An internet information management system and method based on fusion of cloud edge coordination

CN122802367APending Publication Date: 2026-09-22CHONGQING FURY MASON INFORMATION TECHNOLOGY CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610975054.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-01
Publication Date
2026-09-22

AI Technical Summary

Technical Problem

[0005]本发明的目的在于提供一种融合云边协同的互联网信息化管理系统及方法,解决了上述背景中前置网关用固定数据处理规则,流量激增易硬件过载,中心无法动态干预,易崩溃丢包影响全局调度的问题

Benefits of technology

1.针对现有技术中前置采集网关处理规则固定、与硬件状态脱节的问题,本发明通过采集边缘节点处理器温度、内存占用率等硬件参数,动态计算处理阈值并调整数据处理模型激活层数,实现数据处理强度与边缘节点硬件承载能力的精准适配,有效避免并发洪峰时处理器过热、内存溢出导致的设备崩溃和数据丢包,保障边缘节点稳定运行;

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122802367A_ABST
    Figure CN122802367A_ABST
Patent Text Reader

Abstract

The application discloses an internet information management system and method based on cloud-edge cooperation, belongs to the technical field of information management, and aims to solve the problems of fixed data processing rules of a front gateway, hardware overload caused by traffic surge, incapability of dynamic intervention of a center, collapse and packet loss affecting global scheduling, and the like. The internet information management method based on cloud-edge cooperation comprises the following steps: acquiring device operation parameters of an edge node; calculating a dynamic processing threshold value; acquiring a current network bandwidth of the edge node; acquiring internet management data through a data acquisition interface of the edge node at a current acquisition frequency, and performing time sequence segmentation on the internet management data according to a segmentation length, so as to obtain a plurality of data subsegments; determining a target activation layer number of a preset data processing model according to the dynamic processing threshold value; and inputting the data subsegments into the data processing model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of information management technology, specifically to an internet-based information management system and method that integrates cloud-edge collaboration. Background Technology

[0002] Internet-based information management systems typically involve widely distributed data acquisition devices and central processing centers. When these technologies are applied, they often require uploading various logs, access traffic, and status parameters collected at the edge to the center for centralized storage and analysis. In the execution logic of these technologies, fixed upload rules are usually set at the front-end acquisition gateway, or the data is packaged indiscriminately and transmitted unidirectionally to the cloud via network links. To reduce bandwidth pressure, some technologies deploy fixed feature extraction rules at the front-end acquisition gateway to uniformly format or extract features from all incoming data.

[0003] However, the aforementioned mechanisms of these technologies have significant hidden weaknesses. When faced with peak internet concurrency or abnormal access leading to a surge in data flow, the front-end data acquisition gateway, while executing fixed high-load feature extraction rules, can cause a sharp increase in the temperature of the underlying processor and memory overflow. Because the central terminal only passively receives data, it cannot perceive the limits of the underlying physical parameters of the front-end device in real time, nor can it dynamically intervene in the data processing logic. This makes the front-end device highly susceptible to crashing or experiencing severe data loss under full load, which in turn prevents the central terminal from making correct global scheduling and responses due to the lack of data at critical time points.

[0004] To address the above issues, a cloud-edge collaborative internet information management system and method are proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an integrated cloud-edge collaborative Internet information management system and method, which solves the problems in the above-mentioned background where the front-end gateway uses fixed data processing rules, which can easily lead to hardware overload when traffic surges, the inability of the central control to intervene dynamically, and the risk of crashes and packet loss affecting global scheduling.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A cloud-edge collaborative approach to internet-based information management includes: S101: Obtain the device operating parameters of the edge node, including processor temperature and memory usage; S102: Calculate the dynamic processing threshold based on the processor temperature and the memory usage rate; S103: Obtain the current network bandwidth of the edge node, and calculate the segment length based on the current network bandwidth; S104: Obtain Internet management data at the current acquisition frequency through the data acquisition interface of the edge node, and perform time series segmentation on the Internet management data according to the segment length to obtain multiple data sub-segments; S105: Determine the target activation layer of the preset data processing model based on the dynamic processing threshold; S106: Input the data segment into the data processing model, and control the data processing model to execute up to the target activation layer to obtain the local feature vector; S107: The local feature vector, the corresponding timestamp, and the target activation layer number are sent to the cloud server so that the cloud server generates a global scheduling instruction based on the local feature vector and updates the current acquisition frequency according to the global scheduling instruction.

[0007] Further, the calculation of the dynamic processing threshold based on the processor temperature and the memory usage rate includes: Obtain the first weight parameter corresponding to the processor temperature and the second weight parameter corresponding to the memory usage rate; The processor temperature is weighted based on the first weighting parameter to obtain the temperature influence factor; The memory occupancy rate is weighted and calculated based on the second weighting parameter to obtain the memory impact factor; The dynamic processing threshold is calculated based on the temperature influence factor and the memory influence factor.

[0008] Furthermore, the cloud server generates a global scheduling instruction based on the local feature vector, including: Obtain the set of historical feature vectors pre-stored in the cloud server; The local feature vector is compared with each historical feature vector in the set of historical feature vectors to obtain the deviation value; If the deviation value is greater than a preset deviation threshold, a global scheduling instruction is generated. The global scheduling instruction includes a warning indicator and a frequency adjustment step size.

[0009] Further, updating the current acquisition frequency according to the global scheduling instruction includes: Receive the global scheduling instruction sent by the cloud server, and parse the global scheduling instruction to obtain the frequency adjustment step size; Based on the frequency adjustment step size, the current acquisition frequency of the data acquisition interface of the edge node is calculated and updated to obtain the updated acquisition frequency; The edge node's communication module periodically sends heartbeat signals to the cloud server. If no heartbeat response signal is received from the cloud server within a preset time period, the edge node is determined to enter offline mode; In offline mode, the input of the data segment to the data processing model is stopped, and the Internet management data is hashed to obtain a data fingerprint; The data fingerprint and the timestamp are stored in the local emergency cache of the edge node; Before calculating and updating the current acquisition frequency of the data acquisition interface of the edge node based on the frequency adjustment step size, the current remaining power of the edge node is obtained; If the current remaining power is lower than a preset safe power threshold and the frequency adjustment step size is an instruction to increase the sampling frequency, the global scheduling instruction is intercepted. A power level warning signal is generated and fed back to the cloud server, while maintaining the current sampling frequency unchanged.

[0010] Furthermore, it also includes: Upon receiving a heartbeat response signal from the cloud server, the edge node is determined to exit offline mode and enter online mode; The data fingerprint and timestamp in the local emergency cache are sent to the cloud server so that the cloud server can perform synchronous verification based on the data fingerprint.

[0011] Furthermore, the cloud server performs synchronous verification based on the data fingerprint, including: Obtain the cloud fingerprint stored internally on the cloud server at the corresponding timestamp; Compare the data fingerprint with the cloud fingerprint; If the data fingerprint is inconsistent with the cloud fingerprint, a data retransmission instruction for a specified time period is sent to the edge node so that the edge node can re-upload the Internet management data that was in offline mode.

[0012] Furthermore, the data processing model includes multiple processing sub-layers connected in series, and each processing sub-layer is provided with a bypass output interface; The process of controlling the data processing model to execute up to the target activation layer to obtain local feature vectors includes: Starting from the first processing sub-layer, the data segments are sequentially input into the current processing sub-layer to obtain the feature vector of the current layer; Determine whether the number of processed sub-layers that have been executed has reached the target number of activated layers; If the number of executed sub-layers reaches the target number of activated layers, the current layer feature vector is output as the local feature vector through the bypass output interface corresponding to the current sub-layer, and the execution of subsequent sub-layers is stopped.

[0013] Further, the step of calculating the segment length based on the current network bandwidth includes: Obtain the preset baseline segment length and baseline network bandwidth; Calculate the bandwidth ratio between the current network bandwidth and the reference network bandwidth; The primary segment length is obtained by scaling the baseline segment length based on the bandwidth ratio. The current load rate of the edge node processor is obtained, and the initial segment length is corrected based on the current load rate to obtain the segment length.

[0014] Furthermore, before inputting the data segment into the data processing model, the method further includes: Obtain the number of null values ​​and the number of abnormal transition values ​​in the data segment; The noise ratio is obtained by calculating the proportion of the number of null values ​​and the number of abnormal jump values ​​in the total data volume of the data segment. If the noise ratio is less than or equal to a preset noise rejection threshold, the step of inputting the data segment into the data processing model is triggered. If the noise ratio is greater than a preset noise rejection threshold, the data segment is directly discarded and a discard log is recorded.

[0015] An internet-based information management system integrating cloud-edge collaboration, comprising: The parameter acquisition module acquires the device operating parameters of the edge node, including processor temperature and memory usage. The threshold calculation module calculates a dynamic processing threshold based on the processor temperature and the memory usage rate. The segmentation calculation module obtains the current network bandwidth of the edge node and calculates the segment length based on the current network bandwidth. The data segmentation module acquires Internet management data at the current acquisition frequency through the data acquisition interface of the edge node, and performs time-series segmentation of the Internet management data according to the segmentation length to obtain multiple data sub-segments; The layer number determination module determines the target activation layer of the preset data processing model based on the dynamic processing threshold. The local feature extraction module inputs the data segment into the data processing model and controls the data processing model to execute up to the target activation layer to obtain the local feature vector; The data upload and acquisition frequency update module sends the local feature vector, the corresponding timestamp, and the target activation layer number to the cloud server, so that the cloud server generates a global scheduling instruction based on the local feature vector and updates the current acquisition frequency according to the global scheduling instruction.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. In view of the problem that the processing rules of the front-end acquisition gateway are fixed and out of touch with the hardware status in the existing technology, the present invention collects hardware parameters such as edge node processor temperature and memory usage, dynamically calculates the processing threshold and adjusts the number of activation layers of the data processing model, so as to achieve precise matching between data processing intensity and edge node hardware carrying capacity, effectively avoid device crashes and data packet loss caused by processor overheating and memory overflow during concurrent peaks, and ensure the stable operation of edge nodes; 2. This invention realizes a two-way collaborative closed-loop management and control between cloud and edge, breaking the limitation of the cloud passively receiving data in traditional technology. Edge nodes upload local feature vectors, timestamps and model activation layers. The cloud generates a global scheduling instruction with frequency adjustment step size based on historical feature vector comparison, dynamically optimizing the edge node acquisition frequency. This ensures the integrity of Internet management data acquisition, avoids the waste of computing power and network resources caused by excessive acquisition, and improves management efficiency. 3. This invention constructs a full-process data quality control and disaster recovery mechanism, which intercepts inferior data through pre-noise verification to avoid invalid data consuming edge computing power; at the same time, it designs data fingerprint storage in offline mode, online synchronization verification and abnormal retransmission functions, which not only prevent data loss and tampering when offline, but also recover missing data through fingerprint comparison, ensuring the integrity, security and continuity of Internet information management data. 4. This invention optimizes data transmission and processing efficiency by dynamically calculating the data segment length based on current network bandwidth and CPU load, thereby achieving reasonable data stream splitting and avoiding network congestion or wasted computing power. At the same time, through the bypass output interface of the data processing model, the calculation stops at the target activation layer and outputs local feature vectors, achieving lightweight data processing and reducing network transmission pressure and computing power burden on cloud-based deep analysis. 5. It has hardware security protection and flexible adaptation capabilities, and adds a power warning and interception mechanism. When the remaining power of the edge node is insufficient, it intercepts the cloud command to increase the collection frequency to prevent the device from shutting down due to overclocking. Each core parameter can be flexibly adjusted according to the edge node hardware model and application scenario to adapt to the Internet information management needs of different scales and scenarios, and has strong versatility. Attached Figure Description

[0017] Figure 1 This is a flowchart outlining the steps of the present invention. Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: As Figure 1 As shown, the following preferred technical solutions are provided: This embodiment provides an internet-based information management method that integrates cloud-edge collaboration, including: S101, Edge Node Device Operating Parameter Acquisition: Acquire the operating parameters of the edge node devices, including processor temperature and memory usage. In some embodiments, edge nodes serve as the local data processing core in a cloud-edge collaborative architecture. Their hardware operating status directly determines data processing capabilities and device operational security. Therefore, it is crucial to prioritize collecting key parameters reflecting hardware load—processor temperature and memory utilization—to ensure that subsequent data processing flows match the hardware's capacity, providing fundamental data support for dynamic adjustment. Specifically, processor temperature characterizes the thermal load status of the edge node's core computing unit; excessively high temperatures can lead to processor throttling, computational errors, or even hardware damage. Memory utilization characterizes the current use of memory resources; excessively high utilization can cause data processing lag and memory overflow. Together, these two metrics constitute the core evaluation indicators of edge node hardware load.

[0020] Specifically, by calling the low-level hardware monitoring interfaces of the edge node operating system, such as the / sys / class / thermal / thermal_zone0 / temp interface and the free command interface of the Linux system, the current physical temperature data of the microprocessor, i.e., the CPU, is periodically extracted in degrees Celsius, with a sampling period of 100ms to ensure the real-time nature of the temperature data. At the same time, memory usage data is retrieved from the memory controller in percentage form, including the specific values ​​of total memory, used memory, and free memory. The accurate memory usage rate is calculated by using the ratio of used memory to total memory, eliminating the interference of cache memory on the usage rate and ensuring the accuracy of the data.

[0021] This step provides accurate and reliable underlying data for the calculation of subsequent dynamic processing thresholds by accurately collecting the core hardware operating parameters of edge nodes. It solves the problem of ignoring the hardware status of edge nodes and blindly processing data in existing technologies, which leads to device overheating and memory overflow. This lays the foundation for lightweight management of cloud-edge collaboration.

[0022] S102, Dynamic processing threshold calculation: The dynamic processing threshold is calculated based on the processor temperature and memory usage. Since the processor temperature and memory usage of edge nodes are changing in real time, a fixed processing threshold cannot adapt to the dynamically changing hardware load. Therefore, it is necessary to combine the two core parameters collected and generate a dynamic processing threshold through weighted calculation. This threshold is used to constrain the execution depth of the subsequent data processing model to ensure that the data processing intensity matches the hardware carrying capacity.

[0023] In some embodiments, a first weighting parameter corresponding to the processor temperature and a second weighting parameter corresponding to the memory utilization rate are obtained; the processor temperature is weighted based on the first weighting parameter to obtain a temperature influence factor; the memory utilization rate is weighted based on the second weighting parameter to obtain a memory influence factor; and a dynamic processing threshold is calculated based on the temperature influence factor and the memory influence factor.

[0024] Specifically, a pre-programmed first weight parameter (range 0.4-0.6, preferably 0.5) is read from the configuration area of ​​local non-volatile memory, such as Flash. This is combined with a second weight parameter (range 0.4-0.6, preferably 0.5). The weight parameters can be flexibly adjusted according to the edge node hardware model and application scenario to ensure that the weight allocation matches the actual hardware characteristics. The processor temperature is multiplied by the first weight parameter to obtain a temperature influence factor characterizing the thermal effect cost. For example, if the processor temperature is 65℃ and the first weight is 0.5... When the temperature influence factor is 32.5, the memory occupancy rate is multiplied by the second weight parameter to obtain the memory influence factor, which represents the cost of space stacking. For example, when the memory occupancy rate is 70% and the second weight is 0.5, the memory influence factor is 35. The temperature influence factor and the memory influence factor are arithmetically summed to output a dynamic processing threshold with clear numerical boundaries, ranging from 0 to 100. In the example above, the dynamic processing threshold is 67.5. The higher the threshold, the smaller the hardware load of the edge node and the higher the processing intensity it can withstand, and vice versa.

[0025] This step, through a refined weighted calculation process of hardware parameters, directly transforms physical dimension indicators into quantifiable and identifiable constraints at the algorithm level. This avoids subjective experience-based judgments, ensures the objectivity and accuracy of threshold settings, and solves the problem in existing technologies where fixed thresholds cannot adapt to dynamic hardware loads.

[0026] S103, Data segment length calculation: Obtain the current network bandwidth of the edge node and calculate the segment length based on the current network bandwidth; Internet management data is typically a continuous data stream. Direct transmission can lead to network congestion, data loss, and is not conducive to segmented processing at edge nodes. Therefore, it is necessary to calculate a reasonable segment length based on the current network bandwidth to achieve efficient data transmission and lightweight processing. The core influencing factor for segment length is network bandwidth, while also considering the processor load of edge nodes to ensure that segmented data can be transmitted quickly without placing excessive computational pressure on the edge nodes.

[0027] In some embodiments, a preset baseline segment length and a baseline network bandwidth are obtained; the bandwidth ratio between the current network bandwidth and the baseline network bandwidth is calculated; the baseline segment length is scaled based on the bandwidth ratio to obtain the primary segment length; the current load rate of the edge node processor is obtained, and the primary segment length is corrected based on the current load rate to obtain the segment length.

[0028] Specifically, the baseline segment length is extracted from the factory configuration parameter table and preferentially set to 1024 bytes, along with the baseline network bandwidth, which is preferentially set to 100Mbps. The baseline parameters can be adjusted according to the actual application scenario, such as data transmission priority and data type. The current network bandwidth is divided by the baseline network bandwidth to obtain the bandwidth ratio reflecting the congestion of the external transmission channel. For example, when the current network bandwidth is 50Mbps, the bandwidth ratio is 0.5. The baseline segment length is multiplied by the bandwidth ratio and scaled to form a preliminary segment length adapted to the external environment. In the above example, the preliminary segment length is 512 bytes. The processor monitoring probe is called to extract the current CPU load rate in percentage form. The current load rate is used as an attenuation coefficient. The higher the load rate, the smaller the attenuation coefficient. The preliminary segment length is multiplied and corrected for compensation. For example, when the CPU load rate is 60%, the attenuation coefficient is 0.8, and the final segment length is 512 × 0.8 = 409.6 bytes, rounded to 410 bytes, ensuring that the segment length adapts to the dual requirements of network bandwidth and CPU load.

[0029] This step rigorously connects and verifies external network transmission constraints with internal computing power constraints, dynamically controlling the size of data blocks sent into the processing flow each time. This effectively prevents edge devices from crashing due to a concentrated influx of long segmented data in special circumstances where bandwidth is abundant but computing power is scarce. At the same time, it avoids the problem of low transmission efficiency caused by short segmented transmission when bandwidth is insufficient.

[0030] S104, Internet management data collection and segmentation: Internet management data is acquired through the data collection interface of the edge node at the current collection frequency, and the Internet management data is segmented into time series segments according to the segment length to obtain multiple data sub-segments; Internet management data includes, but is not limited to, network traffic data, device access logs, and data transmission status data. This data needs to be acquired in real time through the data acquisition interface of the edge node. At the same time, it is segmented into time series according to the calculated segment length, breaking the continuous data stream into multiple independent data segments to facilitate subsequent batch processing, reduce the computing power consumption of single data processing, and improve processing efficiency. The current acquisition frequency is an initial preset value, which can be initially configured by the cloud server and subsequently updated through global scheduling commands. It is initially set to 100 times per minute to ensure the real-time performance and integrity of data acquisition.

[0031] In some embodiments, the number of null values ​​and the number of abnormal transition values ​​in the data segment are obtained; the proportion of the number of null values ​​and the number of abnormal transition values ​​in the total data volume of the data segment is calculated to obtain the noise ratio; if the noise ratio is less than or equal to a preset noise rejection threshold, the step of inputting the data segment into the data processing model is triggered; if the noise ratio is greater than the preset noise rejection threshold, the data segment is directly discarded and a discard log is recorded.

[0032] Specifically, before the data is officially sent to the model calculation area, each row of field records in the data segment is traversed; the number of missing key identifiers, such as timestamps and data type identifiers, is identified and accumulated, as well as the number of abnormal jump values ​​that exceed the predetermined physical extreme value boundary, such as network traffic data suddenly jumping from 100Mbps to 10000Mbps, exceeding the normal range; the two are added together and divided by the total packet length of the current data segment to calculate the exact noise ratio; the noise ratio is sent to the comparator and compared with the noise rejection threshold in the configuration area, which is initially set to 5%; if the threshold boundary is not exceeded, the subsequent bus is opened to allow the data to be successfully imported into the model layer; if the boundary is exceeded, an interrupt clear instruction is issued through the microprocessor to directly release the random access storage space (RAM) occupied by the data segment, and an exception handling log is appended to the file system to record the discard time, data segment number, noise ratio, and other information for subsequent investigation.

[0033] This step constructs a lightweight data noise reduction interception wall at the model input end. By calculating the proportion, it cuts off the flow path of invalid or polluting data from the physical bottom layer, avoiding the needless consumption of the extremely limited computing resources of edge nodes by inferior data, while ensuring the accuracy of subsequent data processing.

[0034] S105, Determining the target activation layer of the data processing model: Determine the preset target activation layer of the data processing model based on the dynamic processing threshold; The preset data processing model is a serial, multi-layered processing structure, such as a convolutional neural network or a lightweight feature extraction model. The computational complexity of each processing sub-layer varies. The more activation layers there are, the more refined the data processing becomes, but the greater the computational cost. Therefore, it is necessary to dynamically determine the target number of activation layers of the model based on a dynamic processing threshold to achieve a balance between data processing accuracy and edge node computing power, avoiding wasted computing power or insufficient processing.

[0035] Specifically, the preset data processing model contains 3 to 5 processing sub-layers, with 4 layers being the preferred setting. Each layer corresponds to a different dynamic processing threshold range: when the dynamic processing threshold is ≥80, the target activation layer is 4 layers, the hardware load is low, and deep processing can be performed; when 60≤dynamic processing threshold<80, the target activation layer is 3 layers; when 40≤dynamic processing threshold<60, the target activation layer is 2 layers; when the dynamic processing threshold<40, the target activation layer is 1 layer, the hardware load is high, and only basic feature extraction is performed. The threshold range can be flexibly adjusted according to the model complexity and hardware performance.

[0036] Specifically, the dynamic processing threshold calculated by S102 is input into the layer matching module, and the corresponding target activation layer is quickly matched through the preset threshold layer mapping table; for example, when the dynamic processing threshold is 67.5, the target activation layer is matched as 3 layers; at the same time, the target activation layer is stored in the local cache for easy subsequent model control and cloud feedback.

[0037] This step achieves precise matching between computing power requirements and edge node hardware capacity by dynamically adjusting the number of activation layers in the data processing model. It solves the problems of fixed model processing depth and disconnect between computing power consumption and hardware status in existing technologies, and balances data processing accuracy and device operation stability.

[0038] S106, Local Feature Vector Extraction: Input the data segment into the data processing model and control the data processing model to execute to the target activation layer to obtain the local feature vector; Local feature vectors are a condensed form of core features of Internet management data. Through partial activation layer operations of the data processing model, key features of data segments are extracted, which reduces data redundancy and retains core information, making it easier to transmit to the cloud server for global analysis. This enables lightweight processing of cloud-edge collaboration—edge nodes are responsible for local feature extraction, while the cloud is responsible for global scheduling and in-depth analysis, reducing cloud computing power pressure and network transmission pressure.

[0039] In some embodiments, the data processing model includes multiple processing sub-layers connected in series, each processing sub-layer being provided with a bypass output interface; starting from the first processing sub-layer, data segments are sequentially input into the current processing sub-layer to obtain the current layer feature vector; it is determined whether the number of executed processing sub-layers has reached the target activation layer number; if the number of executed processing sub-layers has reached the target activation layer number, the current layer feature vector is output as a local feature vector through the bypass output interface corresponding to the current processing sub-layer, and the execution of subsequent processing sub-layers is stopped.

[0040] Specifically, a concatenated network comprising four matrix convolutional structures, namely processing sublayers 1 to 4, is constructed. The functions of each layer are as follows: data preprocessing, basic feature extraction, feature optimization, and deep feature fusion. In terms of hardware or logical flow architecture, each concatenated matrix processing sublayer is equipped with an independent bypass data bus interface for pre-outputting feature vectors to avoid computation in subsequent useless layers. The data segments that pass verification in S104 are imported into the first processing sublayer for matrix operations, completing data denoising and normalization, and outputting the first layer feature vector. After each layer of operation is completed, the count value of the step register is incremented by one, and the count value of the step register is compared with the target activation layer number in real time. When the count value is equal to the target activation layer number, a control switch is triggered to forcibly strip the feature data of the current matrix layer to the bypass bus, define it as a local feature vector, and push it into the transmission queue. At the same time, instructions to force sleep or block the clock source are sent to the remaining processing sublayers that are still at the back of the queue to stop their operations and reduce computing power consumption.

[0041] This step enables the abstract algorithm truncation mechanism to have practical underlying interface-level data flow support. Through physical-level dynamic dimensionality reduction and bypass output, the execution depth of the model is strictly and strongly bound to the physical indicators of the hardware, realizing a close coordination and adaptation between the algorithm's computing power requirements and the underlying carrying capacity.

[0042] S107, Data Upload and Acquisition Frequency Update: Send the local feature vector, the corresponding timestamp, and the target activation layer to the cloud server so that the cloud server can generate a global scheduling instruction based on the local feature vector and update the current acquisition frequency according to the global scheduling instruction; The local feature vectors extracted from edge nodes need to be uploaded to the cloud server. The cloud server combines the local feature vectors of all edge nodes for global analysis, generates global scheduling instructions, and feeds them back to the edge nodes to adjust the collection frequency, realizing closed-loop control of cloud-edge collaboration. Through global coordination in the cloud, the collection frequency of edge nodes is dynamically optimized, ensuring the integrity of data collection while avoiding the waste of computing power and network resources caused by over-collection. The timestamp is used to identify the collection time of data segments, which facilitates time-series analysis in the cloud. The target activation layer is used by the cloud to determine the hardware load status of edge nodes and assists in the generation of global scheduling instructions.

[0043] In some embodiments, a set of historical feature vectors pre-stored in a cloud server is obtained; the local feature vectors are compared with each historical feature vector in the set of historical feature vectors to obtain a deviation value; if the deviation value is greater than a preset deviation threshold, a global scheduling instruction is generated, which includes a warning indicator and a frequency adjustment step size.

[0044] Specifically, in the distributed relational database of the cloud server, the system retrieves a set of historical feature vectors corresponding to the current edge node's time period, such as the same time period, date, or business context, for example, local feature vectors from the same time period over the past 7 days. It then calls the cosine similarity calculation module to measure the spatial distance between the returned local feature vectors and each historical vector in the set. Based on this distance, it calculates a deviation value representing the degree of data mutation, ranging from 0 to 1. The larger the deviation, the more significant the data mutation. The deviation value is then compared with the deviation threshold residing in memory, initially set to 0.3 for logical verification. If an over-limit judgment is triggered (deviation > 0.3), the system combines the difference in deviation value with a mapping table to calculate the frequency adjustment step size for intervening in the action cycle of the underlying collector of the edge node. For example, a deviation of 0.4 corresponds to an adjustment step size of +20 times per minute, and a deviation of 0.2 corresponds to an adjustment step size of -10 times per minute. This frequency adjustment step size, along with a warning flag, is used to indicate that there is an anomaly in the edge node data. This is encapsulated in a specific control protocol message, such as the MQTT protocol, to form a global scheduling instruction and sent to the corresponding edge node.

[0045] In some embodiments, before calculating and updating the current acquisition frequency of the data acquisition interface of the edge node based on the frequency adjustment step size, the current remaining power of the edge node is obtained; if the current remaining power is lower than the preset safe power threshold and the frequency adjustment step size is an instruction to increase the acquisition frequency, the global scheduling instruction is intercepted; a power warning signal is generated and fed back to the cloud server, and the current acquisition frequency is kept unchanged.

[0046] Specifically, within the time slot after the edge communication module receives the data packet sent from the cloud but before performing a write operation to the acquisition interface register, it calls the battery management system (BMS) interface through the internal serial bus; reads the current remaining power percentage returned by the battery module, such as the remaining power of the lithium battery; serially compares this power percentage with the factory-set safe power threshold, preferentially setting it to 20%, and simultaneously analyzes the positive and negative signs of the frequency adjustment step size data, with positive numbers increasing the acquisition frequency and negative numbers decreasing the acquisition frequency; if it is detected that the power has triggered a low power warning (i.e., the remaining power is <20%) and the instruction intent is to increase the peripheral operating frequency to increase the acquisition density, the local microprocessor triggers a forced reload mechanism; clears the global scheduling instruction packet temporarily stored in the cache, blocking the write operation to the register; subsequently, it packages an alarm frame containing the underlying physical power status and pushes it back to the cloud server, maintaining the original underlying sampling clock unchanged to avoid rapid power consumption causing device shutdown.

[0047] In some embodiments, a global scheduling instruction issued by a cloud server is received and parsed to obtain a frequency adjustment step size; based on the frequency adjustment step size, the current acquisition frequency of the data acquisition interface of the edge node is calculated and updated to obtain the updated acquisition frequency.

[0048] Specifically, the process receives global scheduling instructions in the form of data packets through a socket listening process; without triggering the local power interception mechanism, it unpacks and extracts the frequency adjustment step size from the payload; it performs algebraic addition and subtraction operations on the current acquisition hardware's reference sampling frequency and the frequency adjustment step size, for example, if the current acquisition frequency is 100 times per minute and the adjustment step size is +20 times per minute, then the updated acquisition frequency is 120 times per minute; using the underlying device driver, the calculated updated acquisition frequency is directly overwritten into the timer control register of the analog-to-digital converter (ADC) of the data acquisition interface, completing the real-time update of the acquisition frequency and ensuring that subsequent data acquisition adapts to the global scheduling requirements of the cloud.

[0049] In some embodiments, the edge node's communication module periodically sends heartbeat signals to the cloud server; if no heartbeat response signal is received from the cloud server within a preset time period, the edge node is determined to enter offline mode; in offline mode, the input of data segments to the data processing model is stopped, and the Internet management data is hashed to obtain a data fingerprint; the data fingerprint and timestamp are stored in the edge node's local emergency cache.

[0050] Specifically, the internal microcontroller (MCU) is configured with a built-in watchdog timer, which, according to a predetermined period, is set to drive the RF or wired communication module to send extremely short probe data packets (heartbeat signals) to the peer cloud server every 10 seconds. A listening socket is enabled and a timeout timer is started, with a timeout period set to 30 seconds. When the timer overflows and no heartbeat confirmation packet is captured from the cloud, the system-level network status flag is set, marking the system as offline. The input bus to the processing model is cut off, completely blocking energy-intensive matrix operations to reduce computing power and power consumption. A hardware encryption chip is invoked to perform SHA-256 hash folding mapping on the original internet management data source accumulated in the cache, extracting a fixed-length 256-bit data fingerprint. The extracted data fingerprint, along with the acquisition interrupt timestamp obtained from the system clock, is written into a non-volatile Flash local emergency buffer to ensure that the data is not lost or tampered with.

[0051] In some embodiments, upon receiving a heartbeat response signal from the cloud server, the edge node is determined to exit the offline mode and enter the online mode; the data fingerprint and timestamp in the local emergency cache are sent to the cloud server so that the cloud server can perform synchronous verification based on the data fingerprint.

[0052] Specifically, when the physical layer interrupt service routine of the communication module re-intercepts the heartbeat confirmation sequence from the cloud, it resets the network status flag, causing the device's logical state to reset back to online mode; wakes up the background data synchronization thread that is in a suspended state; locates and reads the data fingerprint and timestamp stored in Flash through the file system function pointer; repackages this part of the historical evidence data into a synchronization message and pushes it to the sending queue, and sends it to the cloud server through the restored network channel to complete the synchronization of data during the offline period.

[0053] In some embodiments, a cloud fingerprint stored internally on a cloud server at a corresponding timestamp is obtained; the data fingerprint is compared with the cloud fingerprint; if the data fingerprint and the cloud fingerprint are inconsistent, a data retransmission instruction for a specified time period is sent to the edge node so that the edge node re-uploads the Internet management data during the offline mode.

[0054] Specifically, after the cloud server takes over the synchronization message, it parses the corresponding timestamp and queries the database to find the cloud fingerprint credentials that the system estimated or has received at that time point, i.e., the fingerprint of the last synchronization before disconnection or the fingerprint estimated based on historical data; it then initiates string comparison logic and performs bit-by-bit XOR verification to determine the matching degree of the data fingerprint; once a non-zero verification result is found, it means that the local record at the edge has been tampered with or that the critical data was not effectively captured by the cloud before the network disconnection; subsequently, it automatically assembles a data retransmission instruction message containing the fault timestamp interval, such as the data from the start time of disconnection to the end time of disconnection; it then sends the message to the communication gateway, triggering the designated edge node to directly retrieve the original plaintext data from its local large-capacity storage disk and initiate a segmented secondary backhaul to ensure the integrity and consistency of the data.

[0055] This step establishes a deep control feedback channel from the cloud to the edge, realizing closed-loop management of cloud-edge collaboration. It not only solves the problem of data loss in offline scenarios, but also ensures data security through fingerprint verification. At the same time, by dynamically adjusting the collection frequency, it optimizes the allocation of computing power and network resources of edge nodes, thereby improving the overall efficiency and reliability of Internet information management.

[0056] Example 2: Figure 2 As shown, the following preferred technical solutions are provided: Based on Example 1, this embodiment also provides an Internet information management system that integrates cloud-edge collaboration, including: The parameter acquisition module acquires the device operating parameters of the edge nodes, including processor temperature and memory usage. The threshold calculation module calculates the dynamic processing threshold based on processor temperature and memory usage. The segmentation calculation module obtains the current network bandwidth of the edge node and calculates the segment length based on the current network bandwidth. The data segmentation module acquires Internet management data at the current acquisition frequency through the data acquisition interface of the edge node, and performs time-series segmentation of the Internet management data according to the segment length to obtain multiple data sub-segments; The layer determination module determines the target activation layer of the preset data processing model based on the dynamic processing threshold. The local feature extraction module inputs data segments into the data processing model and controls the data processing model to execute up to the target activation layer to obtain local feature vectors. The data upload and acquisition frequency update module sends the local feature vector, the corresponding timestamp, and the target activation layer number to the cloud server, so that the cloud server can generate a global scheduling instruction based on the local feature vector and update the current acquisition frequency according to the global scheduling instruction.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud-edge collaborative internet information management method, characterized in that, include: S101: Obtain the device operating parameters of the edge node, including processor temperature and memory usage; S102: Calculate the dynamic processing threshold based on the processor temperature and the memory usage rate; S103: Obtain the current network bandwidth of the edge node, and calculate the segment length based on the current network bandwidth; S104: Obtain Internet management data at the current acquisition frequency through the data acquisition interface of the edge node, and perform time series segmentation on the Internet management data according to the segment length to obtain multiple data sub-segments; S105: Determine the target activation layer of the preset data processing model based on the dynamic processing threshold; S106: Input the data segment into the data processing model, and control the data processing model to execute up to the target activation layer to obtain the local feature vector; S107: The local feature vector, the corresponding timestamp, and the target activation layer number are sent to the cloud server, so that the cloud server generates a global scheduling instruction based on the local feature vector and updates the current acquisition frequency according to the global scheduling instruction.

2. The Internet information management method integrating cloud-edge collaboration according to claim 1, characterized in that, The calculation of the dynamic processing threshold based on the processor temperature and the memory usage includes: Obtain the first weight parameter corresponding to the processor temperature and the second weight parameter corresponding to the memory usage rate; The processor temperature is weighted based on the first weighting parameter to obtain the temperature influence factor; The memory occupancy rate is weighted and calculated based on the second weighting parameter to obtain the memory impact factor; The dynamic processing threshold is calculated based on the temperature influence factor and the memory influence factor.

3. The Internet information management method integrating cloud-edge collaboration according to claim 1, characterized in that, The cloud server generates global scheduling instructions based on the local feature vectors, including: Obtain the set of historical feature vectors pre-stored in the cloud server; The local feature vector is compared with each historical feature vector in the set of historical feature vectors to obtain the deviation value; If the deviation value is greater than a preset deviation threshold, a global scheduling instruction is generated. The global scheduling instruction includes a warning indicator and a frequency adjustment step size.

4. The Internet information management method integrating cloud-edge collaboration according to claim 1, characterized in that, The step of updating the current acquisition frequency according to the global scheduling instruction includes: Receive the global scheduling instruction sent by the cloud server, and parse the global scheduling instruction to obtain the frequency adjustment step size; Based on the frequency adjustment step size, the current acquisition frequency of the data acquisition interface of the edge node is calculated and updated to obtain the updated acquisition frequency; The edge node's communication module periodically sends heartbeat signals to the cloud server. If no heartbeat response signal is received from the cloud server within a preset time period, the edge node is determined to enter offline mode; In offline mode, the input of the data segment to the data processing model is stopped, and the Internet management data is hashed to obtain a data fingerprint; The data fingerprint and the timestamp are stored in the local emergency cache of the edge node; Before calculating and updating the current acquisition frequency of the data acquisition interface of the edge node based on the frequency adjustment step size, the current remaining power of the edge node is obtained; If the current remaining power is lower than a preset safe power threshold and the frequency adjustment step size is an instruction to increase the sampling frequency, the global scheduling instruction is intercepted. A power level warning signal is generated and fed back to the cloud server, while maintaining the current sampling frequency unchanged.

5. The Internet information management method integrating cloud-edge collaboration according to claim 4, characterized in that, Also includes: Upon receiving a heartbeat response signal from the cloud server, the edge node is determined to exit offline mode and enter online mode; The data fingerprint and timestamp in the local emergency cache are sent to the cloud server so that the cloud server can perform synchronous verification based on the data fingerprint.

6. The Internet information management method integrating cloud-edge collaboration according to claim 5, characterized in that, The cloud server performs synchronous verification based on the data fingerprint, including: Obtain the cloud fingerprint stored internally on the cloud server at the corresponding timestamp; Compare the data fingerprint with the cloud fingerprint; If the data fingerprint is inconsistent with the cloud fingerprint, a data retransmission instruction for a specified time period is sent to the edge node so that the edge node can re-upload the Internet management data that was in offline mode.

7. The Internet information management method integrating cloud-edge collaboration according to claim 1, characterized in that, The data processing model includes multiple processing sub-layers connected in series, and each processing sub-layer is equipped with a bypass output interface. The process of controlling the data processing model to execute up to the target activation layer to obtain local feature vectors includes: Starting from the first processing sub-layer, the data segments are sequentially input into the current processing sub-layer to obtain the feature vector of the current layer; Determine whether the number of processed sub-layers that have been executed has reached the target number of activated layers; If the number of executed sub-layers reaches the target number of activated layers, the current layer feature vector is output as the local feature vector through the bypass output interface corresponding to the current sub-layer, and the execution of subsequent sub-layers is stopped.

8. The Internet information management method integrating cloud-edge collaboration according to claim 1, characterized in that, The calculation of the segment length based on the current network bandwidth includes: Obtain the preset baseline segment length and baseline network bandwidth; Calculate the bandwidth ratio between the current network bandwidth and the reference network bandwidth; The primary segment length is obtained by scaling the baseline segment length based on the bandwidth ratio. The current load rate of the edge node processor is obtained, and the initial segment length is corrected based on the current load rate to obtain the segment length.

9. The Internet information management method integrating cloud-edge collaboration according to claim 1, characterized in that, Before inputting the data segment into the data processing model, the method further includes: Obtain the number of null values ​​and the number of abnormal transition values ​​in the data segment; The noise ratio is obtained by calculating the proportion of the number of null values ​​and the number of abnormal jump values ​​in the total data volume of the data segment. If the noise ratio is less than or equal to a preset noise rejection threshold, the step of inputting the data segment into the data processing model is triggered. If the noise ratio is greater than a preset noise rejection threshold, the data segment is directly discarded and a discard log is recorded.

10. An internet-based information management system integrating cloud-edge collaboration, characterized in that, A method for implementing an internet-based information management system integrating cloud-edge collaboration as described in any one of claims 1-9, comprising: The parameter acquisition module acquires the device operating parameters of the edge node, including processor temperature and memory usage. The threshold calculation module calculates a dynamic processing threshold based on the processor temperature and the memory usage rate. The segmentation calculation module obtains the current network bandwidth of the edge node and calculates the segment length based on the current network bandwidth. The data segmentation module acquires Internet management data at the current acquisition frequency through the data acquisition interface of the edge node, and performs time-series segmentation of the Internet management data according to the segmentation length to obtain multiple data sub-segments; The layer number determination module determines the target activation layer of the preset data processing model based on the dynamic processing threshold. The local feature extraction module inputs the data segment into the data processing model and controls the data processing model to execute up to the target activation layer to obtain the local feature vector; The data upload and acquisition frequency update module sends the local feature vector, the corresponding timestamp, and the target activation layer number to the cloud server, so that the cloud server generates a global scheduling instruction based on the local feature vector and updates the current acquisition frequency according to the global scheduling instruction.