Remote diagnosis and maintenance system and method for auxiliary engine control system of thermal power plant

Abstract

The invention relates to the technical field of fault diagnosis, and particularly discloses a remote diagnosis and maintenance system and method for a thermal power plant auxiliary machine control system, and the method comprises the steps: firstly collecting point location data of a power plant auxiliary network air compressor in real time, carrying out the time-parameter curve fitting of all parameters in the data based on a local time domain after the data is cleaned, and carrying out the remote diagnosis and maintenance of the thermal power plant auxiliary machine control system; the method comprises the following steps: firstly, acquiring time-parameter curves of various parameters in each local time domain, splicing the time-parameter curves of the various parameters in each local time domain into a multi-dimensional curve graph, and then, carrying out feature extraction and significance aggregation based on time sequence context perception on each multi-dimensional curve graph so as to excavate long-term time sequence relevance and potential fault features of the multi-dimensional parameters of the air compressor; therefore, the fault of the auxiliary engine control system of the thermal power plant can be accurately predicted. In this way, multi-dimensional parameter time sequence information of a long time span can be comprehensively considered, the influence of instantaneous interference and abnormal values on fault prediction is reduced, and the stability and accuracy of fault prediction are further improved.

Description

Technical Field

[0001] This application relates to the technical field of fault diagnosis, and more specifically, to a remote diagnosis and maintenance system and method for an auxiliary control system of a thermal power plant. Background Art

[0002] In the operation and management of thermal power plants, the auxiliary control system, as a key component supporting the stable operation of power generation equipment, its reliability and efficiency are directly related to the overall operation benefit and safety performance of the power plant. With the continuous progress of industrial automation and intelligent technologies, higher requirements are put forward for the monitoring and maintenance of the auxiliary control system. Most traditional fault detection and diagnosis methods rely on manual inspections and experience judgments, which are not only inefficient but also difficult to capture the early signals of faults, resulting in a lag in fault response, increased maintenance costs, and downtime.

[0003] In this regard, the invention patent with the publication number CN116859853A proposes an intelligent auxiliary control DCS system for power plants, which collects the point data of the air compressor in the auxiliary network of the power plant in real time, and performs data cleaning, local time period division, curve fitting and splicing of various parameters on the collected point data to obtain the multi-dimensional curve graphs of each local time period, and then predicts the fault information of the air compressor by extracting features from the multi-dimensional curve graphs.

[0004] In the prior art, deep learning technology has been used to achieve automatic prediction of faults in the auxiliary control system, effectively improving the accuracy and efficiency of fault prediction. However, when the prior art performs fault prediction, it uses a neural network model to perform independent feature pattern recognition on the multi-dimensional curve features of each local time period to obtain the fault prediction result. Since some faults in the auxiliary control system are gradually evolving dynamic processes, if only short-term data is considered for fault prediction while ignoring potential problems spanning multiple time periods, it may be affected by instantaneous interference or outliers and generate false alarms. Especially for slowly developing fault modes, the prediction accuracy may be limited due to the lack of in-depth mining of global time series information.

[0005] Therefore, an optimized remote diagnosis and maintenance system and method for an auxiliary control system of a thermal power plant are expected. Summary of the Invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a remote diagnosis and maintenance system and method for the auxiliary control system of a thermal power plant, which first collects the point data of the auxiliary network air compressor of the power plant in real time, cleans the data, and then performs time-parameter curve fitting based on the local time domain on the various parameters in the data, and splices the time-parameter curves of the various parameters in each local time domain into a multi-dimensional curve graph. Then, feature extraction and significance aggregation based on temporal context perception are performed on each multi-dimensional curve graph to mine the long-term temporal correlation and potential fault characteristics of the multi-dimensional parameters of the air compressor, thereby realizing accurate prediction of the fault of the auxiliary control system of the thermal power plant. In this way, the multi-dimensional parameter time series information with a long time span can be comprehensively considered, the influence of instantaneous interference and outliers on fault prediction can be reduced, and the stability and accuracy of fault prediction can be further improved.

[0007] Accordingly, according to one aspect of the present application, a remote diagnosis and maintenance method for a thermal power plant auxiliary equipment control system is provided, comprising:

[0008] Real-time collection of point data of each air compressor in the power plant auxiliary network, including various parameters of the real-time operation of the air compressor;

[0009] Performing data cleaning on the point data to obtain cleaned point data;

[0010] The parameters in the cleaned point data are divided according to the set time interval to obtain the time period data of each parameter;

[0011] Fitting the time period data of each parameter into a curve, and splicing the curves of each parameter in each time period into a multidimensional curve graph containing the curves of each parameter to obtain a time series of the multidimensional curve graph;

[0012] Performing feature extraction processing on each multidimensional curve graph in the time series of the multidimensional curve graph to obtain a time series of multidimensional feature tensor graphs;

[0013] Performing graph walk autocorrelation gated aggregation on the time series of the multidimensional feature tensor graph to obtain a multidimensional feature tensor time series significant aggregation feature vector;

[0014] A fault prediction result is determined based on the multi-dimensional feature tensor time series significant aggregation feature vector.

[0015] According to another aspect of the present application, a remote diagnosis and maintenance system for a thermal power plant auxiliary equipment control system is provided, comprising:

[0016] The data acquisition module is used to collect the point data of each air compressor in the auxiliary network of the power plant in real time. The point data includes various parameters of the real-time operation of the air compressor;

[0017] A data cleaning module, configured to clean the point data to obtain cleaned point data;

[0018] The data local time domain segmentation module is used to segment the various parameters in the point data after data cleaning according to the set time interval to obtain the time period data of each parameter;

[0019] A curve fitting module is used to fit the time period data of each parameter into a curve, and to splice the curves of each parameter in each time period into a multidimensional curve graph containing the curves of each parameter to obtain a time series of the multidimensional curve graph;

[0020] A multidimensional curve feature extraction module, configured to perform feature extraction processing on each multidimensional curve graph in the time series of the multidimensional curve graph to obtain a time series of multidimensional feature tensor graphs;

[0021] An autocorrelation gated aggregation module is used to perform graph walk autocorrelation gated aggregation on the time series of the multidimensional feature tensor graph to obtain a multidimensional feature tensor time series significant aggregation feature vector;

[0022] A fault prediction module is used to determine a fault prediction result based on the multi-dimensional feature tensor time series significant aggregation feature vector.

[0023] Compared with the existing technology, the remote diagnosis and maintenance system and method of the auxiliary control system of a thermal power plant provided by this application first collects the point data of the auxiliary network air compressor of the power plant in real time, cleans the data, and then fits the time-parameter curve of each parameter in the data based on the local time domain, and splices the time-parameter curves of each parameter in each local time domain into a multidimensional curve graph. Then, feature extraction and significance aggregation based on time series context perception are performed on each multidimensional curve graph to mine the long-term time series correlation and potential fault characteristics of the multidimensional parameters of the air compressor, thereby realizing accurate prediction of the fault of the auxiliary control system of the thermal power plant. In this way, the multidimensional parameter time series information with a long time span can be comprehensively considered, the influence of instantaneous interference and outliers on fault prediction can be reduced, and the stability and accuracy of fault prediction can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended 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 drawings, the same reference numerals generally represent the same components or steps.

[0025] Figure 1 Flowchart of a remote diagnosis and maintenance method for a thermal power plant auxiliary equipment control system according to an embodiment of the present application.

[0026] Figure 2 Schematic diagram of data flow for the remote diagnosis and maintenance method of the auxiliary equipment control system of a thermal power plant according to an embodiment of the present application.

[0027] Figure 3 Flowchart of step S6 in the remote diagnosis and maintenance method of the auxiliary equipment control system of a thermal power plant according to an embodiment of the present application.

[0028] Figure 4 Block diagram of the remote diagnosis and maintenance system of the auxiliary equipment control system of a thermal power plant according to an embodiment of the present application. Detailed implementation manners

[0029] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0030] Figure 1 Flowchart of the remote diagnosis and maintenance method of the auxiliary equipment control system of a thermal power plant according to an embodiment of the present application. Figure 2 Schematic diagram of data flow for the remote diagnosis and maintenance method of the auxiliary equipment control system of a thermal power plant according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the remote diagnosis and maintenance method of the auxiliary equipment control system of a thermal power plant according to an embodiment of the present application includes the steps of: S1, collecting in real time the point data of each air compressor in the auxiliary network of the power plant, where the point data includes various parameters of the real-time operation of the air compressor; S2, performing data cleaning on the point data to obtain the point data after data cleaning; S3, dividing the various parameters in the point data after data cleaning at set time intervals to obtain the time period data of each parameter; S4, fitting the time period data of each parameter into a curve, and splicing the curves of the various parameters within each time period into a multi-dimensional curve graph containing the curves of each parameter to obtain the time series of the multi-dimensional curve graph; S5, performing feature extraction processing on each multi-dimensional curve graph in the time series of the multi-dimensional curve graph to obtain the time series of the multi-dimensional feature tensor graph; S6, performing graph walk self-correlation gated aggregation on the time series of the multi-dimensional feature tensor graph to obtain the multi-dimensional feature tensor time series significant aggregation feature vector; S7, determining the fault prediction result based on the multi-dimensional feature tensor time series significant aggregation feature vector.

[0031] In the above remote diagnosis and maintenance method for the auxiliary equipment control system of a thermal power plant, in step S1, the point data of each air compressor in the auxiliary network of the power plant is collected in real time, and the point data includes various parameters of the real-time operation of the air compressor. It should be understood that the various operating parameters of the air compressor are key information reflecting its operating state. By using devices such as sensors and monitoring and collecting parameters such as the current, exhaust pressure, exhaust temperature, outlet pressure, outlet temperature, and outlet dew point of the air compressor at a certain sampling frequency, the operating state information of the air compressor can be obtained comprehensively and in a timely manner, providing basic data for subsequent diagnosis and maintenance, so as to facilitate the discovery of potential operating problems and avoid the occurrence or expansion of faults.

[0032] In specific implementation, to ensure the accuracy, integrity, and timeliness of the data, first, at the hardware level, various high-precision sensors need to be carefully selected and installed. For the monitoring of the air compressor current, a Hall current sensor is usually selected. This type of sensor is based on the Hall effect principle. When current passes through a current-carrying conductor, a magnetic field will be generated around the conductor. The Hall element will generate a Hall potential proportional to the current magnitude in this magnetic field. By measuring this potential, the current value can be accurately obtained. It has the advantages of high precision, fast response speed, and good linearity, and can meet the requirements of real-time collection. During installation, it needs to be connected in series to the power supply line of the air compressor to ensure that the current change during the operation of the air compressor can be accurately measured.

[0033] The measurement of the exhaust pressure and outlet pressure of the air compressor mainly relies on pressure sensors, and common ones are piezoresistive pressure sensors. Its working principle is to utilize the piezoresistive effect of semiconductor materials. When pressure acts on the sensitive element of the sensor, it will cause a change in its resistance value. By measuring the change in the resistance value and passing it through a corresponding conversion circuit, the pressure value can be obtained. During the installation process, the pressure sensor needs to be installed on the exhaust port and outlet pipeline of the air compressor, and it is necessary to ensure that the installation position of the sensor can accurately reflect the pressure situation and avoid pressure fluctuations or measurement errors. For example, it is necessary to ensure that the sensor is installed on the straight pipe section of the pipeline and away from components such as valves, elbows, and tees that may cause pressure disturbances. At the same time, the tightness of the connection between the sensor and the pipeline needs to be ensured to prevent gas leakage from affecting the measurement accuracy.

[0034] Temperature sensors are crucial for collecting the exhaust temperature and outlet temperature of air compressors. Generally, thermocouple or thermal resistance temperature sensors are used. A thermocouple is based on the Seebeck effect, that is, when a closed loop is composed of two conductors of different materials and the two junctions are at different temperatures, a thermoelectric potential will be generated in the loop, and the temperature can be deduced by measuring this thermoelectric potential. A thermal resistance measures temperature by utilizing the characteristic that the resistance value of a metal or semiconductor material changes with temperature. During installation, the temperature sensor should be in close contact with the exhaust pipe and outlet pipe of the air compressor, and heat insulation measures should be taken to prevent heat dissipation or interference from the external ambient temperature, so as to ensure that the measured temperature can truly reflect the internal operating temperature of the air compressor.

[0035] For measuring the outlet dew point of an air compressor, a dew point sensor is usually adopted. Its principle is to use capacitive or optical measurement methods to determine the dew point temperature by measuring the water vapor content in the gas. During installation, it is necessary to ensure that the sensor can accurately collect the sample of the outlet gas, and it should be installed in a dry and clean environment to avoid the influence of impurities such as dust and oil on the measurement results.

[0036] At the software level, an efficient data acquisition system needs to be constructed. This system should have strong data reception and processing capabilities, be able to receive analog signals or digital signals from various sensors in real time, and convert them into digital formats that can be recognized and processed by a computer. For example, a data acquisition card is used to collect and convert the signals of the sensors. The data acquisition card usually has multiple analog input channels and digital input channels, can receive data from multiple sensors simultaneously, and convert analog signals into digital signals through high-speed A / D conversion.

[0037] At the same time, the data acquisition system also needs to have a real-time clock function to add accurate timestamps to the collected data, so as to clearly understand the changes of each parameter at different time points and discover potential fault trends. For example, a high-precision clock chip is adopted to ensure that the accuracy of the timestamp can reach the millisecond or even microsecond level to meet the requirements for monitoring the real-time operating status of the air compressor.

[0038] In addition, to ensure the stability and reliability of data transmission, industrial Ethernet or fieldbus technology is usually adopted for data transmission. Industrial Ethernet has the advantages of high speed, reliability, and easy expansion, and can meet the real-time transmission requirements of a large amount of data. The fieldbus has the characteristics of simplicity, low cost, and strong anti-interference ability, and is suitable for some occasions with high real-time requirements and relatively harsh environments. During data transmission, appropriate communication protocols such as Modbus TCP, Profibus, etc. should be adopted to ensure that data can be accurately transmitted from the sensor to the upper computer or data processing center.

[0039] In terms of the frequency setting of data acquisition, it is necessary to make a reasonable selection according to the operating characteristics of the air compressor and the requirements of fault diagnosis. Generally speaking, for some parameters that change relatively slowly, such as the outlet pressure and outlet temperature of the air compressor, the acquisition frequency can be appropriately reduced, for example, once every 10 seconds; while for some parameters that change relatively quickly and are more critical for fault diagnosis, such as the air compressor current and exhaust temperature, the acquisition frequency needs to be increased, and it can even reach several times per second. By setting such a differentiated acquisition frequency, while ensuring the effectiveness of the data, the pressure of data storage and processing can be reduced.

[0040] In addition, in order to ensure the long-term stable operation of the data acquisition system, regular maintenance and calibration are also required. For sensors, they need to be cleaned and inspected regularly to prevent impurities such as dust and oil stains from affecting the performance of the sensors. At the same time, according to the calibration cycle of the sensors, use standard calibration equipment to calibrate the sensors to ensure that their measurement accuracy always meets the requirements. For the hardware devices of the data acquisition system, such as data acquisition cards, communication modules, etc., regular inspections and maintenance should be carried out to detect and handle possible hardware failures in a timely manner. In terms of software, the data acquisition program needs to be optimized and updated regularly to adapt to the changing operating environment and new functional requirements.

[0041] During the process of data acquisition, it is also necessary to consider the security and confidentiality of the data. Since the operating data of the air compressor may involve important information such as the production safety and operating efficiency of the thermal power plant, corresponding security measures should be taken, such as data encryption and access control. For example, use encryption protocols such as SSL / TLS to encrypt the data transmission to prevent the data from being stolen or tampered with during the transmission process. In terms of data storage, strict access permissions should be set, and only authorized personnel can access and process the acquired data to ensure the security and confidentiality of the data.

[0042] In the above remote diagnosis and maintenance method of the auxiliary control system of the thermal power plant, in step S2, the point data is subjected to data cleaning to obtain the point data after data cleaning. It should be understood that in the actual operating environment, due to the existence of various interference factors, the acquired data may contain some outliers or incorrect data, which will affect the accuracy of subsequent analysis. By data cleaning, these interferences can be excluded and the data quality can be improved. In specific implementation, thresholds can be set according to the normal operating ranges of the parameters of the air compressor, and the parameter values exceeding the threshold range are regarded as outliers and cleared. For example, for the exhaust temperature of the air compressor, according to its equipment manual and operating experience, the normal range is set to 80°C - 100°C, and the data outside this range may be caused by sensor abnormalities or transmission noise and should be excluded.

[0043] In the above remote diagnosis and maintenance method for the auxiliary control system of a thermal power plant, in step S3, the various parameters in the point data after data cleaning are segmented at set time intervals to obtain the time period data for each parameter. It should be understood that by segmenting the operation data with a long time span at certain time intervals, the change trends of various parameters in different time periods can be better analyzed, which helps to capture more subtle parameter fluctuations and discover potential fault hazards. At the same time, it also facilitates subsequent data processing and model training, reducing the complexity of the data. In specific implementation, an appropriate time interval can be selected according to the size of the data volume and the analysis requirements, and the continuous operation data is divided into multiple time periods. The parameter data within each time period is processed as an independent data set. For example, if the segmentation is performed at hourly intervals, 24 data sets can be obtained in a day, and each data set reflects the operation of the air compressor within that hour.

[0044] In the above remote diagnosis and maintenance method for the auxiliary control system of a thermal power plant, in step S4, the time period data for each parameter is fitted into a curve, and the curves of the various parameters within each time period are spliced together to form a multi-dimensional curve graph containing the curves of each parameter to obtain the time series of the multi-dimensional curve graph. It should be understood that by fitting the parameter data into curves, the change law of the parameters over time can be intuitively displayed, so as to better mine potential fault characteristics based on the time series change trends of the various parameters. At the same time, in actual operation, the various parameters of the auxiliary control system are interrelated and jointly reflect the operation state of the system. Therefore, by splicing the curves of the various parameters in the same time period into a multi-dimensional curve graph, it helps to comprehensively analyze the mutual relationship between the various parameters and understand the operation state of the air compressor more comprehensively. In specific implementation, for each parameter within each time period, a data fitting algorithm (such as the least squares method, etc.) is used to fit multiple parameter value points into a time-domain curve of time and parameter values. Then, through frequency-domain conversion (such as the FFT algorithm), the time-domain curve is converted into the frequency-domain region, and the frequency-domain curves of the various parameters are spliced in the same channel graph according to different channels to form a multi-dimensional curve graph. For different time periods, the above operations are repeated, so as to obtain the time series of the multi-dimensional curve graph. In this way, it helps to more effectively represent the change trends of the various parameters and their mutual relationships within a specific time period, thus providing more reliable data support for subsequent fault diagnosis.

[0045] In the above method for remote diagnosis and maintenance of the auxiliary control system of a thermal power plant, in step S5, feature extraction processing is performed on each multi-dimensional curve graph in the time series of the multi-dimensional curve graphs to obtain a time series of multi-dimensional feature tensor graphs. It should be understood that the multi-dimensional curve graphs contain various operation parameter information of the air compressor during a certain local time period. In order to extract the temporal dynamic change characteristics of various operation parameters and the correlation between parameters from them, so as to reflect the operation state and potential fault characteristics of the air compressor, it is necessary to further perform feature extraction processing on the multi-dimensional curve graphs. In a specific implementation, for the feature extraction processing of the multi-dimensional curve graphs, the method disclosed in Patent CN116859853A can be adopted. First, the direct current is removed from each parameter curve in the multi-dimensional curve graph to eliminate the influence of the direct current component on feature extraction, and a direct-current-removed multi-dimensional curve graph is formed. Then, the direct-current-removed multi-dimensional curve graph is converted into a wavelet time-frequency graph, and the time-frequency analysis characteristics of wavelet transform are used to better capture the local features of the signal. Finally, the gray-scale image pixels of the wavelet time-frequency graph are extracted, and the image information is converted into numerical features to obtain a multi-dimensional feature tensor graph. Correspondingly, other effective feature extraction methods can also be adopted, such as using a convolutional neural network model to perform feature learning on the multi-dimensional curve graphs, automatically extracting the key features that can reflect the operation state of the air compressor, and generating a multi-dimensional feature tensor graph. However, this is not limited to the present application.

[0046] In the above method for remote diagnosis and maintenance of the auxiliary control system of a thermal power plant, in step S6, graph-walking autocorrelation gated aggregation is performed on the time series of the multi-dimensional feature tensor graphs to obtain a multi-dimensional feature tensor time-series significant aggregation feature vector. In particular, considering that although the multi-dimensional parameter feature extraction method based on local time domain can capture more subtle parameter fluctuations, it is difficult to grasp the operation state of the air compressor and its change trend as a whole, and it is easily interfered by noise information, resulting in insufficient robustness of fault prediction. To this end, in order to effectively capture the long-distance dependence relationship and global time-series structure information between parameters, the present application proposes a graph-walking autocorrelation gated aggregation method, which represents the relationship between multi-dimensional parameter features in each local time domain by constructing a graph structure, uses the graph-walking algorithm to search and traverse in the graph structure, and combines the autocorrelation gated mechanism to automatically learn the correlation and importance weights between features at different time steps, so as to aggregate and screen the time series of the multi-dimensional feature tensor graphs, thereby obtaining a feature representation that can reflect the overall operation state of the air compressor and its significant time-series change trend.

[0047] Figure 3 It is a flowchart of step S6 in the method for remote diagnosis and maintenance of the auxiliary control system of a thermal power plant according to an embodiment of the present application. As Figure 3As shown, the step S6 includes: S61, performing feature flattening processing on each multidimensional feature tensor graph in the time series of the multidimensional feature tensor graph to obtain a time series of multidimensional parameter feature vectors; S62, performing graph walking context-aware enhancement on the time series of the multidimensional parameter feature vectors to obtain a time series of multidimensional parameter time series context-aware enhancement feature vectors; S63, based on the time series of the multidimensional parameter time series context-aware enhancement feature vectors, performing significance analysis and aggregation on the time series of the multidimensional parameter feature vectors to obtain the multidimensional feature tensor time series significant aggregation feature vector.

[0048] Specifically, the step S62 includes: first, inputting each multidimensional parameter feature vector in the time series of the multidimensional parameter feature vector into a hyperbolic space mapper to obtain a time series of the multidimensional parameter feature vector after hyperbolic space mapping, which is expressed as follows:

[0049] V={v1,v2,...,v i ,...,v n},

[0050] h i =W1v i W2,

[0051] Wherein, V represents the time series of the multidimensional parameter feature vector, v1, v2, v i and v n represents the first, second, i-th and n-th multidimensional parameter feature vectors in the time series of the multidimensional parameter feature vectors, n is the number of the multidimensional parameter feature vectors, W1 and W2 represent the first linear mapping matrix and the second linear mapping matrix, respectively, h i Represents the i-th multidimensional parameter feature vector after hyperbolic space mapping in the time series of multidimensional parameter feature vectors after hyperbolic space mapping.

[0052] Here, each multidimensional feature tensor graph is first flattened to convert it into a vector form to facilitate the subsequent graph walk context-aware enhancement processing. Next, given that the volume of the hyperbolic space grows exponentially compared to the Euclidean space, this characteristic enables it to better capture and represent data sets with a hierarchical structure. Based on this, in order to strengthen the correlation between the multidimensional parameter features in each local time domain and mine their potential hierarchical structure information, the present application first maps the time series of the multidimensional parameter feature vectors into the hyperbolic space, so as to maintain the hierarchical structure and long-tail distribution state in the high-dimensional data with the help of the unique properties of hyperbolic geometry, thereby enhancing the expressiveness of the relative distance between the multidimensional parameter time series features.

[0053] Next, extract the topological structure features of the time series of the multi-dimensional parameter feature vectors after hyperbolic space mapping to obtain the multi-dimensional parameter time series graph walking topological feature matrix. More specifically, first, calculate the Poincaré distance between any two multi-dimensional parameter feature vectors after hyperbolic space mapping in the time series of the multi-dimensional parameter feature vectors after hyperbolic space mapping to obtain the multi-dimensional parameter time series graph walking topological matrix; then, perform dilated convolution encoding on the multi-dimensional parameter time series graph walking topological matrix to obtain the multi-dimensional parameter time series graph walking topological feature matrix, which is expressed by the formula:

[0054]

[0055] D t =DilatedConv(D),

[0056] where h j represents the j-th multi-dimensional parameter feature vector after hyperbolic space mapping in the time series of the multi-dimensional parameter feature vectors after hyperbolic space mapping, ‖·‖ represents calculating the norm of a vector, arccosh(·) represents the inverse cosine function, d P (·,·) represents the Poincaré distance metric function, D represents the multi-dimensional parameter time series graph walking topological matrix, D ij represents the element value at the (i,j) position in the multi-dimensional parameter time series graph walking topological matrix, DilatedConv(·) represents the dilated convolution operation, and D t represents the multi-dimensional parameter time series graph walking topological feature matrix.

[0057] That is, by calculating the Poincaré distance between each multi-dimensional parameter feature vector after hyperbolic space mapping, the relative semantic proximity between each multi-dimensional parameter time series feature is measured, and the multi-dimensional parameter time series graph walking topological matrix is constructed to reveal the global semantic topological structure of the multi-dimensional parameter time series features. Subsequently, the dilated convolution encoding technology is used to perform dilated convolution processing on the multi-dimensional parameter time series graph walking topological matrix, so as to capture the latent semantic dependence relationship between the multi-dimensional parameter time series features in the long-distance range by virtue of the sparse connection characteristics of the dilated convolution, thereby further enhancing its feature expression ability.

[0058] Then, input the multi-dimensional parameter time series graph walking topological feature matrix and the time series of the multi-dimensional parameter feature vectors after hyperbolic space mapping into the global context walking encoder based on the graph convolutional neural network model to obtain the time series of the multi-dimensional parameter time series context-aware enhanced feature vectors, which is expressed by the formula:

[0059]

[0060] where GCN(·,·) represents the graph convolutional neural network, s i represents the vi The corresponding multi-dimensional parameter time-series context-aware enhanced feature vector.

[0061] Here, each multi-dimensional parameter feature vector is used as a node in the graph structure, and the multi-dimensional parameter time-series graph walk topological feature matrix is used as the edge of the graph walk. By performing a walk on the entire graph structure through graph convolution operations, information transfer and accumulation are carried out among the time-series features of each multi-dimensional parameter, so as to make full use of the global time-series context information of the multi-dimensional parameters to enhance their feature representations, thereby extracting a time series of multi-dimensional parameter time-series context-aware enhanced feature vectors with rich context information.

[0062] Specifically, step S63 includes: First, measure the feature correlation between the time series of the multi-dimensional parameter time-series context-aware enhanced feature vectors and the time series of the multi-dimensional parameter feature vectors to obtain a time series of multi-dimensional parameter time-series autocorrelation gating significant confidence factors. More specifically, input each group of corresponding multi-dimensional parameter time-series context-aware enhanced feature vectors and multi-dimensional parameter feature vectors in the time series of the multi-dimensional parameter time-series context-aware enhanced feature vectors and the time series of the multi-dimensional parameter feature vectors into the autocorrelation gating unit to obtain the time series of the multi-dimensional parameter time-series autocorrelation gating significant confidence factors, which is expressed by the formula:

[0063]

[0064] where c i represents the corresponding multi-dimensional parameter time-series autocorrelation gating significant confidence factor of v i , g(·,·) represents the autocorrelation gating unit, represents subtraction by position, and softmax[·] represents a classifier based on the Softmax function.

[0065] That is, the context-aware enhanced multi-dimensional parameter time-series features and the original multi-dimensional parameter time-series features are fed into the autocorrelation gating unit to measure the consistency and correlation between the two. It should be understood that in the same local time domain, if the original multi-dimensional parameter time-series features and the context-aware enhanced feature representations have a high degree of consistency and correlation, it can be considered that the multi-dimensional parameter time-series features in this local time domain have a high degree of credibility and importance in the characterization of the global operating state of the air compressor. In this way, the relative importance among the multi-dimensional parameter time-series features in each local time domain can be effectively revealed, generating a time series of multi-dimensional parameter time-series autocorrelation gating significant confidence factors, so as to guide the subsequent feature aggregation process.

[0066] Next, based on the time series of the multi-dimensional parameter temporal autocorrelation gated significant confidence factors, guide the time series of the multi-dimensional parameter feature vectors to perform significant aggregation to obtain the multi-dimensional feature tensor temporal significant aggregation feature vectors. More specifically, first, input the time series of the multi-dimensional parameter temporal autocorrelation gated significant confidence factors into a normalization unit based on the Softmax function to obtain the time series of the multi-dimensional parameter temporal autocorrelation gated significant confidence weight factors; then, using the time series of the multi-dimensional parameter temporal autocorrelation gated significant confidence weight factors as the weight distribution, calculate the position-wise weighted sum of the time series of the multi-dimensional parameter feature vectors to obtain the multi-dimensional feature tensor temporal significant aggregation feature vectors, which is expressed by the formula:

[0067]

[0068] where exp(·) represents the exponential function with the natural constant as the base, w i represents the multi-dimensional parameter temporal autocorrelation gated significant confidence weight factor corresponding to the v i , ⊙ represents element-wise multiplication, and v f represents the multi-dimensional feature tensor temporal significant aggregation feature vectors.

[0069] That is, by normalizing the time series of the multi-dimensional parameter temporal autocorrelation gated significant confidence factors, the sum of the weight factors corresponding to each multi-dimensional parameter temporal semantic feature is 1, ensuring the rationality and standardization of the weight distribution. Finally, based on the generated weight factors, perform weighted aggregation processing on the time series of the multi-dimensional parameter feature vectors to emphasize the key multi-dimensional parameter time series and filter redundant information, obtaining the multi-dimensional feature tensor temporal significant aggregation feature vectors. In this way, not only is the local temporal fluctuation information of the multi-dimensional parameters fully utilized, but the global dependence relationship and temporal correlation structure among the parameters are also deeply explored, thus achieving a more accurate and robust description of the operating state and its change trend of the air compressor, providing a richer and more effective feature basis for subsequent air compressor fault prediction and health management.

[0070] In the above-mentioned remote diagnosis and maintenance method of the auxiliary control system of the thermal power plant, the step S7 determines the fault prediction result based on the multi-dimensional feature tensor time series significant aggregation feature vector. In a specific example of the present application, the multi-dimensional feature tensor time series significant aggregation feature vector is input into the trained fault prediction neural network model to obtain the fault prediction result. Specifically, the fault prediction neural network model uses its internal neuron structure and learning algorithm to perform a nonlinear transformation on the input multi-dimensional feature tensor time series significant aggregation feature vector, so as to make a classification decision based on the air compressor operating status information contained in the multi-dimensional feature tensor time series significant aggregation feature vector, thereby outputting the corresponding fault prediction result. During the training process, the model learns the pattern of the feature vector under different fault states through a large amount of sample data, so that when applied, it can output the corresponding fault prediction result based on the input feature representation, so as to achieve accurate early warning of air compressor failure.

[0071] In summary, the remote diagnosis and maintenance method of the auxiliary control system of a thermal power plant according to the embodiment of the present application is explained, which first collects the point data of the auxiliary network air compressor of the power plant in real time, cleans the data, and then performs time-parameter curve fitting based on the local time domain on the various parameters in the data, and splices the time-parameter curves of the various parameters in each local time domain into a multi-dimensional curve graph. Then, feature extraction and significance aggregation based on temporal context perception are performed on each multi-dimensional curve graph to mine the long-term temporal correlation and potential fault characteristics of the multi-dimensional parameters of the air compressor, thereby realizing accurate prediction of the fault of the auxiliary control system of the thermal power plant. In this way, the multi-dimensional parameter time series information with a long time span can be comprehensively considered, the influence of instantaneous interference and outliers on fault prediction can be reduced, and the stability and accuracy of fault prediction can be further improved.

[0072] Furthermore, the present application also provides a remote diagnosis and maintenance system for an auxiliary equipment control system of a thermal power plant.

[0073] Figure 4 FIG is a block diagram of a remote diagnosis and maintenance system for a thermal power plant auxiliary equipment control system according to an embodiment of the present application. Figure 4As shown in the figure, the remote diagnosis and maintenance system 100 of the auxiliary machine control system of a thermal power plant according to an embodiment of the present application includes: a data acquisition module 110, configured to collect in real time the point data of each air compressor in the auxiliary network of the power plant, where the point data includes various parameters of the real-time operation of the air compressor; a data cleaning module 120, configured to clean the point data to obtain the point data after data cleaning; a data local time domain segmentation module 130, configured to segment each parameter in the point data after data cleaning at a set time interval to obtain the time period data of each parameter; a curve fitting module 140, configured to fit the time period data of each parameter into a curve, and splice the curves of each parameter within each time period into a multi-dimensional curve graph including the curves of each parameter to obtain a time series of the multi-dimensional curve graph; a multi-dimensional curve feature extraction module 150, configured to perform feature extraction processing on each multi-dimensional curve graph in the time series of the multi-dimensional curve graph to obtain a time series of multi-dimensional feature tensor graphs; an autocorrelation gated aggregation module 160, configured to perform graph walk autocorrelation gated aggregation on the time series of the multi-dimensional feature tensor graphs to obtain a multi-dimensional feature tensor time series significant aggregation feature vector; and a fault prediction module 170, configured to determine a fault prediction result based on the multi-dimensional feature tensor time series significant aggregation feature vector.

[0074] Here, those skilled in the art can understand that the specific operations of each module in the above remote diagnosis and maintenance system of the auxiliary machine control system of a thermal power plant have been introduced in detail in the description of the Figures 1 to 3 remote diagnosis and maintenance method of the auxiliary machine control system of a thermal power plant, and therefore, the repeated description thereof will be omitted.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A remote diagnosis and maintenance method for the auxiliary control system of a thermal power plant, characterized in that, Including: Real-time collect the point data of each air compressor in the auxiliary network of the power plant, and the point data includes various parameters of the real-time operation of the air compressor; Perform data cleaning on the point data to obtain the point data after data cleaning; Divide the parameters in the point data after data cleaning at set time intervals to obtain the time period data of each parameter; Fit the time period data of each parameter into a curve, and splice the curves of each parameter within each time period into a multi-dimensional curve graph containing the curves of each parameter to obtain the time series of the multi-dimensional curve graph; Perform feature extraction processing on each multi-dimensional curve graph in the time series of the multi-dimensional curve graph to obtain the time series of the multi-dimensional feature tensor graph; Perform graph walk autocorrelation gated aggregation on the time series of the multi-dimensional feature tensor graph to obtain the multi-dimensional feature tensor time series significant aggregation feature vector; Based on the multi-dimensional feature tensor time series significant aggregation feature vector, determine the fault prediction result.

2. The remote diagnosis and maintenance method of the auxiliary control system of a thermal power plant according to claim 1, characterized in that Performing graph walk autocorrelation gated aggregation on the time series of the multi-dimensional feature tensor graph to obtain the multi-dimensional feature tensor time series significant aggregation feature vector includes: Perform feature flattening processing on each multi-dimensional feature tensor graph in the time series of the multi-dimensional feature tensor graph to obtain the time series of the multi-dimensional parameter feature vector; Perform graph walk context-aware reinforcement on the time series of the multi-dimensional parameter feature vector to obtain the time series of the multi-dimensional parameter time series context-aware reinforcement feature vector; Based on the time series of the multi-dimensional parameter time series context-aware reinforcement feature vector, perform significance analysis aggregation on the time series of the multi-dimensional parameter feature vector to obtain the multi-dimensional feature tensor time series significant aggregation feature vector.

3. The remote diagnosis and maintenance method of the auxiliary control system of a thermal power plant according to claim 2, characterized in that, Performing graph walk context-aware reinforcement on the time series of the multi-dimensional parameter feature vector to obtain the time series of the multi-dimensional parameter time series context-aware reinforcement feature vector includes: Input each multi-dimensional parameter feature vector in the time series of the multi-dimensional parameter feature vector into a hyperbolic space mapper to obtain the time series of the multi-dimensional parameter feature vector after hyperbolic space mapping; Extract the topological structure features of the time series of the multi-dimensional parameter feature vector after hyperbolic space mapping to obtain the multi-dimensional parameter time series graph walk topological feature matrix; Input the multi-dimensional parameter time series graph walk topological feature matrix and the time series of the multi-dimensional parameter feature vector after hyperbolic space mapping into the global context walk encoder based on the graph convolutional neural network model to obtain the time series of the multi-dimensional parameter time series context-aware reinforcement feature vector.

4. The remote diagnosis and maintenance method of the auxiliary control system of a thermal power plant according to claim 3, characterized in that, Extracting the topological structure features of the time series of the multi-dimensional parameter feature vector after hyperbolic space mapping to obtain the multi-dimensional parameter time series graph walk topological feature matrix includes: Calculate the Poincaré distance between any two multi-dimensional parameter feature vectors after hyperbolic space mapping in the time series of the multi-dimensional parameter feature vector after hyperbolic space mapping to obtain the multi-dimensional parameter time series graph walk topological matrix; Perform dilated convolution encoding on the multi-dimensional parameter time series graph walk topological matrix to obtain the multi-dimensional parameter time series graph walk topological feature matrix.

5. The remote diagnosis and maintenance method of the auxiliary control system of a thermal power plant according to claim 4, characterized in that, Based on the time series of the multi-dimensional parameter time-series context-aware enhanced feature vectors, performing significance analysis aggregation on the time series of the multi-dimensional parameter feature vectors to obtain the multi-dimensional feature tensor time-series significant aggregation feature vectors, including: Performing feature correlation measurement on the time series of the multi-dimensional parameter time-series context-aware enhanced feature vectors and the time series of the multi-dimensional parameter feature vectors to obtain the time series of the multi-dimensional parameter time-series autocorrelation gated significant confidence factors; Based on the time series of the multi-dimensional parameter time-series autocorrelation gated significant confidence factors, guiding the time series of the multi-dimensional parameter feature vectors to perform significance aggregation to obtain the multi-dimensional feature tensor time-series significant aggregation feature vectors.

6. The remote diagnosis and maintenance method of the auxiliary control system of a thermal power plant according to claim 5, characterized in that, Performing feature correlation measurement on the time series of the multi-dimensional parameter time-series context-aware enhanced feature vectors and the time series of the multi-dimensional parameter feature vectors to obtain the time series of the multi-dimensional parameter time-series autocorrelation gated significant confidence factors, including: Inputting each group of corresponding multi-dimensional parameter time-series context-aware enhanced feature vectors and multi-dimensional parameter feature vectors in the time series of the multi-dimensional parameter time-series context-aware enhanced feature vectors and the time series of the multi-dimensional parameter feature vectors into the autocorrelation gated unit to obtain the time series of the multi-dimensional parameter time-series autocorrelation gated significant confidence factors.

7. The remote diagnosis and maintenance method of the auxiliary control system of a thermal power plant according to claim 6, characterized in that, Based on the time series of the multi-dimensional parameter time-series autocorrelation gated significant confidence factors, guiding the time series of the multi-dimensional parameter feature vectors to perform significance aggregation to obtain the multi-dimensional feature tensor time-series significant aggregation feature vectors, including: Inputting the time series of the multi-dimensional parameter time-series autocorrelation gated significant confidence factors into the normalization unit based on the Softmax function to obtain the time series of the multi-dimensional parameter time-series autocorrelation gated significant confidence weight factors; Using the time series of the multi-dimensional parameter time-series autocorrelation gated significant confidence weight factors as the weight distribution, calculating the position-wise weighted sum of the time series of the multi-dimensional parameter feature vectors to obtain the multi-dimensional feature tensor time-series significant aggregation feature vectors.

8. The remote diagnosis and maintenance method of the auxiliary control system of a thermal power plant according to claim 7, characterized in that, Based on the multi-dimensional feature tensor time-series significant aggregation feature vectors, determining the fault prediction result, including: Inputting the multi-dimensional feature tensor time-series significant aggregation feature vectors into the trained fault prediction neural network model to obtain the fault prediction result.

9. A remote diagnosis and maintenance system for the auxiliary control system of a thermal power plant, characterized in that, Including: A data acquisition module, configured to collect the point data of each air compressor in the auxiliary network of the power plant in real time, where the point data includes various parameters of the real-time operation of the air compressor; A data cleaning module, configured to clean the point data to obtain the point data after data cleaning; A data local time domain segmentation module, configured to segment each parameter in the point data after data cleaning at a set time interval to obtain the period data of each parameter; A curve fitting module, configured to fit the period data of each parameter into a curve, and splice the curves of each parameter in each time period into a multi-dimensional curve graph including the curves of each parameter to obtain the time series of the multi-dimensional curve graph; A multi-dimensional curve feature extraction module, configured to perform feature extraction processing on each multi-dimensional curve graph in the time series of the multi-dimensional curve graph to obtain the time series of the multi-dimensional feature tensor graph; The autocorrelation gating aggregation module is used to perform graph-walking autocorrelation gating aggregation on the time series of the multi-dimensional feature tensor graph to obtain a multi-dimensional feature tensor time-series significant aggregation feature vector; The fault prediction module is used to determine the fault prediction result based on the multi-dimensional feature tensor time-series significant aggregation feature vector.

Images