State monitoring and early warning method and system for vibration power signal fusion of steam turbine generator unit

Abstract

The invention relates to the technical field of monitoring and early warning, and discloses a state monitoring and early warning method and system for vibration power signal fusion of a steam turbine generator unit, and the method comprises the steps: collecting vibration signal data and power signal data of the steam turbine generator unit in a set period, and processing the collected vibration signal data, extracting and analyzing to obtain a vibration fault mapping characteristic value of the steam turbine generator unit; processing the collected power signal data through a pre-trained power mapping neural network, extracting and analyzing to obtain a power fault mapping characteristic value, and combining the power fault mapping characteristic value with the vibration fault mapping characteristic value to perform fusion analysis to obtain an energy vibration cooperative fault characteristic value; and performing fault early warning processing on the steam turbine generator unit through the energy vibration cooperative fault characteristic value. The technical problem that in the prior art, collaborative analysis of power and vibration signals is difficult to achieve, and consequently fault recognition is lack of accuracy is solved.

Description

Technical Field

[0001] This invention relates to the field of monitoring and early warning technology, specifically to a state monitoring and early warning method and system for fusion of vibration power signals of steam turbine generator sets. Background Technology

[0002] With the rapid development of the power industry, large steam turbine generator sets, as the core power equipment of the power system, are constantly evolving towards high-parameter, high-load, and wide-load regulation operation. This development trend makes the operating conditions of the units increasingly complex, placing more stringent requirements on the real-time monitoring of the unit's operating status and early fault identification. Under typical complex operating conditions such as rapid start-up and shutdown, deep peak shaving, and frequent load changes, there is a significant dynamic coupling relationship between the unit's mechanical vibration response and power output. Abnormalities in any link may cause overall vibration imbalance, power fluctuations, and in severe cases, even system instability, resulting in significant economic losses and safety hazards.

[0003] Currently, the industry primarily uses traditional monitoring methods for monitoring the operational status of steam turbine generator sets, relying on either a single vibration signal or a single power signal. While these methods have proven effective under steady-state operating conditions, they are insufficient to meet the practical needs of accurate early fault identification under the complex dynamic conditions described above. The core issue lies in the fact that in existing monitoring systems, vibration signal monitoring and power signal monitoring largely operate independently, lacking the capability for dynamic correlation analysis and fusion modeling of these two cross-domain data sets. This results in significant deficiencies in both monitoring accuracy and early warning timeliness.

[0004] Specifically, the limitations of existing technologies are mainly reflected in the following aspects: During the operation of steam turbine generator sets, power output and mechanical vibration naturally exhibit complex dynamic coupling characteristics. When the unit load fluctuates or operating conditions switch, the responses of power signals and vibration signals are significantly asynchronous, which can easily lead to deviations in the state assessment of the unit's energy transfer chain, thereby affecting the accurate judgment of the unit's stable operating state. For example, when the unit's power is disturbed in a short period of time, relying solely on a single vibration signal or a single power signal for state identification cannot accurately establish the correspondence between energy fluctuations and mechanical vibrations, causing some early abnormal features of faults to be ignored or misjudged as normal operating condition fluctuations. Under extreme and complex operating conditions such as deep peak shaving and rapid start-up and shutdown, the difference between the characteristic change patterns of vibration signals and the trend changes of power signals is even more prominent. If the joint and coordinated analysis of the two signals cannot be achieved, it will be difficult to capture potential fault signs under the coupling effect of the two, ultimately leading to a lag in fault warning response and missing the best opportunity for fault intervention.

[0005] In summary, existing turbine generator monitoring technologies based on single signals lack the ability to fuse and analyze cross-domain data, making them unsuitable for the complex operating conditions of high-parameter, wide-load regulating units. There is an urgent need for a monitoring technology that can achieve dynamic correlation modeling of power and vibration data across domains to improve the accuracy and timeliness of early fault identification. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, the present invention aims to provide a condition monitoring and early warning method and system for generator set vibration power signal fusion, so as to solve the technical problem that it is difficult to achieve collaborative analysis of power and vibration signals in the existing technology, resulting in a lack of accuracy in fault identification.

[0007] This invention is achieved through the following technical solution: In a first aspect, the present invention provides a condition monitoring and early warning method for turbine generator set vibration power signal fusion, comprising: Vibration signal data and power signal data of the steam turbine generator set are collected within a set period. The vibration signal data includes the vibration amplitude value at each vibration time point, and the power signal data includes the power amplitude value at each power time point. The collected vibration signal data is processed to extract and analyze the vibration fault mapping feature values ​​of the steam turbine generator set. The collected power signal data is processed by a pre-trained power mapping neural network to extract and analyze power fault mapping feature values. The power fault mapping feature values ​​are then combined with the vibration fault mapping feature values ​​to obtain energy-vibration synergistic fault feature values. Fault early warning processing of steam turbine generator sets is carried out by using energy vibration coordinated fault characteristic values.

[0008] Preferably, the specific steps for processing the collected vibration signal data and extracting and analyzing the vibration fault mapping feature values ​​of the steam turbine generator set are as follows: The vibration signal data is divided into time periods to obtain vibration sub-signal data for several time periods. The vibration sub-signal data includes the vibration response value at each vibration time point. The vibration sub-signal data for each time period are deconstructed to obtain the energy-frequency mapping feature set corresponding to each time period. The energy-frequency mapping feature set includes vibration impulse energy feature value and frequency-energy co-mapping feature value. Based on the energy-frequency mapping feature set of each time period, the vibration fault mapping feature value of the steam turbine generator set is obtained by analysis.

[0009] Furthermore, the specific steps for dividing the vibration signal data into time periods to obtain vibration sub-signal data for several time periods are as follows: Based on the vibration amplitude value at each vibration time point, calculate the vibration energy change values ​​of several sets of adjacent vibration time points; Based on the vibration energy change values ​​of each group of adjacent vibration time points, the vibration sub-signal data for each time period are obtained.

[0010] Furthermore, the specific steps for deconstructing the vibrator signal data for each time period to obtain the energy-frequency mapping feature set corresponding to each time period are as follows: The vibration sub-signal data for each time period are processed in the time domain to obtain the vibration impulse energy characteristic value for that time period; Frequency domain extraction processing is performed on the vibrator signal data for each time period to obtain the frequency-energy co-encoding characteristic value for that time period.

[0011] Preferably, the pre-trained power mapping neural network includes an input layer, a convolutional feature extraction layer, a dimensionality reduction and fusion layer, and an output mapping layer.

[0012] Furthermore, the specific steps for analyzing and obtaining the power fault mapping characteristic values ​​are as follows: The power signal data is input into a pre-trained power mapping neural network, and the power anomaly mapping feature set is obtained by analysis. The power anomaly mapping feature set includes power disturbance and instability feature values, power drift feature values, and power complexity feature values. Based on the power anomaly mapping feature set, the power fault mapping feature value of the steam turbine generator set is obtained through analysis.

[0013] Furthermore, the specific steps for inputting the power signal data into a pre-trained power mapping neural network and analyzing it to obtain the power anomaly mapping feature set are as follows: In the input layer, the power signal data is received and preprocessed; In the convolutional feature extraction layer, a power mapping feature vector is extracted based on the preprocessed power signal data; In the dimensionality reduction fusion layer, the power mapping feature vector is correlated to obtain the dimensionality reduction feature vector; In the output mapping layer, a power anomaly mapping feature set is output based on the dimensionality-reduced feature vector.

[0014] Preferably, the specific steps for fusing the power fault mapping feature value with the vibration fault mapping feature value to obtain the energy-vibration coordinated fault feature value are as follows: Read the vibration signal data and power signal data, divide the vibration amplitude value and power amplitude value corresponding to each time point, perform correlation analysis on the two, and obtain the signal coordinated regulation coefficient; The power fault mapping feature value and vibration fault mapping feature value are read, and combined with the signal coordinated adjustment coefficient, the energy-vibration coordinated fault feature value is obtained by analysis; The specific formula for calculating the characteristic value of the energy-vibration coordinated fault of the steam turbine generator set is as follows: ; in, The characteristic value of the energy-vibration coordinated fault of the steam turbine generator set. For vibration fault mapping characteristic values ​​of steam turbine generator sets, The vibration fault adjustment coefficients are stored in the database. For power fault mapping characteristic values ​​of steam turbine generator sets, The power fault adjustment coefficients are stored in the database. This refers to the signal coordination regulation coefficient of the steam turbine generator set.

[0015] Preferably, the specific steps for fault early warning processing of steam turbine generator sets using energy-vibration coordinated fault characteristic values ​​are as follows: The energy-vibration coordinated fault characteristic values ​​of the steam turbine generator set are compared with the preset energy-vibration coordinated fault characteristic thresholds. Fault warnings are issued for steam turbine generator sets based on the comparison and processing results.

[0016] Secondly, the present invention also provides a state monitoring and early warning system for turbine generator set vibration power signal fusion, used to implement the above-mentioned state monitoring and early warning method for turbine generator set vibration power signal fusion, comprising: The data acquisition module is used to collect vibration signal data and power signal data of the steam turbine generator set within a set period. The vibration signal data includes the vibration amplitude value at each vibration time point; the power signal data includes the power amplitude value at each power time point. The vibration analysis module is used to process the collected vibration signal data, extract and analyze the vibration fault mapping feature values ​​of the steam turbine generator set; The fusion analysis module is used to process the collected power signal data through a pre-trained power mapping neural network, extract and analyze the power fault mapping feature values, and combine the power fault mapping feature values ​​with the vibration fault mapping feature values ​​to obtain the energy-vibration synergistic fault feature values. The early warning processing module is used to perform fault early warning processing on steam turbine generator sets based on the energy vibration coordinated fault characteristic values.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects: This invention provides a state monitoring and early warning method for generator set vibration and power signal fusion. By jointly analyzing the power signal and vibration signal of the turbine generator set, a fusion model of power fault mapping feature value and vibration fault mapping feature value is constructed, realizing multi-dimensional collaborative representation of the energy domain and mechanical domain. Based on the signal collaborative adjustment coefficient, a dynamic correlation relationship between the two types of signals is established, and the energy-vibration collaborative fault feature value is used as a unified index to comprehensively reflect the abnormal coupling state of the unit at the energy transfer and mechanical response levels. This enables a holistic judgment of unit faults, and can accurately identify hidden abnormal trends even when there are differences between power fluctuations and vibration changes. This achieves refined monitoring of the operating status of the turbine generator set, thereby significantly improving the reliability of the state judgment.

[0018] Furthermore, by introducing a pre-trained power mapping neural network into power signal analysis, the automatic extraction and fusion of power disturbance instability features, power drift features, and power complexity features are achieved. This network model can perform feature dimensionality reduction and nonlinear mapping processing on the original power signal, thereby effectively capturing the multi-scale dynamic change features of the power signal under complex operating conditions. Then, combined with the power anomaly mapping feature set, power fault mapping feature values ​​are generated, which significantly enhances the fault identification accuracy under multiple operating conditions.

[0019] Furthermore, by deconstructing the vibration signal data of the steam turbine generator set into time-segmented segments, multiple vibration sub-signals are formed. Vibration impulse energy characteristic values ​​and frequency-energy co-mapping characteristic values ​​are extracted in both the time and frequency domains, constructing a complete energy-frequency mapping characteristic system. This enables the capture of vibration energy distribution characteristics of the unit under different load stages and operating periods, achieving dynamic tracking and identification of weak and progressive mechanical anomalies. In turn, it can comprehensively reflect the time-frequency coupling characteristics of vibration signals, effectively distinguishing between random disturbances and fault-induced structural vibration changes, thereby achieving refined diagnosis of the unit's vibration state. Attached Figure Description

[0020] Figure 1 This is a flowchart of the state monitoring and early warning method for turbine generator set vibration and power signal fusion according to the present invention; Figure 2 This is a flowchart illustrating the specific steps involved in obtaining the vibration fault mapping feature value of a steam turbine generator set in the state monitoring and early warning method for fusing vibration and power signals of a steam turbine generator set according to the present invention. Figure 3 This is a flowchart illustrating the specific steps involved in analyzing the power anomaly mapping feature set of a steam turbine generator set in the state monitoring and early warning method for fusing vibration and power signals of a steam turbine generator set according to the present invention. Figure 4 This is a schematic diagram of the state monitoring and early warning system for the turbine generator set, which integrates vibration and power signals according to the present invention. In the diagram: 1. Data acquisition module; 2. Vibration analysis module; 3. Fusion analysis module; 4. Early warning processing module. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] The purpose of this invention is to provide a condition monitoring and early warning method and system for generator set vibration and power signal fusion, so as to solve the technical problem that it is difficult to achieve collaborative analysis of power and vibration signals in the prior art, resulting in a lack of accuracy in fault identification.

[0024] The present invention will now be described in further detail with reference to the accompanying drawings: See Figure 1 In one embodiment of the present invention, a state monitoring and early warning method for fusion of vibration power signals of a steam turbine generator set is provided, comprising: Step 1: Collect vibration signal data and power signal data of the steam turbine generator set within a set period. The vibration signal data includes the vibration amplitude value at each vibration time point; the power signal data includes the power amplitude value at each power time point. In this embodiment, the duration of the set period is the least common multiple of the sampling frequencies of the vibration signal data and the power signal data. For example, it can be 50-100 times the least common multiple time. In this embodiment, it can be 80 times the least common multiple time.

[0025] Step 2: Process the collected vibration signal data, extract and analyze the vibration fault mapping feature values ​​of the steam turbine generator set; Step 3: Process the collected power signal data through a pre-trained power mapping neural network, extract and analyze the power fault mapping feature values, and combine the power fault mapping feature values ​​with the vibration fault mapping feature values ​​to obtain the energy-vibration synergistic fault feature values. Step 4: Perform fault early warning processing on the steam turbine generator set by using the energy vibration coordinated fault characteristic value.

[0026] Specifically, the steps for fault early warning processing of steam turbine generator sets based on energy-vibration coordinated fault characteristic values ​​are as follows: The energy-vibration coordinated fault characteristic values ​​of the steam turbine generator sets are compared with preset energy-vibration coordinated fault characteristic thresholds; based on the comparison results, fault early warning is issued for the steam turbine generator sets, specifically: if the energy-vibration coordinated fault characteristic value of the steam turbine generator sets is higher than the preset energy-vibration coordinated fault characteristic threshold, a fault early warning notification is sent to relevant personnel; if the energy-vibration coordinated fault characteristic value of the steam turbine generator sets is lower than or equal to the preset energy-vibration coordinated fault characteristic threshold, monitoring of the steam turbine generator sets continues.

[0027] Specifically, such as Figure 2 As shown, the vibration signal data includes the vibration amplitude value at each vibration time point (the time interval between adjacent vibration time points is the sampling frequency of the vibration signal data, such as 2kHz to 10kHz, which can be taken as 10kHz in this embodiment). The specific steps to obtain the vibration fault mapping feature value of the steam turbine generator set are as follows: The vibration signal data of the steam turbine generator set is divided into time periods to obtain vibration sub-signal data for several time periods; the vibration sub-signal data for each time period of the steam turbine generator set is deconstructed to obtain the energy-frequency mapping feature set for its corresponding time period, including vibration impulse characteristics. Characteristic values ​​and frequency-energy co-mapping characteristic values; based on the energy-frequency mapping characteristic set of each time period of the steam turbine generator set, the vibration fault mapping characteristic value of the steam turbine generator set is analyzed. Specifically, based on the vibration impulse energy characteristic value and frequency-energy co-mapping characteristic value of each time period of the steam turbine generator set, the time-frequency co-mapping characteristic value of the corresponding time period is analyzed (used to characterize the multi-domain energy anomaly response intensity of the steam turbine generator set; the larger the value, the greater the possibility of fault occurrence), and a moving average processing is performed to obtain the vibration fault mapping characteristic value of the steam turbine generator set (used to characterize the probability of fault occurrence mapped by the vibration signal of the steam turbine generator set).

[0028] The specific formula for calculating the time-frequency coordinated characteristic value of a steam turbine generator set at a certain time period is as follows: ;in, This represents the time-frequency coordinated characteristic value of the steam turbine generator set at a certain time period. This represents the characteristic value of the vibration impulse energy of the steam turbine generator unit at a certain time period. This represents the frequency-energy co-occurrence characteristic value of the steam turbine generator unit at a certain time period. These are the interaction adjustment coefficients stored in the database. The difference adjustment coefficients are stored in the database. This is the smoothing adjustment coefficient stored in the database (and in this implementation example, the value is approximately 0.001).

[0029] It should be noted that the interaction adjustment coefficients stored in the database The acquisition steps are as follows: Read the vibration impulse energy characteristic value and frequency energy co-emission characteristic value of the steam turbine generator set for each time period, and extract the correlation value (absolute value) of the two based on the Pearson correlation coefficient method, and use it as the interaction adjustment coefficient. .

[0030] Difference adjustment coefficients stored in the database The acquisition steps are as follows: Read the vibration impulse energy characteristic value and frequency energy co-occurrence characteristic value of the steam turbine generator set for each time period, and perform difference processing (take the absolute value) to obtain the difference value for the corresponding time period. Extract the difference variance and difference mean, and perform ratio processing. Use the result as the difference adjustment coefficient. .

[0031] The specific steps to obtain the vibration sub-signal data of the steam turbine generator set for several time periods are as follows: Based on the vibration amplitude value of each vibration time point of the steam turbine generator set, analyze the vibration energy change value of several adjacent vibration time points. Specifically, the vibration amplitude value of each vibration time point of the steam turbine generator set is squared to obtain the vibration energy value of the corresponding time point, and the vibration energy values ​​of adjacent vibration time points are successively processed by difference (taking the absolute value) to obtain the vibration energy change value of several adjacent vibration time points. Based on the vibration energy change values ​​of each group of adjacent vibration time points of the steam turbine generator set, the vibration sub-signal data of each time period of the steam turbine generator set is analyzed. Specifically, the vibration energy change values ​​of each group of adjacent vibration time points are compared with a preset vibration energy change threshold. If there are multiple consecutive groups of adjacent vibration time points (e.g., 3 to 5 consecutive groups) with vibration energy change values ​​higher than the preset vibration energy change threshold, the first time point in the first group of adjacent vibration time points with vibration energy change values ​​higher than the threshold is taken as the starting point of the time period. If there are multiple consecutive groups of adjacent vibration time points (e.g., 3 to 5 consecutive groups) with vibration energy change values ​​lower than or equal to the preset vibration energy change threshold, the time period is determined accordingly. The change threshold is then used to determine the end point of the time period by taking the first time point among the adjacent vibration time points that are lower than or equal to the vibration energy change threshold in the first group. This results in several time periods (and when the duration of a certain time period, i.e., the total number of time points, is lower than the preset minimum duration threshold, the time compensation is extended to both sides of the start and end points of the time period by a preset amount, for example, by extending forward by 20% to 30% of the minimum duration threshold and backward by 50% to 70% of the minimum duration threshold, forming an expanded effective analysis time period). The vibration amplitude value of each vibration time point within this time period is marked as the vibration response value of the corresponding vibration time point, i.e., the vibration sub-signal data.

[0032] It should be noted that when no vibration energy change value is detected at any adjacent vibration time point higher than the preset vibration energy change threshold, it indicates that the turbine generator set is in a stable operating state. In this case, the vibration signal data of the turbine generator set can be divided into fixed durations, that is, the entire vibration signal is divided into several fixed duration periods by a preset time interval (e.g., 0.5s, 1s or other set values), and the vibration signal data in each period is used as vibration sub-signal data.

[0033] The vibration sub-signal data includes the vibration response value at each vibration time point. The specific steps to obtain the energy-frequency mapping feature set of the steam turbine generator set for each time period are as follows: Perform time-domain extraction processing on the vibration sub-signal data of the steam turbine generator set for each time period to obtain the vibration impulse energy feature value of the corresponding time period. Specifically, for the vibration response value at each vibration time point of each time period, calculate the maximum value of the vibration response in the corresponding time period and take it as the peak value. At the same time, extract the root mean square value of the vibration response in the corresponding time period (i.e., perform root mean square processing on the vibration response value at each vibration time point in the corresponding time period). Based on the vibration response value at each vibration time point in the corresponding period, the fourth-order central moment of vibration and the square of the second-order central moment of vibration are extracted and the ratio is processed to obtain the vibration kurtosis factor of the corresponding period. The kurtosis factor is then weighted with the peak value and root mean square value of the vibration response of the corresponding period to obtain the vibration impulse energy characteristic value of the corresponding period. This characteristic value is used to characterize the impact release intensity of the vibration energy of the turbine generator set in the period. When the unit experiences mechanical abnormalities such as bearing wear, rubbing, imbalance, or loosening, the energy of the vibration signal will be released instantaneously or fluctuate irregularly, resulting in a significant increase in this characteristic value. Frequency domain extraction processing is performed on the vibration sub-signal data of the steam turbine generator set for each time period to obtain the frequency-energy co-mapping characteristic value of the corresponding time period. Specifically, the fast Fourier transform is performed on the vibration sub-signal data of each time period to obtain each frequency component and its corresponding amplitude value for the corresponding time period. The frequency component corresponding to the maximum amplitude is taken as the dominant frequency. Based on this, the theoretical frequency positions corresponding to each integer harmonic and fractional harmonic of the dominant frequency (e.g., 1 / 2 harmonic, 3 / 2 harmonic, 2nd harmonic, 3rd harmonic, etc.) are statistically analyzed. A preset frequency search window (e.g., ±0.5Hz or ±1Hz range) is set near the frequency positions corresponding to each harmonic and fractional harmonic. Peak search processing is performed on the amplitude value within this search window, and the maximum amplitude value within each window is determined as the amplitude value of the corresponding harmonic or fractional harmonic. In this way, the harmonic amplitude and fractional harmonic amplitude of the steam turbine generator set in that time period are obtained. The amplitude values ​​of all frequency components in each time period are read and squared to obtain the spectral energy value of the corresponding frequency component. Taking the frequency component corresponding to the main frequency as the center, the energy values ​​within the adjacent preset frequency band range (e.g., the main frequency ±1Hz to 3Hz) are summed to obtain the main frequency band energy. At the same time, the spectral energy values ​​of all frequency components are summed to obtain the total energy. The ratio of the main frequency band energy to the total energy is processed to obtain the spectral energy concentration characteristic value, which characterizes the degree of energy concentration of the vibration signal near the main frequency in that time period. When the unit experiences abnormal energy diffusion due to structural loosening, bearing wear, or uneven meshing, this characteristic value decreases significantly. The amplitude values ​​of all frequency components in each time period are read, and the root mean square value of the dominant frequency amplitude is extracted. The amplitude value corresponding to the dominant frequency is then compared with the root mean square value of the dominant frequency amplitude to extract the dominant frequency impact factor, which characterizes the prominence of the dominant frequency component in the overall spectrum. When the turbine generator set experiences faults such as structural loosening, rotor imbalance, or local resonance, the dominant frequency impact factor increases significantly. The octave amplitude, sub-octave amplitude, spectral energy concentration feature value, and main frequency impact factor of each time period are weighted. In this weighting process, the spectral energy concentration feature value is transformed using the reciprocal suppression mapping function f(x)=1 / (1+x), i.e., 1 / (1+spectral energy concentration feature value), to obtain the frequency energy co-mapping feature value of the corresponding time period. This value is used to characterize the degree of spectral energy distribution diffusion of the turbine generator set vibration signal in that time period. The larger the value, the greater the probability of the unit experiencing structural loosening, rotor imbalance, or meshing abnormalities.

[0034] In this implementation scheme, by dividing the vibration signal of the steam turbine generator set into multiple time periods and extracting features in both the energy and frequency domains, a fine-grained characterization of the unit's mechanical state is achieved. Secondly, by dividing the vibration sub-signals by the energy change values ​​of adjacent vibration time points, abnormal energy release intervals can be identified in the early stages of vibration signal fluctuations, effectively capturing early signs of mechanical faults. Furthermore, by jointly calculating the vibration impulse energy feature value and the frequency energy co-reflection feature value, the energy impact characteristics of the vibration signal in the time domain and the energy diffusion characteristics in the frequency domain are considered simultaneously. This ensures that the extracted feature values ​​can reflect both mechanical impact anomalies, such as bearing wear and rotor imbalance, and structural anomalies, such as loosening and uneven meshing. Overall, this step achieves dynamic adaptive segmentation of the vibration signal, fine analysis of energy changes, and multi-domain fusion processing, thereby maintaining high stability and high sensitivity under complex operating conditions and improving the accuracy of fault identification.

[0035] Specifically, the power signal data includes the power amplitude value of each power time point (the time interval between adjacent power time points is the sampling frequency of the power signal data, such as 50Hz to 200Hz, which can be taken as 50Hz in this embodiment), and the power mapping neural network includes an input layer, a convolutional feature extraction layer, a dimensionality reduction fusion layer, and an output mapping layer.

[0036] The specific steps for analyzing the power fault mapping feature values ​​of a steam turbine generator set are as follows: Input the power signal data of the steam turbine generator set into a pre-trained power mapping neural network to analyze the power anomaly mapping feature set of the steam turbine generator set, including power disturbance and instability imbalance feature values, power drift feature values, and power complexity feature values; Based on the power anomaly mapping feature set of the steam turbine generator set, analyze the power fault mapping feature values ​​of the steam turbine generator set, specifically by weighting the power disturbance and instability imbalance feature values, power drift feature values, and power complexity feature values ​​of the steam turbine generator set to obtain the power fault mapping feature values ​​(used to characterize the probability of fault occurrence mapped by the power signal of the steam turbine generator set).

[0037] It should be noted that in this implementation example, the weight coefficients of each parameter in the weighted processing can be obtained by using sample entropy weights. Taking the weighted processing of power fault mapping feature values ​​as an example, the power disturbance and instability feature values, power drift feature values, and power complexity feature values ​​of several historical cycles are obtained, and their corresponding information entropy values ​​are extracted respectively. Then, their corresponding information entropy values ​​are transformed using the reciprocal suppression mapping function f(x)=1 / (1+x), such as 1 / (1+information entropy value of power disturbance and instability feature value), and summed to obtain the information entropy sum value. The corresponding transformed information entropy values ​​are then compared with the information entropy sum value to obtain the weight coefficients corresponding to each parameter.

[0038] like Figure 3 As shown, the specific steps for analyzing the power anomaly mapping feature set of the steam turbine generator set are as follows: In the input layer of the power mapping neural network, the power signal data of the steam turbine generator set is received and preprocessed, such as linear normalization or Z-Score standardization of the power signal data, which is converted to the standardized range (e.g., [0,1]), and the power signal data is smoothed by moving average filtering, low-pass filtering or wavelet threshold denoising, so as to filter out electrical measurement sampling noise and random disturbance signals; In the convolutional feature extraction layer of the power mapping neural network, based on the preprocessed power signal data of the steam turbine generator set, the power mapping feature vector of the steam turbine generator set is extracted. Specifically, it analyzes the power signal sequence through multiple convolutional kernels of different scales, extracts the local variation features of the power signal in the time dimension, and captures the dynamic variation pattern of the power signal. The convolution process can identify features such as rapid fluctuations, slow changes, and periodic responses in the power signal. The multidimensional feature results output by the convolutional feature extraction layer are concatenated to form the power mapping feature vector, such as: Read the power amplitude value at each power time point and perform mean processing to extract the power average response feature; sequentially perform trend processing on the power amplitude values ​​of two adjacent power time points, such as (power amplitude value of the first and second power time points / power amplitude value of the second power time point), and take the mean to extract the power change rate feature; read the result of the trend processing, that is, the power response change rate of several groups of adjacent power time points, count the maximum and minimum power response change rates, and perform difference processing to extract the power fluctuation feature; The power amplitude value at each power time point is read and autocorrelation processing is performed. That is, the autocorrelation coefficient is extracted under different time delays (the time delays can be set to integer multiples of the power signal sampling frequency, such as 1, 2, 3 times the sampling frequency up to the preset maximum delay range, such as 0.1s to 1s). The corresponding autocorrelation coefficients under all time delays are traversed, and the time delay corresponding to the autocorrelation coefficient changing from high to low is counted (that is, the first time delay among adjacent time delays when the difference between the autocorrelation coefficients corresponding to adjacent time delays is lower than the preset correlation threshold). This is used as a power stability feature to characterize the continuous stability of the turbine generator power signal. When the value is small, it indicates that the unit has abnormal disturbances in the regulation link, energy feedback or load response. The power response change rate of each group of adjacent power time points is read, and the mean of the rising power response change rate during the rising phase (i.e., the mean of the power response change rate of all adjacent power time points with a power response change rate higher than 0 is calculated and taken as the mean of the rising power response change rate) and the mean of the falling power response change rate during the falling phase (i.e., the mean of the power response change rate of all adjacent power time points with a power response change rate lower than 0 is calculated and taken as the mean of the falling power response change rate, and the absolute value is taken) are calculated. The mean of the rising power response change rate and the mean of the falling power response change rate are comprehensively processed, i.e., |mean of rising power response change rate - mean of falling power response change rate| / (mean of rising power response change rate + mean of falling power response change rate), to extract the power imbalance feature, which is used to characterize the dynamic response asymmetry of the turbine generator set during the power output change process. When the unit experiences abnormalities such as load disturbance response lag, uneven energy feedback, or inconsistent control links, the response difference between the rising and falling power phases increases, and this feature increases. Based on the power signal data of the steam turbine generator set, the power signal data is reconstructed in phase space according to the preset time delay and embedding dimension. That is, the power amplitude sequence of continuous power time points is segmented according to the time delay, and each power time point and its delayed multiple power amplitudes are combined into a set of multi-dimensional state vectors to form the evolution trajectory of the power signal in phase space. After obtaining the phase space evolution trajectory of the power signal, the distance between any two sets of adjacent state vectors in the trajectory is calculated to characterize the change amplitude of the power signal in multi-dimensional space. The distance change between the two sets of state vectors is continuously tracked in subsequent time intervals to reflect the divergence trend of the power signal in the time evolution process. The distance change rate of all adjacent state vectors is statistically processed to calculate its average divergence rate, and the average divergence rate is used as the power complexity feature value. When the unit is under the state of energy feedback coupling, control oscillation or nonlinear disturbance of the excitation system, the divergence speed of the power signal trajectory in phase space increases significantly, and this feature is significantly increased. The power average response characteristics, power change rate characteristics, power fluctuation characteristics, power stability characteristics, power imbalance characteristics, and power complexity characteristics are concatenated into a power mapping feature vector. In the dimensionality reduction fusion layer of the power mapping neural network, the power mapping feature vector of the steam turbine generator set is correlated to obtain the dimensionality reduction feature vector of the steam turbine generator set, which is as follows: The power average response feature, power change rate feature, and power fluctuation feature in the power mapping feature vector are weighted to extract the power disturbance and instability imbalance feature, which is used to characterize the degree of stability destruction of the power signal of the steam turbine generator set during the change process. When the unit experiences faults such as load disturbance, regulation lag, or uneven energy coupling, this feature increases significantly. The power stability feature and power imbalance feature in the power mapping feature vector are weighted. In this weighting process, the power stability feature is transformed using the reciprocal suppression mapping function f(x)=1 / (1+x), i.e., 1 / (1+power stability feature), to extract the power drift feature, which is used to characterize the degree of time-series drift of the turbine generator power signal during the time evolution process. When the unit experiences abnormalities such as energy feedback lag, control link mismatch, or uneven load response, this feature increases significantly. The power disturbance and imbalance feature, power drift feature, and power complexity feature are then concatenated into a dimensionality-reduced feature vector. In the output mapping layer of the power mapping neural network, based on the dimensionality reduction feature vector of the steam turbine generator set, the power anomaly mapping feature set of the steam turbine generator set is output. Specifically, the power disturbance and instability features, power drift features, and power complexity features in the dimensionality reduction feature vector are activated by the Sigmoid function to obtain power disturbance and instability feature values, power drift feature values, and power complexity feature values ​​with results between 0 and 1.

[0039] The pre-training steps of the power mapping neural network are as follows: A labeled dataset is obtained, which consists of power signal samples from several steam turbine generator units under different operating conditions. Each sample includes power time series data over multiple consecutive operating cycles and its corresponding operating status label. The operating status label is formed by operation and maintenance engineering experts based on unit operation logs, fault records, and on-site detection results. Each sample in the labeled dataset contains a power amplitude change curve and corresponding abnormal state indicators, such as stable operation, slight disturbance, and nonlinear disturbance. The original samples are preprocessed, including time series alignment, denoising, normalization, and sample balancing. The dataset is then divided into training, validation, and test sets proportionally, for example, 80% of the data is used for training, 10% for validation, and 10% for testing, to ensure the generalization ability of the model.

[0040] During the training phase, the power signal sequences of each sample are input into the input layer of the power mapping neural network. The power response change features at different time scales are extracted through the convolutional feature extraction layer, and then feature reduction and correlation enhancement are performed through the dimensionality reduction and fusion layer. During training, the Adam optimizer is used to iteratively optimize the network weights, and the loss function adopts mean squared error (MSE) or cross-entropy loss to minimize the deviation between the predicted and labeled power feature values. At the same time, an early stopping mechanism and regularization constraints are introduced to prevent overfitting. In order to balance model performance and training cost, the network size is controlled within a lightweight range (e.g., the number of convolutional layers ≤ 3 layers, feature dimension ≤ 128 dimensions) to ensure that model training and optimization can be completed on a conventional industrial server or GPU workstation, avoiding excessive hardware costs.

[0041] By monitoring the convergence trend of the training process in real time using the validation set, the hyperparameters such as learning rate and batch size are dynamically adjusted until the power anomaly mapping features output by the network are stably converged and the error rate of the test set is lower than the preset threshold. After training is completed, the optimal network parameters and structure configuration are saved to the database for rapid loading and real-time inference in the online monitoring system, so as to realize the feature mapping and anomaly identification of power signals.

[0042] In this implementation scheme, a power mapping neural network is constructed to achieve multi-layer feature learning and intelligent fault mapping of the power signal of the steam turbine generator set. By standardizing, denoising, and smoothing the power signal at the input layer, random interference and sampling errors are effectively filtered out. The convolutional feature extraction layer can automatically extract multi-scale features such as power change rate, power fluctuation characteristics, and response trends from the time series, achieving accurate capture of the dynamic laws of the power signal. Secondly, the dimensionality reduction and fusion layer forms a comprehensive characterization of power fluctuation stability, energy feedback delay, and system nonlinear disturbance by weighted combination of power disturbance and instability features, power drift features, and power complexity features. In addition, sample entropy weight and reciprocal suppression mapping function are introduced in this process, so that the feature weighting result can adaptively reflect the differences in information contribution between features, thereby avoiding judgment bias caused by feature redundancy. Finally, this step can identify potential abnormal trends in the early stage of power output fluctuation, thereby effectively revealing the disturbance source and regulation imbalance characteristics in the energy transfer chain, and thus achieving high-precision identification and dynamic mapping of power signal faults.

[0043] Specifically, the steps to obtain the energy-vibration coordinated fault characteristic value of the steam turbine generator set are as follows: Read the vibration signal data and power signal data of the steam turbine generator set, and perform segmentation processing (that is, use the least common multiple of the sampling frequencies of the vibration signal data and power signal data as the segmentation time step to obtain the corresponding values ​​of several time points, that is, the time interval between two adjacent time points is its least common multiple time step), obtain the vibration amplitude value and power amplitude value of the steam turbine generator set at each time point, and perform correlation analysis (that is, normalize the vibration amplitude value and power amplitude value of the steam turbine generator set at each time point, and use the Pearson correlation coefficient to extract the signal correlation coefficient value of the vibration amplitude value and power amplitude value of each time point after normalization, and it should be noted that the signal correlation coefficient value is taken as the absolute value), and obtain the signal coordinated adjustment coefficient of the steam turbine generator set; Read the power fault mapping characteristic value and vibration fault mapping characteristic value of the steam turbine generator set, and combine them with the signal correlation coefficient value to analyze the energy-vibration coordinated fault characteristic value of the steam turbine generator set (used to characterize the probability of fault occurrence mapped by the vibration signal and power signal of the steam turbine generator set).

[0044] The specific formula for calculating the characteristic value of the energy-vibration coordinated fault of the steam turbine generator set is as follows: ;in, The characteristic value of the energy-vibration coordinated fault of the steam turbine generator set. For vibration fault mapping characteristic values ​​of steam turbine generator sets, The vibration fault adjustment coefficients are stored in the database. For power fault mapping characteristic values ​​of steam turbine generator sets, The power fault adjustment coefficients are stored in the database. This refers to the signal coordination regulation coefficient of the steam turbine generator set.

[0045] It needs to be explained that the vibration fault adjustment coefficients stored in the database Power fault regulation coefficient The acquisition steps are as follows: Obtain the vibration fault mapping feature values ​​and power fault mapping feature values ​​for several historical cycles. Extract the mean values ​​of the vibration fault mapping features and power fault mapping features respectively, and sum them to obtain the fault sum value. Ratio the mean values ​​of the vibration fault mapping features and power fault mapping features to the fault sum value, and use the corresponding results as the vibration fault adjustment coefficient. Power fault regulation coefficient .

[0046] In this implementation scheme, the vibration and power signals of the steam turbine generator set are fused to achieve dynamic coupling analysis of the energy and mechanical domains. The least common multiple of the sampling frequencies of the two types of signals is used as the time step to achieve data alignment and synchronization on the time scale, thereby ensuring that the vibration amplitude and power amplitude values ​​are comparable under the same time reference. The signal coordination adjustment coefficient is extracted using the Pearson correlation coefficient, which can quantitatively reflect the correlation coupling strength between the power signal and the vibration signal. Subsequently, the vibration fault mapping feature value and the power fault mapping feature value are fused according to their respective adjustment coefficients to obtain the energy-vibration coordinated fault feature value. The fusion result can comprehensively reflect the coupling deviation between the power-side disturbance and the vibration-side response, thereby accurately revealing the coordinated anomaly between energy fluctuation and mechanical response, and thus achieving comprehensive identification of multi-domain faults and improving the reliability of early warning.

[0047] Example 2 according to Figure 4 As shown, this embodiment also provides a state monitoring and early warning system for turbine generator set vibration power signal fusion, used to implement the above-described state monitoring and early warning method for turbine generator set vibration power signal fusion, including: Data acquisition module 1 is used to collect vibration signal data and power signal data of steam turbine generator set within a set period. The vibration signal data includes the vibration amplitude value at each vibration time point; the power signal data includes the power amplitude value at each power time point. Vibration analysis module 2 is used to process the collected vibration signal data, extract and analyze the vibration fault mapping feature values ​​of the steam turbine generator set; The fusion analysis module 3 is used to process the collected power signal data through a pre-trained power mapping neural network, extract and analyze the power fault mapping feature value, and combine the power fault mapping feature value with the vibration fault mapping feature value to obtain the energy-vibration synergistic fault feature value. The early warning processing module 4 is used to perform fault early warning processing on the steam turbine generator set through the energy vibration coordinated fault characteristic value.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A condition monitoring and early warning method for turbine generator set vibration power signal fusion, characterized in that, include: Vibration signal data and power signal data of the steam turbine generator set are collected within a set period. The vibration signal data includes the vibration amplitude value at each vibration time point, and the power signal data includes the power amplitude value at each power time point. The collected vibration signal data is processed to extract and analyze the vibration fault mapping feature values ​​of the steam turbine generator set. The collected power signal data is processed by a pre-trained power mapping neural network to extract and analyze power fault mapping feature values. The power fault mapping feature values ​​are then combined with the vibration fault mapping feature values ​​to obtain energy-vibration synergistic fault feature values. Fault early warning processing of steam turbine generator sets is carried out by using energy vibration coordinated fault characteristic values.

2. The method for condition monitoring and early warning of a steam turbine generator set vibration power signal fusion according to claim 1, characterized in that, The specific steps for processing the collected vibration signal data and extracting and analyzing the vibration fault mapping feature values ​​of the steam turbine generator set are as follows: The vibration signal data is divided into time periods to obtain vibration sub-signal data for several time periods. The vibration sub-signal data includes the vibration response value at each vibration time point. The vibration sub-signal data for each time period are deconstructed to obtain the energy-frequency mapping feature set corresponding to each time period. The energy-frequency mapping feature set includes vibration impulse energy feature value and frequency-energy co-mapping feature value. Based on the energy-frequency mapping feature set of each time period, the vibration fault mapping feature value of the steam turbine generator set is obtained by analysis.

3. The method for state monitoring and early warning of vibration power signal fusion of a steam turbine generator set according to claim 2, characterized in that, The specific steps for dividing the vibration signal data into time periods to obtain vibration sub-signal data for several time periods are as follows: Based on the vibration amplitude value at each vibration time point, calculate the vibration energy change values ​​of several sets of adjacent vibration time points; Based on the vibration energy change values ​​of each group of adjacent vibration time points, the vibration sub-signal data for each time period are obtained.

4. A state monitoring and early warning method for vibration power signal fusion of a steam turbine generator set according to claim 2, characterized in that, The specific steps for deconstructing the vibrator signal data for each time period to obtain the corresponding energy-frequency mapping feature set are as follows: The vibration sub-signal data for each time period are processed in the time domain to obtain the vibration impulse energy characteristic value for that time period; Frequency domain extraction processing is performed on the vibrator signal data for each time period to obtain the frequency-energy co-encoding characteristic value for that time period.

5. The method for condition monitoring and early warning of a steam turbine generator set vibration power signal fusion according to claim 1, characterized in that, The pre-trained power mapping neural network includes an input layer, a convolutional feature extraction layer, a dimensionality reduction and fusion layer, and an output mapping layer.

6. The method for state monitoring and early warning of vibration power signal fusion of a steam turbine generator set according to claim 5, characterized in that, The specific steps for analyzing and obtaining the power fault mapping characteristic values ​​are as follows: The power signal data is input into a pre-trained power mapping neural network, and the power anomaly mapping feature set is obtained by analysis. The power anomaly mapping feature set includes power disturbance and instability feature values, power drift feature values, and power complexity feature values. Based on the power anomaly mapping feature set, the power fault mapping feature value of the steam turbine generator set is obtained through analysis.

7. The method for condition monitoring and early warning of vibration power signal fusion of a steam turbine generator set according to claim 6, characterized in that, The specific steps for inputting the power signal data into a pre-trained power mapping neural network and analyzing it to obtain the power anomaly mapping feature set are as follows: In the input layer, the power signal data is received and preprocessed; In the convolutional feature extraction layer, a power mapping feature vector is extracted based on the preprocessed power signal data; In the dimensionality reduction fusion layer, the power mapping feature vector is correlated to obtain the dimensionality reduction feature vector; In the output mapping layer, a power anomaly mapping feature set is output based on the dimensionality-reduced feature vector.

8. The method for condition monitoring and early warning of a steam turbine generator set vibration power signal fusion according to claim 1, characterized in that, The specific steps for fusing the power fault mapping feature values ​​with the vibration fault mapping feature values ​​to obtain the energy-vibration coordinated fault feature values ​​are as follows: Read the vibration signal data and power signal data, divide the vibration amplitude value and power amplitude value corresponding to each time point, perform correlation analysis on the two, and obtain the signal coordinated regulation coefficient; The power fault mapping feature value and vibration fault mapping feature value are read, and combined with the signal coordinated adjustment coefficient, the energy-vibration coordinated fault feature value is obtained by analysis; The specific formula for calculating the characteristic value of the energy-vibration coordinated fault of the steam turbine generator set is as follows: ； in, The characteristic value of the energy-vibration coordinated fault of the steam turbine generator set. For vibration fault mapping characteristic values ​​of steam turbine generator sets, The vibration fault adjustment coefficients are stored in the database. For power fault mapping characteristic values ​​of steam turbine generator sets, The power fault adjustment coefficients are stored in the database. This refers to the signal coordination regulation coefficient of the steam turbine generator set.

9. The method for state monitoring and early warning of vibration power signal fusion of a steam turbine generator set according to claim 1, characterized in that, The specific steps for fault early warning processing of steam turbine generator sets based on energy-vibration coordinated fault characteristic values ​​are as follows: The energy-vibration coordinated fault characteristic values ​​of the steam turbine generator set are compared with the preset energy-vibration coordinated fault characteristic thresholds. Fault warnings are issued for steam turbine generator sets based on the comparison and processing results.

10. A condition monitoring and early warning system for fusion of vibration power signals of a steam turbine generator set, characterized in that, A condition monitoring and early warning method for implementing the vibration power signal fusion of a steam turbine generator set as described in any one of claims 1-9 includes: The data acquisition module is used to collect vibration signal data and power signal data of the steam turbine generator set within a set period. The vibration signal data includes the vibration amplitude value at each vibration time point; the power signal data includes the power amplitude value at each power time point. The vibration analysis module is used to process the collected vibration signal data, extract and analyze the vibration fault mapping feature values ​​of the steam turbine generator set; The fusion analysis module is used to process the collected power signal data through a pre-trained power mapping neural network, extract and analyze the power fault mapping feature values, and combine the power fault mapping feature values ​​with the vibration fault mapping feature values ​​to obtain the energy-vibration synergistic fault feature values. The early warning processing module is used to perform fault early warning processing on steam turbine generator sets based on the energy vibration coordinated fault characteristic values.

Images