Multifunctional integrated digital network broadcasting system based on intelligent algorithm

Abstract

The invention relates to the technical field of digital network broadcasting, in particular to a multifunctional integrated digital network broadcasting system based on an intelligent algorithm. Comprising a multi-modal data acquisition module, a broadcast content generation module, a broadcast scheduling decision module, an abnormal interference suppression module and a terminal collaborative feedback module. The multi-modal data acquisition module collects multi-source information of a coverage area in real time. And the broadcast content generation module performs intelligent clustering analysis based on the information to generate structured content. And the broadcast scheduling decision module adopts a graph attention scheduling algorithm to optimize a scheduling strategy. And the abnormal interference suppression module identifies an interference source by using a residual convolutional network, and repairs an interfered signal through a recurrent neural network. And the terminal collaborative feedback module executes playing and data feedback, and dynamically adjusts receiving parameters and pushing strategies by means of federal reinforcement learning. The intelligent level, the adaptability, the user satisfaction and the overall service efficiency of the digital network broadcasting system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital network broadcasting, and in particular to a multifunctional integrated digital network broadcasting system based on intelligent algorithms. Background Art

[0002] With the rapid development of information and communication technologies, digital network broadcasting systems have been widely used in a variety of scenarios, including public safety warnings, emergency response instructions, rail transit guidance, and smart campus information releases. Traditional broadcasting systems primarily rely on central servers to push preset content on a scheduled basis. Their core mechanisms typically employ static scheduling strategies, fixed playlists, and one-way transmission channels, making them difficult to adapt to dynamic changes in the actual application environment and user needs. This rigid mechanism not only affects the timeliness and accuracy of information delivery, but also presents significant limitations in the context of network fluctuations, terminal diversity, and content diversity. Currently, the following problems still exist: the existing system mainly relies on manual settings or static templates for the generation and distribution of broadcast content, making it difficult to dynamically adjust the content structure according to the actual environment, and lacks comprehensive perception and intelligent analysis of multi-source heterogeneous data; traditional broadcast scheduling decision-making methods are mostly based on preset rules or simple priority sorting, and cannot achieve dynamic adaptive resource allocation and task optimization based on terminal location, regional congestion status, and user historical response behavior; in complex electromagnetic environments, the existing system has limited ability to identify and suppress abnormal factors such as channel conflicts, signal attenuation, and malicious interference, and lacks efficient anomaly detection and signal repair mechanisms, which can easily lead to distortion, interruption, or even loss of broadcast signals, affecting the continuity and reliability of broadcast services. Summary of the Invention

[0003] To solve the above problems, the present invention provides a multifunctional integrated digital network broadcasting system based on intelligent algorithms, which solves the problem of how to break through the limitations of traditional digital broadcasting systems in content intelligent generation, scheduling adaptive optimization and abnormal interference suppression, and realize the accurate distribution of broadcast content and full-process intelligent management and control under multi-source heterogeneous data fusion, thereby improving the intelligence level, adaptability, user satisfaction and overall service efficiency of the digital network broadcasting system.

[0004] To achieve the above object, the technical solution adopted by the present invention is: A multifunctional integrated digital network broadcasting system based on intelligent algorithms, comprising a multimodal data acquisition module, a broadcast content generation module, a broadcast scheduling decision module, an abnormal interference suppression module and a terminal collaborative feedback module, which are sequentially connected in communication; The multimodal data acquisition module is used to collect multi-source information in the broadcast coverage area in real time; The broadcast content generation module is used to perform cluster analysis based on the semantic features of the content using a sparse tensor semantic clustering algorithm based on the multi-source information, and to construct a content-context label mapping relationship in combination with historical playback data to generate structured broadcast content; The broadcast scheduling decision module is used to construct a broadcast task flow graph with time window constraints based on the structured broadcast content, combined with terminal location, regional congestion status and historical broadcast response data, and adopt a graph attention scheduling algorithm based on asynchronous advantage update to jointly optimize the scheduling strategy of broadcast time window, content priority and terminal scheduling order to generate broadcast scheduling decision instructions; The abnormal interference suppression module is used to identify potential interference sources during broadcast signal transmission using a residual convolutional neural network algorithm and reconstruct the disturbed broadcast segment through a signal repair mechanism driven by a recurrent neural network; the potential interference sources include channel conflicts, signal attenuation, and malicious interference; The terminal collaborative feedback module is used to receive the broadcast signal controlled by the broadcast scheduling instruction, perform playback operations on the terminal, and simultaneously collect terminal playback status data, user interaction feedback data and environmental response data, and dynamically adjust terminal reception parameters and content push strategies through a federated reinforcement learning algorithm.

[0005] Furthermore, the multi-source information includes environmental sound signals, real-time voice signals collected by microphones, video surveillance image data, regional meteorological parameters, electromagnetic interference signal detection data, terminal device geographic location and motion status data, user interaction behavior data and network link status parameters.

[0006] Furthermore, the operation process of the broadcast content generation module includes the following steps: Performing standardized preprocessing on the multi-source information and constructing a multi-order feature tensor including semantic dimension, temporal dimension and modal dimension; Based on the multi-order feature tensor, a sparse tensor semantic clustering algorithm is used to compress and represent semantic units and perform topic clustering. The semantic core units in the multi-source content are identified through low-rank representation and semantic sparsity constraints to form a set of candidate broadcast semantic segments. Based on the candidate broadcast semantic segment set, combined with historical broadcast playback data and terminal response records, a label transfer modeling method is used to construct a content-context label mapping relationship, and a multi-scenario adaptation model based on label weight fitting is used to generate a content-context matching matrix; Based on the candidate broadcast semantic segment set and the content-context matching matrix, a broadcast structure generation mechanism based on constraint satisfaction problem solving is adopted to sequentially reconstruct and format optimize the candidate content to generate structured broadcast content including time series logic, context adaptability and multimodal fusion features.

[0007] Furthermore, the formula of the sparse tensor semantic clustering algorithm is as follows: in, represents the optimal semantic clustering result set; S represents the set of all clustering partitioning schemes; Represents multi-order feature tensor elements; represents the content-context label weight matrix; i represents the semantic unit number; j represents the time window number; k represents the modality number; Represents the data index set within the c-th cluster; Indicates the number of multi-source data in the c-th cluster; represents the indicator vector of the c-th cluster; represents the number of feature items of the cth topic cluster; C represents the final number of topic categories; Represents the sparsity adjustment parameter.

[0008] Furthermore, the operation process of the broadcast scheduling decision module includes the following steps: Based on the structured broadcast content, combined with the spatial distribution of terminals, congestion status within the region, and historical broadcast response data, a broadcast task flow graph with time window constraints is constructed, where nodes represent specific broadcast content, edges represent temporal or logical dependencies between content, and the time window is used to limit the executable time range of each task node; Collect the link accessibility, reception capability, interactive behavior characteristics and content preference information of each terminal, construct the terminal state tensor, and fuse it with the broadcast task graph to generate a state-enhanced scheduling graph; Based on the state-enhanced scheduling graph, a graph attention scheduling algorithm based on an asynchronous advantage update mechanism is adopted to dynamically calculate the scheduling value function of each broadcast task node. Based on the structural correlation between nodes, content priority, and terminal feedback potential, attention weights are assigned. In addition, an asynchronous strategy is used to jointly optimize the allocation of broadcast time windows, content delivery priority sorting, and target terminal scheduling order. Based on the optimization results, a broadcast scheduling decision instruction including content identification, target terminal ID list, scheduled time window and frequency resource allocation information is generated, and the instruction is packaged and issued through the heterogeneous link control module.

[0009] Furthermore, the asynchronous advantage update mechanism specifically includes the following steps: Based on the state-enhanced scheduling graph, for each broadcast task node, based on the node's historical scheduling status, current environment characteristics and terminal feedback data, asynchronously sample the node state transition sequence and dynamically update the scheduling value function of the task node; By taking weighted differences in the scheduling value functions at different time steps, the advantage valuation function is calculated. Combined with the contextual attention weights of each node in the graph structure, local sensitivity adjustment of broadcast content scheduling priority and target terminal selection is achieved. In scenarios with multi-tasking concurrency and resource competition, distributed parallel computing is used to independently optimize the scheduling parameters of each node and update the global scheduling policy parameters through asynchronous feedback. Based on the global scheduling policy parameter update completed by the asynchronous advantage update mechanism, the latest scheduling priority and resource allocation recommendation of the node are integrated into the final broadcast scheduling decision output, including content distribution order, target terminal grouping and time slot resource allocation information, and scheduling instructions are issued to the heterogeneous link control module.

[0010] Furthermore, the formula of the graph attention scheduling algorithm is as follows: in, represents the final scheduling priority score of broadcast task node i; represents the content importance score corresponding to the i-th task node in the structured broadcast content; represents the user activity index in the target area of ​​task i; represents the historical feedback score of content i; represents the current link bandwidth occupancy rate of the geographical area where task i is located; Indicates the maximum link bandwidth capacity; represents the congestion risk coefficient of task i in the state-enhanced scheduling graph; represents the remaining schedulable time window of task i; 、 and represents the score weighting coefficient; represents the exponential amplification factor of the response feedback score; and They represent the slope and center position parameters of the Sigmoid function in the time window urgency score, respectively.

[0011] Furthermore, the operation process of the abnormal interference suppression module includes the following steps: Real-time acquisition of multi-channel signal streams in the broadcast signal transmission link, and the use of multi-scale decomposition methods to extract multi-dimensional characteristic parameters such as signal energy spectrum distribution, instantaneous frequency characteristics, channel utilization and interference index; Based on the residual convolutional neural network algorithm, the above characteristic parameters are subjected to time series modeling and spatial feature fusion. Combined with channel historical data and statistical thresholds, it can automatically detect abnormal signal segments and distinguish and classify interference types. For the detected abnormal sections, an adaptive signal repair mechanism driven by recurrent neural networks is used to reconstruct the disturbed signal content based on the timing pattern and amplitude characteristics of historical reference segments; By comparing the signals before and after reconstruction, combined with the current link dynamics and the target terminal status, an interference suppression control vector is generated, including policy parameters for forwarding delay adjustment, link remapping, frequency band avoidance, and content reconstruction level. The anomaly suppression processing results are synchronously fed back to the terminal collaborative feedback module.

[0012] Furthermore, the terminal collaborative feedback module constructs a policy adaptation mechanism based on the federated reinforcement learning architecture, uses the playback status data, interaction feedback data and environmental perception parameters collected locally by the terminal to periodically optimize the receiving strategy, and realizes cross-terminal policy updates through global policy aggregation.

[0013] The beneficial effects of the present invention are: The present invention uses a multimodal data acquisition module to acquire multi-source heterogeneous information such as audio, video, environment and user behavior in the broadcast area in real time. Combined with the sparse tensor semantic clustering algorithm, it can extract core content features from the semantic level and construct a semantic mapping model between content and context, so that the broadcast content is more in line with the audience's needs and environmental context, significantly improving the effectiveness and acceptance rate of information dissemination. The broadcast scheduling decision module combines the terminal location distribution, regional congestion status and historical response status to construct a task flow graph with time window constraints, and identifies key nodes through the graph attention mechanism. It performs dynamic scheduling optimization through the asynchronous advantage update strategy, effectively avoiding conflicts and duplications in broadcast content, improving the timeliness and accuracy of scheduling, and ensuring that high-priority content is delivered first at key nodes. The abnormal interference suppression module integrates the residual convolutional neural network algorithm to identify potential abnormal sources such as channel conflicts and malicious interference, and uses the time series feature modeling mechanism driven by the recursive neural network to achieve real-time repair and reconstruction of disturbed signal segments, thereby significantly improving the stability and reliability of the broadcast link in a complex transmission environment and reducing the risk of content interruption. The terminal collaborative feedback module not only realizes the real-time collection of terminal playback status and user behavior data, but also makes personalized dynamic adjustments to terminal reception parameters and push content through the federated reinforcement learning mechanism, which not only ensures user privacy and security, but also improves the individual user's receiving experience, enabling the system to have a "perception-feedback-regulation" closed-loop optimization capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a module diagram of a multifunctional integrated digital network broadcasting system based on intelligent algorithms of the present invention.

[0015] Figure 2 It is a flowchart of the operation process of the broadcast content generation module provided by one embodiment of the present invention.

[0016] Figure 3 It is a flowchart of the operation process of the broadcast scheduling decision module provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0017] See also Figure 1-3 As shown, the present invention relates to a multifunctional integrated digital network broadcasting system based on intelligent algorithms.

[0018] Example A multifunctional integrated digital network broadcasting system based on intelligent algorithms, comprising a multimodal data acquisition module, a broadcast content generation module, a broadcast scheduling decision module, an abnormal interference suppression module and a terminal collaborative feedback module, which are sequentially connected in communication; The multimodal data acquisition module is used to collect multi-source information within the broadcast coverage area in real time; the multi-source information includes ambient sound signals, real-time voice signals collected by microphones, video surveillance image data, regional meteorological parameters, electromagnetic interference signal detection data, terminal device geographic location and motion status data, user interaction behavior data and network link status parameters.

[0019] Specifically, this module integrates multiple types of sensors and data collection terminals and is deployed at key nodes within the broadcast coverage area (such as teaching buildings, subway stations, business districts, community entrances and exits, etc.).

[0020] Environmental sound collection: High-sensitivity digital environmental sound sensors are installed at key locations in broadcast coverage areas (such as subway passages, shopping malls, campuses, etc.). A mixed directional and omnidirectional layout is adopted to achieve full-area coverage, and background noise and abnormal sound sources (such as alarms, noise, vehicle horns, etc.) are collected in real time 24 hours a day.

[0021] Microphone real-time voice collection: High-sensitivity array microphones are installed in crowded areas to collect the voices of on-site personnel and sudden verbal commands in real time, providing voice input for incident response and emergency broadcasts.

[0022] Video surveillance image data: Integrates high-definition cameras to collect video information such as crowd distribution, abnormal behavior, and environmental changes in the area, supporting content recognition and event-driven.

[0023] Regional meteorological parameter collection: Deploy multi-parameter sensors such as temperature and humidity, air pressure, wind speed, and rainfall to dynamically collect environmental meteorological data, providing a basis for content push and scheduling optimization in extreme weather conditions.

[0024] Electromagnetic interference signal detection: Establish electromagnetic environment monitoring points, deploy dedicated RF detectors and broadband signal collectors, scan key operating frequency bands and backup frequency bands, and monitor and record signal strength, noise levels, and interference events within specific frequency bands.

[0025] Terminal device geographic and motion data: Through the built-in GPS / Beidou module and accelerometer of the broadcast terminal, the terminal location and movement trajectory are collected in real time to achieve spatial precision of content distribution.

[0026] User interaction behavior data: The terminal side integrates touch, button, and voice interaction interfaces to collect user feedback, instructions, and interaction data on broadcast content.

[0027] Network link status parameters: Periodically detect the bandwidth, latency, and packet loss rate of wireless and wired network links in the area to assess network quality and provide real-time support for scheduling and push strategies.

[0028] All collected data is initially aggregated and pre-processed (such as data cleaning, anomaly removal, and feature compression) at the local edge node and securely uploaded to the system data center.

[0029] The broadcast content generation module is used to perform cluster analysis based on the semantic features of the content using a sparse tensor semantic clustering algorithm based on the multi-source information, and to construct a content-context label mapping relationship in combination with historical playback data to generate structured broadcast content; The operation process of the broadcast content generation module includes the following steps: Performing standardized preprocessing on the multi-source information and constructing a multi-order feature tensor including semantic dimension, temporal dimension and modal dimension; Specifically, data from different sources (such as text, audio, video, and environmental parameters) undergoes format unification, missing value completion, anomaly detection, and filtering. For example, audio signals undergo noise suppression and normalization, video frames undergo resolution unification, text data undergoes unified character encoding, and environmental parameters undergo anomaly removal and interpolation completion.

[0030] Align all modal features to form a three-dimensional or higher-dimensional tensor with the following axes: Semantic dimension (different semantic units, such as keywords, topic tags, etc.) Time dimension (timing information of data segments, ensuring event order and context) Modality (data source type, such as text, audio, video, environment, etc.) Sparsity detection and indexing of tensor data structures facilitate subsequent efficient clustering and topic discovery.

[0031] Based on the multi-order feature tensor, a sparse tensor semantic clustering algorithm is used to compress and represent semantic units and perform topic clustering. The semantic core units in the multi-source content are identified through low-rank representation and semantic sparsity constraints to form a set of candidate broadcast semantic segments. Specifically, on the data center server, the constructed multi-order feature tensor is first checked for structural integrity, removing extremely sparse dimensions caused by acquisition anomalies to ensure the accuracy of subsequent clustering. Using efficient sparse tensor analysis toolkits (such as Tensorly), the tensor data is gradually decomposed, automatically selecting the most representative thematic feature components with a high signal-to-noise ratio, achieving high-density aggregation and noise reduction of semantic information.

[0032] Based on the topic feature space derived from tensor decomposition, the system automatically distinguishes between global topics (such as emergency alerts, traffic reports, and regular notifications) and localized topic areas (such as temporary events at specific locations or during specific time periods). It employs various distribution-independent clustering strategies, such as spectral clustering and density clustering, to ensure that even small but important marginal events or niche content can be identified and extracted, preventing mainstream data topics from drowning out long-tail information.

[0033] The system traces the source of each clustering result, automatically locating high-weighted segments within the original multi-source data (e.g., a speech clip that elicits a high response, a video clip containing unusual behavior, or the range of parameter changes corresponding to a sudden meteorological event). A complete metadata structure is generated for each candidate segment, including the source modality, start and end times, key semantic tags, and relevant contextual descriptions. Some segments support multimodal cross-verification (e.g., the same event is simultaneously tagged with audio, video, and environmental sensor data). All candidate segments are pushed to the review interface for manual review or automated filtering, resulting in a "semantic segment whitelist" that can ultimately participate in content generation.

[0034] Based on the candidate broadcast semantic segment set, combined with historical broadcast playback data and terminal response records, a label transfer modeling method is used to construct a content-context label mapping relationship, and a multi-scenario adaptation model based on label weight fitting is used to generate a content-context matching matrix; Specifically, the system automatically captures all recent historical broadcast content and its corresponding terminal feedback data (such as user click-through rates, completion rates, interaction frequency, complaints or likes, etc.). Using the "content segment - contextual tag - feedback performance" key, a high-dimensional data table is created within the knowledge base, mapping content to actual application scenarios. External data sources (such as holiday schedules, weather service APIs, and city event calendars) are automatically incorporated to enrich the semantic breadth and adaptability of the contextual tag library.

[0035] For the newly generated candidate semantic segments, the system uses methods such as semantic embedding similarity analysis and contextual behavior feature comparison to migrate the effective context labels of historical content to the new segments. Each segment can obtain multiple alternative context labels and their adaptation confidence, and the system automatically adjusts the label weights based on historical data. For example, if a certain content has a high playback response during the "rainy morning rush hour", the corresponding context label weight will be increased. For cold-start content that cannot be automatically adapted, the system can use a weak label recommendation mechanism to prioritize the context labels with the highest similarity, and simultaneously collect terminal responses after the first push to quickly complete the label weights.

[0036] The system constructs a real-time, updatable, multidimensional adaptation matrix, with content segments as rows and contextual tags as columns. This matrix not only incorporates historical statistical fitness but also dynamically weights factors such as real-time terminal responses and external events, ensuring that matrix weights are continuously optimized as the environment and user needs change. This content-context matching matrix supports fast retrieval and reverse indexing, facilitating the subsequent scheduling module's real-time push and distribution based on scenario-specific content fitness.

[0037] Based on the candidate broadcast semantic segment set and the content-context matching matrix, a broadcast structure generation mechanism based on constraint satisfaction problem solving is adopted to sequentially reconstruct and format optimize the candidate content to generate structured broadcast content including time series logic, context adaptability and multimodal fusion features.

[0038] Specifically, the system defines multiple constraints for each broadcast content generation task, such as total duration (must not exceed the limit), topic coverage (must include key events), redundancy (the same topic cannot be broadcast consecutively), and content mutual exclusion (for example, different types of warnings cannot be broadcast simultaneously). User groups, scene types, and regional special needs are input as additional constraints to guide content assembly and editing.

[0039] Based on constraints, a priority queue and multi-round screening algorithm are used to optimally sort the candidate semantic segments, ensuring that emergency and high-priority content is prioritized, regular content is appropriately interspersed, and less-focused content is automatically delayed or consolidated. Auxiliary segments such as intelligent transitional phrases (such as "The following is a weather warning, please be safe"), time prompts, and associated sound effects are inserted to enhance the overall playback logic and auditory / visual experience. Specialized adaptations are made to content combinations for special scenarios (such as emergency broadcasts and nighttime silent mode), such as prioritizing text content, adjusting speech speed, and automatically reducing noise.

[0040] Based on the type and capabilities of the push device (e.g., smart speakers, mobile apps, LED screens, etc.), each piece of content is assigned an appropriate multimodal output format. This allows for the co-packaging of text, audio, images, and video (e.g., text with audio, video with ambient sound, and image carousels), supporting simultaneous display or differentiated presentation across multiple devices. Detailed metadata, such as content ID, topic, priority, push timeframe, target device, applicable context, and redundancy checksum, is embedded within each structured content package, facilitating downstream system scheduling, distribution, monitoring, and feedback loops.

[0041] All generated structured broadcast content is archived in the content library. The system automatically adds a "recommended" tag to highly responsive content, enabling automatic reuse and continuous iterative optimization. Archived content also supports tag updates and context-adaptive historical tracking, facilitating subsequent system evaluation of push strategy effectiveness, enabling self-evolution and precise service delivery.

[0042] Furthermore, the formula of the sparse tensor semantic clustering algorithm is as follows: in, represents the optimal semantic clustering result set; S represents the set of all clustering partitioning schemes; Represents multi-order feature tensor elements; represents the content-context label weight matrix; a represents the semantic unit number; e represents the time window number; d represents the modality number; Represents the data index set within the c-th cluster; Indicates the number of multi-source data in the c-th cluster; represents the indicator vector of the c-th cluster; represents the number of feature items of the cth topic cluster; C represents the final number of topic categories; Represents the sparsity adjustment parameter.

[0043] The broadcast scheduling decision module is used to construct a broadcast task flow graph with time window constraints based on the structured broadcast content, combined with terminal location, regional congestion status and historical broadcast response data, and adopt a graph attention scheduling algorithm based on asynchronous advantage update to jointly optimize the scheduling strategy of broadcast time window, content priority and terminal scheduling order to generate broadcast scheduling decision instructions; The operation process of the broadcast scheduling decision module includes the following steps: Based on the structured broadcast content, combined with the spatial distribution of terminals, congestion status within the region, and historical broadcast response data, a broadcast task flow graph with time window constraints is constructed, where nodes represent specific broadcast content, edges represent temporal or logical dependencies between content, and the time window is used to limit the executable time range of each task node; It should be noted that the system first automatically parses the attributes of each structured broadcast content, including content theme, target audience, priority, expected duration, recommended playback scenario, compatible terminal type, emergency or general tags, etc. Each broadcast content is treated as an independent task node, with all scheduling-related metadata embedded in the node and assigned a globally unique identifier. For example, the "Emergency Weather Alert" node is marked as high priority and required to be pushed within 10 minutes. Content with repeated themes or overlapping time is deduplicated and merged to reduce redundant nodes in the task flow graph and optimize subsequent scheduling efficiency.

[0044] The system aggregates the spatial locations of all terminals (using GPS, WiFi, and Bluetooth positioning) and their corresponding physical areas (such as buildings, subway stations, and shopping malls) in real time. It collects data on the number of online terminals, the current active terminal ratio, wireless link channel utilization, and historical broadcast peak periods within each area to assess local resource pressure. Areas with high terminal density, severe channel congestion, or high mobility are automatically marked and prioritized for scheduling.

[0045] For content that needs to be played sequentially (such as announcements, event descriptions, and emergency response procedures), the system automatically identifies logical dependencies between content through semantic analysis and historical playback sequences, establishing directed edge associations. Constraints such as "only one emergency broadcast allowed at a time" are annotated to prevent content mixing and audience interference. Based on the urgency of the content, historical response time distribution, and user habits, the earliest and latest available push times are set for each task node, forming a task execution window. For example, traffic information should be pushed during the morning rush hour, while silent / text notifications should be pushed at night.

[0046] The system retrieves terminal response data for historical broadcast tasks, including actual task completion time, playback success rate by audience area, and user feedback (such as likes and complaints). Based on historically slow response areas or time periods, the time window for relevant task nodes is shifted forward or extended. For tasks with excellent historical feedback, the time window is appropriately shortened, improving overall system timeliness and satisfaction. The system supports real-time visualization of task flow diagrams, allowing dispatchers to easily identify key task links, bottleneck nodes, and high-risk areas.

[0047] Collect the link accessibility, reception capability, interactive behavior characteristics and content preference information of each terminal, construct the terminal state tensor, and fuse it with the broadcast task graph to generate a state-enhanced scheduling graph; Specifically, the terminal's signal strength, current bandwidth, available network type (such as 5G, WiFi, wired, etc.), latency and packet loss rate are periodically detected to identify terminals with poor or restricted networks. The terminal's CPU, memory, storage utilization, battery power and temperature rise are collected, and devices with tight resources or possible failures are automatically downgraded to avoid pushing high-load content. The terminal regularly uploads user interaction behaviors with the terminal (such as active playback, pause, skipping, repeat playback, rejection / closing of content, etc.). The system collects statistics on terminal activity, response delay, preferred content type and other indicators, and automatically grades the terminal's response capabilities to broadcasts. The terminal's historical content response curve is continuously tracked (such as liking news, music, emergency information, advertising, etc.), and preference tags and weighted recommendation parameters are assigned to the terminal.

[0048] Each terminal's state characteristics (network, device, interaction, preferences, etc.) are standardized and mapped into a state vector. All terminal state vectors are stacked by terminal ID primary key, forming a high-dimensional tensor with the structure [terminal ID, feature category, time slice]. This allows tracking of historical state changes and periodic behavior, facilitating anomaly detection and trend analysis. Terminals with extreme anomalies in the tensor (such as prolonged offline or persistent load abnormalities) are automatically signaled, temporarily excluding them from subsequent scheduling or reducing their task push priority.

[0049] For each content task node, a group of the most matching terminals is dynamically screened based on its push target, adapted terminal type, and historical terminal response records, and a content-terminal mapping relationship is established. Each task node is associated with the real-time status characteristics of its target terminal. The graph node attributes are no longer a single content label, but are embedded with multi-dimensional contextual information such as the actual reachability, load capacity, and preference matching of the current terminal. The entire scheduling graph supports visual display, and the system can refresh the attributes of each node and edge in real time. For example, node color changes indicate status changes, and edge thickness reflects terminal availability or congestion. For terminals that may fail or are prone to errors, the system can automatically specify alternative terminals to achieve fault-tolerant redundancy of scheduling tasks, improve broadcast arrival rate and system robustness.

[0050] Based on the state-enhanced scheduling graph, a graph attention scheduling algorithm based on an asynchronous advantage update mechanism is adopted to dynamically calculate the scheduling value function of each broadcast task node. Based on the structural correlation between nodes, content priority, and terminal feedback potential, attention weights are assigned. In addition, an asynchronous strategy is used to jointly optimize the allocation of broadcast time windows, content delivery priority sorting, and target terminal scheduling order. Specifically, in the state-enhanced scheduling graph, the system assigns initial attention weights to each content node, terminal node, and the edges between them, reflecting the structural relevance between nodes, content priority, historical feedback potential, etc. If a content node is strongly connected to a highly active and high-feedback terminal, its scheduling priority will be increased.

[0051] The system utilizes a multi-threaded / distributed architecture, independently sampling the scheduling status of each content node (including historical scheduling, current terminal status, regional congestion changes, etc.), and asynchronously updating the node's scheduling value function. For each task node, the system asynchronously samples its state transition sequence at different time points and for different target terminals, accumulating scheduling feedback information derived from scheduling history and environmental changes.

[0052] The advantage valuation function is calculated by taking a weighted differential of the scheduling value function (e.g., "new policy value minus average historical value"). This result, combined with the node's contextual attention weight, dynamically adjusts the content push order, time window allocation, and target terminal scheduling sequence, achieving adaptive sensitivity optimization for emergencies, hot topics, and special areas. For example, if a terminal has been particularly responsive to similar content in recent times, its scheduling weight is increased, and its assigned content time window and priority are simultaneously optimized.

[0053] In scenarios like regional hotspots and high content concurrency, the system concurrently optimizes the scheduling parameters for each content node and corresponding terminal. All optimization results are asynchronously transmitted back to the global scheduling parameter library, eliminating the need to wait for all nodes to synchronize, improving the system's real-time performance and scalability. After each global parameter update, node priorities, content terminal assignments, and resource allocation recommendations are reorganized to prepare for scheduling instruction generation.

[0054] Based on the optimization results, a broadcast scheduling decision instruction including content identification, target terminal ID list, scheduled time window and frequency resource allocation information is generated, and the instruction is packaged and issued through the heterogeneous link control module.

[0055] Specifically, the system generates a broadcast scheduling instruction set based on the optimization results. This includes a unique identifier for each piece of content, a list of target terminal IDs, the specific time window allocated, and resource configuration information such as frequency / bandwidth. The instruction set can be grouped by multiple criteria, including region, terminal type, and content urgency.

[0056] Scheduling commands are encapsulated through the heterogeneous link control module, ensuring compatibility with various terminals (such as 4G / 5G, WiFi, wired Ethernet, LoRa, and other links) and device protocols from different manufacturers. During encapsulation, the command format is automatically adjusted based on the target terminal capabilities and real-time network conditions. If some terminals only support text content, the system automatically transcodes and prunes non-essential content.

[0057] After encapsulation is complete, the instructions are dispatched to the target terminal via the dispatch gateway and link distribution mechanism, pushed in batches, and support breakpoint resumption and retransmission fault tolerance. The system monitors the delivery of instructions and the execution status of the terminal in real time, collects execution logs and feedback information, and automatically issues an alarm or reissues in the event of anomalies, forming a complete closed loop.

[0058] The asynchronous advantage update mechanism specifically includes the following steps: Based on the state-enhanced scheduling graph, for each broadcast task node, based on the node's historical scheduling status, current environment characteristics and terminal feedback data, asynchronously sample the node state transition sequence and dynamically update the scheduling value function of the task node; Specifically, each broadcast task node (e.g., a piece of content to be broadcast) is managed by an independent thread or scheduling agent, allowing each node to initiate state sampling without waiting for all nodes to synchronize. Nodes periodically access the scheduling graph, terminal state tensors, and the latest environmental parameters to autonomously collect their own scheduling status, such as the time of the most recent push, the reception success rate of associated terminals, congestion in the region, and the urgency of the content. If the content belonging to the node experienced a failure or response delay in the previous push, the node can autonomously mark the current status as "pending optimization," triggering a self-correction strategy.

[0059] After collecting data, nodes compare the current scheduling status with a historical sampling sequence (e.g., the performance of each scheduling cycle within a week) to identify recent trends in the environment and terminal response, such as improved terminal response, improved regional network performance, or changes in content demand. Through comparative analysis, nodes can promptly identify fluctuations in their own scheduling value (such as push completion rate, timeliness, and user feedback), providing data support for subsequent dynamic adjustments to the value function.

[0060] By taking weighted differences in the scheduling value functions at different time steps, the advantage valuation function is calculated. Combined with the contextual attention weights of each node in the graph structure, local sensitivity adjustment of broadcast content scheduling priority and target terminal selection is achieved. Specifically, each broadcast task node is equipped with a local state cache that automatically records the scheduling performance of the most recent (e.g., 10 or 30) scheduling cycles, including multi-dimensional data such as the push period, the number of assigned terminals, actual coverage, user feedback scores, and terminal response speed. After a new scheduling, the scheduling node will automatically compare the current scheduling performance with the historical average, historical maximum / minimum performance, etc. For example, if the user feedback for this content push is significantly better than the historical average, it will be recorded as a "positive advantage" event. For new content, cold start nodes, or nodes whose performance changes drastically due to sudden changes in the environment, the system allows automatic adjustment of the comparison window length to improve the speed of adaptation to emergencies.

[0061] Nodes dynamically calculate the "weighted difference" between the current scheduling value and the historical average, focusing not only on absolute improvement but also adjusting the differential influence based on the actual weight of the content, time sensitivity, and regional specificity. For example, if a piece of content pushed during the morning rush hour receives better feedback than similar content in the past, the differential weight will be increased more significantly, and the scheduling system will prioritize pushing this content during subsequent peak hours. For certain situations where traffic surges briefly but then declines over a long period, the system combines moving averages, sliding windows, and threshold mechanisms to prevent occasional data noise from affecting the overall scheduling strategy.

[0062] When scheduling a node, the system proactively reads the recent scheduling status and attention weights of adjacent nodes in the task flow graph (e.g., logical dependencies, topic associations, temporal proximity, etc.). If a node's push is delayed due to a recent increase in the priority of a preceding node, the system automatically identifies this and contextually fine-tunes the node's scheduling priority to prevent critical content from being delayed for extended periods. For nodes with hot topics, urgent content, or frequent user feedback, the attention weights of related nodes are also dynamically increased, forming a multi-dimensional scheduling association based on semantics, logic, and time.

[0063] Combining weighted differentials and contextual attention, the system dynamically updates node priorities, target terminal combinations, and content distribution windows during each round of scheduling decisions. If a terminal is detected to have recently provided multiple positive responses to similar content, the system automatically increases its distribution weight for that content; otherwise, it devalues ​​it, ensuring that scheduling resources are allocated to high-potential targets. All fine-tuning and policy changes are recorded with detailed reasons and supporting data to facilitate subsequent system evaluation and policy optimization.

[0064] In scenarios with multi-tasking concurrency and resource competition, distributed parallel computing is used to independently optimize the scheduling parameters of each node and update the global scheduling policy parameters through asynchronous feedback. Specifically, each broadcast task node operates independently within a distributed architecture and can be deployed on different servers, edge nodes, or cloud platforms, achieving true physical parallelism and fault-tolerant isolation. Each node has a local optimization process responsible for collecting required environmental data, terminal status, and scheduling feedback in real time, and autonomously adjusting parameters (such as push timing, terminal allocation, content order, and resource pre-allocation). Nodes only need to asynchronously upload optimization results at a set frequency (e.g., every 5 seconds, every minute) or upon conditional triggers (e.g., changes in key indicators), without having to wait for global synchronization.

[0065] The latest optimization results of all nodes, including scheduling priorities, resource allocation recommendations, and attention weights, are transmitted back to the global scheduling center in real time via secure, low-latency message queues or API interfaces. The scheduling center automatically maintains parameter version numbers and timestamps for each node. When multiple nodes compete for the same terminal resource, the highest priority or most recent allocation request is prioritized, and conflict resolution results are synchronously fed back to the relevant nodes. For large-scale scenarios (such as hundreds of broadcast task nodes and thousands of terminals), the system supports asynchronous batch transmission and rapid local parameter updates, greatly improving overall scheduling responsiveness and concurrency capabilities.

[0066] If a node temporarily fails to transmit its latest parameters due to network failures, unexpected crashes, or other reasons, the dispatch center will temporarily use its most recently transmitted valid parameters, or use the locally optimal historical parameters, as a temporary scheduling basis to ensure global scheduling continuity and uninterrupted service. Upon recovery, the node automatically retransmits the missing parameters and scheduling logs. The dispatch center reassesses the validity of the parameters based on actual performance to ensure the long-term optimal global strategy. All anomalies and compensation actions are automatically recorded in the log system for easy retrospective review and system health assessment.

[0067] The dispatch center periodically aggregates, resolves conflicts, and maintains consistency across all returned optimization parameters, applying rules such as maximum priority, timing consistency, and resource mutual exclusion. This aggregated global scheduling policy is used to generate broadcast scheduling instructions for the next cycle, covering all key parameters such as content push order, terminal grouping, time slot allocation, and resource reservation. Changes to the global policy are immediately pushed to all nodes and terminals, ensuring a dynamic, controllable, and highly adaptive scheduling strategy.

[0068] Based on the global scheduling policy parameter update completed by the asynchronous advantage update mechanism, the latest scheduling priority and resource allocation recommendation of the node are integrated into the final broadcast scheduling decision output, including content distribution order, target terminal grouping and time slot resource allocation information, and scheduling instructions are issued to the heterogeneous link control module.

[0069] Specifically, the global scheduling center regularly or on demand aggregates the latest scheduling parameters for all nodes, comprehensively ranking and centrally planning scheduling priorities, target terminal grouping, and time slot resource allocation for all nodes. The system considers inter-node resource competition and current environmental constraints (such as frequency, bandwidth, and total task duration) to rationally coordinate resource requests from each node, preventing resource conflicts and scheduling bottlenecks.

[0070] The dispatch center generates the final set of scheduling instructions based on the latest global parameters, including detailed information such as the content distribution order, push grouping for each terminal, specific playback time windows, and link allocation strategies. All instructions are encapsulated by the heterogeneous link control module and then issued to each terminal. The system monitors the issuance and execution results. Execution feedback (such as push success, terminal non-response, and content loss) is collected and analyzed in real time to provide basic data for the next round of asynchronous advantage updates.

[0071] The system continuously tracks scheduling feedback, terminal behavior, and network environment changes. Through continuous asynchronous updates, it enables self-learning and adaptive evolution of scheduling strategies. In the event of emergencies, significant traffic changes, or the addition or removal of new terminals, the scheduling system can quickly adjust itself to ensure that broadcast tasks consistently and reliably reach the target terminal population.

[0072] Furthermore, the formula of the graph attention scheduling algorithm is as follows: in, represents the final scheduling priority score of broadcast task node i; represents the content importance score corresponding to the i-th task node in the structured broadcast content; represents the user activity index in the target area of ​​task i; represents the historical feedback score of content i; represents the current link bandwidth occupancy rate of the geographical area where task i is located; Indicates the maximum link bandwidth capacity; represents the congestion risk coefficient of task i in the state-enhanced scheduling graph; represents the remaining schedulable time window of task i; 、 and represents the score weighting coefficient; represents the exponential amplification factor of the response feedback score; and They represent the slope and center position parameters of the Sigmoid function in the time window urgency score, respectively.

[0073] The calculation formula is as follows: in, represents the context label set corresponding to broadcast content i; Indicates the semantic matching strength between content i and label k; Represents the historical response frequency index corresponding to label k; Represents the weighting coefficient (0-1) of semantic similarity and historical click frequency, which is used to control the weight ratio of semantic structure and behavioral feedback.

[0074] The calculation formula is as follows: in, represents the dependency path depth of task i in the broadcast task flow graph; represents the local task density of task i and the number of its adjacent nodes (upstream and downstream tasks), which is used to evaluate the logical coupling between tasks; Indicates the number of broadcast tasks concurrent with task i in the current period; and Represents the adjustment weight parameter that depends on depth and local density.

[0075] The abnormal interference suppression module is used to identify potential interference sources during broadcast signal transmission using a residual convolutional neural network algorithm and reconstruct the disturbed broadcast segment through a signal repair mechanism driven by a recurrent neural network; the potential interference sources include channel conflicts, signal attenuation, and malicious interference; The operation process of the abnormal interference suppression module includes the following steps: Real-time acquisition of multi-channel signal streams in the broadcast signal transmission link, and the use of multi-scale decomposition methods to extract multi-dimensional characteristic parameters such as signal energy spectrum distribution, instantaneous frequency characteristics, channel utilization and interference index; Specifically, multi-channel, high-sensitivity acquisition units are deployed at key nodes in the broadcast signal transmission chain (such as base stations, relays, and terminal front-ends). These units support the simultaneous acquisition of raw signal streams from the primary, backup, and ambient channels. Each signal acquisition channel includes essential parameters such as timestamp, frequency band identifier, signal amplitude, and phase. The system adaptively adjusts the sampling rate, automatically increasing sampling density during periods of abnormality or high risk.

[0076] For the raw signals collected from each channel, multi-scale analysis methods such as wavelet decomposition and empirical mode decomposition (EMD) are used to decompose the signal energy at different time scales, capturing both sudden anomalies and long-term trends. Features such as the energy spectrum distribution, signal dominant frequency, and instantaneous frequency drift are calculated for each decomposition layer, with a focus on extracting anomalies such as local amplitude surges and frequency spurious signals related to interference. Combined with link status data, channel utilization (such as spectrum occupancy and signal passband utilization) and interference indices (such as SINR and BER) are simultaneously extracted to form a complete set of multi-dimensional feature parameters that serve as input for subsequent intelligent recognition.

[0077] Based on the residual convolutional neural network algorithm, the above characteristic parameters are subjected to time series modeling and spatial feature fusion. Combined with channel historical data and statistical thresholds, it can automatically detect abnormal signal segments and distinguish and classify interference types. Specifically, multidimensional feature parameters are encoded separately by time series and spatial channels, forming a three-dimensional tensor of [time, channel, feature category]. A multi-layer convolutional neural network (ResNet-CNN) with a residual structure is used to simultaneously extract temporal patterns (such as abnormal waveforms and signal mutations) and spatial collaborative features (such as coherence between multiple channels and noise coupling between different links in the same frequency band). This residual structure effectively avoids the degradation of deep models and improves detection robustness in complex interference environments.

[0078] During the training phase, the neural network model is calibrated using historical channel data (e.g., typical examples of channel conflict, signal attenuation, and malicious interference). During runtime, the model identifies the real-time input feature tensors and outputs anomaly probabilities and classification results for each time segment in the signal stream. If the anomaly probability exceeds a dynamic threshold, the system automatically locks onto the corresponding signal segment and labels the interference type (e.g., channel conflict, signal attenuation, malicious injection) based on the model's identification results.

[0079] For each abnormal event detected, the system automatically records the start and end time, associated channels, main abnormal characteristics and preliminary classification results, and archives them in the anomaly detection log for subsequent tracing and statistical analysis.

[0080] For the detected abnormal sections, an adaptive signal repair mechanism driven by recurrent neural networks is used to reconstruct the disturbed signal content based on the timing pattern and amplitude characteristics of historical reference segments; Specifically, for each signal segment marked as abnormal, the system retrieves recent historical "healthy" signal segments for that link, prioritizing samples with amplitude and frequency patterns similar to those used in the current task. These historical reference segments are used to assist in signal reconstruction, improving the similarity and coherence of the repair.

[0081] Using recurrent neural networks such as LSTM or GRU, the model is fed with contextual data from historical reference segments and the current anomalous segment. Based on temporal correlation and amplitude dynamics, the model generates or corrects the signal content of the anomalous segment frame by frame, including amplitude restoration, missing data completion, and transition smoothing. The repair process supports collaborative modeling of multi-channel signals, ensuring consistency and synchronization of signals across all links.

[0082] The repair strategy is dynamically adjusted based on the anomaly type and impact range. For minor noise interference, interpolation and smoothing can be used; for severe signal loss, deep generative repair or switching to backup channel data is used. The repaired signal is evaluated for differences with the original reference signal, triggering a secondary repair if necessary or flagging a severe anomaly requiring manual intervention.

[0083] By comparing the signals before and after reconstruction, combined with the current link dynamics and the target terminal status, an interference suppression control vector is generated, including policy parameters for forwarding delay adjustment, link remapping, frequency band avoidance, and content reconstruction level. The anomaly suppression processing results are synchronously fed back to the terminal collaborative feedback module.

[0084] Specifically, the system automatically compares key signal indicators (such as energy spectrum, signal integrity, and bit error rate) before and after repair, and combines current link dynamics (such as remaining channel bandwidth and latency changes) with the target terminal status (such as playback failures and feedback anomalies) to evaluate the effectiveness of signal repair and determine whether subsequent link strategies need to be dynamically adjusted.

[0085] Based on the comprehensive evaluation results, the interference suppression control vector is automatically generated, which contains multi-level strategy parameters, such as: Forwarding delay adjustment: Extend or shorten the link waiting time of broadcast tasks to avoid high-interference time slots.

[0086] Link remapping: Temporarily switch to a backup link, redundant channel, or multi-link concurrent forwarding to improve content delivery rate.

[0087] Frequency band avoidance: Dynamically hops to available frequency bands with lower interference, reducing the probability of channel conflicts.

[0088] Content reconstruction level: Set the content reconstruction depth (such as full-clip repair, partial smoothing, degraded push, etc.) to balance content integrity and system resource overhead.

[0089] All control parameters are encapsulated as standard control vectors and sent to relevant broadcast nodes and link scheduling modules.

[0090] Anomaly suppression results are synchronized in real time to the terminal collaborative feedback module, supporting terminal-side collection of anomaly recovery status and user feedback. The system periodically collects terminal feedback to evaluate the effectiveness of suppression strategies, providing a closed-loop data loop for subsequent model training and strategy fine-tuning.

[0091] The terminal collaborative feedback module is used to receive the broadcast signal controlled by the broadcast scheduling instruction, perform playback operations on the terminal, and simultaneously collect terminal playback status data, user interaction feedback data and environmental response data, and dynamically adjust the terminal reception parameters and content push strategy through the federated reinforcement learning algorithm; the terminal collaborative feedback module constructs a strategy adaptation mechanism based on the federated reinforcement learning architecture, uses the playback status data, interaction feedback data and environmental perception parameters collected locally by the terminal to periodically optimize the reception strategy, and realizes cross-terminal policy updates through global strategy aggregation.

[0092] Specifically, after receiving the system's broadcast scheduling instructions, the terminal device automatically parses the pushed content and calls the corresponding broadcast control component based on the content type (text, audio, video, mixed graphics, etc.). It supports various broadcast strategies such as timing, interstitial, and looping to ensure that content is played on time and according to priority. It collects real-time information about the terminal's current playback status (playback progress, content type, volume setting, playback anomalies), device health (power supply, battery, storage space, CPU / memory utilization), and network connection quality (bandwidth, signal strength, latency, packet loss, etc.). It also integrates environmental acoustic sensors, temperature and humidity sensors, and collects the real-time acoustic environment (such as background noise, user conversations, emergency alarm sounds, etc.), temperature, humidity, and light intensity of the space where the terminal is located, providing an environmental baseline for subsequent content adaptation and interference detection.

[0093] The terminal features multiple interactive channels, including a touch screen, physical buttons, and voice input, allowing users to provide feedback on broadcast content (e.g., confirming content, skipping, repeating playback, making complaints, liking, rating, and tagging). Using algorithms like voice recognition and semantic analysis, it captures key intent and emotional tone (e.g., satisfaction, dissatisfaction, urgency, and confusion) in user feedback. The video terminal can also be equipped with an optional camera for facial expression recognition to assess user attention and engagement (subject to regulatory compliance). For specialized scenarios like emergency broadcasts and disaster warnings, it collects user emergency response actions (e.g., confirming escape routes, reporting locations, and making SOS calls) and feedback, enhancing the broadcast system's emergency response capabilities.

[0094] Based on a federated reinforcement learning architecture, each terminal utilizes locally collected playback data, interactive feedback, and environmental parameters to build a user preference model and content push strategy. Real-time feedback signals (such as user-initiated playback, positive feedback, and playback interruptions) are used to adaptively adjust parameters such as content category, playback time, volume control, and push frequency to continuously improve the user experience. Parameters and push strategies are dynamically adjusted based on specific scenarios for different environments (such as noisy environments, quiet spaces, and outdoors), different user groups (such as the elderly, students, and merchants), and different terminal forms (mobile devices, wall-mounted terminals, and public large screens). The system periodically detects playback anomalies (such as freezes, frame drops, silence, and playback failures) and network anomalies (such as disconnections, low speeds, and handoffs) locally, automatically switching to locally cached content, adjusting playback plans, attempting to resume playback from breakpoints, and promptly reporting any anomalies.

[0095] The terminal side only uploads local model parameters or policy gradients, and does not upload original user data. The central server aggregates model updates from multiple terminals, and after completing global model aggregation, it simultaneously sends the optimization results to each terminal to achieve system-level collaborative adaptation. When the global strategy is significantly optimized or specific content strategies need to be urgently adjusted, the system supports real-time distribution of policy parameters to the target terminal to ensure the timeliness and effectiveness of push notifications for emergencies and dynamic hot content. Terminals are grouped based on factors such as the terminal's region, user profile, and historical performance to achieve personalized and differentiated management of content push, thereby improving the overall push efficiency and satisfaction of the system.

[0096] Automatically generate security logs and report any locally detected abnormalities and system security risks (such as forged commands, terminal tampering, and unauthorized access) to the backend, enabling rapid system response and secure self-healing. Backend administrators can view the status of each terminal in real time through the system platform and remotely issue control commands, such as terminal reboots, content updates, policy switching, and blacklisting, ensuring controllable and robust system operation.

[0097] In summary, the present invention realizes comprehensive dynamic perception of the broadcast coverage area through the multimodal data acquisition module, which integrates multi-source information such as environmental audio, real-time voice, video images, meteorological parameters, electromagnetic interference signals, user behavior and terminal status.

[0098] This paper introduces a sparse tensor semantic clustering algorithm, constructing a semantic-temporal-modal third-order tensor based on multimodal features, mining representative candidate broadcast segments, and achieving precise matching of content and scenarios through label transfer modeling, thereby generating structured and context-adaptive broadcast content. The scheduling module uses a graph attention scheduling algorithm based on asynchronous advantage updates, integrating terminal distribution, network status, and historical feedback to construct a state-enhanced task graph, achieving joint optimization of time windows, push order, and terminal resources, improving the accessibility of broadcast tasks and the overall load balancing capability of the system.

[0099] The abnormal interference suppression module in this invention combines a residual convolutional neural network with a recurrent neural network to identify multiple types of interference sources in the signal and perform adaptive repair, ensuring the stable transmission of broadcast content in complex electromagnetic environments. The terminal collaborative feedback module builds a federated reinforcement learning architecture, combining local playback status, user feedback, and environmental parameters to optimize reception strategies while protecting data privacy. Through global model aggregation, terminal-side self-learning and continuous optimization of strategies are achieved.

[0100] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A multifunctional integrated digital network broadcasting system based on intelligent algorithm, characterized in that: It includes a multimodal data acquisition module, a broadcast content generation module, a broadcast scheduling decision module, an abnormal interference suppression module and a terminal collaborative feedback module that are sequentially connected in communication; The multimodal data acquisition module is used to collect multi-source information in the broadcast coverage area in real time; The broadcast content generation module is used to perform cluster analysis based on the semantic features of the content using a sparse tensor semantic clustering algorithm based on the multi-source information, and to construct a content-context label mapping relationship in combination with historical playback data to generate structured broadcast content; The broadcast scheduling decision module is used to construct a broadcast task flow graph with time window constraints based on the structured broadcast content, combined with terminal location, regional congestion status and historical broadcast response data, and adopt a graph attention scheduling algorithm based on asynchronous advantage update to jointly optimize the scheduling strategy of broadcast time window, content priority and terminal scheduling order to generate broadcast scheduling decision instructions; The abnormal interference suppression module is used to identify potential interference sources during broadcast signal transmission using a residual convolutional neural network algorithm and reconstruct the disturbed broadcast segment through a signal repair mechanism driven by a recurrent neural network; the potential interference sources include channel conflicts, signal attenuation, and malicious interference; The terminal collaborative feedback module is used to receive the broadcast signal controlled by the broadcast scheduling instruction, perform playback operations on the terminal, and simultaneously collect terminal playback status data, user interaction feedback data and environmental response data, and dynamically adjust terminal reception parameters and content push strategies through a federated reinforcement learning algorithm.

2. The multifunctional integrated digital network broadcasting system based on intelligent algorithm according to claim 1, characterized in that: The multi-source information includes environmental sound signals, real-time voice signals collected by microphones, video surveillance image data, regional meteorological parameters, electromagnetic interference signal detection data, terminal device geographic location and motion status data, user interaction behavior data and network link status parameters.

3. The multifunctional integrated digital network broadcasting system based on intelligent algorithm according to claim 1 is characterized in that: The operation process of the broadcast content generation module includes the following steps: Performing standardized preprocessing on the multi-source information and constructing a multi-order feature tensor including semantic dimension, temporal dimension and modal dimension; Based on the multi-order feature tensor, a sparse tensor semantic clustering algorithm is used to compress and represent semantic units and perform topic clustering. The semantic core units in the multi-source content are identified through low-rank representation and semantic sparsity constraints to form a set of candidate broadcast semantic segments. Based on the candidate broadcast semantic segment set, combined with historical broadcast playback data and terminal response records, a label transfer modeling method is used to construct a content-context label mapping relationship, and a multi-scenario adaptation model based on label weight fitting is used to generate a content-context matching matrix; Based on the candidate broadcast semantic segment set and the content-context matching matrix, a broadcast structure generation mechanism based on constraint satisfaction problem solving is adopted to sequentially reconstruct and format optimize the candidate content to generate structured broadcast content including time series logic, context adaptability and multimodal fusion features.

4. The multifunctional integrated digital network broadcasting system based on intelligent algorithm according to claim 3 is characterized in that: The formula of the sparse tensor semantic clustering algorithm is as follows: in, represents the optimal semantic clustering result set; S represents the set of all clustering partitioning schemes; Represents multi-order feature tensor elements; represents the content-context label weight matrix; i represents the semantic unit number; j represents the time window number; k represents the modality number; Represents the data index set within the c-th cluster; Indicates the number of multi-source data in the c-th cluster; represents the indicator vector of the c-th cluster; represents the number of feature items of the cth topic cluster; C represents the final number of topic categories; Represents the sparsity adjustment parameter.

5. The multifunctional integrated digital network broadcasting system based on intelligent algorithm according to claim 1 is characterized in that: The operation process of the broadcast scheduling decision module includes the following steps: Based on the structured broadcast content, combined with the spatial distribution of terminals, congestion status within the region, and historical broadcast response data, a broadcast task flow graph with time window constraints is constructed, where nodes represent specific broadcast content, edges represent temporal or logical dependencies between content, and the time window is used to limit the executable time range of each task node; Collect the link accessibility, reception capability, interactive behavior characteristics and content preference information of each terminal, construct the terminal state tensor, and fuse it with the broadcast task graph to generate a state-enhanced scheduling graph; Based on the state-enhanced scheduling graph, a graph attention scheduling algorithm based on an asynchronous advantage update mechanism is adopted to dynamically calculate the scheduling value function of each broadcast task node. Based on the structural correlation between nodes, content priority, and terminal feedback potential, attention weights are assigned. In addition, an asynchronous strategy is used to jointly optimize the allocation of broadcast time windows, content delivery priority sorting, and target terminal scheduling order. Based on the optimization results, a broadcast scheduling decision instruction including content identification, target terminal ID list, scheduled time window and frequency resource allocation information is generated, and the instruction is packaged and issued through the heterogeneous link control module.

6. The multifunctional integrated digital network broadcasting system based on intelligent algorithm according to claim 5, characterized in that: The asynchronous advantage update mechanism specifically includes the following steps: Based on the state-enhanced scheduling graph, for each broadcast task node, based on the node's historical scheduling status, current environment characteristics and terminal feedback data, asynchronously sample the node state transition sequence and dynamically update the scheduling value function of the task node; By taking weighted differences in the scheduling value functions at different time steps, the advantage valuation function is calculated. Combined with the contextual attention weights of each node in the graph structure, local sensitivity adjustment of broadcast content scheduling priority and target terminal selection is achieved. In scenarios with multi-tasking concurrency and resource competition, distributed parallel computing is used to independently optimize the scheduling parameters of each node and update the global scheduling policy parameters through asynchronous feedback. Based on the global scheduling policy parameter update completed by the asynchronous advantage update mechanism, the latest scheduling priority and resource allocation recommendation of the node are integrated into the final broadcast scheduling decision output, including content distribution order, target terminal grouping and time slot resource allocation information, and scheduling instructions are issued to the heterogeneous link control module.

7. The multifunctional integrated digital network broadcasting system based on intelligent algorithm according to claim 5, characterized in that: The formula of the graph attention scheduling algorithm is as follows: in, represents the final scheduling priority score of broadcast task node i; represents the content importance score corresponding to the i-th task node in the structured broadcast content; represents the user activity index in the target area of ​​task i; represents the historical feedback score of content i; represents the current link bandwidth occupancy rate of the geographical area where task i is located; Indicates the maximum link bandwidth capacity; represents the congestion risk coefficient of task i in the state-enhanced scheduling graph; represents the remaining schedulable time window of task i; 、 and represents the score weighting coefficient; represents the exponential amplification factor of the response feedback score; and They represent the slope and center position parameters of the Sigmoid function in the time window urgency score, respectively.

8. The multifunctional integrated digital network broadcasting system based on intelligent algorithm according to claim 1 is characterized in that: The operation process of the abnormal interference suppression module includes the following steps: Real-time acquisition of multi-channel signal streams in the broadcast signal transmission link, and the use of multi-scale decomposition methods to extract multi-dimensional characteristic parameters such as signal energy spectrum distribution, instantaneous frequency characteristics, channel utilization and interference index; Based on the residual convolutional neural network algorithm, the above characteristic parameters are subjected to time series modeling and spatial feature fusion. Combined with channel historical data and statistical thresholds, it can automatically detect abnormal signal segments and distinguish and classify interference types. For the detected abnormal sections, an adaptive signal repair mechanism driven by recurrent neural networks is used to reconstruct the disturbed signal content based on the timing pattern and amplitude characteristics of historical reference segments; By comparing the signals before and after reconstruction, combined with the current link dynamics and the target terminal status, an interference suppression control vector is generated, including policy parameters for forwarding delay adjustment, link remapping, frequency band avoidance, and content reconstruction level. The anomaly suppression processing results are synchronously fed back to the terminal collaborative feedback module.

9. The multifunctional integrated digital network broadcasting system based on intelligent algorithm according to claim 1, characterized in that: The terminal collaborative feedback module builds a strategy adaptation mechanism based on the federated reinforcement learning architecture, uses the playback status data, interaction feedback data and environmental perception parameters collected locally by the terminal to periodically optimize the receiving strategy, and realizes cross-terminal strategy updates through global strategy aggregation.

Images