Cloud server operating system for remote data processing
By constructing a three-layer collaborative architecture, dynamically dividing tasks and combining IPv4 and IPv6 communication modes, the problem of balancing accuracy and efficiency in traditional architectures is solved, and efficient and secure remote data processing capabilities are achieved.
Patent Information
- Application Number
- CN202511183671.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-22
- Publication Date
- 2025-11-18
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In existing technologies, traditional architectures only have a two-layer architecture of terminal-cloud or edge-cloud. Task allocation and resource scheduling rely heavily on empirical rules, making it difficult to balance accuracy and efficiency. Traditional systems often use static bandwidth allocation, which cannot adapt to task fluctuations.
A three-tier collaborative architecture is constructed, including a core data processing module, a cloud service collaboration module, and a remote terminal interaction module. Tasks are dynamically segmented through intelligent decision-making algorithms to achieve dynamic bandwidth allocation and fine-grained data offloading. Combined with IPv4 and IPv6 communication modes, task segmentation points and resource allocation are optimized.
It achieves real-time local processing, precision cloud computing, and ease of user interaction, improving the overall average accuracy and resource utilization efficiency of the system, enhancing security and adaptability, and preventing direct attacks.
Smart Images
Figure CN120980074A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of cloud server operating systems, in particular to a cloud server operating system for remote data processing. BACKGROUND
[0002] The cloud server operating system for remote data processing is a key infrastructure in the digital era. Its development background is rooted in the limitations of traditional solutions, the evolution of cloud computing technology, the customization needs of industry scenarios, and the rigid requirements of security compliance. By integrating cloud-native, distributed, and hardware acceleration technologies, the system focuses on the whole-process optimization of remote data processing, aiming to solve the pain points of high cost, low efficiency, and security risks, and to provide a "safe, efficient, and customizable" data processing foundation for various industries, promoting the further development of digital transformation.
[0003] At present, the Chinese invention application with the application number PCT / CN2016 / 099771 discloses a cloud computing method, a cloud server and a terminal. The cloud server receives a cloud computing request sent by the terminal, and the cloud computing request carries a cloud computing keyword. The cloud server analyzes the cloud computing keyword to obtain a target cloud computing category. The cloud server calls at least one target service program to perform cloud computing according to the cloud computing keyword and the target cloud computing category, and obtains at least one cloud computing result matched with the cloud computing keyword. The cloud server returns the at least one cloud computing result to the terminal for display. The cloud computing process can be simplified, and the efficiency of cloud computing can be improved. However, the traditional architecture in the prior art only has two-layer architecture of terminal-cloud or edge-cloud, and the task allocation and resource scheduling of the traditional remote data processing system mostly rely on experience rules, such as fixed bandwidth allocation and full-amount data uploading, which makes it difficult to balance precision and efficiency. The traditional system mostly adopts "static bandwidth allocation" (such as fixed uplink bandwidth for data uploading and downlink bandwidth for instruction receiving), which cannot adapt to task fluctuations. SUMMARY
[0004] The technical problem solved by the present application is that the traditional architecture in the prior art only has two-layer architecture of terminal-cloud or edge-cloud, and the task allocation and resource scheduling of the traditional remote data processing system mostly rely on experience rules, such as fixed bandwidth allocation and full-amount data uploading, which makes it difficult to balance precision and efficiency. The traditional system mostly adopts "static bandwidth allocation" (such as fixed uplink bandwidth for data uploading and downlink bandwidth for instruction receiving), which cannot adapt to task fluctuations.
[0005] To solve the above technical problems, the present application provides the following technical solution: a cloud server operating system for remote data processing, comprising a core data processing module, a cloud service collaboration module and a remote terminal interaction module.
[0006] The core data processing module is used for obtaining raw data from front-end devices, and performing preliminary processing, feature extraction and data distribution, while receiving and executing control instructions of the cloud and remote terminal, wherein the front-end devices include unmanned aerial vehicles, sensors and industrial cameras;
[0007] The cloud service coordination module is used for security access management, intelligent task decision, complex data reasoning, model coordination evolution and historical data management operations through the cloud server with a public network fixed IP address;
[0008] The remote terminal interaction module is used for remote access to a remote display terminal, and performs task monitoring, instruction issuing and result interpretation operations, wherein the remote display terminal includes workstations, computers and tablet computers.
[0009] Preferably, the operation process of the remote data processing cloud server system based on the coordinated evolution intelligent decision includes:
[0010] Step S100: A technician initiates a data processing or monitoring request through a remote display terminal of the remote terminal interaction module, and sends the request to the cloud server of the cloud service coordination module through an HTTPS secure domain name. After receiving the request, the cloud server of the cloud service coordination module performs security verification and permission check on the request, confirms the execution operation permission of the remote display terminal sending the request, and when the execution operation permission meets the condition of the current security permission, issues an initial instruction for preparing collection to a specified core server included in the core data processing module according to the task demand included in the request;
[0011] Step S200: The specified core server collects front-end raw data in response to the instruction, and immediately detects the network environment of the specified core server itself, wherein the network environment includes IPV4 and IPV6. The network state information including the current network environment, available uplink and downlink bandwidth, current computing load of the core server, task type requested by the remote terminal and expected reasoning accuracy target is reported to the cloud server. The cloud server performs communication mode decision according to the network state reported by the specified core server and the remote terminal, wherein the communication mode decision is: if both communication parties support IPV6, start IPV6 direct connection tunnel, otherwise start IPV4 data forwarding;
[0012] Step S300: the cloud server runs an intelligent decision algorithm to maximize the average precision, analyzes the current bandwidth and task requirement information comprehensively, generates an optimal task segmentation point, data offloading strategy and transmission resource allocation scheme, issues the data offloading strategy instruction to the core server, the core server executes the received data offloading strategy instruction, performs local processing and feature extraction on the collected raw data, packages the feature stream and residual data stream to be uploaded, and uploads the data to the cloud server through the established communication link, the communication link including IPV4 forwarding and IPV6 direct connection;
[0013] Step S400: the cloud server performs collaborative reasoning on the received data, generates a high-precision analysis result, and sends the final reasoning result to the remote display terminal initiating the task through a secure link;
[0014] Step S500: when the collaborative reasoning is completed, the cloud server optimizes the model using the data of this task, and generates a lightweight model incremental update package, the model incremental update package is a model stream downloaded to the core server, the core server receives and applies the model incremental update package, and upgrades the local communication model.
[0015] Preferably, the step S100 includes establishing a secure communication link that hides the real service port to the outside and strictly controls the permission to the inside;
[0016] Sending the request to the cloud server of the cloud service collaboration module through the HTTPS secure domain name includes:
[0017] Configuring a reverse proxy service on the cloud server in the cloud service collaboration module, associating an HTTPS secure domain name with a certificate applied to the public IP address of the cloud server, setting all external communication requests to access the standard secure port corresponding to the public IP address, configuring the reverse proxy service inside, and pointing the pre-specified access path to the actual IP address and internal communication port of the core server in the core data processing module.
[0018] Preferably, when the core server and the remote display terminal are started for the first time, a connection request is initiated to the cloud server through the HTTPS secure domain name, the cloud server performs identity authentication on the requesting device through a pre-set key, when the identity authentication is passed, a unique terminal ID in the whole system is allocated to the core server and each remote display terminal respectively, and the identity information of the terminal ID is recorded, the identity information including ID name, MAC address, IP address, device type and device model, the device type including core server and remote display terminal;
[0019] The network permission controller in the cloud server generates a unique network ID according to a data processing task to be executed, and a task administrator adds the authorized core server and remote display terminal to a virtual task network with the same network ID through an access interface of the cloud server based on a remote display terminal ID.
[0020] Preferably, the step S200 comprises:
[0021] When the core server and the remote display terminal successfully access the cloud server, the network environment in which they are located is detected and reported immediately.
[0022] When at least one of the core server or any remote display terminal supports only the IPV4 protocol, an IPV4 data forwarding mode is started, and the data forwarding logic of the IPV4 data forwarding mode comprises:
[0023] The cloud server is the first sender of all data between the core server and the remote display terminal, and the cloud server transmits and distributes the data, wherein the data includes characteristic data, original data segments, and result data.
[0024] Preferably, when the core server and the remote display terminal detect and report available public IPV6 addresses respectively, the cloud server starts an IPV6 direct connection tunnel mode, and the data processing steps of the IPV6 direct connection tunnel mode comprise:
[0025] An initial connection is established based on IPV4, and the cloud server pushes the public IPV6 address information of the communication parties to each other through the initial connection, wherein the communication parties are the core server and the remote display terminal currently establishing the connection.
[0026] After the core server and the remote display terminal obtain the IPV6 address of the other party, a point-to-point encrypted data tunnel is established based on the IPV6 addresses of the communication parties, and when the encrypted data tunnel is successfully established, the high-resolution data segments and the inference results in the transmission information are transmitted through the IPV6 direct connection tunnel.
[0027] When the IPV6 address changes, each remote display terminal performs self-checking of the address at a preset period, and reports the updated IPV6 address to the cloud server, and the cloud server broadcasts the updated IPV6 address to all terminal devices in the current communication network.
[0028] Preferably, the step S300 comprises:
[0029] The intelligent decision algorithm takes maximizing the overall average precision mAP of the final output of the cloud server operating system as the core optimization target, takes network state information as input for multi-dimensional collaborative decision-making, generates the optimal task segmentation point, data offloading strategy and transmission resource allocation scheme, and the collaborative decision-making process specifically includes the following steps:
[0030] Step S301: The communication task is segmented, and the segmentation process includes: based on the computing load characteristics and data characteristics of the communication task, the current processing capacity and historical processing performance data of the core server, based on the computer local operating system, the part of the communication task processed locally on the core server and the complex reasoning part are dynamically segmented, the complex reasoning part is the part that needs to be unloaded to the cloud server for processing due to high computing complexity, poor core server processing effect or inability to complete, and the task segmentation point is dynamically determined;
[0031] Step S302: According to the processing state of the core server to the data stream, the data is offloaded and uploaded based on the fine transmission strategy;
[0032] The fine transmission strategy includes feature stream transmission and residual data stream transmission;
[0033] The feature stream transmission specifically includes:
[0034] From the original data of the communication task and the intermediate data generated by the core server local processing, the key feature data is obtained by feature extraction through a lightweight neural network model, and all key data forms a feature stream according to the timestamp, and the feature stream is transmitted to the cloud server;
[0035] The residual data stream transmission specifically includes:
[0036] The intelligent decision algorithm determines the demand for supplementing the inference accuracy of the residual data stream according to the current uplink bandwidth, the specific requirements of the communication task for inference accuracy and the bandwidth required for feature stream transmission, the specific requirements include a preset mAP threshold, and the residual data stream is a random part of the original data segment of the communication task and a data segment after preliminary processing by the core server, and the determination includes:
[0037] When the feature stream does not reach the expected accuracy, the residual data stream is compressed;
[0038] When the feature stream reaches the expected accuracy, the residual data stream is transmitted, and the quality and quantity of the transmitted residual data are determined, and the determination includes:
[0039] The ratio of the bandwidth required for transmitting the residual data stream to the current uplink bandwidth is obtained to obtain a bandwidth utilization rate, a bandwidth utilization rate threshold is set, when the current bandwidth utilization rate is less than the bandwidth utilization rate threshold, half of the residual data stream with higher quality is transmitted, the selected original data segment is compressed and encoded by using the first high quantization bit number, and the top 50% of the residual data stream is sorted and extracted for transmission;
[0040] When the current bandwidth utilization rate is greater than the bandwidth utilization rate threshold, the quantization bit number of the residual data is reduced to reduce the amount of transmission data, or the residual data stream is not transmitted;
[0041] Step S303: The uplink bandwidth is allocated to the feature stream transmission channel and the residual data stream transmission channel according to the proportion of the required transmission data, and the downlink bandwidth is used for receiving the update instruction;
[0042] Preferably, after the decision is made, the cloud server issues a generated composite decision instruction to the core server, the composite decision instruction including the determined task segmentation point, the data offloading content and the transmission bandwidth, the task segmentation point being the segmentation boundary of the data processed by the core server and the data for generating the to-be-uploaded data stream, the data offloading content including the specific content of the feature stream, the decision of transmitting the residual data stream, the original data segment included in the residual data stream, the compression algorithm used and the quantization bit number.
[0043] Preferably, the step S400 comprises:
[0044] The core server performs the local part of the computing task according to the received instruction, and prepares the feature stream and the residual data stream according to the refined transmission strategy, and sends the prepared feature stream and residual data stream to the cloud server through the established communication link, and the cloud server receives the feature stream and residual data stream, and performs deep analysis and inference calculation on the received data to obtain an inference result, and returns the inference result to the designated remote display terminal through the communication link.
[0045] Preferably, the step S500 comprises:
[0046] The cloud server analyzes the performance bottleneck of the local inference model, and generates a model incremental update package, the model incremental update package including an optimized network layer and a feature extractor;
[0047] The cloud server transmits the model incremental update package as a model stream to the core server through the downlink, and the core server receives the model stream and automatically applies the model stream to the local inference model to complete the seamless upgrade of the local inference model.
[0048] The application has the beneficial effects that the application breaks through the tradition by a three-layer synergistic architecture, constructs a three-layer architecture of core data processing (edge)-cloud service synergy (cloud)-remote terminal interaction (user side), realizes the end-edge-cloud deep integration of local processing for real-time, cloud computing for accuracy and user interaction for convenience, takes the average precision (mAP) as the core optimization target, balances the local low latency and the cloud high precision through dynamic task segmentation, transmits the refined data unloading feature flow+residual flow on demand, and allocates resources according to the task priority / data type through dynamic bandwidth allocation, to realize the global optimization of data processing. The real address of the core server is hidden through the reverse proxy to prevent direct attacks; through the synergistic innovation of architecture, algorithm, security and model, the system realizes the remote data processing capability with high accuracy, high security and high efficiency, and breaks through the limitations of traditional schemes in resource utilization, accuracy and efficiency balance, security protection and adaptability. BRIEF DESCRIPTION OF DRAWINGS
[0049] Figure 1 A basic flow diagram of a cloud server operating system for remote data processing is provided for an embodiment of the application. DETAILED DESCRIPTION
[0050] To make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments.
[0051] REFERENCE Figure 1 For an embodiment of the application, a cloud server operating system for remote data processing is provided, including a core data processing module, a cloud service synergy module and a remote terminal interaction module:
[0052] The core data processing module is used to obtain raw data from the front-end device, and to perform preliminary processing, feature extraction and data distribution, while receiving and executing control instructions from the cloud and the remote terminal. The front-end device includes a drone, a sensor and an industrial camera.
[0053] The cloud service synergy module is used to perform security access management, intelligent task decision, complex data reasoning, model collaborative evolution and historical data management operations through the cloud server with a public network fixed IP address.
[0054] The remote terminal interaction module is used to remotely access the remote display terminal, perform task monitoring, instruction issuing and result interpretation operations, and the remote display terminal includes a workstation, a computer and a tablet computer.
[0055] The system aims to provide a remote data processing cloud server operating system with high accuracy, high security and high efficiency. It realizes the safe access, intelligent processing and dynamic optimization of remote data by building a three-layer architecture composed of a core data processing module, a cloud service collaboration module and a remote terminal interaction module. In particular, the system uses intelligent decision algorithms to optimize inference accuracy by maximizing the mean average precision (mAP) index of the system, and combines with adaptive switching of communication mode according to network environment to realize efficient use of resources.
[0056] The operation process of the remote data processing cloud server operating system based on collaborative evolution intelligent decision includes:
[0057] The system includes five core stages: system initialization and secure access, network environment adaptive communication, intelligent task processing based on maximum average precision, collaborative reasoning and result return, and model collaborative evolution update.
[0058] Step S100: The technician initiates a request for data processing or monitoring through the remote display terminal of the remote terminal interaction module, sends the request to the cloud server of the cloud service collaboration module through the HTTPS secure domain name, and the cloud server of the cloud service collaboration module receives the request, performs security verification and permission check, confirms the execution operation permission of the remote display terminal sending the request, and when the execution operation permission meets the current security permission condition, issues the initial instruction for preparation collection to the specified core server included in the core data processing module according to the task demand included in the request;
[0059] Step S200: The specified core server collects the front-end original data in response to the instruction, and immediately detects the network environment of the specified core server itself, including IPV4 and IPV6, reports the network status information to the cloud server, including the current network environment, available uplink and downlink bandwidth, current computing load of the core server, task type requested by the remote terminal and expected inference accuracy target, the cloud server makes communication mode decision according to the network status reported by the specified core server and the remote terminal, the communication mode decision is: if both communication parties support IPV6, start IPV6 direct connection tunnel, otherwise start IPV4 data forwarding;
[0060] Step S300: The cloud server runs an intelligent decision algorithm to maximize average precision, analyzes the current bandwidth and task requirement information, and generates an optimal task segmentation point, data offloading strategy and transmission resource allocation scheme, i.e., decides which feature data and how much residual data to upload, and issues the data offloading strategy instruction to the core server. The core server executes the received data offloading strategy instruction, performs local processing and feature extraction on the collected raw data, packages the feature stream and residual data stream to be uploaded, and uploads the data to the cloud server through the established communication link, including IPV4 forwarding and IPV6 direct connection.
[0061] Step S400: The cloud server performs collaborative reasoning on the received data, generates high-precision analysis results, and sends the final reasoning results to the remote display terminal that initiated the task through a secure link for technicians to view and interpret.
[0062] Step S500: When the collaborative reasoning is completed, the cloud server optimizes the model using the data of this task and generates a lightweight model incremental update package. The model incremental update package is a model stream that is transmitted downward to the core server. The core server receives and applies the model incremental update package to upgrade the local communication model. This enables it to perform local computation more efficiently in the next operation cycle, thereby forming a continuously optimized intelligent closed loop.
[0063] Step S100 includes establishing a secure communication link that hides the real service port from the outside and strictly controls the permissions of the inside.
[0064] Sending the request to the cloud server of the cloud service collaboration module through the HTTPS secure domain name includes:
[0065] Configuring a reverse proxy service on the cloud server in the cloud service collaboration module, associating an HTTPS secure domain name with a certificate to the public IP address of the cloud server, setting all external communication requests to access the standard secure port corresponding to the public IP address, configuring the reverse proxy service internally to point the pre-specified access path to the actual IP address and internal communication port of the core server in the core data processing module. In this way, the real address and port of the core server are invisible to the public network, hiding the secure port and preventing direct attacks.
[0066] When the core server and the remote display terminal are started for the first time, a connection request is initiated to the cloud server through an HTTPS secure domain name, the cloud server authenticates the requesting device through a preset key, when the authentication is passed, a terminal ID unique in the whole system is allocated to the core server and each remote display terminal respectively, and the identity information of the terminal ID is recorded, the identity information includes ID name, MAC address, IP address, device type and device model, the device type includes the core server and the remote display terminal;
[0067] The network permission controller in the cloud server generates a unique network ID according to a data processing task to be executed, the task administrator adds the authorized core server and remote display terminal to a virtual task network with the same network ID based on the remote display terminal ID by accessing the management interface of the cloud server, only the devices with the same network ID are allowed to perform subsequent data interaction, thereby isolating different tasks and preventing data crosstalk.
[0068] Step S200 includes:
[0069] Step S200 aims to intelligently select the optimal communication mode according to the actual network condition, so as to reduce the delay and the cloud server resource consumption.
[0070] When the core server and the remote display terminal successfully access the cloud server, the network environment in which the core server and the remote display terminal are located is detected and reported immediately;
[0071] When at least one of the core server or any remote display terminal only supports IPV4 protocol, IPV4 data forwarding mode is started, the data forwarding logic of the IPV4 data forwarding mode includes:
[0072] All data between the core server and the remote display terminal is first sent in order to the cloud server, and is distributed through the cloud server, all data between the core server and the remote display terminal includes feature data, original data segment and result data.
[0073] When the core server and the remote display terminal respectively detect and report the available public IPV6 address, the cloud server starts IPV6 direct connection tunnel mode, the data processing steps of the IPV6 direct connection tunnel mode include:
[0074] The initial connection is established based on IPV4, the cloud server pushes the public IPV6 address information of the communication parties to each other through the initial connection, and the communication parties are the core server and the remote display terminal currently establishing the connection;
[0075] The core server and the remote display terminal establish a point-to-point encrypted data tunnel based on the IPV6 addresses of the two parties after obtaining the IPV6 addresses of each other. When the encrypted data tunnel is successfully established, the high-resolution data segments and inference results in the transmission information are transmitted through the IPV6 direct tunnel without the need of forwarding through the cloud server.
[0076] The cloud server is only responsible for transmitting lightweight control signaling, task allocation instructions and dynamically changing IPV6 address update information in this mode, thereby greatly saving the bandwidth resources of the cloud server and significantly reducing the data transmission delay.
[0077] When the IPV6 address changes, each remote display terminal performs self-checking address at a preset period and reports the updated IPV6 address to the cloud server. The cloud server broadcasts the updated IPV6 address to all terminal devices in the current communication network to maintain the effectiveness of the direct tunnel.
[0078] Step S300 includes:
[0079] Step S300 aims to maximize the overall inference accuracy of the cloud server operating system for remote data processing through intelligent decision-making, dynamic allocation of computing tasks and transmission resources. The overall inference accuracy is measured by the mean average precision (mAP);
[0080] The intelligent decision-making algorithm takes maximizing the overall average precision mAP of the final output of the cloud server operating system as the core optimization target, and uses network state information as input for multi-dimensional collaborative decision-making to generate the optimal task segmentation point, data offloading strategy and transmission resource allocation scheme. The collaborative decision-making process specifically includes the following steps:
[0081] Step S301: The communication task is segmented, and the segmentation process includes: based on the computing load characteristics and data characteristics of the communication task, the current processing capacity and historical processing performance data of the core server, the communication task is dynamically segmented into a part suitable for local processing of the core server and a complex inference part based on the local operating system of the computer. The complex inference part is the part that needs to be offloaded to the cloud server for processing due to high computing complexity, poor processing effect of the core server or inability to complete. The task segmentation point is dynamically determined to balance the low latency advantage of local processing and the high precision and strong computing power advantage of cloud processing.
[0082] Step S302: Data is offloaded and uploaded based on a refined transmission strategy according to the processing state of the core server for the data stream;
[0083] Specifically, for the complex inference part determined to be offloaded to the cloud server in step S301, the intelligent decision algorithm determines the data content and data volume that need to be offloaded to the cloud server, and cooperatively decides not to simply upload all relevant raw data or intermediate data, but to formulate and implement a refined transmission strategy according to the real-time processing state of the core server, the current network state information (especially the available uplink bandwidth), and the requirement of the task on the average accuracy of the final inference;
[0084] The refined transmission strategy includes feature stream transmission and residual data stream transmission.
[0085] The feature stream transmission specifically includes:
[0086] From the raw data of the communication task and the intermediate data generated by the local processing of the core server, key feature data is obtained through feature extraction by a lightweight neural network model, and all key data forms a feature stream according to the timestamp, and the feature stream is transmitted to the cloud server.
[0087] The key feature data extracted by the lightweight neural network model carries the most important discriminative information with the smallest data volume, and for the subsequent cloud server, it is the most compact and critical feature data for inference among all transmitted data.
[0088] The residual data stream transmission specifically includes:
[0089] The intelligent decision algorithm determines the need for supplementing the residual data stream to further improve the inference accuracy of the cloud server according to the current uplink bandwidth, the specific requirements of the communication task on the inference accuracy, and the bandwidth required for the feature stream transmission, and the specific requirements include a preset mAP threshold. The residual data stream is a random part of the raw data of the communication task and the data segment processed by the core server, and the determination includes:
[0090] When the feature stream does not reach the expected accuracy, the residual data stream is compressed, and the purpose is to provide the cloud server with detailed information or context information that the feature stream cannot fully represent.
[0091] When the feature stream reaches the expected accuracy, the residual data stream is transmitted, and the quality and quantity of the transmitted residual data are determined, including:
[0092] The ratio of the bandwidth required for transmitting the residual data stream to the current uplink bandwidth is calculated to obtain the bandwidth utilization rate, and a bandwidth utilization rate threshold is set. When the current bandwidth utilization rate is less than the bandwidth utilization rate threshold, half of the residual data stream with higher quality is transmitted, and the selected raw data segment is compressed and encoded by using the first high quantization bit number, and the top 50% of the residual data stream is extracted and sorted for transmission.
[0093] When the current bandwidth utilization is greater than the bandwidth utilization threshold, the quantization bit number of the residual data is reduced to reduce the amount of transmitted data, that is, the compression degree is increased, the resolution is reduced, or the residual data stream is not transmitted, so as to ensure that more concurrent task feature streams can be effectively processed, or to ensure the timely transmission of the current task core feature stream. The selection of the quantization bit number includes dynamically determining a discrete or continuous decision variable according to an optimization target by an intelligent decision algorithm;
[0094] Step S303: The available uplink bandwidth is allocated to the feature stream transmission channel and the residual data stream transmission channel according to the proportion of the required transmission data, and the downlink bandwidth is used for receiving update instructions.
[0095] Specifically, the intelligent decision algorithm not only determines which data to transmit and the quantization scheme of the data, but also dynamically allocates the available uplink communication bandwidth resources between the core server and the cloud server to the feature stream and the residual data stream (if the decision is to transmit) under the premise of meeting the task processing delay requirement (for example, by comprehensively evaluating the predicted transmission time and cloud processing time). This allocation is intended to ensure that the combination of data (feature stream plus possible residual data stream) finally transmitted to the cloud server can enable the large AI model carried by the cloud server to produce the highest overall average precision mAP. In some embodiments, if the local model of the core server also participates in collaborative inference or needs to be updated by the cloud to improve its processing capability (and thus affect the offloading decision and overall mAP), part of the downlink bandwidth is also reserved or dynamically allocated for receiving update data from the cloud server. This update itself is one of the means to improve the overall mAP of the system.
[0096] When the decision is completed, the cloud server issues the generated composite decision instruction to the core server. The composite decision instruction includes the determined task segmentation point, the data offloading content, and the transmission bandwidth. The task segmentation point is the segmentation boundary between the data processed by the core server and the data used to generate the data stream to be uploaded. The data offloading content includes the specific content of the feature stream, the decision to transmit the residual data stream, the original data segments included in the residual data stream, the compression algorithm used, and the quantization bit number.
[0097] The combination of the transmitted data enables the large model of the cloud server to produce the highest mAP.
[0098] Step S400 includes:
[0099] The core server performs a calculation task of a local part according to the received instruction, and prepares a feature stream and a residual data stream according to a refined transmission strategy, and sends the prepared feature stream and residual data stream to the cloud server through the established completed communication link, the cloud server receives the feature stream and residual data stream, and performs deep analysis and inference calculation on the received data to obtain a high-precision inference result, and returns the inference result to the designated remote display terminal through the communication link for a technical personnel to perform monitoring, interpretation and subsequent operation.
[0100] The step S500 comprises:
[0101] The cloud server analyzes a performance bottleneck of the local inference model, and generates a targeted and lightweight model incremental update package, the model incremental update package comprising an optimized network layer and a feature extractor;
[0102] The cloud server transmits the model incremental update package as a model stream to the core server through a downlink, and the core server automatically applies the model stream to the local inference model after receiving the model stream, thereby completing seamless upgrading of the local inference model, and the upgraded local model can more efficiently and accurately complete the local processing part when performing next task segmentation, so that a more optimal segmentation point can be obtained in next intelligent decision-making, and the collaborative efficiency and the final average precision (mAP) of the entire system are further improved. In this stage, the self-optimization and continuous evolution capability of the cloud server operating system of remote data processing are realized.
[0103] The application breaks through the tradition through a three-layer collaborative architecture, constructs a three-layer architecture of core data processing (edge)-cloud service collaboration (cloud)-remote terminal interaction (user side), realizes the end-edge-cloud deep integration of real-time local processing, accurate cloud computing and convenient user interaction, takes the average precision (mAP) as a core optimization target, balances the local low latency and the cloud high precision through dynamic task segmentation, and realizes the global optimization of data processing through on-demand transmission of refined data unloading feature stream+residual stream, dynamic bandwidth allocation and resource allocation according to task priority / data type. The real address of the core server is hidden through reverse proxy to prevent direct attack, and the system realizes the remote data processing capability with high accuracy, high security and high efficiency through collaborative innovation of architecture, algorithm, security and model, and breaks through the limitations of traditional schemes in resource utilization, precision and efficiency balance, security protection and adaptability.
[0104] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (or computer- readable storage media) having computer-usable program code embodied in the medium. The medium can be any available medium or combination thereof that is accessible by a general purpose or special purpose computer. By way of example, such computer-usable storage media can include a volatile memory, such as a random access memory (RAM), a non-volatile memory, such as a read-only memory (ROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a floppy diskette, a compact disk, a tape, a flash memory, etc. The computer-usable program code can include any suitable set of instructions directly readable by a computer or executable by the computer which, when executed, implements the functions described below. The computer program can be downloaded to the computer from an external computer or to a device that is connected to the computer via a network or a communication link. Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or a plurality of blocks.
[0105] It should be noted that the above-mentioned embodiments are only used to illustrate but not to limit the technical solutions of the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.
Claims
1. A cloud server operating system for remote data processing, characterized in that, The system comprises a core data processing module, a cloud service coordination module and a remote terminal interaction module. The core data processing module is used for obtaining raw data from front-end devices, including unmanned aerial vehicles, sensors and industrial cameras, and performing preliminary processing, feature extraction and data distribution, while receiving and executing control instructions from the cloud and remote terminals. The cloud service coordination module is used for security access management, intelligent task decision-making, complex data reasoning, model collaborative evolution and historical data management through a cloud server with a public network fixed IP address. The remote terminal interaction module is used for remote access to remote display terminals, including workstations, computers and tablets, for task monitoring, instruction issuance and result interpretation.
2. The cloud server operating system for remote data processing of claim 1, wherein, The operation process of the remote data processing cloud server operating system based on collaborative evolution intelligent decision-making comprises the following steps: Step S100: A technician initiates a data processing or monitoring request through a remote display terminal of the remote terminal interaction module, and sends the request to the cloud server of the cloud service coordination module through an HTTPS secure domain name. Step S200: The specified core server responds to the instruction to collect front-end raw data, and immediately detects the network environment of the specified core server itself, including IPV4 and IPV6. Step S300: The cloud server runs an intelligent decision-making algorithm to maximize the average accuracy, analyzes the current bandwidth and task requirement information, generates the optimal task segmentation point, data offloading strategy and transmission resource allocation scheme, and issues the data offloading strategy instruction to the core server. Step S400: The cloud server performs collaborative reasoning on the received data to generate high-precision analysis results, and sends the final reasoning results to the remote display terminal that initiated the task through a secure link. Step S500: when the collaborative reasoning is completed, the cloud server optimizes the model using the data of this task, and generates a lightweight model incremental update package, which is transmitted to the core server as a model stream, the core server receives and applies the model incremental update package, and upgrades the local communication model.
3. The cloud server operating system for remote data processing of claim 2, wherein, The step S100 includes establishing a secure communication link that hides the real service port from the outside and strictly controls the permissions internally; Sending the request to the cloud server of the cloud service collaboration module through the HTTPS secure domain name includes: Configuring a reverse proxy service on the cloud server in the cloud service collaboration module, associating an HTTPS secure domain name with a certificate applied to the public IP address of the cloud server, setting all external communication requests to access the standard security port corresponding to the public IP address, and configuring the internal reverse proxy service to point the pre-specified access path to the actual IP address and internal communication port of the core server in the core data processing module.
4. The cloud server operating system for remote data processing of claim 3, wherein, When the core server and the remote display terminal are started for the first time, a connection request is initiated to the cloud server through the HTTPS secure domain name, the cloud server authenticates the request device through a pre-set key, and when the authentication is passed, a unique terminal ID is assigned to the core server and each remote display terminal in the whole system, and the identity information of the terminal ID is recorded, including ID name, MAC address, IP address, device type and device model, and the device type includes core server and remote display terminal. The network permission controller in the cloud server generates a unique network ID according to the data processing task to be executed, and the task administrator adds the authorized core server and remote display terminal to the virtual task network with the same network ID through the management interface of the cloud server based on the remote display terminal ID.
5. The cloud server operating system for remote data processing of claim 4, wherein, The step S200 includes: When the core server and the remote display terminal successfully access the cloud server, the network environment in which they are located is detected and reported immediately; When at least one of the core server or any remote display terminal only supports IPV4 protocol, start IPV4 data forwarding mode, the data forwarding logic of IPV4 data forwarding mode includes: All data between the core server and the remote display terminal is first sent in order to the cloud server, which is transferred and distributed through the cloud server, and all data between the core server and the remote display terminal includes feature data, raw data segment and result data.
6. The cloud server operating system for remote data processing of claim 5, wherein: When the core server and the remote display terminal respectively detect and report the available public IPV6 address, the cloud server starts the IPV6 direct connection tunnel mode, and the data processing steps of the IPV6 direct connection tunnel mode include: Based on IPV4, an initial connection is established, and the cloud server pushes the public IPV6 address information of the communication parties to each other through the initial connection, and the communication parties are the core server and the remote display terminal currently establishing the connection; When the core server and the remote display terminal respectively detect and report the available public IPV6 address, the cloud server starts the IPV6 direct connection tunnel mode, and the data processing steps of the IPV6 direct connection tunnel mode include: The core server and the remote display terminal establish a point-to-point encrypted data tunnel based on the IPV6 addresses of the two parties after obtaining the IPV6 addresses of each other, and when the encrypted data tunnel is successfully established, the high-resolution data segments and inference results in the transmission information are transmitted through the IPV6 direct tunnel. When the IPV6 address changes, each remote display terminal performs self-checking at a preset period, and reports the updated IPV6 address to the cloud server, and the cloud server broadcasts the updated IPV6 address to all terminal devices in the current communication network.
7. The cloud server operating system for remote data processing of claim 6, wherein, The step S300 comprises: The intelligent decision algorithm takes the maximization of the overall average precision mAP of the final output of the cloud server operating system as the core optimization target, takes the network state information as the input for multi-dimensional collaborative decision, generates the optimal task segmentation point, data offloading strategy and transmission resource allocation scheme, and the collaborative decision process specifically comprises the following steps: Step S301: The communication task is segmented, and the segmentation process comprises: based on the computing load characteristics and data characteristics of the communication task, the current processing capacity and historical processing performance data of the core server, the communication task is dynamically segmented into a part processed locally on the core server and a complex inference part based on the computer local operating system, the complex inference part is a part that needs to be offloaded to the cloud server for processing due to high computing complexity, poor processing effect of the core server or inability to complete, and the task segmentation point is dynamically determined; Step S302: Based on the processing state of the core server to the data stream, the data is offloaded and uploaded based on the fine transmission strategy; The fine transmission strategy comprises feature stream transmission and residual data stream transmission; The feature stream transmission specifically comprises: From the original data of the communication task and the intermediate data generated by the core server local processing, key feature data is obtained by feature extraction through a lightweight neural network model, all key data forms a feature stream according to the time stamp, and the feature stream is transmitted to the cloud server; The residual data stream transmission specifically comprises: The intelligent decision algorithm judges the demand for supplementing the inference accuracy of the residual data stream according to the current uplink bandwidth, the specific requirements of the communication task for inference accuracy and the bandwidth required for feature stream transmission, the specific requirements include a preset mAP threshold, the residual data stream is a random part of the original data segments of the communication task and the data segments after the preliminary processing of the core server, and the judgment comprises: When the feature stream does not reach the expected accuracy, the residual data stream is compressed; When the feature stream reaches the expected accuracy, the residual data stream is transmitted, and the quality and quantity of the transmitted residual data are decided, and the decision comprises: The bandwidth utilization rate is obtained by calculating the ratio of the bandwidth required for transmitting the residual data stream to the current uplink bandwidth, and a bandwidth utilization rate threshold is set, when the current bandwidth utilization rate is less than the bandwidth utilization rate threshold, half of the residual data stream with higher quality is transmitted, the selected original data segments are compressed and encoded by using a first high quantization bit number, and the residual data stream with a high ranking of 50% is extracted and transmitted by sorting. When the current bandwidth utilization is greater than the bandwidth utilization threshold, the quantization bit number of residual data is reduced to reduce the amount of data transmission, or the residual data stream is not transmitted. Step S303: The uplink bandwidth is allocated to the feature stream transmission channel and the residual data stream transmission channel according to the proportion of the required data transmission, and the downlink bandwidth is used for receiving the update instruction.
8. The cloud server operating system for remote data processing of claim 7, wherein, After the decision is made, the cloud server issues the generated composite decision instruction to the core server, and the composite decision instruction includes the determined task segmentation point, the data offloading content and the transmission bandwidth. The task segmentation point is the segmentation boundary of the data processed by the core server and the data for generating the to-be-uploaded data stream. The data offloading content includes the specific content of the feature stream, the decision of transmitting the residual data stream, the original data segment included in the residual data stream, the compression algorithm used and the quantization bit number.
9. The cloud server operating system for remote data processing of claim 8, wherein, The step S400 includes: The core server performs the computing task of the local part according to the received instruction, and prepares the feature stream and the residual data stream according to the refined transmission strategy. The prepared feature stream and residual data stream are sent to the cloud server through the established communication link. After receiving the feature stream and the residual data stream, the cloud server performs deep analysis and inference calculation on the received data to obtain an inference result. The inference result is returned to the designated remote display terminal through the communication link.
10. The cloud server operating system for remote data processing of claim 9, wherein, The step S500 includes: The cloud server analyzes the performance bottleneck of the local inference model and generates a model incremental update package. The model incremental update package includes an optimized network layer and a feature extractor. The cloud server transmits the model incremental update package to the core server as a model stream through the downlink. After receiving the model stream, the core server automatically applies the model stream to the local inference model to complete the seamless upgrade of the local inference model.