Event- and context-aware cross-modal communication resource scheduling methods, systems, electronic devices, and storage media.

By collecting multi-source heterogeneous context data and constructing a standardized unified context view, combined with a context-event joint representation model and a lightweight scheduling model, the problems of rigid routing strategies, lack of context awareness, and weak cross-modal collaboration in existing communication scheduling mechanisms are solved, and adaptive scheduling strategy generation and efficient communication are realized.

CN123074995APending Publication Date: 2026-11-10XINJIANG NORTH-WEST STAR INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611136016.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-29
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing communication scheduling mechanisms cannot adapt to dynamic and heterogeneous converged environments, resulting in rigid routing strategies, lack of context awareness, weak cross-modal collaboration, and delayed emergency response, making it impossible to achieve intelligent handover decisions and efficient communication.

Method used

By collecting multi-source heterogeneous contextual data, a standardized and unified contextual view is constructed using a three-level progressive multi-source data fusion method. Combined with a context-event joint representation model and a lightweight scheduling model, an adaptive scheduling strategy is generated to achieve context awareness and cross-modal collaboration.

Benefits of technology

It achieves adaptive scheduling strategies in complex scenarios, ensuring the compliance, security, and reliability of command and dispatch, and providing precise communication decision support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123074995A_ABST
    Figure CN123074995A_ABST
Patent Text Reader

Abstract

This invention relates to the field of communication resource scheduling technology, specifically a cross-modal communication resource scheduling method, system, electronic device, and storage medium based on event and context awareness. The method includes collecting multi-source heterogeneous contextual data related to events; introducing a three-level progressive multi-source data fusion method to obtain a standardized unified context view; obtaining a context-event joint representation vector based on a context-event joint representation model; obtaining mandatory scheduling constraints based on a rule engine; obtaining adaptive scheduling decisions based on a lightweight scheduling model; and fusing these to generate a final standardized scheduling strategy. This invention collects four-dimensional contextual awareness data and uses a three-level progressive multi-source data fusion method to construct a standardized unified context view, providing accurate data for decision-making in converged communication scheduling. The introduction of a dual decision engine enables the final standardized scheduling strategy to be obtained within the framework of mandatory scheduling constraints, ensuring the compliance, security, and reliability of command and dispatch.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of communication resource scheduling technology, specifically a cross-modal communication resource scheduling method, system, electronic device, and storage medium based on event and context awareness. Background Technology

[0002] In modern converged communications applications, users' communication needs have evolved from simple call connections to complex demands involving multiple terminals, multiple modalities, and multiple scenarios. A user often simultaneously possesses multiple terminals such as desktop phones, smartphones, dedicated walkie-talkies, and in-vehicle devices, and their communication activities encompass rich modalities including instant messaging, voice calls, video conferencing, and emergency broadcasting. However, existing communication scheduling mechanisms mostly remain in a static, pre-configured stage, making it difficult to adapt to this dynamic and heterogeneous converged environment. Their shortcomings are mainly reflected in the following four dimensions: First, the routing strategy is rigid. Call routing is usually based on simple number analysis or static routing tables, which cannot dynamically select the optimal path according to real-time network conditions (such as bandwidth congestion and latency jitter), which may lead to a decline in communication quality.

[0003] Secondly, the system lacks context awareness, failing to effectively perceive and understand the user's current environment. For example, it cannot determine whether the user is participating in a high-priority video conference, performing inspection operations on the production floor, or being in the core area of ​​an emergency. Without this information, the system cannot make intelligent follow-up decisions, potentially disrupting the user's current work or failing to prioritize tasks in emergencies.

[0004] Third, cross-modal collaboration is weak, lacking the ability to intelligently switch between different business formats. When a user initiates a voice call, the called party may be better suited to receiving video, instant messages, or text translation due to scenario limitations (such as being in a noisy environment, driving, or being unable to answer a voice call). The existing system lacks a scenario-based automatic modality judgment and seamless switching mechanism, leading to a mismatch between communication intent and receiving method, thus reducing communication efficiency.

[0005] Fourth, the emergency response is lagging behind. When faced with sudden emergency events, the system still relies heavily on the intervention of human administrators, which is inefficient and can easily lead to missed opportunities. Summary of the Invention

[0006] This invention provides an event- and context-aware cross-modal communication resource scheduling method, system, electronic device, and storage medium, which overcomes the shortcomings of the prior art. It can effectively solve the problems that existing cross-modal communication resource scheduling methods are mostly in the static pre-configuration stage, lack context awareness, and are highly dependent on manual intervention.

[0007] One of the technical solutions of this invention is achieved through the following measures: a cross-modal communication resource scheduling method based on event and context awareness, comprising: In response to event triggering in converged communication scenarios, multi-source heterogeneous contextual data related to the event is collected. A three-level progressive multi-source data fusion method is introduced to fuse the multi-source heterogeneous contextual data to obtain a standardized and unified contextual view. Input the feature vector of the standardized unified context view and the standardized event feature vector into the context-event joint representation model to obtain the context-event joint representation vector, where the context-event joint representation model is constructed through deep learning; The standardized unified context view and standardized events are input into the rule engine to obtain mandatory scheduling constraints. The context-event joint representation vector is input into the lightweight scheduling model to obtain adaptive scheduling decisions. The mandatory scheduling constraints and adaptive scheduling decisions are then fused to generate the final standardized scheduling strategy. The lightweight scheduling model is constructed through a joint training framework of knowledge distillation and attention distillation.

[0008] The following are further optimizations and / or improvements to the above-mentioned technical solution: The above response to event triggering in converged communication scenarios involves collecting multi-source heterogeneous contextual data related to the event, and introducing a three-level progressive multi-source data fusion method to fuse the multi-source heterogeneous contextual data to obtain a standardized and unified contextual view, including: In response to events triggered in converged communication scenarios, multi-source heterogeneous contextual data related to the events are collected, including network contextual data, terminal contextual data, user contextual data, and business contextual data. Differentiated cleaning is performed on multi-source heterogeneous context data, and four-level association is carried out in terms of time dimension, spatial dimension, hierarchical dimension and subject dimension; A three-level progressive multi-source data fusion method is introduced to fuse multi-source heterogeneous context data. The three-level progressive multi-source data fusion method includes data layer fusion, feature layer fusion, and decision layer fusion.

[0009] The above-mentioned input standardized unified context view feature vector and standardized event feature vector are fed into the context-event joint representation model to obtain the context-event joint representation vector, including: The events in the converged communication scenario are identified, and their features are extracted and standardized sequentially to obtain standardized event feature vectors. Input the feature vectors of the standardized unified context view and the standardized event feature vectors into the context-event joint representation model to obtain the context-event joint representation vector, where the context-event joint representation model is constructed through deep learning.

[0010] The above-mentioned input standardized unified context view and standardized events are fed into the rule engine to obtain mandatory scheduling constraints. The input context-event joint representation vector is fed into the lightweight scheduling model to obtain adaptive scheduling decisions. Finally, the mandatory scheduling constraints and adaptive scheduling decisions are fused to generate the final standardized scheduling strategy, including: Input standardized unified context view and standardized events into the rule engine to obtain mandatory scheduling constraints; Inputting the context-event joint representation vector into the lightweight scheduling model yields adaptive scheduling decisions, which include optimal terminal selection, optimal communication mode decision, and optimal media path planning. Within the framework of mandatory scheduling constraints, mandatory scheduling constraints and adaptive scheduling decisions are integrated to generate the final standardized scheduling strategy, which is then distributed after passing security permission and compliance verification.

[0011] The construction process of the above lightweight scheduling model includes: Several samples were obtained and divided into training set, validation set and test set. Each sample includes a context-event joint representation vector obtained after the event is triggered in the converged communication scenario and a corresponding adaptive scheduling decision label. The adaptive scheduling decision includes optimal terminal selection, optimal communication mode decision and optimal media path planning. The training set is input into the initial student model and the pre-trained teacher model. The student model is then trained using knowledge distillation with the teacher model to obtain the trained student model. The teacher model is learned using a number of samples. The total loss function when using the teacher model to train the student model using knowledge distillation is... ,in For cross-entropy loss, For knowledge distillation loss, For distillation loss, For priority-weighted loss, These are the weighting coefficients; The model parameters of the trained student model are optimized using the validation set and the test set, and the student model that meets the test evaluation requirements is output as a lightweight scheduling model.

[0012] The above also includes converting standardized scheduling strategies into standardized executable instruction objects, issuing and executing them according to priority, monitoring key indicators during the execution process after execution, and optimizing scheduling decisions and model parameters in reverse.

[0013] The second technical solution of the present invention is achieved through the following measures: a cross-modal communication resource scheduling system based on event and context awareness, comprising: The input perception layer includes: The multi-source heterogeneous context data acquisition module, in response to event triggering in the converged communication scenario, acquires multi-source heterogeneous context data related to the event; The multi-source heterogeneous context data fusion module introduces a three-level progressive multi-source data fusion method to fuse multi-source heterogeneous context data and obtain a standardized and unified context view; The context-event joint representation module takes the feature vector of the standardized unified context view and the standardized event feature vector as input to the context-event joint representation model to obtain the context-event joint representation vector. The context-event joint representation model is constructed through deep learning. The intelligent decision-making layer inputs a standardized unified context view and standardized events into the rule engine to obtain mandatory scheduling constraints. It inputs a context-event joint representation vector into the lightweight scheduling model to obtain adaptive scheduling decisions. Finally, it integrates the mandatory scheduling constraints and the adaptive scheduling decisions to generate the final standardized scheduling strategy. The lightweight scheduling model is constructed through a joint training framework of knowledge distillation and attention distillation.

[0014] The following are further optimizations and / or improvements to the above-mentioned technical solution: The aforementioned intelligent decision-making layer includes: The rules engine takes standardized unified context views and standardized events as input and obtains mandatory scheduling constraints. The lightweight scheduling decision engine takes the context-event joint representation vector as input to the lightweight scheduling model and obtains adaptive scheduling decisions, which include optimal terminal selection, optimal communication mode decision and optimal media path planning. The scheduling decision fusion and adjustment module integrates mandatory scheduling constraints and adaptive scheduling decisions within the framework of mandatory scheduling constraints to generate the final standardized scheduling strategy, which is then distributed after passing security permission and compliance verification.

[0015] The above also includes: a strategy execution layer, which converts standardized scheduling strategies into standardized executable instruction objects and issues them for execution according to priority; and a closed-loop optimization layer, which monitors the core indicators during the execution process after execution and optimizes the scheduling decisions and model parameters in reverse.

[0016] The third technical solution of the present invention is achieved through the following measures: an electronic device, characterized in that it includes a processor and a memory, wherein the memory stores a computer program, which is loaded and executed by the processor to implement the steps in the event- and context-aware cross-modal communication resource scheduling method.

[0017] The fourth technical solution of the present invention is achieved through the following measures: a storage medium, characterized in that the storage medium stores a computer program that can be read by a computer, the computer program being configured to execute the steps in the event- and context-aware cross-modal communication resource scheduling method at runtime.

[0018] This invention establishes a dynamic, automated cross-modal communication resource scheduling link integrating context acquisition, event processing, intelligent decision-making, strategy execution, and feedback optimization, significantly reducing manual decision-making intervention. Specifically, for key fields such as emergency management, judicial administration, and national defense, it collects and integrates four-dimensional contextual awareness of communication scenarios and uses a three-level progressive multi-source data fusion method to construct a standardized and unified context view. This addresses the problems of spatiotemporal misalignment and semantic gaps in multi-domain data in existing technologies, providing accurate data basis for integrated communication scheduling decisions. Deep learning is introduced to construct a context-event joint representation model, fully extracting context-event joint representation vectors to provide an accurate data foundation for the dual decision engine. The dual decision engine, incorporating a rule engine and a lightweight scheduling model, enables the final standardized scheduling strategy to be obtained within a framework of mandatory scheduling constraints, ensuring the compliance, security, and reliability of command and dispatch, and achieving adaptive scheduling strategies in complex scenarios. Attached Figure Description

[0019] Appendix Figure 1 This is a schematic flowchart of a cross-modal communication resource scheduling method provided in an embodiment of the present invention.

[0020] Appendix Figure 2 This is a schematic diagram of the standardized and unified context view acquisition method provided in the embodiments of the present invention.

[0021] Appendix Figure 3 This is a schematic diagram of the process for obtaining the context-event joint representation vector provided in an embodiment of the present invention.

[0022] Appendix Figure 4 This is a schematic diagram of the standardized scheduling strategy generation method provided in an embodiment of the present invention.

[0023] Appendix Figure 5 This is a schematic diagram of another cross-modal communication resource scheduling method provided in an embodiment of the present invention.

[0024] Appendix Figure 6 This is a schematic diagram of the cross-modal communication resource scheduling system provided in an embodiment of the present invention. Detailed Implementation

[0025] The present invention is not limited to the following embodiments, and specific implementation methods can be determined according to the technical solutions and actual conditions of the present invention.

[0026] Those skilled in the art will understand that, unless specifically stated otherwise, in the embodiments of the present invention, a "module" or "unit" refers to a computer program or part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0027] In addition, in the embodiments of the present invention, "multiple" refers to two or more, and "first" and "second" are used to distinguish descriptions and should not be construed as implying relative importance.

[0028] This invention provides an event- and context-aware cross-modal communication resource scheduling method, system, electronic device, and storage medium. This event- and context-aware cross-modal communication resource scheduling system can be integrated into a computer device, which can be a server, a terminal, or other similar device; it can also be executed jointly by a terminal and a server. The above examples should not be construed as limiting the invention.

[0029] The aforementioned terminals may include mobile phones, wearable smart devices, tablet computers, laptops, personal computers (PCs), and in-vehicle computers, etc., and this invention does not limit them. This invention also does not limit the number of terminal devices.

[0030] The aforementioned server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. This invention does not limit these features.

[0031] For example, in response to events triggered in a converged communication scenario, a computer device collects multi-source heterogeneous contextual data related to the event. A three-level progressive multi-source data fusion method is introduced to fuse the multi-source heterogeneous contextual data to obtain a standardized unified context view. The feature vectors of the standardized unified context view and the standardized event feature vectors are input into a context-event joint representation model to obtain a context-event joint representation vector, where the context-event joint representation model is constructed through deep learning. The standardized unified context view and standardized events are input into a rule engine to obtain mandatory scheduling constraints. The context-event joint representation vectors are input into a lightweight scheduling model to obtain adaptive scheduling decisions. The mandatory scheduling constraints and adaptive scheduling decisions are then fused to generate the final standardized scheduling strategy, where the lightweight scheduling model is constructed through a joint training framework of knowledge distillation and attention distillation.

[0032] Based on this, the technical solution of the present invention will be described and explained below with reference to several examples.

[0033] Example 1: As shown in the attached document Figure 1 As shown, this embodiment of the invention discloses a cross-modal communication resource scheduling method based on event and context awareness, including: Step S110: In response to the triggering of an event in the converged communication scenario, collect multi-source heterogeneous contextual data related to the event, and introduce a three-level progressive multi-source data fusion method to fuse the multi-source heterogeneous contextual data to obtain a standardized unified contextual view. Step S120: Input the feature vector of the standardized unified context view and the standardized event feature vector into the context-event joint representation model to obtain the context-event joint representation vector, wherein the context-event joint representation model is constructed through deep learning; In step S130, the standardized unified context view and standardized events are input into the rule engine to obtain mandatory scheduling constraints. The context-event joint representation vector is input into the lightweight scheduling model to obtain adaptive scheduling decisions. The mandatory scheduling constraints and adaptive scheduling decisions are then fused to generate the final standardized scheduling strategy. The lightweight scheduling model is constructed through a joint training framework of knowledge distillation and attention distillation.

[0034] This invention discloses a cross-modal communication resource scheduling method based on event and context awareness. It integrates deep context awareness after an event is triggered in a communication scenario, collects multi-source heterogeneous context data related to the event, and uses a three-level progressive multi-source data fusion method to fuse four-dimensional context-aware data, constructing a standardized and unified context view. This solves the problems of spatiotemporal misalignment and semantic gaps in multi-domain data in existing technologies, providing accurate data basis for fusion communication scheduling. Deep learning is introduced to construct a context-event joint representation model and a lightweight scheduling model. A dual decision engine, consisting of a rule engine and a lightweight scheduling model, is set up to accurately acquire standardized scheduling strategies, ensuring the timeliness and effectiveness of event handling.

[0035] Example 2: As shown in the attached document Figure 2 As shown, this embodiment of the invention is a further optimization of the above embodiment. In response to an event triggering in a converged communication scenario, multi-source heterogeneous contextual data related to the event is collected. A three-level progressive multi-source data fusion method is introduced to fuse the multi-source heterogeneous contextual data, resulting in a standardized unified contextual view, including: Step S210: In response to the triggering of an event in the converged communication scenario, collect multi-source heterogeneous context data related to the event, wherein the multi-source heterogeneous context data includes network context data, terminal context data, user context data and business context data; In this embodiment, the events in the converged communication scenario include emergency trigger events, command and dispatch events, status change events, and instruction execution events. The corresponding trigger source and trigger priority can be set according to the event type, such as an early warning signal transmitted to the converged communication platform through the park intranet.

[0036] In this embodiment, the multi-source heterogeneous context data is set up with four core datasets. Before collection, the collection objects, methods, update frequency and format need to be clearly defined to adapt to the characteristics of multi-network convergence and multi-level command scenarios. Here, multi-source heterogeneous context data related to events is collected, as follows: Network contextual data: The data collection targets are multi-domain devices including government intranets, emergency private networks, 4G / 5G public networks, satellite communication networks, and dedicated communication networks, including SDN controllers, core switches, domestically produced routing equipment, QoS gateways, and communication base stations. The collection method is set to a dual-mode approach of active probing and passive mirroring. Active probing sends probe packets at a period of 50ms, while passive collection synchronizes data through the standard interface of domestically produced network equipment. The collection content includes end-to-end available bandwidth, transmission latency, packet loss rate, jitter, network topology, node load, link encryption status, and network domain isolation boundaries. The update frequency can be set to 100ms. This forms a corresponding time-series structured dataset, i.e., network contextual data. ; Terminal contextual data: The data collection targets are set as law enforcement recorders, digital trunking radios, individual soldier combat terminals, vehicle-mounted command terminals, desktop command PCs, classified special terminals, and portable command tablets; the data collection method is set as terminal agent periodic reporting + domestic access gateway status synchronization; the data collection content includes terminal type, hardware encoding / decoding capability, encryption level, remaining battery power, BeiDou / GPS positioning, signal strength, online status, user ID, and access network domain; the update frequency can be set to 1 second; a terminal attribute dataset, i.e., terminal contextual data, is formed. ; User contextual data: The data collection targets are set as command personnel at all levels, law enforcement personnel, emergency response personnel, combat personnel, and support personnel; the data collection method is set as user management system interface synchronization + duty system integration + manual status reporting; the data collection content is set as user role / permission level, department / command level, online / on duty / duty / meeting status, duty arrangement, geographical location, communication preferences, classified permissions, and do-not-disturb mode; the update frequency can be set as converged communication scenario event trigger + 5-second cycle synchronization; forming a user profile dataset, i.e., user contextual data. ; Business Context Data: The data collection targets are set as emergency command platforms, judicial service systems, video surveillance systems, alarm linkage systems, and GIS geographic information platforms; the collection method is set as event-triggered collection in converged communication scenarios + synchronization with business process interfaces; the collection content is set as business type (emergency response / disaster relief / judicial escort / security / command and dispatch), event priority, time sensitivity, associated emergency plans, service specifications, event ID, target disposal object, and GIS geographic coordinates; the update frequency can be set to real-time update triggered by converged communication scenarios, forming a business attribute dataset, i.e., business context data. .

[0037] Step S220: Perform differentiated cleaning on multi-source heterogeneous context data and perform four-level associations in terms of time dimension, spatial dimension, hierarchical dimension and subject dimension; In this embodiment, cleaning rules are set according to the confidentiality characteristics and multi-domain differences of various data scenarios. This not only solves the problems of outliers, missing values, permission conflicts, and logical contradictions in the original data, but also ensures the efficiency and accuracy of cleaning. The specific differentiated cleaning rules include: Network context data cleaning employs the 3σ criterion and sliding window filtering to remove network performance anomalies, linear interpolation and fitting of the same route links to complete missing data, and deduplication based on timestamps and link IDs. At the same time, it verifies the compliance of network domain isolation, removes illegal cross-domain data, and ensures the security and validity of multi-network data. Terminal context data cleaning involves establishing a terminal type-capability-encryption level mapping rule base, verifying the matching of terminal-reported data with model number, supplementing standardized timestamps, and resolving conflicts in status reports from multiple network domains for the same terminal according to the priority of "dedicated network > government intranet > emergency network > public network", while retaining high-confidence data. User context data cleaning: Establish a user status mutual exclusion rule base and permission-status matching rules to resolve status conflicts such as "in meeting" and "on duty", remove abnormal status data that do not meet the confidentiality permissions, and determine the final user status according to the priority of "manual reporting > duty system synchronization > automatic identification". Business context data cleaning: Establish a business type-priority-command level mapping standard, map non-standard business descriptions to priority levels 1-5 and standardized business types, remove invalid data without event ID or target object, and fill in missing fields according to emergency plan specifications to form business data in a unified format.

[0038] In this embodiment, precise binding of multi-source heterogeneous contextual data is achieved through a four-level association system encompassing time, space, hierarchy, and subject dimensions. The specific four-level association settings include: Time dimension correlation: Set a 200ms sliding correlation window, use the event trigger timestamp of the converged communication scenario as the benchmark, align all multi-dimensional data with timestamps in the window to ensure that the correlation data are in the same time segment, and solve the problem of latency difference in collection from multiple network domains; Spatial dimension association: Based on the GIS geographic information system, a hierarchical geofence is established to match terminal location, user location, and event location with the geofence, and associate spatial data within the same fence to achieve spatial binding of "event site - handling personnel - terminal location - network coverage". The hierarchical geofence includes national / provincial / municipal / county area fences, event site fences, service area fences, and key control area fences. By establishing a hierarchical relationship, a five-level command hierarchy mapping relationship is established, namely "national-provincial-municipal-county-on-site". According to the event handling authority, the corresponding command departments, handling personnel and communication resources are associated to realize cross-level command data. By establishing a subject-based mapping relationship database with "User ID - Terminal ID - Department ID - Event ID - Command Group ID", the database associates the corresponding terminal and department with the user ID as the core, and the handling process, emergency plan, and command group with the event ID as the core, thereby achieving full-link association from "event - command level - user - terminal - network - business".

[0039] Step S230: A three-level progressive multi-source data fusion method is introduced to fuse multi-source heterogeneous context data. The three-level progressive multi-source data fusion method includes: Data layer fusion involves assigning weights to time series data of the same dimension and type based on domain confidence, and then using a weighted average algorithm to perform data layer fusion to obtain effective values ​​for fused indicators. The above-mentioned weighted average algorithm for data layer fusion can eliminate data bias from multiple networks. When setting weights for time-series data of the same dimension and type according to the network domain confidence level, the weight of the dedicated network can be set to 0.7, the weight of the government intranet can be set to 0.2, and the weight of the public network can be set to 0.1.

[0040] Feature layer fusion utilizes a feature layer fusion model to fuse features from multiple heterogeneous contextual data. This model is trained on an initial network, which employs an improved Transformer feature encoder. The initial network performs one-hot encoding and embedding on each of the heterogeneous contextual data, and learns the semantic relationships between cross-dimensional features through a multi-head cross-attention mechanism, outputting a fused global contextual feature vector. ; The feature layer fusion model introduced in the above feature layer fusion process obtains a fused global contextual feature vector, which solves the problem of semantic gap in heterogeneous data.

[0041] Decision-level fusion, based on the improved DS evidence theory, performs decision-level fusion on multi-source heterogeneous situational data to obtain high-confidence situational decision conclusions; The above-mentioned decision-level fusion of multi-source heterogeneous contextual data based on the improved DS evidence theory specifically includes constructing a basic probability allocation function for fused communication command scenarios based on the improved DS evidence theory, calculating the support of network, terminal, user, and service status for the current scheduling decision, fusing multi-source evidence through DS combination rules, resolving data conflicts, and outputting high-confidence contextual decision conclusions to provide a basis for scheduling decisions.

[0042] Step S240: Construct a standardized unified context view based on the fused multi-source heterogeneous context data; In this embodiment, the constructed standardized unified context view is a five-dimensional seven-tuple standardized unified context view, which can provide a unique, standardized, and real-time updated decision basis for scheduling. The structure of the standardized unified context view is as follows: in, EventID , TimeStamp , GeoFence , CommandLevel All are core primary keys, namely, unique event ID, view update timestamp, event-associated geofence, and corresponding command level; These are four-dimensional data volumes, namely, the fused network context data, terminal context data, user context data, and business context data; Furthermore, the update mechanism in the aforementioned standardized unified context view is a real-time reconstruction triggered by events in the converged communication scenario, plus incremental updates every 500ms, ensuring real-time performance and consistency. The storage method uses a domestically produced classified database for storage, a memory database to cache the currently effective view, and a time-series database to store historical data, supporting full-link backtracking and auditing of scheduling.

[0043] This invention presents a three-level progressive multi-source data fusion method. After an event is triggered in a converged communication scenario, multi-source heterogeneous contextual data related to the event is collected. The three-level progressive multi-source data fusion method is used to fuse the multi-source heterogeneous contextual data to obtain a standardized and unified contextual view. This solves the problems of spatiotemporal misalignment and insufficient semantic gap in multi-domain data in the prior art, and provides a data basis for accurate decision-making in converged communication scheduling.

[0044] Example 3: As shown in the attached document Figure 3 As shown, this embodiment of the invention is a further optimization of the above embodiment, wherein the standardized unified context view feature vector and the standardized event feature vector are input into the context-event joint representation model to obtain the context-event joint representation vector, including: Step S310: Determine the events in the converged communication scenario, and perform feature extraction and standardization processing on them in sequence to obtain a standardized event feature vector; In this embodiment, the converged communication scenario events are determined by full monitoring and classification of these events. These events are divided into four categories, each with a trigger source and trigger priority, as detailed below: Emergency-triggered events: Triggering sources include emergency alarms, rescue requests, emergency reporting, equipment alarms, and public opinion warnings, with a priority level of 1-2. Command and dispatch events: Triggering sources are the issuance of command instructions, duty requests, call notifications, meeting convening, group call intercoms, and broadcast announcements, with a priority level of 2-3; Status change events: Triggering sources are user status changes, terminal online / offline status changes, network status changes, duty process updates, and command level adjustments, with a priority of 4-5. Command execution events: The triggering sources are scheduling command execution feedback, task progress updates, and emergency plan node execution reports, with a priority level of 3-4.

[0045] In this embodiment, the process of sequentially extracting features and standardizing events in a converged communication scenario includes: Event parsing involves extracting the core fields of the event, including the event type. timestamp Triggering source, initiator Target Event Level Related Event Description Attachment data (Video stream address, GIS coordinates, on-site data); Standardized formatting: Convert the core event fields to the standard event format, the structure of which is shown below: Event validation and sorting: Validate the completeness of required fields and compliance with permissions, and remove invalid events; sort by event priority + command level + timestamp, with high-priority and high-command-level events entering the dispatch queue first to ensure that emergency command instructions are processed first.

[0046] Step S320: Input the feature vector of the standardized unified context view and the standardized event feature vector into the context-event joint representation model to obtain the context-event joint representation vector, wherein the context-event joint representation model is constructed through deep learning; In this embodiment, the construction process of the context-event joint representation model is a general deep learning process, that is, it is obtained by training a pre-defined network structure using multiple training samples. The pre-defined network structure includes an input layer, a 6-layer Transformer encoder, and an output layer. The input layer standardizes the feature vectors of the unified context view. and standardized event feature vectors The layers are concatenated; a 6-layer Transformer encoder performs cross-modal feature interaction to extract deep correlation features. Specifically, the shallow layers (layers 1-2) capture low-level statistical correlations between modalities, the middle layers (layers 3-4) construct structured mappings, mapping the temporal logic of the context to the evolutionary logic of events, and the deep layers (layers 5-6) refine high-order semantic correlations. Residual connections and layer normalization are also used to ensure that the deep network does not experience gradient degradation, thus preserving the original modal information. The output layer outputs a context-event joint representation vector. .

[0047] Example 4: As shown in the appendix Figure 4 As shown, this embodiment of the invention is a further optimization of the above embodiment. It inputs a standardized unified context view and standardized events to the rule engine to obtain mandatory scheduling constraints, inputs a context-event joint representation vector to the lightweight scheduling model to obtain adaptive scheduling decisions, and fuses the mandatory scheduling constraints and adaptive scheduling decisions to generate the final standardized scheduling strategy, including: Step S410: Input the standardized unified context view and standardized events into the rule engine to obtain mandatory scheduling constraints. The rule engine uses a standardized policy language to implement IF-THEN rule encapsulation and has a built-in three-level rule library specifically for converged communication command scenarios, including: Level 1 mandatory rules are unbreakable mandatory rules, such as "For Level 1 emergency events, the terminals of the corresponding level of command leaders and on-site handling leaders must be forcibly connected, their current calls must be forcibly interrupted, and the highest priority encrypted communication channel must be established" and "Emergency command traffic must enable end-to-end QoS protection and reserve 80% dedicated bandwidth". Secondary scenario rules adapt to scheduling rules for different business scenarios, such as "for emergency rescue site alarm events, automatically link the on-site monitoring camera, push real-time video streams to the personnel in charge, and initiate two-way video intercom" and "for cross-departmental collaborative command events, automatically establish multi-party encrypted video conferences and simultaneously push event-related information to all participating terminals." Level 3 routine rules are the scheduling rules for daily duties and routine communications. For example, "When a user is in a meeting, non-emergency command instructions are automatically converted into encrypted instant message pushes without triggering voice ringing" and "When a user is in a field duty state, their handheld walkie-talkie / soldier terminal is dispatched first, followed by mobile phone terminals."

[0048] It should also be noted that the three-level rule base for converged communication command scenarios features visual configuration, version management, offline simulation, and canary release, and can be dynamically updated according to command needs.

[0049] Step S420: Input the context-event joint representation vector into the lightweight scheduling model to obtain the adaptive scheduling decision, wherein the adaptive scheduling decision includes: Optimal terminal selection predicts the optimal terminal for a user in the current context, ranking them by connection probability, response speed, encryption level, and terminal capabilities to solve the terminal selection problem in heterogeneous multi-terminal scenarios. The optimal communication mode decision-making system automatically selects the optimal communication mode (voice / video / encrypted messaging / truncate intercom / emergency broadcast / conference consultation) based on network status, terminal capabilities, user status, and event priority, enabling cross-modal adaptive switching. Optimal media path planning predicts the media transmission path with the lowest packet loss rate, lowest latency, and matching encryption level based on real-time network status and domain isolation requirements. It automatically decides whether transcoding is needed, whether to use a conference bridge, and whether to enable a dedicated encrypted channel to ensure the stability and security of command and control communications.

[0050] In this embodiment, the lightweight scheduling model is constructed through a joint training framework of knowledge distillation and attention distillation, specifically including: (1) Obtain several samples and divide the samples into training set, validation set and test set. Each sample includes the context-event joint representation vector obtained after the event is triggered in the converged communication scenario and the corresponding adaptive scheduling decision label. The adaptive scheduling decision includes optimal terminal selection, optimal communication mode decision and optimal media path planning. (2) Input the training set into the initial student model and the pre-trained teacher model, and use the teacher model to perform knowledge distillation training on the student model to obtain the trained student model, wherein the teacher model is learned using several samples; In this step, the network structure of the teacher model is mainly based on the Transformer architecture, and the teacher model is pre-trained with several samples, each of which includes a context-event joint representation vector obtained after the event is triggered in the fusion communication scenario and the corresponding adaptive scheduling decision label.

[0051] In this step, the student model is a lightweight model, built upon the Transformer architecture of the teacher model with a significantly reduced number of parameters. To adapt to domestically produced hardware, the compression strategy is as follows: Encoder lightweighting: The 12-layer Transformer encoder was replaced with a 4-layer lightweight MobileTransformer structure, reducing the hidden layer dimension from 768 to 384 and the number of parameters by 75%; Attention layer compression: The number of multi-head attention heads per layer is reduced from 12 to 6, redundant parameters of the feedforward network are removed, and only the core scheduling decision-related feature channels are retained; Branch pruning: Retain core command, scheduling, and decision-making branches, and eliminate unnecessary auxiliary branches to further reduce computational load.

[0052] After compression, the number of parameters in the student model can be reduced to 1 / 10 of that in the teacher model, and the memory usage for single-model inference is ≤2G. Millisecond-level inference can be achieved on domestically produced servers, edge gateways, and embedded terminals.

[0053] Furthermore, in the knowledge distillation training of the student model using the teacher model, each iteration executes the process of forward propagation of the teacher model, forward propagation of the student model, total loss calculation, backpropagation and parameter update, with the teacher model parameters frozen throughout the process, thus completing the training of the student model.

[0054] Here, the teacher model and the student model jointly perform forward propagation, specifically by using the same batch of fused communication scheduling training data. Teacher model with input freeze parameters With trainable student models Record the output layer logits and the attention weight matrix of each layer.

[0055] The total loss function here is The weighting coefficients satisfy The weighting coefficient can be set to, but is not limited to, 1. , , , ; Among them, the cross-entropy loss based on the real labels This is used to ensure the basic accuracy of the student model in scheduling tasks, and the formula is as follows: ; Among them, the knowledge distillation loss of the output layer soft labels Introducing a temperature coefficient The output distribution is softened, and the fine-grained decision-making "hidden knowledge" of the teacher model is learned, as shown in the following formula: ; Among them, the distillation loss of multi-head attention in the intermediate layer Multi-head soft-aligned attention distillation is employed to establish an equidistant mapping relationship between the teacher and student layers, transferring the attention distribution of the teacher model to the student model, as shown in the following formula: ; Among them, priority-weighted loss in emergency scenarios Higher weights are assigned to scheduling losses for Level 1 and 2 emergency events (e.g., Level 1 events have a weight of 3, and Level 2 events have a weight of 2) to ensure the accuracy of emergency scenario scheduling. The formula is as follows: .

[0056] (3) Optimize the model parameters of the trained student model using the validation set and test set, and output the student model that meets the test evaluation requirements as a lightweight scheduling model.

[0057] Furthermore, the distilled lightweight scheduling model can be quantized and compressed to adapt to domestic hardware environments, specifically including: Post-training static quantization: The weights and activation values ​​of the Transformer layer and fully connected layer in the model are quantized from FP32 to INT8; Symmetric linear quantization implementation: Employs a symmetric linear quantization scheme with a scaling factor. To reduce accuracy loss, channel-by-channel quantization is used for the weights, and 500-1000 calibration samples covering various command scenarios are used to determine the activation scaling factor. Quantization-aware fine-tuning: Gradient backpropagation is achieved through a pass-through estimator, in order to Fine-tune with a low learning rate for 3-5 epochs to ensure quantization accuracy loss ≤2%; Domestic Deployment Adaptation: After fine-tuning, the model is converted into a format compatible with domestic chips and servers, supporting deployment and operation on domestic CPUs such as Phytium and Kunpeng, domestic edge chips such as Rockchip and HiSilicon, as well as domestic converged communication gateway servers and integrated access gateways.

[0058] Step S430: Within the framework of mandatory scheduling constraints, mandatory scheduling constraints and adaptive scheduling decisions are integrated to generate the final standardized scheduling strategy, which is then distributed after passing security permission and compliance verification.

[0059] The embodiments of this invention introduce a dual decision engine consisting of a rule engine and a lightweight scheduling model, which enables the final standardized scheduling strategy to be obtained within the framework of mandatory scheduling constraints, ensuring the compliance, security and reliability of command and dispatch, and realizing the adaptability of scheduling strategies in complex scenarios.

[0060] Example 5: As shown in the attached document Figure 5 As shown, the embodiments of the present invention are further optimizations of the above embodiments, and also include converting the standardized scheduling strategy into standardized executable instruction objects, issuing and executing them according to priority, monitoring the core indicators during the execution process after execution, and optimizing the scheduling decision and model parameters in reverse.

[0061] In this embodiment, the standardized scheduling strategy is converted into a standardized executable instruction object, such that the standardized executable instruction object satisfies the following instruction structure; in, This includes the instruction's unique ID, execution target, instruction type, core parameters, execution priority, encryption level, and timeout. During the above parsing process, instruction compatibility verification is performed simultaneously, and incompatible instructions are automatically converted into compatible alternative instructions.

[0062] In this embodiment, standardized executable instructions are distributed to the corresponding underlying communication subsystems according to priority, adapting to domestically produced converged communication gateways and integrated access gateways, and using IP softswitches as the core to complete the corresponding operations: Call control services: perform operations such as call setup, call insertion, call disconnection, group call, and encrypted conference setup; Media server: Performs operations such as media transcoding, audio mixing, encrypted recording, and conference bridging; Domestically produced network equipment: performs operations such as route adjustment, QoS configuration, bandwidth reservation, and encrypted channel establishment; Emergency broadcasting system: performs operations such as zone / global broadcasting and text-to-speech announcements; Message push service: Real-time message push for executing encrypted command instructions, event information, and handling requirements.

[0063] The system transmits instruction status back in real time during execution. If the instruction is not executed within a timeout period, it will automatically retry. If the retry fails, the system will trigger the backup scheduling strategy.

[0064] In this embodiment, core indicators during the execution process are monitored after execution, and scheduling decisions and model parameters are optimized in reverse. The core indicators include monitoring call connection rate, average response latency, communication quality MOS value, emergency event handling time, network resource utilization, instruction execution success rate, and encryption compliance. Furthermore, the core indicators can be stored in the feedback dataset.

[0065] The embodiments of the present invention form a complete closed-loop technical link of "context acquisition, event processing, intelligent decision-making, strategy execution, and feedback optimization", which ensures the accuracy of cross-modal communication resource scheduling.

[0066] Example 6: As shown in the appendix Figure 6 As shown, this embodiment of the invention discloses an event- and context-aware cross-modal communication resource scheduling system, comprising: The input perception layer includes: The multi-source heterogeneous context data acquisition module, in response to event triggering in the converged communication scenario, acquires multi-source heterogeneous context data related to the event; The multi-source heterogeneous context data fusion module introduces a three-level progressive multi-source data fusion method to fuse multi-source heterogeneous context data and obtain a standardized and unified context view; The context-event joint representation module takes the feature vector of the standardized unified context view and the standardized event feature vector as input to the context-event joint representation model to obtain the context-event joint representation vector. The context-event joint representation model is constructed through deep learning.

[0067] The intelligent decision-making layer includes: The multi-source heterogeneous context data acquisition module, in response to event triggering in the converged communication scenario, acquires multi-source heterogeneous context data related to the event; The multi-source heterogeneous context data fusion module introduces a three-level progressive multi-source data fusion method to fuse multi-source heterogeneous context data and obtain a standardized and unified context view; The context-event joint representation module takes the feature vector of the standardized unified context view and the standardized event feature vector as input to the context-event joint representation model to obtain the context-event joint representation vector. The context-event joint representation model is constructed through deep learning. The intelligent decision-making layer inputs a standardized unified context view and standardized events into the rule engine to obtain mandatory scheduling constraints. It inputs a context-event joint representation vector into the lightweight scheduling model to obtain adaptive scheduling decisions. Finally, it integrates the mandatory scheduling constraints and the adaptive scheduling decisions to generate the final standardized scheduling strategy. The lightweight scheduling model is constructed through a joint training framework of knowledge distillation and attention distillation.

[0068] The strategy execution layer converts standardized scheduling strategies into standardized executable instruction objects and issues them for execution according to priority.

[0069] The closed-loop optimization layer monitors the core metrics during the execution process after execution and reverse-optimizes scheduling decisions and model parameters.

[0070] Example 7: Deploying the event- and context-aware cross-modal communication resource scheduling system in a completely domestically developed and controllable environment as described in this embodiment of the invention. Specifically: Hardware platform: Domestic edge server equipped with Phytium 2000 / 4 processor, domestic converged communication gateway, and multi-protocol integrated access gateway; Software platform: Tongxin UOS server operating system, Renmin University Kingbase domestic time-series database, domestic IP softswitch core, national cryptographic middleware, AI video analysis engine; Access terminals: digital intercoms (DMR protocol) for the park, security patrol terminals, high-definition surveillance cameras (GB / T28181 protocol), access control system terminals, and desktop PCs for the property command center; Network environment: Enterprise campus intranet, 4G public network, with national cryptographic SM2 / SM3 / SM4 encryption devices deployed throughout the network.

[0071] The system architecture fully corresponds to the four core layers of the present invention: perception input layer, intelligent decision-making layer, strategy execution layer, and closed-loop optimization layer. It has been integrated with the park property management platform, video surveillance system, GIS geographic information platform, access control management system, and security duty system.

[0072] The event- and context-aware cross-modal communication resource scheduling method corresponding to the embodiments of the present invention includes: (1) Event triggering and initial reporting in converged communication scenarios On [Date] at 19:25:42, in a remote green belt on the northwest side of a certain enterprise park (a non-personnel passage area), the AI ​​video analysis system detected that a person lingered for more than 30 minutes, triggering an abnormal lingering warning.

[0073] The warning signal is transmitted to the converged communication platform via the park's intranet, triggering a Level 3 general emergency event. The AI ​​system simultaneously pushes event information: event type "abnormal personnel stay warning", priority level 3, location "northwest green belt A07 area", personnel characteristics "male, wearing dark top", stay duration "32 minutes".

[0074] (2) The perception input layer performs multi-source heterogeneous context data acquisition and fusion. The system automatically initiates a full-dimensional context-aware process, completing data collection, cleaning, correlation, and fusion within 200ms: (a) Targeted data acquisition in multi-source heterogeneous contexts; Network scenario: Actively probe the link bandwidth (45Mbps), latency (25ms), and packet loss rate (0.4%) in the northwest area of ​​the park's intranet to confirm that the 4G public network backup link is normal, with an update frequency of 100ms; Terminal scenario: Collect the status of the digital walkie-talkies (online, 68% battery, signal strength -70dBm) of the two security guards closest to the incident site, the inspection terminal (online, Beidou positioning accuracy 3 meters), and the four surrounding surveillance cameras (online, supporting H.264 encoding), with an update frequency of 1 second; User scenario: The system retrieves the status of two individuals, Li and Wang, from the security duty system as "on night patrol," their department as "Park Security Department," and their command level as "on-site response level." Simultaneously, it retrieves the status and location of the property management duty leader and the monitoring room duty officer, updating every 5 seconds. Business Scenario: Retrieve the "Park Abnormal Personnel Handling Plan" from the property management system, associate the command level with "Park Level", and the handling requirements are "arrive at the scene within 5 minutes to verify and transmit the entire process via video". The event is triggered and updated in real time.

[0075] (b) Differentiated cleaning and multi-dimensional correlation of multi-source heterogeneous context data; The 3σ criterion is used to remove network anomaly detection data, and the status conflicts reported by multiple networks from the same terminal are resolved according to the priority of "internal network of the park > public network". Based on the event trigger timestamp of 19:25:42, set a 200ms sliding window to align all data; Based on GIS geofencing, security terminals, surveillance cameras, access control devices, and events within a 200-meter radius of the A07 area in the northwest green belt are linked together. Establish a principal mapping relationship of "Event ID - Command Level - Response Personnel - Terminal ID - Network Link".

[0076] (c) Introduce a three-level progressive multi-source data fusion method to fuse multi-source heterogeneous contextual data to obtain a standardized and unified contextual view; Data layer fusion: Calculate the effective value of network performance with a weight of 0.7 for the intranet and 0.3 for the public network; Feature layer fusion: Generate global contextual feature vectors through an improved Transformer encoder; Decision-making level integration: Based on the DS evidence theory, the high-confidence conclusions of "normal on-site network, closest personnel, available terminals, and abnormal personnel stay" are confirmed; Finally, a standardized unified context view is generated, with the following core keys: EventID=20260X0X192542, TimeStamp=19:25:42.200, GeoFence=Northwest Green Belt A07 Area, CommandLevel=Park Level.

[0077] (3) Standardized event processing and joint representation Event parsing and encapsulation: Converting raw early warning events into standardized event structures, including core fields such as event type, priority, command level, associated EventID, GIS coordinates, and personnel characteristics; Event sorting: The event is inserted into the scheduling queue according to a 3-level priority, taking precedence over routine communication tasks; Context-Event Joint Representation: Input the feature vector of the standardized unified context view and the standardized event feature vector into the context-event joint representation model to obtain the context-event joint representation vector.

[0078] (4) Utilize the intelligent decision-making layer to generate the final standardized scheduling strategy, including: (a) Input standardized unified context view and standardized events into the rule engine to obtain mandatory scheduling constraints; Second-level scenario rule matching: triggering the rules of "automatic linkage of 4 surveillance cameras around the scene to push real-time video stream to the personnel in charge and initiate two-way voice communication" and "automatic establishment of multi-party temporary voice conference for cross-position collaboration"; Level 3 routine rule matching: triggering the rules "when a user is in patrol mode, prioritize dispatching their digital walkie-talkie, then dispatch the inspection terminal", "do not force call interruption in non-emergency events, send warning messages first", and "automatically turn on the terminal flashlight and law enforcement recorder during nighttime operations".

[0079] (b) Input the context-event joint representation vector into the lightweight scheduling model to obtain adaptive scheduling decisions, which include: Optimal terminal selection: The digital walkie-talkie of the nearest security guard, Li, is the optimal voice terminal, and the patrol terminal is the optimal video transmission terminal; Optimal communication mode: The decision adopts a hybrid mode of "voice intercom + video transmission + encrypted messages"; Optimal media path: Select the direct connection path within the park's intranet, and enable end-to-end SM4 encryption, without the need for transcoding.

[0080] (d) Scheduling decision integration and compliance verification By integrating mandatory scheduling constraints and adaptive scheduling decisions, a final standardized scheduling strategy is generated, specifically as follows: The system initiates a voice call to Li's digital walkie-talkie and simultaneously sends a warning message, on-site location, and personnel characteristics to his inspection terminal. Automatically call the property management duty leader and the monitoring room duty officer to establish a multi-party temporary voice conference; The system integrates four on-site surveillance cameras to push real-time video streams to the large screen in the property management command center and the terminals of the participants. An encrypted message outlining the handling procedures and precautions is sent to all participating terminals. Reserve 30% of the bandwidth for this session on the park's intranet and enable QoS protection; The system automatically triggered the recording of the law enforcement recorder and the flashlight activation command on Li's inspection terminal.

[0081] After passing security permission and compliance verification, it is issued to the policy execution layer.

[0082] (5) The standardized scheduling policy is converted into a standardized executable instruction object using the policy execution layer, and then executed according to priority, including: Policy parsing: The abstract scheduling policy is converted into 8 standardized executable instructions, including instruction ID, target, type, parameters, priority, and encryption level; Command issuance and execution: The call control service initiated a voice call to Li's digital walkie-talkie, with a call connection delay of 270ms. The media server pulls real-time video streams from four surveillance cameras and pushes them to the command center's large screen. The system automatically triggered the recording of Li's patrol terminal's law enforcement recorder and the activation of the flashlight; Automatically calls the property management and monitoring room duty terminals, and completes the establishment of a multi-party conference of 5 people within 16 seconds; Domestically produced network equipment has completed bandwidth reservation and QoS configuration; The encrypted message push service sends processing instructions and on-site information to all participating terminals.

[0083] (6) Utilize the closed-loop optimization layer to monitor core metrics during execution after execution, and reverse-optimize scheduling decisions and model parameters, including: Real-time monitoring metrics: 100% call connection rate, 270ms average response latency, MOS (Mean Offset Syndrome) for communication quality, 4.3, and 100% command execution success rate. Data storage: The entire process scheduling data, execution indicators, and handling results are associated with EventIDs and stored in a domestically produced time-series database; Strategy optimization: This scheduling verified the effectiveness of the rule "automatically activate law enforcement recorders during nighttime operations," and the system automatically increased the confidence level of this rule in the rule engine to 0.95; Contingency Plan Optimization: Based on the results of this implementation, the parameter "early warning stay duration in remote areas" in the "Contingency Plan for Handling Abnormal Personnel in the Park" has been optimized from 30 minutes to 25 minutes.

[0084] Example 8: This embodiment of the invention discloses a storage medium storing a computer program that can be read by a computer. The computer program is configured to execute steps in an event- and context-aware cross-modal communication resource scheduling method at runtime.

[0085] The aforementioned storage media may include, but are not limited to, USB flash drives, read-only memory, portable hard drives, magnetic disks, optical disks, and other media capable of storing computer programs.

[0086] Example 9: This embodiment of the invention discloses an electronic device, including a processor and a memory, wherein the memory stores a computer program, which is loaded and executed by the processor to implement the steps in the event- and context-aware cross-modal communication resource scheduling method.

[0087] The processor described above can be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an ASIC, an FPGA, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. It can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The memory can include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, portable hard drives, magnetic disks, or optical disks.

[0088] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0089] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0090] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0091] The above content is only a specific embodiment of the present invention, which has strong adaptability and implementation effect. However, the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention. Therefore, equivalent changes made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A cross-modal communication resource scheduling method based on event and context awareness, characterized in that, include: In response to event triggering in converged communication scenarios, multi-source heterogeneous contextual data related to the event is collected. A three-level progressive multi-source data fusion method is introduced to fuse the multi-source heterogeneous contextual data to obtain a standardized and unified contextual view. Input the feature vector of the standardized unified context view and the standardized event feature vector into the context-event joint representation model to obtain the context-event joint representation vector, where the context-event joint representation model is constructed through deep learning; The standardized unified context view and standardized events are input into the rule engine to obtain mandatory scheduling constraints. The context-event joint representation vector is input into the lightweight scheduling model to obtain adaptive scheduling decisions. The mandatory scheduling constraints and adaptive scheduling decisions are then fused to generate the final standardized scheduling strategy. The lightweight scheduling model is constructed through a joint training framework of knowledge distillation and attention distillation.

2. The cross-modal communication resource scheduling method based on event and context awareness according to claim 1, characterized in that, In response to an event triggered in a converged communication scenario, multi-source heterogeneous contextual data related to the event is collected. A three-level progressive multi-source data fusion method is introduced to fuse the multi-source heterogeneous contextual data to obtain a standardized and unified contextual view, including: In response to events triggered in converged communication scenarios, multi-source heterogeneous contextual data related to the events are collected, including network contextual data, terminal contextual data, user contextual data, and business contextual data. Differentiated cleaning is performed on multi-source heterogeneous context data, and four-level association is carried out in terms of time dimension, spatial dimension, hierarchical dimension and subject dimension; A three-level progressive multi-source data fusion method is introduced to fuse multi-source heterogeneous contextual data to obtain a standardized and unified contextual view. The three-level progressive multi-source data fusion method includes data layer fusion, feature layer fusion, and decision layer fusion.

3. The cross-modal communication resource scheduling method based on event and context awareness according to claim 1 or 2, characterized in that, The input standardized unified context view feature vector and standardized event feature vector are fed into the context-event joint representation model to obtain the context-event joint representation vector, including: The events in the converged communication scenario are identified, and their features are extracted and standardized sequentially to obtain standardized event feature vectors. Input the feature vectors of the standardized unified context view and the standardized event feature vectors into the context-event joint representation model to obtain the context-event joint representation vector, where the context-event joint representation model is constructed through deep learning.

4. The cross-modal communication resource scheduling method based on event and context awareness according to claim 1 or 2, characterized in that, The input standardized unified context view and standardized events are fed into the rule engine to obtain mandatory scheduling constraints. The input context-event joint representation vector is fed into the lightweight scheduling model to obtain adaptive scheduling decisions. The mandatory scheduling constraints and adaptive scheduling decisions are then fused to generate the final standardized scheduling strategy, including: Input standardized unified context view and standardized events into the rules engine to obtain mandatory scheduling constraints; Inputting the context-event joint representation vector into the lightweight scheduling model yields adaptive scheduling decisions, which include optimal terminal selection, optimal communication mode decision, and optimal media path planning. Within the framework of mandatory scheduling constraints, mandatory scheduling constraints and adaptive scheduling decisions are integrated to generate the final standardized scheduling strategy, which is then distributed after passing security and compliance verification.

5. The cross-modal communication resource scheduling method based on event and context awareness according to claim 4, characterized in that, The construction process of the lightweight scheduling model includes: Several samples were obtained and divided into training set, validation set and test set. Each sample includes a context-event joint representation vector obtained after the event is triggered in the converged communication scenario and a corresponding adaptive scheduling decision label. The adaptive scheduling decision includes optimal terminal selection, optimal communication mode decision and optimal media path planning. The training set is input into the initial student model and the pre-trained teacher model. The student model is then trained using knowledge distillation with the teacher model to obtain the trained student model. The teacher model is learned using a number of samples. The total loss function when using the teacher model to train the student model using knowledge distillation is... ,in For cross-entropy loss, For knowledge distillation loss, For distillation loss, For priority-weighted loss, These are the weighting coefficients; The model parameters of the trained student model are optimized using the validation set and the test set, and the student model that meets the test evaluation requirements is output as a lightweight scheduling model.

6. The cross-modal communication resource scheduling method based on event and context awareness according to any one of claims 1 to 5, characterized in that, It also includes converting standardized scheduling strategies into standardized executable instruction objects, issuing and executing them according to priority, monitoring key indicators during execution, and optimizing scheduling decisions and model parameters in reverse.

7. A cross-modal communication resource scheduling system based on event and context awareness, applying the method described in any one of claims 1 to 6, characterized in that, include: The input perception layer includes: The multi-source heterogeneous context data acquisition module, in response to event triggering in the converged communication scenario, acquires multi-source heterogeneous context data related to the event; The multi-source heterogeneous context data fusion module introduces a three-level progressive multi-source data fusion method to fuse multi-source heterogeneous context data and obtain a standardized and unified context view; The context-event joint representation module takes the feature vector of the standardized unified context view and the standardized event feature vector as input to the context-event joint representation model to obtain the context-event joint representation vector. The context-event joint representation model is constructed through deep learning. The intelligent decision-making layer inputs a standardized unified context view and standardized events into the rule engine to obtain mandatory scheduling constraints. It inputs a context-event joint representation vector into the lightweight scheduling model to obtain adaptive scheduling decisions. Finally, it integrates the mandatory scheduling constraints and the adaptive scheduling decisions to generate the final standardized scheduling strategy. The lightweight scheduling model is constructed through a joint training framework of knowledge distillation and attention distillation.

8. The event- and context-aware cross-modal communication resource scheduling system according to claim 7, characterized in that, The intelligent decision-making layer includes: The rules engine takes standardized unified context views and standardized events as input and obtains mandatory scheduling constraints. The lightweight scheduling decision engine takes the context-event joint representation vector as input to the lightweight scheduling model and obtains adaptive scheduling decisions, which include optimal terminal selection, optimal communication mode decision and optimal media path planning. The scheduling decision fusion and adjustment module integrates mandatory scheduling constraints and adaptive scheduling decisions within the framework of mandatory scheduling constraints to generate the final standardized scheduling strategy, which is then distributed after passing security permission and compliance verification. or / and, Also includes: The strategy execution layer converts standardized scheduling strategies into standardized executable instruction objects and issues them for execution according to priority. The closed-loop optimization layer monitors the core metrics during the execution process after execution and reverse-optimizes scheduling decisions and model parameters.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, which is loaded and executed by the processor to implement the steps of the method as claimed in any one of claims 1 to 6.

10. A storage medium, characterized in that, The storage medium stores a computer program that can be read by a computer, the computer program being configured to execute the steps of the method as described in any one of claims 1 to 6 when it is run.