Cable fault monitoring method and device based on multivariate time series data fusion

Abstract

The invention discloses a cable fault monitoring method and device based on multivariate time series data fusion, and relates to the technical field of cable multi-parameter fault monitoring, and the method comprises the steps: obtaining key time series characteristics representing the health state of a cable; dividing health state grades of the cable according to the electrical parameters and the environmental parameters, performing model training by taking the key time sequence characteristics and the health state grades as input parameters based on a multivariate neural network algorithm, and constructing a health assessment model; predicting the real-time electrical parameters and the environment time sequence parameters of the to-be-tested cable based on the health assessment model, and outputting a health state grade result of the to-be-tested cable; responding to the health state grade result and sending out a corresponding fault monitoring alarm instruction; the problem that the fault judgment condition is unreliable due to the fact that the time sequence dynamic characteristics of monitoring data are not considered in a monitoring method is solved, and the reliability, accuracy and detection efficiency of cable fault monitoring are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cable multi-parameter fault monitoring, and in particular to a cable fault monitoring method and device based on multivariate time series data fusion. Background Art

[0002] With the continuous expansion of the coverage of power networks, the role of power cables in power supply systems has become increasingly significant. Once a power cable fails, it will not only have a serious impact on the power supply system, but may also cause large-scale power outages and even safety accidents. At present, the monitoring of power cables is mainly focused on the temperature detection of cable joints. However, the monitoring of the surrounding environment is often neglected during the implementation process. For example, there may be water accumulation in the underground environment, leading to the risk of leakage. At the same time, the presence of a variety of harmful gases also brings safety hazards to manual maintenance, resulting in the inability of existing cable joint monitoring devices to fully monitor these environmental factors and meet the monitoring needs of complex underground environments. In addition, the conventional technology of cable fault monitoring uses a threshold comparison method to compare the collected monitoring data with the threshold to determine whether the cable has failed. However, the cable monitoring data is essentially a time series. If a single threshold comparison is used, there will be occasional errors. Such limitations make it difficult to accurately evaluate the changes in the actual working conditions of the cable over time.

[0003] Chinese patent, publication number: CN116989844A, publication date: November 3, 2023, discloses a cable well monitoring system and a cable well monitoring method; the cable well temperature information is obtained through the first temperature detection module, the cable well humidity information is obtained through the humidity detection module, the cable well water level information is obtained through the water level detection module, and the data analysis module performs a first analysis based on the cable well temperature information, the cable well humidity information and the cable well water level information, and sends the analysis results to the monitoring terminal to identify cable well abnormalities. However, the invention uses threshold comparison to obtain the analysis results when performing the first analysis. Since the collected monitoring information is a continuous time series, the threshold comparison has a single limitation and cannot adapt to the dynamic changes of the data. It may not accurately reflect the actual situation of the data, and it is difficult to ensure the accuracy of the cable fault judgment results. Summary of the invention

[0004] The purpose of the present invention is to address the problem that conventional cable fault monitoring methods do not take into account the timing dynamic characteristics of monitoring data, resulting in unreliable fault judgment conditions; a cable fault monitoring method and device based on multivariate time series data fusion are proposed, and a health assessment model is trained using the environmental timing parameters and electrical timing parameters of the collected cable through a multivariate neural network algorithm, and the real-time electrical timing parameters and environmental timing parameters of the cable are analyzed based on the trained model, the cable operation status category is output, the cable fault monitoring is completed, and the fault alarm is executed in real time in response to the fault monitoring result; the problem that conventional cable fault monitoring methods do not take into account the timing dynamic characteristics of monitoring data, resulting in unreliable fault judgment conditions, is overcome, and the reliability, accuracy and detection efficiency of cable fault monitoring are significantly improved.

[0005] To solve the above technical problems, according to a first aspect of an embodiment of the present invention, a cable fault monitoring method based on multivariate time series data fusion is provided, comprising the following steps: S1. Obtain electrical parameters of the cable and environmental parameters of the corresponding cable well, preprocess and extract features of the electrical parameters and the environmental parameters, and obtain key timing features that characterize the health status of the cable; S2, classifying the health status level of the cable according to the electrical parameters and the environmental parameters, using the key timing characteristics and the health status level as input parameters for model training based on a multivariate neural network algorithm, and constructing a health assessment model; S3, predicting the real-time electrical parameters and environmental timing parameters of the cable to be tested based on the health assessment model, and outputting the health status level result of the cable to be tested; S4. In response to the health status level result, a corresponding fault monitoring alarm instruction is issued.

[0006] In this scheme, by extracting the key time series features that characterize the health status of the cable, the core information reflecting the health status of the cable can be intuitively obtained, which helps the health assessment model to efficiently learn the corresponding characteristics of various operating states of the cable; at the same time, it reduces the data dimension and significantly improves the computational efficiency and accuracy of the health assessment model, as well as the prediction performance and generalization performance of the model; by dividing the health status level of the cable, the health assessment model can comprehensively learn the relationship between the electrical parameters and environmental parameters of the cable under various operating states, automatically identify the key characteristics of the cable under different operating states, and realize that when the evaluation model analyzes real-time data, it directly outputs the classification results of the operating status of the cable to be tested, realizing intelligent evaluation and fault classification of the cable operating status, that is, simultaneously realizing real-time cable fault detection and health assessment, significantly improving the efficiency and accuracy of cable fault monitoring.

[0007] Preferably, S1 comprises the following sub-steps: S11, continuously collecting the environmental parameters and the electrical parameters based on the sensor network, the environmental parameters at least including environmental temperature parameters, humidity parameters and air pressure parameters, and the electrical parameters at least including cable current sequence and cable temperature sequence; S12, conditioning and converting the environmental parameters and the electrical parameters and performing data standardization processing to obtain a target parameter sequence; S13. Using the target parameter sequence as the input of a convolutional neural network algorithm, extracting the local spatiotemporal dependency of the target parameter sequence through convolution and pooling, and obtaining key timing features.

[0008] In this scheme, the collected electrical parameters and environmental parameters are standardized to ensure the quality and consistency of the input data for later construction of the health assessment model. The target parameter sequence obtained by the standardized processing is processed by the convolutional neural network algorithm, which can capture the local spatiotemporal dependencies of the data and convert the high-dimensional raw data into a more compact feature representation, which helps to reduce the computational complexity of the model while retaining key information, helping to improve the accuracy of the multivariate neural network algorithm, further improve the accuracy and reliability of the health assessment model, and achieve accurate and efficient cable fault monitoring.

[0009] Preferably, the key timing features include at least: The historical changes in the current inside the cable, the fluctuation range of the cable temperature, and the environmental changes in the cable well.

[0010] Preferably, in S2, the step of classifying the health status of the cable according to the electrical parameters and the environmental parameters comprises the following steps: S21. Divide the cable status into a level sequence according to the electrical parameter characteristics and environmental parameter characteristics of the cable under different operating conditions; S22. Establish corresponding status level labels based on the level sequence, obtain the status level label sequence, and complete the division of the health status level.

[0011] In this scheme, by dividing the levels of cable status and establishing the corresponding status level label sequence, a core learning object is provided for building a health assessment model, which helps the model learn and distinguish the characteristic differences between each state; in the model training process, historical data with health level labels are used, and supervised learning methods are used for multiple rounds of iterative training to ensure that the model can accurately classify the health status of the cable, so that the model can more efficiently and accurately capture the changing laws of the cable health status, without calculating the similarity of each data to define the type value, which helps the model to be effectively applied to cable monitoring scenarios in various complex environments, and can make accurate prediction responses in the face of real-time data, output the status level label of the test cable, and thus judge the current health status of the test cable according to the correspondence between the status level label and the status type, significantly improving the adaptability and generalization ability of the model.

[0012] Preferably, in S2, the multivariate neural network algorithm is used to perform model training using the key time series features and the health status level as input parameters to obtain a health assessment model, including the following steps: S23, mapping the key time series features and the state level labels and configuring model parameters based on the LSTM neural network algorithm; S24, based on the mapping relationship between the key time series features and the state level labels and the configuration parameters, the state level labels are predicted, the error sequence between the prediction results and the actual state level labels is calculated using the cross entropy loss function, and the LSTM network is iteratively trained with the error sequence to construct an initial health assessment network model; S25. Calculate the attention score of the time step of the model input parameter based on the attention mechanism, dynamically adjust the model weight according to the calculation result, and construct the health assessment network model.

[0013] Preferably, the S23 comprises the following steps: Input the key timing features and the state level labels into the LSTM neural network model, retrieve the key timing features one by one through the LSTM neural network, map them to the matching state level labels, and establish a mapping relationship between the key timing features and the state level labels; The LSTM neural network model is iteratively trained according to the mapping relationship to obtain an initial health assessment network model.

[0014] Preferably, executing S25 includes the following sub-steps: The attention score is calculated by using the time step and model weight of the initial health assessment network model through an attention mechanism; The attention score is summed with the attention weight to obtain an attention score vector, and the attention score vector is used as the input of the fully connected layer of the initial health assessment network model. The weights of different time steps and input parameters of the model are dynamically adjusted to construct the health assessment network model.

[0015] In this scheme, the long-term and short-term dependencies in the collected time series data are processed by the long short-term memory network LSTM, which can effectively capture the historical change trends of the cable joint parameters, so that the evaluation model can accurately simulate the long-term operation mode of the cable system, especially in complex scenarios involving long-term data dependence. It helps to predict the operating status of the cable according to the real-time operating parameters of the cable, give warnings to the staff, and make corresponding cable repair and maintenance preparations in advance; the attention mechanism can dynamically adjust the weights according to the importance of data at different time steps. The attention mechanism allows the model to focus more on changes at key time points, so that the model will not be disturbed by irrelevant data when processing long time series, thereby improving the accuracy of fault prediction.

[0016] According to another aspect of an embodiment of the present invention, a cable fault monitoring device based on multivariate time series data fusion is provided, comprising: A sensor detection module is used to collect electrical parameters of the cable and environmental parameters of the corresponding cable well, and transmit the electrical parameters and environmental parameters to the main control module after pre-processing; The main control module is used to store the parameters collected by the sensor detection module and to perform data transmission and control between the modules; the monitoring and alarm module is used to retrieve the parameters stored in the main control module, perform feature extraction, classify the health status of the cable, and build a health assessment model, and identify the real-time cable operation status and predict faults, and issue an audible and visual alarm signal based on the identified faults; Display module: used to display the parameters collected in real time by the sensor detection module, the historical monitoring data stored by the main control module, and the alarm log; Power supply module: used to obtain cable energy and provide working power to the cable fault monitoring device.

[0017] Preferably, the sensor detection module comprises: Current sensor detection unit, used to collect current parameters of cable connectors in real time; Gas sensor detection unit, used to collect gas concentration data and air pressure data of the cable well; A water immersion sensor detection unit is used to collect water immersion conditions in the cable well; A first temperature sensor detection unit is used to collect temperature and humidity data of the environment in the cable well; The second temperature sensor detection unit is used to collect cable surface temperature data.

[0018] Preferably, the main control module includes: A microcontroller, used to manage the real-time collection and processing of data from each sensor detection unit through a multi-tasking scheduling mechanism; A clock module, used to manage and maintain the time base of the cable fault monitoring device and the synchronization status of each module; The storage module is used to store the parameters collected by the sensor detection module and the alarm log generated by the monitoring alarm module.

[0019] In this solution, the temperature, humidity, water level, harmful gases, ground current and other parameters are collected by the sensor module, and sent to the main control module after data processing and conversion. Through dynamic memory allocation and management technology, timely response and processing of sensor data are ensured, thereby improving the operation efficiency of the system; on the one hand, the main control module sends the display command to the display module to realize the display of all parameters in the display module; on the other hand, the main control module stores the data to ensure the security and reliability of the data, and sends the data to the monitoring and alarm module for abnormal detection. When an abnormal situation is found, an alarm signal is issued through the monitoring and alarm module to remind the operator to take corresponding measures, thereby realizing early warning of cable joint failure and ensuring the safety and reliability of the cable system; the power supply module obtains energy from the cable through the CT energy extraction module and stores it in the battery, so that the device can operate stably for a long time in the underground environment. Even in the case of power outage, the device can still work normally to ensure the stable operation of the cable detection work.

[0020] Beneficial effects of the present invention: 1. Through the time series LSTM network, CNN network and attention mechanism, combined with multiple time series data such as environmental temperature, humidity, and current, a health assessment model is constructed. A comprehensive assessment of the cable status is achieved by integrating multiple data sources and advanced models, providing decision-making basis for staff, realizing real-time cable fault detection and health assessment, and significantly improving the efficiency and accuracy of cable fault monitoring; 2. By extracting the key time series features that characterize the health status of the cable, the core information reflecting the health status of the cable can be intuitively obtained, which helps the health assessment model to efficiently learn the corresponding characteristics of various operating states of the cable; 3. By dividing the cable status into levels and establishing the corresponding status level label sequence, the model can be helped to learn and distinguish the characteristic differences between the states, ensuring that the model can accurately classify the health status of the cable, so as to more efficiently and accurately capture the changing laws of the cable health status and improve the reliability and robustness of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings. The drawings are only for the purpose of illustrating preferred embodiments and are not to be considered as limiting the present invention. Also, the same reference symbols are used throughout the drawings to represent the same parts.

[0022] Figure 1 This is a flow chart of a cable fault monitoring method based on multivariate time series data fusion according to this embodiment.

[0023] Figure 2 This is an architecture diagram of a cable fault monitoring device based on multivariate time series data fusion according to this embodiment.

[0024] Figure 3 The figure is a functional flow diagram of a cable fault monitoring device of this embodiment.

[0025] Figure 4 This is a physical diagram of the deployment of a cable fault monitoring device of this embodiment. DETAILED DESCRIPTION

[0026] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0027] Example 1: Figure 1 As shown, a cable fault monitoring method based on multivariate time series data fusion includes steps S1-S4, wherein: S1. Obtain the electrical parameters of the cable and the environmental parameters of the corresponding cable well, preprocess and extract features of the electrical parameters and environmental parameters, and obtain key timing features that characterize the health status of the cable.

[0028] Specifically, S1 includes the following sub-steps: S11, continuously collecting environmental parameters and electrical parameters based on the sensor network, the environmental parameters at least including environmental temperature parameters, humidity parameters and air pressure parameters, and the electrical parameters at least including cable current sequence and cable temperature sequence; S12, conditioning and converting the environmental parameters and electrical parameters and performing data standardization processing to obtain a target parameter sequence; S13. Use the target parameter sequence as the input of the convolutional neural network algorithm, extract the local spatiotemporal dependencies of the target parameter sequence through convolution and pooling, and obtain key timing features.

[0029] In this embodiment, the collected electrical parameters and environmental parameters are standardized to ensure the quality and consistency of the input data for later construction of the health assessment model. The target parameter sequence obtained by the standardized processing is processed by the convolutional neural network algorithm, which can capture the local spatiotemporal dependencies of the data and convert the high-dimensional raw data into a more compact feature representation, which helps to reduce the computational complexity of the model while retaining key information, helping to improve the accuracy of the multivariate neural network algorithm, further improve the accuracy and reliability of the health assessment model, and achieve accurate and efficient cable fault monitoring.

[0030] Specifically, the key timing characteristics include at least: The historical changes in the current inside the cable, the fluctuation range of the cable temperature, and the environmental changes in the cable well.

[0031] S2. The health status level of the cable is divided according to the electrical parameters and environmental parameters. The key timing characteristics and health status level are used as input parameters for model training based on the multivariate neural network algorithm to build a health assessment model.

[0032] Specifically, in S2, the health status level of the cable is divided according to the electrical parameters and the environmental parameters, including the following steps: S21, dividing the level sequence of the cable status according to the electrical parameter characteristics and the environmental parameter characteristics under different operating states of the cable; S22, establishing the corresponding status level label based on the level sequence, obtaining the status level label sequence, and completing the division of the health status level.

[0033] As an implementation method, the cable status is divided into different levels according to the electrical parameters and environmental parameters of the cable under different operating conditions, and each level is identified by a unique label. For example, the data sequence of the combination of electrical parameters and environmental parameters is X = {x1, x2, ..., x n}, x n The electrical parameters of the cable and the environmental parameters of the cable well under the same time and the same operating state are represented. According to the X sequence, the cable state is divided into k levels, and each level is represented by a label y, where y∈{1,2,...,k}, and each label corresponds to a state characteristic of the cable. For example, 1 represents a normal state, 2 represents an abnormal state, 3 represents a slight fault, 4 represents a moderate fault, and 5 represents a serious fault. The mapping relationship between the parameters of the cable and the cable well and the state level label is expressed as {(X1,y1),(X2,y2),...(X n ,y n )}.

[0034] In this embodiment, by dividing the levels of cable status and establishing corresponding status level label sequences, a core learning object is provided for building a health assessment model, helping the model to learn and distinguish the characteristic differences between various states; in the model training process, historical data with health level labels are used, and supervised learning methods are used for multiple rounds of iterative training to ensure that the model can accurately classify the health status of the cable, so that the model can more efficiently and accurately capture the changing laws of the cable health status, without calculating the similarity of each data to define the type value, which helps the model to be effectively applied to cable monitoring scenarios in various complex environments, and can make accurate prediction responses in the face of real-time data, output the status level label of the test cable, and thus judge the current health status of the test cable according to the corresponding relationship between the status level label and the status type, significantly improving the adaptability and generalization ability of the model.

[0035] Specifically, in S2, based on the multivariate neural network algorithm, key time series features and health status levels are used as input parameters for model training to obtain a health assessment model, including the following steps: S23, mapping key time series features and state level labels and configuring model parameters based on LSTM neural network algorithm; S24, predicting state level labels based on the mapping relationship between key time series features and state level labels and configuration parameters, using cross entropy loss function to calculate the error sequence between the prediction result and the actual state level label, and iteratively training the LSTM network with the error sequence to construct an initial health assessment network model; S25. Calculate the attention score of the time step of the model input parameter based on the attention mechanism, dynamically adjust the model weight according to the calculation result, and build a health assessment network model.

[0036] Specifically, S23 includes the following steps: The key time series features and state level labels are input into the LSTM neural network model, and the key time series features are retrieved one by one through the LSTM neural network, mapped to the matching state level labels, and a mapping relationship between the key time series features and the state level labels is established; The LSTM neural network model is iteratively trained according to the mapping relationship to obtain the initial health assessment network model.

[0037] Specifically, executing S25 includes the following sub-steps: The attention score is calculated by the attention mechanism using the time step and model weight of the initial health assessment network model; The attention score is summed with the attention weight to obtain the attention score vector, and the attention score vector is used as the input of the fully connected layer of the initial health assessment network model. The weights of different time steps and input parameters of the model are dynamically adjusted to construct the health assessment network model.

[0038] Based on identifying the long-term operation mode of the cable system and building a health assessment model.

[0039] As an implementation method, the calculation formula of the attention score vector c is as follows: e t =tanh(W e ·h t +b e ); Where T represents the time step of the model, W e is the model weight, b e is the bias parameter, h t is the hidden state, e t and a t They are attention scores; The attention score vector c is used as the input of the fully connected layer of the initial health assessment network model, and the output of the model can be expressed as: Both the model weights and bias parameters are converted to be associated with the state level labels. In this way, the attention mechanism can dynamically adjust the weights of different time steps and features, thereby improving the predictive performance of the model.

[0040] In this embodiment, since LSTM has the ability to remember past data states and predict future trends, it can accurately simulate the long-term operation mode of the cable system, especially in complex scenarios involving long-term data dependence. Therefore, by processing the long-term and short-term dependencies in the collected time series data through the long short-term memory network LSTM, it can effectively capture the historical change trends of the cable joint parameters, which is helpful to predict the operating status of the cable according to the real-time operating parameters of the cable, give warnings to the staff, and make corresponding cable repair and maintenance preparations in advance; the attention mechanism is introduced in this embodiment, which can dynamically adjust the weights according to the importance of data at different time steps. The attention mechanism allows the model to focus more on changes at key time points, so that the model will not be disturbed by irrelevant data when processing long time series, thereby improving the accuracy of fault prediction.

[0041] S3. Predict the real-time electrical parameters and environmental timing parameters of the cable to be tested based on the health assessment model, and output the health status level result of the cable to be tested.

[0042] S4. In response to the health status level result, a corresponding fault monitoring alarm instruction is issued.

[0043] As an implementation method, the electrical parameters of the real-time cable and the environmental parameters of the cable well are input into the health assessment model for prediction, and the output result is a status level label. For example, output '1' indicates that the operating status of the cable under test is normal, and output '5' indicates that a serious fault occurs in the cable under test, so that the staff can make corresponding cable repair decisions.

[0044] The beneficial effects of the embodiment: by extracting the key timing features that characterize the health status of the cable, the core information reflecting the health status of the cable can be intuitively obtained, which helps the health assessment model to efficiently learn the corresponding characteristics of various operating states of the cable; at the same time, the data dimension is reduced, and the computational efficiency and accuracy of the health assessment model, as well as the prediction performance and generalization performance of the model are significantly improved; by dividing the health status level of the cable, the health assessment model can comprehensively learn the relationship between the electrical parameters and environmental parameters of the cable under various operating states, automatically identify the key characteristics of the cable under different operating states, and realize that when the evaluation model analyzes real-time data, it directly outputs the operating status classification results of the cable to be tested, realizing intelligent evaluation and fault classification of the cable operating status, that is, simultaneously realizing real-time cable fault detection and health assessment, significantly improving the efficiency and accuracy of cable fault monitoring.

[0045] Embodiment 2, as Figure 2 As shown, a cable fault monitoring device based on multivariate time series data fusion includes: a sensor detection module, which is used to collect electrical parameters of the cable and environmental parameters of the corresponding cable well, and transmit the electrical parameters and environmental parameters to the main control module after pre-processing; The main control module is used to store the parameters collected by the sensor detection module and to transmit and control data between the modules; the monitoring and alarm module is used to retrieve the parameters stored in the main control module, perform feature extraction, classify the health status of the cable, and build a health assessment model, and identify the real-time cable operation status and predict faults, and issue an audible and visual alarm signal based on the identified faults; Display module: used to display the parameters collected in real time by the sensor detection module, the historical monitoring data stored in the main control module, and the alarm log; Power supply module: used to obtain cable energy and provide working power to the cable fault monitoring device.

[0046] Specifically, the sensor detection module includes: Current sensor detection unit, used to collect current parameters of cable connectors in real time; Gas sensor detection unit, used to collect gas concentration data and air pressure data of the cable well; A water immersion sensor detection unit is used to collect water immersion conditions in the cable well; A first temperature sensor detection unit is used to collect temperature and humidity data of the environment in the cable well; The second temperature sensor detection unit is used to collect cable surface temperature data.

[0047] As an implementation method, the main control module adopts a low-power STM32 microcontroller, which is compatible with sensor types of various protocols; the sensor module realizes accurate data collection of various harmful gases through multi-channel ADC, and greatly improves the sampling rate through the DMA mode cyclic scanning method of the ADC; at the same time, the cable surface temperature is collected through the SPI protocol temperature measurement module; the temperature and humidity data of the underground environment are obtained through the single bus protocol temperature and humidity sensor; the data of the water immersion sensor is collected through the RS485 protocol to monitor possible water immersion.

[0048] The display module uses the SPI protocol to drive the LCD display, and the LVGL UI is used to design a UI interface to display historical curves of various sensors and low-power working modes. In this implementation, in order to more effectively coordinate the working efficiency of different modules, the FreeRTOS operating system is used for management, and resources are allocated through multi-task priority design to ensure the real-time and stability of the system.

[0049] The monitoring and alarm module adopts a cable operation status health assessment strategy. Through the time series LSTM network, CNN network and attention mechanism, combined with multiple time series data such as environmental temperature, humidity, and current, a health assessment model is constructed. By integrating multiple data sources and advanced models, a comprehensive assessment of the cable status is achieved, providing a basis for decision-making for staff.

[0050] The power supply module converts electrical energy from the transportation line in the cable well to charge the battery, so as to solve the problem that the embedded equipment is in the cable well for a long time and the battery cannot provide power for a long time.

[0051] Furthermore, if Figure 3 As shown in the figure, the construction process of the FreeRTOS operating system is as follows: after the system is started, multiple tasks are created, including tasks for collecting data from different sensors and LVGL tasks for displaying module parameters. The data collection tasks of each sensor run in a time slice rotation or priority driven manner under the FreeRTOS scheduling mechanism to ensure real-time data collection; in the LVGL task, the graphics library is first initialized and the display driver of the LCD screen is set. This task is given a higher priority to ensure that it can be executed in time in task scheduling, thereby achieving smooth screen refresh and user interface update.

[0052] In this embodiment, the process of LVGL designing the UI interface is as follows: the LVGL task implements the screen refresh operation by regularly calling the LVGL task processing function, and displays the parameters of each module and its change curve in real time on the LCD screen. Due to the non-preemptive nature of LVGL, a mutex is added to the task in this embodiment for protection to ensure the consistency of operation in a multi-tasking environment. In addition, the LVGL task is also responsible for processing user input events. In order to achieve a low-power solution, the device automatically reduces the screen brightness until the display is turned off when there is no operation for a long time, and triggers the screen wake-up through a key event to restore the normal display state.

[0053] Specifically, the main control module includes: A microcontroller, used to manage the real-time collection and processing of data from each sensor detection unit through a multi-tasking scheduling mechanism; A clock module is used to manage and maintain the time base of the cable fault monitoring device and the synchronization status of each module; The storage module is used to store the parameters collected by the sensor detection module and the alarm log generated by the monitoring alarm module.

[0054] Specifically, the monitoring alarm module includes: a sound alarm unit, a light flash unit and a fault diagnosis unit; The sound alarm unit and the light flash unit are used to issue corresponding alarm information according to the diagnosis result of the fault diagnosis unit, that is, the output result of the health assessment model; the fault diagnosis unit is used to identify faults based on the real-time cable electrical parameters and the environmental parameters of the cable well.

[0055] In this embodiment, parameters such as temperature and humidity, water level, harmful gases, and ground current are collected through the sensor module, and sent to the main control module after data processing and conversion. Through dynamic memory allocation and management technology, timely response and processing of sensor data are ensured, thereby improving the operating efficiency of the system; on the one hand, the main control module sends a display command to the display module to display all parameters in the display module; on the other hand, the main control module stores the data to ensure the security and reliability of the data, and sends the data to the monitoring and alarm module for abnormal detection. When an abnormal situation is found, an alarm signal is issued through the monitoring and alarm module to remind the operator to take corresponding measures, thereby achieving early warning of cable joint failures and ensuring the safety and reliability of the cable system; the power supply module obtains energy from the cable through the CT energy extraction module and stores it in the battery, so that the device can operate stably for a long time in the underground environment. Even in the event of a power outage, the device can still work normally to ensure the stable operation of the cable detection work.

[0056] The beneficial effects of the embodiment are as follows: the temperature, humidity, water level, harmful gases, ground current and other parameters are collected by the sensor module, and after data processing and conversion, they are sent to the main control module for storage, and sent to the display module through a display command, so that all parameters can be displayed in the display module, ensuring the real-time, security and effectiveness of the data and avoiding the possibility of data tampering; the data is sent to the monitoring and alarm module to extract the key timing features that characterize the health status of the cable, and to divide the health status level of the cable, so as to ensure that the model efficiently learns the corresponding characteristics of various operating states of the cable, and after completing the construction of the health assessment, it is applied to the cable monitoring scenario to realize intelligent evaluation of the cable operating status and fault classification; when an abnormal situation is found, the monitoring and alarm module sends an alarm signal to remind the operator to take corresponding measures, thereby realizing early warning of cable joint failures, ensuring the safety and reliability of the cable system, and significantly improving the efficiency and accuracy of cable fault monitoring.

[0057] Embodiment 3, as Figure 4 As shown in the figure, the constructed health assessment model is applied to the actual cable fault monitoring device, that is, the trained model is deployed on the STM32 microcontroller to achieve real-time cable fault detection and health assessment. The specific deployment process is as follows: First, export the trained health assessment model to the Keras.h5 format supported by the STM32 microcontroller. Create an STM32 project using the STM32CubeMX software and configure the chip’s system clock, peripherals, and AI expansion functions to ensure that the neural network can run correctly on the STM32. Then specify the model file path and related compression parameters to successfully load the converted LSTM model file into the project.

[0058] Secondly, the running code of the STM32 microcontroller is written to realize the functions of model loading, data input, model reasoning, result saving, data display and transmission, etc. This includes receiving the cable environment data collected by the sensor, inputting it into the health assessment model for prediction, and further processing according to the cable status prediction results output by the model.

[0059] In order to ensure that the model can run efficiently in an embedded environment, it is necessary to optimize the model based on the hardware limitations of the STM32. For example, quantization technology is used to compress the model size to reduce memory usage and computational burden. In addition, the model deployed on the STM32 microcontroller is fully tested and verified to ensure that its performance in the embedded environment is as consistent as possible with that in the training environment.

[0060] Finally, the deployed STM32 microcontroller is integrated into the actual cable fault monitoring device. In order to ensure the long-term stability of the monitoring device, the power consumption and real-time performance of STM32 should be fully considered in practical applications, the operating efficiency of the model should be optimized, and the entire device should be able to stably and efficiently perform the real-time monitoring and prediction tasks of the cable status.

[0061] The above specific embodiments are preferred embodiments of the present invention, and are not intended to limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to the present specific embodiments. All equivalent changes made in accordance with the shape, structure, and method of the present invention are within the protection scope of the present invention.

Claims

1. A cable fault monitoring method based on multivariate time series data fusion, characterized in that: The steps include: S1. Obtain electrical parameters of the cable and environmental parameters of the corresponding cable well, preprocess and extract features of the electrical parameters and the environmental parameters, and obtain key timing features that characterize the health status of the cable; S2, classifying the health status level of the cable according to the electrical parameters and the environmental parameters, using the key timing characteristics and the health status level as input parameters for model training based on a multivariate neural network algorithm, and constructing a health assessment model; S3, predicting the real-time electrical parameters and environmental timing parameters of the cable to be tested based on the health assessment model, and outputting the health status level result of the cable to be tested; S4. In response to the health status level result, a corresponding fault monitoring alarm instruction is issued.

2. The cable fault monitoring method based on multivariate time series data fusion according to claim 1 is characterized in that: The S1 comprises the following sub-steps: S11, continuously collecting the environmental parameters and the electrical parameters based on a sensor network, the environmental parameters at least including an environmental temperature parameter, a humidity parameter and an air pressure parameter, and the electrical parameters at least including a cable current sequence and a cable temperature sequence; S12, conditioning and converting the environmental parameters and the electrical parameters and performing data standardization processing to obtain a target parameter sequence; S13. Using the target parameter sequence as the input of a convolutional neural network algorithm, extracting the local spatiotemporal dependency of the target parameter sequence through convolution and pooling, and obtaining key timing features.

3. The cable fault monitoring method based on multivariate time series data fusion according to claim 1 is characterized in that: The key timing features include at least: The historical changes in the current inside the cable, the fluctuation range of the cable temperature, and the environmental changes in the cable well.

4. The cable fault monitoring method based on multivariate time series data fusion according to claim 1 is characterized in that: In S2, the step of classifying the health status of the cable according to the electrical parameters and the environmental parameters includes the following steps: S21. Classify the cable status according to the electrical parameter characteristics and environmental parameter characteristics of the cable under different operating conditions; S22: Establish corresponding status level labels based on the level sequence, obtain a status level label sequence, and complete the division of the health status levels.

5. The cable fault monitoring method based on multivariate time series data fusion according to claim 4 is characterized in that: In S2, the multivariate neural network algorithm is used to perform model training using the key time series features and the health status level as input parameters to obtain a health assessment model, including the following steps: S23, mapping the key time series features and the state level labels and configuring model parameters based on the LSTM neural network algorithm; S24, based on the mapping relationship between the key time series features and the state level labels and the configuration parameters, the state level labels are predicted, the error sequence between the prediction results and the actual state level labels is calculated using the cross entropy loss function, and the LSTM network is iteratively trained with the error sequence to construct an initial health assessment network model; S25. Calculate the attention score of the time step of the model input parameter based on the attention mechanism, dynamically adjust the model weight according to the calculation result, and construct the health assessment network model.

6. The cable fault monitoring method based on multivariate time series data fusion according to claim 5 is characterized in that: The S23 comprises the following steps: Input the key timing features and the state level labels into the LSTM neural network model, retrieve the key timing features one by one through the LSTM neural network, map them to the matching state level labels, and establish a mapping relationship between the key timing features and the state level labels; The LSTM neural network model is iteratively trained according to the mapping relationship to obtain an initial health assessment network model.

7. The cable fault monitoring method based on multivariate time series data fusion according to claim 5 is characterized in that: Executing S25 includes the following sub-steps: The attention score is calculated by using the time step and model weight of the initial health assessment network model through an attention mechanism; The attention score is summed with the attention weight to obtain an attention score vector, and the attention score vector is used as the input of the fully connected layer of the initial health assessment network model. The weights of different time steps and input parameters of the model are dynamically adjusted to construct the health assessment network model.

8. A cable fault monitoring device based on multivariate time series data fusion, applicable to the cable fault monitoring method based on multivariate time series data fusion as claimed in any one of claims 1 to 7, characterized in that: include: A sensor detection module is used to collect electrical parameters of the cable and environmental parameters of the corresponding cable well, and transmit the electrical parameters and environmental parameters to the main control module after pre-processing; The main control module is used to store the parameters collected by the sensor detection module and to perform data transmission and control between the modules; the monitoring and alarm module is used to retrieve the parameters stored in the main control module, perform feature extraction, classify the health status of the cable, and build a health assessment model, and identify the real-time cable operation status and predict faults, and issue an audible and visual alarm signal based on the identified faults; Display module: used to display the parameters collected in real time by the sensor detection module, the historical monitoring data stored by the main control module, and the alarm log; Power supply module: used to obtain cable energy and provide working power to the cable fault monitoring device.

9. The cable fault monitoring device based on multivariate time series data fusion according to claim 8, characterized in that: The sensor detection module comprises: Current sensor detection unit, used to collect current parameters of cable connectors in real time; Gas sensor detection unit, used to collect gas concentration data and air pressure data of the cable well; A water immersion sensor detection unit is used to collect water immersion conditions in the cable well; A first temperature sensor detection unit is used to collect temperature and humidity data of the environment in the cable well; The second temperature sensor detection unit is used to collect cable surface temperature data.

10. The cable fault monitoring device based on multivariate time series data fusion according to claim 8, characterized in that: The main control module comprises: A microcontroller, used to manage the real-time collection and processing of data from each sensor detection unit through a multi-tasking scheduling mechanism; A clock module, used to manage and maintain the time base of the cable fault monitoring device and the synchronization status of each module; The storage module is used to store the parameters collected by the sensor detection module and the alarm log generated by the monitoring alarm module.

Images