Network box abnormality prediction method, system, device and medium based on edge computing

Through the fusion of edge computing technology and multi-sensor data, a neural network model is built for network box status monitoring, which solves the problems of low accuracy and unpredictable monitoring of traditional alarm methods, and achieves efficient, real-time and reliable early warnings, improving Security of networked box operation.

CN118568644BActive Publication Date: 2024-11-05BEIJING YINGCHUANGLIHE ELECTRONIC TECH CO LTD
View PDF 10 Cites 0 Cited by

Patent Information

Application Number
CN202411034855.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-05
Estimated Expiration
2044-07-31

AI Technical Summary

Technical Problem

The traditional networked box alarm method is based on single data, with poor environmental adaptability and anti-interference ability, and cannot perform predictive real-time monitoring, affecting operational safety.

Method used

Using edge computing-based methods, intelligent monitoring and early warning through multi-sensor data fusion and forward multi-layer network model, build a neural network model including input layer, hidden layer and output layer, train the model using BP algorithm, and use the edge device. Deploy, perform feature extraction and data fusion, and output abnormal prediction results.

Benefits of technology

It improves the accuracy and efficiency of early warning, enhances environmental adaptability and anti-interference capabilities, ensures efficient, real-time and reliability of networked boxes, and provides guarantees for safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118568644B_ABST
    Figure CN118568644B_ABST
Patent Text Reader

Abstract

The present invention provides a network box abnormality prediction method, system, device and medium based on edge computing, which relates to the field of data processing technology. The method constructs a forward multi-layer network model, trains the constructed model based on the BP algorithm and deploys the model to the edge device; the edge device obtains the sensor data of the network box, and extracts and fuses the acquired data as input data; the input data is input into the forward multi-layer network model, and the model outputs the result of whether the network box is abnormal; if the output result is that the network box is abnormal, the alarm information is sent to the user. This application realizes intelligent monitoring and early warning of abnormal states of network boxes through multi-sensor data fusion and optimized neural network models, with efficient, real-time and reliable effects, and solves the technical problems of low accuracy of alarms in the prior art and inability to perform predictive real-time monitoring of the operation of network boxes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, system, device and medium for predicting abnormality of a networked box based on edge computing. Background Art

[0002] The traditional alarm method for abnormal network cabinets is to directly measure through sensors, set alarm thresholds, and trigger alarms when the measured value reaches the alarm threshold. Alarm methods include tilt / vibration alarms: when the host or cabinet detects vibrations exceeding the threshold intensity, a vibration alarm can be triggered; when the host or cabinet detects a tilt angle exceeding the threshold, a tilt alarm can be triggered. Cabinets are usually monitored and alarmed for smoke and temperature.

[0003] Although the traditional alarm method is intuitive and easy to implement, it is only based on a single data, and has poor environmental adaptability and anti-interference ability, which affects the accuracy of the alarm. In addition, the alarm is only issued when the measured value reaches the alarm threshold. It is an alarm after an abnormal situation occurs, and it is unable to perform predictive real-time monitoring of abnormal conditions of the cabinet. Therefore, it does not contribute much to the safety of the operation of the network box. Summary of the invention

[0004] In view of this, the present invention provides a network box abnormality prediction method based on edge computing to solve the technical problems of low alarm accuracy and inability to perform predictive real-time monitoring of the operation of the network box in the prior art. The method includes:

[0005] Acquiring sensor data of a plurality of sensors, wherein the plurality of sensors are used to monitor environmental parameters of the network connection box;

[0006] Constructing a forward multi-layer network model including an input layer, multiple hidden layers and an output layer, training the constructed forward multi-layer network model based on the BP algorithm, and deploying the trained forward multi-layer network model to an edge device, wherein the edge device is arranged in the network connection box;

[0007] Extracting features from the sensor data of the multiple sensors respectively, and fusing the extracted feature data to obtain a data vector containing multiple elements as input data;

[0008] The input data is input into the trained forward multi-layer network model, and a prediction result of whether the network box is abnormal is output. If the prediction result is that the network box is abnormal, an alarm message is output.

[0009] The second aspect of the present invention provides a network box abnormality prediction system based on edge computing to solve the technical problems of low alarm accuracy and inability to perform predictive real-time monitoring of the operation of the network box in the prior art. The system includes:

[0010] A data acquisition module, used to acquire sensor data of a plurality of sensors, wherein the plurality of sensors are used to monitor environmental parameters of the network connection box;

[0011] A model deployment module, used to construct a forward multi-layer network model including an input layer, multiple hidden layers and an output layer, train the constructed forward multi-layer network model based on the BP algorithm, and deploy the trained forward multi-layer network model to an edge device, wherein the edge device is arranged in the network connection box;

[0012] A feature extraction module, used for respectively extracting features from the sensor data of the multiple sensors, and fusing the extracted feature data to obtain a data vector containing multiple elements as input data;

[0013] The early warning module is used to input the input data into the trained forward multi-layer network model, output a prediction result of whether the network box is abnormal, and output an alarm message if the prediction result is that the network box is abnormal.

[0014] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the above-mentioned networked box abnormality prediction methods based on edge computing is implemented.

[0015] The present invention also provides a computer-readable storage medium, which stores a computer program for executing any of the above-mentioned network box abnormality prediction methods based on edge computing.

[0016] Compared with the prior art, the beneficial effects that can be achieved by at least one of the above-mentioned technical solutions adopted in this specification include at least: providing a network-connected box abnormality prediction method based on edge computing, by fusing the sensors of multiple sensors combined with the forward multi-layer network model, the abnormal state of the network-connected box is intelligently monitored and warned. The comprehensive consideration of the sensor data of multiple sensors is conducive to improving the accuracy of the warning, and is conducive to improving the environmental adaptability and anti-interference ability. The use of the model not only further improves the accuracy of the warning, but also helps to improve the efficiency of the warning, and has efficient, real-time and reliable effects; the edge computing method is adopted to ensure the requirements of high performance and low latency, and provide a strong guarantee for the safety of the operation of the network-connected box. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flow chart of a network connection box abnormality prediction method based on edge computing provided by an embodiment of the present invention;

[0019] Figure 2 A schematic diagram of the structure of a network box abnormality prediction system based on edge computing provided by an embodiment of the present invention;

[0020] Figure 3 A schematic diagram of a computer device provided by an embodiment of the present invention.

[0021] Reference numerals in the figure: 200, networked box abnormality prediction system based on edge computing; 201, data acquisition module; 202, model deployment module; 203, feature extraction module; 204, early warning module; 301, memory; 302, processor. DETAILED DESCRIPTION

[0022] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0023] The following describes the implementation methods of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.

[0024] In an embodiment of the present invention, a network box abnormality prediction method based on edge computing is provided. Through multi-sensor data fusion and optimized neural network algorithm, intelligent monitoring and alarm of the network box status are realized, with efficient, real-time and reliable effects. The edge computing method is adopted to ensure high performance and low latency requirements, providing a strong guarantee for the safety of the network box operation.

[0025] like Figure 1As shown, the network connection box abnormality prediction method based on edge computing provided in this embodiment specifically includes the following steps:

[0026] Step S100, acquiring sensor data of a plurality of sensors, wherein the plurality of sensors are used to monitor environmental parameters of the network connection box;

[0027] Specifically, in this embodiment, data is collected from multiple sensors, and the sensors may be different types of sensors for collecting different environmental data. For example, the sensors may include temperature sensors, power sensors, smoke sensors, humidity sensors, flammable and explosive gas sensors, etc. The multiple sensors are used to capture data information about the surrounding and / or internal environment of the network connection box. The data of multiple sensors are collected in real time, and the collected sensor data is processed.

[0028] Step S200, constructing a forward multi-layer network model including an input layer, multiple hidden layers and an output layer, training the constructed forward multi-layer network model based on the BP algorithm, and deploying the trained forward multi-layer network model to an edge device, wherein the edge device is arranged in the network connection box;

[0029] Specifically, the constructed forward multi-layer network model can be a neural network model, using an adaptive network structure, and the number of hidden layers and the number of nodes in the input layer can be dynamically adjusted, for example, based on data complexity and real-time performance requirements. The number of nodes in the input layer can be equal to the number of sensors, and each node receives data from a sensor; there can be one or more hidden layers, each containing several nodes, and activation functions (such as ReLU) can be used to increase nonlinearity and improve model complexity. The number of nodes and layers in the hidden layer can be determined according to actual data processing requirements; the number of nodes in the output layer is equal to the dimension of the expected output.

[0030] Preferably, in this embodiment, a forward multi-layer network model is constructed, which includes an input layer with 5 input nodes (corresponding to 5 sensors), two hidden layers (several nodes in each layer) and an output layer with one output node (device status: normal or abnormal).

[0031] Furthermore, in step S200, in order to improve the performance of the forward multi-layer network model and the accuracy of early warning, the forward multi-layer network model constructed based on the BP algorithm is trained, including:

[0032] Step S201, defining a loss function;

[0033] Step S202: inputting historical sensor data into the forward multi-layer network model for iterative training;

[0034] Step S203, in the iterative training process, dynamically calculating the gradient of the loss function, and adjusting the weights, biases and hyperparameters of the forward multi-layer network model according to the gradient in combination with the BP algorithm and the selected optimization algorithm, wherein the iterative training process includes forward propagation and back propagation;

[0035] Step S204: When it is determined that the relevant indicators of the forward multi-layer network model meet the performance evaluation requirements, a trained forward multi-layer network model is obtained.

[0036] Furthermore, when it is judged that the relevant indicators of the forward multi-layer network model meet the performance evaluation, in the process of obtaining the trained forward multi-layer network model, the test set can be used to perform performance evaluation on the forward multi-layer network model, and the performance evaluation indicators include accuracy, recall rate and F1 score. When the performance evaluation indicators do not meet the evaluation criteria, the structure and training process of the forward multi-layer network model are optimized until the performance evaluation indicators meet the evaluation criteria, thereby obtaining the trained forward multi-layer network model.

[0037] Specifically, after completing the construction of the model, you can use the BP algorithm (Backpropagation Algorithm), adopt improved optimization algorithms (such as AdamW, Ranger) for training, and adjust weights and biases to improve training efficiency and model performance.

[0038] For example, the process of training a forward multi-layer network model may include the following steps:

[0039] a) Initialize model parameters, including weights and biases;

[0040] b) Define the loss function and use mean square error (MSE) as the loss function;

[0041] c) Perform forward propagation to calculate the model’s predicted output;

[0042] d) Calculate the gradient of the loss function using the back-propagation algorithm;

[0043] e) Adjust the weights and biases of the model according to the gradient, which can be done using optimization algorithms such as AdamW or Ranger;

[0044] f) Iterate forward propagation and back propagation until the performance of the model on the validation set meets the predetermined criteria;

[0045] g) Adjust model hyperparameters such as learning rate, number and size of hidden layers to optimize model performance;

[0046] h) Evaluate the model’s accuracy, recall, and F1 score using an independent test set. The F1 score is the harmonic mean of accuracy and recall.

[0047] i) Optimize the model structure and training process based on the evaluation results;

[0048] j) Deploy the trained model to edge devices for real-time monitoring and early warning.

[0049] Step S300: extracting features from the sensor data of the multiple sensors respectively, and fusing the extracted feature data to obtain a data vector containing multiple elements as input data;

[0050] Furthermore, the edge device may first perform data cleaning, standardization and normalization on the collected sensor data to facilitate neural network model processing. The data cleaning may include denoising and missing value processing, by introducing an automated data cleaning pipeline, using a machine learning model to perform data repair; and performing feature extraction on the sensor data after data cleaning, standardization and normalization.

[0051] Specifically, normalization: scaling the data to a certain range (such as [0, 1] or [-1, 1]) so that neural networks can process it more effectively; automated denoising and missing value processing: introducing automated data cleaning pipelines and using machine learning models (such as KNN interpolation or autoencoders) for data repair.

[0052] Furthermore, after data cleaning, standardization and normalization of the sensor data, useful feature data such as temperature threshold, power consumption pattern, smoke concentration and humidity level are extracted from the data of each sensor. After feature extraction, the feature data of different sensors are fused, and the data fusion method is used to improve the data quality and prediction accuracy to form a single input vector for input into the neural network model.

[0053] Preferably, when extracting features from the sensor data, a principal component analysis (PCA) or an independent component analysis (ICA) method may be used.

[0054] Furthermore, in step S300, the extracted feature data is fused to obtain a data vector containing multiple elements as input data, including:

[0055] Step S301, determining a fusion method according to the characteristics of the feature data of multiple sensors, and using the fusion method to fuse the feature data to generate a comprehensive feature vector;

[0056] Step S302: determine whether the data dimension of the comprehensive feature vector exceeds a preset threshold value. If yes, perform dimensionality reduction processing on the comprehensive feature vector, and construct the comprehensive feature vector after dimensionality reduction processing into a data vector containing multiple elements as input data, wherein the number of elements of the data vector is the same as the number of sensors;

[0057] Step S303: If the data dimension of the comprehensive feature vector does not exceed the preset threshold, construct the comprehensive feature vector into a data vector containing multiple elements as input data.

[0058] Specifically, we can select a suitable data fusion method based on the type, amount, and application requirements of feature data, and fuse the feature data of different sensors according to the selected fusion method to generate a comprehensive feature vector. If the fused feature vector data dimension is too high, we can use dimensionality reduction techniques such as PCA to reduce the number of features while retaining important feature information as much as possible. The fused feature vector is constructed as a data vector containing multiple elements as input data.

[0059] Step S400: input the input data into the trained forward multi-layer network model, output a prediction result of whether the network box is abnormal, and output an alarm message if the prediction result is that the network box is abnormal.

[0060] Furthermore, in order to further improve the accuracy of early warning, it is proposed that the edge devices adopt a distributed architecture to correct and comprehensively evaluate the prediction results of the network connection box, for example,

[0061] Constructing a distributed network topology structure for the edge devices of the plurality of network connection boxes and the central server, for example, the plurality of edge devices respectively establish communication connections with the same central server;

[0062] Based on the network topology, the plurality of edge devices respectively send the prediction results and the sensor data of the network connection boxes where they are located to the central server, and the plurality of edge devices respectively monitor different network connection boxes, and one edge device can be arranged in each network connection box;

[0063] The central server combines the prediction results of multiple network boxes and the sensor data for comprehensive evaluation to obtain comprehensive warning information, wherein the comprehensive warning information includes the comprehensive prediction results of each network box and abnormal situation statistics of the network boxes in a preset area.

[0064] Specifically, "edge devices" refer to computing devices deployed at the edge of the network, that is, close to the data source. These devices are capable of performing data processing, storage and certain decision-making tasks, reducing the transmission distance of data in the network, thereby reducing latency and increasing response speed. Wireless communication technologies (such as LoRa, NB-IoT) are used to transmit the fused sensor data, that is, the input data, to the edge device. The forward multi-layer network model is deployed on the edge device to reduce latency and increase response speed, thereby achieving local data processing and decision-making. By processing data locally, edge devices can reduce bandwidth requirements for central data centers. Edge devices can operate independently even if the connection to the central server is disconnected. Security measures can be implemented locally to protect data from unauthorized access.

[0065] By deploying neural network models on edge devices, real-time processing and low latency are ensured, and sensor data is processed in real time; GPU or dedicated hardware such as TPU is used to accelerate the calculation of neural networks to meet the needs of real-time processing. The output results of the forward multi-layer network model are used to evaluate the status of the network connection box to determine whether there is an abnormality; when an abnormality is predicted, the alarm system can be triggered through mobile applications or web interfaces to achieve real-time monitoring and alarm notifications, and can be sent to users through various alarm methods such as sound, light, electricity, or mobile terminals.

[0066] Preferably, in order to further improve the accuracy of the early warning, it is also possible to detect whether the sensor is abnormal so that the abnormal sensor can be repaired in time or the impact of the abnormal sensor on the early warning can be eliminated. For example, redundant sensors of the same type as the multiple sensors are deployed respectively; the data of the redundant sensors are cross-validated with the sensor data to determine the consistency of the sensor data of each sensor, and the cross-validation includes mean value comparison, variance analysis and correlation analysis; sensors with inconsistent sensor data are determined to be abnormal sensors.

[0067] Preferably, the setting of redundant sensors can deploy multiple sensors of the same type at key monitoring points, so that when a sensor fails or is abnormal, other sensors can continue to provide data. The consistency of the data of each sensor is judged based on the cross-validation of the data of each sensor and the data of the redundant sensor; if the sensor data is judged to be inconsistent, the sensor with abnormal data is marked as an abnormal sensor and processed according to a preset processing scheme. The threshold of data consistency can be adjusted according to environmental changes and system learning results, and an alarm is triggered when data inconsistency is detected, and appropriate response measures are taken.

[0068] Preferably, a redundant sensor configuration may be provided, and hardware redundancy or software redundancy may also be provided. A backup communication system may be provided to ensure that the network box can still communicate with other terminals or infrastructure when the primary communication link fails.

[0069] In a specific implementation, the edge device may be any form of device with edge computing capabilities, for example, a gateway.

[0070] Based on the same inventive concept, an edge computing-based network box abnormality prediction system is also provided in an embodiment of the present invention, as described in the following embodiments. Since the principle of solving the problem by an edge computing-based network box abnormality prediction system is similar to that of an edge computing-based network box abnormality prediction method, the implementation of an edge computing-based network box abnormality prediction system can refer to the implementation of an edge computing-based network box abnormality prediction method, and the repeated parts will not be repeated. As used below, the term "unit" or "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.

[0071] Figure 2 is a structural block diagram of a network box abnormality prediction system 200 based on edge computing according to an embodiment of the present invention, such as Figure 2 As shown, it includes: a data acquisition module 201, which is used to acquire sensor data of multiple sensors, and the multiple sensors are used to monitor the environmental parameters of the network connection box;

[0072] A model deployment module 202 is used to construct a forward multi-layer network model including an input layer, multiple hidden layers and an output layer, train the constructed forward multi-layer network model based on the BP algorithm, and deploy the trained forward multi-layer network model to an edge device, wherein the edge device is arranged in the network connection box;

[0073] A feature extraction module 203 is used to extract features from the sensor data of the multiple sensors respectively, and fuse the extracted feature data to obtain a data vector containing multiple elements as input data;

[0074] The early warning module 204 is used to input the input data into the trained forward multi-layer network model, output a prediction result of whether the network box is abnormal, and output an alarm message if the prediction result is that the network box is abnormal.

[0075] In one embodiment, the forward multi-layer network model is an adaptive network structure, which dynamically adjusts the number of hidden layers and the number of nodes in the input layer according to the number of sensors.

[0076] In one embodiment, the feature extraction module 203 is used to determine a fusion method according to the characteristics of the feature data of multiple sensors, use the fusion method to fuse the feature data, and generate a comprehensive feature vector; determine whether the data dimension of the comprehensive feature vector exceeds a preset threshold, and if so, perform dimensionality reduction processing on the comprehensive feature vector, and construct the comprehensive feature vector after dimensionality reduction processing into a data vector containing multiple elements as input data, wherein the number of elements of the data vector is the same as the number of sensors; if the data dimension of the comprehensive feature vector does not exceed the preset threshold, construct the comprehensive feature vector into a data vector containing multiple elements as input data.

[0077] In one embodiment, the model deployment module 202 is used to define a loss function; input historical sensor data into the forward multi-layer network model for iterative training; during the iterative training process, dynamically calculate the gradient of the loss function, and adjust the weights, biases and hyperparameters of the forward multi-layer network model according to the gradient in combination with the BP algorithm and the selected optimization algorithm, and the iterative training process includes forward propagation and back propagation; when it is judged that the relevant indicators of the forward multi-layer network model meet the performance evaluation, obtain the trained forward multi-layer network model.

[0078] In one embodiment, the model deployment module 202 is used to perform performance evaluation on the forward multi-layer network model using a test set, and the performance evaluation indicators include accuracy, recall rate and F1 score. When the performance evaluation indicators do not meet the evaluation criteria, the structure and training process of the forward multi-layer network model are optimized until the performance evaluation indicators meet the evaluation criteria, thereby obtaining a trained forward multi-layer network model.

[0079] In one embodiment, the early warning module 204 is also used to construct a distributed network topology structure for the edge devices of multiple network boxes and the central server; based on the network topology structure, the multiple edge devices respectively send the prediction results and the sensor data of the network boxes where they are located to the central server, and the multiple edge devices respectively monitor different network boxes; the central server combines the prediction results and the sensor data of the multiple network boxes for comprehensive evaluation to obtain comprehensive early warning information, wherein the comprehensive early warning information includes the comprehensive prediction results of each network box and the abnormal situation statistics of the network boxes in a preset area.

[0080] In one embodiment, the system further comprises:

[0081] The sensor anomaly determination module is used to respectively deploy redundant sensors of the same type as the multiple sensors; cross-validate the data of the redundant sensors with the sensor data to determine the consistency of the sensor data of each sensor, wherein the cross-validation includes mean value comparison, variance analysis and correlation analysis; and determine that the sensor with inconsistent sensor data is an abnormal sensor.

[0082] In this embodiment, a computer device is provided, such as Figure 3 As shown, it includes a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302. When the processor 302 executes the computer program, any of the above-mentioned network box abnormality prediction methods based on edge computing is implemented.

[0083] Specifically, the computer device may be a computer terminal, a server or a similar computing device.

[0084] In this embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program for executing any of the above-mentioned network box abnormality prediction methods based on edge computing.

[0085] Specifically, computer-readable storage media include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer-readable storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable storage media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0086] The embodiments of the present invention achieve the following technical effects:

[0087] 1. A network-connected box abnormality prediction method based on edge computing is provided. Through multi-sensor data fusion and optimized neural network algorithm, intelligent monitoring and alarm of the network-connected box status are realized, which has efficient, real-time and reliable effects. The edge computing method is adopted to ensure the requirements of high performance and low latency, and provide a strong guarantee for the safety of network-connected box operation.

[0088] 2. Safety redundancy design can significantly improve the robustness and reliability of the system in the face of various failures and abnormal situations, ensuring the continuity and safety of critical tasks.

[0089] Obviously, those skilled in the art should understand that the modules or steps of the above-mentioned embodiments of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the embodiments of the present invention are not limited to any specific combination of hardware and software.

[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A network box abnormality prediction method based on edge computing, the method runs on an edge device, characterized in that: include: Acquiring sensor data from a plurality of sensors for monitoring environmental parameters of an environment around and / or within the network connection box; Constructing a forward multi-layer network model including an input layer, multiple hidden layers and an output layer, training the constructed forward multi-layer network model based on the BP algorithm, and deploying the trained forward multi-layer network model to an edge device, wherein the edge device is arranged in the network connection box, the forward multi-layer network model is an adaptive network structure, and the number of hidden layers and the number of nodes in the input layer are dynamically adjusted according to the number of sensors, the number of nodes in the input layer is equal to the number of sensors, and each node receives data from one sensor; Extracting features from the sensor data of the multiple sensors respectively, and fusing the extracted feature data to obtain a data vector containing multiple elements as input data; Inputting the input data into the trained forward multi-layer network model, outputting a prediction result of whether the network box environment is abnormal, and outputting an alarm message if the prediction result is that the network box environment is abnormal; The extracted feature data is fused to obtain a data vector containing multiple elements as input data, including: Determine a fusion method according to the characteristics of the feature data of the multiple sensors, and use the fusion method to perform fusion processing on the feature data to generate a comprehensive feature vector; Determine whether the data dimension of the comprehensive feature vector exceeds a preset threshold value, and if so, perform dimensionality reduction processing on the comprehensive feature vector, and construct the comprehensive feature vector after dimensionality reduction processing into a data vector containing multiple elements as input data, wherein the number of elements of the data vector is the same as the number of sensors; If the data dimension of the comprehensive feature vector does not exceed the preset threshold, constructing the comprehensive feature vector into a data vector containing multiple elements as input data; The method further comprises: Constructing a distributed network topology structure within a preset area for the edge devices of the plurality of network connection boxes and the central server; Based on the network topology, the plurality of edge devices respectively send the prediction results and the sensor data of the network connection boxes where they are located to the central server, and the plurality of edge devices respectively monitor different network connection boxes; The central server performs a comprehensive evaluation based on the prediction results and sensor data of the plurality of network-connected boxes to obtain comprehensive warning information, wherein the comprehensive warning information includes the comprehensive prediction results of each network-connected box and statistical data on environmental anomalies of the network-connected boxes in a preset area.

2. According to the method for predicting abnormality of a network connection box based on edge computing in claim 1, it is characterized in that: The method of training the constructed forward multi-layer network model based on the BP algorithm includes: Define the loss function; Inputting historical sensor data into the forward multi-layer network model for iterative training; In an iterative training process, dynamically calculating the gradient of the loss function, and adjusting the weights, biases and hyperparameters of the forward multi-layer network model according to the gradient in combination with the BP algorithm and the selected optimization algorithm, wherein the iterative training process includes forward propagation and back propagation; When it is judged that the relevant indicators of the forward multi-layer network model meet the performance evaluation criteria, the trained forward multi-layer network model is obtained.

3. The method for predicting abnormality of a network connection box based on edge computing according to claim 2 is characterized in that: When it is determined that the relevant indicators of the forward multi-layer network model meet the performance evaluation criteria, the trained forward multi-layer network model is obtained, including: The performance of the forward multi-layer network model is evaluated using a test set, and the performance evaluation indicators include accuracy, recall rate and F1 score. When each of the indicators does not meet the performance evaluation standard, the structure and training process of the forward multi-layer network model are optimized until each of the performance evaluation indicators meets the performance evaluation standard, thereby obtaining the trained forward multi-layer network model.

4. A network box abnormality prediction method based on edge computing according to any one of claims 1 to 3, characterized in that: Also includes: respectively deploying redundant sensors of the same type as the multiple sensors; Cross-validating the data of each redundant sensor and its corresponding sensor data to determine the consistency of the sensor data of each sensor, wherein the cross-validation includes mean value comparison, variance analysis and correlation analysis; A sensor whose sensor data is inconsistent is determined to be an abnormal sensor.

5. A network box abnormality prediction system based on edge computing, characterized in that: include: A data acquisition module, used to acquire sensor data of a plurality of sensors, wherein the plurality of sensors are used to monitor environmental parameters of the surrounding and / or internal environment of the network connection box; A model deployment module, used to construct a forward multi-layer network model including an input layer, multiple hidden layers and an output layer, train the constructed forward multi-layer network model based on the BP algorithm, and deploy the trained forward multi-layer network model to an edge device, wherein the edge device is arranged in the network connection box, the forward multi-layer network model is an adaptive network structure, and the number of hidden layers and the number of nodes in the input layer are dynamically adjusted according to the number of sensors, the number of nodes in the input layer is equal to the number of sensors, and each node receives data from one sensor; A feature extraction module, used for respectively extracting features from the sensor data of the multiple sensors, and fusing the extracted feature data to obtain a data vector containing multiple elements as input data; An early warning module, used for inputting the input data into the trained forward multi-layer network model, outputting a prediction result of whether the network connection box is abnormal, and outputting an alarm message if the prediction result is that the network connection box is abnormal; A feature extraction module is used to determine a fusion method according to the characteristics of the feature data of multiple sensors, use the fusion method to fuse the feature data, and generate a comprehensive feature vector; determine whether the data dimension of the comprehensive feature vector exceeds a preset threshold, and if so, perform dimensionality reduction processing on the comprehensive feature vector, and construct the comprehensive feature vector after dimensionality reduction processing into a data vector containing multiple elements as input data, wherein the number of elements of the data vector is the same as the number of sensors; if the data dimension of the comprehensive feature vector does not exceed the preset threshold, construct the comprehensive feature vector into a data vector containing multiple elements as input data; The early warning module is also used to construct a distributed network topology structure for the edge devices of multiple network boxes and the central server; based on the network topology structure, the multiple edge devices respectively send the prediction results and the sensor data of the network boxes where they are located to the central server, and the multiple edge devices respectively monitor different network boxes; the central server combines the prediction results and the sensor data of the multiple network boxes for comprehensive evaluation to obtain comprehensive early warning information, wherein the comprehensive early warning information includes the comprehensive prediction results of each network box and the abnormal situation statistics of the network boxes in a preset area.

6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the network box abnormality prediction method based on edge computing described in any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program for executing the network box abnormality prediction method based on edge computing according to any one of claims 1 to 4.

Citation Information

Patent Citations

  • Sensor cross validation fault diagnosis method and device and computer equipment

    CN111947702A

  • Environmental monitoring instrument fault prediction method based on edge calculation and BP neural network

    CN114638290A

  • Equipment abnormity early warning method and system based on edge computing

    CN115643159A

  • Pressure transmitter data calibration method, device and system based on edge calculation

    CN117782431A

  • Energy storage method based on open source gap system

    CN118017564A