Gas generator set monitoring system and method

Abstract

The invention discloses a gas generator set monitoring system and method, and the system collects various state parameters during the operation of a gas generator set through a sensor group, transmits the state parameters to a remote monitoring server, and arranges the state parameters into a format suitable for machine learning analysis at a server terminal. Further performing time sequence correlation analysis on the state parameters by using a machine learning algorithm based on deep learning and a data processing technology so as to excavate a time sequence dynamic correlation mode among various state parameters in the operation process of the gas generator set, thereby intelligently determining whether the working state of the gas generator set is abnormal or not; and efficient monitoring of the working state of the gas generator set is realized. Through the mode, the remote monitoring capability of the fault mode of the gas generator set can be improved, the operation condition of the gas generator set can be known in time, any potential problem can be quickly responded, and meanwhile, a solid technical foundation is provided for the remote monitoring technology.

Description

Technical Field

[0001] The present application relates to the field of intelligent monitoring, and more specifically, to a gas generator set monitoring system and method. Background Art

[0002] With the development of industrial automation and informatization, gas generator sets, as key energy conversion equipment, play an indispensable role in power production, distributed energy systems and emergency power supply. In order to ensure the reliability and efficiency of these devices, real-time status monitoring becomes very important.

[0003] However, traditional monitoring methods usually rely on engineers' experience and fixed thresholds to judge the status of equipment. This method has limitations, such as being unable to adapt to complex working conditions and difficulty in capturing early fault characteristics. With the development of the Internet of Things, big data and artificial intelligence technologies, the status monitoring and fault diagnosis of gas generator sets have gradually shifted from traditional rule-based methods to data-driven methods. However, due to limited data processing capabilities and the lack of effective data analysis methods, current data-driven technologies are difficult to capture subtle changes in machines under complex and changeable working conditions. At the same time, there is a problem of slow response speed, and it is impossible to achieve real-time or near-real-time fault detection, which may lead to delayed fault detection, thereby affecting the normal operation of the equipment and even causing serious safety accidents.

[0004] Therefore, in order to realize intelligent monitoring of the working status of the gas generator set, a gas generator set monitoring system and method are 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 gas generator set monitoring system and method, which collects various state parameters of the gas generator set during operation through a sensor group, and transmits them to a remote monitoring server, which is organized into a format suitable for machine learning analysis at the server terminal, and further uses a machine learning algorithm based on deep learning and data processing technology to perform time series correlation analysis on the state parameters to dig out the time series dynamic correlation pattern between various state parameters during the operation of the gas generator set, so as to intelligently determine whether there is an abnormality in the working state of the gas generator set, and realize efficient monitoring of the working state of the gas generator set. In this way, the remote monitoring capability of the fault mode of the gas generator set can be improved, which helps to timely understand the operating status of the gas engine group to quickly respond to any potential problems. At the same time, it also provides a solid technical foundation for remote monitoring technology.

[0006] According to one aspect of the present application, a gas generator set monitoring system is provided, comprising: A data acquisition module, used to acquire the state parameters of the gas generator set collected by the sensor to obtain a data set of the state parameters; A data transmission module, used for compressing the data set of the state parameters and transmitting it to the monitoring server; A data decompression and sorting module is used to decompress and sort the compressed data packets of the state parameters on the monitoring server to obtain the time queue of the main variable and the time series joint matrix of the covariate; A time series correlation feature extraction module is used to extract time series correlation features from the time queue of the main variable and the time series joint matrix of the covariate to obtain time series correlation implicit features of the main variable and time series correlation implicit features of the covariate; A semantic interaction encoding module, used for performing fine-grained feature interaction based on external knowledge guidance on the main variable time series associated implicit features and the covariate time series associated implicit features to obtain the main-covariate time series fine-grained interaction features of the gas generator set; The abnormality monitoring module is used to determine whether there is an abnormality in the working state of the gas generator set based on the fine-grained interaction characteristics of the main and covariate time series of the gas generator set.

[0007] According to another aspect of the present application, a gas generator set monitoring method is provided, which includes: Acquiring state parameters of the gas generator set collected by the sensor to obtain a data set of state parameters; Compressing the data set of the state parameters and transmitting it to the monitoring server; On the monitoring server, the compressed data packets of the state parameters are decompressed and data is sorted to obtain the time queue of the main variables and the time series joint matrix of the covariates; Extracting time series correlation features from the time queue of the main variable and the time series joint matrix of the covariate to obtain time series correlation implicit features of the main variable and time series correlation implicit features of the covariate; Performing fine-grained feature interaction based on external knowledge guidance on the main variable time series associated implicit features and the covariate time series associated implicit features to obtain the main-covariate time series fine-grained interaction features of the gas generator set; Based on the time series fine-grained interaction characteristics of the main and covariates of the gas-fired generator set, it is determined whether the working state of the gas-fired generator set is abnormal.

[0008] Compared with the prior art, the present application provides a gas generator set monitoring system and method, which collects various state parameters of the gas generator set during operation through a sensor group and transmits them to a remote monitoring server, which is organized into a format suitable for machine learning analysis at the server terminal, and further uses a machine learning algorithm based on deep learning and data processing technology to perform time series correlation analysis on the state parameters to dig out the time series dynamic correlation pattern between various state parameters during the operation of the gas generator set, so as to intelligently determine whether there is an abnormality in the working state of the gas generator set, and realize efficient monitoring of the working state of the gas generator set. In this way, the remote monitoring capability of the fault mode of the gas generator set can be improved, which helps to timely understand the operating status of the gas engine set to quickly respond to any potential problems, and at the same time, it also provides a solid technical foundation for remote monitoring technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 is a block diagram of a gas generator set monitoring system according to an embodiment of the present application; Figure 2 A data flow diagram of a gas generator set monitoring system according to an embodiment of the present application; Figure 3 is a block diagram of a semantic interaction encoding module in a gas generator set monitoring system according to an embodiment of the present application; Figure 4 The present invention is a flow chart of a method for monitoring a gas generator set according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] 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.

[0012] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0015] 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.

[0016] In the technical solution of the present application, a gas generator set monitoring system is proposed. Figure 1 4 is a block diagram of a gas generator set monitoring system according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a gas generator set monitoring system according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the gas generator monitoring system 300 includes: a data acquisition module 310, which is used to acquire the state parameters of the gas generator set collected by the sensor to obtain a data set of the state parameters; a data transmission module 320, which is used to compress the data set of the state parameters and transmit it to the monitoring server; a data decompression and sorting module 330, which is used to decompress and sort the compressed data packet of the state parameters on the monitoring server to obtain the time queue of the main variable and the time series joint matrix of the covariate; a time series correlation feature extraction module 340, which is used to extract the time series correlation feature of the main variable The time queue of the quantity and the covariate time series joint matrix are respectively subjected to time series correlation feature extraction to obtain the main variable time series correlation implicit features and the covariate time series correlation implicit features; the semantic interaction encoding module 350 is used to perform fine-grained feature interaction based on external knowledge guidance on the main variable time series correlation implicit features and the covariate time series correlation implicit features to obtain the main-covariate time series fine-grained interaction features of the gas generator set; the abnormality monitoring module 360 ​​is used to determine whether there is an abnormality in the working state of the gas generator set based on the main-covariate time series fine-grained interaction features of the gas generator set.

[0017] In particular, the data acquisition module 310 is used to acquire the state parameters of the gas generator set collected by the sensor to obtain a data set of the state parameters. Among them, the state parameters include engine temperature, exhaust temperature, coolant temperature, intake pressure, fuel pressure, lubricating oil pressure, output current value, output voltage value, engine speed value and vibration amplitude value. It should be understood that the operation of the gas generator set is a dynamic process, and its state parameters may change rapidly and continuously. Collecting the state parameters of the gas generator set in a predetermined sampling frequency can ensure that the data points collected each time are evenly distributed in time. By setting an appropriate sampling frequency, transient changes in the operation of the gas generator set can be captured, so as to more accurately reflect the actual working condition of the gas generator set. In particular, the state parameters include engine temperature, exhaust temperature, coolant temperature, intake pressure, fuel pressure, lubricating oil pressure, output current value, output voltage value, engine speed value and vibration amplitude value. Here, the three temperature parameters of engine temperature, exhaust temperature and coolant temperature can reflect the heat exchange inside the engine and whether it is in the normal working temperature range; the three pressure parameters of intake pressure, fuel pressure and lubricating oil pressure have a direct impact on engine performance. Any abnormality in pressure may lead to unstable power output or increased wear of mechanical parts; the output current value and output voltage value are electrical parameters of the gas generator, which are used to evaluate the working status of the generator part, including load distribution and power quality; the speed is an important indicator to measure the engine output power and is also the key feedback information in the control strategy; and the vibration characteristics of the mechanical structure are an important indicator of the health status of the equipment. Excessive vibration may indicate imbalance or other mechanical problems.

[0018] In one example, multiple types of sensors are installed at key locations of a gas generator set to capture various status information during equipment operation. These sensors include temperature sensors (used to measure engine temperature, exhaust temperature, and coolant temperature), pressure sensors (monitoring intake pressure, fuel pressure, and lubricating oil pressure), current and voltage sensors (evaluating the working status of the generator part), speed sensors (measuring engine output power), and vibration sensors (detecting the vibration characteristics of the mechanical structure). Each sensor is precisely arranged according to the specific working characteristics of the gas generator set and the required monitoring accuracy to ensure that it can accurately reflect the actual working conditions of the equipment.

[0019] In order to ensure that the data points collected each time are evenly distributed in time and can more accurately reflect the actual working conditions of the equipment, the system sets a suitable sampling frequency for each type of sensor. For example, for fast-changing signals such as vibration or transient current, a higher sampling rate is used; while for relatively stable parameters such as average temperature, a lower sampling rate is selected. Such settings not only help capture subtle changes in the operation of the equipment, but also reduce the amount of unnecessary data and improve the efficiency of subsequent data analysis.

[0020] The sensor continuously collects the state parameters of the gas generator set during operation at the set sampling frequency and transmits the raw data to the local data acquisition device. During this process, the raw data is pre-processed, such as removing noise and correcting errors, to improve the quality of subsequent analysis. In addition, additional information such as timestamps, environmental conditions and other auxiliary information may be added to better understand the data context. In this way, the system ensures that the acquired data is both comprehensive and reliable, providing a solid foundation for subsequent deep learning algorithms and other advanced analysis tools.

[0021] The preprocessed state parameters are integrated into a structured data set. This data set contains time series data of all relevant variables, each row represents an observation at a time point, and each column corresponds to a specific state parameter. By constructing such a comprehensive data set, the system can not only record the state of the gas generator set at different time points, but also reveal the intrinsic relationship between the various state parameters, providing rich materials for further feature extraction and anomaly detection.

[0022] Finally, the organized data set is securely stored in an efficient database system for subsequent data access, query, and further analysis and processing. Considering that the amount of data may be very large, a distributed file system or relational / non-relational database with good scalability and high read / write performance is usually selected to meet the needs. Through scientific and reasonable data management and storage solutions, the system ensures the security and availability of data, and also lays a good foundation for possible data mining and applications in the future.

[0023] In particular, the data transmission module 320 is used to compress the data set of the state parameter and transmit it to the monitoring server. Considering that the amount of state parameter data generated during the operation of the gas generator set is likely to be too large, resulting in the risk of transmission delay or packet loss when data transmission occurs, in the technical solution of the present application, the data set of the state parameter is compressed to reduce the amount of data to be transmitted, thereby reducing communication costs and bandwidth requirements, especially when using mobile networks such as 5G for remote transmission. Smaller data packets are usually faster and more stable during transmission, which is crucial to ensure the efficiency of data transmission. Then, in order to achieve a large amount of state parameter data can be quickly and timely transmitted from the site to the remote monitoring server to support real-time monitoring of the operating status of the gas generator set, the compressed state parameter data packet is transmitted to the monitoring server through the 5G communication network. The 5G communication network has high reliability, stability and security, which ensures that even in complex and changeable working environments, such as severe weather conditions or remote areas, a stable data transmission connection can be maintained, reducing the risk of data loss or transmission interruption.

[0024] In one example, in order to effectively reduce the amount of data required during transmission while maintaining the quality and integrity of the original data, the system first selected an efficient lossless compression algorithm. Common compression algorithms include Gzip, Bzip2, LZMA, etc., which can significantly reduce the file size without losing any information. For the state parameter data set of a gas generator set, the most suitable compression algorithm is usually selected based on the data characteristics and actual needs. For example, if the data set contains a large amount of repeated or redundant information, using an LZ77-type compression algorithm (such as Gzip) may achieve better compression results; for data with strong structured characteristics, more advanced compression methods such as dictionary-based compression algorithms can be considered. By carefully selecting compression algorithms, the system can minimize the amount of data transmission while ensuring data integrity, thereby reducing communication costs and bandwidth requirements.

[0025] Once the compression algorithm is determined, the next step is to compress the collected state parameter data set. This process is generally completed by the data acquisition device located on site, which is equipped with a dedicated hardware accelerator or software library to perform the compression operation. Before compression, the system will perform necessary preprocessing on the data, such as format conversion, noise removal, etc., to improve compression efficiency and ensure the quality of subsequent analysis. The compressed data is encapsulated into one or more compressed packages, each of which contains time series data from different sensors, and is accompanied by relevant metadata information such as compression method, timestamp, etc., so that the receiving end can correctly decompress and parse. This compression mechanism not only helps to reduce the amount of data transmitted, but also ensures the accuracy and availability of the data after decompression.

[0026] Considering that the amount of state parameter data generated during the operation of the gas generator set may be very large, in order to avoid the risk of delay or packet loss during data transmission, the system uses the 5G communication network as the main transmission channel. With its high bandwidth and low latency, the 5G network is particularly suitable for industrial application scenarios with high real-time requirements. In addition, in order to further improve the transmission performance, the system has also optimized the transmission protocol for specific application scenarios. For example, the TCP / IP protocol is used to ensure the reliability of data transmission, and when large quantities of data need to be transmitted quickly, the UDP protocol may be used to sacrifice a certain degree of reliability in exchange for a higher speed. By flexibly adjusting the transmission protocol, the system can achieve the best data transmission effect under different network conditions, ensuring a stable data flow even in poor network conditions.

[0027] In order to protect the security of data in transmission and prevent sensitive information from being leaked or tampered with, the system has taken a series of encryption measures. First, the compressed data packets are encrypted at the sending end, and strong encryption algorithms such as AES (Advanced Encryption Standard) are used to ensure that even if they are intercepted during network transmission, third parties cannot read or modify the content. Secondly, after receiving the data packet at the receiving end, the corresponding key must be used to decrypt and restore the original data. In addition, in order to verify the authenticity and integrity of the data, the system also applies digital signature technology and checksum mechanism to ensure that only authorized devices and services can participate in data exchange, and each transmitted data can be accurately identified and processed. Through these encryption methods, the system not only ensures the security of data, but also enhances users' trust in the system.

[0028] In order to ensure that a large amount of status parameter data can be quickly and timely transmitted from the site to the remote monitoring server, and support real-time monitoring of the operating status of the gas generator set, the system has designed several key measures to ensure the efficiency and stability of transmission. First, the system implements segmented transmission, that is, dividing large data packets into several small fragments and sending them in sequence to avoid a single large data packet occupying too many network resources and causing congestion. This method can not only improve transmission efficiency, but also reduce the risk of overall data loss due to the loss of a single large packet. Secondly, different transmission priorities are set according to the importance of different types of data to ensure that key data arrives at the destination first. This can maximize the transmission speed of important information within a limited bandwidth and ensure the response timeliness of the monitoring system. Furthermore, the system introduces automatic retransmission request (ARQ), which automatically triggers the retransmission mechanism in the case of poor network conditions or data loss to ensure that all data can eventually be delivered intact and without error. In addition, when there are multiple monitoring servers, the system reasonably allocates the task volume of each server to avoid overloading a server and affecting the overall performance, and improves the robustness and scalability of the entire system through load balancing. Finally, the system has established redundant paths and backup solutions to ensure that even if there is a problem with the main link, it can quickly switch to the backup link to continue transmission without affecting business continuity. The existence of the fault-tolerant mechanism enables the system to maintain efficient and stable operation in the face of unforeseen network failures.

[0029] In particular, the data decompression and sorting module 330 is used to decompress and sort the compressed data packet of the state parameters on the monitoring server to obtain the time queue of the main variable and the time series joint matrix of the covariate. Specifically, first, the compressed data packet of the state parameters is decompressed to obtain the data set of the state parameters; then, the vibration amplitude value is used as the main variable and the engine temperature, the exhaust temperature, the coolant temperature, the intake pressure, the fuel pressure, the lubricating oil pressure, the output current value, the output voltage value and the engine speed value are used as covariates, and the data set of the state parameters is sorted to obtain the time queue of the main variable and the time series joint matrix of the covariate. Here, in order to carry out subsequent analysis and processing, the compressed data packet of the state parameters is decompressed on the monitoring server to obtain the data set of the state parameters. Further, the vibration amplitude value is used as the main variable. Here, it is considered that the imbalance, wear or looseness of mechanical parts often lead to abnormal vibration. Therefore, in the technical solution of the present application, the vibration amplitude is used as the main variable to focus on possible mechanical failures; and the engine temperature, the exhaust temperature, the coolant temperature, the intake pressure, the fuel pressure, the lubricating oil pressure, the output current value, the output voltage value and the engine speed value are used as covariates, and a series of covariates are integrated according to the time dimension and the parameter sample dimension to construct a time series joint matrix that integrates multiple covariates, revealing the complex dynamic behavior inside the gas generator set. In this way, a richer feature representation of the operating status of the gas generator set can be established in the time dimension, thereby supporting more in-depth status monitoring and fault diagnosis.

[0030] Specifically, when the compressed data packet of the state parameters is transmitted to the remote monitoring server through the 5G communication network, the first stage is the decompression stage. The decompression unit on the monitoring server is responsible for receiving the data packets from the site and restoring them to the original state parameter data set using the corresponding decompression algorithm (such as Gzip, Bzip2 or LZMA, etc.). During the decompression process, the server will accurately reconstruct the data sequence collected by each sensor based on the accompanying metadata information (such as compression method, timestamp, etc.). In order to ensure the efficiency and accuracy of decompression, the server is usually equipped with a dedicated hardware accelerator or optimized software library to perform the decompression operation. This means that decompression is not just a simple file expansion, but a process involving many considerations, including data integrity verification, format conversion, and preliminary error detection.

[0031] After decompression, the next step is to organize the data. This step aims to transform the messy raw data into a structured form that is easy to analyze. Specifically, the data organization module classifies, sorts, and aggregates the decompressed state parameter data set according to pre-defined rules and standards, and finally forms two important data structures: the time queue of the main variable and the time series joint matrix of the covariate. These two structures are crucial for subsequent feature extraction, anomaly detection, and model training.

[0032] The main variables refer to those key indicators that can directly reflect the core operating conditions of the gas generator set, such as the vibration amplitude value. Therefore, when constructing the time queue of the main variables, the system will extract all observations related to the vibration amplitude from the decompressed data set and arrange them in chronological order to form a continuous time series. This time queue not only records the trend of the vibration amplitude over time, but also retains the timestamp corresponding to each observation point, so that subsequent analysis can accurately trace back to the equipment status at a specific moment. In addition, in order to facilitate correlation analysis with other covariates, each data point in the time queue will be assigned a unique identifier to ensure that its position in the entire data set is clear and unique.

[0033] Covariates refer to all other influencing factors other than the main variables, which work together to determine the overall performance of the gas generator set, including engine temperature, exhaust temperature, coolant temperature, intake pressure, fuel pressure, lubricating oil pressure, output current value, output voltage value, and engine speed value. In order to fully capture the relationship between these covariates and their impact on the system, the system needs to build a comprehensive time series joint matrix. In this matrix, each row represents a fixed time point, and each column corresponds to a specific covariate. In this way, not only can we intuitively see the values ​​of each covariate at the same time point, but also discover potential trends or patterns by comparing data at different time points. More importantly, such a matrix structure is very suitable for machine learning algorithms, especially those tasks that rely on multidimensional time series data for prediction and diagnosis. In the process of constructing the covariate time series joint matrix, the system must not only ensure the data integrity and consistency of each covariate, but also take into account the relative importance of different covariates, so as to provide high-quality data support for subsequent feature interactions and fine-grained analysis.

[0034] In the process of constructing the above two data structures, the system will also carry out a series of refined data sorting work to ensure the quality and consistency of the data. For example, for missing values ​​or outliers, the system will use interpolation or other statistical methods to fill or correct them; for inconsistent units, they will be uniformly converted to standard units; for duplicate records or redundant information, they will be cleaned and simplified. In addition, in order to facilitate subsequent feature extraction and model training, the system will also add some auxiliary information, such as equipment number, geographic location, etc. Although this information is not directly involved in the core data of the analysis, it can provide additional background support when necessary to help better understand the physical meaning behind the data. Through these meticulous data sorting measures, the system not only improves the quality of the data, but also lays a good foundation for subsequent advanced analysis.

[0035] In particular, the timing correlation feature extraction module 340 is used to extract timing correlation features from the time queue of the main variable and the time series joint matrix of the covariate to obtain the main variable timing correlation implicit features and the covariate timing correlation implicit features. More specifically, in the technical solution of the present application, first, the time queue of the main variable is input into the timing encoder based on the LSTM model to obtain the main variable timing correlation implicit feature vector as the main variable timing correlation implicit feature; it should be understood that the operation process of the gas generator set involves complex nonlinear dynamic characteristics. During the operation, the vibration amplitude of the gas generator often shows a certain periodicity or trend. For example, the vibration of the engine may show regular fluctuations with changes in workload. The LSTM model can better adapt to these nonlinear change patterns and capture their subtle dynamic characteristics through its internal memory unit structure, which is particularly important for accurately predicting faults. Therefore, in the technical solution of the present application, the time queue of the main variable is input into a timing encoder based on the LSTM model to extract the implicit characteristics of the timing correlation of the vibration amplitude parameters during operation. By capturing the complex dynamic characteristics of the vibration amplitude in the time dimension, the operating status of the gas generator set can be better understood, providing support for subsequent abnormal monitoring.

[0036] Furthermore, since the covariate time series joint matrix is ​​constructed by multiple covariates and is a multidimensional time series data set, it is difficult for traditional univariate time series models to effectively process such complex data structures. The self-attention mechanism in the converter structure can reveal the interaction patterns between different covariates and the impact of each covariate on the overall system performance, for example, how changes in engine temperature affect lubricating oil pressure, or the relationship between exhaust temperature and fuel pressure, etc., which is particularly important for capturing long-distance dependencies between covariates. In addition, when the number of covariates is large, the data may become sparse, that is, not all covariates have significant changes at all times. The converter structure can automatically weight different covariates according to the importance of each time step, while ignoring unimportant information, to effectively handle this situation and ignore unimportant information. Therefore, in the technical solution of the present application, the covariate time series joint matrix is ​​input into the covariate time series association feature extractor based on the converter structure to extract the potential time series connection between each covariate, and obtain the covariate time series association implicit feature vector, which is helpful to achieve fine-grained state monitoring and fault diagnosis.

[0037] In particular, the semantic interaction encoding module 350 is used to perform fine-grained feature interaction based on external knowledge guidance on the main variable time series associated implicit features and the covariate time series associated implicit features to obtain the main-covariate time series fine-grained interaction features of the gas generator set. That is, considering that when a gas generator set has abnormal vibration, it is usually accompanied by changes in certain specific covariates (such as excessive temperature or abnormal pressure). Therefore, in order to further combine the time series association information between the main variable and the covariate to mine the subtle change pattern of the gas generator set in the operating state. The present application proposes a fine-grained feature interaction method based on external knowledge guidance, which introduces external knowledge, such as the working principle, failure mode and cause of the gas generator set, to optimize the feature interaction process of the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector, so as to capture the deeper interactive relationship between the main variable and the covariate, thereby strengthening the monitoring of the working state of the gas generator set and improving the accuracy of abnormal diagnosis and prediction. In a specific example of the present application, such as Figure 3As shown, the semantic interaction encoding module 350 includes: a feature fine-grained interaction optimization unit 351, which is used to perform attention optimization on the fine-grained interaction features between the main variable time series association implicit feature vector and the covariate time series association implicit feature vector based on external knowledge to obtain an external knowledge optimized gas generator set main-covariate time series feature fine-grained interaction matrix; a feature guided modulation unit 352, which is used to optimize the gas generator set main-covariate time series feature fine-grained interaction matrix based on the external knowledge, and perform feature modulation optimization on the main variable time series association implicit feature vector and the covariate time series association implicit feature vector respectively to obtain an optimized main variable time series association implicit feature vector and an optimized covariate time series association implicit feature vector; a position-by-position semantic interaction encoding unit 353, which is used to perform position-by-position semantic response encoding on the optimized main variable time series association implicit feature vector and the optimized covariate time series association implicit feature vector to obtain a gas generator set main-covariate time series fine-grained interaction feature vector as the gas generator set main-covariate time series fine-grained interaction feature.

[0038] Specifically, the feature fine-grained interaction optimization unit 351 is used to optimize the fine-grained interaction features between the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector based on external knowledge to obtain the main-covariate time series feature fine-grained interaction matrix of the external knowledge optimized gas generator set. Specifically, first, the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector are input into the fine-grained feature interaction network to understand the time series correlation between the main variable and the covariate from a micro level, capture the fine-grained time series interaction information between the two, and generate the main-covariate time series feature fine-grained interaction matrix. Then, the main-covariate time series feature fine-grained interaction matrix is ​​input into the attention unit based on external knowledge to obtain the main-covariate time series feature fine-grained interaction matrix of the external knowledge optimized gas generator set. That is, the generated main variable-covariate time series feature fine-grained interaction matrix is ​​sent to the attention unit based on external knowledge, and the main variable-covariate time series feature fine-grained interaction matrix is ​​further optimized using external knowledge. In one example, the parameter matrix of the attention unit is trained by introducing external knowledge, such as the working principle of the gas generator set, the failure modes of each group type and their causes, so that the attention unit can adjust the weight according to the external knowledge, more accurately identify and strengthen the highly correlated time series features between the gas main variable and the covariate, and suppress the interference of irrelevant or unimportant features, thereby enhancing the accuracy of fault monitoring of the gas generator. More specifically, based on external knowledge, the fine-grained interaction features between the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector are optimized with the following attention optimization formula to obtain the external knowledge optimized gas generator set main-covariate time series feature fine-grained interaction matrix; wherein, the attention optimization formula is: in, represents the implicit eigenvector of the main variable time series association, represents the implicit eigenvector of the time series association of the covariate, represents the transpose of a vector, represents the matrix multiplication operation, represents the fine-grained interaction matrix of the main-covariate time series features, and represents the learnable memory parameter matrix of the external knowledge-based attention unit, represents the normalization function, Represents the fine-grained interaction matrix of the main-covariate time series characteristics of the gas generator set for external knowledge optimization.

[0039] Specifically, the feature-guided modulation unit 352 is used to optimize the fine-grained interaction matrix of the main-covariate time series characteristics of the gas generator set based on the external knowledge, and perform feature modulation optimization on the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector respectively to obtain the optimized main variable time series associated implicit feature vector and the optimized covariate time series associated implicit feature vector. Specifically, first, the main variable time series associated implicit feature vector is linearly transformed to obtain a first query feature vector and a first value feature vector, and the external knowledge is used to optimize the main-covariate time series feature fine-grained interaction matrix of the gas generator set as a key matrix, and the first query feature vector, the first value feature vector and the key matrix are input into a fine-grained modulation module based on the Transformer structure to obtain the optimized main variable time series associated implicit feature vector; similarly, the covariate time series associated implicit feature vector is linearly transformed to obtain a second query feature vector and a second value feature vector, and the external knowledge is used to optimize the main-covariate time series feature fine-grained interaction matrix of the gas generator set as a key matrix, and the second query feature vector, the second value feature vector and the key matrix are input into the fine-grained modulation module based on the Transformer structure to obtain the optimized covariate time series associated implicit feature vector. That is, the fine-grained interaction matrix of the main-covariate time series features of the gas generator set optimized by external knowledge is used as the key matrix, and the first query feature vector and the first value feature vector are constructed based on the main variable time series associated implicit feature vector, and the second query feature vector and the second value feature vector are constructed based on the covariate time series associated implicit feature vector, and the self-attention mechanism of the Transformer structure is used to realize the information exchange and integration between the internal and external knowledge of the features, so as to ensure that the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector can obtain a deeper level of semantic understanding from external knowledge to improve the quality of their feature expression. More specifically, the fine-grained interaction matrix of the main-covariate time series features of the gas generator set is optimized based on the external knowledge, and the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector are respectively subjected to feature modulation optimization by the following feature modulation optimization to obtain the optimized main variable time series associated implicit feature vector and the optimized covariate time series associated implicit feature vector; wherein, the feature modulation optimization formula is: in, and denote the first query embedding matrix and the first value embedding matrix respectively, and denote the first query feature vector and the first value feature vector respectively, and denote the second query embedding matrix and the second value embedding matrix respectively, and denote the second query feature vector and the second value feature vector respectively, , , and Represent different bias terms, represents the normalized exponential function, represents the characteristic scale value of the fine-grained interaction matrix of the main-covariate time series characteristics of the external knowledge-optimized gas generator set, and They respectively represent the optimized main variable time series associated latent eigenvector and the optimized covariate time series associated latent eigenvector.

[0040] Specifically, the position-by-position semantic interaction encoding unit 353 is used to perform position-by-position semantic response encoding on the optimized main variable time series associated implicit feature vector and the optimized covariate time series associated implicit feature vector to obtain the main-covariate time series fine-grained interaction feature vector of the gas generator set as the main-covariate time series fine-grained interaction feature of the gas generator set. That is, in the technical solution of the present application, the position-by-position feature response interaction is performed by calculating the position point division between the optimized main variable time series associated implicit feature vector and the optimized covariate time series associated implicit feature vector to obtain the main-covariate time series fine-grained interaction feature vector of the gas generator set, thereby more accurately reflecting the fine-grained interaction information of the working state of the gas generator set. More specifically, the optimized main variable time series associated implicit feature vector and the optimized covariate time series associated implicit feature vector are subjected to position-by-position semantic response encoding using the following position-by-position semantic response encoding formula to obtain the main-covariate time series fine-grained interaction feature vector of the gas generator set; wherein, the position-by-position semantic response encoding formula is: in, and They represent the optimized implicit eigenvector of the main variable time series association and the optimized implicit eigenvector of the covariate time series association, respectively. Represents the time series fine-grained interaction feature vector of the main and covariate variables of the gas generator set.

[0041] In particular, the abnormality monitoring module 360 ​​is used to determine whether the working state of the gas generator set is abnormal based on the main-covariate time series fine-grained interaction characteristics of the gas generator set. That is, in the technical solution of the present application, the main-covariate time series fine-grained interaction feature vector of the gas generator set is input into the state monitor based on the classifier to obtain the monitoring result, and the monitoring result is used to indicate whether the working state of the gas generator set is abnormal. In this process, the main task of the classifier is to classify the input main-covariate time series fine-grained interaction feature vector to determine the working state of the gas generator set. In particular, in the technical solution of the present application, the abnormality monitoring principle of the classifier is based on machine learning by training the features under a large number of known states (normal and abnormal) so that the model learns to distinguish the differences between different states. In the training stage, a large amount of historical data needs to be prepared first, which includes the state parameters of the gas generator set in normal operation and when a fault occurs. Through these labeled data sets, the classifier is trained to identify the subtle differences between normal and abnormal states. In this process, cross-validation and other methods will be used to optimize the hyperparameters of the classifier to ensure that it has good generalization ability, that is, it can make accurate judgments on unseen data. Commonly used classification algorithms include support vector machine (SVM), random forest, etc.

[0042] Preferably, the process of passing the gas generator set main-covariate time series fine-grained interactive feature vector through a classifier-based state monitor to obtain a monitoring result includes: The absolute value sum of all eigenvalues ​​of the main-covariate time series fine-grained interaction eigenvector of the gas generator set is calculated to obtain the main-covariate time series fine-grained interaction space structure value of the first gas generator set, and the square root of the sum of squares is calculated to obtain the main-covariate time series fine-grained interaction space structure value of the second gas generator set, that is: in, represents each eigenvalue of the fine-grained interactive eigenvector of the main-covariate time series of the gas generator set, represents the number of eigenvalues ​​of the fine-grained interaction eigenvector of the main-covariate time series of the gas generator set, represents the time series fine-grained interactive spatial structure value of the main-covariate of the first gas-fired generator set, represents the time series fine-grained interactive spatial structure value of the main-covariate of the second gas-fired generator set; For each eigenvalue of the main-covariate time series fine-grained interaction eigenvector of the gas generator set, the product of the eigenvalue and the total number of eigenvalues ​​of the main-covariate time series fine-grained interaction eigenvector of the gas generator set is calculated to obtain the main-covariate time series fine-grained interaction long-range dependency value of the first gas generator set, that is: in, Indicates the long-range dependency value of the main-covariate time series fine-grained interaction of the first gas-fired generator set corresponding to each eigenvalue of the main-covariate time series fine-grained interaction eigenvector of the gas-fired generator set; The square root of the total number of eigenvalues ​​multiplied by the product of the eigenvalues ​​minus the time series fine-grained interaction spatial structure value of the main-covariate of the second gas-fired generator set is calculated to obtain the time series fine-grained interaction long-range dependency value of the main-covariate of the second gas-fired generator set, that is: in, Indicates the second gas generator set main-covariate time series fine-grained interaction long-range dependency value corresponding to each eigenvalue of the gas generator set main-covariate time series fine-grained interaction eigenvector; The index value calculated by taking the time series fine-grained interaction long-range dependency value of the main-covariate of the first gas-fired generator set as the exponent of the natural constant and the inverse of the time series fine-grained interaction long-range dependency value of the main-covariate of the second gas-fired generator set are weighted summed to obtain the optimized eigenvalue corresponding to each eigenvalue, that is: in, represents the optimized eigenvalues ​​corresponding to the eigenvalues ​​of the fine-grained interaction eigenvector of the main-covariate time series of the gas generator set, and is the preset weight; The optimized gas generator set main-covariate time series fine-grained interactive feature vector composed of the optimized feature values ​​is passed through a classifier-based state monitor to obtain a monitoring result.

[0043] Here, when the main variable time-series association implicit feature vector and the covariate time-series association implicit feature vector respectively represent the time-series association implicit features of the vibration amplitude value in the state parameters of the gas generator set and the time-series association implicit features between parameters of other state parameters, when performing feature fine-grained interaction based on external knowledge modulation, the inconsistent external modulation between the main variable and each covariate will cause the final main-covariate time-series fine-grained interaction feature vector of the gas generator set to have a difference in the fusion distribution space structure, affecting the classification regression convergence consistency, and thus affecting the accuracy of the monitoring results obtained by the classifier-based state monitor.

[0044] Based on this, in view of the possible spatial structure loss in the high-dimensional space of the feature set of the main-covariate time series fine-grained interaction feature vector of the gas generator set, which leads to the inconsistent convergence of the implicit inference spatial structure information based on the feature during classification regression, a long-distance feature dependency relationship is established based on the overall feature scale of the main-covariate time series fine-grained interaction feature vector of the gas generator set relative to the spatial structure representation of the main-covariate time series fine-grained interaction feature vector of the gas generator set, so as to establish the feature local connectivity of the main-covariate time series fine-grained interaction feature vector of the gas generator set, and capture the spatial ambiguous information of the object feature value through the unstructured feature value point prediction of the main-covariate time series fine-grained interaction feature vector of the gas generator set, thereby improving the spatial inductive deviation perception ability of the feature set of the main-covariate time series fine-grained interaction feature vector of the gas generator set, improving the convergence consistency of classification regression, and improving the accuracy of the monitoring results obtained by the main-covariate time series fine-grained interaction feature vector of the gas generator set through a classifier-based state monitor.

[0045] As described above, the gas generator set monitoring system 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a gas generator set monitoring algorithm. In a possible implementation, the gas generator set monitoring system 300 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the gas generator set monitoring system 300 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the gas generator set monitoring system 300 can also be one of the many hardware modules of the wireless terminal.

[0046] Alternatively, in another example, the gas generator set monitoring system 300 and the wireless terminal may also be separate devices, and the gas generator set monitoring system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0047] Furthermore, a gas generator set monitoring method is also provided.

[0048] Figure 4 FIG. 1 is a flow chart of a gas generator set monitoring method according to an embodiment of the present application. Figure 4 As shown, according to the gas generator set monitoring method of the embodiment of the present application, the steps include: S1, acquiring the state parameters of the gas generator set collected by the sensor to obtain a data set of the state parameters; S2, compressing the data set of the state parameters and transmitting it to the monitoring server; S3, at the monitoring server, decompressing and sorting the compressed data packet of the state parameters to obtain the time queue of the main variable and the time series joint matrix of the covariate; S4, extracting the time series correlation features of the time queue of the main variable and the time series joint matrix of the covariate to obtain the time series correlation implicit features of the main variable and the time series correlation implicit features of the covariate; S5, performing fine-grained feature interaction based on external knowledge guidance on the time series correlation implicit features of the main variable and the time series correlation implicit features of the covariate to obtain the main-covariate time series fine-grained interaction features of the gas generator set; S6, based on the main-covariate time series fine-grained interaction features of the gas generator set, determining whether the working state of the gas generator set is abnormal.

[0049] In summary, the gas generator set monitoring method according to the embodiment of the present application is explained, which collects various state parameters of the gas generator set during operation through a sensor group and transmits them to a remote monitoring server, which is organized into a format suitable for machine learning analysis at the server terminal, and further uses a machine learning algorithm based on deep learning and data processing technology to perform time series correlation analysis on the state parameters to dig out the time series dynamic correlation pattern between various state parameters during the operation of the gas generator set, so as to intelligently determine whether there is an abnormality in the working state of the gas generator set, and realize efficient monitoring of the working state of the gas generator set. In this way, the remote monitoring capability of the fault mode of the gas generator set can be improved, which helps to timely understand the operating status of the gas engine set to quickly respond to any potential problems, and at the same time, it also provides a solid technical foundation for remote monitoring technology.

[0050] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A gas generator set monitoring system, characterized in that: include: A data acquisition module, used to acquire the state parameters of the gas generator set collected by the sensor to obtain a data set of the state parameters; A data transmission module, used for compressing the data set of the state parameters and transmitting it to the monitoring server; A data decompression and sorting module is used to decompress and sort the compressed data packets of the state parameters on the monitoring server to obtain the time queue of the main variable and the time series joint matrix of the covariate; A time series correlation feature extraction module is used to extract time series correlation features from the time queue of the main variable and the time series joint matrix of the covariate to obtain time series correlation implicit features of the main variable and time series correlation implicit features of the covariate; A semantic interaction encoding module, used for performing fine-grained feature interaction based on external knowledge guidance on the main variable time series associated implicit features and the covariate time series associated implicit features to obtain the main-covariate time series fine-grained interaction features of the gas generator set; The abnormality monitoring module is used to determine whether there is an abnormality in the working state of the gas generator set based on the fine-grained interaction characteristics of the main and covariate time series of the gas generator set.

2. The gas generator set monitoring system according to claim 1, characterized in that: The data acquisition module is used to: A sensor group is used to collect state parameters of a gas generator set at a predetermined sampling frequency to obtain a data set of the state parameters, wherein the state parameters include engine temperature, exhaust temperature, coolant temperature, intake pressure, fuel pressure, lubricating oil pressure, output current value, output voltage value, engine speed value and vibration amplitude value.

3. The gas generator set monitoring system according to claim 2, characterized in that: The data transmission module comprises: A data compression unit, used for compressing the data set of the state parameters to obtain a state parameter compressed data packet; A data transmission unit is used to transmit the compressed data packet of the status parameter to the monitoring server via the 5G communication network.

4. The gas generator set monitoring system according to claim 3, characterized in that: The data decompression and sorting module includes: A decompression unit, used for decompressing the compressed data packet of the state parameter to obtain a data set of the state parameter; A data sorting unit is used to sort the data set of the state parameters using the vibration amplitude value as the main variable and the engine temperature, the exhaust temperature, the coolant temperature, the intake pressure, the fuel pressure, the lubricating oil pressure, the output current value, the output voltage value and the engine speed value as covariates to obtain a time queue of the main variable and a time series joint matrix of the covariates.

5. The gas generator set monitoring system according to claim 4, characterized in that: The temporal correlation feature extraction module comprises: A main variable time series correlation feature extraction unit, used for inputting the time queue of the main variable into a time series encoder based on the LSTM model to obtain a main variable time series correlation implicit feature vector as the main variable time series correlation implicit feature; The covariate time series correlation feature extraction unit is used to input the covariate time series joint matrix into the covariate time series correlation feature extractor based on the converter structure to obtain the covariate time series correlation implicit feature vector as the covariate time series correlation implicit feature.

6. The gas generator set monitoring system according to claim 5, characterized in that: The semantic interaction encoding module comprises: A feature fine-grained interaction optimization unit is used to optimize the fine-grained interaction features between the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector based on external knowledge to obtain an external knowledge optimized gas generator set main-covariate time series feature fine-grained interaction matrix; A feature guided modulation unit is used to optimize the main-covariate time series feature fine-grained interaction matrix of the gas generator set based on the external knowledge, and perform feature modulation optimization on the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector respectively to obtain an optimized main variable time series associated implicit feature vector and an optimized covariate time series associated implicit feature vector; The position-by-position semantic interaction encoding unit is used to perform position-by-position semantic response encoding on the optimized main variable time series associated implicit feature vector and the optimized covariate time series associated implicit feature vector to obtain the main-covariate time series fine-grained interaction feature vector of the gas generator set as the main-covariate time series fine-grained interaction feature of the gas generator set.

7. The gas generator set monitoring system according to claim 6, characterized in that: The feature fine-grained interactive optimization unit includes: A fine-grained feature interaction subunit, used for inputting the main variable time series associated implicit feature vector and the covariate time series associated implicit feature vector into a fine-grained feature interaction network to obtain a main-covariate time series feature fine-grained interaction matrix; The external knowledge attention optimization subunit is used to input the main-covariate time series feature fine-grained interaction matrix into the attention unit based on external knowledge to obtain the external knowledge optimized gas generator set main-covariate time series feature fine-grained interaction matrix.

8. The gas generator set monitoring system according to claim 7, characterized in that: The feature-guided modulation unit comprises: A main variable feature fine-grained modulation subunit is used to perform a linear transformation on the main variable time series associated implicit feature vector to obtain a first query feature vector and a first value feature vector, and use the external knowledge to optimize the main-covariate time series feature fine-grained interaction matrix of the gas generator set as a key matrix, and input the first query feature vector, the first value feature vector and the key matrix into a fine-grained modulation module based on a Transformer structure to obtain the optimized main variable time series associated implicit feature vector; The covariate feature fine-grained modulation subunit is used to perform a linear transformation on the covariate time series associated implicit feature vector to obtain a second query feature vector and a second value feature vector, and use the external knowledge to optimize the main-covariate time series feature fine-grained interaction matrix of the gas generator set as a key matrix, and input the second query feature vector, the second value feature vector and the key matrix into the Transformer structure-based fine-grained modulation module to obtain the optimized covariate time series associated implicit feature vector.

9. The gas generator set monitoring system according to claim 8, characterized in that: The abnormality monitoring module is used to: The main-covariate time series fine-grained interactive feature vector of the gas generator set is input into a classifier-based state monitor to obtain a monitoring result, and the monitoring result is used to indicate whether there is an abnormality in the working state of the gas generator set.

10. A gas generator set monitoring method, characterized in that: include: Acquiring state parameters of the gas generator set collected by the sensor to obtain a data set of state parameters; Compressing the data set of the state parameters and transmitting it to the monitoring server; On the monitoring server, the compressed data packets of the state parameters are decompressed and data is sorted to obtain the time queue of the main variables and the time series joint matrix of the covariates; Extracting time series correlation features from the time queue of the main variable and the time series joint matrix of the covariate to obtain time series correlation implicit features of the main variable and time series correlation implicit features of the covariate; Performing fine-grained feature interaction based on external knowledge guidance on the main variable time series associated implicit features and the covariate time series associated implicit features to obtain the main-covariate time series fine-grained interaction features of the gas generator set; Based on the time series fine-grained interaction characteristics of the main and covariates of the gas-fired generator set, it is determined whether the working state of the gas-fired generator set is abnormal.

Images