Edge computing system and method for data processing and transmission of industrial flow meters

Abstract

The present application discloses an edge computing system and method for data processing and transmission of industrial flow meters, and relates to the field of data processing technology. The system includes an edge computing gateway and a blockchain network. The edge computing gateway is used for collecting, processing and uploading device data; the blockchain network is used to store the device data uploaded by the edge computing gateway and the monitoring information of the generated device data. The present application also discloses an edge computing method for data processing and transmission of industrial flow meters applicable to the above-mentioned system. The edge computing system and method for data processing and transmission of industrial flow meters of the present application integrate intelligent edge computing and blockchain technology, and provide intelligent, secure and scalable data management technology for equipment monitoring systems, which can improve the data security, availability and transmission efficiency of industrial flow meters, and can reflect high practicality and reliability in the fields of remote equipment monitoring and smart industrial monitoring.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and specifically to an edge computing system and method for data processing and transmission of industrial flow meters. Background Art

[0002] For industrial flow meters, the collected equipment data can generate clear data indications for applications such as industrial operation monitoring. Therefore, proper management and storage of industrial flow meter data is an effective means of protecting the interests of equipment manufacturers and enterprises.

[0003] The existing industrial flow meter data processing and transmission simply collects data through a collector, and then transmits the collected data through a proprietary flow meter interaction protocol and stores it in the respective servers. However, due to the importance of this part of data, once data tampering or malicious data intrusion occurs during the data collection and transmission process, it may lead to the possibility of a complete collapse of server data. Therefore, an efficient and secure industrial flow meter data transmission and processing management technology is urgently needed to deal with such problems. Summary of the invention

[0004] The purpose of this application is to provide an edge computing system and method for data processing and transmission of industrial flow meters to solve the technical problems raised in the above background technology.

[0005] To achieve the above objectives, this application discloses the following technical solutions:

[0006] In a first aspect, the present application discloses an edge computing system for data processing and transmission of industrial flow meters, including an edge computing gateway and a blockchain network;

[0007] The edge computing gateway is connected between the industrial flow meter and the blockchain network, and is used for collecting, processing and uploading device data;

[0008] The blockchain network is used to store the device data uploaded by the edge computing gateway and the monitoring information of the generated device data;

[0009] The edge computing gateway includes a plurality of edge nodes, an edge manager and an edge server;

[0010] The edge node is configured to be connected to the industrial flow meter and collect the equipment data of the industrial flow meter, and is also configured to process the collected equipment data. The edge manager is configured to call and manage multiple edge nodes. The edge server is configured to provide computing power for the edge node, and is also configured to upload the equipment data processed by the edge node to the blockchain network for storage and monitoring, and to call the equipment data in the blockchain network.

[0011] Preferably, the processing of the collected device data specifically includes:

[0012] Verify the device source: The edge node verifies the device source corresponding to the received device data to determine the legitimacy of the device data. When it is detected that the device source corresponding to the device data is a trusted device, it means that the device source verification of the device data has passed. Otherwise, it means that the device source verification of the device data has failed, and the upload of the device data is blocked.

[0013] Verify the authenticity of the data: The edge node verifies the device data that has passed the device source verification to determine the authenticity of the device data. When it is detected that the data content corresponding to the device data is data that is permitted to be transmitted and stored, it is defined that the authenticity verification of the device data has passed. Otherwise, it means that the authenticity verification of the device data has failed, and the storage of the device data is blocked.

[0014] Preferably, the verification of data authenticity also includes: the edge node performs data verification on the device data that has passed the authenticity verification to determine the validity of the device data, and when it is detected that the data content corresponding to the device data is the required data, it is defined that the validity verification of the device data has passed, otherwise, it indicates that the validity verification of the device data has failed and the device data is discarded.

[0015] Preferably, the edge node is further configured to: perform deep learning based on the device data that fails the device source verification and the device data that fails the authenticity verification, build a data contract verification model, and perform data verification on the received device data through the data contract verification model.

[0016] Preferably, the device source verification is performed by identifying a device source identifier carried in the device data. When the device source identifier is not identified, it indicates that the device source verification of the device data has failed. When the device source identifier is identified, the identified device source identifier is matched with the device source identifier corresponding to the stored trusted device. When the match is successful, it indicates that the device source verification of the device data has passed. Otherwise, it indicates that the device source verification of the device data has failed.

[0017] Preferably, the data verification of the device data that has passed the device source verification specifically includes:

[0018] Performing data recognition and feature extraction on the device data;

[0019] Based on the identified data type, the extracted features are compared with the data features corresponding to the preset data permitted to be transmitted and stored corresponding to the data type, and the obtained calculation result is compared with the preset matching threshold value. When the obtained calculation result is greater than the matching threshold value, it indicates that the data content corresponding to the device data is data permitted to be transmitted and stored, otherwise, it indicates that the data content corresponding to the device data is not data permitted to be transmitted and stored;

[0020] The calculation formula of the degree of fit AccW is:

[0021]

[0022] Where m is the feature extracted from the device data, ∑ρ is the total number of data features corresponding to the data type that is permitted to be transmitted and stored, i The sum of the weights of each extracted feature in the data features corresponding to the data permitted to be transmitted and stored.

[0023] Preferably, the data verification of the device data that has passed the authenticity verification specifically includes:

[0024] Based on the features extracted from the device data, a matching degree MaD is calculated between the extracted features and the data features corresponding to the preset required data, and the obtained calculation result is compared with the preset matching threshold value. When the obtained calculation result is greater than the matching threshold value, it indicates that the device data is the required data, otherwise, it indicates that the device data is not the required data;

[0025] The calculation formula of the matching degree MaD is:

[0026]

[0027] Among them, n i is the number of data features of type i among the extracted features, γ i is the weight value of the preset feature type i in the required data, and N is the total number of data features set in the required data.

[0028] In a second aspect, the present application discloses an edge computing method for data processing and transmission of an industrial flow meter, which is applicable to the above-mentioned edge computing system for data processing and transmission of an industrial flow meter, and the method comprises the following steps:

[0029] Connecting the industrial flow meter to an edge computing gateway, and connecting the edge computing gateway to a blockchain network, wherein the edge computing gateway includes a plurality of edge nodes, an edge manager, and an edge server;

[0030] The edge node collects the equipment data of the industrial flow meter and processes the collected equipment data;

[0031] The edge manager calls and manages the multiple edge nodes, provides computing power for the edge nodes, uploads the device data processed by the edge nodes to the blockchain network for storage and monitoring, and calls the device data in the blockchain network.

[0032] Preferably, the processing of the collected device data specifically includes:

[0033] Verify the device source: The edge node verifies the device source corresponding to the received device data to determine the legitimacy of the device data. When it is detected that the device source corresponding to the device data is a trusted device, it means that the device source verification of the device data has passed. Otherwise, it means that the device source verification of the device data has failed, and the upload of the device data is blocked.

[0034] Verify the authenticity of the data: The edge node verifies the device data that has passed the device source verification to determine the authenticity of the device data. When it is detected that the data content corresponding to the device data is data that is permitted to be transmitted and stored, it is defined that the authenticity verification of the device data has passed. Otherwise, it means that the authenticity verification of the device data has failed, and the storage of the device data is blocked.

[0035] Preferably, the verification of data authenticity also includes: the edge node performs data verification on the device data that has passed the authenticity verification to determine the validity of the device data, and when it is detected that the data content corresponding to the device data is the required data, it is defined that the validity verification of the device data has passed, otherwise, it indicates that the validity verification of the device data has failed and the device data is discarded.

[0036] Beneficial effects: The edge computing system and method for data processing and transmission of industrial flow meters in this application integrate intelligent edge computing and blockchain technology, provide intelligent, secure and scalable data management technology for equipment monitoring systems, and can improve the data security, availability and transmission efficiency of industrial flow meters. In the fields of remote equipment monitoring and smart industrial monitoring, it can demonstrate high practicality and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is a structural block diagram of an edge computing system for data processing and transmission of an industrial flow meter in an embodiment of the present application;

[0039] Figure 2 This is a flowchart of an edge computing method for data processing and transmission of an industrial flow meter in an embodiment of the present application. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present application are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0041] In this article, the term "comprising" is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, the elements defined by the sentence "comprising..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0042] This embodiment discloses Figure 1 An edge computing system for data processing and transmission of an industrial flow meter is shown, comprising an edge computing gateway and a blockchain network. The edge computing gateway is connected between the industrial flow meter and the blockchain network, and is used for collecting, processing and uploading device data. The blockchain network is used to store the device data uploaded by the edge computing gateway and the monitoring information of the generated device data.

[0043] Specifically, the edge computing gateway includes multiple edge nodes, an edge manager, and an edge server. The edge node is configured to connect to the industrial flow meter and collect the device data of the industrial flow meter, and is also configured to process the collected device data. The edge manager is configured to call and manage the multiple edge nodes. The edge server is configured to provide computing power for the edge node, and is also configured to upload the device data processed by the edge node to the blockchain network for storage and monitoring, and to call the device data in the blockchain network.

[0044] In this embodiment, the processing of the collected device data specifically includes:

[0045] Verify the device source: The edge node verifies the device source corresponding to the received device data to determine the legitimacy of the device data. When it is detected that the device source corresponding to the device data is a trusted device, it means that the device source verification of the device data has passed. Otherwise, it means that the device source verification of the device data has failed, and the upload of the device data is blocked.

[0046] Verify the authenticity of the data: The edge node verifies the device data that has passed the device source verification to determine the authenticity of the device data. When it is detected that the data content corresponding to the device data is data that is permitted to be transmitted and stored, it is defined that the authenticity verification of the device data has passed. Otherwise, it means that the authenticity verification of the device data has failed, and the storage of the device data is blocked.

[0047] As a preferred implementation mode of this embodiment, the verification of data authenticity also includes: the edge node performs data verification on the device data that has passed the authenticity verification to determine the validity of the device data, and when it is detected that the data content corresponding to the device data is the required data, it is defined that the validity verification of the device data has passed, otherwise, it indicates that the validity verification of the device data has failed and the device data is discarded.

[0048] As a preferred implementation manner of this embodiment, the edge node is also configured to: perform deep learning based on the device data that fails the device source verification and the authenticity verification, build a data contract verification model, and perform data verification on the received device data through the data contract verification model.

[0049] Among them, the verification of the device source is performed by identifying the device source identifier carried in the device data. When the device source identifier is not identified, it means that the device source verification of the device data has failed. When the device source identifier is identified, the identified device source identifier is matched with the device source identifier corresponding to the stored trusted device. When the match is successful, it means that the device source verification of the device data has passed. Otherwise, it means that the device source verification of the device data has failed.

[0050] Furthermore, the data verification of the device data that has passed the device source verification specifically includes:

[0051] Performing data recognition and feature extraction on the device data;

[0052] Based on the identified data type, the extracted features are compared with the data features corresponding to the preset data permitted to be transmitted and stored corresponding to the data type, and the obtained calculation result is compared with the preset matching threshold value. When the obtained calculation result is greater than the matching threshold value, it indicates that the data content corresponding to the device data is data permitted to be transmitted and stored, otherwise, it indicates that the data content corresponding to the device data is not data permitted to be transmitted and stored;

[0053] The calculation formula of the degree of fit AccW is:

[0054]

[0055] Where m is the feature extracted from the device data, ∑ρ is the total number of data features corresponding to the data type that is permitted to be transmitted and stored, i The sum of the weights of each extracted feature in the data features corresponding to the data permitted to be transmitted and stored.

[0056] Furthermore, the data verification of the device data that has passed the authenticity verification specifically includes:

[0057] Based on the features extracted from the device data, a matching degree MaD is calculated between the extracted features and the data features corresponding to the preset required data, and the obtained calculation result is compared with the preset matching threshold value. When the obtained calculation result is greater than the matching threshold value, it indicates that the device data is the required data, otherwise, it indicates that the device data is not the required data;

[0058] The calculation formula of the matching degree MaD is:

[0059]

[0060] Among them, n i is the number of data features of type i among the extracted features, γ i is the weight value of the preset feature type i in the required data, and N is the total number of data features set in the required data.

[0061] This embodiment also discloses Figure 2 The edge computing method for data processing and transmission of industrial flow meters shown is applicable to the above-mentioned edge computing system for data processing and transmission of industrial flow meters. The method comprises the following steps:

[0062] Connecting the industrial flow meter to an edge computing gateway, and connecting the edge computing gateway to a blockchain network, wherein the edge computing gateway includes a plurality of edge nodes, an edge manager, and an edge server;

[0063] The edge node collects the equipment data of the industrial flow meter and processes the collected equipment data;

[0064] The edge manager calls and manages the multiple edge nodes, provides computing power for the edge nodes, uploads the device data processed by the edge nodes to the blockchain network for storage and monitoring, and calls the device data in the blockchain network.

[0065] It should be noted that the above method is applicable to the aforementioned system. Therefore, the contents such as data transmission and management between the edge computing gateway and the blockchain network are the same and will not be elaborated here. In particular, in some feasible implementations of this embodiment, the overall system supports data interaction between multiple blockchain networks, promotes the access and utilization of device data in different networks, and at the same time, the system has the ability to monitor device status and data in real time. Once an abnormal situation is found, the system can immediately trigger the relevant event response mechanism, and use encryption technology to protect privacy during data upload. Only authorized personnel can access sensitive data. Since this part of the content is not the key protection content of this application, it will not be described in detail in this text.

[0066] It should be noted that the edge computing system and method for data processing and transmission of industrial flowmeters disclosed in this embodiment are suitable for application scenarios with high requirements for industrial flow stability. In the edge computing system, the data of industrial flowmeters are analyzed in real time, and the abnormal flow detection method is specifically as follows: data collection, algorithm denoising → data modeling, establishment of a digital twin model of the flowmeter → abnormality detection, if the difference exceeds the preset threshold, it is determined to be an abnormal situation → alarm and response, providing diagnostic suggestions and direct dialogue responses for the cause of the abnormality → model optimization module, collecting abnormality detection results, feedback adjustment, and continuous optimization of the digital twin model.

[0067] Mainly involves the following technologies:

[0068] 1. Data collection: Deploy the edge node to collect the real-time data of the industrial flow meter, and perform necessary preprocessing to improve data quality and reduce the false alarm rate of anomaly detection. The preprocessing of the data is mainly denoising. This solution uses a denoising autoencoder. The reasons for this are: the data of the flow meter has obvious time series characteristics, and the denoising autoencoder can learn time dependencies; the noise type of the flow meter may be more complex, including high-frequency noise, pulse noise, etc.; the denoising autoencoder can adapt to different types of noise through training; compared with other algorithms, the denoising autoencoder can directly perform denoising end-to-end without manual feature extraction; the denoising autoencoder is built based on a deep neural network, with strong learning ability and good denoising effect. Specifically, the construction method of denoising technology includes:

[0069] (1) Input layer: receives the original flow meter signal with added noise;

[0070] (2) Encoding layer: contains multiple Dense layers to extract features;

[0071] (3) Decoding layer: reconstructs the features of the encoding layer and outputs the denoised traffic signal;

[0072] (4) Loss function: Minimize the MSE loss between input and output.

[0073] The corresponding algorithm framework is as follows:

[0074] (1) Input layer: Receives the traffic time series data xt after adding noise, which contains T time step inputs.

[0075] (2) Coding layer: The coding mapping is realized through a multi-layer fully connected network, and the feature representation is extracted: h = f(W1xt+b1), W1 and b1 are the network parameters in the coding layer, and f is a nonlinear activation function such as ReLU;

[0076] (3) Bottleneck layer: It is the last layer of the encoding layer. It has a small number of units and is used for feature compression.

[0077] (4) Decoding layer: Decoding mapping is performed through network parameters W2, b2 to reconstruct the original signal: x′t=g(W2h+b2);

[0078] (5) Output layer: The decoding layer outputs the denoised traffic signal x';

[0079] (6) Loss function: Calculate the mean square error between input xt and output x′t as the loss function for optimization: L(θ) = 1 / T*Σ(xt-x′t)2.

[0080] By end-to-end training and minimizing the reconstruction loss L, the optimized denoising autoencoder model can be obtained to achieve denoising of flow meter data.

[0081] 2. Data modeling: RNN is used to build a digital twin model of the normal flow of each flow meter. The reasons for this are: flow meter data has time series dependency, and RNN can model time series data; RNN memorizes history through hidden layers and learns complex dynamic change rules; compared with linear models, RNN can model the nonlinear characteristics of flow meters; RNN has high computational efficiency and is suitable for real-time edge computing requirements; based on big data training, RNN has high prediction accuracy. Specifically, the construction method includes:

[0082] (1) Input layer: receives the traffic values ​​of the past n moments as sequence data;

[0083] (2) Hidden layer: Set the GRU or LSTM structure to capture the temporal dependency pattern;

[0084] (3) Output layer: predict the flow value at the next moment;

[0085] (4) Loss function: optimize the MSE of the predicted flow value and the actual value;

[0086] (5) Training network: Use historical traffic data for repeated training;

[0087] (6) Model evaluation: verify the accuracy of the model on the test set;

[0088] (7) Parameter tuning: improve network structure and hyperparameters to improve performance;

[0089] (8) Deployment model: Build a real-time prediction system to realize the digital twin of the flow meter.

[0090] By building, training, and continuously optimizing the RNN model, an accurate digital twin can be obtained to achieve real-time state prediction and monitoring of the flow meter. The RNN algorithm framework for flow meter digital twin modeling is as follows:

[0091] (1) Input layer: receiving historical traffic data xt of length L, t = 1, 2, ..., L;

[0092] (2) Hidden layer: The GRU structure is used to implement RNN, and its framework is as follows: rt = σ(Wrxt + Urht-1), zt = σ(Wzxt + Uzht-1), ht = tanh(Wxt + rt⊙(Uht-1)), where rt and zt are reset gates and update gates, ⊙ represents element-wise multiplication, and ht is the current hidden layer state;

[0093] (3) Output layer: fully connected layer obt = g(Whoht), predicting the traffic value at the next moment;

[0094] (4) Loss function: calculate the MSE loss between the predicted value and the true value and optimize it;

[0095] (5)L(θ)=1 / T*∑(y^t-yt)2;

[0096] (6) Model training: Use optimization algorithms such as Adam to minimize the loss function and train network parameters;

[0097] (7) Model evaluation: Evaluate accuracy indicators such as RMSE on the test set;

[0098] (8) Deployment: Obtain traffic data online and perform recursive prediction to realize digital twins.

[0099] 3. Anomaly detection: Compare the collected data and the predicted value of the digital twin model in real time. If the difference exceeds the preset threshold, it is judged as an abnormal situation. The threshold setting problem of flow meter anomaly detection adopts the autoencoder algorithm. The reasons are as follows: the autoencoder can reconstruct normal flow data through the network and establish anomaly judgment; the network training goal is to minimize the reconstruction error and directly generate the threshold; instances where the reconstruction error exceeds the threshold are judged as abnormal; it can be trained end-to-end by a large amount of historical data without manually setting the threshold; the calculation efficiency is high and can meet the real-time detection needs of the flow meter. Among them, the construction method includes:

[0100] (1) Collect a large amount of normal traffic historical data;

[0101] (2) Design encoder-decoder network structure and bottleneck features;

[0102] (3) Train the network to reconstruct normal samples and minimize the reconstruction error;

[0103] (4) Test data traffic and calculate network reconstruction errors;

[0104] (5) Errors exceeding the training threshold are considered abnormal;

[0105] (6) Continuous online learning to optimize network performance;

[0106] (7) Reduce training reconstruction error and improve detection sensitivity.

[0107] The algorithm framework is as follows:

[0108] (1) Collect historical traffic data {X1, X2, ...XN};

[0109] (2) Training the digital twin model DS based on the above traffic data;

[0110] (3) Use DS to predict historical traffic and obtain {X'1,X'2,...X'N};

[0111] (4) Take {X1,X2,...XN} and {X'1,X'2,...X'N} as training data;

[0112] (5) Construct the autoencoder network SEC;

[0113] (6) Input layer: accepts raw traffic data Xi;

[0114] (7) Coding layer: extract features;

[0115] (8) Decoding layer: reconstructed into predicted traffic data X'i;

[0116] (9) Output layer: predicted flow X'i;

[0117] (10) Loss function: MSE(Xi,X'i);

[0118] (11) Train SEC, minimize the reconstruction loss, and determine the threshold ε;

[0119] (12) For new traffic data X: use DS to generate predicted traffic X', SEC to reconstruct X", calculate MSE(X,X"), if MSE>ε, it is detected as an anomaly.

[0120] 4. Alarm and response: Once an abnormality is detected, an alarm is immediately sent to the operator, and combined with the semantic information of the industrial site, diagnostic suggestions and direct dialogue responses are provided for the cause of the abnormality. It can also be connected to the industrial control system for automated response. Method for constructing dialogue responses for flow meter abnormality alarms using large language models:

[0121] (1) Build an abnormal condition knowledge base to collect verbal descriptions of various abnormal modes of flow meters, such as "flow surge" and "sensor failure", and record them as text;

[0122] (2) Fine-tune the large language model based on the knowledge base in step 1, using the pre-trained language model to fine-tune it to adapt it to the abnormal semantics of the flow meter;

[0123] (3) Construct a question-and-answer dataset around abnormal situations and compile question-and-answer interaction corpus, including dialogues such as abnormal alarms, cause inquiries, and processing suggestions;

[0124] (4) Training the dialogue model Use the question-answering dataset to continue fine-tuning the language model so that it can conduct fluent conversations in abnormal situations;

[0125] (5) Integration into the response system: The dialogue model is deployed into the actual flow meter monitoring system and integrated with the anomaly detection module;

[0126] (6) Model optimization: Through actual interaction with users, feedback data is collected to continuously optimize the conversational capabilities of the large language model and the digital twin model of the flow meter.

[0127] The framework is as follows:

[0128] Pre-trained language model: GPT model is used, which has the ability to learn semantic representation;

[0129] Fine-tuning dataset: Use the anomaly knowledge base built in step 1 as text data;

[0130] Input representation: convert text into word vector sequence and input it into the model;

[0131] Model fine-tuning: retain the pre-trained parameters and add an output layer that adapts to abnormal semantics;

[0132] Loss function: Design cross entropy and other losses for abnormal semantic understanding tasks;

[0133] Optimized training: Use low learning rate to fine-tune pre-training parameters to learn abnormal semantics;

[0134] Application model: Deployed to the exception response system to generate semantic responses from exception data;

[0135] Continuous optimization: Collect user feedback to optimize model output.

[0136] 5. Model optimization: Collect anomaly detection results, feedback and adjust the digital twin model, continuously optimize the digital twin model, and improve the accuracy of anomaly detection. The Transformer model is used to optimize the digital twin model of the flow meter, and the results of anomaly detection in the online operation of the digital twin are collected. These detected abnormal data can be regarded as error cases of the model. The encoder-decoder structure of the Transformer is used, in which the encoder inputs abnormal data and the decoder aims to predict normal data. Through end-to-end training, the Transformer learns the mapping relationship between abnormal data and normal data. After obtaining the trained Transformer model, it is applied to the optimization of the digital twin model. When the newly collected data is detected as abnormal, the Transformer model is first used for prediction and correction, and then input into the digital twin. This can improve the digital twin's ability to model abnormal situations and reduce the false alarm rate. At the same time, continue to collect new abnormal data and continuously fine-tune the Transformer model. Using the parallel computing performance of the Transformer, the rapid online optimization of the digital twin can be achieved.

[0137] The algorithm framework of the flow meter digital twin model generated by using Transformer to optimize RNN includes:

[0138] (1) Use DSrnn (the constructed flow meter digital twin model) to generate the prediction result Y' of normal flow;

[0139] (2) For the detected abnormal traffic data X, DSrnn will have a large error;

[0140] (3) Collect these {X,Y'} as training data for the Transformer model;

[0141] (4) Build Transformer model T:

[0142] (41) Input: Abnormal flow X

[0143] (42) Encoder: Use self-attention mechanism to learn features

[0144] (43) Decoder: Output predicted normal traffic Y″

[0145] (44) Loss function: MSE(Y”,Y’);

[0146] (5) Train the Transformer model T, minimize the loss function, and obtain the optimized digital twin DSrnn': For new traffic data X, use DSrnn to generate Y'. If it is detected as an anomaly, use T(X) to obtain Y", and use Y" as a new sample of DSrnn' to continue optimizing. Continue to use new anomaly data to train T to improve the accuracy of DSrnn'.

[0147] The key to this algorithm is to build an accurate flow meter digital twin model and adopt appropriate anomaly detection algorithms and flow meter digital twin model optimization algorithms. Then the model is trained through the cloud platform / AI software, and the trained model is deployed to the edge computing system for data processing and transmission of industrial flow meters.

[0148] In the embodiments provided in the present application, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: application specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field programmable gate array (FPGA), processor, controller, microcontroller, microprocessor, other electronic units designed to implement the functions described herein or their combination. For software implementation, part or all of the flow of the embodiment can be completed by instructing the relevant hardware through a computer program. When implemented, the above program can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein the communication medium includes any medium that is convenient for transmitting a computer program from one place to another. The storage medium can be any available medium that a computer can access. The computer-readable storage medium can include but is not limited to RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer.

[0149] To sum up, the edge computing system and method for data processing and transmission of industrial flow meters in this embodiment integrate intelligent edge computing and blockchain technology, provide intelligent, secure and scalable data management technology for the equipment monitoring system, and can improve the data security, availability and transmission efficiency of industrial flow meters. In the fields of remote equipment monitoring and smart industrial monitoring, it can demonstrate high practicality and reliability.

[0150] Finally, it should be noted that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An edge computing system for data processing and transmission of industrial flow meters, characterized in that: Includes edge computing gateways and blockchain networks; The edge computing gateway is connected between the industrial flow meter and the blockchain network, and is used for collecting, processing and uploading device data; The blockchain network is used to store the device data uploaded by the edge computing gateway and the monitoring information of the generated device data; The edge computing gateway includes a plurality of edge nodes, an edge manager and an edge server; The edge node is configured to be connected to the industrial flow meter and collect the device data of the industrial flow meter, and is also configured to process the collected device data. The edge manager is configured to call and manage multiple edge nodes. The edge server is configured to provide computing power for the edge node, and is also configured to upload the device data processed by the edge node to the blockchain network for storage and monitoring, and to call the device data in the blockchain network; The processing of the collected device data specifically includes: Verify the device source: The edge node verifies the device source corresponding to the received device data to determine the legitimacy of the device data. When it is detected that the device source corresponding to the device data is a trusted device, it means that the device source verification of the device data has passed. Otherwise, it means that the device source verification of the device data has failed, and the upload of the device data is blocked. Verify the authenticity of the data: The edge node verifies the device data that has passed the device source verification to determine the authenticity of the device data. When it is detected that the data content corresponding to the device data is data that is permitted to be transmitted and stored, it is defined that the authenticity verification of the device data has passed. Otherwise, it means that the authenticity verification of the device data has failed, and the storage of the device data is blocked; The verifying the authenticity of the data further includes: the edge node verifies the device data that has passed the authenticity verification, determines the validity of the device data, and when it is detected that the data content corresponding to the device data is the required data, defines that the validity verification of the device data has passed, otherwise, it indicates that the validity verification of the device data has failed, and the device data is discarded; The data verification of the device data that has passed the device source verification specifically includes: Performing data recognition and feature extraction on the device data; Based on the identified data type, the extracted features are compared with the data features corresponding to the preset data permitted to be transmitted and stored corresponding to the data type, and the obtained calculation result is compared with the preset matching threshold value. When the obtained calculation result is greater than the matching threshold value, it indicates that the data content corresponding to the device data is data permitted to be transmitted and stored, otherwise, it indicates that the data content corresponding to the device data is not data permitted to be transmitted and stored; The calculation formula of the degree of fit AccW is: Wherein, m is the feature extracted from the device data, M is the total number of data features corresponding to the data permitted to be transmitted and stored corresponding to the data type, The sum of the weights of each extracted feature in the data features corresponding to the data permitted to be transmitted and stored; The data verification of the device data that has passed the authenticity verification specifically includes: Based on the features extracted from the device data, a matching degree MaD is calculated between the extracted features and the data features corresponding to the preset required data, and the obtained calculation result is compared with the preset matching threshold value. When the obtained calculation result is greater than the matching threshold value, it indicates that the device data is the required data, otherwise, it indicates that the device data is not the required data; The calculation formula of the matching degree MaD is: in, is the number of data features of type i among the extracted features, is the weight value of the preset feature type i in the required data, The total number of data features set in the required data; The edge computing gateway also performs real-time analysis on the data of the industrial flow meter to detect abnormal flow, wherein the method for detecting abnormal flow includes: data collection, algorithm denoising, establishing a digital twin model of the flow meter, anomaly detection, alarm and response, and model optimization; The data collection specifically includes: collecting real-time data of the industrial flow meter through the edge node, and preprocessing the real-time data; The algorithm denoising specifically includes: denoising the collected real-time data through a denoising autoencoder, wherein the algorithm framework corresponding to the denoising includes: (1) Input layer: receives the traffic time series data xt after adding noise, including T time step inputs; (2) Coding layer: The coding mapping is realized through a multi-layer fully connected network to extract feature representation: h = f(W1*xt + b1), W1 and b1 are the network parameters in the coding layer, and f is a nonlinear activation function such as ReLU; (3) Bottleneck layer: It is the last layer of the encoding layer. It has a small number of units and is used for feature compression. (4) Decoding layer: Decoding mapping is performed through network parameters W2, b2 to reconstruct the original signal: x′t = g(W2*h + b2); (5) Output layer: The decoding layer outputs the denoised traffic signal x't; (6) Loss function: Calculate the mean square error between input xt and output x′t as the loss function for optimization: ; Through end-to-end training, the reconstruction loss L is minimized to obtain the optimized denoising autoencoder model and realize the denoising of flow meter data; The establishment of the digital twin model of the flow meter specifically includes: using RNN to establish a digital twin model of the normal flow of each flow meter, wherein the RNN algorithm framework for the digital twin modeling of the flow meter includes: (1) Input layer: receiving historical traffic data pt, t=1,2,...,N with a length of N; (2) Hidden layer: The GRU structure is used to implement RNN. Its framework is as follows: rt = σ(Wr*pt + Ur*ht-1), zt = σ(Wz*pt + Uz*ht-1), ht = tanh(W*pt + rt⊙(U*ht-1)), where rt and zt are the reset gate and update gate, ⊙ represents element-wise multiplication, and ht is the current hidden layer state; (3) Output layer: fully connected layer obt = g(Who*Ht), predicting the traffic value at the next moment; (4) Loss function: calculate the MSE loss of the predicted value and the true value and optimize it; （5） ； (6) Model training: Use the Adam optimization algorithm to minimize the loss function and train network parameters; (7) Model evaluation: Evaluate accuracy indicators such as RMSE on the test set; (8) Deployment: Obtain traffic data online and perform recursive prediction to realize digital twins; The anomaly detection specifically includes: comparing the collected real-time data with the predicted value of the digital twin model in real time. If the difference between the real-time data and the predicted value exceeds a preset threshold, it is determined to be an abnormal situation. The algorithm framework corresponding to the anomaly detection includes: (1) Collect raw traffic data {X1, X2,...XN}; (2) Digital twin model based on flow meter; (3) Use the digital twin model to predict the original traffic data and obtain {X'1, X'2,...X'N}; (4) Take {X1, X2,...XN} and {X'1, X'2,...X'N} as training data; (5) Construct the autoencoder network SEC; (51) Input layer: accepts training data {X1, X2,...XN} and {X'1, X'2,...X'N}; (52) Coding layer: extract features; (53) Decoding layer: reconstructed into predicted traffic data X''i; (54) Output layer: predicted flow X''i; (6) Loss function: MSE(Xi, X'i); (7) Train SEC, minimize the reconstruction loss, and determine the threshold ε; (8) For new traffic data X: use the digital twin model to generate predicted traffic X', reconstruct X'' using the autoencoder network SEC, calculate the actual value MSE(X, X''), and if MSE > ε, it is detected as an anomaly; The alarm and response specifically include: when an abnormal situation is detected, an alarm is immediately sent to the operator, and combined with the semantic information of the industrial site, a diagnosis suggestion and a direct dialogue response are provided for the cause of the abnormality; wherein the algorithm framework corresponding to the alarm and response includes: (1) Pre-trained language model: The GPT model is used, which has the ability to learn semantic representations; (2) Fine-tuning dataset: collect verbal descriptions of various abnormal modes of flow meters to obtain an abnormal condition knowledge base, and use the abnormal condition knowledge base as text data; (3) Input representation: convert the text into a word vector sequence and input it into the model; (4) Model fine-tuning: retain the pre-trained parameters and add an output layer that adapts to abnormal semantics; (5) Loss function: Design cross entropy loss for abnormal semantic understanding tasks; (6) Optimized training: Use low learning rate to fine-tune pre-training parameters to learn abnormal semantics; (7) Application model: deployed to the exception response system to generate semantic responses from exception data; (8) Continuous optimization: Collect user feedback to optimize model output; The model optimization specifically includes: collecting anomaly detection results, adjusting the digital twin model through feedback, and continuously optimizing the digital twin model; wherein the algorithm framework of the model optimization includes: (1) After the digital twin model is established, the current real-time data can be input and the model will generate a predicted normal flow value Y'; (2) For the detected abnormal traffic data X, the digital twin model will have errors; (3) Collect these {X,Y'} as training data for the Transformer model; (4) Build Transformer model TR: (41) Input: Abnormal traffic X (42) Encoder: Using self-attention mechanism to learn features (43) Decoder: Output predicted normal traffic Y′′ (44) Loss function: MSE(Y'',Y'); (5) Train the Transformer model TR, minimize the loss function, and obtain the optimized digital twin DSrnn': For the new traffic data Q, use the digital twin model to generate Q'. If it is detected as an anomaly, use TR(Q) to obtain Q'', and use Q'' as a new sample of the digital twin DSrnn' to continue optimizing. Continue to use new abnormal data to train TR to improve the accuracy of the digital twin DSrnn'.

2. The edge computing system for data processing and transmission of industrial flow meters according to claim 1, characterized in that: The edge node is also configured to: perform deep learning based on the device data that fails the device source verification and the device data that fails the authenticity verification, build a data contract verification model, and perform data verification on the received device data through the data contract verification model.

3. The edge computing system for data processing and transmission of industrial flow meters according to claim 1, characterized in that: The device source verification is performed by identifying the device source identifier carried in the device data. When the device source identifier is not identified, it indicates that the device source verification of the device data has failed. When the device source identifier is identified, the identified device source identifier is matched with the device source identifier corresponding to the stored trusted device. When the match is successful, it indicates that the device source verification of the device data has passed. Otherwise, it indicates that the device source verification of the device data has failed.

4. An edge computing method for data processing and transmission of industrial flow meters, characterized in that: The method comprises the following steps: Connecting the industrial flow meter to an edge computing gateway, and connecting the edge computing gateway to a blockchain network, wherein the edge computing gateway includes a plurality of edge nodes, an edge manager, and an edge server; The edge node collects the equipment data of the industrial flow meter and processes the collected equipment data; The edge manager calls and manages the plurality of edge nodes, provides computing power for the edge nodes, uploads the device data processed by the edge nodes to the blockchain network for storage and monitoring, and calls the device data in the blockchain network; The processing of the collected device data specifically includes: Verify the device source: The edge node verifies the device source corresponding to the received device data to determine the legitimacy of the device data. When it is detected that the device source corresponding to the device data is a trusted device, it means that the device source verification of the device data has passed. Otherwise, it means that the device source verification of the device data has failed, and the upload of the device data is blocked. Verify the authenticity of the data: The edge node verifies the device data that has passed the device source verification to determine the authenticity of the device data. When it is detected that the data content corresponding to the device data is data that is permitted to be transmitted and stored, it is defined that the authenticity verification of the device data has passed. Otherwise, it means that the authenticity verification of the device data has failed, and the storage of the device data is blocked; The verifying the authenticity of the data further includes: the edge node verifies the device data that has passed the authenticity verification, determines the validity of the device data, and when it is detected that the data content corresponding to the device data is the required data, defines that the validity verification of the device data has passed, otherwise, it indicates that the validity verification of the device data has failed, and the device data is discarded; The data verification of the device data that has passed the device source verification specifically includes: Performing data recognition and feature extraction on the device data; Based on the identified data type, the extracted features are compared with the data features corresponding to the preset data permitted to be transmitted and stored corresponding to the data type, and the obtained calculation result is compared with the preset matching threshold value. When the obtained calculation result is greater than the matching threshold value, it indicates that the data content corresponding to the device data is data permitted to be transmitted and stored, otherwise, it indicates that the data content corresponding to the device data is not data permitted to be transmitted and stored; The calculation formula of the degree of fit AccW is: Wherein, m is the feature extracted from the device data, M is the total number of data features corresponding to the data permitted to be transmitted and stored corresponding to the data type, The sum of the weights of each extracted feature in the data features corresponding to the data permitted to be transmitted and stored; The data verification of the device data that has passed the authenticity verification specifically includes: Based on the features extracted from the device data, a matching degree MaD is calculated between the extracted features and the data features corresponding to the preset required data, and the obtained calculation result is compared with the preset matching threshold value. When the obtained calculation result is greater than the matching threshold value, it indicates that the device data is the required data, otherwise, it indicates that the device data is not the required data; The calculation formula of the matching degree MaD is: in, is the number of data features of type i among the extracted features, is the weight value of the preset feature type i in the required data, The total number of data features set in the required data; The edge computing gateway also performs real-time analysis on the data of the industrial flow meter to detect abnormal flow, wherein the method for detecting abnormal flow includes: data collection, algorithm denoising, establishing a digital twin model of the flow meter, anomaly detection, alarm and response, and model optimization; The data collection specifically includes: collecting real-time data of the industrial flow meter through the edge node, and preprocessing the real-time data; The algorithm denoising specifically includes: denoising the collected real-time data through a denoising autoencoder, wherein the algorithm framework corresponding to the denoising includes: The algorithm framework corresponding to denoising includes: (1) Input layer: receives the traffic time series data xt after adding noise, including T time step inputs; (2) Coding layer: The coding mapping is realized through a multi-layer fully connected network to extract feature representation: h = f(W1*xt + b1), W1 and b1 are the network parameters in the coding layer, and f is a nonlinear activation function such as ReLU; (3) Bottleneck layer: It is the last layer of the encoding layer. It has a small number of units and is used for feature compression. (4) Decoding layer: Decoding mapping is performed through network parameters W2, b2 to reconstruct the original signal: x′t = g(W2*h + b2); (5) Output layer: The decoding layer outputs the denoised traffic signal x't; (6) Loss function: Calculate the mean square error between input xt and output x′t as the loss function for optimization: ; Through end-to-end training, the reconstruction loss L is minimized to obtain the optimized denoising autoencoder model and realize the denoising of flow meter data; The establishment of the digital twin model of the flow meter specifically includes: using RNN to establish a digital twin model of the normal flow of each flow meter, wherein the RNN algorithm framework for the digital twin modeling of the flow meter includes: (1) Input layer: receiving historical traffic data pt, t=1,2,...,N with a length of N; (2) Hidden layer: The GRU structure is used to implement RNN. Its framework is as follows: rt = σ(Wr*pt + Ur*ht-1), zt = σ(Wz*pt + Uz*ht-1), ht = tanh(W*pt + rt⊙(U*ht-1)), where rt and zt are the reset gate and update gate, ⊙ represents element-wise multiplication, and ht is the current hidden layer state; (3) Output layer: fully connected layer obt = g(Who*Ht), predicting the traffic value at the next moment; (4) Loss function: calculate the MSE loss of the predicted value and the true value and optimize it; （5） ； (6) Model training: Use the Adam optimization algorithm to minimize the loss function and train network parameters; (7) Model evaluation: Evaluate accuracy indicators such as RMSE on the test set; (8) Deployment: Obtain traffic data online and perform recursive prediction to realize digital twins; The anomaly detection specifically includes: comparing the collected real-time data with the predicted value of the digital twin model in real time. If the difference between the real-time data and the predicted value exceeds a preset threshold, it is determined to be an abnormal situation. The algorithm framework corresponding to the anomaly detection includes: (1) Collect raw traffic data {X1, X2,...XN}; (2) Digital twin model based on flow meter; (3) Use the digital twin model to predict the original traffic data and obtain {X'1, X'2,...X'N}; (4) Take {X1, X2,...XN} and {X'1, X'2,...X'N} as training data; (5) Construct the autoencoder network SEC; (51) Input layer: accepts the training data from (4); (52) Coding layer: extract features; (53) Decoding layer: reconstructed into predicted traffic data X''i; (54) Output layer: predicted flow X''i; (6) Loss function: MSE(Xi, X'i); (7) Train SEC, minimize the reconstruction loss, and determine the threshold ε; (8) For new traffic data X: use the digital twin model to generate predicted traffic X', reconstruct X'' using the autoencoder network SEC, calculate the actual value MSE(X, X''), and if MSE > ε, it is detected as an anomaly; The alarm and response specifically include: when an abnormal situation is detected, an alarm is immediately sent to the operator, and combined with the semantic information of the industrial site, a diagnosis suggestion and a direct dialogue response are provided for the cause of the abnormality; wherein the algorithm framework corresponding to the alarm and response includes: (1) Pre-trained language model: The GPT model is used, which has the ability to learn semantic representations; (2) Fine-tuning dataset: collect verbal descriptions of various abnormal modes of flow meters to obtain an abnormal condition knowledge base, and use the abnormal condition knowledge base as text data; (3) Input representation: convert the text into a word vector sequence and input it into the model; (4) Model fine-tuning: retain the pre-trained parameters and add an output layer that adapts to abnormal semantics; (5) Loss function: Design cross entropy loss for abnormal semantic understanding tasks; (6) Optimized training: Use low learning rate to fine-tune pre-training parameters to learn abnormal semantics; (7) Application model: deployed to the exception response system to generate semantic responses from exception data; (8) Continuous optimization: Collect user feedback to optimize model output; The model optimization specifically includes: collecting anomaly detection results, adjusting the digital twin model through feedback, and continuously optimizing the digital twin model; wherein the algorithm framework of the model optimization includes: (1) After the digital twin model is established, the current real-time data can be input and the model will generate a predicted normal flow value Y'; (2) For the detected abnormal traffic data X, the digital twin model will have errors; (3) Collect these {X,Y'} as training data for the Transformer model; (4) Build Transformer model TR: (41) Input: Abnormal traffic X (42) Encoder: Using self-attention mechanism to learn features (43) Decoder: Output predicted normal traffic Y′′ (44) Loss function: MSE(Y'',Y'); (5) Train the Transformer model TR, minimize the loss function, and obtain the optimized digital twin DSrnn': For the new traffic data Q, use the digital twin model to generate Q'. If it is detected as an anomaly, use TR(Q) to obtain Q'', and use Q'' as a new sample of the digital twin DSrnn' to continue optimizing. Continue to use new abnormal data to train TR to improve the accuracy of the digital twin DSrnn'.

Images