the blockchain verification model outputs a trusted verification result y. v =0.

[0040] A further improvement to the technical solution of this invention lies in the following: the process of dynamically weighting and optimizing the weight parameters and the reliable verification results in the status monitoring and early warning module, and outputting a comprehensive judgment value, includes:

[0041] Based on the dynamic weight factor γ, the optimized weight parameters y output by the quantum optimization model are... q The trusted verification result y output by the blockchain verification model v Dynamic weighting is performed, and γ is optimized based on backpropagation of historical errors to output a comprehensive judgment value y. The calculation process is as follows:

[0042] y = γy q +(1-γ)y v ;

[0043]

[0044] Where η is the learning rate and L is the cross-entropy loss function. Accelerate convergence through mixed-precision computing techniques.

[0045] A further improvement to the technical solution of this invention lies in the following: In the status monitoring and early warning module, the process of setting an adaptive threshold for the comprehensive judgment value, judging the status of the AI ​​data stream, and performing graded early warning includes:

[0046] Based on the sliding window entropy fluctuation range and covariance matrix stability, the threshold interval [T] is dynamically set. min ,T max The calculation process is as follows:

[0047] T min =μ y -k·σ y ;

[0048] T max =μ y +k·σ y ;

[0049]

[0050] Among them, μ y is the mean value of the recent comprehensive judgment value, σ y is the standard deviation, and k is the warning sensitivity;

[0051] If T min <y < T max , it is determined as a normal data stream and output to the execution end. If y < T min , y > T max , and the deviation amplitude Δ = |y - T 边界 | ≤ 0.2σ y , a first-level warning is triggered and it is determined as a mildly abnormal data stream. If Δ > 0.2σ y , a second-level warning and an emergency response instruction are triggered, and it is determined as a severely risky data stream;

[0052] After each batch of data stream processing is completed, the ratio of the number of false alarms to the total number of warnings is used as the error rate E, and the threshold interval is corrected based on the error rate. The correction process includes that if E > 5%, the threshold interval is expanded to [T min -0.1σ y , T max +0.1σ y . If E < 1%, the threshold interval is reduced to [T min +0.05σ y [[ID=3S]], T max -0.05σ y . If 1% < E < 5%, the threshold interval remains unchanged.

[0053] A further improvement of the technical solution of the present invention is that in the visualization traceability module, the process of generating a three-dimensional space heat map and an abnormal event traceability map includes:

[0054] Dynamically weight the covariance matrix elements and the sliding window entropy to generate a three-dimensional space heat density value. Based on the collected spatial coordinate data, map the three-dimensional space heat density value to a three-dimensional grid, and fill the heat distribution between discrete points through an interpolation algorithm to generate a three-dimensional space heat map;

[0055] Taking the spatio-temporal correlation matrix, the hierarchical warning signal output by the status monitoring and warning module, and the corresponding time window identifier as inputs, according to the warning time window identifier, trace back to the original operation log batch in the cache gateway, extract the abnormal event type distribution and timestamp sequence, and screen the coordinate points in the spatio-temporal correlation matrix with elements greater than 0.7 and mark them as the locations where abnormal events occur;

[0056] Construct an abnormal event source map and define abnormal event nodes that include timestamps, sliding window entropy, warning levels, and associated spatial coordinates. If two abnormal event nodes share the same coordinates in the spatiotemporal correlation matrix, the edge weight of the abnormal event source map is determined by the difference in sliding window entropy between the two abnormal events. Nodes and edges are rendered in layers according to their weights, with high-weight edges highlighted in red and low-weight edges faded in gray.

[0057] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows:

[0058] 1. This invention provides a cluster aggregation communication system for AI data stream processing. By deploying a star-ring hybrid topology network and combining it with a low-latency protocol, it effectively solves the contradiction between scalability and fault tolerance in traditional single topology structures, realizes efficient synchronization and load balancing of AI data streams between modules, reduces communication latency to the microsecond level, and significantly improves cluster throughput.

[0059] 2. This invention provides a cluster aggregation communication system for AI data stream processing. Based on spatiotemporal correlation data, it extracts sliding window entropy and covariance matrix, constructs quantum optimization model and blockchain verification model, breaks through the local optimum limitation of classical algorithms, ensures decision credibility through distributed ledger, and makes the optimized weight parameters and system verification results tamper-resistant, thereby improving the accuracy and security of high-dimensional dynamic data stream processing.

[0060] 3. This invention provides a cluster aggregation communication system for AI data stream processing. It innovatively integrates dynamic weighted comprehensive judgment and adaptive threshold learning algorithm to realize real-time hierarchical early warning of AI data stream status. Through three-dimensional spatial heat map and abnormal event tracing map, it shortens the fault location time by more than 80%, providing intelligent transformation support for operation and maintenance from passive response to proactive prediction. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0062] Figure 1 This is a block diagram of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Examples, such as Figure 1 As shown, the present invention provides a cluster aggregation communication system for AI data stream processing, including a topology network deployment module, a data stream spatiotemporal data acquisition module, an optimization verification module, a status monitoring and early warning module, and a visualization and tracing module;

[0065] The topology network deployment module deploys a star-ring hybrid topology network within the cluster aggregation communication system. Combined with low-latency protocols, it synchronously processes AI data streams between modules. Within the cluster aggregation communication system, a star backbone network is constructed using high-performance routers, serving as core nodes connecting AI data stream processing terminals and execution terminals. A low-latency protocol-driven ring topology subnet is deployed between the quantum optimization model and the blockchain verification model, forming a closed-loop redundant link. Heterogeneous data caching gateways are deployed at the star-ring intersection nodes to dynamically distribute peak and average traffic of the AI ​​data streams. A dynamic load balancing algorithm optimizes the load threshold of the intersection nodes. A priority scheduling mechanism is adopted to dynamically load balance and distribute AI data streams to target terminals. Transmission latency is constrained by the size of a single data packet and the number of routing hops. Each node in the ring subnet uses an alternating verification mechanism to process data. After the quantum-optimized node processes the data, the blockchain verification node verifies the integrity of the result by going back along the ring path. The verification cycle is dynamically adjusted by the real-time load feedback factor. When the star backbone load exceeds the limit, the cache gateway triggers the ring diversion path to directly connect the data to the target node via the ring link. The optimal path is selected based on the weighted average of the square root of the load rate and the load rate of the ring link. When a node in the ring subnet fails, the low-latency protocol initiates the reverse link retransmission.

[0066] The spatiotemporal data acquisition module, based on a star-ring hybrid topology network, collects and preprocesses operation log data and spatial coordinate data, extracting sliding window entropy and covariance matrices for each dimension. Distributed fiber optic sensors are deployed at the edge nodes of the star backbone network, capturing system events at millisecond-level frequencies. Real-time acquisition of timestamped operation log data from the heterogeneous data stream is transmitted to the cache gateway via the star backbone network. Intelligent edge nodes are deployed on the execution terminal side of the ring topology subnet. These intelligent edge nodes, based on spatial sensor clusters, collect spatial coordinate data of device locations and environmental monitoring points in the communication system, synchronously generating environmental status labels, and pushing them to the cache gateway via the low-latency protocol of the ring subnet. A noise reduction threshold is dynamically set by signal energy, and a wavelet threshold denoising algorithm is used to filter out high-frequency noise in the operation log data. The operation log data and spatial coordinate data are matched by timestamps to construct a spatiotemporal correlation matrix. A dual-channel data interface is configured on the heterogeneous data cache gateway. Channel one receives the preprocessed star backbone operation log data and uses a sliding time window segmentation technique to divide the continuous log stream into discrete batches, determined by the peak data stream rate V. peak Dynamically adjust the time window length W t Within a single window, statistically analyze the distribution of event types and calculate the frequency p of the i-th type of event. i Based on event frequency, the sliding window entropy S is obtained through a parallel computing framework. The calculation process is as follows:

[0067]

[0068] Among them, D d Where n is the size of a single data packet, N is the number of parallel processing windows, and n i Let m be the number of events of type i, and m be the total number of event types within the window. Channel 2 receives the preprocessed spatial coordinate data of the ring subnet and inputs it into the covariance matrix algorithm to calculate the mean and covariance of dimension X and dimension Y, and construct the covariance matrix of each dimension. If the coordinate value of a spatial coordinate data point deviates from the mean by more than 3 times the standard deviation of the dimension, it is judged as an outlier and removed.

[0069] The optimization verification module, based on sliding window entropy and covariance matrix, constructs a quantum optimization model and a blockchain verification model, outputting optimized weight parameters and reliable verification results. It employs a quantum annealing algorithm to map the sliding window entropy to a quantum Hamiltonian H, and constructs a coupling coefficient J. ij With bias term h i The calculation process is as follows:

[0070] H = Σ i<j J ij σ i σ j +∑ i hi σ i ;

[0071] Where, σ i ∈{-1,+1} represents the spin state of the qubit. Initialize the qubit spin state by setting the initial temperature T. c and final temperature T z Based on the real-time load feedback factor α, the annealing rate T(t) is dynamically adjusted according to an exponential function to generate a quantum optimization model. Through quantum tunneling, the energy barrier is traversed, allowing the cluster communication system to approach the ground state. The final quantum bit state is measured, and the final quantum bit state σ is... i Mean converted to optimized weight parameter y q The calculation process is as follows:

[0072] T(t) = T c ·e -αt ;

[0073]

[0074] Where t is the annealing time and N is the total number of qubits, based on smart contract technology, the covariance matrix M is broadcast to each blockchain verification node through a ring topology subnet. Each blockchain verification node calculates the hash value of the covariance matrix H(M) = SHA256(SHA256(M)) using a dual SHA-256 chain structure. If more than 2 / 3 of the blockchain verification nodes return the same H(M), the data is considered complete. A distributed consensus mechanism is adopted, and smart contracts are executed in three phases: proposal, verification, and submission, to build a blockchain verification model. In the proposal phase, the main blockchain verification node encapsulates the covariance matrix and hash value into a block proposal and broadcasts it along the ring path. In the verification phase, each node... The consistency of the covariance matrix distribution is verified based on the 3σ anomaly removal rule, generating a verification result. If the absolute value of the k-th data point's coordinate value in dimension i minus the mean of dimension i does not exceed three times the standard deviation of dimension i, the verification result is 1; if the absolute value of the k-th data point's coordinate value in dimension i minus the mean of dimension i exceeds three times the standard deviation of dimension i, the verification result is 0. During the submission phase, if more than 2 / 3 of the blockchain verification nodes return a verification result of 1, the smart contract generates a digital signature N = ECDSA(H(M), private key) and writes the signed result (M, N) into the blockchain. The consensus round is dynamically adjusted based on the real-time load rate L of the ring subnet and the load feedback factor β. If consensus is reached, the blockchain verification model outputs a trusted verification result y. v =1, if consensus fails, the blockchain verification model outputs a trusted verification result y. v =0;

[0075] The status monitoring and early warning module dynamically weights and optimizes the weight parameters and reliable verification results, outputs a comprehensive judgment value, sets an adaptive threshold for the comprehensive judgment value, judges the status of the AI ​​data stream, and provides tiered early warnings. Based on the dynamic weight factor γ, it optimizes the weight parameters y output by the quantum optimization model. q The trusted verification result y output by the blockchain verification model v Dynamic weighting is performed, and γ is optimized based on backpropagation of historical errors to output a comprehensive judgment value y. The calculation process is as follows:

[0076] y = γy q +(1-γ)y v ;

[0077]

[0078] Where η is the learning rate and L is the cross-entropy loss function. Convergence is accelerated through mixed-precision computation techniques, and the threshold interval [T] is dynamically set based on the sliding window entropy fluctuation range and the stability of the covariance matrix. min ,T max The calculation process is as follows:

[0079] T min =μ y -k·σ y ;

[0080] T max =μ y +k·σ y ;

[0081]

[0082] Where, μ y σ is the average of recent comprehensive judgment values. y Let T be the standard deviation, k be the early warning sensitivity, and if T min <y<T max If y is normal, it is considered a normal data stream and output to the execution end. <T min y>T max And the deviation Δ=|yT 边界 |≤0.2σ y If this occurs, a Level 1 warning is triggered, indicating a slightly abnormal data stream. If Δ > 0.2σ y If a data stream is deemed to be at high risk, a Level II early warning and emergency response command will be triggered. After each batch of data streams is processed, the error rate E is calculated as the ratio of false alarms to total early warnings. A threshold range is then adjusted based on this error rate. The adjustment process includes expanding the threshold range to [T] if E > 5%. min -0.1σ y ,T max+0.1σy ] , if E < 1%, then narrow the threshold interval to [T min +0.05σ y , T max -0.05σ y , if 1% < E < 5%, then keep the threshold interval unchanged;

[0083] The visualization and traceability module generates a 3D spatial heat map and an abnormal event traceability map. It dynamically weights the covariance matrix elements and the sliding window entropy to generate the 3D spatial heat density value. Based on the collected spatial coordinate data, it maps the 3D spatial heat density value to a 3D grid, fills the heat distribution between discrete points through an interpolation algorithm, and generates a 3D spatial heat map. Taking the spatio-temporal correlation matrix, the hierarchical warning signals output by the status monitoring and warning module, and the corresponding time window identifiers as inputs, according to the warning time window identifier, it traces back to the original operation log batches in the cache gateway, extracts the abnormal event type distribution and time stamp sequence, filters the coordinate points with the spatio-temporal correlation matrix elements greater than 0.7, marks them as the occurrence locations of abnormal events, constructs an abnormal event traceability map, defines abnormal event nodes including time stamps, sliding window entropy, warning levels, and associated spatial coordinates. If two abnormal event nodes share the same coordinate points in the spatio-temporal correlation matrix, the edge weight of the abnormal event traceability map is determined by the difference in the sliding window entropy of the two abnormal events. The nodes and edges are rendered in layers according to the weights, with high-weight edges highlighted in red and low-weight edges faded in gray.

[0084] First, the star-ring hybrid network architecture is constructed through the topology network deployment module, and the low-latency protocol is used to establish the communication channels between modules to ensure the basic support for the synchronous transmission of AI data streams. Subsequently, the data stream spatio-temporal data acquisition module collects operation logs and node spatial coordinates in real time on the hybrid topology, extracts the sliding window entropy and multi-dimensional covariance matrix after preprocessing, and forms a feature set reflecting the dynamic characteristics and spatial distribution of data flow. Immediately afterwards, the optimization and verification module drives the quantum optimization algorithm to solve the optimal weight parameters based on the extracted spatio-temporal features, and at the same time uses the blockchain distributed ledger to perform trusted verification on the decision-making process, and outputs an optimized result with both efficiency and security. Further, the status monitoring and warning module dynamically fuses the optimized weights and verification results, calculates the comprehensive determination value through the adaptive threshold learning algorithm, and realizes the real-time hierarchical warning of the AI data stream state. Finally, the visualization and traceability module maps the monitoring data and the topology structure to the 3D space, generates a heat map to intuitively display the abnormal distribution, and constructs a traceability map in combination with the graph neural network to support the operation and maintenance personnel to quickly locate the root cause of the fault from a global perspective. Each module forms a complete solution from network deployment, feature extraction, intelligent optimization to state perception and abnormal traceability through closed-loop data flow and collaborative decision-making.

[0085] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A cluster aggregation communication system for AI data stream processing, characterized in that: It includes a topology network deployment module, a data flow spatiotemporal data acquisition module, an optimization and verification module, a status monitoring and early warning module, and a visualization and tracing module; The topology network deployment module deploys a star-ring hybrid topology network in the cluster aggregation communication system, and combines a low-latency protocol to synchronously process AI data streams between modules. The data stream spatiotemporal data acquisition module, based on a star-ring hybrid topology network, collects and preprocesses operation log data and spatial coordinate data, and extracts the sliding window entropy and covariance matrix of each dimension from them. The optimization verification module constructs a quantum optimization model and a blockchain verification model based on the sliding window entropy and covariance matrix, and outputs optimized weight parameters and reliable verification results. The status monitoring and early warning module dynamically weights and optimizes the weight parameters and the trusted verification results, outputs a comprehensive judgment value, sets an adaptive threshold for the comprehensive judgment value, judges the status of the AI ​​data stream, and provides hierarchical early warnings. The visualization and tracing module generates a three-dimensional spatial heat map and an abnormal event tracing map. In the topology network deployment module, the process of synchronously processing AI data streams between modules includes: In the cluster aggregation communication system, a star-shaped backbone network is constructed through high-performance routers, which serve as core nodes to connect AI data stream processing terminals and execution terminals. A ring topology subnet driven by a low-latency protocol is deployed between the quantum optimization model and the blockchain verification model to form a closed-loop redundant link. Heterogeneous data caching gateways are deployed at the star-ring intersection nodes to dynamically distribute the peak and average traffic of AI data streams, and the load threshold of the intersection nodes is optimized through a dynamic load balancing algorithm. The star backbone employs a priority scheduling mechanism to dynamically load balance AI data streams to target terminals. Transmission latency is constrained by the size of a single data packet and the number of routing hops. Within the ring subnet, each node uses an alternating verification mechanism to process data. After the quantum-optimized node processes the data, the blockchain verification node verifies the integrity of the result along the ring path. The verification cycle is dynamically adjusted by a real-time load feedback factor. When the star backbone exceeds its load limit, the cache gateway triggers a ring-based traffic splitting path, directly connecting the data to the target node via the ring link. The optimal path is selected based on a weighted average of the square root of the load rate and the ring link load rate. When a node within the ring subnet fails, a low-latency protocol initiates a reverse link retransmission. The data stream spatiotemporal data acquisition module includes the following processes for acquiring and preprocessing operation log data and spatial coordinate data: Distributed fiber optic sensors are deployed at the edge nodes of the star backbone network. The distributed fiber optic sensors capture system events at millisecond frequency, collect timestamped operation log data from heterogeneous data streams in real time, and transmit them to the cache gateway through the star backbone network. Intelligent edge nodes are deployed on the execution terminal side of the ring topology subnet. The intelligent edge nodes collect spatial coordinate data of device locations and environmental monitoring points in the communication system based on spatial sensors, synchronously generate environmental status labels, and push them to the cache gateway through the low-latency protocol of the ring subnet. The noise threshold is dynamically set by the signal energy, and the wavelet threshold noise denoising algorithm is used to filter out high-frequency noise in the operation log data. The operation log data and spatial coordinate data are matched by timestamps to construct a spatiotemporal correlation matrix. In the data stream spatiotemporal data acquisition module, the process of extracting the sliding window entropy from the preprocessed operation log data includes: A dual-channel data interface is configured on the heterogeneous data caching gateway. Channel 1 receives preprocessed star-topology backbone operation log data and uses a sliding time window segmentation technique to divide the continuous log stream into discrete batches, based on the peak data stream rate. Dynamically adjust the time window length Within a single window, statistically analyze the distribution of event types and calculate the frequency of occurrence of the i-th type of event. Based on event frequency, the sliding window entropy is obtained through a parallel computing framework. ; In the data stream spatiotemporal data acquisition module, the process of extracting the covariance matrix of each dimension from the preprocessed spatial coordinate data includes: Channel 2 receives the preprocessed spatial coordinate data of the ring subnet, inputs it into the covariance matrix algorithm, calculates the mean and covariance of dimension X and dimension Y, and constructs the covariance matrix of each dimension. If the coordinate values ​​of a spatial coordinate data point deviate from the mean by more than 3 times the standard deviation of the dimension, it is identified as an outlier and removed. In the optimization verification module, the process of constructing the quantum optimization model and outputting the optimization weight parameters includes: The quantum annealing algorithm is used to map the sliding window entropy to a quantum Hamiltonian. Construct coupling coefficients With bias term ; Initialize the spin state of the qubit and set the initial temperature. and final temperature Based on real-time load feedback factor The annealing rate is dynamically adjusted according to an exponential function. A quantum optimization model is generated, and the energy barrier is traversed through the quantum tunneling effect to make the cluster communication system approximate the ground state. The final quantum bit state is then measured, and the final quantum bit state is determined. Mean converted to optimized weight parameters ; The optimization verification module includes the following process for constructing a blockchain verification model and outputting trusted verification results: Based on smart contract technology, the covariance matrix is ​​broadcast to each blockchain verification node through a ring topology subnet. Each blockchain verification node uses a dual SHA-256 chain structure to calculate the hash value of the covariance matrix. If more than 2 / 3 of the blockchain verification nodes return If they match, the data is considered complete. A distributed consensus mechanism is adopted to execute smart contracts in three phases: proposal, verification, and submission, thereby constructing a blockchain verification model. During the proposal phase, the main blockchain validator encapsulates the covariance matrix and hash value into a block proposal and broadcasts it along a circular path; During the verification phase, each node verifies the distribution consistency of the covariance matrix based on the 3σ anomaly removal rule and generates a verification result. If the absolute value of the coordinate value of the k-th data point in dimension i minus the mean of dimension i does not exceed three times the standard deviation of dimension i, the verification result is 1. If the absolute value of the coordinate value of the k-th data point in dimension i minus the mean of dimension i exceeds three times the standard deviation of dimension i, the verification result is 0. During the submission phase, if more than two-thirds of the blockchain verification nodes return a verification result of 1, the smart contract generates a digital signature. And write the signed result (M,N) into the blockchain; Based on the real-time load rate of the ring subnet and load feedback factor Dynamically adjust consensus rounds If consensus is reached, the blockchain verification model outputs a trusted verification result. If consensus fails, the blockchain verification model outputs a trusted verification result. .

2. The cluster aggregation communication system for AI data stream processing according to claim 1, characterized in that: The process of dynamically weighting and optimizing the weight parameters and the trusted verification results in the status monitoring and early warning module, and outputting a comprehensive judgment value, includes: Based on dynamic weighting factors The optimized weight parameters output by the quantum optimization model Trusted verification results output by the blockchain verification model Dynamic weighting is applied, and optimization is performed based on backpropagation of historical errors. Output comprehensive judgment value .

3. A cluster aggregation communication system for AI data stream processing according to claim 2, characterized in that: The status monitoring and early warning module includes the following process: setting an adaptive threshold for the comprehensive judgment value, judging the status of the AI ​​data stream, and issuing tiered early warnings: Based on the sliding window entropy fluctuation range and covariance matrix stability, the threshold interval is dynamically set. ; like If it is, then it is determined to be a normal data stream and output to the execution end. , And the deviation range If this triggers a Level 1 warning, it is determined to be a slightly abnormal data stream. If so, a Level II warning and emergency response order will be triggered, and the data stream will be identified as a serious risk. After each batch of data stream is processed, the error rate E is defined as the ratio of false alarms to total warnings. Based on the error rate correction threshold range, the correction process includes the following: Then the threshold interval is expanded to be ,like Then the threshold range is narrowed down to ,like If so, the threshold range remains unchanged.

4. A cluster aggregation communication system for AI data stream processing according to claim 3, characterized in that: The process of generating a three-dimensional spatial heat map and anomaly event source map in the visualization and tracing module includes: The elements of the covariance matrix are dynamically weighted with the sliding window entropy to generate three-dimensional spatial thermal density values. Based on the collected spatial coordinate data, the three-dimensional spatial thermal density values ​​are mapped to a three-dimensional grid. The thermal distribution between discrete points is filled in by an interpolation algorithm to generate a three-dimensional spatial heat map. Using the spatiotemporal correlation matrix, the hierarchical early warning signal output by the status monitoring and early warning module, and the corresponding time window identifier as input, the system traces back to the original operation log batch in the cache gateway based on the early warning time window identifier, extracts the distribution of abnormal event types and timestamp sequences, filters out coordinate points with spatiotemporal correlation matrix elements greater than 0.7, and marks them as the location where the abnormal event occurred. Construct an abnormal event source map and define abnormal event nodes that include timestamps, sliding window entropy, warning levels, and associated spatial coordinates. If two abnormal event nodes share the same coordinates in the spatiotemporal correlation matrix, the edge weight of the abnormal event source map is determined by the difference in sliding window entropy between the two abnormal events. Nodes and edges are rendered in layers according to their weights, with high-weight edges highlighted in red and low-weight edges faded in gray.

Images