Intelligent pipeline detection system based on Internet of Things technology

Abstract

The invention provides an intelligent pipeline detection system based on the Internet of Things technology, and relates to the field of intelligent pipeline management, and the system collects the time sequence of state parameter data such as a temperature value, a pressure value and a vibration amplitude value of a monitored pipeline through a sensor network, and transmits the time sequence to a cloud data center through the Internet of Things technology. And the cloud data center processes the time sequence data of the state parameters to obtain a pipeline state detection result of the monitored pipeline, and if the pipeline state detection result shows that the pipeline state is abnormal, an early warning signal and a detection result are sent to a manager. Thus, intelligent monitoring of the pipeline state can be achieved, abnormal conditions can be found in time, and then safety and stability of pipeline operation can be ensured.

Description

Technical Field

[0001] The present application relates to the field of pipeline intelligent management, and more specifically, to a pipeline intelligent detection system based on Internet of Things technology. Background Art

[0002] Pipelines are used to transport various hazardous substances, such as oil, natural gas, chemical raw materials, etc. Once a pipeline leaks, it may cause serious consequences such as fire, explosion, environmental pollution, etc., threatening the lives and health of surrounding residents and the safety of the ecological environment. Through testing, hidden dangers such as corrosion and damage in the pipeline that may cause leakage can be discovered in time, and repair or preventive measures can be taken in advance to avoid such catastrophic accidents.

[0003] In traditional pipeline inspection, the detection method may rely on a simple threshold alarm mechanism. This method often lags behind the occurrence of actual problems and cannot detect problems in real time. At the same time, the threshold setting is too simple, which may ignore certain potential abnormal patterns and lead to missed detection. In addition, in traditional pipeline inspection, the correlation and interaction information between different pipeline status parameters are often not deeply mined. Temperature, pressure, vibration and other parameters are only viewed in isolation, ignoring the importance of their mutual influence in judging the pipeline status, which easily leads to misjudgment of the pipeline status.

[0004] Therefore, an optimized pipeline detection system is expected. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a pipeline intelligent detection system based on Internet of Things technology.

[0006] According to one aspect of the present application, a pipeline intelligent detection system based on the Internet of Things technology is provided, which includes: a pipeline state parameter data acquisition module, which is used to obtain a time series of state parameter data of a monitored pipeline collected by a sensor network, wherein the state parameter data includes a temperature value, a pressure value, and a vibration amplitude value; a pipeline state parameter data transmission module, which is used to transmit the time series of the state parameter data of the monitored pipeline to a cloud data center through the Internet of Things technology; a pipeline state parameter data processing module, which is used to process the time series of the state parameter data of the monitored pipeline in the cloud data center to obtain a pipeline state detection result, wherein the pipeline state parameter data processing module includes: a pipeline state parameter data feature encoding unit, which is used to encode the time series of the state parameter data of the monitored pipeline; The pipeline comprises a pipeline temperature and pressure time series feature interaction unit, which is used to perform attention interaction fusion on the temperature time series feature and the pressure time series feature to obtain the pipeline temperature-pressure time series interaction representation; a pipeline multi-source information interaction response unit, which is used to perform fine-grained semantic interaction encoding on the pipeline temperature-pressure time series interaction representation and the vibration amplitude time series feature to obtain the pipeline multi-source information response representation; a pipeline state detection unit, which is used to obtain the pipeline state detection result based on the pipeline multi-source information response representation; and a pipeline state detection result generation module, which is used to send an early warning signal and the pipeline state detection result to the management personnel in response to the pipeline state detection result indicating that the state of the monitored pipeline is abnormal.

[0007] Compared with the prior art, the pipeline intelligent detection system based on the Internet of Things technology provided by the present application collects the time series of state parameter data such as temperature value, pressure value, vibration amplitude value, etc. of the monitored pipeline through the sensor network, and transmits it to the cloud data center using the Internet of Things technology. The time series data of these state parameters are processed in the cloud data center to obtain the pipeline state detection result of the monitored pipeline. If the pipeline state detection result shows that the pipeline state is abnormal, an early warning signal and the detection result are sent to the management personnel. In this way, intelligent monitoring of the pipeline state can be realized, which is conducive to timely detection of abnormal conditions, and thus can ensure the safety and stability of pipeline operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0009] In the attached picture:Figure 1 4 is a system block diagram of a pipeline intelligent detection system based on Internet of Things technology according to an embodiment of the present application.

[0010] Figure 2 The block diagram is a pipeline state parameter data processing module in the pipeline intelligent detection system based on the Internet of Things technology according to an embodiment of the present application.

[0011] Figure 3 1 is a block diagram of a pipeline state parameter data feature encoding unit in a pipeline intelligent detection system based on Internet of Things technology according to an embodiment of the present application.

[0012] Figure 4 The present invention is a block diagram of a pipeline multi-source information interaction response unit in a pipeline intelligent detection system based on Internet of Things technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0013] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0014] In response to the technical problems raised in the above background technology, the present application provides a pipeline intelligent detection system based on Internet of Things technology. Figure 1 : is a system block diagram of a pipeline intelligent detection system based on Internet of Things technology according to an embodiment of the present application. Figure 1 As shown, in the pipeline intelligent detection system 100 based on the Internet of Things technology, it includes: a pipeline state parameter data acquisition module 110, which is used to obtain the time series of the state parameter data of the monitored pipeline collected by the sensor network, wherein the state parameter data includes a temperature value, a pressure value and a vibration amplitude value; a pipeline state parameter data transmission module 120, which is used to transmit the time series of the state parameter data of the monitored pipeline to a cloud data center through the Internet of Things technology; a pipeline state parameter data processing module 130, which is used to process the time series of the state parameter data of the monitored pipeline in the cloud data center to obtain a pipeline state detection result; a pipeline state detection result generation module 140, which is used to send an early warning signal and the pipeline state detection result to the management personnel in response to the pipeline state detection result indicating that the state of the monitored pipeline is abnormal.

[0015] In the embodiment of the present application, the pipeline state parameter data acquisition module 110 is used to obtain the time series of the state parameter data of the monitored pipeline collected by the sensor network, wherein the state parameter data includes temperature value, pressure value and vibration amplitude value. It should be understood that when the pipeline transports various types of hazardous substances (such as oil, natural gas, etc.), the flow of internal substances and the interaction with the pipeline wall will cause heat exchange. Under normal operation, the temperature will be in a relatively stable and reasonable range. Once the pipeline has abnormal conditions such as corrosion, blockage, leakage, etc., the heat transfer process will change, resulting in abnormal temperature increase or decrease. For example, when corrosion and leakage occur locally in the pipeline, the internal substances leak into the external environment, which may cause a sudden change in local temperature. By continuously monitoring the time series of temperature values, this change trend can be captured. The pressure in the pipeline is generally maintained in a specific stable range during normal transportation. When problems such as pipeline rupture and leakage, valve failure or internal blockage occur, the pressure will fluctuate immediately. For example, when the pipeline ruptures, the internal substances leak out and the pressure will drop sharply; and when there is a blockage, the pressure before the blockage point will increase abnormally. By continuously acquiring the time series of pressure values, it is possible to understand in real time whether the flow of materials in the pipeline is smooth and normal, and then determine whether there are any abnormal conditions. Pipeline systems are often accompanied by the operation of power equipment such as pumps and compressors. Under normal circumstances, the vibration amplitude is in a stable and small range. However, when abnormalities such as loose pipe connections, damaged supporting structures, or collisions with foreign objects inside occur, the vibration amplitude will increase significantly. In addition, external interference from the surrounding environment will also affect the vibration of the pipeline. By monitoring the time series of vibration amplitude values, it is possible to distinguish between vibrations caused by normal operation and vibration changes caused by abnormalities, and to promptly detect possible problems in pipeline machinery. In general, by obtaining the time series of temperature values, pressure values, and vibration amplitude values, it is possible to understand the coordinated changes of these parameters at different times, and a comprehensive analysis of them can more accurately determine whether there are abnormalities in the pipeline.

[0016] The following is a detailed description of the specific implementation process of "obtaining the time series of state parameter data of the monitored pipeline collected by the sensor network": First, temperature sensors, pressure sensors and vibration sensors are deployed at key locations of the monitored pipeline to form a sensor network. These sensors should have high precision, high sensitivity and stability to ensure that the collected data is accurate and reliable. For example, the temperature sensor can use a high-precision thermocouple or thermistor sensor to accurately measure the temperature change inside the pipeline; the pressure sensor can select a piezoelectric or piezoresistive sensor suitable for the pipeline pressure range to monitor the pressure fluctuations in the pipeline in real time; the vibration sensor can use an accelerometer to detect the vibration of the pipeline caused by fluid flow, mechanical vibration or potential failure.

[0017] Then, according to the pipeline operation characteristics and safety monitoring requirements, set a reasonable data collection frequency. For pipelines with relatively stable operation status and relatively low risks, the collection frequency can be relatively low, such as collecting data once a minute; while for pipelines that transport hazardous substances, have complex operating conditions or are in high-risk areas, the collection frequency needs to be increased, such as collecting data several times per second. This can ensure that changes in pipeline status are captured in a timely manner, while avoiding the generation of too much redundant data, which affects data transmission and processing efficiency.

[0018] When each sensor collects data, the synchronization of timestamps is crucial. The Network Time Protocol (NTP) can be used to achieve accurate synchronization of the time of each node in the sensor network. Accurate timestamps can accurately record the collection time of each state parameter data, which provides a key basis for the time series analysis of subsequent data and helps to analyze the change trend of pipeline state parameters over time, periodic laws, and time correlation between different parameters.

[0019] Then, the collected data is pre-processed at the sensor end or on edge devices close to the sensor. This includes data cleaning to remove obviously erroneous or abnormal data points, such as outliers caused by sensor failure or electromagnetic interference; data verification to ensure the integrity and accuracy of the data; and data compression to reduce the amount of data to reduce the transmission bandwidth requirements.

[0020] Through the above-mentioned multiple implementation steps, accurate time series data of the status parameters of the monitored pipeline can be obtained, which can provide reliable analysis data for subsequent accurate judgment of the pipeline status.

[0021] In the embodiment of the present application, the pipeline state parameter data transmission module 120 is used to transmit the time series of the state parameter data of the monitored pipeline to the cloud data center through the Internet of Things technology. It should be understood that the Internet of Things technology is a network technology that connects any object to the Internet through sensing devices (such as sensors, radio frequency identification tags, etc.), network transmission (wired network, wireless network, etc.) and application platforms, etc., according to the agreed protocol, to exchange information and communicate, so as to achieve intelligent monitoring and management. Pipelines are often distributed in a wide geographical area, and may cross different regions or even areas with complex terrain. Relying on traditional data collection methods is extremely inefficient and difficult to achieve real-time performance. With its powerful network connection capabilities, the Internet of Things technology can enable each sensor installed at different positions of the pipeline to automatically and in real time send the collected data, meeting the requirements for real-time acquisition of pipeline state data scattered everywhere. After the time series of pipeline state parameter data is transmitted to the cloud data center, the powerful computing power of the cloud can be used to analyze the acquired pipeline state parameter data, which is conducive to timely discovering abnormal conditions of the pipeline and responding in a targeted manner.

[0022] The following is a detailed description of the specific implementation process of "transmitting the time series of the state parameter data of the monitored pipeline to the cloud data center through the Internet of Things technology": First, select appropriate Internet of Things devices for pipeline monitoring scenarios. These devices should have low power consumption, high stability and good compatibility to adapt to complex industrial environments. For example, an Internet of Things gateway with a multi-sensor interface can be selected, which can simultaneously connect various sensors such as temperature, pressure and vibration, and perform preliminary integration and processing of the collected data. In terms of configuration, according to the type and number of sensors, the communication parameters of the Internet of Things device are reasonably set to ensure stable and efficient communication between it and the sensor. At the same time, a unique identifier is assigned to the device so that it can be accurately identified and managed in the entire Internet of Things system.

[0023] Next, considering the real-time and accuracy requirements of pipeline monitoring data, as well as possible interference factors in industrial environments, it is necessary to select a reliable network communication protocol. A common example is MQTT, which is designed for IoT devices. It is lightweight, has low bandwidth usage, and reliable message delivery, making it very suitable for real-time transmission of pipeline status parameter data to the cloud. In addition, the CoAP protocol can also be used as an alternative. It performs well in resource-constrained network environments and can effectively reduce device power consumption while ensuring the reliability of data transmission. In actual applications, the most suitable communication protocol should be comprehensively evaluated and selected based on factors such as the scale of the pipeline monitoring system, the network environment, and data transmission requirements.

[0024] Then, in order to ensure the confidentiality, integrity and availability of pipeline status parameter data during transmission, strict data encryption and secure transmission mechanisms must be implemented. On the IoT device side, advanced encryption algorithms (such as AES, RSA, etc.) are used to encrypt the collected data and convert sensitive information into ciphertext to prevent the data from being stolen or tampered with during transmission. At the same time, a secure communication channel is established, such as using the TLS (Transport Layer Security) protocol to encrypt the transmission link to ensure that the data is in a secure encrypted state during transmission between the IoT device and the cloud data center. In addition, identity authentication and access control strategies should be implemented so that only authorized devices and users can access and transmit pipeline monitoring data, further enhancing the security of the system.

[0025] The IoT device encapsulates the encrypted time series of the pipeline state parameter data according to the selected communication protocol and sends it to the cloud data center through a wireless network (such as Wi-Fi, 4G / 5G, etc.) or a wired network (such as Ethernet). During the data transmission process, in order to cope with possible network fluctuations, signal interference or equipment failures, a variety of reliability assurance measures need to be adopted. For example, a data retransmission mechanism is set up. When data transmission fails or is lost, the IoT device automatically resends the data to ensure that the cloud data center can receive the data completely. At the same time, data verification technology, such as cyclic redundancy check (CRC), is used to perform integrity verification on the received data to promptly detect and correct erroneous data that may occur during the transmission process. In addition, through flow control and congestion avoidance algorithms, the data transmission efficiency is optimized to avoid data transmission delays or losses due to network congestion, ensuring that the pipeline state parameter data can be transmitted to the cloud data center in real time and accurately.

[0026] The cloud data center is deployed with corresponding servers and services to receive pipeline status parameter data from IoT devices. On the cloud server side, communication interfaces and protocol parsing modules that match IoT devices are configured to ensure that the received data can be correctly parsed and processed. At the same time, an efficient data storage and management system is established to perform operations such as classified storage, index creation, and data backup on the received data. In addition, through monitoring and management tools, the connection status of IoT devices, data transmission, and the operating status of cloud servers are monitored in real time to promptly discover and solve possible problems to ensure the stable operation of the entire data transmission link, thereby providing reliable data support for subsequent pipeline status analysis and processing.

[0027] In the embodiment of the present application, the pipeline state parameter data processing module 130 is used to process the time series of the state parameter data of the monitored pipeline in the cloud data center to obtain a pipeline state detection result.

[0028] Specifically, in the pipeline state parameter data processing module, the technical concept of the present application is to use an artificial intelligence-based data analysis method to first sort out the time series data of the state parameter data of the monitored pipeline containing temperature values, pressure values ​​and vibration amplitude values ​​and extract time series features to obtain temperature time series features, pressure time series features and vibration amplitude time series features, and then perform attention interaction fusion on the temperature time series features and pressure time series features to obtain the pipeline temperature-pressure time series interaction representation, and then perform fine-grained semantic interaction encoding on the pipeline temperature-pressure time series interaction representation and the vibration amplitude time series features to obtain the feature response representation to intelligently determine whether there is an abnormality in the state of the monitored pipeline. Compared with simple threshold alarms, this can improve the accuracy and real-time performance of pipeline state monitoring by deeply mining the correlation and interaction information between different pipeline state parameters in real time, which in turn helps to discover potential abnormal patterns and timely discover potential problems in the pipeline.

[0029] Figure 2 FIG. 1 is a block diagram of a pipeline state parameter data processing module in a pipeline intelligent detection system based on Internet of Things technology according to an embodiment of the present application. Figure 2 As shown, the pipeline state parameter data processing module 130 includes: a pipeline state parameter data feature encoding unit 131, which is used to perform data sorting and time series feature extraction on the time series of the state parameter data of the monitored pipeline to obtain temperature time series features, pressure time series features and vibration amplitude time series features; a pipeline temperature and pressure time series feature interaction unit 132, which is used to perform attention interaction fusion on the temperature time series features and the pressure time series features to obtain a pipeline temperature-pressure time series interaction representation; a pipeline multi-source information interaction response unit 133, which is used to perform fine-grained semantic interaction encoding on the pipeline temperature-pressure time series interaction representation and the vibration amplitude time series features to obtain a pipeline multi-source information response representation; a pipeline state detection unit 134, which is used to obtain the pipeline state detection result based on the pipeline multi-source information response representation.

[0030] In the embodiment of the present application, the pipeline state parameter data feature encoding unit 131 is used to sort the time series of the state parameter data of the monitored pipeline and extract the time series features to obtain the temperature time series features, pressure time series features and vibration amplitude time series features. Specifically, Figure 3 FIG. 1 is a block diagram of a pipeline state parameter data feature encoding unit in a pipeline intelligent detection system based on Internet of Things technology according to an embodiment of the present application. Figure 3As shown, the pipeline state parameter data feature encoding unit 131 includes: a pipeline state parameter data structured sorting subunit 1311, which is used to sort the time series of the state parameter data of the monitored pipeline according to the state parameter sample dimension to obtain the time series of temperature values, the time series of pressure values ​​and the time series of vibration amplitude values; a pipeline state parameter data timing encoding subunit 1312, which is used to extract timing features of the time series of temperature values, the time series of pressure values ​​and the time series of vibration amplitude values ​​respectively to obtain the temperature timing features, the pressure timing features and the vibration amplitude timing features.

[0031] In the embodiment of the present application, the pipeline state parameter data structured arrangement subunit 1311 is used to arrange the time series of the state parameter data of the monitored pipeline according to the state parameter sample dimension to obtain the time series of temperature value, the time series of pressure value and the time series of vibration amplitude value. It should be understood that the state parameters of the pipeline include multiple dimensions such as temperature value, pressure value and vibration amplitude value, and the data of each dimension has a unique role in judging the state of the pipeline. By arranging the data according to the state parameter sample dimension, the time series of the mixed original state parameter data can be split into the time series of temperature value, the time series of pressure value and the time series of vibration amplitude value, which can provide a clear and independent data basis for the subsequent feature extraction, change trend analysis and other operations for each specific parameter. In this way, the feature information of each different state parameter can be more accurately mined without being disturbed by other parameter data. That is, compared with directly extracting features from mixed data, this targeted feature extraction method can more accurately capture the key feature information contained in each parameter itself and reflecting the pipeline state, reduce the feature extraction deviation caused by data mixing, and thus help improve the accuracy of the entire pipeline state detection.

[0032] In the embodiment of the present application, the pipeline state parameter data time series encoding subunit 1312 is used to extract time series features from the time series of the temperature value, the time series of the pressure value and the time series of the vibration amplitude value to obtain the temperature time series features, the pressure time series features and the vibration amplitude time series features. Specifically, in the embodiment of the present application, the pipeline state parameter data time series encoding subunit is used to: respectively pass the time series of the temperature value, the time series of the pressure value and the time series of the vibration amplitude value through the pipeline state parameter time series feature extractor based on the void causal convolution network to obtain the temperature time series feature vector as the temperature time series feature, the pressure time series feature vector as the pressure time series feature and the vibration amplitude time series feature vector as the vibration amplitude time series feature. It should be understood that although the original time series data of the temperature value, the pressure value and the vibration amplitude value intuitively reflect the changes of these parameters over time, the deep information such as data fluctuations, periodic changes and correlation between data at different times contained therein can better reveal the real operation state mode of the pipeline. In order to mine the hidden information in the original time series data to better characterize the actual operating status of the pipeline, in the technical solution of the present application, it is necessary to respectively pass the time series of the temperature value, the time series of the pressure value and the time series of the vibration amplitude value through a pipeline state parameter time series feature extractor based on a void causal convolutional network to obtain a temperature time series feature vector, a pressure time series feature vector and a vibration amplitude time series feature vector. Those skilled in the art should know that the atrous causal convolution network uses a causal convolution operation, which means that during convolution calculation, the convolution kernel only processes the data at the current moment and the previous moment, and does not involve the data at the future moment. This makes the analysis of the pipeline state more in line with the actual physical process, avoids the feature extraction that does not conform to the actual logic, and makes the time series features finally extracted more physically meaningful and reasonable, which can provide a reliable basis for the subsequent accurate judgment of the pipeline state. At the same time, this model introduces atrous convolution, which can capture longer-distance data dependencies in the time series without increasing too much calculation and the number of parameters. This is conducive to more comprehensive mining of long-term dependent features in time series data, that is, grasping the health status of the pipeline from a more macroscopic time scale, which helps to discover potential abnormal risks in advance.

[0033] The following is a detailed description of the specific implementation process of "passing the time series of the temperature value, the time series of the pressure value and the time series of the vibration amplitude value through the pipeline state parameter time series feature extractor based on the void causal convolutional network to obtain the temperature time series feature vector as the temperature time series feature, the pressure time series feature vector as the pressure time series feature and the vibration amplitude time series feature vector as the vibration amplitude time series feature": First, according to the characteristics of the pipeline state parameter data and the analysis requirements, the architecture of the void causal convolutional network is designed. Determine the key parameters such as the number of network layers, the size of the convolution kernel of each layer, and the void rate. For example, the network is set to contain multiple convolutional layers. The shallow convolutional layer can use a smaller convolution kernel (such as 3x3) to capture local features. As the network depth increases, the void rate is gradually increased to expand the receptive field and obtain dependencies on longer time series. At the same time, ensure that the network structure is a causal structure, that is, the convolution operation only depends on the data of the past and current moments, to ensure the real-time processing, which is suitable for real-time analysis scenarios of pipeline state monitoring.

[0034] After building the network architecture, initialize the network. Use appropriate initialization methods for the weight parameters of the convolutional layer, such as He initialization or Xavier initialization, so that the network has good stability and convergence in the early stage of training. According to the distribution range and feature importance of the pipeline state parameter data, set the bias parameters reasonably. At the same time, determine the activation function (such as ReLU or LeakyReLU) to enhance the nonlinear expression ability of the network in order to better fit the complex time series feature relationship. In addition, according to the computing resources and real-time requirements, adjust the network's hyperparameters, such as learning rate, batch size, etc., to optimize the network training and inference process.

[0035] Before inputting the time series of temperature, pressure, and vibration amplitude values ​​into the network, necessary data preprocessing operations are performed. Normalizing the time series data and mapping the data to a specific interval (such as [0,1] or [-1,1]) helps improve the network training efficiency and feature extraction effect. At the same time, according to the network input requirements, the time series data is dimensionally adapted, such as adjusting the data shape, padding or truncating the sequence length, etc., to ensure that the data can be correctly input into the network for processing.

[0036] Next, the preprocessed temperature value time series is input into the atrous causal convolutional network. In each convolution operation of the network, the convolution kernel performs interval convolution calculations on the time series according to the set atrous rate to extract temperature features at different time scales. Through the stacking of multiple convolutional layers, the network gradually learns a more advanced and abstract temperature time series feature representation. Due to the limitation of the causal structure, the convolution operation only uses the temperature data of the past and current moments to ensure that the extracted features reflect the real-time temperature change trend and periodic law without being interfered by future data. Similarly, the time series of pressure values ​​and vibration amplitude values ​​are respectively input into independent atrous causal convolutional networks for feature extraction. For the pressure time series feature extraction, the network focuses on the fluctuation pattern of pressure values ​​in time, the rate of pressure change, and the correlation with pressure values ​​at other moments. The vibration amplitude time series feature extraction network focuses on the change trend of vibration amplitude, the vibration frequency component, and the energy distribution of vibration signals in the time series.

[0037] After multiple layers of processing in the dilated causal convolutional network, the time series feature vectors corresponding to temperature, pressure, and vibration amplitude are obtained in the last layer of the network. These feature vectors are highly abstract representations of the original time series data at different time scales and feature dimensions.

[0038] It should be noted that in order for the void causal convolutional network to accurately extract the time series features of pipeline state parameters, a large amount of historical pipeline state data is needed for model training. The time series data of pipeline temperature, pressure and vibration amplitude marked with normal and abnormal pipeline states are divided into training set, validation set and test set. During the training process, the difference between the network prediction result and the true label is calculated according to the set loss function (such as mean square error, cross entropy, etc.), and the weight parameters of the network are updated through the back propagation algorithm to optimize the feature extraction ability of the network. At the same time, the training process is monitored using the validation set to prevent overfitting. According to the performance of the validation set, the network's hyperparameters, optimization algorithms (such as stochastic gradient descent, Adam, etc.) or regularization terms (such as L1, L2 regularization) are adjusted to continuously improve the network's extraction accuracy and generalization ability of pipeline state parameter time series features. Finally, the fully trained and optimized model can accurately extract effective time series feature vectors from the real-time collected pipeline state parameter time series, which can provide strong support for pipeline state detection.

[0039] In the embodiment of the present application, the pipeline temperature and pressure time series feature interaction unit 132 is used to perform attention interaction fusion on the temperature time series feature and the pressure time series feature to obtain the pipeline temperature-pressure time series interaction representation. Specifically, in the embodiment of the present application, the pipeline temperature and pressure time series feature interaction unit is used to: pass the temperature time series feature vector and the pressure time series feature vector through the pipeline multi-parameter time series feature interactor based on the X-linear attention network to obtain the pipeline temperature-pressure time series interaction representation vector as the pipeline temperature-pressure time series interaction representation. It should be understood that the two parameters of the pipeline temperature and pressure have a close internal connection in the actual operation process. For example, when the pressure in the pipeline changes, it often affects the flow state of the material and the heat exchange process with the pipeline wall, thereby causing a corresponding change in temperature; conversely, an abnormal increase or decrease in temperature may also change the physical properties of the material, thereby affecting the pressure in the pipeline. However, after the previous separate time series feature extraction, although the feature vectors that can each reflect a certain pipeline state are obtained, these feature vectors are relatively independent reflections of the changes in temperature and pressure. In order to be able to deeply mine the hidden information of the mutual influence and correlation between temperature and pressure parameters, and to grasp the operation status of the pipeline more comprehensively as a whole, in the technical solution of the present application, it is necessary to pass the temperature time series feature vector and the pressure time series feature vector through the pipeline multi-parameter time series feature interactor based on the X-linear attention network to obtain the pipeline temperature-pressure time series interaction representation vector. Specifically, the X-linear attention network is based on bilinear pooling technology, is good at capturing the interaction between two-order features, and can mine the nonlinear relationship between the temperature time series feature vector and the pressure time series feature vector. It is not just a simple merger or linear combination of two feature vectors, but through operations such as calculating the outer product, it fuses the information of the two and generates a new pipeline temperature-pressure time series interaction representation vector. This vector contains rich feature information reflecting the complex interaction of temperature and pressure, which can provide more valuable clues for a deeper and more accurate understanding of the pipeline status.

[0040] In the embodiment of the present application, the pipeline multi-source information interactive response unit 133 is used to perform fine-grained semantic interactive encoding on the pipeline temperature-pressure time series interactive representation and the vibration amplitude time series feature to obtain the pipeline multi-source information response representation. Specifically, Figure 4 FIG. 1 is a block diagram of a pipeline multi-source information interaction response unit in a pipeline intelligent detection system based on Internet of Things technology according to an embodiment of the present application. Figure 4As shown, the pipeline multi-source information interaction response unit 133 includes: a pipeline multi-source information interaction semantic flow field construction subunit 1331, which is used to construct a pipeline multi-source information interaction semantic flow field between the pipeline temperature-pressure timing interaction representation vector and the vibration amplitude timing feature vector; a pipeline multi-source information feature alignment subunit 1332, which is used to perform feature alignment on the pipeline temperature-pressure timing interaction representation vector and the vibration amplitude timing feature vector based on the pipeline multi-source information interaction semantic flow field to obtain an aligned pipeline temperature-pressure timing interaction representation vector and an aligned vibration amplitude timing feature vector; a pipeline multi-source information response representation generation subunit 1333, which is used to perform cross-domain fine-grained semantic interaction on the aligned pipeline temperature-pressure timing interaction representation vector and the aligned vibration amplitude timing feature vector to obtain a pipeline multi-source information response representation vector as the pipeline multi-source information response representation.

[0041] It should be understood that the pipeline temperature-pressure time series interaction representation vector and the vibration amplitude time series feature vector carry key feature information reflecting the pipeline state from different sources, and there is a complex association of time series granularity between the two in different time dimensions. When processing feature data from different sources, since the feature vectors of different parameters may differ in feature meaning and data magnitude, if these features are directly fused, it is likely that some valuable information will be ignored or lost during the fusion process. Specifically, some feature information generated by the interaction of temperature and pressure may not be fully reflected in the final fusion result due to the failure to reasonably handle the correlation and complementarity between each other during conventional fusion operations such as direct addition of vibration amplitude features, resulting in inaccurate representation information used to judge the pipeline state, affecting the accuracy of the pipeline state detection results. Therefore, in the technical solution of the present application, it is necessary to perform fine-grained semantic interaction encoding on the pipeline temperature-pressure time series interaction representation and the vibration amplitude time series feature to obtain the pipeline multi-source information response representation.

[0042] Specifically, in an embodiment of the present application, the pipeline multi-source information interaction semantic flow field construction subunit is used to: perform point convolution encoding operations and activation operations based on the Sigmoid function on the pipeline temperature-pressure timing interaction representation vector and the vibration amplitude timing feature vector respectively to obtain the pipeline temperature-pressure timing interaction representation modulation vector and the vibration amplitude timing feature modulation vector; multiply the pipeline temperature-pressure timing interaction representation modulation vector and the transposed vector of the vibration amplitude timing feature modulation vector and divide the multiplication by the square root of the length of the vibration amplitude timing feature modulation vector to obtain the pipeline multi-source information interaction matrix; perform convolution operation on the pipeline multi-source information interaction matrix based on the convolution kernel of The convolution coding operation and the activation operation based on the Sigmoid function are performed to obtain the pipeline multi-source information interaction modulation matrix; the pipeline multi-source information interaction modulation matrix is ​​subjected to the convolution kernel based on The convolutional encoding operation and upsampling operation are performed to obtain the pipeline multi-source information interaction semantic flow field. The above process can be expressed as: ;in, represents the pipeline temperature-pressure time series interaction representation vector, represents point convolutional coding, express function, represents the pipeline temperature-pressure time series interaction characterization modulation vector, represents the time series eigenvector of the vibration amplitude, represents the vibration amplitude temporal characteristic modulation vector, represents matrix multiplication, express The transposed vector of express Length, Indicates that the convolution kernel is The convolutional coding operation, Indicates that the convolution kernel is The convolutional coding operation represents the upsampling operation, express and The semantic flow field of pipeline multi-source information interaction between them.

[0043] It should be understood that the pipeline temperature-pressure time series interaction representation vector and the vibration amplitude time series feature vector each contain information at different levels, and the relationship between them is often complex and nonlinear. By constructing the pipeline multi-source information interaction semantic flow field between the pipeline temperature-pressure time series interaction representation vector and the vibration amplitude time series feature vector, this complex semantic relationship space can be effectively captured, and the associated information that may be difficult to intuitively perceive and scattered in two different types of vectors can be integrated and presented, so that these internal connections can be systematically analyzed and utilized in this flow field in the future, providing a basis for a comprehensive understanding of the multi-source information interaction involved in the pipeline operation status.

[0044] Next, based on the pipeline multi-source information interaction semantic flow field, feature alignment is performed on the pipeline temperature-pressure time series interaction representation vector and the vibration amplitude time series feature vector to obtain the aligned pipeline temperature-pressure time series interaction representation vector and the aligned vibration amplitude time series feature vector. The above process can be expressed as: ;in, represents the pipeline temperature-pressure time series interaction characterization modulation vector, express and The semantic flow field of pipeline multi-source information interaction between them, represents the vibration amplitude temporal characteristic modulation vector, represents the pipeline temperature-pressure time series interaction representation vector after alignment, Represents the time series feature vector of vibration amplitude after alignment.

[0045] It should be understandable that the pipeline temperature-pressure time series interaction representation vector and the vibration amplitude time series feature vector come from different aspects and have different forms of expression and data characteristics. Feature alignment based on the constructed pipeline multi-source information interaction semantic flow field can allow these two feature vectors that originally belonged to different "semantic worlds" to enter the same semantic space. In this unified space, they can be accurately matched to eliminate obstacles to understanding and association caused by different sources and modal differences. In other words, through feature alignment, the specific points of mutual response and correlation between temperature, pressure-related features and vibration amplitude features can be found, which can make the subsequent comprehensive analysis of the overall condition of the pipeline more reasonable and accurate.

[0046] Specifically, in the embodiment of the present application, the pipeline multi-source information response representation generating subunit is used to: perform a linear transformation on the aligned pipeline temperature-pressure time series interaction representation vector to obtain a pipeline temperature-pressure time series interaction representation query vector and a pipeline temperature-pressure time series interaction representation value vector. This process can be expressed as: ;in, represents the pipeline temperature-pressure time series interaction representation vector after alignment, represents matrix multiplication, represents the query matrix, represents the query bias vector, represents the pipeline temperature-pressure time series interaction representation query vector, represents the value matrix, represents the value bias vector, Represents the pipeline temperature-pressure time series interaction characterization value vector.

[0047] The aligned vibration amplitude time series feature vector is linearly transformed to obtain a vibration amplitude time series feature key vector. The process can be expressed as: ;in, represents the time series feature vector of vibration amplitude after alignment, represents matrix multiplication, represents the key matrix, represents the key bias vector, Represents the vibration amplitude time series feature key vector.

[0048] The pipeline temperature-pressure time series interaction characterization query vector, the pipeline temperature-pressure time series interaction characterization value vector, and the vibration amplitude time series feature key vector are subjected to fine-grained response encoding based on a converter structure to obtain the pipeline multi-source information response characterization vector. This process can be expressed as: ;in, represents the pipeline temperature-pressure time series interaction representation query vector, represents the pipeline temperature-pressure time series interaction characterization value vector, represents the vibration amplitude time series feature key vector, represents matrix multiplication, express The transposed vector of express function, express Length, Represents the pipeline multi-source information response representation vector.

[0049] It should be understood that although the aligned pipeline temperature-pressure time series interaction representation vector and the aligned vibration amplitude time series feature vector have been aligned, their original feature spaces may not be directly consistent with the specific tasks to be carried out later. Through linear transformation, these feature spaces can be remapped. Specifically, for the pipeline temperature-pressure time series interaction representation vector, it is split into a pipeline temperature-pressure time series interaction representation query vector and a pipeline temperature-pressure time series interaction representation value vector, so that different parts of the vector can perform their respective functions. The pipeline temperature-pressure time series interaction representation query vector can be used specifically to search and locate other features related to the current task, and the pipeline temperature-pressure time series interaction representation value vector is responsible for carrying specific information content, such as specific temperature and pressure values ​​and the actual meaning of their interaction. Similarly, the vibration amplitude time series feature key vector obtained by transforming the vibration amplitude time series feature vector can also accurately assist in locating the content associated with it among many features. In this way, different task requirements can be effectively supported by corresponding vectors, which can make the entire analysis process more targeted and organized.

[0050] Accordingly, considering that there are various interactions between the features of different sources of the pipeline, changes in temperature and pressure may affect the vibration amplitude, and changes in the vibration amplitude will in turn affect the temperature, pressure, etc., and these interactions are complicated. In order to understand the interactive relationship between these pipeline state parameters, in this application, it is necessary to perform fine-grained response encoding operations based on the converter structure on the pipeline temperature-pressure time series interaction representation query vector, the pipeline temperature-pressure time series interaction representation value vector, and the vibration amplitude time series feature key vector. Fine-grained response encoding based on the converter structure can use its unique mechanism to effectively model the interaction between these features. Specifically, the multi-head self-attention mechanism can examine the association between different features from multiple angles, that is, observe these interactions from different "perspectives", thereby constructing a more detailed and comprehensive interaction state. This can fully integrate these interactive information into the generated pipeline multi-source information response representation vector, which can better reflect the real state of mutual influence and mutual restriction of multi-source information during the actual operation of the pipeline, and help to accurately judge and deeply analyze the pipeline state in the future.

[0051] In an embodiment of the present application, the pipeline state detection unit 134 is used to obtain the pipeline state detection result based on the pipeline multi-source information response representation. Specifically, in an embodiment of the present application, the pipeline state detection unit is used to: pass the pipeline multi-source information response representation vector through a pipeline state detector based on a classifier to obtain the pipeline state detection result, and the pipeline state detection result is used to indicate whether the state of the monitored pipeline is abnormal. More specifically, in an embodiment of the present application, the pipeline state detection unit is used to: use the fully connected layer of the classifier to fully connect the pipeline multi-source information response representation vector to obtain a pipeline multi-source information response representation fully connected encoding feature vector; input the pipeline multi-source information response representation fully connected encoding feature vector into the Softmax classification function of the classifier to obtain the probability value of the pipeline multi-source information response representation vector belonging to each classification label, and the classification label includes a label for indicating that the state of the monitored pipeline is abnormal and a label for indicating that the state of the monitored pipeline is not abnormal; the classification label corresponding to the largest probability value is determined as the pipeline state detection result. It should be understood that the classifier itself has a strong discriminative ability. It can be trained based on a large amount of sample data to discover the characteristic difference rules between different categories (normal state and various abnormal states). Specifically, by learning the state parameter data corresponding to many normally operating pipelines and pipelines that have experienced different types of faults (such as leakage, corrosion, blockage, etc.), the classifier can grasp the key discriminative information such as the distribution range of feature vectors under normal conditions and the trend of feature vector changes corresponding to different faults under abnormal conditions. When faced with a newly input pipeline multi-source information response representation vector, the classifier can accurately determine whether the pipeline belongs to the normal or abnormal category based on these previously learned rules, so as to give full play to its advantage of accurately distinguishing different states from complex feature data, so as to make the pipeline state detection results more scientific and reliable.

[0052] In particular, when the pipeline temperature-pressure temporal interaction representation vector and the vibration amplitude temporal feature vector respectively represent the local correlation temporal interaction features of the pipeline temperature-pressure and the local temporal correlation features of the vibration amplitude value, when performing feature interaction response based on semantic flow field modulation, the misalignment of the temporal semantic flow field modulation under the differentiation of source temporal association patterns will also cause the interactive response richness of the pipeline multi-source information response representation vector, resulting in redundancy of the mapping domain probability distribution information, affecting the accuracy of the classification results.

[0053] Preferably, when the pipeline multi-source information response characterization vector is passed through a pipeline state detector based on a classifier to obtain the pipeline state detection result, optimizing the pipeline multi-source information response characterization vector includes: determining the distance from adjacent eigenvalues ​​and the number of isolated eigenvalues ​​of the pipeline multi-source information response greater than a predetermined threshold in the pipeline multi-source information response characterization vector. and its corresponding pipeline multi-source information response smoothness factor , that is: ;in, They represent the first , , The eigenvalues ​​at the positions, Indicates that the eligible To count, Indicates a predetermined threshold.

[0054] Calculate the square root of the sum of squares of all eigenvalues ​​of the pipeline multi-source information response characterization vector and divide it by the length of the pipeline multi-source information response characterization vector to obtain the main component domain value of the pipeline multi-source information response ,in, Represents the length of the pipeline multi-source information response representation vector.

[0055] The number of isolated eigenvalues ​​in response to the multi-source information of the pipeline The power function of the pipeline multi-source information response characterization vector is calculated as an exponent and the pipeline multi-source information response smoothing factor factor Perform point multiplication to obtain the orthogonal basis eigenvector of the pipeline multi-source information response ,in, Represents dot product.

[0056] Calculate the pipeline multi-source information response characterization vector and the pipeline multi-source information response smoothness factor and the pipeline multi-source information response principal component domain value The dot product of the inverse of obtains the pipeline multi-source information response projection channel vector .

[0057] Calculate the orthogonal basis eigenvector of the pipeline multi-source information response The pipeline multi-source information response projection channel vector to obtain the optimized pipeline multi-source information response representation vector.

[0058] That is, based on the vector field approximation theorem, the discrete singular points and smoothness factors in the pipeline multi-source information response characterization vector are used as the differentiable dimensions of the multi-dimensional topological structure, and the normal directions of the orthogonal basis vectors are fixed in the local coordinate system of the principal component aggregation domain of the pipeline multi-source information response characterization vector, and an orthogonal projection channel from the feature space to the probability measure manifold is established. Thus, the random noise interference caused by the multi-modal coupling strength is eliminated through the effective alignment of the principal component aggregation domain of the pipeline multi-source information response characterization vector, thereby avoiding the feature distortion problem caused by the failure of information redundancy elimination in the probability mapping process, and improving the accuracy of the pipeline state detection result obtained by the pipeline multi-source information response characterization vector through the classifier-based pipeline state detector.

[0059] In summary, the pipeline state parameter data processing module 130 is clearly explained. It uses an artificial intelligence-based data analysis method to first sort out the time series data of the state parameter data of the monitored pipeline including temperature values, pressure values ​​and vibration amplitude values ​​and extract time series features to obtain temperature time series features, pressure time series features and vibration amplitude time series features, and then perform attention interaction fusion on the temperature time series features and pressure time series features to obtain the pipeline temperature-pressure time series interaction representation, and then perform fine-grained semantic interaction encoding on the pipeline temperature-pressure time series interaction representation and the vibration amplitude time series features to obtain the feature response representation to intelligently determine whether the state of the monitored pipeline is abnormal. Compared with simple threshold alarms, this can improve the accuracy and real-time performance of pipeline state monitoring by deeply mining the correlation and interaction information between different pipeline state parameters in real time, thereby helping to discover potential abnormal patterns and timely discover potential problems in the pipeline.

[0060] In the embodiment of the present application, the pipeline state detection result generating module 140 is used to send an early warning signal and the pipeline state detection result to the management personnel in response to the pipeline state detection result that the state of the monitored pipeline is abnormal. It should be understood that by quickly sending an early warning signal and a specific pipeline state detection result to the management personnel when the pipeline state is detected to be abnormal, the relevant personnel responsible for pipeline operation and maintenance and safety management can be informed of the pipeline problem as soon as possible, avoiding delays in processing due to untimely information transmission, so that they can start response measures as soon as possible to minimize the harm that may be caused by abnormal conditions. Specifically, after receiving the early warning signal, the management personnel can respond quickly and organize forces in time to investigate and deal with the abnormal situation. For example, if the early warning indicates that there is a risk of leakage in the pipeline, the maintenance personnel can immediately rush to the scene and take emergency measures such as plugging the leakage point and adjusting the pipeline operation parameters in the shortest time to prevent the abnormal situation from further deteriorating, effectively avoid major safety accidents that may be caused, and maximize the normal operation of the pipeline and the safety of the surrounding environment and personnel.

[0061] In summary, the pipeline intelligent detection system 100 based on the Internet of Things technology according to the embodiment of the present application is explained, which collects the time series of state parameter data such as temperature value, pressure value, vibration amplitude value, etc. of the monitored pipeline through the sensor network, and transmits it to the cloud data center using the Internet of Things technology, and processes the time series data of these state parameters in the cloud data center to obtain the pipeline state detection result of the monitored pipeline. If the pipeline state detection result shows that the pipeline state is abnormal, an early warning signal and the detection result are sent to the management personnel. In this way, intelligent monitoring of the pipeline state can be realized, which is conducive to timely detection of abnormal conditions, and thus can ensure the safety and stability of pipeline operation.

Claims

1. A pipeline intelligent detection system based on Internet of Things technology, characterized in that: include: The pipeline state parameter data acquisition module is used to obtain the time series of the state parameter data of the monitored pipeline collected by the sensor network, wherein the state parameter data includes temperature value, pressure value and vibration amplitude value; the pipeline state parameter data transmission module is used to transmit the time series of the state parameter data of the monitored pipeline to the cloud data center through the Internet of Things technology; the pipeline state parameter data processing module is used to process the time series of the state parameter data of the monitored pipeline in the cloud data center to obtain the pipeline state detection result, wherein the pipeline state parameter data processing module includes: a pipeline state parameter data feature encoding unit, which is used to sort the time series of the state parameter data of the monitored pipeline and extract the time series features to obtain to temperature timing features, pressure timing features and vibration amplitude timing features; a pipeline temperature and pressure timing feature interaction unit, used to perform attention interaction fusion on the temperature timing features and the pressure timing features to obtain a pipeline temperature-pressure timing interaction representation; a pipeline multi-source information interaction response unit, used to perform fine-grained semantic interaction encoding on the pipeline temperature-pressure timing interaction representation and the vibration amplitude timing features to obtain a pipeline multi-source information response representation; a pipeline state detection unit, used to obtain the pipeline state detection result based on the pipeline multi-source information response representation; a pipeline state detection result generation module, used to send a warning signal and the pipeline state detection result to the management personnel in response to the pipeline state detection result that the state of the monitored pipeline is abnormal.

2. According to claim 1, the intelligent pipeline detection system based on Internet of Things technology is characterized in that: The pipeline state parameter data feature encoding unit includes: a pipeline state parameter data structured sorting subunit, which is used to sort the time series of the state parameter data of the monitored pipeline according to the state parameter sample dimension to obtain the time series of temperature values, the time series of pressure values ​​and the time series of vibration amplitude values; a pipeline state parameter data time series encoding subunit, which is used to extract time series features from the time series of temperature values, the time series of pressure values ​​and the time series of vibration amplitude values ​​respectively to obtain the temperature time series features, the pressure time series features and the vibration amplitude time series features.

3. The pipeline intelligent detection system based on Internet of Things technology according to claim 2 is characterized in that: The pipeline state parameter data timing encoding subunit is used to: respectively pass the time series of the temperature value, the time series of the pressure value and the time series of the vibration amplitude value through a pipeline state parameter timing feature extractor based on a void causal convolutional network to obtain a temperature timing feature vector as the temperature timing feature, a pressure timing feature vector as the pressure timing feature and a vibration amplitude timing feature vector as the vibration amplitude timing feature.

4. The pipeline intelligent detection system based on Internet of Things technology according to claim 3 is characterized in that: The pipeline temperature-pressure time series feature interaction unit is used to: pass the temperature time series feature vector and the pressure time series feature vector through a pipeline multi-parameter time series feature interactor based on an X-linear attention network to obtain a pipeline temperature-pressure time series interaction representation vector as the pipeline temperature-pressure time series interaction representation.

5. The pipeline intelligent detection system based on Internet of Things technology according to claim 4 is characterized in that: The pipeline multi-source information interaction response unit includes: a pipeline multi-source information interaction semantic flow field construction subunit, which is used to construct a pipeline multi-source information interaction semantic flow field between the pipeline temperature-pressure time series interaction representation vector and the vibration amplitude time series feature vector; a pipeline multi-source information feature alignment subunit, which is used to perform feature alignment on the pipeline temperature-pressure time series interaction representation vector and the vibration amplitude time series feature vector respectively based on the pipeline multi-source information interaction semantic flow field to obtain an aligned pipeline temperature-pressure time series interaction representation vector and an aligned vibration amplitude time series feature vector; a pipeline multi-source information response representation generation subunit, which is used to perform cross-domain fine-grained semantic interaction on the aligned pipeline temperature-pressure time series interaction representation vector and the aligned vibration amplitude time series feature vector to obtain a pipeline multi-source information response representation vector as the pipeline multi-source information response representation.

6. The pipeline intelligent detection system based on Internet of Things technology according to claim 5 is characterized in that: The pipeline multi-source information interaction semantic flow field construction subunit is used to: perform point convolution encoding operations and activation operations based on Sigmoid function on the pipeline temperature-pressure time series interaction representation vector and the vibration amplitude time series feature vector respectively to obtain the pipeline temperature-pressure time series interaction representation modulation vector and the vibration amplitude time series feature modulation vector; multiply the pipeline temperature-pressure time series interaction representation modulation vector and the transposed vector of the vibration amplitude time series feature modulation vector and divide the multiplication by the square root of the length of the vibration amplitude time series feature modulation vector to obtain the pipeline multi-source information interaction matrix; perform convolution operation on the pipeline multi-source information interaction matrix based on the convolution kernel of The convolution coding operation and the activation operation based on the Sigmoid function are performed to obtain the pipeline multi-source information interaction modulation matrix; the pipeline multi-source information interaction modulation matrix is ​​subjected to the convolution kernel based on The convolutional encoding operation and upsampling operation are performed to obtain the pipeline multi-source information interaction semantic flow field.

7. The pipeline intelligent detection system based on Internet of Things technology according to claim 6 is characterized in that: The pipeline multi-source information response representation generating subunit is used to: perform linear transformation on the aligned pipeline temperature-pressure time series interaction representation vector to obtain a pipeline temperature-pressure time series interaction representation query vector and a pipeline temperature-pressure time series interaction representation value vector; perform linear transformation on the aligned vibration amplitude time series feature vector to obtain a vibration amplitude time series feature key vector; Fine-grained response encoding based on a converter structure is performed on the pipeline temperature-pressure time series interaction characterization query vector, the pipeline temperature-pressure time series interaction characterization value vector, and the vibration amplitude time series feature key vector to obtain the pipeline multi-source information response characterization vector.

8. The pipeline intelligent detection system based on Internet of Things technology according to claim 7 is characterized in that: The pipeline state detection unit is used to: pass the pipeline multi-source information response characterization vector through a classifier-based pipeline state detector to obtain the pipeline state detection result, and the pipeline state detection result is used to indicate whether the state of the monitored pipeline is abnormal.

9. The pipeline intelligent detection system based on Internet of Things technology according to claim 8 is characterized in that: The pipeline state detection unit is used to: use the fully connected layer of the classifier to fully connect the pipeline multi-source information response representation vector to obtain a pipeline multi-source information response representation fully connected encoding feature vector; input the pipeline multi-source information response representation fully connected encoding feature vector into the Softmax classification function of the classifier to obtain the probability value of the pipeline multi-source information response representation vector belonging to each classification label, wherein the classification label includes a label for indicating that the state of the monitored pipeline is abnormal and a label for indicating that the state of the monitored pipeline is not abnormal; and determine the classification label corresponding to the largest one of the probability values ​​as the pipeline state detection result.

Images