Intelligent service migration methods, systems, devices, and media based on distributed devices
By capturing application and physical world state information on distributed devices and optimizing data transmission using artificial intelligence models and neural processing units (NPUs), intelligent and rapid service migration is achieved, solving the problems of high migration latency and poor task continuity in existing technologies and improving user experience.
Patent Information
- Application Number
- CN202611273592.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-21
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, service migration between devices relies on static rules, which cannot accurately perceive the user's application state and the physical world state, resulting in high migration latency, poor task continuity, and high false migration rate, thus failing to meet the needs of real-time interaction.
By capturing application state information and physical world state information, using artificial intelligence models to predict migration intentions and timing, and only migrating the minimum state increment dataset necessary to maintain task continuity, combined with neural processing units (NPUs) for inference and direct memory access (DMA) technology to optimize data transmission, intelligent and fast service migration is achieved.
It significantly reduces migration latency to the millisecond level, improves migration accuracy and task continuity, reduces mismigration rate, optimizes user experience, and runs efficiently on resource-constrained embedded devices.
Smart Images

Figure CN123114656A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of distributed computing technology, and in particular to a method, system, device and medium for intelligent service migration based on distributed devices. Background Technology
[0002] With the development of smart office and educational informatization, smart terminal devices, represented by interactive whiteboards, have been widely used in scenarios such as conference rooms and classrooms. These devices usually exist in the form of distributed devices, and users may move between multiple devices and expect tasks to be seamlessly connected.
[0003] In existing technologies, service migration between devices typically relies on preset static rules. For example, the system determines the distance between the user and the device by monitoring the strength of wireless signals (such as Bluetooth or Wi-Fi), and triggers service migration when the signal strength exceeds a certain fixed threshold. This decision-making method based on a single physical quantity and a fixed threshold is relatively simple and rigid.
[0004] However, this simplistic, rule-based migration strategy ignores complex real-world scenarios and the user's true intentions. It fails to perceive the user's application state (e.g., whether they are focused on a demonstration or simply passing by the device) and cannot understand the dynamic changes in the physical world state between the user and the device. Therefore, this mechanism often leads to unnecessary migrations or fails to migrate in a timely manner when the user truly needs it, causing task interruptions and severely impacting the user experience.
[0005] Furthermore, existing general migration solutions often fail to adequately consider the resource constraints of embedded devices such as interactive flat panels. Migrating a complete application state snapshot consumes significant network bandwidth and computing resources, resulting in excessively high migration latency and failing to meet the demands of real-time interaction. Deploying complex decision-making logic on resource-constrained embedded platforms also presents challenges in terms of power consumption and performance. Therefore, there is an urgent need for a service migration solution that can intelligently perceive user intent, achieve rapid and seamless migration, and adapt to the resource limitations of embedded devices, in order to address the problems of unintelligent migration, high latency, and poor user experience. Summary of the Invention
[0006] This invention aims to address the problems in existing technologies where service migration decisions rely on static rules, cannot accurately perceive application and physical world state information, resulting in high migration latency, poor task continuity, and high false migration rates. To this end, this invention provides an intelligent service migration method, system, electronic device, and computer-readable storage medium based on distributed devices.
[0007] This invention provides an intelligent service migration method based on distributed devices, comprising: capturing application state information required to maintain task continuity on the source device; combining the application state information with physical world state information characterizing the physical proximity trend between a user and at least one target device to predict the migration intention and timing to the target device; and migrating the minimum state increment dataset necessary to maintain task continuity from the source device to the target device in response to the predicted migration intention and timing. This method, by comprehensively analyzing both application and physical world state information, can intelligently determine the migration intention and timing, and significantly improves the accuracy and efficiency of migration by migrating the minimum state increment dataset.
[0008] In one implementation, the application state information includes at least one of the following: the content of the document being edited, the cursor position, the playback progress of the media file, and the current running application state; the physical world state information includes: the user's movement trajectory and / or the gradual trend of the communication signal strength between the target device and the user. By specifying the types of state information, the system can more accurately capture the details of the user's task and the intent of physical movement, further improving the accuracy of prediction.
[0009] Furthermore, the steps of predicting migration intent and timing are performed using an artificial intelligence model deployed on the source device. By replacing traditional static rules with an AI model, the system gains the ability to learn and reason from complex, multi-dimensional information, thereby achieving more intelligent and human-like migration decisions.
[0010] Optionally, the artificial intelligence model is deployed on the neural processing unit (NPU) of the source device for inference. Offloading computationally intensive AI inference tasks to dedicated NPU hardware can significantly improve computational efficiency and reduce CPU load and power consumption, making it particularly suitable for resource-constrained embedded devices.
[0011] In one implementation, the method further includes using Direct Memory Access (DMA) technology to transfer the data required for the computation of the artificial intelligence model between processor cores. DMA technology enables "zero-copy" data transfer between different processing units, avoiding CPU intervention, reducing data transfer overhead and latency, and further improving the overall processing performance of the system.
[0012] Furthermore, it also includes: using unified memory pool management technology to uniformly allocate and reclaim intermediate data generated during the computation of the artificial intelligence model. Adopting unified memory pool management can effectively reduce memory fragmentation, improve memory utilization efficiency, avoid performance fluctuations caused by frequent memory allocation and deallocation operations, and ensure system stability.
[0013] Optionally, the source device or the target device is an interactive flat panel. This solution is particularly suitable for multi-device collaboration of interactive flat panels in scenarios such as education and meetings, solving the pain point of seamless service connection when users move between different interactive flat panels.
[0014] This invention also provides an intelligent service migration system based on distributed devices, comprising: an application state capture module for capturing application state information required to maintain task continuity on the source device; a migration information prediction module for combining the application state information with physical world state information characterizing the physical proximity trend between a user and at least one target device to predict the migration intention and timing to the target device; and an incremental data migration module for migrating the minimum incremental state dataset necessary to maintain task continuity from the source device to the target device in response to the predicted migration intention and timing. Through a modular design, this system achieves the functions corresponding to the above methods, and can be deployed as a whole in electronic devices, providing stable and reliable intelligent service migration capabilities.
[0015] The present invention also provides an electronic device including a processor coupled to a memory; the processor is configured to execute a computer program stored in the memory to cause the electronic device to perform the method as described in any of the preceding claims.
[0016] The present invention also provides a computer-readable storage medium including a computer program or instructions that, when executed on a computer, cause the computer to perform the method as described in any of the preceding claims.
[0017] In summary, this application includes at least one of the following beneficial technical effects: First, by predicting users' migration intentions and migrating only the minimum incremental state dataset necessary to maintain task continuity, rather than a complete application state snapshot, the amount of data to be migrated is greatly reduced, thereby reducing service migration latency from seconds to milliseconds and achieving near-seamless fast migration.
[0018] Secondly, this invention uses an artificial intelligence model to comprehensively analyze application state information and physical world state information, which can intelligently and accurately determine migration intentions and timing, avoiding erroneous migrations and inappropriate migrations caused by simple rules, significantly improving the continuity of tasks and reducing the erroneous migration rate, thereby optimizing the user experience in multi-device collaboration scenarios.
[0019] Furthermore, the technical solution of this invention takes into account the deployment optimization on embedded devices (such as interactive flat panels). By utilizing the Neural Processing Unit (NPU) for inference acceleration, employing Direct Memory Access (DMA) technology for data transmission, and implementing unified memory management, it achieves efficient operation with limited computing and storage resources, balancing high performance and low power consumption, and has strong practicality and scenario adaptability. Attached Figure Description
[0020] Figure 1 This is a schematic diagram of an intelligent service migration system architecture provided in an embodiment of the present invention.
[0021] Figure 2 This is a core flowchart of an intelligent service migration method provided in an embodiment of the present invention.
[0022] Figure 3 This is a schematic diagram comparing the key performance indicators of the method of the present invention with those of existing technologies.
[0023] Figure 4 This is a hardware architecture block diagram provided for an embodiment of the present invention. Detailed Implementation
[0024] The present application will be further described in detail below with reference to the accompanying drawings.
[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0026] In current distributed device applications, such as smart conference rooms or smart classrooms, task handover between multiple devices often relies on static rules, such as triggering service migration based on communication signal strength reaching a preset threshold. This approach cannot accurately perceive the user's true intent, application state information, and physical world state information, often leading to problems such as inappropriate migration timing, interrupted migration experience, or erroneous migration. This invention aims to solve this problem by proposing a context-aware intelligent service migration method. By comprehensively analyzing the task state at the application layer and the user state in the physical world, it achieves accurate prediction of migration intent and timing, and uses the minimum state increment dataset necessary for maintaining task continuity to ensure fast and seamless service transition. This solution fully considers the resource limitations of embedded platforms and, through hardware and software co-optimization, significantly reduces system overhead while ensuring low latency and high continuity.
[0027] To ensure a consistent understanding throughout the specification, the core terms involved in this invention are explained uniformly here. Distributed device refers to a set of physically separate but interconnected devices that can work collaboratively, such as multiple interactive tablets or smartphones on the same local area network. Application state information refers to application-level information required to ensure seamless user task transitions from a source device to a target device, such as the content of the document being edited, cursor position, media file playback progress, and the current running application state. Physical world state information refers to data characterizing the changing trends of the physical relationship between the user and the device, used to determine the user's spatial movement intentions, such as the user's movement trajectory or the gradual trend of communication signal strength between the device and the user. Migration intention and migration timing refer to the system's prediction based on input information whether the user wishes to migrate services and the optimal migration time. Minimal state increment dataset necessary to maintain task continuity refers to data containing only the changes in the task state since the last synchronization, rather than a complete application state snapshot, designed to achieve rapid migration. Artificial intelligence model refers to the algorithmic model used to process application state information and physical world state information, and to predict migration intentions and timing. A Neural Processing Unit (NPU) is a hardware processing unit specifically designed to accelerate artificial intelligence calculations such as those involving neural networks. Direct Memory Access (DMA) technology allows hardware subsystems to directly read and write to main memory without going through the central processing unit (CPU).
[0028] like Figure 1 As shown, this embodiment of the invention provides an intelligent service migration system based on distributed devices. The system is deployed in a distributed device environment comprising at least one source device 10 and one target device 20. In a typical application scenario, a user initially operates on the source device 10 and then physically moves to the vicinity of the target device 20. The system of this invention can intelligently sense this process and seamlessly migrate the service from the source device 10 to the target device 20. The system mainly includes three core functional modules deployed on the source device 10: an application state capture module 101, a migration information prediction module 102, and an incremental data migration module 103.
[0029] Specifically, the application state capture module 101 is responsible for real-time monitoring and capturing application state information related to the user's current task on the source device 10. This module delves into the application to obtain key data necessary to maintain task continuity. The migration information prediction module 102, as the system's decision-making core, obtains application state information from the application state capture module 101 and actively collects physical world state information representing the spatial relationship between the user and the device. By fusing these two types of heterogeneous information, this module can accurately predict the user's migration intention and the optimal migration time. The incremental data migration module 103, as the execution unit, immediately generates a minimum state increment dataset necessary to maintain task continuity based on the current application state upon receiving a migration instruction from the migration information prediction module 102, and quickly sends it to the target device 20 via the network. These three modules work together to form a complete closed loop of perception, decision-making, and execution, thereby achieving efficient and intelligent service migration.
[0030] The following is combined Figure 2 The specific process of the intelligent service migration method provided by this invention will be described in detail. For example... Figure 2 As shown, the method mainly includes the following steps: First, step S201 is executed, which involves capturing application state information and physical world state information. In this step, the system acquires two types of key context data in parallel. For capturing application state information, the system extracts specific state data based on the currently running application type. For example, when a user is editing a document using office software, the application state capture module 101 records the latest content of the document, format changes, and the specific position of the cursor and the text selection state. If the user is watching a video, it captures the media file identifier and the playback progress bar position accurate to milliseconds. This fine-grained application state information is the foundation for ensuring that the task can be restored as is on the target device, achieving seamless continuity. While capturing application state information, the system also continuously acquires physical world state information to determine the user's spatial movement intentions. Specifically, this can be achieved by analyzing the wireless communication signal strength between the user's smart device (such as a smartwatch or mobile phone) and the source device 10 and potential target device 20. The system does not rely on simple signal strength thresholds but analyzes the gradual trend of signal strength changes. For example, when the system detects a continuous decrease in the Bluetooth (BLE) or Wi-Fi signal strength (RSSI, Received Signal Strength Indicator) between the user device and the source device 10, while the signal strength between the user device and the target device 20 continues to increase, the system can determine with high confidence that the user is moving from the source device 10 to the target device 20. This trend-based analysis method can effectively filter out transient changes caused by signal fluctuations, thereby accurately identifying the user's physical approach trend and movement trajectory.
[0031] Next, step S202 is executed, which involves predicting the migration intention and timing. This step is the core of the invention and is accomplished by an artificial intelligence model deployed on the source device 10. This AI model receives the application state information and physical world state information captured in step S201 as input. By combining the context of the task (e.g., whether the user is performing an uninterrupted critical operation) with the user's physical movement intention, the model can make more accurate judgments than any single source of information. For example, even if the user is near the target device, if the application state information indicates that the user is performing an important file saving operation, the model may delay the migration timing to avoid interruption of the operation. As a specific implementation, the AI model can employ a time series analysis model (such as a Long Short-Term Memory network LSTM) to process the signal strength sequence in the physical world state information to accurately capture its gradual trend. The model's output can be a classification result indicating whether migration has occurred, and a regression result predicting the optimal migration time. By training on a large amount of real-world scenario data, the AI model can learn complex spatiotemporal correlation patterns, thereby achieving a highly intelligent prediction of migration intention and timing. This design transforms migration decisions from rigid rule-triggered processes into dynamic, adaptive intelligent behaviors, significantly improving migration accuracy and user experience.
[0032] Then, step S203 is executed, which involves migrating the minimum incremental state dataset necessary to maintain task continuity. This step is triggered when the artificial intelligence model in step S202 predicts the migration intention and determines the migration timing. Unlike traditional methods that require packaging the entire application's data and state (i.e., application snapshots) for migration, this invention employs a more efficient strategy. The incremental data migration module 103 compares the current application state with the state at the last synchronization point or when the task started, extracting only the changed parts to generate a minimum incremental state dataset necessary to maintain task continuity. For example, in a document editing scenario, this dataset might only contain newly entered text, deleted paragraphs, and formatting instructions, rather than the entire document file of several megabytes. In a video playback scenario, this dataset might only contain the video file identifier and the new playback timestamp. Because the incremental dataset is extremely small, its transmission from the source device 10 to the target device 20 over the network is very rapid. This design significantly reduces data transmission time, network bandwidth usage, and device power consumption, and is key to achieving ultra-low latency migration of less than 200 milliseconds. After receiving this incremental dataset, the target device 20 can quickly apply these changes to its existing local application copy or on top of it, thereby instantly restoring the user's task context and bringing the user a seamless experience with continuous and uninterrupted service.
[0033] To efficiently execute the above methods on resource-constrained embedded devices such as interactive flat panels, especially when running artificial intelligence models, this invention also proposes a series of targeted hardware deployment and optimization schemes. The following example uses a device equipped with a multi-core heterogeneous embedded processor 40 (SoC) and combines... Figure 4 The hardware architecture block diagram is used to illustrate this. The embedded processor 40 can be a mainstream model in the industry, such as Rockchip's RK3588, which integrates a high-performance CPU core, a powerful neural processing unit NPU 402, and other hardware modules.
[0034] In a preferred implementation, the AI model used to predict migration intent and timing is specifically deployed on the source device's Neural Processing Unit (NPU) 402 for inference. Before deployment, the model undergoes quantization, for example, from 32-bit floating-point (FP32) precision to 8-bit integer (INT8) precision. The quantized model is smaller, faster, and requires less hardware computing power. Delegating the inference task to the dedicated NPU 402 fully utilizes its parallel computing power of up to 6 TOPS (trillion operations per second), with measured utilization consistently between 75% and 85%. This design makes the AI model's inference process extremely efficient, while freeing the general-purpose CPU 401 (e.g., its high-performance A76 cores) from heavy AI computation, allowing it to focus on high-priority real-time tasks, while low-power cores (e.g., A55) handle background services, thus ensuring the overall system's smoothness and responsiveness.
[0035] Furthermore, to optimize data transfer efficiency between heterogeneous computing units, this invention also utilizes Direct Memory Access (DMA) technology. In the architecture of the embedded processor 40, data (such as raw data collected by sensors, model weight parameters, and intermediate data generated during NPU computation) needs to be frequently transferred between memory 403, CPU 401, and the neural processing unit (NPU) 402. Using the traditional CPU copy method would consume a significant number of CPU cycles and introduce unnecessary latency. This solution, by configuring a Direct Memory Access (DMA) controller 404, achieves "zero-copy" operation when transferring data between these processor cores. Data can be directly transferred from the address space of one hardware unit to another without the intervention of CPU 401. This design significantly reduces latency in the data path and is one of the key technologies to ensure that the end-to-end latency of the entire migration decision process meets the requirements of real-time interaction.
[0036] Furthermore, addressing the issue of a large amount of temporary intermediate data (i.e., tensors) generated during the computation process of artificial intelligence models, this invention also optimizes memory usage through a unified memory pool management technique. Frequently requesting and releasing small blocks of memory from the operating system to store these intermediate tensors easily leads to memory fragmentation, reduces memory usage efficiency, and can even cause system instability after prolonged operation. This solution pre-allocates a large and contiguous block of physical memory as a unified memory pool during the system initialization phase. During model inference, all memory allocation and reclamation of intermediate data are handled within this pool by a dedicated memory manager. This unified allocation and reclamation mechanism avoids frequent system call overhead and memory fragmentation, ensuring efficient and predictable memory usage. According to actual tests, after adopting this technique, the system's peak memory usage can be controlled below 30% of the total dynamic random access memory (DRAM), providing a strong guarantee for the long-term stable operation of the system.
[0037] To more comprehensively demonstrate the technical solution and beneficial effects of this invention, a complete application scenario will be used to connect all the above technical points. Imagine a "smart classroom" scenario: a teacher is using an interactive whiteboard A (source device 10) to give a lecture, displaying a PPT document on the screen. At this time, to interact with students on the other side of the classroom, the teacher walks to another interactive whiteboard B (target device 20) at the other end of the classroom while explaining. With the support of this invention, the service migration process will occur intelligently and seamlessly as follows: While the teacher is using interactive whiteboard A, its application status capture module 101 runs silently in the background, recording application status information in real time, such as the current PPT page number being page 15, and the teacher having just circled a note on the page with a red pen. Simultaneously, the system continuously collects physical world status information through the smart bracelet worn by the teacher. At this time, the BLE signal between the bracelet and interactive whiteboard A is strong, while the signal with interactive whiteboard B is weak. When the teacher begins to move, the system detects that the signal strength of the wristband with A begins to steadily decrease, while the signal strength with B steadily increases. This clear gradual trend is sent to the migration information prediction module 102.
[0038] The artificial intelligence model deployed on the Neural Processing Unit (NPU) 402 in the migration information prediction module 102 receives the aforementioned application state information and physical world state information. The model analysis concludes that the user is performing a continuous presentation task (PPT lecturing), and their movement trajectory clearly points towards the interactive whiteboard B, thus predicting the user's migration intention with a high probability. The model further determines that the optimal migration time is the instant the teacher arrives in front of the interactive whiteboard B, preparing to perform the next operation. At this precisely predicted moment, the system triggers the incremental data migration module 103. This module quickly generates a minimal incremental dataset necessary to maintain task continuity, containing only a very small amount of information such as "Application: PPT; Page Number: 15; Incremental Data: Vector coordinates of a red circle."
[0039] This incremental dataset is sent to interactive whiteboard B via a wireless network. Upon receiving the data, interactive whiteboard B immediately opens a PowerPoint application on it, jumps directly to page 15, and renders the red annotations just drawn by the teacher. The entire process, from the teacher leaving A to seeing the perfectly synchronized screen on B, takes far less time than the latency perceptible to the human eye. The teacher does not need to perform any manual operations, such as plugging or unplugging a USB drive, scanning a QR code to log in, or transferring files; they can naturally continue their lecture on interactive whiteboard B. This experience is made possible by the synergistic effect of a series of technologies, including context awareness, AI model prediction, Neural Processing Unit (NPU) acceleration, Direct Memory Access (DMA) technology, unified memory pool management, and the transfer of the minimum state incremental dataset necessary to maintain task continuity.
[0040] like Figure 3 As shown, the technical effect of the method of the present invention is significantly improved compared with the prior art in terms of key performance indicators. Figure 3 These advantages are visually illustrated through comparative graphs. Regarding service migration latency, this invention successfully controls end-to-end service migration latency to within 200 milliseconds (ms) by migrating the minimum state increment dataset necessary to maintain task continuity, providing users with a near-instantaneous switching experience. In contrast, traditional full application snapshot migration solutions, due to the need to transfer large amounts of data, typically have latency exceeding 2 seconds (s), causing noticeable lag. Regarding post-migration task continuity, because this invention accurately captures and restores the critical state of the task, its task continuity retention rate is over 95%, ensuring seamless continuity of user work. Regarding migration accuracy, intelligent decision-making based on an artificial intelligence model reduces the false migration rate (e.g., a user mistakenly triggering migration simply by passing by a device) to below 3%, far superior to the typically high false migration rate of up to 12% for static rule-based solutions. These data, verified through real-world testing, fully demonstrate the significant value of this invention in improving the service migration experience between distributed devices.
[0041] The present invention also provides an electronic device, which may be the interactive tablet mentioned in the foregoing embodiments, or any intelligent device capable of executing the method of the present invention, such as a smartphone, tablet computer, laptop computer, or vehicle central control system. The internal structure of the electronic device includes, but is not limited to, a processor and a memory coupled to the processor. The memory stores computer programs or instructions. When the processor executes the computer program stored in the memory, it causes the electronic device to execute the intelligent service migration method. The processor may be a general-purpose processor, a digital signal processor (D32SP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, etc.
[0042] The present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon. When these computer programs or instructions are executed on a computer or electronic device, they cause the computer or electronic device to perform an intelligent service migration method. The computer-readable storage medium can be any available medium, as long as it is accessible to a computer, including volatile and non-volatile media, removable media and non-removable media. For example, it can be, but is not limited to, random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), solid-state drive (SSD), optical disc (such as CD-ROM) or other optical disc storage, disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing instructions or data structures in the form of desired program code and accessible by a computer.
[0043] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing application concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions claimed in this application.
Claims
1. A method for intelligent service migration based on distributed devices, characterized in that, include: Capture application state information on the source device required to maintain task continuity; By combining the application state information with physical world state information that characterizes the physical proximity trend between the user and at least one target device, the migration intention and timing to the target device can be predicted. In response to the predicted migration intention and timing, the minimum state increment dataset necessary to maintain mission continuity is migrated from the source device to the target device.
2. The method according to claim 1, characterized in that, The application status information includes: At least one of the following: the content of the document being edited, the cursor position, the playback progress of the media file, and the status of the currently running application; The physical world state information includes: the user's movement trajectory and / or the gradual trend of the communication signal strength between the target device and the user.
3. The method according to claim 1, characterized in that, The steps of predicting migration intent and timing are performed using an artificial intelligence model deployed on the source device.
4. The method according to claim 3, characterized in that, The artificial intelligence model is deployed on the neural processing unit (NPU) of the source device for inference.
5. The method according to claim 4, characterized in that, It also includes: using Direct Memory Access (DMA) technology to transfer the data required for the computation of the artificial intelligence model between processor cores.
6. The method according to claim 4, characterized in that, Also includes: By using unified memory pool management technology, intermediate data generated during the computation of the artificial intelligence model can be uniformly allocated and recycled.
7. The method according to claim 1, characterized in that, The source device or the target device is an interactive flat panel.
8. A smart service migration system based on distributed devices, characterized in that, include: The application state capture module is used to capture application state information required on the source device to maintain task continuity; The migration information prediction module is used to combine the application state information with physical world state information that characterizes the physical proximity trend between the user and at least one target device to predict the migration intention and timing to migrate to the target device. An incremental data migration module is used to migrate the minimum state incremental dataset necessary to maintain mission continuity from the source device to the target device in response to the predicted migration intention and migration timing.
9. An electronic device, characterized in that, Includes a processor, which is coupled to a memory; The processor is configured to execute a computer program stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It includes a computer program or instructions that, when run on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.
