Search

CN-121397083-B - Edge router quick acquisition decision method and system

CN121397083BCN 121397083 BCN121397083 BCN 121397083BCN-121397083-B

Abstract

The invention relates to the technical field of the Internet of things and discloses a quick acquisition decision method of an edge router, which comprises the following steps that step 1, a pre-training model is loaded, the pre-training model is obtained by constructing a meta model by a cloud server based on a meta learning algorithm and training the meta model by adopting historical task data; the method comprises the steps of step 2, collecting a data sample of target equipment, carrying out iterative adjustment on a pre-training model by adopting the data sample to obtain a collection strategy model, and outputting a collection strategy mapping table by the collection strategy model, step 3, obtaining the current running state of the target equipment, mapping the current running state information to the collection strategy mapping table, screening out the optimal collection strategy, collecting sensor data of the target equipment according to the optimal collection strategy to obtain target data, and step 4, constructing a quantization filtering rule according to the collection strategy mapping table, and filtering the target data through the quantization filtering rule to obtain reported data. Meanwhile, the invention also discloses a fast mining decision system of the edge router.

Inventors

  • HE KUI
  • CAO LIANFENG
  • WAN YANPENG

Assignees

  • 广州鲁邦通物联网科技股份有限公司

Dates

Publication Date
20260512
Application Date
20251225

Claims (9)

  1. 1. The edge router quick acquisition decision method is characterized by comprising the following steps: step 1, loading a pre-training model, wherein the pre-training model is obtained by constructing a meta model by a cloud server based on a meta learning algorithm and training the meta model by adopting historical task data of various devices; Step 2, acquiring a data sample of target equipment, and performing iterative adjustment on a pre-training model by adopting the data sample to obtain an acquisition strategy model, wherein the acquisition strategy model outputs an acquisition strategy mapping table which comprises acquisition objects, acquisition frequencies and thresholds corresponding to various running states; Step 3, acquiring the current running state of the target equipment, mapping the current running state information to an acquisition strategy mapping table, screening out the optimal acquisition strategy, and acquiring sensor data of the target equipment according to the optimal acquisition strategy to obtain target data; and 4, constructing a quantization filtering rule according to the acquisition strategy mapping table, and filtering target data through the quantization filtering rule to obtain reported data.
  2. 2. The method for quickly acquiring decision-making by the edge router according to claim 1, wherein the historical task data comprise historical running states and corresponding acquired data, and the training process of the pre-training model in step 1 is that the cloud server builds a meta model based on a meta learning algorithm, trains the meta model by taking the historical running states of various devices and the corresponding acquired data as a training set, and optimizes parameters of the meta model by minimizing a strategy utility loss function, thereby obtaining the pre-training model.
  3. 3. The edge router fast mining decision method of claim 2, wherein the formula of the minimization of the policy utility loss function is: ; Wherein, the Is that Is used for the weight of the (c), To the extent to which the collected data contributes to the cloud fault prediction model, Is that Is used for the weight of the (c), The bandwidth consumed for collecting the data, the CPU and the weighted sum of the storage space.
  4. 4. The edge router fast mining decision method according to claim 1, characterized in that step 2 comprises the sub-steps of: A1, collecting a data sample of target equipment, wherein the data sample is a corresponding collecting object and collecting frequency under various running states obtained by running the target equipment for N times, and N is a positive integer; a2, inputting the data sample into a pre-training model, and performing iterative adjustment on the pre-training model by a gradient descent method to obtain an acquisition strategy model; A3, generating an acquisition strategy mapping table by the acquisition strategy model according to various running states of the target equipment, wherein the acquisition strategy mapping table comprises acquisition objects, acquisition frequencies and thresholds corresponding to the various running states; And step A4, storing the collection strategy mapping table in a collection strategy library.
  5. 5. The edge router fast mining decision method according to claim 1, wherein in the step 3, the current operation state of the target device is obtained by: The first mode is that a control signal of a controller of target equipment is obtained, and the current running state of the target equipment is obtained through analysis from the control signal; And secondly, when the control signal of the controller cannot be acquired, acquiring the sensor data of the target equipment, and reversely pushing out the current running state of the target equipment from the sensor data.
  6. 6. The rapid acquisition decision method of the edge router according to claim 1 is characterized in that the step 4 specifically comprises the steps of obtaining an acquisition object and a corresponding threshold value in an acquisition strategy mapping table, wherein the acquisition object is various sensors, constructing a quantization filtering rule based on the acquisition object and the corresponding threshold value, filtering target data by adopting the quantization filtering rule to obtain report data, and the report data is used for predicting the state of target equipment by a prediction model uploaded to a peripheral device.
  7. 7. The edge router fast mining decision method of claim 6, wherein the quantization filtering rule comprises: setting a baseline of target data when an acquisition object is a sensor for monitoring specific actions, wherein the corresponding threshold value is 105-115% of the baseline, and taking the target data as reporting data when the target data exceeds the corresponding threshold value, otherwise discarding the target data; When the acquisition object is a sensor with small fluctuation of the monitoring data, when the absolute value of the difference value between the current target data and the last reported data is larger than the corresponding threshold value, taking the target data as the reported data, otherwise discarding the target data; and thirdly, when the acquisition object is a sensor for monitoring state change, when the target data is larger than the corresponding threshold value, taking the target data as reporting data, otherwise, discarding the target data.
  8. 8. The method for fast acquisition decision-making by an edge router according to claim 1, further comprising step 5 of uploading the reported data to a peripheral predictive model to predict the state of the target device, calculating the resource cost consumed by the reported data, and adaptively updating the acquisition strategy model by a gradient descent method based on the consumed resource cost and the feedback information by the peripheral predictive model output feedback information.
  9. 9. An edge router rapid acquisition decision system, which is used for implementing the edge router rapid acquisition decision method according to any one of claims 1-8, and comprises the following modules: the model loading module is used for loading a pre-training model, wherein the pre-training model is obtained by constructing a meta model by a cloud server based on a meta learning algorithm and training the meta model by adopting historical task data of various devices; the model adaptation module is used for acquiring a data sample of the target equipment, carrying out iterative adjustment on the pre-training model by adopting the data sample to obtain an acquisition strategy model, and outputting an acquisition strategy mapping table by the acquisition strategy model, wherein the acquisition strategy mapping table comprises corresponding acquisition objects, acquisition frequencies and thresholds under various running states; the data acquisition module is used for acquiring the current running state of the target equipment, mapping the current running state information to an acquisition strategy mapping table, screening out the optimal acquisition strategy, and acquiring sensor data of the target equipment according to the optimal acquisition strategy to obtain target data; and the data filtering module is used for constructing a quantization filtering rule according to the acquisition strategy mapping table, and filtering the target data through the quantization filtering rule to obtain the reported data.

Description

Edge router quick acquisition decision method and system Technical Field The application belongs to the technical field of the Internet of things, and particularly relates to a fast mining decision method and system for an edge router. Background With the deep development of industry 4.0 and universal intelligent alliance, an edge router is used as a key node for connecting a physical world and a digital world, and the edge router needs to transmit a large amount of data of various sensors to a cloud server in a network mode and then analyze and predict the data through a cloud prediction model. For the existing edge router, in order to avoid missing key information, high-frequency carpet type acquisition is required to be carried out on data of a sensor, so that a large amount of redundant data is generated, network congestion and waste of cloud storage computing resources are easily caused, and real valuable information is often submerged in massive data to influence the accuracy of prediction of a cloud prediction model. Although some edge routers are provided with filtering rules, the filtering rules are only collected and filtered by means of static rules or preset scripts, the instructions are only passively executed, and the collection and the filtering cannot be actively decided according to the value of the data and the working condition of real-time change. Moreover, when the edge router is applied to a new monitoring scene, a large amount of manual configuration and data samples are required for model training and strategy adjustment, the adaptation period is long, and the corresponding rapid deployment and iteration requirements of flexible manufacturing cannot be met. Therefore, the technical problem solved by the scheme is how to improve the intelligence, adaptability and resource utilization rate of the edge router. Disclosure of Invention The application mainly aims to provide a fast acquisition decision method of an edge router, which comprises the steps of training a pre-training model in a cloud by adopting a meta-learning algorithm, inputting a small amount of data samples of target equipment into the pre-training model, so as to obtain a personalized acquisition strategy model aiming at the target equipment, and outputting an acquisition strategy aiming at the target equipment through the acquisition strategy model, so that the edge router can carry out adaptive data acquisition aiming at the target equipment, reduce the data volume reported to the cloud, and improve the intelligence, adaptability and resource utilization rate of the edge router. Meanwhile, the edge router quick acquisition decision system is also provided. In order to achieve the above purpose, the present application adopts the following technical scheme: The edge router fast acquisition decision method comprises the following steps: step 1, loading a pre-training model, wherein the pre-training model is obtained by constructing a meta model by a cloud server based on a meta learning algorithm and training the meta model by adopting historical task data of various devices; Step 2, acquiring a data sample of target equipment, and performing iterative adjustment on a pre-training model by adopting the data sample to obtain an acquisition strategy model, wherein the acquisition strategy model outputs an acquisition strategy mapping table which comprises acquisition objects, acquisition frequencies and thresholds corresponding to various running states; Step 3, acquiring the current running state of the target equipment, mapping the current running state information to an acquisition strategy mapping table, screening out the optimal acquisition strategy, and acquiring sensor data of the target equipment according to the optimal acquisition strategy to obtain target data; and 4, constructing a quantization filtering rule according to the acquisition strategy mapping table, and filtering target data through the quantization filtering rule to obtain reported data. In the step 1, the training process of the pre-training model specifically comprises the steps that a cloud server builds a meta model based on a meta learning algorithm, the historical running states of various devices and the corresponding acquired data are used as a training set, the meta model is trained, and parameters of the meta model are optimized through a minimized strategy utility loss function, so that the pre-training model is obtained. Preferably, the formula of the minimization policy utility loss function is: ; Wherein, the Is thatIs used for the weight of the (c),To the extent to which the collected data contributes to the cloud fault prediction model,Is thatIs used for the weight of the (c),The bandwidth consumed for collecting the data, the CPU and the weighted sum of the storage space. Preferably, step2 comprises the sub-steps of: A1, collecting a data sample of target equipment, wherein the data sample is a corresponding collecting object and