Hydropower station equipment state monitoring data feature extraction and evaluation method

Abstract

The invention relates to the technical field of hydropower station equipment state monitoring, and discloses a hydropower station equipment state monitoring data feature extraction and evaluation method, which comprises the following steps: firstly, constructing an energy transfer topology network model based on an equipment physical structure; secondly, acquiring a multi-source heterogeneous monitoring signal of each node, and uniformly converting the multi-source heterogeneous monitoring signal into an energy dissipation rate taking watts as a unit; then, based on the topology network model and the energy dissipation rate, extracting system-level state features including total system entropy yield, entropy production topology centrality, main path information fidelity and the like, and component-level state features including node dissipation energy spectrum entropy and the like; and finally, combining the features to generate a comprehensive state evaluation vector, carrying out fault positioning based on the system-level features, and carrying out fault mode judgment based on the component-level features to realize collaborative fault diagnosis. By fusing multi-source information under a unified energy framework, multi-dimensional and hierarchical evaluation and accurate diagnosis of the equipment state are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of hydropower station equipment status monitoring, and in particular to a method for extracting and evaluating features of hydropower station equipment status monitoring data. Background Art

[0002] Hydropower stations are key components of the modern energy system, and their operational stability and safety are crucial. In hydropower stations, turbine generator sets, the core power equipment that converts water energy into electricity, are complex, costly, and operate under high load for extended periods. To ensure the safety of the units, improve their operating efficiency, and implement predictive maintenance, a large number of sensors are typically deployed on the units' key physical components, forming a complex condition monitoring system. This system continuously collects massive amounts of monitoring data, including equipment vibration, operating temperature, acoustic pressure, and electromagnetic parameters. These data are multi-source and heterogeneous, coming from diverse sources and possessing varying physical properties. This data provides fundamental data support for accurately assessing equipment health and providing early warning of potential failures.

[0003] Existing equipment status monitoring data analysis methods face significant challenges in processing this multi-source heterogeneous information. Because signals such as vibration, temperature, and acoustics belong to different physical categories and have completely different dimensions and physical connotations, traditional analysis methods usually treat them as independent, fragmented information streams. This processing method leads to a serious information island phenomenon, that is, it is difficult to effectively integrate and correlate monitoring data from different sources within a unified framework. For example, vibration analysis experts and thermal analysis experts may draw conclusions based on their own data, but lack an objective and quantitative means to comprehensively evaluate the joint impact of these two phenomena on the overall status of the unit, thereby limiting the ability to conduct a comprehensive and holistic assessment of the equipment status.

[0004] Current mainstream condition assessment and fault diagnosis technologies are limited in the depth of their analytical models and diagnostic conclusions. Many methods overly rely on purely data-driven statistical models or artificial intelligence algorithms. While these models can identify correlations within massive amounts of data, they often overlook the physical structure of the equipment itself, the constraints between components, and the laws governing energy transfer and conversion within the system. This results in a lack of physical interpretability and a lack of confidence in the reliability of their diagnostic conclusions. This analytical paradigm often results in monitoring systems remaining limited to threshold alarms for single measurement points, resulting in fragmented condition assessments that fail to reveal overall stability degradation trends or topological instability risks at a macro level. The resulting fault diagnostic conclusions are also relatively simplistic, typically only indicating the approximate location of the fault and failing to distinguish the underlying fault pattern (such as deterministic frequency faults or broadband wear-out faults), thus compromising the accuracy of subsequent maintenance decisions. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a method for extracting and evaluating features from hydropower station equipment status monitoring data. It solves the problems that the existing hydropower station equipment status monitoring methods are difficult to perform fusion analysis under a unified physical dimension when processing multi-source heterogeneous data, lack dynamic evaluation of the overall system status evolution trend and in-depth diagnosis capabilities of the root causes of faults, resulting in insufficient comprehensiveness of status evaluation and accuracy of diagnostic conclusions.

[0006] In order to solve the above technical problems, the present invention provides a method for extracting and evaluating features of equipment status monitoring data of a hydropower station.

[0007] The technical solution provided by the present invention is: a method for extracting and evaluating features from hydropower station equipment status monitoring data, comprising the following steps:

[0008] Step S1: Based on the physical structure of the hydropower station equipment, an energy transfer topology network model is constructed. The energy transfer topology network model consists of multiple nodes representing the physical components of the equipment and multiple directed edges representing the energy transfer paths between the physical components. This model provides physical constraints and a topological basis for subsequent calculations.

[0009] Step S2: Acquire multi-source heterogeneous monitoring signals corresponding to nodes in the energy transfer topology network model, wherein the monitoring signals include vibration signals, temperature signals, and acoustic signals.

[0010] Step S3: Perform a unified energy rate characterization on the multi-source heterogeneous monitoring signals to obtain an energy dissipation rate that characterizes the energy dissipation level of each physical component. This step projects the monitoring signals from different physical dimensions onto the energy rate dimension (unit: watt) to enable fusion analysis of multi-physical information. The specific conversion process is as follows:

[0011] For the vibration signal, the corresponding kinetic energy dissipation power It can be calculated by the following formula:

[0012] ;

[0013] Where, is the kinetic energy dissipation power, is the equivalent mass of the physical component, is the instantaneous acceleration signal of the component, is the instantaneous velocity obtained by integrating the instantaneous acceleration signal. For the temperature signal, the corresponding heat dissipation power is It can be calculated by the following formula:

[0014] ;

[0015] Where, is the heat dissipation power; is the specific heat capacity of the physical component; is the mass of the physical component; is the instantaneous temperature signal of the component The rate of change with respect to time.

[0016] For acoustic signals, the corresponding sound energy radiation power It can be calculated by the following formula:

[0017] ;

[0018] Where, is the sound energy radiation power, is the sound pressure monitoring signal, is the density of the propagation medium, is the speed of sound, is the equivalent acoustic radiation area of ​​the component. The total energy dissipation rate of a single node is the sum of all types of dissipated power.

[0019] Step S4: extracting system-level state characteristics and component-level state characteristics based on the energy transfer topology network model and the energy dissipation rate.

[0020] The system-level status characteristics include:

[0021] Total system entropy production rate: The energy dissipation rates of all nodes in the energy transfer topology network model are summed up, and the result is used as a macro-indicator to quantify the overall level of ineffective energy dissipation of the hydropower station equipment.

[0022] Entropy production topological centrality: Among all the nodes that constitute the energy transfer topological network model, the node with the largest energy dissipation rate is identified, and the identity of the node is used as the entropy production topological centrality to locate the physical component with the most significant energy dissipation.

[0023] Main path information fidelity: First, the main energy transfer path is pre-set in the energy transfer topology network model. Then, for adjacent nodes on the main energy transfer path, the transfer entropy between them is calculated based on the time series of their energy dissipation rates to quantify the causal relationship strength of the upstream node on the energy information flow of the downstream node. Finally, the main path information fidelity is obtained by taking a weighted sum of the calculated transfer entropies between all adjacent nodes on the main energy transfer path.

[0024] Topological instability tendency index: First, the extracted system-level state features are transformed into a state feature vector sequence in the time dimension. Then, by clustering the state feature vector sequence during the normal operation of the hydropower station equipment, the topological stable states corresponding to different typical operating conditions are identified. Then, based on the historical data of the equipment switching between topological stable states, a baseline transition model is established to characterize its normal state transition law. Finally, the actual transition behavior of the current state feature vector sequence between topological stable states is monitored, and the degree of deviation from the baseline transition model is quantified to obtain the topological instability tendency index. The degree of deviation is determined by comprehensively quantifying the frequency of topological stable state transitions per unit time, the proportion of transient time that the state feature vector sequence stays outside any identified topological stable state, and the difference between the probability distribution of the current actual transition behavior and the probability distribution of the baseline transition model.

[0025] The component-level state feature includes the node dissipation energy spectrum entropy. This feature is specifically used to diagnose the fault mode of the node with the most significant energy dissipation determined by the entropy production topological centrality. The extraction steps are as follows: first, for the node determined by the entropy production topological centrality, the corresponding original monitoring signal is obtained; then, the power spectrum of the original monitoring signal is calculated to obtain the dissipation energy spectrum that describes the distribution of the node's energy dissipation in the frequency domain. Finally, the dissipated energy spectrum is normalized to obtain the normalized energy spectrum , and calculate its information entropy to obtain the node dissipated energy spectrum entropy :

[0026] ;

[0027] in, is the power spectrum of the original monitoring signal corresponding to the node, is the normalized energy spectrum obtained by normalizing the power spectrum, is the frequency, is the frequency element.

[0028] Step S5: Combine the system-level status features with the component-level status features to generate a multi-dimensional comprehensive status assessment vector. This step provides a quantitative, multi-dimensional analysis basis for operating status assessment and maintenance decision-making of hydropower station equipment.

[0029] Step S6: Perform collaborative fault diagnosis based on the system-level state characteristics and the component-level state characteristics to generate a diagnostic conclusion containing fault location information and fault mode information. This step is specifically implemented by: using the physical components determined by the entropy-generating topological centrality to determine the fault location information in the diagnostic conclusion; using the value of the node dissipated energy spectrum entropy to determine whether the fault mode in the diagnostic conclusion is a fault mode in which energy is concentrated in a deterministic frequency or a fault mode in which energy is dispersed across a wide frequency band; and correlating the fault location information with the fault mode information to generate the diagnostic conclusion.

[0030] The present invention provides a method for extracting and evaluating features from hydropower station equipment status monitoring data. It has the following beneficial effects:

[0031] 1. This invention addresses the technical challenge of integrating and analyzing data from different physical dimensions by constructing an energy transfer topology network model based on the physical structure of the equipment and uniformly representing heterogeneous monitoring signals from multiple sources, such as vibration, temperature, and acoustics, as energy dissipation rates. This approach enables equipment status assessment within a unified, physically meaningful energy dimension, thereby establishing a systematic analytical framework and improving the comprehensiveness and consistency of status assessments.

[0032] 2. The present invention achieves deep insight into the device status by extracting state characteristics in multiple dimensions such as macro-static, macro-dynamic and micro-patterns. The total system entropy production rate and entropy production topological centrality quantify the energy efficiency of the system as a whole and locate the key dissipation sources; the main path information fidelity uses transfer entropy to capture the causal anomalies of energy flow; in particular, the topological instability tendency index, by analyzing the system's transition behavior between different steady states, realizes a dynamic assessment of operational stability and can provide early warning of slowly developing potential faults.

[0033] 3. The present invention realizes the precise judgment of fault location and fault mode through collaborative diagnosis, and improves the accuracy and guidance of diagnostic conclusions. First, the entropy-generating topological centrality is used to determine the physical component with the most significant energy dissipation, thereby locking the fault location information; then, the node dissipation energy spectrum entropy of the component is extracted, and the disordered distribution of energy in the frequency domain is quantified to determine whether it is a deterministic fault with concentrated energy or a random fault with dispersed energy. Finally, a complete diagnostic conclusion including location and mode is generated, providing a direct basis for subsequent maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a flow chart of the method of the present invention;

[0035] Figure 2 Schematic diagram of the energy transmission topology network model of the generator set of the present invention;

[0036] Figure 3 This is a schematic diagram of the monitoring signal layout and energy rate conversion of the present invention;

[0037] Figure 4 This is a schematic diagram of the multi-dimensional state feature extraction process of the present invention;

[0038] Figure 5 It is a schematic diagram of the comprehensive evaluation and collaborative fault diagnosis process of the present invention. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] Refer to the attached Figure 1 , Figure 1 This is a flow chart of a method for extracting and evaluating features from hydropower station equipment status monitoring data according to an embodiment of the present invention. In a specific embodiment of the technical solution provided by the present invention, a Francis turbine generator set is used as the monitoring object. The method may include the following steps:

[0041] Step S1: Based on the physical structure of the hydropower station equipment, an energy transfer topology network model is constructed, wherein the energy transfer topology network model includes nodes representing physical components of the equipment and edges representing energy transfer paths between the physical components.

[0042] Step S2: Acquire multi-source heterogeneous monitoring signals corresponding to the node.

[0043] Step S3: performing a unified energy rate characterization on the multi-source heterogeneous monitoring signals to obtain an energy dissipation rate characterizing the energy dissipation level of each of the physical components.

[0044] Step S4: extracting system-level state characteristics and component-level state characteristics based on the energy transfer topology network model and the energy dissipation rate.

[0045] Step S5: combining the system-level state characteristics with the component-level state characteristics to generate a multi-dimensional comprehensive state assessment vector.

[0046] Step S6: Perform collaborative fault diagnosis based on the system-level status characteristics and the component-level status characteristics to generate a diagnosis conclusion including fault location information and fault mode information.

[0047] In this embodiment, step S1 is first performed to construct an energy transfer topology network model based on the mechanical structure and energy flow paths of the Francis turbine generator set. This model defines key physical components of the unit, such as the water guide mechanism, runner, main shaft, upper guide bearing, lower guide bearing, thrust bearing, generator rotor, and generator stator, as nodes. The directed edges connecting these nodes represent the paths through which energy is transferred, converted, and dissipated between these components.

[0048] Next, step S2 is executed. For each defined node, a sensor network deployed on the turbine is used to acquire multi-source heterogeneous monitoring signals during operation. These signals include, but are not limited to, vibration acceleration signals from bearing seats, temperature signals from bearings and stator windings, and acoustic pressure signals from the turbine volute or near the draft tube.

[0049] Then, step S3 is executed to uniformly characterize the energy rate of the acquired multi-source heterogeneous monitoring signals, converting the signal data of different physical dimensions into energy dissipation rate in watts W. This step is the basis for subsequent unified analysis. The specific conversion calculation is as follows:

[0050] For the vibration signal, the corresponding kinetic energy dissipation power It can be calculated by the following formula:

[0051] ;

[0052] Where, is the kinetic energy dissipation power, is the equivalent mass of the physical component, is the instantaneous acceleration signal of the component, is the instantaneous velocity obtained by integrating the instantaneous acceleration signal. For the temperature signal, the corresponding heat dissipation power is It can be calculated by the following formula:

[0053] ;

[0054] Where, is the heat dissipation power; is the specific heat capacity of the physical component; is the mass of the physical component; is the instantaneous temperature signal of the component The rate of change with respect to time.

[0055] For acoustic signals, the corresponding sound energy radiation power It can be calculated by the following formula:

[0056] ;

[0057] Where, is the sound energy radiation power, is the sound pressure monitoring signal, is the density of the propagation medium, is the speed of sound, is the equivalent acoustic radiation area of ​​the component. The total energy dissipation rate of a single node is the sum of all types of dissipated power.

[0058] After obtaining the energy dissipation rate of each node, step S4 is executed to extract state features that can characterize the operating status of the unit at different levels. The system-level state features include: total system entropy production rate, entropy production topological centrality, main path information fidelity, and topological instability tendency index. The component-level state features include node dissipation energy spectrum entropy. Node dissipation energy spectrum entropy The calculation method is:

[0059] ;

[0060] in, is the power spectrum of the original monitoring signal corresponding to the node, is the normalized energy spectrum obtained by normalizing the power spectrum, is the frequency, is the frequency element.

[0061] Refer to the attached Figure 2 , Figure 2 This is a schematic diagram of a Francis turbine generator set energy transfer topology network model constructed according to an embodiment of the present invention. Step S1, constructing the energy transfer topology network model, is the foundation of the present invention's method. This step ensures that subsequent state analysis is based on clear physical structural constraints.

[0062] This step may specifically include: first, physical component identification and node abstraction; then, energy path analysis and edge definition.

[0063] During object component identification and node abstraction, the physical entities responsible for the core energy transmission, conversion, support, and dissipation functions of the Francis turbine generator set are precisely identified and abstracted as nodes in the model. Each node uniquely corresponds to a specific, monitorable physical component. For the Francis turbine generator set in this embodiment, the physical components defined as nodes include: the guide mechanism, runner, main shaft, upper guide bearing, lower guide bearing, thrust bearing, generator rotor, and generator stator.

[0064] In energy path analysis and edge definition, the energy flow relationships between the aforementioned nodes, with clear physical directions, are defined as directed edges in the model. These directed edges objectively describe the process of energy flowing from one component to another, and their direction is determined by the causal relationship of energy transfer. These edges are divided into two categories based on their physical properties:

[0065] The first type is the main energy transfer edge. This edge represents the main path for converting hydraulic energy into mechanical energy and then into electrical energy. In this embodiment, the main energy transfer edges include: the edge from the water guide mechanism to the runner, representing the high-pressure water flow impacting the runner, converting hydraulic energy into mechanical energy for the runner's rotation; the edge from the runner to the main shaft, representing the mechanical energy transfer from the runner to the main shaft; and the edge from the main shaft to the generator rotor, representing the mechanical energy transfer from the main shaft to the generator rotor.

[0066] The second type is energy dissipation edges. This type of edge represents the path of energy that is inevitably dissipated in the form of heat, vibration, and sound during the energy transfer and conversion process. In this embodiment, the energy dissipation edges include: edges pointing from the main shaft to the upper guide bearing, lower guide bearing, and thrust bearing, representing the heat and vibration energy dissipated by mechanical friction at each bearing during main shaft rotation; and edges pointing from the generator stator to the external environment or cooling system, representing the heat energy dissipated by the stator winding due to the thermal effects of current and core losses.

[0067] After defining the nodes and edges described above, the energy transfer topology network model can be stored as a computer-readable data structure, such as an adjacency matrix or adjacency table, during computational implementation. This data structure explicitly records the identifier of each node and its corresponding physical component, as well as the direction and type of the edges connecting each node (primarily energy transfer or energy dissipation), providing structured input for subsequent computational program calls.

[0068] Refer to the attached Figure 3 , Figure 3 This is a schematic diagram of the monitoring signal layout and energy rate conversion of a hydro-generator set according to an embodiment of the present invention. After completing the construction of the energy transfer topology network model, the method of the present invention then performs steps S2 and S3, namely, signal acquisition and energy rate unified representation.

[0069] The process may specifically include: first, deploying monitoring points and collecting signals; then, performing multi-source data energy conversion.

[0070] In the monitoring point layout and signal collection step S2, for each node defined in step S1, a corresponding sensor is deployed at its physical location or adjacent area to obtain real-time monitoring signals of its operating status. The type and installation location of the sensor are determined by the physical component it corresponds to and the physical quantity to be monitored. In this embodiment, the specific layout of signal collection is:

[0071] Triaxial accelerometers are installed on the bearing seats of the upper guide bearing, lower guide bearing, and thrust bearing to collect radial, axial, and tangential vibration signals at these nodes. Furthermore, platinum resistance temperature sensors are installed inside the pads or oil tanks of each bearing to collect operating temperature signals.

[0072] A temperature sensor is installed on the stator core or winding end of the generator to collect the temperature signal when the stator is running.

[0073] An acoustic pressure sensor or microphone is installed on the outside of the unit volute or on the wall of the tailwater pipe to collect acoustic signals generated by water flow and unit operation.

[0074] All sensors convert the collected analog signals into digital time series signals through the data acquisition system, providing data input for the next step of calculation.

[0075] In the multi-source data energy rate conversion step S3, the monitoring signals of different physical dimensions collected in step S2 are uniformly converted into energy dissipation rates in watts.

[0076] When calculating energy conversion rates, the required physical component parameters, such as equivalent mass, specific heat capacity, mass, and equivalent acoustic radiation area, can be pre-calculated by consulting equipment design drawings, manufacturer data sheets, or through modeling and simulation methods such as finite element analysis (FEA). Once these parameters are determined, they are stored in the system as fixed values ​​for later use during calculations.

[0077] The vibration acceleration signal collected from each bearing node is converted into kinetic energy dissipation power using the kinetic energy dissipation power formula based on the equivalent mass of its components.

[0078] The temperature signals collected from each bearing node and generator stator node are converted into thermal energy dissipation power using the thermal energy dissipation power formula based on the specific heat capacity and mass of their components.

[0079] The collected acoustic pressure signal is converted into acoustic energy radiation power using the acoustic energy radiation power formula based on the physical properties of the propagation medium and the equivalent acoustic radiation area of ​​the component.

[0080] Through this conversion, signals such as vibration, temperature, and sound pressure, originally expressed in various physical units, are mapped onto the unified dimension of energy dissipation rate. For a node experiencing multiple forms of energy dissipation (for example, a bearing node generating both vibration and heat), its total energy dissipation rate is the arithmetic sum of all energy dissipation powers (kinetic, thermal, and so on) at that node. This step lays the foundation for subsequent extraction of system- and component-level features within a unified framework.

[0081] Refer to the attached Figure 4 , Figure 4 Figure 2 is a schematic diagram of a multi-dimensional state feature extraction process according to one embodiment of the present invention. After calculating the energy dissipation rate of each node in step S3, the method of the present invention proceeds to step S4, which extracts system-level and component-level state features based on the energy transfer topology network model and the energy dissipation rate of each node.

[0082] This step may specifically include: first, extracting system-level state features; and then, extracting component-level state features.

[0083] When extracting system-level state features, this embodiment extracts the following four features:

[0084] Total system entropy production rate. This characteristic is calculated by arithmetically summing the energy dissipation rates of all nodes in the energy transfer topology network model defined in step S1. The result is a scalar value that quantifies the total energy loss incurred by the hydro-generator unit as a whole due to irreversible processes per unit time, providing a macroscopic physical indicator for evaluating the unit's overall operating efficiency.

[0085] Entropy production topological centrality. This feature is extracted by comparing the energy dissipation rates of all nodes in the energy transfer topology network model at the same moment, identifying and outputting the unique identifier of the node with the highest energy dissipation rate. The output of this feature directly indicates the physical component with the highest energy dissipation in the entire unit at the current moment, which is used for subsequent fault source location.

[0086] Main path information fidelity. The method for extracting this feature is as follows: first, based on the energy transfer topology network model defined in step S1, a main energy transfer path is determined, for example, from the runner node to the main shaft node and then to the generator rotor node. Then, for any two adjacent nodes on the path, the time series of their respective energy dissipation rates are obtained, and the transfer entropy of the upstream node to the downstream node is calculated. Finally, the transfer entropies calculated between all adjacent node pairs on the main path are weighted and summed to obtain a scalar value. This value quantifies the causal fidelity of energy flow on the core transfer link. Any change in its value indicates that there is an abnormality in the energy transfer law on the main path.

[0087] Topological instability tendency index. The extraction of this feature is a dynamic evaluation process: first, the system-level state features extracted above are combined into a state feature vector and a vector sequence is formed in chronological order. Then, a clustering algorithm, such as K-means clustering (K-Means), is used to analyze the historical state feature vector sequence of the unit during the period determined to be normal operation, and several stable cluster centers are identified, each center representing a typical topological steady state. Next, based on the transition frequency between each topological steady state in normal historical data, a baseline state transition model, such as a Markov model, is established. Finally, by monitoring the actual transition behavior of the current state feature vector sequence between each topological steady state and comprehensively quantifying its transition frequency, the proportion of transient time outside any steady state, and the difference between the actual transition probability distribution and the baseline transition model (for example, using KL divergence calculation), the topological instability tendency index is obtained.

[0088] When extracting component-level state features, this embodiment extracts the node dissipation energy spectrum entropy. The extraction of this feature is targeted, and the extraction object is the node with the most significant energy dissipation determined by the aforementioned entropy-generating topological centrality. The extraction steps are: first, obtain the original time domain monitoring signal corresponding to the specific node, such as a vibration signal. Then, the power spectrum of the signal is calculated by methods such as Fourier transform to obtain a dissipation energy spectrum that describes the distribution of the node's energy dissipation in the frequency domain. Finally, the dissipation energy spectrum is normalized and calculated according to the calculation formula of information entropy to obtain the node dissipation energy spectrum entropy. The level of this entropy value reflects the orderliness of the energy distribution in the frequency domain, which is used to judge the failure mode of the component in step S6.

[0089] Refer to the attached Figure 5 , Figure 5 1 is a flow chart of comprehensive evaluation and collaborative fault diagnosis according to an embodiment of the present invention. After completing the extraction of multi-dimensional state features in step S4, the method of the present invention then performs steps S5 and S6.

[0090] The process may specifically include: first, generating a comprehensive state assessment vector; and then performing collaborative fault diagnosis based on the features.

[0091] When generating the comprehensive state assessment vector (step S5), all state features calculated at the same point in time in step S4 are combined. Specifically, system-level state features such as the total system entropy production rate, entropy production topological centrality, main path information fidelity, and topological instability tendency index, as well as the component-level state feature (the dissipated energy spectrum entropy of the node determined by the entropy production topological centrality), are arranged in a predefined order to form a high-dimensional comprehensive state assessment vector. This vector provides a quantitative, multi-dimensional digital snapshot of the hydro-generator unit's operating status at that moment, which can be recorded and used for subsequent equipment status trend analysis and maintenance decision-making.

[0092] When performing a weighted summation of transfer entropy, the weight coefficients between each node pair can be set based on their importance in the primary energy transfer path or energy transfer efficiency. In one specific implementation, all weight coefficients can be set to 1, i.e., an arithmetic average or direct summation can be performed. In another specific implementation, nodes with higher energy transfer efficiency can be assigned higher weights.

[0093] When performing the collaborative fault diagnosis step S6, the execution of this step is usually initiated by a trigger condition, such as monitoring that the total system entropy production rate or the topological instability tendency index exceeds the preset normal operation threshold. Once triggered, the specific implementation of the diagnostic process is as follows:

[0094] First, the entropy-generated topological centrality feature is used to determine the fault location information in the diagnostic conclusion. The output of this feature, namely the identifier of the node with the highest energy dissipation rate at the current moment, is directly used to locate the physical location of the fault, for example, the thrust bearing node.

[0095] The node dissipated energy spectrum entropy characteristics are then used to determine the fault mode information in the diagnostic conclusion. The method extracts the corresponding node dissipated energy spectrum entropy value for the thrust bearing node located in the previous step. This value is compared with a pre-established baseline value for the energy spectrum entropy of the component under normal operating conditions.

[0096] If the dissipated energy spectrum entropy value of the node is low, indicating that its energy dissipation is mainly concentrated in a few deterministic frequencies, then its fault mode is judged as a deterministic frequency fault mode, which includes rotor imbalance, shaft misalignment, etc.

[0097] If the dissipated energy spectrum entropy value of the node is high, indicating that its energy dissipation is widely distributed in a wide frequency band, then its fault mode is judged to be a wide-band fault mode, which includes random wear of bearings, deterioration of lubrication status, etc.

[0098] Finally, the aforementioned determined fault location information, such as thrust bearing, is associated and combined with the judged fault mode information, such as broadband fault mode, to generate a final, structured diagnosis conclusion.

[0099] The preset normal operating threshold is determined by collecting long-term operational data from the equipment under a variety of typical, confirmed fault-free operating conditions (e.g., varying loads and water heads) to form a normal operating database. Statistical methods, such as calculating the mean and standard deviation of the total system entropy production rate or topological instability propensity index in the database, are then used to determine the threshold for triggering diagnosis. The mean plus three times the standard deviation is used as the threshold.

[0100] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for extracting and evaluating features from hydropower station equipment status monitoring data, characterized in that: The following steps are involved: Step S1: constructing an energy transfer topology network model based on the physical structure of the hydropower station equipment, wherein the energy transfer topology network model includes nodes representing physical components of the equipment and edges representing energy transfer paths between the physical components; Step S2: Acquire multi-source heterogeneous monitoring signals corresponding to the node; Step S3: performing a unified energy rate characterization on the multi-source heterogeneous monitoring signals to obtain an energy dissipation rate characterizing the energy dissipation level of each physical component; Step S4: extracting system-level state features and component-level state features based on the energy transfer topology network model and the energy dissipation rate; Step S5: combining the system-level state features with the component-level state features to generate a multi-dimensional comprehensive state assessment vector; Step S6: Perform collaborative fault diagnosis based on the system-level status characteristics and the component-level status characteristics to generate a diagnosis conclusion including fault location information and fault mode information.

2. A method for extracting and evaluating features from equipment status monitoring data of a hydropower station according to claim 1, characterized in that: The step of uniformly characterizing the energy rates of the multi-source heterogeneous monitoring signals in step S3 is specifically implemented as follows: for the monitoring signals associated with a single node, uniformly converting them from their respective physical dimensions to an energy rate dimension to obtain the total energy dissipation rate of the node; wherein the conversion includes: Converting the vibration signals in the multi-source heterogeneous monitoring signals into kinetic energy dissipation power according to the equivalent mass of the corresponding physical components; Converting the temperature signal in the multi-source heterogeneous monitoring signal into thermal energy dissipation power according to the specific heat capacity and mass of the corresponding physical component; The acoustic signal in the multi-source heterogeneous monitoring signal is converted into acoustic energy radiation power according to the physical properties of its propagation medium and the equivalent acoustic radiation area.

3. A method for extracting and evaluating features from equipment status monitoring data of a hydropower station according to claim 1, characterized in that: The system-level state characteristics extracted in step S4 include the total system entropy production rate; the method for determining the total system entropy production rate is: summing the energy dissipation rates of all nodes in the energy transfer topology network model, and using the result as a macro indicator to quantify the overall energy ineffective dissipation level of the hydropower station equipment.

4. A method for extracting and evaluating features from equipment status monitoring data of a hydropower station according to claim 1, characterized in that: The system-level state characteristics extracted in step S4 include entropy production topological centrality; the method for determining the entropy production topological centrality is: among all the nodes constituting the energy transfer topological network model, identifying the node with the largest energy dissipation rate, and using the identifier of the node as the entropy production topological centrality to locate the physical component with the most significant energy dissipation.

5. A method for extracting and evaluating features from equipment status monitoring data of a hydropower station according to claim 1, characterized in that: The system-level state features extracted in step S4 include the main path information fidelity; the main path information fidelity extraction step is specifically implemented as follows: First, a main energy transfer path is pre-set in the energy transfer topology network model; Then, for any two adjacent nodes on the main energy transfer path, the time series of their respective energy dissipation rates are obtained, and the transfer entropy between the two time series is calculated to quantify the causal relationship strength of the upstream node on the energy information flow of the downstream node; Finally, the transfer entropies calculated between all adjacent nodes on the main energy transfer path are weightedly summed to obtain the main path information fidelity.

6. A method for extracting and evaluating features from hydropower station equipment status monitoring data according to claim 1, characterized in that: The system-level state features extracted in step S4 also include a topological instability tendency index, and the extraction step of the index is specifically implemented as follows: First, the system-level state characteristics are transformed into a state feature vector sequence in the time dimension; Then, by clustering the state feature vector sequences of the hydropower station equipment during normal operation, the topological steady states corresponding to different typical operating conditions are identified. Next, based on historical data of the device switching normally between the topological stable states, a benchmark transition model is established to characterize its normal state transition law; Finally, the topological instability tendency index is obtained by monitoring the actual transition behavior of the current state eigenvector sequence between the topological stable states and quantifying the degree of deviation between the actual transition behavior and the benchmark transition model.

7. A method for extracting and evaluating features from equipment status monitoring data of a hydropower station according to claim 6, characterized in that: The degree of deviation is determined by comprehensively quantifying the following factors: the frequency of topological steady-state transitions per unit time, the proportion of transient time that the state eigenvector sequence stays outside any identified topological steady state, and the difference between the probability distribution of the current actual transition behavior and the probability distribution of the benchmark transition model.

8. A method for extracting and evaluating features from equipment status monitoring data of a hydropower station according to claim 4, characterized in that: The component-level state features extracted in step S4 include node dissipation energy spectrum entropy, which is specifically used to perform fault mode diagnosis on the node with the most significant energy dissipation determined by the entropy-generating topological centrality. The extraction steps are specifically implemented as follows: First, for the node determined by the entropy production topological centrality, the corresponding original monitoring signal is obtained; Then, the power spectrum of the original monitoring signal is calculated to obtain a dissipated energy spectrum describing the distribution of energy dissipation of the node in the frequency domain; Finally, the dissipated energy spectrum is normalized and its information entropy is calculated to obtain the node dissipated energy spectrum entropy.

9. A method for extracting and evaluating features from hydropower station equipment status monitoring data according to claim 1, characterized in that: The step of generating a multi-dimensional comprehensive status assessment vector in step S5 is to provide a quantitative, multi-dimensional analysis basis for the operating status assessment and maintenance decision-making of the hydropower station equipment. This step is specifically achieved by combining the system-level status features and component-level status features extracted in step S4.

10. A method for extracting and evaluating features from hydropower station equipment status monitoring data according to claim 8, characterized in that: The step of performing collaborative fault diagnosis in step S6 is specifically implemented as follows: Determining fault location information in the diagnosis conclusion using the physical component determined by the entropy-generated topological centrality; Using the value of the node dissipated energy spectrum entropy, it is determined whether the fault mode in the diagnosis conclusion is a fault mode in which energy is concentrated in a deterministic frequency or a fault mode in which energy is dispersed in a wide frequency band; The fault location information is associated with the fault mode information to generate the diagnosis conclusion.

Images