GIL expansion joint multi-dimensional data management system and method based on Internet of Things

Abstract

The invention discloses a GIL expansion joint multi-dimensional data management system and method based on the Internet of Things, and relates to the technical field of power equipment monitoring and data management, and the system comprises a data collection module, an Internet of Things transmission module, a data processing module and a data storage module. The data acquisition module is used for acquiring temperature, pressure, displacement and vibration data of the GIL expansion joint; the Internet of Things transmission module is used for encrypting and transmitting the multi-dimensional data acquired by the data acquisition module; the data processing module establishes a GIL expansion joint operation state prediction model according to the transmitted data; and the data storage module adopts a distributed storage technology, the processed data are stored in a cloud storage platform and a local server at the same time, all units cooperate to ensure effective management of the GIL expansion joint data, and the operation reliability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment monitoring and data management, and in particular to a multidimensional data management system and method for GIL expansion joints based on the Internet of Things. Background Art

[0002] GIL (Geared Interlocking Loop) is a highly efficient transmission line widely used in power transmission. During operation, its expansion joints are affected by a variety of factors, including temperature, pressure, displacement, and vibration. These changes can cause expansion joint failures, impacting the safe and stable operation of the entire transmission line. Traditional monitoring methods are unable to comprehensively and in real time acquire this multidimensional data, making it difficult to effectively analyze and manage the data, and thus unable to promptly identify potential fault hazards. Therefore, the development of a multidimensional data management system and method for GIL expansion joints based on the Internet of Things (IoT) is of great practical significance. Summary of the Invention

[0003] The purpose of the present invention is to provide a GIL expansion joint multidimensional data management system and method based on the Internet of Things to solve the problems raised in the prior art.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] The multidimensional data management system for GIL expansion joints based on the Internet of Things includes a data acquisition module, an Internet of Things transmission module, a data processing module and a data storage module; the data acquisition module is used to collect temperature, pressure, displacement and vibration data of the GIL expansion joint; the Internet of Things transmission module is used to encrypt and transmit the multidimensional data collected by the data acquisition module; the data processing module establishes a GIL expansion joint operation status prediction model based on the transmitted data; the data storage module adopts distributed storage technology to store the processed data in a cloud storage platform and a local server at the same time.

[0006] The data acquisition module includes a sensor unit, a sampling control unit, a data conditioning unit, a data cache unit and a communication interface unit;

[0007] The sensor unit consists of a temperature sensor, a pressure sensor, a displacement sensor, and a vibration sensor. These sensors are deployed on the GIL expansion joint based on the structural characteristics and operating characteristics of the GIL expansion joint to collect temperature T, pressure P, displacement D, vibration frequency f, and vibration amplitude A in real time during the operation of the GIL expansion joint. These sensors convert physical quantities into electrical signals and digital signals, and transmit them to the data processing module for preprocessing.

[0008] The sampling control unit is responsible for sampling according to the preset sampling period T s Collect and control sensor data; sampling period T sThe value range is 1-60 seconds; among them, the sampling frequency f s It is inversely proportional to the sampling period; the formula is as follows:

[0009]

[0010] The data conditioning unit is used to amplify the weak signal output by the sensor to a preset amplitude range, remove noise interference in the signal by filtering, and then perform level conversion to make the signal meet the input requirements of subsequent equipment;

[0011] The data cache unit is responsible for temporarily storing the collected and conditioned data. During the data transmission process, a first-in-first-out storage method is adopted. When the data transmission rate does not match the acquisition rate, the data storage unit is responsible for buffering the data transmission rate to prevent data loss;

[0012] The communication interface unit is responsible for transmitting the collected and processed data to the Internet of Things transmission module.

[0013] The Internet of Things transmission module includes a data encryption unit, a communication protocol adaptation unit, a wireless communication unit and a data retransmission and error correction unit;

[0014] The data encryption unit is used to encrypt the collected multidimensional data using the AES symmetric encryption algorithm before data transmission; the encryption process is as follows, the plaintext data is M, the encryption key is K, and the ciphertext C is obtained after the AES encryption algorithm is processed, and the formula is as follows:

[0015] C=AES K (M);

[0016] The communication protocol adaptation unit is responsible for adapting the encrypted data to the corresponding communication protocol, encapsulating the data, adding protocol headers and verification information, and ensuring that the data can be correctly transmitted in 5G and NB-IoT networks;

[0017] The wireless communication unit is responsible for sending the adapted data frames to the network through the corresponding wireless frequency band and communication mode according to the selected communication technology, and receiving feedback information from the data processing module;

[0018] The data retransmission and error correction unit detects errors by adding check information to the data frame. When an error is detected, a retransmission request is sent to the sender, requesting the data to be resent. At the same time, the unit also uses forward error correction technology to add redundant information to the data, so that the receiving end can correct the error by itself when receiving partially erroneous data, thereby reducing the number of retransmissions.

[0019] The data processing module includes a data decryption unit, a machine learning model training unit, a model evaluation and optimization unit, and a fault prediction and early warning unit;

[0020] After receiving the encrypted data transmitted by the IoT module, the data decryption unit first uses the same key K as used for encryption to decrypt the ciphertext C to obtain the plaintext data M. The decryption formula is as follows:

[0021]

[0022] The machine learning model training unit uses the data transmitted by the Internet of Things transmission module to train the machine model and establish a GIL expansion joint operation status prediction model; the model uses the multi-layer perceptron MLP in the neural network algorithm for model training;

[0023] The model evaluation and optimization unit uses a test data set to evaluate the trained GIL expansion joint operation status prediction model; then adjusts the model structure and model hyperparameters based on the evaluation results; and finally uses the test data set to evaluate and optimize the adjusted model;

[0024] The fault prediction and early warning unit is responsible for inputting the real-time collected and processed data into the trained model to obtain the current operating status prediction result of the equipment; then, based on the preset fault threshold and rules, it is determined whether the prediction result exceeds the normal range. If it exceeds, the early warning machine is triggered. The machine learning model training unit uses the data transmitted by the Internet of Things transmission module to train the machine model. Establishing the GIL expansion joint operating status prediction model includes the following contents:

[0025] First, the network structure of the MLP model is constructed, including the input layer, hidden layer, and output layer. The input layer receives the preprocessed multidimensional data feature vector x, and the hidden layer extracts and transforms the input data through nonlinear transformation. The nonlinear transformation formula is as follows:

[0026] h=f(W1x+b1)

[0027] Where W1 is the weight matrix from the input layer to the hidden layer, b1 is the bias vector, and f is the activation function, which is used to introduce nonlinear factors and enhance the model's expressiveness.

[0028] The output layer obtains the predicted running status label through linear transformation. The linear transformation formula is as follows:

[0029] y=W2h+b2

[0030] Where W2 is the weight matrix from the hidden layer to the output layer, and b2 is the bias vector;

[0031] The MLP model is then trained using historical and real-time data, and the model parameters W1, b1, W2, and b2 are iteratively updated using the backpropagation algorithm. The backpropagation algorithm is based on the idea of gradient descent, which calculates the gradient of the prediction error with respect to each parameter and updates the parameters in the opposite direction of the gradient to minimize the prediction error.

[0032] Finally, after multiple iterations of the model through the back-propagation algorithm, when the prediction error of the GIL expansion joint operation status prediction model is lower than the set threshold, the model construction and training is completed.

[0033] The model evaluation and optimization unit uses a test data set to evaluate the trained GIL expansion joint operation status prediction model; then adjusts the model structure and model hyperparameters based on the evaluation results; and finally uses the test data set to evaluate and optimize the adjusted model, including the following:

[0034] First, the model evaluation and optimization unit uses the test dataset to evaluate the trained model. The evaluation indicators include model prediction accuracy, recall rate, F1 value and mean square error.

[0035] The model prediction accuracy is the ratio of the number of samples correctly predicted by the model to the total number of samples. The calculation formula is as follows:

[0036]

[0037] Among them, TP represents the number of samples correctly predicted as positive, TN represents the number of samples correctly predicted as negative, FP represents the number of samples incorrectly predicted as positive, and FN represents the number of samples incorrectly predicted as negative.

[0038] Recall is the ratio of the number of samples correctly predicted by the model to the actual number of positive samples. The calculation formula is as follows:

[0039]

[0040] The recall rate reflects the model's ability to identify positive samples;

[0041] The F1 value is the harmonic mean of precision and recall, which is used to comprehensively measure the performance of the model. The calculation formula is as follows:

[0042]

[0043] The calculation formula for Precision is as follows:

[0044]

[0045] The range of F1 is {0-1}. The closer F1 is to 1, the better the performance of the model.

[0046] The mean square error is used to measure the average error between the model's predicted value and the true value. The calculation formula is as follows:

[0047]

[0048] Where n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample;

[0049] Secondly, according to the set threshold, the calculated evaluation index is compared with the threshold to determine whether each index of the model reaches the set threshold range;

[0050] Then, based on the evaluation results, when the model performance does not reach the threshold range, the model structure and hyperparameters are adjusted to optimize the model;

[0051] Regarding model structure adjustments, when the model's accuracy, recall, and F1 value are lower than the threshold, and the mean square error is higher than the threshold, the number of hidden layers is increased to improve the model's expressiveness. When the model overfits, the number of hidden layers is gradually reduced to reduce the model's complexity. After adjusting the number of hidden layers, the model needs to be re-evaluated.

[0052] For hyperparameter adjustment, hyperparameters include learning rate, number of iterations, and regularization parameters. For learning rate adjustment, when the various indicators in the model performance do not reach the set threshold, the model learning rate is adjusted using the adaptive learning rate method. This algorithm dynamically adjusts the learning rate based on the first-order moment estimate and second-order moment estimate of the parameter gradient during the model training process.

[0053] The number of iterations is determined by observing the performance changes of the model on the training set and the test set. The performance change of the model refers to the difference between the various indicators of the model and the performance in the training set and the test set. The early stopping strategy is used to select the number of iterations to adjust the model based on the performance change of the model. The number of iterations is determined based on the preset performance evaluation window. When the performance of the test set does not change within multiple consecutive evaluation windows, the training is stopped and the number of iterations at that time is recorded.

[0054] To determine the regularization parameter, the k-fold cross-validation method is used to select the appropriate regularization parameter value. The training data is divided into k mutually non-overlapping subsets, and k-1 subsets are used as training sets and 1 subset is used as validation sets each time. For different regularization parameter values, the average performance index of the model is calculated in k validations, and the regularization parameter value that optimizes the average performance is selected to control the model complexity and prevent overfitting.

[0055] Finally, the adjusted model is evaluated again using the test data set, the values of various evaluation indicators are calculated, the performance indicators of the models before and after adjustment are compared, and the effectiveness of the optimization measures is judged; if the adjusted model still does not meet the requirements, the above steps of model structure adjustment and hyperparameter adjustment are repeated, and continuous iterative optimization is performed to form a closed-loop model optimization mechanism.

[0056] The data storage module includes a cloud storage interaction unit, a local storage management unit, a data classification index unit and a data backup and synchronization unit;

[0057] The cloud storage interaction unit is responsible for docking and data interaction with the cloud storage platform. During the data upload phase, the unit receives data processed by the data processing module, performs adaptability processing on the data, and then transmits the data to the storage location specified by the cloud storage receipt according to the API provided by the cloud storage platform. During the data download phase, based on data requests for the operating status and fault diagnosis of the GIL expansion joint from other modules, the unit initiates a data read request to the cloud storage platform through the API, obtains the required data, and returns it to the requesting module.

[0058] The local storage management unit is responsible for receiving data from the data processing module, storing the data in the corresponding storage location according to the established classification rules of the GIL expansion joint, and updating the data index; when data is requested, it locates the storage location of the data on the local server according to the index, reads the data and returns it to the request module;

[0059] Before data storage, the data classification index unit classifies the data according to the timestamp and data type of the GIL expansion joint data and the preset classification rules; then, according to the category and content of the data, a unique index item is assigned to each data and the index item is stored in the index table; in the data query stage, according to the user and other modules' query conditions for the operating data of a specific part and a specific time range of the GIL expansion joint, the time range and data type information in the query conditions are parsed, and the index table is used to locate the data storage location that meets the conditions and perform data query;

[0060] The data backup and synchronization unit is responsible for backing up and synchronizing data between the cloud storage and the local server; it backs up the data of the local server to the cloud storage platform according to the preset time nodes, and at the same time, monitors the data changes of the cloud storage platform and the local server in real time. When the data of one party is updated, the updated data will be synchronized to the other party to maintain the integrity of the data on both ends.

[0061] The multi-dimensional data management method of the GIL expansion joint based on the Internet of Things includes the following steps:

[0062] S1, the sensor unit is deployed according to the characteristics of the GIL expansion joint, collecting temperature, pressure, displacement and vibration data in real time, which is then processed by the sampling control, conditioning and cache units, and finally transmitted to the IoT transmission module by the communication interface unit;

[0063] S2. The data encryption unit uses the AES algorithm to encrypt the data. The communication protocol adaptation unit adapts it to the corresponding protocol and encapsulates it. Then the wireless communication unit sends the data frame. The data retransmission and error correction unit ensures the transmission accuracy.

[0064] S3: The data decryption unit decrypts the encrypted data. The machine learning model training unit builds and trains a GIL expansion joint operation status prediction model. The model evaluation and optimization unit evaluates and adjusts the model. The fault prediction and warning unit determines whether to trigger a warning based on the model prediction results.

[0065] S4. The cloud storage interaction unit connects with the cloud platform to complete data upload and download. The local storage management unit stores and reads data according to the rules. The data classification and indexing unit classifies the data and creates an index. The data backup and synchronization unit ensures the consistency between cloud storage and local data.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] 1. Comprehensive monitoring and accurate data acquisition: Through the various sensors in the data acquisition module, which are deployed according to the structural characteristics and operating characteristics of the GIL expansion joint, multi-dimensional data such as temperature, pressure, displacement, vibration frequency and vibration amplitude can be collected in real time to achieve comprehensive monitoring of the operating status of the GIL expansion joint; the data conditioning unit amplifies, filters and level converts the sensor output signal to ensure the accuracy and availability of the data, providing a reliable data basis for subsequent analysis.

[0068] 2. Accurate Fault Prediction and Warning: The data processing module uses the multi-layer perceptron (MLP) in the neural network algorithm to establish a GIL expansion joint operating status prediction model. The model evaluation and optimization unit continuously adjusts the model structure and hyperparameters to improve the model's accuracy and reliability. Based on the model's prediction results, combined with preset fault thresholds and rules, the fault prediction and warning unit issues timely and accurate warnings, enabling operations and maintenance personnel to take preemptive measures, reduce the risk of equipment failure, and ensure stable operation of the power system.

[0069] 3. Secure and efficient data transmission: The IoT transmission module uses the AES symmetric encryption algorithm to encrypt data. Combined with communication protocol adaptation, wireless communication, and data retransmission and error correction units, it ensures the secure, accurate, and efficient transmission of data in 5G and NB-IoT networks, effectively preventing data from being stolen, tampered with, or lost during transmission.

[0070] 4. Reliable Data Storage and Management: The data storage module utilizes distributed storage technology. Through the collaborative work of the cloud storage interaction unit, local storage management unit, data classification and indexing unit, and data backup and synchronization unit, data is stored simultaneously on the cloud storage platform and local servers. This enables classified data storage, rapid indexing and querying, and real-time backup and synchronization of data on both ends, ensuring data security, integrity, and traceability, and facilitating subsequent data analysis and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 Schematic diagram of the system architecture of the multi-dimensional data management system for GIL expansion joints based on the Internet of Things of the present invention;

[0072] Figure 2 The present invention is a method flow chart of a multi-dimensional data management method for GIL expansion joints based on the Internet of Things. DETAILED DESCRIPTION

[0073] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0074] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution.

[0075] The multidimensional data management system for GIL expansion joints based on the Internet of Things includes a data acquisition module, an Internet of Things transmission module, a data processing module and a data storage module; the data acquisition module is used to collect temperature, pressure, displacement and vibration data of the GIL expansion joint; the Internet of Things transmission module is used to encrypt and transmit the multidimensional data collected by the data acquisition module; the data processing module establishes a GIL expansion joint operation status prediction model based on the transmitted data; the data storage module adopts distributed storage technology to store the processed data in a cloud storage platform and a local server at the same time.

[0076] The data acquisition module includes a sensor unit, a sampling control unit, a data conditioning unit, a data cache unit and a communication interface unit;

[0077] The sensor unit consists of a temperature sensor, a pressure sensor, a displacement sensor, and a vibration sensor. These sensors are deployed on the GIL expansion joint based on the structural characteristics and operating characteristics of the GIL expansion joint to collect temperature T, pressure P, displacement D, vibration frequency f, and vibration amplitude A in real time during the operation of the GIL expansion joint. These sensors convert physical quantities into electrical signals and digital signals, and transmit them to the data processing module for preprocessing.

[0078] The sampling control unit is responsible for sampling according to the preset sampling period T s Collect and control sensor data; sampling period T s The value range is 1-60 seconds; among them, the sampling frequency f s It is inversely proportional to the sampling period; the formula is as follows:

[0079]

[0080] The data conditioning unit is used to amplify the weak signal output by the sensor to a preset amplitude range, remove noise interference in the signal by filtering, and then perform level conversion to make the signal meet the input requirements of subsequent equipment;

[0081] The data cache unit is responsible for temporarily storing the collected and conditioned data. During the data transmission process, a first-in-first-out storage method is adopted. When the data transmission rate does not match the acquisition rate, the data storage unit is responsible for buffering the data transmission rate to prevent data loss;

[0082] The communication interface unit is responsible for transmitting the collected and processed data to the Internet of Things transmission module.

[0083] The Internet of Things transmission module includes a data encryption unit, a communication protocol adaptation unit, a wireless communication unit and a data retransmission and error correction unit;

[0084] The data encryption unit is used to encrypt the collected multidimensional data using the AES symmetric encryption algorithm before data transmission; the encryption process is as follows, the plaintext data is M, the encryption key is K, and the ciphertext C is obtained after the AES encryption algorithm is processed, and the formula is as follows:

[0085] C=AES K (M);

[0086] The communication protocol adaptation unit is responsible for adapting the encrypted data to the corresponding communication protocol, encapsulating the data, adding protocol headers and verification information, and ensuring that the data can be correctly transmitted in 5G and NB-IoT networks;

[0087] The wireless communication unit is responsible for sending the adapted data frames to the network through the corresponding wireless frequency band and communication mode according to the selected communication technology, and receiving feedback information from the data processing module;

[0088] The data retransmission and error correction unit detects errors by adding check information to the data frame. When an error is detected, a retransmission request is sent to the sender, requesting the data to be resent. At the same time, the unit also uses forward error correction technology to add redundant information to the data, so that the receiving end can correct the error by itself when receiving partially erroneous data, thereby reducing the number of retransmissions.

[0089] The data processing module includes a data decryption unit, a machine learning model training unit, a model evaluation and optimization unit, and a fault prediction and early warning unit;

[0090] After receiving the encrypted data transmitted by the IoT module, the data decryption unit first uses the same key K as used for encryption to decrypt the ciphertext C to obtain the plaintext data M. The decryption formula is as follows:

[0091]

[0092] The machine learning model training unit uses the data transmitted by the Internet of Things transmission module to train the machine model and establish a GIL expansion joint operation status prediction model; the model uses the multi-layer perceptron MLP in the neural network algorithm for model training;

[0093] The model evaluation and optimization unit uses a test data set to evaluate the trained GIL expansion joint operation status prediction model; then adjusts the model structure and model hyperparameters based on the evaluation results; and finally uses the test data set to evaluate and optimize the adjusted model;

[0094] The fault prediction and early warning unit is responsible for inputting the real-time collected and processed data into the trained model to obtain the current operating status prediction result of the equipment; then, based on the preset fault threshold and rules, it is determined whether the prediction result exceeds the normal range. If it exceeds, the early warning machine is triggered. The machine learning model training unit uses the data transmitted by the Internet of Things transmission module to train the machine model. Establishing the GIL expansion joint operating status prediction model includes the following contents:

[0095] First, the network structure of the MLP model is constructed, including the input layer, hidden layer, and output layer. The input layer receives the preprocessed multidimensional data feature vector x, and the hidden layer extracts and transforms the input data through nonlinear transformation. The nonlinear transformation formula is as follows:

[0096] h=f(W1x+b1)

[0097] Where W1 is the weight matrix from the input layer to the hidden layer, b1 is the bias vector, and f is the activation function, which is used to introduce nonlinear factors and enhance the model's expressiveness.

[0098] The output layer obtains the predicted running status label through linear transformation. The linear transformation formula is as follows:

[0099] y=W2h+b2

[0100] Where W2 is the weight matrix from the hidden layer to the output layer, and b2 is the bias vector;

[0101] The MLP model is then trained using historical and real-time data, and the model parameters W1, b1, W2, and b2 are iteratively updated using the backpropagation algorithm. The backpropagation algorithm is based on the idea of gradient descent, which calculates the gradient of the prediction error with respect to each parameter and updates the parameters in the opposite direction of the gradient to minimize the prediction error.

[0102] Finally, after multiple iterations of the model through the back-propagation algorithm, when the prediction error of the GIL expansion joint operation status prediction model is lower than the set threshold, the model construction and training is completed.

[0103] The model evaluation and optimization unit uses a test data set to evaluate the trained GIL expansion joint operation status prediction model; then adjusts the model structure and model hyperparameters based on the evaluation results; and finally uses the test data set to evaluate and optimize the adjusted model, including the following:

[0104] First, the model evaluation and optimization unit uses the test dataset to evaluate the trained model. The evaluation indicators include model prediction accuracy, recall rate, F1 value and mean square error.

[0105] The model prediction accuracy is the ratio of the number of samples correctly predicted by the model to the total number of samples. The calculation formula is as follows:

[0106]

[0107] Among them, TP represents the number of samples correctly predicted as positive, TN represents the number of samples correctly predicted as negative, FP represents the number of samples incorrectly predicted as positive, and FN represents the number of samples incorrectly predicted as negative.

[0108] Recall is the ratio of the number of samples correctly predicted by the model to the actual number of positive samples. The calculation formula is as follows:

[0109]

[0110] The recall rate reflects the model's ability to identify positive samples;

[0111] The F1 value is the harmonic mean of precision and recall, which is used to comprehensively measure the performance of the model. The calculation formula is as follows:

[0112]

[0113] The calculation formula for Precision is as follows:

[0114]

[0115] The range of F1 is {0-1}. The closer F1 is to 1, the better the performance of the model.

[0116] The mean square error is used to measure the average error between the model's predicted value and the true value. The calculation formula is as follows:

[0117]

[0118] Where n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample;

[0119] Secondly, according to the set threshold, the calculated evaluation index is compared with the threshold to determine whether each index of the model reaches the set threshold range;

[0120] Then, based on the evaluation results, when the model performance does not reach the threshold range, the model structure and hyperparameters are adjusted to optimize the model;

[0121] Regarding model structure adjustments, when the model's accuracy, recall, and F1 value are lower than the threshold, and the mean square error is higher than the threshold, the number of hidden layers is increased to improve the model's expressiveness. When the model overfits, the number of hidden layers is gradually reduced to reduce the model's complexity. After adjusting the number of hidden layers, the model needs to be re-evaluated.

[0122] For hyperparameter adjustment, hyperparameters include learning rate, number of iterations, and regularization parameters. For learning rate adjustment, when the various indicators in the model performance do not reach the set threshold, the model learning rate is adjusted using the adaptive learning rate method. This algorithm dynamically adjusts the learning rate based on the first-order moment estimate and second-order moment estimate of the parameter gradient during the model training process.

[0123] The number of iterations is determined by observing the performance changes of the model on the training set and the test set. The performance change of the model refers to the difference between the various indicators of the model and the performance in the training set and the test set. The early stopping strategy is used to select the number of iterations to adjust the model based on the performance change of the model. The number of iterations is determined based on the preset performance evaluation window. When the performance of the test set does not change within multiple consecutive evaluation windows, the training is stopped and the number of iterations at that time is recorded.

[0124] To determine the regularization parameter, the k-fold cross-validation method is used to select the appropriate regularization parameter value. The training data is divided into k mutually non-overlapping subsets, and k-1 subsets are used as training sets and 1 subset is used as validation sets each time. For different regularization parameter values, the average performance index of the model is calculated in k validations, and the regularization parameter value that optimizes the average performance is selected to control the model complexity and prevent overfitting.

[0125] Finally, the adjusted model is evaluated again using the test data set, the values of various evaluation indicators are calculated, the performance indicators of the models before and after adjustment are compared, and the effectiveness of the optimization measures is judged; if the adjusted model still does not meet the requirements, the above steps of model structure adjustment and hyperparameter adjustment are repeated, and continuous iterative optimization is performed to form a closed-loop model optimization mechanism.

[0126] The data storage module includes a cloud storage interaction unit, a local storage management unit, a data classification index unit and a data backup and synchronization unit;

[0127] The cloud storage interaction unit is responsible for docking and data interaction with the cloud storage platform. During the data upload phase, the unit receives data processed by the data processing module, performs adaptability processing on the data, and then transmits the data to the storage location specified by the cloud storage receipt according to the API provided by the cloud storage platform. During the data download phase, based on data requests for the operating status and fault diagnosis of the GIL expansion joint from other modules, the unit initiates a data read request to the cloud storage platform through the API, obtains the required data, and returns it to the requesting module.

[0128] The local storage management unit is responsible for receiving data from the data processing module, storing the data in the corresponding storage location according to the established classification rules of the GIL expansion joint, and updating the data index; when data is requested, it locates the storage location of the data on the local server according to the index, reads the data and returns it to the request module;

[0129] Before data storage, the data classification index unit classifies the data according to the timestamp and data type of the GIL expansion joint data and the preset classification rules; then, according to the category and content of the data, a unique index item is assigned to each data and the index item is stored in the index table; in the data query stage, according to the user and other modules' query conditions for the operating data of a specific part and a specific time range of the GIL expansion joint, the time range and data type information in the query conditions are parsed, and the index table is used to locate the data storage location that meets the conditions and perform data query;

[0130] The data backup and synchronization unit is responsible for backing up and synchronizing data between the cloud storage and the local server; it backs up the data of the local server to the cloud storage platform according to the preset time nodes, and at the same time, monitors the data changes of the cloud storage platform and the local server in real time. When the data of one party is updated, the updated data will be synchronized to the other party to maintain the integrity of the data on both ends.

[0131] The multi-dimensional data management method of the GIL expansion joint based on the Internet of Things includes the following steps:

[0132] S1, the sensor unit is deployed according to the characteristics of the GIL expansion joint, collecting temperature, pressure, displacement and vibration data in real time, which is then processed by the sampling control, conditioning and cache units, and finally transmitted to the IoT transmission module by the communication interface unit;

[0133] S2. The data encryption unit uses the AES algorithm to encrypt the data. The communication protocol adaptation unit adapts it to the corresponding protocol and encapsulates it. Then the wireless communication unit sends the data frame. The data retransmission and error correction unit ensures the transmission accuracy.

[0134] S3: The data decryption unit decrypts the encrypted data. The machine learning model training unit builds and trains a GIL expansion joint operation status prediction model. The model evaluation and optimization unit evaluates and adjusts the model. The fault prediction and warning unit determines whether to trigger a warning based on the model prediction results.

[0135] S4. The cloud storage interaction unit connects with the cloud platform to complete data upload and download. The local storage management unit stores and reads data according to the rules. The data classification and indexing unit classifies the data and creates an index. The data backup and synchronization unit ensures the consistency between cloud storage and local data.

[0136] Example:

[0137] 1. Data collection stage

[0138] At a certain GIL expansion joint monitoring site, temperature sensors, pressure sensors, displacement sensors, and vibration sensors are installed at key locations. Assume that at a certain moment, the temperature sensor collects temperature T = 42 ° C, the pressure sensor collects pressure P = 2.3 MPa, the displacement sensor monitors displacement D = 3 mm, and the vibration sensor measures vibration frequency f = 45 Hz and vibration amplitude A = 0.08 mm. The sampling control unit is set according to the preset sampling period T s =5 seconds for data acquisition. Based on the inverse relationship between sampling frequency and sampling period, the sampling frequency f = 0.2Hz. The data conditioning unit processes the weak signal output by the sensor, for example, amplifying the voltage signal output by the temperature sensor by 2 times to make its amplitude meet the requirements of subsequent equipment, and removes high-frequency noise interference through filtering.

[0139] 2. Data transmission stage

[0140] The collected multidimensional data (here the data is represented in a simplified array form) is encrypted by the data encryption unit using the AES symmetric encryption algorithm. Assuming that the encryption key is K, the ciphertext C is obtained after encryption. The communication protocol adaptation unit adapts the encrypted data to the 5G communication protocol, adds the protocol header and verification information, and encapsulates it into a data frame. The wireless communication unit sends the data frame through a specific 5G frequency band, and the data retransmission and error correction unit adds cyclic redundancy check (CRC) information to the data frame for error detection. If the receiving end detects that the data frame error rate is 5%, it sends a retransmission request to the sending end, and uses forward error correction technology to add 10% redundant information to the data, so that the receiving end can correct some erroneous data and reduce the number of retransmissions.

[0141] Data processing stage

[0142] The data decryption unit receives the ciphertext C from the IoT transmission module and decrypts it using the same key used for encryption. Assume the ciphertext is a binary data string encrypted using AES, with the key K being "abcdefghijklmnop" (in practice, the key is more complex and its length complies with encryption standards). After decryption, the plaintext data M is obtained. In this case, M is presented as multidimensional data, such as {T = 42°C, P = 2.3 MPa, D = 3 mm, f = 45 Hz, A = 0.08 mm}.

[0143] The machine learning model training unit constructs an MLP model, which includes an input layer, a hidden layer, and an output layer. The input layer receives the preprocessed multidimensional data feature vector. Assuming that the input layer of the GIL expansion joint operation status prediction model corresponds to 5 features (temperature, pressure, displacement, vibration frequency, and vibration amplitude), the number of neurons in the input layer is 5. The hidden layer is set to 3 layers, with the number of neurons in the first hidden layer set to 15, the second layer to 10, and the third layer to 5. The activation function f uses the ReLU function, that is, f(x) = max(0,x). The weight matrix W1 (size 5×15) from the input layer to the hidden layer is randomly initialized before the start of training. Assume that the initial value is:

[0144]

[0145] The bias vector b1 (length 15) is also randomly initialized as follows:

[0146] b1=[0.1,-0.2,0.3,-0.1,0.2,-0.3,0.4,-0.2,0.1,-0.4,0.3,-0.2,0.2,-0.1].

[0147] After the nonlinear transformation of the hidden layer, the input data is feature extracted and transformed. The output layer obtains the predicted running status label through linear transformation. Assume that the weight matrix W2 (size is 5×1) from the hidden layer to the output layer is [0.2, -0.3, 0.1, -0.2, 0.3] T , the bias vector b2 is 0.1.

[0148] The MLP model is trained using historical and real-time data. Assume that the historical data contains 5,000 samples, and the real-time data is the currently collected 100 samples. During training, the backpropagation algorithm is used to iteratively update the model parameters W1, b1, W2, and b2 based on the concept of gradient descent. The mean squared error (MSE) is used as the loss function. In each iteration, the gradient of the prediction error with respect to each parameter is calculated. For example, the gradient of the weight matrix is calculated using the chain rule, as follows:

[0149]

[0150] Update the parameters in the opposite direction of the gradient. The formula is as follows:

[0151]

[0152] Where α is the learning rate, which is initially set to 0.01. After multiple iterations of training, the model parameters are continuously adjusted to minimize the prediction error.

[0153] The model evaluation and optimization unit evaluates the trained model using a test dataset containing 800 samples. Evaluation metrics include model prediction accuracy, recall, F1 value, and mean squared error. Assume that in the test dataset, the model correctly predicts 650 samples as positive (device operating normally), 100 samples as negative (device operating abnormally), 30 samples as positive (FP), and 20 samples as negative (FN). Each metric is calculated using the following formula:

[0154] By substituting the data into the calculation, the model prediction accuracy is 0.9375, the recall rate is 0.9701, the precision rate is 0.9559, the F1 value is 0.963, and the mean square error is 0.015;

[0155] Based on system requirements and actual application scenarios, set thresholds for model evaluation metrics. Assume the accuracy threshold is 0.9, the recall threshold is 0.9, the F1 value threshold is 0.9, and the mean squared error threshold is 0.02. The current model metrics meet the requirements. If not, optimize them. For example, if the model's accuracy, recall, and F1 value are low, and the mean squared error is high, increase the number of hidden layers or adjust the number of neurons in the hidden layers to improve the model's expressiveness. If the model overfits (performing well on the training set but poorly on the test set), reduce the number of hidden layers or use regularization to reduce model complexity. For hyperparameter tuning, if model performance is poor, use an adaptive learning rate algorithm (such as the Adam algorithm) to automatically adjust the learning rate based on the model's training progress. Use an early stopping strategy to select an appropriate number of iterations by observing the performance changes on the training and test sets. Use cross-validation to adjust the regularization parameter and select an appropriate value to control the strength of the regularization. The adjusted model is evaluated again using the test dataset, and the values of various evaluation indicators are calculated. The performance indicators of the models before and after the adjustment are compared to determine whether the optimization measures are effective. If the adjusted model still does not meet the requirements, the above steps of model structure adjustment and hyperparameter adjustment are repeated to continue optimizing and evaluating the model, forming a closed-loop model optimization mechanism.

[0156] The fault prediction and early warning unit inputs real-time collected and processed data into a trained model to generate a prediction of the device's current operating status. Assume that the model predicts the currently collected data and outputs a predicted value of 0.9 (a value close to 1 indicates normal device operation, while a value close to 0 indicates abnormal device operation). Based on preset fault thresholds and rules, if the predicted value is less than 0.5, it is considered abnormal. At this point, a predicted value of 0.9 exceeds the threshold, indicating normal device operation. If the predicted value is less than 0.5, the early warning mechanism is triggered, and system prompts (such as text messages and in-site messages) notify relevant personnel to take appropriate measures to make adjustments.

[0157] The data processed by the data processing module is received by the cloud storage interaction unit. Assuming that the processed data is normal (including the operating status label), the cloud storage interaction unit performs adaptability processing on the data and uploads the data to the designated storage location according to the cloud storage platform API. The local storage management unit stores data according to timestamp and data type, such as storing the data in the monitoring data folder of the day, and updates the data index. The data classification index unit assigns a unique index item to the data, assuming the index item is "20241001_001", and stores it in the index table. The data backup and synchronization unit backs up the local server's data of the day to the cloud storage platform at the preset time node of 2 am every day. At the same time, the data changes of the cloud storage platform and the local server are monitored in real time. If a piece of data on the local server is updated, it is synchronized to the cloud storage platform in time, and vice versa to ensure the integrity of the data on both ends.

[0158] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. The multi-dimensional data management system for GIL expansion joints based on the Internet of Things is characterized by: It includes a data acquisition module, an Internet of Things transmission module, a data processing module and a data storage module; the data acquisition module is used to collect the temperature, pressure, displacement and vibration data of the GIL expansion joint; the Internet of Things transmission module is used to encrypt and transmit the multidimensional data collected by the data acquisition module; the data processing module establishes a GIL expansion joint operation status prediction model based on the transmitted data; the data storage module adopts distributed storage technology to store the processed data in the cloud storage platform and the local server at the same time.

2. The multidimensional data management system for GIL expansion joints based on the Internet of Things according to claim 1, characterized in that: The data acquisition module includes a sensor unit, a sampling control unit, a data conditioning unit, a data cache unit and a communication interface unit; The sensor unit consists of a temperature sensor, a pressure sensor, a displacement sensor, and a vibration sensor. These sensors are deployed on the GIL expansion joint based on the structural characteristics and operating characteristics of the GIL expansion joint to collect temperature T, pressure P, displacement D, vibration frequency f, and vibration amplitude A in real time during the operation of the GIL expansion joint. These sensors convert physical quantities into electrical signals and digital signals, and transmit them to the data processing module for preprocessing. The sampling control unit is responsible for sampling according to the preset sampling period T s Collect and control sensor data; sampling period T s The value range is 1-60 seconds; among them, the sampling frequency f s It is inversely proportional to the sampling period; the formula is as follows: The data conditioning unit is used to amplify the weak signal output by the sensor to a preset amplitude range, remove noise interference in the signal by filtering, and then perform level conversion to make the signal meet the input requirements of subsequent equipment; The data cache unit is responsible for temporarily storing the collected and conditioned data. During the data transmission process, a first-in-first-out storage method is adopted. When the data transmission rate does not match the acquisition rate, the data storage unit is responsible for buffering the data transmission rate to prevent data loss; The communication interface unit is responsible for transmitting the collected and processed data to the Internet of Things transmission module.

3. The multi-dimensional data management system for GIL expansion joints based on the Internet of Things according to claim 1, characterized in that: The Internet of Things transmission module includes a data encryption unit, a communication protocol adaptation unit, a wireless communication unit and a data retransmission and error correction unit; The data encryption unit is used to encrypt the collected multidimensional data using the AES symmetric encryption algorithm before data transmission; the encryption process is as follows, the plaintext data is M, the encryption key is K, and the ciphertext C is obtained after the AES encryption algorithm is processed, and the formula is as follows: C=AES K (M); The communication protocol adaptation unit is responsible for adapting the encrypted data to the corresponding communication protocol, encapsulating the data, adding protocol headers and verification information, and ensuring that the data can be correctly transmitted in 5G and NB-IoT networks; The wireless communication unit is responsible for sending the adapted data frames to the network through the corresponding wireless frequency band and communication mode according to the selected communication technology, and receiving feedback information from the data processing module; The data retransmission and error correction unit detects errors by adding check information to the data frame. When an error is detected, a retransmission request is sent to the sender, requesting the data to be resent. At the same time, the unit also uses forward error correction technology to add redundant information to the data, so that the receiving end can correct the error by itself when receiving partially erroneous data, thereby reducing the number of retransmissions.

4. The multi-dimensional data management system for GIL expansion joints based on the Internet of Things according to claim 1, characterized in that: The data processing module includes a data decryption unit, a machine learning model training unit, a model evaluation and optimization unit, and a fault prediction and early warning unit; After receiving the encrypted data transmitted by the IoT module, the data decryption unit first uses the same key K as used for encryption to decrypt the ciphertext C to obtain the plaintext data M. The decryption formula is as follows: The machine learning model training unit uses the data transmitted by the Internet of Things transmission module to train the machine model and establish a GIL expansion joint operation status prediction model; the model uses the multi-layer perceptron MLP in the neural network algorithm for model training; The model evaluation and optimization unit uses a test data set to evaluate the trained GIL expansion joint operation status prediction model; then adjusts the model structure and model hyperparameters based on the evaluation results; and finally uses the test data set to evaluate and optimize the adjusted model; The fault prediction and warning unit is responsible for inputting the real-time collected and processed data into the trained model to obtain the prediction results of the current operating status of the equipment; then, based on the preset fault thresholds and rules, it determines whether the prediction results exceed the normal range. If so, the warning mechanism is triggered and the relevant personnel are notified through system prompts to take appropriate measures to make adjustments.

5. The multi-dimensional data management system for GIL expansion joints based on the Internet of Things according to claim 4 is characterized in that: The machine learning model training unit uses the data transmitted by the Internet of Things transmission module to train the machine model, and establishing a GIL expansion joint operation status prediction model includes the following contents: First, the network structure of the MLP model is constructed, including the input layer, hidden layer, and output layer. The input layer receives the preprocessed multidimensional data feature vector x, and the hidden layer extracts and transforms the input data through nonlinear transformation. The nonlinear transformation formula is as follows: h=f(W1x+b1) Where W1 is the weight matrix from the input layer to the hidden layer, b1 is the bias vector, and f is the activation function, which is used to introduce nonlinear factors and enhance the model's expressiveness. The output layer obtains the predicted running status label through linear transformation. The linear transformation formula is as follows: y=W2h+b2 Where W2 is the weight matrix from the hidden layer to the output layer, and b2 is the bias vector; The MLP model is then trained using historical and real-time data, and the model parameters W1, b1, W2, and b2 are iteratively updated using the backpropagation algorithm. The backpropagation algorithm is based on the idea of gradient descent, which calculates the gradient of the prediction error with respect to each parameter and updates the parameters in the opposite direction of the gradient to minimize the prediction error. Finally, after multiple iterations of the model through the back-propagation algorithm, when the prediction error of the GIL expansion joint operation status prediction model is lower than the set threshold, the model construction and training is completed.

6. The multi-dimensional data management system for GIL expansion joints based on the Internet of Things according to claim 4, characterized in that: The model evaluation and optimization unit uses a test data set to evaluate the trained GIL expansion joint operation status prediction model; then adjusts the model structure and model hyperparameters based on the evaluation results; and finally uses the test data set to evaluate and optimize the adjusted model, including the following: First, the model evaluation and optimization unit uses the test dataset to evaluate the trained model. The evaluation indicators include model prediction accuracy, recall rate, F1 value and mean square error. The model prediction accuracy is the ratio of the number of samples correctly predicted by the model to the total number of samples. The calculation formula is as follows: Among them, TP represents the number of samples correctly predicted as positive, TN represents the number of samples correctly predicted as negative, FP represents the number of samples incorrectly predicted as positive, and FN represents the number of samples incorrectly predicted as negative. Recall is the ratio of the number of samples correctly predicted by the model to the actual number of positive samples. The calculation formula is as follows: The recall rate reflects the model's ability to identify positive samples; The F1 value is the harmonic mean of precision and recall, which is used to comprehensively measure the performance of the model. The calculation formula is as follows: The calculation formula for Precision is as follows: The range of F1 is {0-1}. The closer F1 is to 1, the better the performance of the model. The mean square error is used to measure the average error between the model's predicted value and the true value. The calculation formula is as follows: Where n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample; Secondly, according to the set threshold, the calculated evaluation index is compared with the threshold to determine whether each index of the model reaches the set threshold range; Then, based on the evaluation results, when the model performance does not reach the threshold range, the model structure and hyperparameters are adjusted to optimize the model; Regarding model structure adjustments, when the model's accuracy, recall, and F1 value are lower than the threshold, and the mean square error is higher than the threshold, the number of hidden layers is increased to improve the model's expressiveness. When the model overfits, the number of hidden layers is gradually reduced to reduce the model's complexity. After adjusting the number of hidden layers, the model needs to be re-evaluated. For hyperparameter adjustment, hyperparameters include learning rate, number of iterations, and regularization parameters. For learning rate adjustment, when the various indicators in the model performance do not reach the set threshold, the model learning rate is adjusted using the adaptive learning rate method. This algorithm dynamically adjusts the learning rate based on the first-order moment estimate and second-order moment estimate of the parameter gradient during the model training process. The number of iterations is determined by observing the performance changes of the model on the training set and the test set. The performance change of the model refers to the difference between the various indicators of the model and the performance in the training set and the test set. The early stopping strategy is used to select the number of iterations to adjust the model based on the performance change of the model. The number of iterations is determined based on the preset performance evaluation window. When the performance of the test set does not change within multiple consecutive evaluation windows, the training is stopped and the number of iterations at that time is recorded. To determine the regularization parameter, the k-fold cross-validation method is used to select the appropriate regularization parameter value. The training data is divided into k mutually non-overlapping subsets, and k-1 subsets are used as training sets and 1 subset is used as validation sets each time. For different regularization parameter values, the average performance index of the model is calculated in k validations, and the regularization parameter value that optimizes the average performance is selected to control the model complexity and prevent overfitting. Finally, the adjusted model is evaluated again using the test dataset, the values of various evaluation indicators are calculated, the performance indicators of the models before and after adjustment are compared, and the effectiveness of the optimization measures is judged; if the adjusted model still does not meet the requirements, repeat the above steps of model structure adjustment and hyperparameter adjustment for continuous iterative optimization.

7. The multi-dimensional data management system for GIL expansion joints based on the Internet of Things according to claim 1, characterized in that: The data storage module includes a cloud storage interaction unit, a local storage management unit, a data classification index unit and a data backup and synchronization unit; The cloud storage interaction unit is responsible for docking and data interaction with the cloud storage platform. During the data upload phase, the unit receives data processed by the data processing module, performs adaptability processing on the data, and then transmits the data to the storage location specified by the cloud storage receipt according to the API provided by the cloud storage platform. During the data download phase, based on data requests for the operating status and fault diagnosis of the GIL expansion joint from other modules, the unit initiates a data read request to the cloud storage platform through the API, obtains the required data, and returns it to the requesting module. The local storage management unit is responsible for receiving data from the data processing module, storing the data in the corresponding storage location according to the established classification rules of the GIL expansion joint, and updating the data index; when data is requested, it locates the storage location of the data on the local server according to the index, reads the data and returns it to the request module; The data classification index unit classifies the data according to the timestamp and data type of the GIL expansion joint data and the preset classification rules before the data is stored; Then, according to the category and content of the data, a unique index item is assigned to each piece of data and the index item is stored in the index table; In the data query phase, based on the user and other modules' query conditions for the operating data of a specific part and a specific time range of the GIL expansion joint, the time range and data type information in the query conditions are parsed, and the index table is used to locate the data storage location that meets the conditions for data query; The data backup and synchronization unit is responsible for backing up and synchronizing data between the cloud storage and the local server; The data of the local server is backed up to the cloud storage platform according to the preset time nodes. At the same time, the data changes of the cloud storage platform and the local server are monitored in real time. When the data on one side is updated, the updated data is synchronized to the other side to maintain the integrity of the data on both ends.

8. A multi-dimensional data management method for GIL expansion joints based on the Internet of Things, characterized by: The following steps are involved: S1, the sensor unit is deployed according to the characteristics of the GIL expansion joint, collecting temperature, pressure, displacement and vibration data in real time, which is then processed by the sampling control, conditioning and cache units, and finally transmitted to the IoT transmission module by the communication interface unit; S2. The data encryption unit uses the AES algorithm to encrypt the data. The communication protocol adaptation unit adapts it to the corresponding protocol and encapsulates it. Then the wireless communication unit sends the data frame. The data retransmission and error correction unit ensures the transmission accuracy. S3: The data decryption unit decrypts the encrypted data. The machine learning model training unit builds and trains a GIL expansion joint operation status prediction model. The model evaluation and optimization unit evaluates and adjusts the model. The fault prediction and warning unit determines whether to trigger a warning based on the model prediction results. S4. The cloud storage interaction unit connects with the cloud platform to complete data upload and download, and the local storage management unit stores and reads data according to the rules; The data classification and indexing unit classifies data and creates indexes, and the data backup and synchronization unit ensures the consistency of cloud storage and local data.

Images