End-edge-cloud three-end collaborative low-energy-consumption cognitive artificial intelligence system and implementation method thereof

Abstract

The invention discloses an end-edge-cloud three-end collaborative low-energy-consumption cognitive artificial intelligence system and an implementation method thereof, and belongs to the field of artificial intelligence and edge computing. The system adopts a software and hardware integrated layered decoupling elastic architecture, and comprises an end node, an edge node and a cloud node. The end node is integrated with an intelligent hardware layer and a cognitive AI module group, and can independently undertake full-amount cognitive calculation in a single-node mode. According to the system, energy consumption is reduced through a low-consumption time sequence pulse type calculation mechanism, decision reliability is improved through root multi-node cooperative verification, a new hardware access period is shortened through a computing power non-inductive plug-and-play cross-hardware adaptive interface, and self-iterative optimization is achieved through full-link feedback. And the system can initiate active guide type interaction when the cognitive confidence coefficient is insufficient, so that man-machine collaborative decision making is realized. According to the method, the problems of high end side energy consumption, insufficient decision reliability, high hardware adaptation cost and mechanical one-way interaction mode in the prior art are effectively solved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence edge computing, embodied intelligent hardware, edge-cloud collaboration and cognitive intelligence, and specifically to a low-energy cognitive AI system and its implementation method that integrates edge-cloud collaboration. Background Technology

[0002] With the development of edge-side intelligent hardware and cloud-edge collaboration technologies, edge devices have made significant progress in perception capabilities. However, several problems still exist in actual deployment: common architectures typically adopt a fixed "low-end, mid-edge, high-cloud" computing power hierarchy, making it difficult to adapt to the trend of upgrading edge node computing power; insufficient hardware-software coupling leads to high costs and long integration cycles for new hardware access and adaptation; traditional edge-side cognitive algorithms are mainly based on continuous floating-point neural networks, resulting in high energy consumption and impacting battery life; decision confidence and reliability are still insufficient in high-reliability scenarios (such as automotive and industrial applications); human-computer interaction is mechanized, lacking natural multimodal interaction based on physiological-emotional-scenario fusion. More importantly, existing systems all adopt a unidirectional open-loop paradigm of "perception-computation-output," lacking the ability to engage in bidirectional, collaborative dialogue with users during the cognitive process. When the system's confidence is insufficient or faces multiple possibilities, it cannot clarify intentions, obtain key information, or guide collaboration through proactive questioning, thus limiting the accuracy, personalization, and naturalness of its decision-making and user experience. Therefore, there is a need for an edge-cloud collaborative cognitive AI system and method that can reduce edge power consumption, improve decision reliability, shorten hardware adaptation cycle, and support proactive collaborative cognitive interaction.

[0003] This invention aims to overcome the shortcomings of the prior art and provide a low-energy cognitive AI system and its implementation method that integrates edge, cloud, and end-device collaboration. To achieve the above objective, the technical solution of this invention is as follows: A low-energy cognitive AI system integrating edge, cloud, and end-device collaboration adopts a layered, decoupled, and flexible architecture that integrates hardware and software. The system includes end nodes, edge nodes, and cloud nodes. The end nodes, as the core of perception and execution, integrate a full-stack module from data acquisition to physical interaction, and innovatively introduce a low-power time-series pulse computing unit to reduce energy consumption, and a root-cause collaborative verification module to improve decision reliability. The meta-dynamic decision kernel not only generates the final decision instruction based on the verified features, but also possesses proactive cognitive interaction judgment capabilities. When the kernel determines that the confidence level of the current cognitive state has not reached a preset threshold, or when the user needs to supplement information or make collaborative choices to achieve a better decision goal, the kernel will generate guided, open-ended inquiry or dialogue instructions, and proactively initiate interaction with the user through a multimodal physical interaction layer, thereby upgrading the traditional one-way decision-making closed loop to a human-machine two-way collaborative cognitive enhancement closed loop. The three nodes are connected via a standardized protocol, enabling flexible scheduling and free flow of computing power, data, and models. The beneficial effects of this invention include: 1) significantly reducing edge-side energy consumption and extending device battery life through a flexible computing architecture and pulse computing; 2) greatly improving the reliability and security of cognitive decision-making through multi-node root cause verification; 3) greatly shortening the integration and adaptation cycle of new hardware through plug-and-play cross-hardware adaptation interfaces; 4) enabling the system to continuously self-optimize through end-to-end feedback iteration; and 5) introducing a proactive interaction mechanism based on cognitive confidence, enabling the system to initiate collaborative dialogue when information is insufficient, improving decision-making accuracy and the naturalness of human-machine collaboration, unlike the one-way command mode of existing technologies. Attached Figure Description

[0004] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention. Figure 2 A diagram showing the functional division of labor and data flow among the three nodes: endpoint, edge, and cloud. Figure 3 This is a schematic diagram of the structure of the end node embodying the intelligent hardware layer. Figure 4 This is a flowchart of the method for implementing the present invention. In the diagram: 1-End node, 2-Edge node, 3-Cloud node, 4-Embodied intelligent hardware layer, 5-Cross-hardware adaptation interface layer, 6-Multimodal haptic signal fusion layer, 7-Low-power timing signal computation layer, 8-Scene cognition memory cluster, 9-Root cause cognition verification module, 10-Meta-power decision kernel, 11-Multimodal physical interaction layer.

[0005] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be noted that the following embodiments are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0006] Example 1: Wearable Scenario (Demonstrating Active Cognitive Interaction) This example uses a wearable health device as an end node application scenario. System Configuration: The end node is a smart wearable device integrating sensors such as heart rate and acceleration, and the main controller is a low-power processor. The edge node is a home gateway, and the cloud node is a health cloud service platform. Workflow: Data Acquisition (S1) and Cognitive Computation (S3): The device acquires physiological signals. The low-power time-series signal computation layer performs pulse-based feature extraction on heart rate variability.

[0007] Cognitive Verification (S4): The root cognitive verification module compares the current feature with multiple scene models in the scene cognitive memory cluster. The results show that the matching degree does not reach the high confidence threshold, and there is ambiguity such as "work excitement" or "anxiety".

[0008] Decision Making and Proactive Interaction (S5, S5a): The Metadynamic Decision Kernel determines that the confidence level is insufficient and does not force the output of potentially erroneous conclusions. Instead, it generates guiding inquiry instructions: "Your body shows a high state of focus. Is this efficient work, or does it need adjustment?" Contextualized Interaction (S6): The multimodal physical interaction layer broadcasts the inquiry via screen display and voice.

[0009] Feedback Iteration (S7): The user verbally responds, "I'm concentrating on my work." The system reinforces the current physiological data by associating it with the "efficient work" memory model and suppresses subsequent unnecessary relaxation reminders. This process demonstrates that the system achieves cognitive alignment through proactive dialogue, rather than a mechanical response.

[0010] Example 2: In-vehicle Scenario (Demonstrating High Reliability and Collaborative Verification) System Configuration: End nodes are automotive-grade computing units, integrating cameras and steering wheel grip sensors. Edge nodes are in-vehicle domain controllers. Workflow: The system monitors the driver's state. The low-power timing signal computing layer processes visual features in a continuous pulse mode. When potential signs of distraction are identified, the root cause cognitive verification module simultaneously retrieves local traffic rules and real-time vehicle data (such as vehicle speed and lane) for cross-verification. If the verification confidence level is in the middle range, the meta-dynamic decision kernel may generate a confirmatory question such as "Are you looking for roadside signs?" instead of directly triggering an alarm, thus reducing false alarms while ensuring safety. The final decision instruction (such as a warning) is executed through steering wheel vibration (S6), and the driver's reaction is used to optimize the model (S7).

[0011] Example 3: Industrial Scenario (Demonstrating Flexible Collaboration) System Configuration: End nodes are equipment vibration sensors, and edge nodes are workshop servers. Workflow: Sensors acquire data at high frequency (S1). Simple features are extracted in real time at the end, while complex spectral analysis is performed by the edge nodes (S2, S3). Edge nodes call the global fault model distributed from the cloud for verification (S4). Based on the diagnostic confidence level, the system can autonomously decide on an alarm or generate an inquiry command (such as "Equipment X vibration mode is abnormal, is it recommended to arrange an inspection?") and send it to the maintenance personnel's terminal (S5, S5a, S6). All data flows back to the cloud to update the model (S7).

Claims

1. An end-edge-cloud three-end collaborative low-energy-consumption cognitive artificial intelligence system, characterized in that, It comprises an end node, an edge node and a cloud node; the end node comprises a body intelligent hardware layer, a cross hardware adaptation interface layer, a multi-modal somatosensory signal fusion layer, a low consumption time sequence signal calculation layer, a scene cognitive memory cluster, a root cognitive verification module, a meta-dynamic decision kernel and a multi-modal physical interaction layer connected in sequence; the output end of the multi-modal physical interaction layer is connected with the input end of the body intelligent hardware layer in signal to form a local interaction closed loop; the end node, the edge node and the cloud node realize bidirectional data interaction and elastic computing power scheduling through a standardized communication protocol.

2. The system of claim 1, wherein, The low consumption time sequence signal calculation layer is a pulse neural network calculation module, which adopts a mechanism of time sequence signal accumulation and threshold triggering to perform calculation; the root cognitive verification module is used to trigger retrieval and causal fusion of authoritative data sources of the cloud or the edge node when local verification confidence is lower than a threshold.

3. The system of claim 1 or 2, wherein, The meta-dynamic decision kernel has an active cognitive interaction judgment capability, and can generate active inquiry or cooperative guidance instructions when it is determined that decision confidence does not reach a threshold or information needs to be supplemented, and executes the instructions through the multi-modal physical interaction layer.

4. The system of claim 1, wherein, The end-edge-cloud three-end adopts a function positioning decoupling elastic computing power architecture, supports the end node to independently undertake full amount calculation, or performs two-node or three-node cooperative calculation with the edge node and the cloud node according to computing power load and scene demand.

5. The system of claim 1, wherein, The cross hardware adaptation interface layer is pre-provided with a heterogeneous hardware signal analysis protocol library, supports adaptation of new hardware by adding an analysis protocol, and is a computing power non-inductive structure.

6. The system of claim 1, wherein, The multi-modal somatosensory signal fusion layer constructs a physiological-emotional-scene multi-dimensional correlation network, and supports cross-layer feature fusion of the end, the edge and the cloud.

7. The system of claim 1, wherein, The scene cognitive memory cluster adopts an end-edge-cloud elastic cooperative storage architecture, and the end node can save data of any range from high priority local memory to global memory according to local storage capacity.

8. The system of claim 1, wherein, The body intelligent hardware layer comprises a multi-dimensional sensing module, a local low delay processing module and a scene physical output module, and supports local independent triggering of an emergency scene.

9. A method of implementing a system according to any one of claims 1 to 8, characterized in that, It comprises the following steps: S1: end node scene acquisition; S2: multi-node computing power elastic scheduling judgment; S3: multi-node elastic cognitive calculation; S4: multi-node elastic cognitive verification; S5: meta-dynamic multi-dimensional decision; S5a: active cognitive interaction judgment and generation: when the meta-dynamic decision kernel determines that current decision confidence does not reach a threshold, or needs to supplement information to achieve a more optimal decision, active inquiry or cooperative guidance instructions are generated; S6: end node scene interaction; (this step comprises executing the decision instructions generated in S5 and the active interaction instructions generated in S5a) S7: multi-node feedback iterative optimization.

10. The method of claim 9, wherein, In step S7, the system collects feedback data of the user on the active interaction in step S6, and updates relevant model parameters of the scene cognitive memory cluster and the multi-modal somatosensory signal fusion layer according to the feedback data, to realize individualized closed loop optimization.

Images