Method and system for diagnosing abnormal shaft vibration of a combined boiler-turbine unit in all operating conditions

Abstract

This invention belongs to the field of shaft vibration monitoring technology and provides a method and system for diagnosing shaft vibration anomalies under all operating conditions in combined combustion and steam power units. It solves the problem that existing technologies only establish a limited library of characteristic samples from typical operating conditions, making it difficult to cover the vibration response patterns under all operating conditions and easily leading to misjudgments or omissions. The method includes establishing a mapping relationship between multi-source time-series feature vectors within a multi-source time-series data stream through a full-condition classification system; establishing a shaft vibration spectrum based on multi-source time-series feature vectors in a discrete operating condition parameter space; and adaptively diagnosing real-time vibration vectors based on standard feature templates. In this invention, by establishing a full-condition shaft vibration spectrum library containing normal state templates and typical fault templates, and employing a comprehensive similarity calculation and adaptive diagnosis mechanism, accurate matching between real-time vibration vectors and full-condition benchmark features is achieved, significantly improving the accuracy and reliability of vibration anomaly diagnosis under atypical operating conditions.

Description

Technical Field

[0001] This invention relates to the field of shaft vibration monitoring technology, and in particular to a method and system for diagnosing abnormal shaft vibration under all operating conditions of a combined combustion and steam power unit. Background Technology

[0002] Gas-steam combined cycle units, with their advantages of high efficiency, rapid start-up and shutdown, and low emissions, have become core equipment for grid peak shaving, district heating, and distributed energy systems. Their shafting, as a key structure connecting core components such as the gas turbine, steam turbine, and generator, consists of multiple rotors rigidly or flexibly connected by couplings. During operation, it withstands the coupling effects of multiple physical fields, including thermal stress, mechanical stress, and electromagnetic forces, and is the core guarantee for the safe and stable operation of the unit.

[0003] Throughout the entire lifecycle of a generator unit, the vibration state of the shaft system directly reflects its health level. In actual operation, influenced by factors such as design and manufacturing errors, poor installation alignment, long-term thermal fatigue, component wear, airflow excitation, and oil film instability, shaft system vibration is prone to problems such as excessive amplitude, abrupt phase changes, and abnormal spectral characteristics. If these issues are not detected and addressed in a timely manner, they may further lead to serious accidents such as loose bearing housings, seal failure, rotor rubbing, or even shaft breakage, resulting in unplanned unit shutdowns and causing significant economic losses and safety hazards.

[0004] To effectively monitor shaft vibration, the industry currently widely adopts online monitoring systems. These systems collect vibration signals using accelerometers and eddy current displacement sensors installed in bearing housings, end covers, and other locations. Combined with time-domain and frequency-domain analysis techniques, they achieve real-time display and alarm of vibration status. However, the limitations of existing technologies are becoming increasingly apparent in the complex operating scenarios of combined heat and power units. The operating conditions of combined heat and power units are significantly multidimensional, encompassing cold start, warm start, hot start, steady-state operation, load regulation, shutdown, and special conditions. The stress characteristics of the shaft system vary greatly under different operating conditions. Existing technologies mostly collect vibration data based on single operating conditions, and the resulting spectral libraries contain only a small number of characteristic samples from typical operating conditions, making it difficult to cover the vibration response patterns under all operating conditions. Furthermore, when the unit exhibits abnormalities under atypical operating conditions, existing spectral libraries cannot match the corresponding baseline features, easily leading to misjudgments or omissions. Summary of the Invention

[0005] The present invention aims to solve at least one of the problems existing in the prior art, and provides a method and system for diagnosing abnormal shaft vibration of combined combustion and steam power units under all operating conditions.

[0006] One aspect of the present invention provides a method for diagnosing abnormal shaft vibration in a combined combustion and steam power unit under all operating conditions, comprising: The historical vibration signals and unit operating parameters of the gas-steam combined cycle unit under all operating conditions are collected synchronously. Based on the timestamp alignment method, the historical vibration signals and unit operating parameters are synchronously aligned and processed to output a multi-source time-series data stream. Acquire multi-source time-series data streams, establish a full-condition classification system for gas-steam combined cycle units based on the multi-source time-series data streams, and establish the mapping relationship of multi-source time-series feature vectors within the multi-source time-series data streams through the full-condition classification system; Based on the mapping relationship of multi-source time-series feature vectors, a continuous parameter space for all working conditions is constructed, which contains multi-source time-series feature vectors. The continuous parameter space for all working conditions is then discretized to obtain a discretized working condition parameter space. Based on the multi-source time-series feature vectors in the discretized working condition parameter space, a shaft vibration spectrum is established. At least one set of shaft vibration spectra is merged to obtain a shaft vibration spectrum library for all working conditions. Real-time vibration signals of the gas-steam combined cycle unit are acquired, the real-time vibration signals are preprocessed to obtain real-time vibration vectors, and the real-time vibration vectors are mapped to the shaft vibration spectrum library under all operating conditions. The template similarity between the real-time vibration vectors and the standard feature templates in the shaft vibration spectrum library under all operating conditions is calculated, and standard feature templates that match the real-time vibration vectors are selected. Load a standard feature template that matches the real-time vibration vector, perform adaptive diagnosis on the real-time vibration vector based on the standard feature template, and output the shaft vibration anomaly diagnosis result.

[0007] Optionally, the time-stamp-based alignment process for synchronizing historical vibration signals and unit operating parameters includes: A hardware triggering mechanism based on GPS clock synchronization is used to acquire historical vibration signals and unit operating parameters. The acquired historical vibration signals and unit operating parameters are synchronized and aligned using cubic spline interpolation to obtain historical vibration signals and unit operating parameters that eliminate time scale differences. Historical vibration signals and unit operating parameters with time scale differences eliminated are obtained. Outlier detection based on the isolated forest algorithm is used to detect outliers in the historical vibration signals and unit operating parameters. Outliers in the historical vibration signals and unit operating parameters are removed to obtain historical vibration signals and unit operating parameters after outlier removal. Load the historical vibration signals and unit operating parameters after outlier removal, merge the historical vibration signals and unit operating parameters into a multi-source time series dataset, extract the time domain features, frequency domain features, time-frequency domain features and trend features of the multi-source time series data in the multi-source time series dataset, and obtain a multi-source time series data stream containing multi-source time series feature vectors.

[0008] Optionally, the full-condition classification system for gas-steam combined cycle units based on multi-source time-series data streams includes: A pre-constructed full-condition classification system including sub-condition domains is constructed, where sub-condition domains include start-up condition domain, steady-state condition domain, variable load condition domain, shutdown condition domain, and special condition domain; Principal component analysis was performed on the multi-source time series feature vectors in the multi-source time series data to obtain the comprehensive correlation coefficient between the multi-source time series feature vectors and the sub-working condition domains of the full-working condition classification system; Determine whether the comprehensive correlation coefficient between the multi-source time-series feature vector and the sub-condition domain of the full-condition classification system exceeds the preset correlation coefficient threshold; If the comprehensive correlation coefficient between the multi-source time-series feature vector and the sub-working condition domain of the full-working condition classification system exceeds the preset correlation coefficient threshold, a mapping relationship between the multi-source time-series feature vector and the sub-working condition domain of the full-working condition classification system is established. Based on the mapping relationship between multi-source time-series feature vectors and sub-condition domains of the full-condition classification system, the multi-source time-series feature vectors that have a mapping relationship with the sub-condition domains are subjected to dimensionality reduction processing to obtain the multi-source time-series feature vector matrix.

[0009] Optionally, the full-condition classification system for gas-steam combined cycle units based on multi-source time-series data streams further includes: Obtain the multi-source time-series feature vector matrix, and perform secondary clustering analysis on the multi-source time-series feature vector matrix based on hierarchical clustering method to obtain at least one set of matrix clusters; Calculate the cluster similarity between the matrix clustering clusters and the sub-condition domains of the full-condition classification system; Determine whether the cluster similarity between the matrix clustering cluster and the sub-condition domain of the full-condition classification system exceeds a preset cluster similarity threshold; If the cluster similarity between the matrix cluster and the sub-working condition domain of the full-working-condition classification system exceeds the preset cluster similarity threshold, a quadratic mapping relationship between the matrix cluster and the sub-working condition domain of the full-working-condition classification system is established. Based on the quadratic mapping relationship of the sub-operating condition domains of the full-operating condition classification system and the mapping relationship between the multi-source time-series feature vectors and the sub-operating condition domains of the full-operating condition classification system, a full-operating condition data flow mapping table is constructed to obtain a full-operating condition classification system containing the full-operating condition data flow mapping table.

[0010] Optionally, the step of establishing the shaft vibration spectrum based on multi-source time-series feature vectors in the discrete chemical condition parameter space includes: Obtain multi-source time series feature vectors, perform feature statistical analysis on the multi-source time series feature vectors, calculate the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical anomalies, and generate standard feature templates based on the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical anomalies. The discrete chemical condition parameter space is associated with the standard feature template, and a graph structure of the shaft vibration spectrum corresponding to the discrete chemical condition parameter space is constructed. The graph structure includes node attributes, edge attributes, and fault evolution paths. Multi-source time series feature vectors are mapped to node attributes, edge attributes, and fault evolution paths. The shaft vibration spectra of all discrete working condition parameter spaces are merged according to working condition category, uniformly coded, and an indexing mechanism is established to form a shaft vibration spectra library covering all working conditions.

[0011] Optionally, the adaptive diagnosis of real-time vibration vectors based on standard feature templates includes: The real-time vibration vector is obtained and the template similarity between the real-time vibration vector and the standard feature template is identified. The template similarity is used as the pattern confidence. The local density of the real-time vibration vector is calculated based on the density peak clustering method. The cluster center of the real-time vibration vector is determined based on the local density of the real-time vibration vector. The weighted Euclidean distance between the cluster center of the real-time vibration vector and the standard feature template is calculated. The comprehensive similarity between the real-time vibration vector and the standard feature template is calculated based on the weighted Euclidean distance and the pattern confidence. The standard feature template with the largest comprehensive similarity is selected as the main feature template. Determine whether the overall similarity between the real-time vibration vector and the main feature template exceeds the preset normal similarity threshold; If the combined similarity between the real-time vibration vector and the main feature template exceeds the preset normal similarity threshold, the real-time vibration vector is determined to be a normal vibration mode, and the standard feature template recognition result corresponding to the real-time vibration vector is output.

[0012] Optionally, the adaptive diagnosis of real-time vibration vectors based on standard feature templates further includes: If the overall similarity between the real-time vibration vector and the main feature template does not exceed the preset normal similarity threshold, a graph convolutional network is used to classify the nodes of the real-time vibration vector and calculate the probability distribution of the abnormal fault type corresponding to the real-time vibration vector. By combining the local attention mechanism, the real-time vibration signal that contributes the most to the abnormal fault type is traced back, and the vibration feature change gradient of the real-time vibration signal that contributes the most to the abnormal fault type is extracted based on the graph attention network. Based on the gradient of vibration characteristic changes, the fault evolution path is determined. Using the fault evolution path as prior information, the vibration mode transfer path within the future time window is predicted. The output includes the probability distribution of abnormal fault types, the real-time vibration signal with the largest contribution of abnormal fault types, the fault evolution path, and the vibration mode transfer path, which are the shaft vibration anomaly diagnosis results.

[0013] In another aspect, the present invention provides a full-condition shaft vibration anomaly diagnosis system for combined heat and steam power units, used to implement the full-condition shaft vibration anomaly diagnosis method for combined heat and steam power units described above. The full-condition shaft vibration anomaly diagnosis system for combined heat and steam power units includes: The time-series data stream generation module is used to synchronously collect historical vibration signals and unit operating parameters under all operating conditions of the gas-steam combined cycle unit. Based on the timestamp alignment method, the historical vibration signals and unit operating parameters are synchronously aligned and processed to output multi-source time-series data streams. The classification system construction module is used to acquire multi-source time-series data streams, establish a full-condition classification system for gas-steam combined cycle units based on the multi-source time-series data streams, and establish the mapping relationship of multi-source time-series feature vectors within the multi-source time-series data streams through the full-condition classification system; The graph library construction module is used to construct a continuous parameter space containing multi-source time-series feature vectors based on the mapping relationship of multi-source time-series feature vectors, and to discretize the continuous parameter space to obtain a discrete parameter space. Based on the multi-source time-series feature vectors in the discrete parameter space, a shaft vibration graph is established, and at least one set of shaft vibration graphs is merged to obtain a shaft vibration graph library for all working conditions. The real-time signal analysis module is used to acquire real-time vibration signals of the gas-steam combined cycle unit, preprocess the real-time vibration signals to obtain real-time vibration vectors, map the real-time vibration vectors to the shaft vibration spectrum library for all operating conditions, calculate the template similarity between the real-time vibration vectors and the standard feature templates in the shaft vibration spectrum library for all operating conditions, and select the standard feature templates that match the real-time vibration vectors. The anomaly diagnosis module is used to load standard feature templates that match the real-time vibration vectors, adaptively diagnose the real-time vibration vectors based on the standard feature templates, and output the shaft vibration anomaly diagnosis results.

[0014] Optionally, the time-series data stream generation module includes: The data alignment unit uses a hardware triggering mechanism based on GPS clock synchronization to acquire historical vibration signals and unit operating parameters. It then uses cubic spline interpolation to synchronize and align the acquired historical vibration signals and unit operating parameters, thereby eliminating time scale differences. The outlier removal unit is used to acquire historical vibration signals and unit operating parameters with time scale differences eliminated. It uses an outlier detection unit based on the isolated forest algorithm to remove outliers from the historical vibration signals and unit operating parameters, thus obtaining the historical vibration signals and unit operating parameters after outlier removal. The feature extraction unit is used to load the historical vibration signals and unit operating parameters after outlier removal, merge the historical vibration signals and unit operating parameters into a multi-source time series dataset, and extract the time domain features, frequency domain features, time-frequency domain features and trend features of the multi-source time series data in the multi-source time series dataset to obtain a multi-source time series data stream containing multi-source time series feature vectors.

[0015] Optionally, the atlas library construction module includes: The standard template generation unit is used to obtain multi-source time series feature vectors, perform feature statistical analysis on the multi-source time series feature vectors, calculate the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical anomalies, and generate standard feature templates based on the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical anomalies. The graph structure construction unit is used to establish a connection between the discrete chemical condition parameter space and the standard feature template, and to construct a graph structure of the shaft vibration spectrum corresponding to the discrete chemical condition parameter space. The graph structure includes node attributes, edge attributes, and fault evolution paths, and maps multi-source time series feature vectors to node attributes, edge attributes, and fault evolution paths. The image library merging unit is used to merge the shaft vibration images of all discrete working condition parameter spaces according to working condition categories, uniformly encode them and establish an indexing mechanism to form a shaft vibration image library covering all working conditions.

[0016] Compared with the prior art, the present invention has the following advantages: 1. By constructing a comprehensive classification system covering startup, steady state, variable load, shutdown, and special operating conditions, and achieving accurate synchronization and feature fusion of multi-source heterogeneous data based on timestamp alignment, this effectively solves the limitation of coverage caused by existing technologies that only establish vibration maps for a single or a small number of typical operating conditions. Simultaneously, by mapping multi-source time-series feature vectors to a discretized continuous parameter space for all operating conditions, a shaft vibration map library containing normal state templates and typical fault templates is established. Furthermore, by employing a comprehensive similarity calculation and adaptive diagnostic mechanism, accurate synchronization of real-time vibration vectors with full-condition benchmark features is achieved. The near-matching method significantly improves the accuracy and reliability of vibration anomaly diagnosis under atypical operating conditions, avoiding misjudgments or omissions caused by incomplete coverage of operating conditions in traditional methods. It can accurately identify various fault types such as rotor imbalance, misalignment, oil film instability, airflow excitation, and bearing wear under typical and atypical operating conditions, effectively avoiding misjudgments and omissions caused by missing operating conditions in existing atlas libraries. At the same time, by using the operating condition transition probability, vibration characteristic change gradient, and fault evolution path in the shaft vibration atlas library covering all operating conditions, it can predict and warn of fault development trends, providing a reliable basis for unit operation and maintenance decisions. 2. When synchronizing and aligning historical vibration signals and unit operating parameters based on the timestamp alignment method, the organic combination of the hardware triggering mechanism of GPS clock synchronization and the cubic spline interpolation method achieves spatiotemporal unification of high-frequency vibration signals and low-frequency operating parameters with millisecond-level accuracy. At the same time, the isolated forest algorithm is used for intelligent detection and removal of outliers, which significantly improves data quality and avoids interference from abnormal data on subsequent feature extraction and operating condition classification. Furthermore, through multi-dimensional fusion extraction of time domain, frequency domain, time-frequency domain, and trend features, a multi-source time-series feature vector containing rich dynamic information is constructed, laying a high-quality data foundation for the accurate establishment of the full operating condition classification system and the reliable construction of the shaft vibration spectrum library. 3. When establishing a full-condition classification system for gas-steam combined cycle units based on multi-source time-series data streams, a secondary mapping relationship is established through hierarchical clustering, secondary clustering analysis, and cluster similarity calculation. This achieves refined condition classification by combining data-driven approaches with prior knowledge, overcoming the shortcomings of single clustering algorithms that are prone to getting trapped in local optima and have blurred condition boundaries. The final constructed full-condition data stream mapping table forms a structured feature vector-condition domain bidirectional indexing mechanism. Combining primary and secondary mapping relationships, a structured and traceable classification system is formed. The full-condition data stream mapping table not only supports the condition indexing of the graph library but also provides a clear condition background and feature source path for subsequent anomaly diagnosis, achieving full interpretability from raw data to diagnostic results. 4. When establishing shaft vibration maps based on multi-source time-series feature vectors in the discrete working condition parameter space, adaptive standard feature templates containing normal state templates and typical fault templates are generated through feature statistical analysis. This achieves accurate quantitative characterization of the vibration state benchmark of the combined gas-fired power unit under all working conditions. Furthermore, by constructing a graph structure containing node attributes, edge attributes, and fault evolution paths, multi-dimensional vibration features, working condition transition laws, and fault development time sequences are organically integrated, forming a structured knowledge expression with physical interpretability. This solves the shortcomings of existing map libraries that only store static features and lack evolutionary correlations. Finally, through multi-map merging and unified coding indexing, a shaft vibration map library covering all working conditions is constructed. This upgrades the shaft vibration map from single-working-condition, static, and experience-driven to full-working-condition, dynamic, data-driven, and interpretable, providing a solid map foundation for the intelligent operation and maintenance of combined gas-fired power units. 5. The specific implementation method of adaptive diagnosis of real-time vibration vector based on standard feature templates proposed in this invention achieves accurate matching and dynamic threshold determination of real-time vibration vectors and standard feature templates for all working conditions by integrating template similarity and density peak clustering through a comprehensive similarity calculation mechanism. This effectively solves the problem of false alarms and missed alarms caused by fixed threshold diagnosis being unable to adapt to working condition drift. Furthermore, a deep inference framework combining graph convolutional networks and local attention mechanisms is adopted. Under abnormal conditions, it can not only output the probability distribution of multiple fault types, but also trace back the key vibration signal source and extract the feature change gradient. Combined with the fault evolution path, it can realize the forward prediction of vibration mode transfer. Attached Figure Description

[0017] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0018] Figure 1 A flowchart illustrating a method for diagnosing abnormal shaft vibration in a combined combustion and steam power unit under all operating conditions, provided as an embodiment of the present invention; Figure 2 A schematic diagram illustrating the implementation process of synchronously aligning historical vibration signals and unit operating parameters based on the timestamp alignment method, as provided in another embodiment of the present invention; Figure 3 A schematic diagram illustrating the implementation process of establishing a full-condition classification system for a gas-steam combined cycle unit based on multi-source time-series data streams, as provided in another embodiment of the present invention; Figure 4 A schematic diagram illustrating the implementation process of establishing a shaft vibration spectrum based on multi-source temporal feature vectors in a discrete chemical condition parameter space, as provided in another embodiment of the present invention; Figure 5 A schematic diagram illustrating the implementation process of adaptive diagnosis of real-time vibration vectors based on standard feature templates, provided for another embodiment of the present invention; Figure 6 This is a schematic diagram of a shaft vibration anomaly diagnosis system for a combined combustion and steam power unit under all operating conditions, provided as another embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details are presented in the embodiments of the present invention to facilitate a better understanding of the invention. However, the technical solutions claimed in the present invention can be implemented even without these technical details and various variations and modifications based on the following embodiments. The division of the following embodiments is for ease of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined with and referenced by each other without contradiction.

[0020] Existing technologies mostly rely on vibration data collected under single operating conditions, resulting in a spectrum library containing only a small number of feature samples from typical operating conditions. This makes it difficult to cover the vibration response patterns under all operating conditions. Furthermore, when the unit experiences anomalies under atypical operating conditions, existing spectra cannot match the corresponding baseline features, easily leading to misdiagnosis or missed diagnosis. To address these issues, this invention proposes a method and system for diagnosing shaft vibration anomalies under all operating conditions in combined combustion and steam power units. In short, the method first aligns historical vibration signals and unit operating parameters synchronously using a timestamp alignment method, outputting a multi-source time-series data stream. A mapping relationship between multi-source time-series feature vectors within the multi-source time-series data stream is established through a full-condition classification system. A shaft vibration spectrum is then established based on the multi-source time-series feature vectors in the discrete operating condition parameter space. At least one set of shaft vibration spectra is merged to obtain a full-condition shaft vibration spectrum library. The template similarity between the real-time vibration vector and the standard feature templates in the full-condition shaft vibration spectrum library is calculated. Standard feature templates matching the real-time vibration vector are selected. Adaptive diagnosis of the real-time vibration vector is performed based on the standard feature templates, outputting the shaft vibration anomaly diagnosis result. In this embodiment of the invention, a comprehensive classification system covering startup, steady state, variable load, shutdown, and special operating conditions is constructed. Based on timestamp alignment, accurate synchronization and feature fusion of multi-source heterogeneous data are achieved, effectively solving the limitation in coverage caused by existing technologies that only establish vibration maps for a single or a small number of typical operating conditions. Simultaneously, by mapping multi-source time-series feature vectors to a discretized continuous parameter space for all operating conditions, a shaft vibration map library containing normal state templates and typical fault templates is established. Furthermore, a comprehensive similarity calculation and adaptive diagnostic mechanism are employed to achieve real-time vibration vector and full-condition benchmark feature mapping. The precise matching of characteristics significantly improves the accuracy and reliability of vibration anomaly diagnosis under atypical operating conditions, avoiding misjudgments or omissions caused by incomplete coverage of operating conditions in traditional methods. It can accurately identify various fault types such as rotor imbalance, misalignment, oil film instability, airflow excitation, and bearing wear under typical and atypical operating conditions, effectively avoiding misjudgments and omissions caused by missing operating conditions in existing atlas libraries. At the same time, by using the operating condition transition probability, vibration characteristic change gradient, and fault evolution path in the shaft vibration atlas library covering all operating conditions, it is possible to predict and warn of fault development trends, providing a reliable basis for unit operation and maintenance decisions.

[0021] One embodiment of the present invention provides a method for diagnosing abnormal shaft vibration in a combined combustion and steam power unit under all operating conditions, the process of which is as follows: Figure 1 As shown, the process includes steps S10 to S50.

[0022] S10 synchronously collects historical vibration signals and unit operating parameters under all operating conditions of the gas-steam combined cycle unit. Based on the timestamp alignment method, it synchronously aligns and processes the historical vibration signals and unit operating parameters, and outputs a multi-source time-series data stream. The full operating conditions of the gas-steam combined cycle unit include, but are not limited to, the start-up phase, steady-state operation phase, load change phase, shutdown phase, and special operating conditions. In the start-up phase, it is divided into cold start, warm start, and hot start based on the gas turbine metal temperature and steam parameters. In the steady-state operation phase, it is divided into low load, medium load, and rated load based on the load rate. In the load change phase, it is divided into rapid load increase, slow load decrease, and step load change based on the load change rate. Special operating conditions include, but are not limited to, load shedding, unstable gas turbine combustion, and sudden drop in steam pressure.

[0023] S20: Acquire multi-source time-series data streams, establish a full-condition classification system for gas-steam combined cycle units based on the multi-source time-series data streams, and establish the mapping relationship of multi-source time-series feature vectors within the multi-source time-series data streams through the full-condition classification system.

[0024] S30. Based on the mapping relationship of multi-source time-series feature vectors, construct a continuous parameter space for all working conditions containing multi-source time-series feature vectors, and discretize the continuous parameter space for all working conditions to obtain a discretized parameter space for working conditions. Based on the multi-source time-series feature vectors in the discretized parameter space for working conditions, establish a shaft vibration spectrum, and merge at least one set of shaft vibration spectra to obtain a shaft vibration spectrum library for all working conditions. S40: Real-time vibration signals of the gas-steam combined cycle unit are acquired. The real-time vibration signals are preprocessed to obtain real-time vibration vectors, which are then mapped to a full-condition shaft vibration spectrum library. The template similarity between the real-time vibration vectors and standard feature templates in the full-condition shaft vibration spectrum library is calculated, and standard feature templates matching the real-time vibration vectors are selected. In this embodiment, when acquiring real-time vibration signals of the gas-steam combined cycle unit, eddy current displacement sensors and acceleration sensors can be used to acquire vibration signals at various measuring points in the shaft system, and thermodynamic parameters can be obtained through the unit's DCS system.

[0025] S50 loads a standard feature template that matches the real-time vibration vector, performs adaptive diagnosis of the real-time vibration vector based on the standard feature template, and outputs the diagnosis result of shaft vibration anomaly.

[0026] This embodiment constructs a comprehensive classification system covering startup, steady state, variable load, shutdown, and special operating conditions. It achieves accurate synchronization and feature fusion of multi-source heterogeneous data based on timestamp alignment, effectively solving the coverage limitation problem caused by existing technologies that only establish vibration maps for single or a small number of typical operating conditions. Simultaneously, by mapping multi-source time-series feature vectors to a discretized continuous parameter space for all operating conditions, a shaft vibration map library containing normal state templates and typical fault templates is established. Furthermore, a comprehensive similarity calculation and adaptive diagnostic mechanism are employed to achieve real-time vibration vector and full-condition baseline feature mapping. Precise matching significantly improves the accuracy and reliability of vibration anomaly diagnosis under atypical operating conditions, avoiding misjudgments or omissions caused by incomplete coverage of operating conditions in traditional methods. It can accurately identify various fault types such as rotor imbalance, misalignment, oil film instability, airflow excitation, and bearing wear under typical and atypical operating conditions, effectively avoiding misjudgments and omissions caused by missing operating conditions in existing atlas libraries. At the same time, by using the operating condition transition probability, vibration characteristic change gradient, and fault evolution path in the shaft vibration atlas library covering all operating conditions, it is possible to predict and warn of fault development trends, providing a reliable basis for unit operation and maintenance decisions.

[0027] For example, combined Figure 2 The historical vibration signals and unit operating parameters are synchronously aligned based on the timestamp alignment method, including steps S101 to S103.

[0028] S101 uses a hardware triggering mechanism based on GPS clock synchronization to acquire historical vibration signals and unit operating parameters. The acquired historical vibration signals and unit operating parameters are synchronized and aligned using cubic spline interpolation to obtain historical vibration signals and unit operating parameters that eliminate time scale differences.

[0029] S102: Acquire historical vibration signals and unit operating parameters after eliminating time-scale differences. Use an isolated forest algorithm to detect outliers in the historical vibration signals and unit operating parameters, removing outliers to obtain the outlier-removed historical vibration signals and unit operating parameters. After outlier removal, the statistical characteristics of the multi-source time-series data are more stable, and the estimates of the mean, covariance, and frequency domain components are closer to the actual operating state. This improves the reliability of the full-condition classification system and the representativeness of the baseline features of the spectral library, enhancing the anti-interference capability of subsequent anomaly diagnosis.

[0030] S103, load the historical vibration signal and unit operating parameters after outlier removal, merge the historical vibration signal and unit operating parameters into a multi-source time series dataset, extract the time domain features, frequency domain features, time-frequency domain features and trend features of the multi-source time series data in the multi-source time series dataset, and obtain a multi-source time series data stream containing multi-source time series feature vectors.

[0031] In this embodiment, when synchronizing historical vibration signals and unit operating parameters based on the timestamp alignment method, the hardware triggering mechanism of GPS clock synchronization and the cubic spline interpolation method are organically combined to achieve spatiotemporal unification of high-frequency vibration signals and low-frequency operating parameters with millisecond-level accuracy. At the same time, the isolated forest algorithm is used for intelligent detection and removal of outliers, which significantly improves data quality and avoids interference from abnormal data on subsequent feature extraction and operating condition classification. Furthermore, through multi-dimensional fusion extraction of time domain, frequency domain, time-frequency domain and trend features, a multi-source time-series feature vector containing rich dynamic information is constructed, laying a high-quality data foundation for the accurate establishment of the full operating condition classification system and the reliable construction of the shaft vibration spectrum library.

[0032] For example, combined Figure 3 A full-condition classification system for gas-steam combined cycle units is established based on multi-source time-series data streams, including steps S201 to S2010.

[0033] S201, a comprehensive operating condition classification system including sub-operating condition domains is pre-constructed. These sub-operating condition domains include, but are not limited to, startup operating condition domain, steady-state operating condition domain, variable load operating condition domain, shutdown operating condition domain, and special operating condition domain. This embodiment pre-constructs a comprehensive operating condition classification system including startup operating condition domain, steady-state operating condition domain, variable load operating condition domain, shutdown operating condition domain, and special operating condition domain, subdividing the entire unit operation process into multiple mutually exclusive and comprehensive sub-operating condition domains, providing a structured operating condition framework for subsequent feature mapping and atlas library construction. This fine-grained division ensures that vibration characteristics under different operating modes have corresponding classifications.

[0034] S202, Principal component analysis (PCA) is performed on the multi-source time-series feature vectors within the multi-source time-series data to obtain the comprehensive correlation coefficient between the multi-source time-series feature vectors and the sub-load-condition domains of the full-load-condition classification system. In this embodiment, PCA analysis is performed on the multi-source time-series feature vectors to extract principal components to reduce dimensionality and retain the main energy distribution information. Then, the comprehensive correlation coefficient between the feature vectors and each sub-load-condition domain is calculated (based on the Mahalanobis correlation coefficient form of principal component spatial projection). This method can effectively measure the similarity of vibration characteristics in the principal directions of each load-condition domain.

[0035] The formula for calculating the comprehensive correlation coefficient is as follows: ; in, Represents multi-source temporal feature vectors Sub-operating condition domains of the full operating condition classification system The comprehensive correlation coefficient, These are multi-source time-series feature vectors. Mean, full-condition classification system sub-condition domain The mean, These are multi-source time-series feature vectors. The covariance inverse matrix, the sub-working condition domain of the full-condition classification system The inverse covariance matrix, with the superscript T indicating transpose.

[0036] S203, determine whether the comprehensive correlation coefficient between the multi-source time-series feature vector and the sub-work condition domain of the full-work condition classification system exceeds a preset correlation coefficient threshold. Considering that existing technologies often directly classify all feature vectors into the most recent work condition, ignoring weak correlations and leading to incorrect labels, this embodiment introduces a correlation coefficient threshold. A mapping relationship between the multi-source time-series feature vector and the sub-work condition domain is established only when the comprehensive correlation coefficient exceeds this threshold; otherwise, no mapping is established. This strategy of mapping only when there is a strong correlation can filter out abnormal or transitional data that does not match any work condition domain, preventing noisy or cross-work condition mixed data from entering the spectral library.

[0037] S204. If the comprehensive correlation coefficient between the multi-source time-series feature vector and the sub-operating condition domain of the full-condition classification system exceeds a preset correlation coefficient threshold, a mapping relationship between the multi-source time-series feature vector and the sub-operating condition domain of the full-condition classification system is established. If the comprehensive correlation coefficient between the multi-source time-series feature vector and the sub-operating condition domain of the full-condition classification system does not exceed the preset correlation coefficient threshold, a mapping relationship between the multi-source time-series feature vector and the sub-operating condition domain of the full-condition classification system is not established.

[0038] S205, based on the mapping relationship between multi-source time-series feature vectors and sub-operating condition domains of the full-condition classification system, performs dimensionality reduction on multi-source time-series feature vectors that have a mapping relationship with the sub-operating condition domains to obtain a multi-source time-series feature vector matrix. The multi-source time-series feature vector matrix compresses the data size while retaining the main feature distribution information of the operating condition domain.

[0039] S206, Obtain the multi-source time-series feature vector matrix, and perform secondary clustering analysis on the multi-source time-series feature vector matrix based on hierarchical clustering to obtain at least one set of matrix clusters. Multiple operating modes or latent fault states may still exist within a single operating condition domain. Therefore, this embodiment uses hierarchical clustering to perform secondary clustering on the feature vector matrix, dividing the data into several matrix clusters. This can discover potential sub-class operating modes within the operating condition domain. Compared to one-time hard classification, hierarchical clustering can reveal the tree-like structural relationship of the data, helping to capture subtle but important differences in vibration characteristics.

[0040] S207 calculates the cluster similarity between the matrix clustering cluster and the sub-condition domain of the full-condition classification system.

[0041] The formula for calculating cluster similarity is as follows: ; ; in, This represents the similarity between the matrix clustering clusters and the clusters in the sub-working condition domains of the full-working-condition classification system. These are the feature center vectors of the matrix cluster and the feature center vectors of the sub-operating condition domain, respectively. The weighted Euclidean distance represents the feature center vector of the matrix cluster and the feature center vector of the sub-condition domain. This is the feature dimension weight matrix.

[0042] S208, determine whether the cluster similarity between the matrix clustering cluster and the sub-condition domain of the full-condition classification system exceeds the preset cluster similarity threshold.

[0043] S209, If the cluster similarity between the matrix cluster and the sub-working condition domain of the full-working-condition classification system exceeds the preset cluster similarity threshold, establish a quadratic mapping relationship between the matrix cluster and the sub-working condition domain of the full-working-condition classification system.

[0044] S2010: Based on the quadratic mapping relationship of the sub-operating condition domains of the full-operating condition classification system and the mapping relationship between the multi-source time-series feature vectors and the sub-operating condition domains of the full-operating condition classification system, a full-operating condition data flow mapping table is constructed to obtain a full-operating condition classification system containing the full-operating condition data flow mapping table.

[0045] In this embodiment, when establishing a full-condition classification system for gas-steam combined cycle units based on multi-source time-series data streams, a secondary mapping relationship is established through hierarchical clustering, secondary clustering analysis, and cluster similarity calculation. This achieves refined condition classification by combining data-driven approaches with prior knowledge, overcoming the shortcomings of single clustering algorithms that are prone to getting trapped in local optima and have blurred condition boundaries. The final constructed full-condition data stream mapping table forms a structured feature vector-condition domain bidirectional indexing mechanism. Combining primary and secondary mapping relationships, a structured and traceable classification system is formed. The full-condition data stream mapping table not only supports the condition indexing of the graph library but also provides a clear condition background and feature source path for subsequent anomaly diagnosis, achieving full interpretability from raw data to diagnostic results.

[0046] For example, combined Figure 4 The vibration spectrum of the shaft system is established based on the multi-source time-series feature vectors in the discrete chemical condition parameter space, including steps S301 to S303.

[0047] S301: Obtain multi-source time-series feature vectors, perform feature statistical analysis on these vectors, calculate the feature dimension mean vector, the inter-feature covariance matrix, and the distribution characteristics of typical abnormal features, and generate standard feature templates based on these features. The standard feature templates include normal state templates and abnormal state templates. The normal state template uses the feature dimension mean vector as its center and combines it with the standard deviation to set upper and lower tolerance intervals, forming a multi-dimensional normal boundary range. The abnormal state template forms an adaptive fault boundary range based on the fault feature mean, variance, spectrum, and shaft center trajectory of known typical fault types (rotor imbalance, misalignment, oil film instability, airflow excitation, bearing wear).

[0048] S302 establishes a connection between the discrete-time parameter space and the standard feature template, and constructs a graph structure for the vibration spectrum of the shaft system corresponding to the discrete-time parameter space. The graph structure includes node attributes, edge attributes, and fault evolution paths, mapping multi-source temporal feature vectors to node attributes, edge attributes, and fault evolution paths. Node attributes include sub-condition domain identifiers, vibration feature vector centers, and mode confidence scores. Edge attributes include condition transition probabilities, condition transition conditions, and vibration feature change gradients during condition transitions. Fault evolution paths include fault development trajectories extracted from known typical fault types, including temporal chains of initial abnormal modes, deterioration modes, and severe fault modes.

[0049] S303 merges the shaft vibration spectra from all discrete working condition parameter spaces according to working condition categories, unifies the coding, and establishes an indexing mechanism to form a full-working-condition shaft vibration spectra library.

[0050] In this embodiment, when establishing the shaft vibration spectrum based on multi-source time-series feature vectors in the discrete operating condition parameter space, an adaptive standard feature template containing normal state templates and typical fault templates is generated through feature statistical analysis. This achieves accurate quantitative characterization of the vibration state benchmark of the combined gas-fired power unit under all operating conditions. Furthermore, by constructing a graph structure containing node attributes, edge attributes, and fault evolution paths, multi-dimensional vibration features, operating condition transition laws, and fault development time sequences are organically integrated, forming a structured knowledge expression with physical interpretability. This addresses the shortcomings of existing spectrum libraries that only store static features and lack evolutionary correlations. Finally, through multi-spectrum merging and unified coding indexing, a shaft vibration spectrum library covering all operating conditions is constructed. Ultimately, the shaft vibration spectrum is upgraded from single-condition, static, and experience-driven to full-condition, dynamic, data-driven, and interpretable, providing a solid spectrum foundation for the intelligent operation and maintenance of combined gas-fired power units.

[0051] For example, combined Figure 5 The real-time vibration vector is adaptively diagnosed based on a standard feature template, including steps S401 to S407.

[0052] S401: Acquire real-time vibration vectors and identify the template similarity between the real-time vibration vectors and standard feature templates, using the template similarity as the pattern confidence level. First, the template similarity between the real-time vibration vectors and the standard feature templates is calculated and used as the pattern confidence level, objectively reflecting the degree of conformity between the current vibration pattern and the known normal operating condition template. This provides a quantifiable basic indicator for subsequent comprehensive judgment, avoiding misjudgments caused by relying solely on absolute value thresholds, and maintaining diagnostic stability, especially when operating conditions fluctuate significantly.

[0053] S402: Calculate the local density of the real-time vibration vector based on the density peak clustering method, determine the cluster center of the real-time vibration vector based on the local density of the real-time vibration vector, calculate the weighted Euclidean distance between the cluster center of the real-time vibration vector and the standard feature template, calculate the comprehensive similarity between the real-time vibration vector and the standard feature template based on the weighted Euclidean distance and the pattern confidence, and select the standard feature template with the maximum comprehensive similarity as the main feature template.

[0054] The formula for calculating the comprehensive similarity between the real-time vibration vector and the standard feature template is as follows: ; ; ; in, Represents the real-time vibration vector With the i-th standard feature template The overall similarity These represent the distance similarity between the cluster centers of the real-time vibration vectors and the standard feature template, and the pattern confidence, respectively. This is a balance coefficient between distance similarity and pattern credibility. Represents the real-time vibration vector With the i-th standard feature template The weighted Euclidean distance; This represents the normalization coefficient, and its specific value can be determined according to actual needs. Let be the adaptive weight of the i-th template, the i-th element in the real-time vibration vector, and the center feature vector of the i-th standard feature template, respectively, and j represent the total number of standard feature templates.

[0055] S403, determine whether the overall similarity between the real-time vibration vector and the main feature template exceeds the preset normal similarity threshold.

[0056] S404, if the overall similarity between the real-time vibration vector and the main feature template exceeds a preset normal similarity threshold, the real-time vibration vector is determined to be a normal vibration mode, and the standard feature template recognition result corresponding to the real-time vibration vector is output. This embodiment sets a preset normal similarity threshold. When the overall similarity between the real-time vibration vector and the main feature template exceeds this threshold, it is directly determined to be a normal vibration mode, and the matching template recognition result is output. This determination method based on overall similarity reflects the overall operating status better than a single image value or frequency threshold, significantly reducing the false alarm rate caused by normal fluctuations and improving the usability of the diagnostic results.

[0057] S405. If the overall similarity between the real-time vibration vector and the main feature template does not exceed the preset normal similarity threshold, a graph convolutional network is used to classify the nodes of the real-time vibration vector and calculate the probability distribution of the abnormal fault type corresponding to the real-time vibration vector.

[0058] S406, combining a local attention mechanism to trace back the real-time vibration signal that contributes most to the abnormal fault type, and extracting the vibration feature change gradient of the real-time vibration signal that contributes most to the abnormal fault type based on a graph attention network. This embodiment combines a local attention mechanism to trace back the real-time vibration signal that contributes most to the abnormal fault type, and uses a graph attention network to extract the vibration feature change gradient of this signal, accurately locating key feature changes. Employing a highly interpretable feature focusing method, it not only helps maintenance personnel pinpoint fault-sensitive sections, but also provides high-value input for subsequent trend analysis, avoiding the uninterpretable problems of "black box" models.

[0059] S407 determines the fault evolution path based on the gradient of vibration characteristic changes, uses the fault evolution path as prior information, predicts the vibration mode transfer path within the future time window, and outputs shaft vibration anomaly diagnosis results including the probability distribution of abnormal fault types, the real-time vibration signal with the largest contribution of abnormal fault types, the fault evolution path, and the vibration mode transfer path.

[0060] The proposed method for adaptive diagnosis of real-time vibration vectors based on standard feature templates in this embodiment achieves accurate matching and dynamic threshold determination of real-time vibration vectors with standard feature templates for all operating conditions through a comprehensive similarity calculation mechanism that integrates template similarity and density peak clustering. This effectively solves the problem of false alarms and missed alarms caused by fixed threshold diagnosis being unable to adapt to operating condition drift. Furthermore, a deep inference framework combining graph convolutional networks and local attention mechanisms is adopted. Under abnormal conditions, it can not only output the probability distribution of multiple fault types, but also trace back the key vibration signal source and extract the feature change gradient. Combined with the fault evolution path, it can realize forward prediction of vibration mode transition.

[0061] Another embodiment of the present invention provides a full-condition shaft vibration anomaly diagnosis system for combined combustion and steam power units, used to implement the full-condition shaft vibration anomaly diagnosis method for combined combustion and steam power units described in the above embodiments. Figure 6 As shown, the shaft vibration anomaly diagnosis system for the combined combustion and steam power unit under all operating conditions includes a time-series data stream generation module 100, a classification system construction module 200, a spectrum library construction module 300, a real-time signal analysis module 400, and an anomaly diagnosis module 500.

[0062] The time-series data stream generation module 100 is used to synchronously collect historical vibration signals and unit operating parameters under all operating conditions of the gas-steam combined cycle unit. Based on the timestamp alignment method, the historical vibration signals and unit operating parameters are synchronously aligned and processed to output multi-source time-series data streams.

[0063] The time-series data stream generation module 100 includes a data alignment unit 110, an outlier removal unit 120, and a feature extraction unit 130.

[0064] The data alignment unit 110 uses a hardware triggering mechanism based on GPS clock synchronization to acquire historical vibration signals and unit operating parameters. It then uses cubic spline interpolation to synchronize and align the acquired historical vibration signals and unit operating parameters, thereby eliminating time scale differences.

[0065] The outlier removal unit 120 is used to acquire historical vibration signals and unit operating parameters with time scale differences eliminated. It uses an outlier detection system based on the isolated forest algorithm to remove outliers from the historical vibration signals and unit operating parameters, thus obtaining the historical vibration signals and unit operating parameters after outlier removal.

[0066] The feature extraction unit 130 is used to load the historical vibration signal and unit operating parameters after outlier removal, merge the historical vibration signal and unit operating parameters into a multi-source time series dataset, extract the time domain features, frequency domain features, time-frequency domain features and trend features of the multi-source time series data in the multi-source time series dataset, and obtain a multi-source time series data stream containing multi-source time series feature vectors.

[0067] The classification system construction module 200 is used to acquire multi-source time-series data streams, establish a full-condition classification system for gas-steam combined cycle units based on the multi-source time-series data streams, and establish the mapping relationship of multi-source time-series feature vectors within the multi-source time-series data streams through the full-condition classification system.

[0068] The spectrum library construction module 300 constructs a continuous parameter space containing multi-source time-series feature vectors based on the mapping relationship of multi-source time-series feature vectors. It then discretizes the continuous parameter space to obtain a discrete parameter space. Based on the multi-source time-series feature vectors in the discrete parameter space, it establishes a shaft vibration spectrum and merges at least one set of shaft vibration spectra to obtain a full-condition shaft vibration spectrum library.

[0069] The graph library construction module 300 includes a standard template generation unit 310, a graph structure construction unit 320, and a graph library merging unit 330.

[0070] The standard template generation unit 310 is used to obtain multi-source time series feature vectors, perform feature statistical analysis on the multi-source time series feature vectors, calculate the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical abnormal features, and generate a standard feature template based on the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical abnormal features.

[0071] The graph structure construction unit 320 is used to establish a connection between the discrete chemical condition parameter space and the standard feature template, and to construct a graph structure of the shaft vibration spectrum corresponding to the discrete chemical condition parameter space. The graph structure includes node attributes, edge attributes, and fault evolution paths, and maps multi-source time-series feature vectors to node attributes, edge attributes, and fault evolution paths.

[0072] The image library merging unit 330 is used to merge the shaft vibration images of all discrete working condition parameter spaces according to working condition categories, uniformly encode them and establish an indexing mechanism to form a shaft vibration image library covering all working conditions.

[0073] The real-time signal analysis module 400 is used to acquire real-time vibration signals of the gas-steam combined cycle unit, preprocess the real-time vibration signals to obtain real-time vibration vectors, map the real-time vibration vectors to the shaft vibration spectrum library for all operating conditions, calculate the template similarity between the real-time vibration vectors and the standard feature templates in the shaft vibration spectrum library for all operating conditions, and select the standard feature templates that match the real-time vibration vectors.

[0074] The anomaly diagnosis module 500 is used to load a standard feature template that matches the real-time vibration vector, adaptively diagnose the real-time vibration vector based on the standard feature template, and output the shaft vibration anomaly diagnosis result.

[0075] The specific implementation method of the shaft vibration anomaly diagnosis system for combined heat and steam power units under all operating conditions provided in this embodiment of the invention is described in the method for diagnosing shaft vibration anomalies under all operating conditions for combined heat and steam power units provided in this embodiment of the invention, and will not be repeated here.

[0076] In summary, this invention provides a method and system for diagnosing shaft vibration anomalies in combined combustion and steam power units under all operating conditions. By constructing a comprehensive classification system covering startup, steady state, variable load, shutdown, and special operating conditions, and achieving accurate synchronization and feature fusion of multi-source heterogeneous data based on timestamp alignment, it effectively solves the problem of limited coverage caused by existing technologies that only establish vibration maps for single or a few typical operating conditions. Simultaneously, by mapping multi-source time-series feature vectors to a discretized continuous parameter space for all operating conditions, a shaft vibration map library containing normal state templates and typical fault templates is established for all operating conditions. Furthermore, by employing a comprehensive similarity calculation and adaptive diagnostic mechanism, it achieves practical... The precise matching of the vibration vector with the full-condition benchmark features significantly improves the accuracy and reliability of vibration anomaly diagnosis under atypical conditions, avoiding misjudgments or omissions caused by incomplete condition coverage in traditional methods. It can accurately identify various fault types such as rotor imbalance, misalignment, oil film instability, airflow excitation, and bearing wear under typical and atypical operating conditions, effectively avoiding misjudgments and omissions caused by missing conditions in existing atlas libraries. At the same time, by using the condition transition probability, vibration feature change gradient, and fault evolution path in the full-condition shaft vibration atlas library, it is possible to predict and warn of fault development trends, providing a reliable basis for unit operation and maintenance decisions.

[0077] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0078] In the embodiments provided by this invention, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or communication connections shown or discussed may be through some interfaces; the indirect coupling or communication connections between devices or units may be telecommunications or other forms.

[0079] Those skilled in the art will understand that the above embodiments are specific implementations of the present invention, and in practical applications, various changes can be made in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A method for diagnosing abnormal shaft vibration in a combined combustion and steam power unit under all operating conditions, characterized in that, include: The historical vibration signals and unit operating parameters of the gas-steam combined cycle unit under all operating conditions are collected synchronously. Based on the timestamp alignment method, the historical vibration signals and unit operating parameters are synchronously aligned and processed to output a multi-source time-series data stream. Acquire multi-source time-series data streams, establish a full-condition classification system for gas-steam combined cycle units based on the multi-source time-series data streams, and establish the mapping relationship of multi-source time-series feature vectors within the multi-source time-series data streams through the full-condition classification system; Based on the mapping relationship of multi-source time-series feature vectors, a continuous parameter space for all working conditions is constructed, which contains multi-source time-series feature vectors. The continuous parameter space for all working conditions is then discretized to obtain a discretized working condition parameter space. Based on the multi-source time-series feature vectors in the discretized working condition parameter space, a shaft vibration spectrum is established. At least one set of shaft vibration spectra is merged to obtain a shaft vibration spectrum library for all working conditions. Real-time vibration signals of the gas-steam combined cycle unit are acquired, the real-time vibration signals are preprocessed to obtain real-time vibration vectors, and the real-time vibration vectors are mapped to the shaft vibration spectrum library under all operating conditions. The template similarity between the real-time vibration vectors and the standard feature templates in the shaft vibration spectrum library under all operating conditions is calculated, and standard feature templates that match the real-time vibration vectors are selected. Load a standard feature template that matches the real-time vibration vector, perform adaptive diagnosis on the real-time vibration vector based on the standard feature template, and output the shaft vibration anomaly diagnosis result.

2. The method for diagnosing abnormal shaft vibration in a combined combustion and steam power unit under all operating conditions according to claim 1, characterized in that, The time-stamp alignment method for synchronizing historical vibration signals and unit operating parameters includes: A hardware triggering mechanism based on GPS clock synchronization is used to acquire historical vibration signals and unit operating parameters. The acquired historical vibration signals and unit operating parameters are synchronized and aligned using cubic spline interpolation to obtain historical vibration signals and unit operating parameters that eliminate time scale differences. Historical vibration signals and unit operating parameters with time scale differences eliminated are obtained. Outlier detection based on the isolated forest algorithm is used to detect outliers in the historical vibration signals and unit operating parameters. Outliers in the historical vibration signals and unit operating parameters are removed to obtain historical vibration signals and unit operating parameters after outlier removal. Load the historical vibration signals and unit operating parameters after outlier removal, merge the historical vibration signals and unit operating parameters into a multi-source time series dataset, extract the time domain features, frequency domain features, time-frequency domain features and trend features of the multi-source time series data in the multi-source time series dataset, and obtain a multi-source time series data stream containing multi-source time series feature vectors.

3. The method for diagnosing abnormal shaft vibration in a combined combustion and steam power unit under all operating conditions according to claim 1, characterized in that, The full-condition classification system for gas-steam combined cycle units based on multi-source time-series data streams includes: A pre-constructed full-condition classification system including sub-condition domains is constructed, where sub-condition domains include start-up condition domain, steady-state condition domain, variable load condition domain, shutdown condition domain, and special condition domain; Principal component analysis was performed on the multi-source time series feature vectors in the multi-source time series data to obtain the comprehensive correlation coefficient between the multi-source time series feature vectors and the sub-working condition domains of the full-working condition classification system; Determine whether the comprehensive correlation coefficient between the multi-source time-series feature vector and the sub-condition domain of the full-condition classification system exceeds the preset correlation coefficient threshold; If the comprehensive correlation coefficient between the multi-source time-series feature vector and the sub-working condition domain of the full-working condition classification system exceeds the preset correlation coefficient threshold, a mapping relationship between the multi-source time-series feature vector and the sub-working condition domain of the full-working condition classification system is established. Based on the mapping relationship between multi-source time-series feature vectors and sub-condition domains of the full-condition classification system, the multi-source time-series feature vectors that have a mapping relationship with the sub-condition domains are subjected to dimensionality reduction processing to obtain the multi-source time-series feature vector matrix.

4. The method for diagnosing abnormal shaft vibration in a combined combustion and steam power unit under all operating conditions according to claim 3, characterized in that, The full-condition classification system for gas-steam combined cycle units based on multi-source time-series data streams also includes: Obtain the multi-source time-series feature vector matrix, and perform secondary clustering analysis on the multi-source time-series feature vector matrix based on hierarchical clustering method to obtain at least one set of matrix clusters; Calculate the cluster similarity between the matrix clustering clusters and the sub-condition domains of the full-condition classification system; Determine whether the cluster similarity between the matrix clustering cluster and the sub-condition domain of the full-condition classification system exceeds a preset cluster similarity threshold; If the cluster similarity between the matrix cluster and the sub-working condition domain of the full-working-condition classification system exceeds the preset cluster similarity threshold, a quadratic mapping relationship between the matrix cluster and the sub-working condition domain of the full-working-condition classification system is established. Based on the quadratic mapping relationship of the sub-operating condition domains of the full-operating condition classification system and the mapping relationship between the multi-source time-series feature vectors and the sub-operating condition domains of the full-operating condition classification system, a full-operating condition data flow mapping table is constructed to obtain a full-operating condition classification system containing the full-operating condition data flow mapping table.

5. The method for diagnosing abnormal shaft vibration of a combined combustion and steam power unit under all operating conditions according to claim 4, characterized in that, The method of establishing a shaft vibration spectrum based on multi-source time-series feature vectors in the discrete chemical condition parameter space includes: Obtain multi-source time series feature vectors, perform feature statistical analysis on the multi-source time series feature vectors, calculate the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical anomalies, and generate standard feature templates based on the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical anomalies. The discrete chemical condition parameter space is associated with the standard feature template, and a graph structure of the shaft vibration spectrum corresponding to the discrete chemical condition parameter space is constructed. The graph structure includes node attributes, edge attributes, and fault evolution paths. Multi-source time series feature vectors are mapped to node attributes, edge attributes, and fault evolution paths. The shaft vibration spectra of all discrete working condition parameter spaces are merged according to working condition category, uniformly coded, and an indexing mechanism is established to form a shaft vibration spectra library covering all working conditions.

6. The method for diagnosing abnormal shaft vibration in a combined combustion and steam power unit under all operating conditions according to claim 5, characterized in that, The adaptive diagnosis of real-time vibration vectors based on standard feature templates includes: The real-time vibration vector is obtained and the template similarity between the real-time vibration vector and the standard feature template is identified. The template similarity is used as the pattern confidence. The local density of the real-time vibration vector is calculated based on the density peak clustering method. The cluster center of the real-time vibration vector is determined based on the local density of the real-time vibration vector. The weighted Euclidean distance between the cluster center of the real-time vibration vector and the standard feature template is calculated. The comprehensive similarity between the real-time vibration vector and the standard feature template is calculated based on the weighted Euclidean distance and the pattern confidence. The standard feature template with the largest comprehensive similarity is selected as the main feature template. Determine whether the overall similarity between the real-time vibration vector and the main feature template exceeds the preset normal similarity threshold; If the combined similarity between the real-time vibration vector and the main feature template exceeds the preset normal similarity threshold, the real-time vibration vector is determined to be a normal vibration mode, and the standard feature template recognition result corresponding to the real-time vibration vector is output.

7. The method for diagnosing abnormal shaft vibration of a combined combustion and steam power unit under all operating conditions according to claim 6, characterized in that, The adaptive diagnosis of real-time vibration vectors based on standard feature templates also includes: If the overall similarity between the real-time vibration vector and the main feature template does not exceed the preset normal similarity threshold, a graph convolutional network is used to classify the nodes of the real-time vibration vector and calculate the probability distribution of the abnormal fault type corresponding to the real-time vibration vector. By combining the local attention mechanism, the real-time vibration signal that contributes the most to the abnormal fault type is traced back, and the vibration feature change gradient of the real-time vibration signal that contributes the most to the abnormal fault type is extracted based on the graph attention network. Based on the gradient of vibration characteristic changes, the fault evolution path is determined. Using the fault evolution path as prior information, the vibration mode transfer path within the future time window is predicted. The output includes the probability distribution of abnormal fault types, the real-time vibration signal with the largest contribution of abnormal fault types, the fault evolution path, and the vibration mode transfer path, which are the shaft vibration anomaly diagnosis results.

8. A system for diagnosing abnormal shaft vibration in a combined combustion and steam power unit under all operating conditions, characterized in that, For implementing the method for diagnosing abnormal shaft vibration under all operating conditions of a combined heat and steam power unit according to any one of claims 1 to 7, the system for diagnosing abnormal shaft vibration under all operating conditions of a combined heat and steam power unit comprises: The time-series data stream generation module is used to synchronously collect historical vibration signals and unit operating parameters under all operating conditions of the gas-steam combined cycle unit. Based on the timestamp alignment method, the historical vibration signals and unit operating parameters are synchronously aligned and processed to output multi-source time-series data streams. The classification system construction module is used to acquire multi-source time-series data streams, establish a full-condition classification system for gas-steam combined cycle units based on the multi-source time-series data streams, and establish the mapping relationship of multi-source time-series feature vectors within the multi-source time-series data streams through the full-condition classification system; The graph library construction module is used to construct a continuous parameter space containing multi-source time-series feature vectors based on the mapping relationship of multi-source time-series feature vectors, and to discretize the continuous parameter space to obtain a discrete parameter space. Based on the multi-source time-series feature vectors in the discrete parameter space, a shaft vibration graph is established, and at least one set of shaft vibration graphs is merged to obtain a shaft vibration graph library for all working conditions. The real-time signal analysis module is used to acquire real-time vibration signals of the gas-steam combined cycle unit, preprocess the real-time vibration signals to obtain real-time vibration vectors, map the real-time vibration vectors to the shaft vibration spectrum library for all operating conditions, calculate the template similarity between the real-time vibration vectors and the standard feature templates in the shaft vibration spectrum library for all operating conditions, and select the standard feature templates that match the real-time vibration vectors. The anomaly diagnosis module is used to load standard feature templates that match the real-time vibration vectors, adaptively diagnose the real-time vibration vectors based on the standard feature templates, and output the shaft vibration anomaly diagnosis results.

9. The combined combustion and steam power unit full-condition shaft vibration anomaly diagnosis system according to claim 8, characterized in that, The time-series data stream generation module includes: The data alignment unit uses a hardware triggering mechanism based on GPS clock synchronization to acquire historical vibration signals and unit operating parameters. It then uses cubic spline interpolation to synchronize and align the acquired historical vibration signals and unit operating parameters, thereby eliminating time scale differences. The outlier removal unit is used to acquire historical vibration signals and unit operating parameters with time scale differences eliminated. It uses an outlier detection unit based on the isolated forest algorithm to remove outliers from the historical vibration signals and unit operating parameters, thus obtaining the historical vibration signals and unit operating parameters after outlier removal. The feature extraction unit is used to load the historical vibration signals and unit operating parameters after outlier removal, merge the historical vibration signals and unit operating parameters into a multi-source time series dataset, and extract the time domain features, frequency domain features, time-frequency domain features and trend features of the multi-source time series data in the multi-source time series dataset to obtain a multi-source time series data stream containing multi-source time series feature vectors.

10. The combined combustion and steam power unit full-condition shaft vibration anomaly diagnosis system according to claim 8, characterized in that, The atlas library construction module includes: The standard template generation unit is used to obtain multi-source time series feature vectors, perform feature statistical analysis on the multi-source time series feature vectors, calculate the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical anomalies, and generate standard feature templates based on the feature dimension mean vector, the feature covariance matrix, and the distribution features of typical anomalies. The graph structure construction unit is used to establish a connection between the discrete chemical condition parameter space and the standard feature template, and to construct a graph structure of the shaft vibration spectrum corresponding to the discrete chemical condition parameter space. The graph structure includes node attributes, edge attributes, and fault evolution paths, and maps multi-source time series feature vectors to node attributes, edge attributes, and fault evolution paths. The image library merging unit is used to merge the shaft vibration images of all discrete working condition parameter spaces according to working condition categories, uniformly encode them and establish an indexing mechanism to form a shaft vibration image library covering all working conditions.

Images