Cross-domain multi-modal flow security interaction method based on intranet AI engine

Abstract

The invention belongs to the technical field of data interaction, and particularly relates to a cross-domain multi-mode flow safety interaction method based on an intranet AI engine. The method comprises the following steps: after receiving an extranet service request, an AI engine middle station analyzes the power service attribute of the service request, predicts the core demand characteristic of the service request for data transmission, and preloads and initializes an optimal transmission strategy template according to the core demand characteristic; continuously monitoring the real-time state of the whole transmission process, and dynamically deciding and commanding transmission link adjustment according to the real-time state; and active safety protection is realized through dynamic key management and intelligent flow analysis. According to the method, the real-time performance, reliability and efficiency of key monitoring data transmission are remarkably improved in a complex power network environment.

Description

Technical Field

[0001] This invention belongs to the field of data interaction technology, specifically relating to a cross-domain multimodal stream secure interaction method based on an intranet AI engine. Background Technology

[0002] As power grid companies deepen their integrated marketing efforts, the shift from "internal business-centric" to "customer experience-centric" and from "extensive operation" to "lean service" has become the core guiding principle for service upgrades. However, under the centralized customer service operation framework, the demand from electricity customers for services such as visualized interpretation of electricity pricing policies, fault repair, and electricity bill inquiries continues to surge, posing a severe challenge to traditional service channels: the 95598 hotline is under constant pressure during peak hours, business halls are operating beyond capacity for inquiries, and electronic channels, lacking immersive interactive capabilities, cannot effectively alleviate the pressure on manual channels; in particular, the internal and external network security isolation mechanism prevents visualized business content from penetrating to the internet side, leaving customers' needs for real-time inquiries and handling of fault reports and electricity bill inquiries via mobile devices unmet for a long time.

[0003] Traditional remote service models have fundamental limitations: their workflow relies on customers proactively calling to inquire, and agents must cross systems to query scattered data such as fault reporting progress and electricity bill details before providing feedback. This passive response mechanism not only increases the pressure on customer service but also leads to excessively long customer wait times and complaints. Summary of the Invention

[0004] To overcome the problems in the existing technology, this invention proposes a cross-domain multimodal stream secure interaction method based on an intranet AI engine.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This invention provides a cross-domain multimodal stream secure interaction method based on an intranet AI engine, comprising the following steps: After receiving an external network service request, the AI ​​engine platform parses the power service attributes of the service request and predicts the core data transmission requirements of the service request. Based on this, it preloads and initializes the optimal transmission strategy template. It continuously monitors the real-time status of the entire transmission process and makes dynamic decisions and directs the transmission link to adjust based on the real-time status; it achieves proactive security protection through dynamic key governance and intelligent traffic analysis.

[0006] Further, the power service attributes of the service request are parsed, including: After receiving a business request from an external client, the AI ​​engine platform performs security verification and format parsing on the business request. After verifying the legality of the business request, the platform performs word segmentation, entity recognition, and intent classification on the text of the requested business, and extracts key elements, including business type, device information, and operation requirements. The extracted key elements are matched and mapped with the built-in power business knowledge base. Through fuzzy matching algorithm, unstructured business requests are transformed into structured power business attributes.

[0007] Furthermore, the security verification includes: using a verification algorithm or message authentication code to verify whether the data packet has been damaged or tampered with during transmission; if it has been damaged or tampered with, the security verification fails, an error message is returned, and a resend request is required; otherwise, the security verification passes.

[0008] Furthermore, prior to security verification, the source of the business request is authenticated, verifying whether the sender's identity information matches the authorization information pre-stored in the system; if they match, the source authentication is successful; if they do not match, the business request is rejected and an abnormal event is recorded.

[0009] Furthermore, the core characteristics of the data transmission requirements predicted by the service request include: Regularly conduct in-depth mining of historical data in the power business knowledge base to analyze the changing trends of core business requirements. These core requirements include the required data modes to be transmitted, the maximum end-to-end tolerable latency for each data mode, reliability requirements and bandwidth budget, and the correlation between data from multiple data modes.

[0010] Furthermore, the optimal transport strategy template is preloaded and initialized, including: For text commands, the pre-configured WebSocket connection is set to the highest message priority, minimum protocol overhead, and low heartbeat interval mode. For video streams, the congestion control algorithm parameters of WebRTC PeerConnection are pre-configured to aggressive mode, low-latency TURN server candidate paths are pre-selected, and an initial forward error correction strategy is set.

[0011] Furthermore, the system continuously monitors the real-time status of the entire transmission process and dynamically makes decisions and directs adjustments to the transmission link based on the real-time status, including: The AI ​​engine platform continuously monitors the real-time status of the entire transmission process, including network performance indicators and data transmission efficiency. Based on the real-time status, the AI ​​engine platform dynamically makes decisions and directs the adjustment of the transmission link, which includes intelligent switching of WebRTC path, adaptive WebSocket transmission mode, and cross-modal resource collaborative scheduling.

[0012] Furthermore, WebSocket transport modes are adaptive, including: When network jitter J exceeds the preset threshold, the AI ​​engine platform decides to switch the WebSocket connection from streaming mode to message mode and enable a stricter retransmission confirmation mechanism. When the network jitter J is less than the preset threshold, the network will stabilize before switching to backflow mode to reduce the transmission delay of text in normal state.

[0013] Furthermore, cross-modal resource collaborative scheduling includes: Monitor the load of the TURN server in real time. When it is found that the TURN server is overloaded due to carrying high-priority video streams, assess the currently available bandwidth resources B. araiaule And the bandwidth requirements of various services B require The resource assessment results were obtained. Based on the resource assessment results, the decision was made to route some low-relevance, low-priority text WebSocket connections to other available gateways.

[0014] Furthermore, intelligent WebRTC path switching includes: Set path quality evaluation metrics, including packet loss rate threshold L. threahol and delay value threshold D threshold ; If the detected packet loss rate L>L on the public network path threahold And the delay D > Dt hreshold Meanwhile, current business requirements necessitate delay D require When the latency is less than 100ms, the AI ​​engine platform makes a decision and instructs the relevant WebRTC PeerConnection to bypass the congested path and quickly switch to the pre-selected low-latency, low-load TURN server relay path.

[0015] Compared with the prior art, the present invention has the following technical effects: This invention provides a cross-domain multimodal stream security interaction method based on an intranet AI engine. Its core lies in acting as an intelligent decision-making hub and real-time optimization engine, deeply intervening in the entire communication lifecycle. By continuously sensing business needs, network status, and data characteristics, and making dynamic decisions based on power knowledge, it achieves collaborative orchestration, adaptive reconstruction, and intelligent resource scheduling of WebRTC (for audio and video streams) and WebSocket (for text and control commands) transmission links. This significantly improves the real-time performance, reliability, and efficiency of critical monitoring data transmission in complex power network environments such as strong NAT / firewall isolation, bandwidth fluctuations, and high security requirements. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a schematic diagram illustrating an embodiment of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solutions proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. Specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] In one embodiment of the present invention, reference is made to... Figures 1-2 This paper provides a method for secure cross-domain multimodal stream interaction based on an intranet AI engine, including the following steps: Step 100: After receiving the external network service request, the AI ​​engine platform parses the power service attributes of the service request and determines the core requirements of the service request for data transmission. Based on this, it preloads and initializes the optimal transmission strategy template. Step 200: The AI ​​engine platform continuously monitors the real-time status of the entire transmission process, and makes dynamic decisions and directs the adjustment of the transmission link based on the real-time status. Step 300: The AI ​​engine platform achieves proactive security protection through dynamic key governance and intelligent traffic analysis.

[0020] The following is a detailed explanation of each of the above steps: Step 100: After receiving an external network service request, the AI ​​engine platform parses out the power service attributes of the service request and determines the core requirements of the service request for data transmission. Based on this, it preloads and initializes the optimal transmission strategy template.

[0021] As an example, step 100 specifically includes: Step 101: When the AI ​​engine platform receives a service request initiated by an external network sender, it immediately parses the power service attributes of the service request.

[0022] The core objective of this step is to transform users' unstructured natural language requests into machine-readable structured power business attributes.

[0023] The power AI engine platform utilizes WebRTC (Web Real-Time Communication) components and WebSocket technology to construct a multimodal internal and external network interaction architecture. WebRTC is used for real-time audio and video communication and data sharing, while WebSocket is a TCP-based protocol that allows full-duplex communication between the server and client.

[0024] Step 101 specifically includes the following sub-steps: Step 1011: After receiving a business request initiated from the external network, the AI ​​engine platform performs security verification and format parsing on the business request and extracts key elements.

[0025] Perform security checks and format parsing on business requests to verify the legitimacy of the external sender's identity and the integrity of the data, ensuring that only compliant requests can proceed to the next stage of the process.

[0026] The source of business requests is authenticated using technologies such as digital certificates and identity tokens to ensure that the requests originate from legitimate users or devices; the sender's identity information is verified to match the authorization information pre-stored in the system. For example, for an electricity bill inquiry request, the system checks whether the identity token provided by the user matches the user's identity recorded in the system; if they do not match, the request is rejected and the abnormal event is recorded.

[0027] Verification algorithms (such as CRC checksums) or Message Authentication Codes (MACs) are used to verify whether data packets have been corrupted or tampered with during transmission. If the verification fails, an error message is returned to the sender on the external network, requesting a retransmission. For example, in a fault reporting request, the system checks whether key fields such as user ID, fault address, mobile phone number, and user name are complete and correctly formatted; if the user ID is missing or incorrectly formatted, an error message is immediately returned.

[0028] After verifying the legitimacy of the business request, natural language processing (NLP) technology is used to segment the text of the request, identify entities, and classify intents. Key elements such as business type, equipment information, and operational requirements are extracted, laying the foundation for subsequent business attribute mapping. For example, through a series of NLP operations such as segmentation, entity recognition and extraction, and intent classification, key elements such as "electricity meter" (equipment information), "tripped" (fault phenomenon), "repair" (operational requirement), "city and district" (address), and "130xxxxxxx" (phone number) are extracted from a sentence like "My electricity meter tripped, I need to report it for repair. My address is [city name], [district name], and my mobile phone number is 130xxxxxxx." Ultimately, the user's intent is determined to belong to the "fault repair" business type.

[0029] Step 1012: Match the key elements with the built-in power business knowledge base to transform them into structured power business attributes.

[0030] The extracted key elements are matched and mapped with the built-in power business knowledge base. Since the description of the business request may be ambiguous, the AI ​​engine platform matches and maps the extracted key elements with the built-in power business knowledge base when parsing the power business attributes.

[0031] For services with a high degree of matching, further confirmation of service details with the user is conducted to avoid misunderstandings. For example, for a request like "My electricity meter has tripped and needs to be repaired," NLP is used to extract key information such as "electricity meter tripped" and "repair," and the engine uses fuzzy matching to determine that it may belong to a fault repair service, that is, transforming unstructured user input into structured power service attributes.

[0032] This step transforms scattered key elements into a standardized, structured dictionary of business attributes, such as specifying the business type as fault reporting. This structured attribute provides accurate and reliable input for the AI ​​engine to subsequently predict data transmission needs and formulate optimal transmission strategies.

[0033] Step 102: Based on the structured power business attributes, predict the core requirements of the business request for data transmission using a prediction method based on multimodal time series deep learning.

[0034] Multi-dimensional features of historical business data, including business attributes, network performance indicators, and environmental context data, are extracted from the power business knowledge base to form a time-series sample set.

[0035] Structured power business attributes are the direct output of the AI ​​engine after parsing and standardizing the currently received real-time business requests. They originate from the instant analysis of individual business requests and represent the explicit business context of this specific request. For example, this is a "fault reporting" business, associated with a specific device, requiring "low-latency video stream" and "high-reliability control commands".

[0036] Multidimensional features are massive amounts of historical data extracted by the AI ​​engine from a historically accumulated power business knowledge base, used to train and drive predictive models. They are no longer attributes of a single request, but rather time-series datasets recorded during numerous past business executions, containing business attributes, network performance, and environmental context.

[0037] Structured power business attributes serve as real-time input instances of multi-dimensional features, which in turn form the basis for predictions based on historical business data. The AI ​​engine uses these structured power business attributes as the current input to match and query patterns and rules learned from historical multi-dimensional feature data, thereby predicting the core demand characteristics of the current request in the future. In short, the real-time parsing of structured power business attributes is like "asking a question," while the model learned from historical multi-dimensional features provides the basis for the answer; only by combining both can accurate predictions be achieved.

[0038] By capturing long-term dependencies of network metrics through multi-scale temporal convolutional networks (TCNs), contextual modeling of business semantic features is performed, and cross-modal attention mechanisms are used to fuse business, network, and environmental features to generate a unified deep representation vector.

[0039] Based on this, a multi-task learning model is constructed, using a shared feature extraction layer and multiple task-specific output heads to simultaneously predict core requirements such as end-to-end latency tolerance, reliability level, bandwidth requirement range, and multimodal synchronization accuracy.

[0040] The model training employs an uncertainty-weighted strategy to balance the multi-task loss and optimizes the parameters.

[0041] The system also integrates an online learning mechanism, regularly fine-tunes the model using newly added transmission data, and dynamically corrects the prediction results by combining real-time network status data. Finally, the power business rules engine performs compliance verification and adjustment on the output to ensure that the prediction results conform to both data-driven principles and business security requirements.

[0042] Regularly conduct in-depth data mining of historical data in the power business knowledge base to analyze the changing trends of core business requirements. These core requirements include the required data modes to be transmitted, the maximum end-to-end tolerable latency for each data mode, reliability requirements and bandwidth budgets, and the coordination and synchronization requirements between multiple data modes. Data modes include video streams, audio streams, structured alarm text, and control commands; the maximum end-to-end tolerable latency for each data mode is, for example, fault video <100ms, trip command <50ms; and reliability requirements include zero loss of control commands.

[0043] Step 103: Based on the predicted core requirement characteristics, preload and initialize the optimal transport strategy template for WebRTC PeerConnection and Web Socket connections.

[0044] PeerConnection is a key component of WebRTC technology, responsible for establishing a direct peer-to-peer connection between two browsers or devices. This connection allows audio and video streams and data to be transmitted directly between devices without going through an intermediate server, thereby reducing latency, improving communication efficiency, and enhancing communication security.

[0045] For ultra-low latency trip text commands, the pre-configured WebSocket connection is set to the highest message priority, minimum protocol overhead, and low heartbeat interval mode.

[0046] For video streams from high-bandwidth devices, the WebRTC PeerConnection congestion control algorithm parameters are pre-configured to aggressive mode, low-latency TURN server candidate paths are pre-selected, and an initial FEC (Forward Error Correction) strategy is set. The congestion control algorithm is GCC (Google Congestion Control). Specifically, by calling the WebRTC API, parameters of the congestion control algorithm are set, such as increasing the magnitude of the transmission rate adjustment; selecting low-latency servers as candidate paths from the pre-configured TURN server list; and setting parameters such as the redundancy of FEC encoding.

[0047] Step 200: The AI ​​engine platform continuously monitors the real-time status of the entire transmission process and makes dynamic decisions and directs the adjustment of the transmission link based on the real-time status.

[0048] As an example, step 200 specifically includes: Step 201: The AI ​​engine platform continuously monitors the real-time status of the entire transmission process, including network performance indicators and data transmission efficiency.

[0049] Use the WebRTC Stats API to obtain metrics such as end-to-end RTT (Round-Trip Time), jitter, packet loss rate, and available bandwidth. Simultaneously, deploy custom probes to periodically send test packets to key nodes in the network, and calculate network performance metrics based on the returned results. Obtain metrics such as STUN / TURN server load, WebSocket gateway connection count, and CPU utilization through server management interfaces or monitoring tools. For example, measure RTT by sending ICMP (Internet Control Message Protocol) packets and calculate jitter by analyzing the arrival time intervals of data packets; use the SNMP (Simple Network Management Protocol) protocol to obtain server load information.

[0050] Data transmission performance monitoring includes: WebSocket messages: Record the sending and receiving times of WebSocket messages, calculate the end-to-end latency, and calculate the retransmission rate by showing the ratio of the number of retransmitted messages to the total number of messages.

[0051] WebRTC Media Streams: The WebRTC media processing module obtains metrics such as the actual bitrate, rendering latency, and stuttering rate of the media stream. For example, stuttering rate can be calculated by analyzing frame rate changes and rendering time.

[0052] Step 202: Based on the real-time status, the AI ​​engine platform dynamically makes decisions and directs the adjustment of the transmission link, which includes intelligent switching of WebRTC path, adaptive WebSocket transmission mode, and cross-modal resource collaborative scheduling.

[0053] For intelligent switching of WebRTC paths, including: Set path quality evaluation metrics, such as packet loss rate threshold L. threahol =5%, Delay value D threshold =150ms; If the detected packet loss rate L>L on the public network path threahold And the delay D > Dt hreshold Meanwhile, current business requirements necessitate delay D require When the latency is less than 100ms, the AI ​​engine platform immediately makes a decision and instructs the relevant WebRTC PeerConnection to bypass the congested path and quickly switch to the pre-selected low-latency, low-load TURN server relay path to ensure the real-time performance of critical status video of high-voltage equipment.

[0054] Based on real-time load across multiple TURN server nodes and network topology distance The system uses intelligent scheduling algorithms to select the optimal node. For example, a weighted scoring method can be used to calculate the score for each node. ,in, and These are the weighting coefficients. The maximum network topology distance is used to select the node with the highest score as the new transmission path.

[0055] For WebSocket transport mode adaptation, including: When network jitter J exceeds the preset threshold, the AI ​​engine platform decision command WebSocket connection switches from streaming mode to message-based mode and enables a stricter retransmission confirmation mechanism to ensure the reliable and orderly delivery of key alarm texts such as "circuit breaker fault lockout". When the network jitter J is less than the preset threshold, the network will stabilize before switching to backflow mode to reduce the transmission delay of text in normal state.

[0056] For cross-modal resource collaborative scheduling, including: Monitor the load of the core TURN server in real time. When it is found that the TURN server is overloaded due to carrying high-priority video streams, assess the currently available bandwidth resources B. araiaule And the bandwidth requirements of various services B require The resource assessment results were obtained.

[0057] Based on the resource assessment results, the decision was made to route some low-relevance, low-priority text WebSocket connections to other available gateways; simultaneously, the media processing unit was instructed to temporarily reduce the encoding bitrate R or resolution of non-critical video streams. Res For example, according to the relationship model between video quality and bitrate, Q= f (R), where Q is the video quality. Under the premise of ensuring a certain video quality, the bitrate is reduced to free up bandwidth resources.

[0058] Step 300: The AI ​​engine platform achieves proactive security protection through dynamic key governance and intelligent traffic analysis.

[0059] As an example, step 300 specifically includes: Step 310: Based on session security sensitivity, duration, and real-time network risk awareness, dynamically decide and execute the WebSocket session key and WebRTC SRTP key rotation strategy.

[0060] The AI engine middleware dynamically determines the rotation period of the WebSocket session key and the WebRTC SRTP key based on the session security sensitivity S (such as high session security sensitivity for sessions involving critical control instructions), the duration T, and the real-time network risk perception R (such as high risk when a network attack is detected). . For example, using the formula , where is the basic rotation period, , , are the weight coefficients.

[0061] According to the determined rotation period, call the interface of the encryption library to generate a new key, and update the encryption configurations of WebSocket and WebRTC to achieve dynamic key rotation.

[0062] Step 3: Continuously analyze the pattern features of the WebSocket message stream and the WebRTC media stream, give real-time alarms for suspected abnormal behaviors, limit the flow rate or block illegal requests, and dynamically update the access control rules.

[0063] The AI engine middleware continuously analyzes the pattern features of the WebSocket message stream and the WebRTC media stream, such as the message frequency F, the message size Size, the encoding format Code of the media stream, etc. Establish a normal traffic pattern feature library, and detect suspected abnormal behaviors by comparing the real-time traffic features with the normal feature library.

[0064] For the detected suspected abnormal behaviors, such as high-frequency small messages that do not conform to the power service logic (message frequency F > Fthrehold and message size Size < Sizethreahola), media stream requests with abnormal sources, the engine actively makes decisions and executes intervention measures, such as giving real-time alarms, limiting the flow rate of suspicious connections (limiting the maximum bandwidth Bimit of the connection) or directly blocking illegal requests, and dynamically updating the access control rules of the WebSocket gateway and the TURN server, such as adding blacklists or whitelists.

[0065] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. An intra-network AI engine-based cross-domain multi-modal flow security interaction method, characterized in that, The method comprises the following steps: The AI engine platform receives the external network service request, analyzes the power service attributes of the service request, and predicts the core demand characteristics of the data transmission of the service request, thereby preloading and initializing the optimal transmission strategy template; Continuously monitor the real-time state of the entire transmission process, and dynamically decide and command the transmission link adjustment according to the real-time state; Through dynamic key management and intelligent flow analysis, active security protection is realized.

2. The cross-domain multi-modal flow security interaction method based on the intranet AI engine according to claim 1, characterized in that, Analyzing the power service attributes of the service request comprises: After receiving the external network client service request, the AI engine platform performs security verification and format analysis on the service request; after verifying the legality of the service request, the text of the request service is segmented, entity recognition and intent classification are performed, and key elements including service type, device information and operation requirements are extracted; The extracted key elements are matched and mapped with the built-in power service knowledge base, and the unstructured service request is converted into structured power service attributes through a fuzzy matching algorithm.

3. The cross-domain multi-modal flow security interaction method based on the intranet AI engine according to claim 2, characterized in that, The security verification includes: using a verification algorithm or a message authentication code to verify whether the data packet is damaged or tampered with during transmission; if damaged or tampered with, the security verification fails, and an error message is returned, requiring retransmission of the request; otherwise, the security verification is passed.

4. The cross-domain multi-modal flow security interaction method based on the intranet AI engine according to claim 3, characterized in that, Before security verification, the source of the service request is also authenticated to verify whether the identity information of the sender matches the authorized information pre-stored in the system; if matched, the source authentication is passed; if not matched, the service request is rejected and an abnormal event is recorded.

5. The cross-domain multi-modal flow security interaction method based on the intranet AI engine according to claim 2, characterized in that, Predicting the core demand characteristics of the data transmission of the service request comprises: Periodically deep mining historical data in the power service knowledge base to analyze the trend of changes in core demand characteristics, wherein the core demand characteristics include the data modalities required for transmission, the end-to-end maximum tolerable delay of each data modality, reliability requirements and bandwidth budget, and the spatio-temporal correlation between multiple data modalities.

6. The cross-domain multi-modal flow security interaction method based on the intranet AI engine according to claim 5, characterized in that, Preloading and initializing the optimal transmission strategy template comprises: For text instructions, preconfigure the WebSocket connection to the highest message priority, minimum protocol overhead, and low heartbeat interval mode; For video streaming, preconfigure the congestion control algorithm parameters of WebRTC PeerConnection to aggressive mode, preselect low-delay TURN server candidate paths, and set the initial forward error correction strategy.

7. The cross-domain multi-modal flow security interaction method based on the intranet AI engine according to claim 1, characterized in that, Continuously monitoring the real-time state of the entire transmission process and dynamically deciding and commanding the transmission link adjustment according to the real-time state comprises: The AI engine platform continuously monitors the real-time state of the entire transmission process, including network performance indicators and data transmission efficiency; According to the real-time state, the AI engine platform dynamically decides and commands the transmission link adjustment, including WebRTC path intelligent switching, WebSocket transmission mode adaptation, and cross-modal resource collaborative scheduling.

8. The cross-domain multi-modal flow security interaction method based on the intranet AI engine according to claim 7, characterized in that, WebSocket transmission mode adaptation comprises: When network jitter J exceeds the preset threshold, the AI engine middle platform decision instruction WebSocket connection switches from streaming mode to message mode, and enables a more stringent retransmission confirmation mechanism; When network jitter J is less than the preset threshold, after the network recovers smoothly, it is decided to switch back to streaming mode to reduce the transmission delay of regular state text.

9. The cross-domain multi-modal flow security interaction method based on the intranet AI engine according to claim 7, characterized in that, Cross-modal resource collaborative scheduling, including: Real-time monitoring of the load of the TURN server, when the TURN server is found to be overloaded due to carrying high-priority video streams, evaluating the currently available bandwidth resources B araiaule and the bandwidth demand B of each service require , obtaining a resource evaluation result; According to the resource evaluation result, it is decided to route part of the low correlation and low priority text WebSocket connection to other available gateways.

10. The cross-domain multi-modal flow security interaction method based on the intranet AI engine according to claim 7, characterized in that, WebRTC path intelligent switching, including: Set the path quality evaluation index, including the packet loss rate threshold L threahol And the delay value threshold D threshold ; If the packet loss rate L of the public network path is detected to be L>L threahold and the delay D is D>Dt hreshold and the current service requires a delay D require <100ms, the AI engine platform decision and instruct the relevant WebRTC PeerConnection bypass congestion path, fast switching to the pre-selected low delay, low load TURN server relay path.

Images