Compressor surge judgment critical value determination method and system based on multi-source data fusion

Abstract

The present application belongs to the technical field of compressor performance prediction, and specifically proposes a compressor surge discrimination critical value determination method and system based on multi-source data fusion. It includes building and running a compressor surge test system, calculating the first surge critical value; obtaining multi-source historical data related to compressor surge, selecting characteristic variables to construct a historical input vector, fitting a multiple regression model using the historical input vector to obtain a discrimination model; training a machine learning model to obtain a trained machine learning model; constructing a real-time input vector based on real-time data collected from the compressor to be monitored, and then obtaining the second and third surge critical values; and fusing the first, second and third surge critical values to obtain the predicted surge critical value of the compressor to be monitored. The present application combines high-precision dynamic testing and historical data regression analysis, and ultimately quantitatively determines the actual value of the surge critical value through multi-parameter coupling measurement and data mining.

Description

Technical Field

[0001] This invention belongs to the field of compressor performance prediction technology, and in particular relates to a method and system for determining the critical value of compressor surge discrimination based on multi-source data fusion. Background Technology

[0002] Axial flow compressors are the core components of compressed air energy storage power stations, and their performance directly determines the efficiency and stability of the entire power unit. During compressor operation, when the operating conditions deviate from the design point and enter the low-flow range, two main instability phenomena may occur in the gas flow: rotational stall and surge. Both phenomena can lead to a significant decrease in compressor performance, and in severe cases, can cause mechanical vibration, bearing damage, or even complete unit failure, making them one of the key factors affecting the safe operation of compressors.

[0003] Rotating stall manifests as a low-energy air mass formed by airflow detachment from a localized cascade, propagating circumferentially, with its airflow pulsation direction predominantly circumferential. Surge, on the other hand, is a longitudinal oscillation phenomenon characterized by periodic backward flow throughout the entire compressor, exhibiting significant axial flow fluctuations across the entire flow path. Although rotating stall is often considered a precursor to surge, their mechanisms, triggering conditions, and flow field symmetry differ significantly. The former pertains to the aerodynamic stability of the compressor's internal cascades, while the latter is a coupling instability phenomenon between the compressor and the external piping system. A dimensionless criterion is used to determine the cause. This characterizes the dynamic behavior of this complex system. The criterion indicates that when... At this time, the system mainly exhibits rotational stall; when At this time, the compressor-pipeline system will enter a periodic surge state. Critical value. It reflects the critical boundary of the system transitioning from local instability to overall instability and is a key indicator for determining the compressor's stability margin.

[0004] existing The determination of parameters generally relies on experience or single experiments, lacking a unified measurement standard and data processing procedure. The main problems are: single experimental measurement dimension, insufficient capture of dynamic characteristics, underutilization of historical data, large uncertainty of model parameters, and lack of standardized criteria. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this invention provides a method and system for determining the critical value of compressor surge by multi-source data fusion. By combining high-precision dynamic testing and historical data regression analysis, and through multi-parameter coupled measurement and data mining, the aerodynamic stability boundary of the compressor-pipeline system is identified, and the actual value of the surge critical value is finally quantitatively determined.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] The first aspect of this invention provides a method for determining the critical value of compressor surge detection based on multi-source data fusion.

[0008] A method for determining the critical value for compressor surge detection based on multi-source data fusion includes the following steps:

[0009] A compressor surge test system was built and run to identify the surge initiation point during the test. Based on the test data at the surge initiation point, the first surge critical value was calculated.

[0010] Multi-source historical data related to compressor surge are obtained, feature variables are selected to construct historical input vectors, and a multivariate regression model is fitted using the historical input vectors to obtain a discriminant model; at the same time, a machine learning model is trained using the historical input vectors to obtain a trained machine learning model;

[0011] A real-time input vector is constructed based on the real-time collected data of the compressor to be monitored. The real-time input vector is then input into the discrimination model and the trained machine learning model to obtain the second surge threshold and the third surge threshold.

[0012] The first surge threshold, the second surge threshold, and the third surge threshold are fused together to obtain the predicted surge threshold of the compressor to be monitored.

[0013] The second aspect of the present invention provides a compressor surge discrimination critical value determination system based on multi-source data fusion.

[0014] A multi-source data fusion-based compressor surge discrimination threshold determination system includes:

[0015] The first critical value prediction module is configured to: build and run a compressor surge test system, identify the surge start point during the test, and calculate the first surge critical value based on the test data at the surge start point.

[0016] The model determination module is configured to: acquire multi-source historical data related to compressor surge, select feature variables to construct historical input vectors, fit a multivariate regression model using the historical input vectors to obtain a discriminant model; and simultaneously train a machine learning model using the historical input vectors to obtain a trained machine learning model.

[0017] The second critical value prediction module is configured to: construct a real-time input vector based on the real-time collected data of the compressor to be monitored, and input the real-time input vector into the discrimination model and the trained machine learning model to obtain the second surge critical value and the third surge critical value;

[0018] The fusion module is configured to fuse the first surge threshold, the second surge threshold, and the third surge threshold to obtain the predicted surge threshold of the compressor to be monitored.

[0019] The above one or more technical solutions have the following beneficial effects:

[0020] This invention provides a method and system for determining the critical value of compressor surge based on multi-source data fusion. By organically combining experimental physical modeling, regression analysis of historical operating data, and intelligent algorithm discrimination, it brings significant technological progress. Compared with traditional methods that rely on empirical critical curves, pressure ratio-flow fitting, or single experiments, this invention achieves more accurate and reliable quantitative identification and predictable determination of the critical stability boundary of the compressor system. It is a highly reliable solution with both high-precision discrimination capability and early warning effectiveness, providing a solid technical guarantee for the safe, stable, and efficient operation of axial compressors in key facilities such as compressed air energy storage power stations.

[0021] This invention integrates historical operating data to construct a complete technical chain from rigorous theoretical derivation to efficient engineering implementation, thereby achieving higher physical consistency and model scalability. This method not only ensures the repeatability of experimental conclusions but also overcomes the limitations of traditional static thresholds, enabling online adaptive updates and real-time judgment based on the actual operating status of the system, thus improving early warning and protection capabilities under dynamic operating conditions.

[0022] This invention exhibits excellent robustness. Even when faced with complex situations such as unavoidable measurement noise, data errors, external interference, and even partial data loss in the field, its core algorithm can still maintain a low false judgment rate.

[0023] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0025] Figure 1 This is a flowchart of the method in Example 1. Detailed Implementation

[0026] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0027] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0028] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0029] Example 1

[0030] This embodiment discloses a method for determining the critical value of compressor surge based on multi-source data fusion. It establishes a unified methodological system combining high-precision dynamic testing and historical data regression analysis. Through multi-parameter coupled measurement and data mining, the system identifies the aerodynamic stability boundary of the compressor-pipeline system, thereby quantitatively determining the surge critical value. The actual value of .

[0031] like Figure 1 As shown, the method for determining the critical value of compressor surge discrimination based on multi-source data fusion includes the following steps:

[0032] A compressor surge test system was built and run to identify the surge initiation point during the test. Based on the test data at the surge initiation point, the first surge critical value was calculated.

[0033] Multi-source historical data related to compressor surge are obtained, feature variables are selected to construct historical input vectors, and a multivariate regression model is fitted using the historical input vectors to obtain a discriminant model; at the same time, a machine learning model is trained using the historical input vectors to obtain a trained machine learning model;

[0034] A real-time input vector is constructed based on the real-time collected data of the compressor to be monitored. The real-time input vector is then input into the discrimination model and the trained machine learning model to obtain the second surge threshold and the third surge threshold.

[0035] The first surge threshold, the second surge threshold, and the third surge threshold are fused together to obtain the predicted surge threshold of the compressor to be monitored.

[0036] This embodiment establishes a unified and quantitative aerodynamic instability discrimination criterion by multi-dimensional acquisition, dynamic analysis, and data mining of key aerodynamic and structural parameters during compressor operation, achieving accurate identification of the critical state from rotating stall to surge. Through a systematic process of physical modeling, experimental measurement, data regression, and criticality discrimination, it solves the problems found in existing technologies. The problems include reliance on experience for value selection, inconsistent models, and poor repeatability.

[0037] This invention introduces a dynamic measurement mechanism. By deploying high-frequency pressure sensors and flow meters at the rotor blade tip, outlet, and pipeline, multi-dimensional acquisition of airflow pulsations is achieved, ultimately calculating the first surge critical value. Combining historical data regression analysis, information such as rotational speed, pressure, flow rate, and temperature recorded over long-term operation is used to determine the critical interval for system instability through multivariate nonlinear regression, thus obtaining the second and third surge critical values. A unified discrimination algorithm is established to automatically solve the system's problems through coupled analysis of experimental data and theoretical models. .

[0038] The technical solution of this embodiment will be explained in detail below.

[0039] (a) Determination of the first surge threshold value.

[0040] This embodiment establishes a compressor surge test system, which mainly includes an axial compressor test body, a variable capacity pipeline system (including connecting pipelines and gas storage chambers), a multi-channel high-frequency dynamic measurement system, a data synchronous acquisition system, a signal processing system, and a control and regulation subsystem (including throttle valves, speed regulating motors, and data interfaces).

[0041] This system can gradually approach the aerodynamic instability region of the compressor while ensuring safety, thereby systematically measuring the entire process of transition from rotating stall to surge.

[0042] (1) Furthermore, the construction process of the compressor surge test system includes:

[0043] A multi-stage adjustable stationary vane axial flow compressor was selected as the main test unit.

[0044] A pipeline is installed between the multi-stage adjustable stationary vane axial flow compressor and the air receiver, with a throttle valve at the end of the pipeline.

[0045] Multiple sensors are deployed: a pressure sensor is installed at the compressor rotor blade tip outlet, multiple pressure sampling points are arranged in a circumferential distribution at the compressor outlet section, a static pressure sensor is installed in the middle section of the pipeline, pressure and temperature sensors are installed inside the gas storage tank, and flow meters and thermometers are installed in the intake pipeline.

[0046] More detailed:

[0047] 1. Axial compressor body:

[0048] A multi-stage adjustable stationary vane axial flow compressor was selected as the research object. The speed can be precisely adjusted by a variable speed motor with an accuracy of ±0.2%.

[0049] 2. Variable capacity piping system (including connecting pipes and gas storage chambers):

[0050] The compressor outlet is connected to a length of adjustable-diameter pipe with an inner diameter matching the annular flow area of ​​the compressor outlet. The end of the pipe is connected to a variable-volume gas receiver (gas receiver volume...). exist (Adjustable within the range); a throttle valve is installed at the end of the system to control the total flow and regulate the back pressure; temperature and static pressure detection ports are installed in the pipeline for easy sound velocity detection. Calculation of flow rate.

[0051] 3. Multi-channel high-frequency dynamic measurement system:

[0052] A miniature high-frequency pressure sensor is installed at the compressor rotor blade tip outlet to measure the formation and propagation characteristics of local stall cells in the blade cascade. Eight to sixteen circumferentially distributed pressure sampling points are arranged at the compressor outlet cross-section to obtain circumferential asymmetric flow information. A high-sensitivity static pressure sensor is installed in the middle section of the connecting pipe to record pressure fluctuations along the pipe. An integrated pressure and temperature sensor is installed inside the gas receiver to obtain the pressure change rate and gas temperature within the system cavity, thereby calculating the real-time speed of sound. A hot-film flow meter and thermometer are installed in the inlet pipe to calculate the mass flow rate in real time. and flow coefficient The rotational speed is measured in real time by a photoelectric tachometer with a sampling frequency of not less than 1 kHz (kilohertz).

[0053] 4. Data Synchronization Acquisition System:

[0054] All sensor signals are uniformly connected to a high-speed data acquisition system (DAQ), with a sampling frequency set at 20 kHz to ensure accurate capture of high-frequency pressure fluctuations and rotational stall propagation. The data acquisition system employs a multi-channel synchronous sampling mode, with a time synchronization error of less than 1 μs. The system is equipped with an anti-aliasing filter (cutoff frequency 10 kHz) and a differential input interface to ensure signal stability. Pressure fluctuations at the blade tip are recorded in each experiment. Average static pressure at compressor outlet Pressure fluctuations in the middle section of the pipeline Gas storage tank pressure With temperature Rotation speed With traffic , Indicates time. After data is stored in real time, it is uniformly calibrated using synchronized timestamps.

[0055] 5. Signal processing system:

[0056] The signal is denoised by wavelet and low-pass filtered to remove electromagnetic noise and high-frequency interference; amplitude normalization and phase alignment are performed on the signals at each measuring point; and the root mean square value of pressure pulsation is calculated. , main frequency The coherence coefficient CC is used to determine the characteristics of the flow pattern.

[0057] 6. Control and regulation subsystem (including throttle valve, speed control motor, and data interface):

[0058] Using a high-precision electric throttle valve as the main actuator, the compressor back pressure and flow rate are changed by rapidly adjusting the opening degree, and the operating point is precisely controlled to approach the stable boundary; the variable frequency speed control motor serves as the power source to precisely control the compressor speed; the central controller with an integrated data interface collects sensor data in real time, calculates B parameters, and runs the discrimination algorithm.

[0059] (2) Identify the first surge threshold using experiments.

[0060] The experimental and computational discrimination process of the above system is as follows:

[0061] 1. Experimental procedure.

[0062] a. Stable operation phase: Run the compressor at the design flow rate to confirm that all sensor signals are stable and the system temperature and pressure reach a steady state.

[0063] b. Approaching the unstable region stage: Gradually reduce the throttle valve opening to make the flow coefficient... Gradually decrease the pressure. Monitor the pressure fluctuation spectrum at the blade tip and outlet. When obvious rotational stall characteristics (low-amplitude, high-frequency fluctuations) appear, record the current system pressure. value.

[0064] c. Entering the instability zone: Continue to reduce the flow rate. Observe periodic large-amplitude pressure and flow rate declines (low-frequency, high-amplitude fluctuations), indicating that the system has entered the surge zone. At this time, the high-frequency sampling mode of the data acquisition system is automatically triggered to record the entire process.

[0065] d. Safety backoff phase: When the system detects that the pressure pulsation amplitude exceeds the set threshold (such as ±5% pressure ratio change), it automatically increases the throttle valve opening to restore the system to stability and prevent mechanical damage.

[0066] e. Repeatability testing: under different gas reservoir volumes Equivalent length of the outlet pipeline Rotation speed Repeated experiments under different conditions form a multidimensional structure. - Response database.

[0067] 2. Calculation and discrimination parameter extraction.

[0068] a. Sound velocity calculation: calculated in real time based on the temperature and pressure of the gas reservoir:

[0069]

[0070] In the formula, Specific heat ratio; It is the gas constant; This indicates the temperature of the gas storage tank.

[0071] b. Discriminant parameters Real-time calculation:

[0072]

[0073] In the formula, ; This indicates the tip tangential velocity or circumferential velocity of the compressor blades; Indicates the speed of sound; Indicates the volume of the gas storage tank; This indicates the cross-sectional area of ​​the compressor outlet pipe; This indicates the equivalent length of the compressor outlet pipe; Indicates the characteristic diameter of the compressor; Indicates rotational speed. This indicates that the parameter is an instantaneous value that changes over time.

[0074] c. Instability marker extraction:

[0075] By analyzing the energy spectrum of the signal at each measurement point using Short Time Fourier Transform (STFT), when the energy proportion of the low-frequency component exceeds 40%, and it is related to the system's Helmholtz frequency... When approaching, mark it as the surge initiation point; at this time... The value is denoted as .

[0076] All data, after being synchronized with the timeline, is input into the data analysis module for subsequent historical data fusion and... Identification.

[0077] (ii) Determination of the hybrid model based on multiple regression model and machine learning model.

[0078] To improve the accuracy of the discrimination, in addition to the compressor instability characteristics directly measured by the test device, this embodiment of the invention also establishes a discrimination criterion for determining surge criticality by fusing historical operating data and mathematical modeling. The data-driven analysis model combines experimental measurement, theoretical derivation, and machine learning regression techniques. It can automatically calculate the corresponding critical discrimination values ​​under different compressor structures and operating conditions, achieving applicability across different compressor types and parameters.

[0079] Through training and statistical analysis of a large amount of historical data, the embodiments of this invention have established... The mapping relationship between the compressor and multiple sets of aerodynamic and geometric parameters provides a quantitative basis for evaluating compressor stability.

[0080] (1) Historical data structure and feature selection.

[0081] a. Historical data sources:

[0082] Historical data sources include axial compressor bench test data, industrial field operation records (including flow rate, speed, pressure, temperature, sound pressure spectrum, etc.), and instability boundary data under different geometric parameters (such as blade shape, number of stages, and outlet area).

[0083] b. Definition of characteristic parameters, combined with theoretical formulas:

[0084]

[0085] Based on the system's dynamic response characteristics, the following feature variables are selected to form the input vector:

[0086]

[0087] In the formula, This indicates the tip tangential velocity of the compressor blades; Pressure ratio; For flow coefficient; For adiabatic efficiency; It is the Reynolds number; The Strouhal number represents the unsteady characteristics of pressure fluctuations.

[0088] c. Output target: The output variable is defined as follows:

[0089]

[0090] That is, the critical surge criterion value measured under specific working conditions and structural conditions.

[0091] It is the target value of the machine learning model, which is a quantity obtained directly or indirectly through experiments. The method of obtaining it is as follows:

[0092] 1) Perform a single instability test. Fix a set of geometric parameters (e.g., ... The system is controlled by the compressor characteristic curve and the rotational speed n. Starting at a certain stable point, the valve is gradually closed (reducing the flow coefficient φ), and the system moves towards the surge boundary along the compressor characteristic curve. When the system first experiences surge (i.e., a periodic large-scale pressure and flow rate reversal), the high-frequency data acquisition system is triggered.

[0093] 2) Determine the critical point and extract Analyzing the experimentally collected data, the inflection point from stability to surge is located; this point is the critical point. The instantaneous readings of all sensors are extracted from the moment just before this critical point occurs, and these instantaneous values ​​are substituted into the formula for calculating B(t). The calculated value of B(t) is defined as the critical parameter under the experimental conditions. .

[0094] (2) Mathematical modeling and regression analysis.

[0095] To establish The mapping relationship between multiple parameters is determined using a hybrid approach that combines a multi-level regression model with a machine learning nonlinear fitting algorithm.

[0096] The multiple regression model has the following formula:

[0097]

[0098] In the formula, For constant terms; These are the regression coefficients; These are higher-order nonlinear function terms used to characterize coupling effects; For the first One feature parameter; For the first One feature parameter; This is the surge threshold.

[0099] By fitting experimental and historical data using the least squares method, a basic statistical relationship is obtained, forming a first-order discriminant model.

[0100] (3) Machine learning model establishment.

[0101] To improve the prediction accuracy in nonlinear intervals, this embodiment of the invention introduces a multilayer perceptron neural network (MLP) machine learning algorithm based on the statistical model, by processing the input feature vector... After standardization, the model learns from historical samples. The nonlinear variation law.

[0102] The training objective function is:

[0103]

[0104] In the formula, Indicates the number of samples; and This represents the model weights and biases; This represents the regularization coefficient, used to prevent overfitting. Indicates the first One sample; Indicates the first The true surge threshold value for each sample; Indicates the first The surge threshold value output by the machine learning model for each sample.

[0105] (4) Model validation and cross-calibration.

[0106] use Cross-validation is used to evaluate the generalization ability of the models. The coefficient of determination is calculated for each model. Root mean square error (RMSE) and relative deviation (MAPE). If the following conditions are met:

[0107]

[0108] The decision model can then be used predict.

[0109] (iii) Determination of the second surge threshold and the third surge threshold.

[0110] To ensure the accuracy and universality of the judgment, this embodiment designs a data fusion algorithm, which realizes an adaptive, data-driven method for identifying surge critical discrimination criteria based on theoretical criteria.

[0111] a. Input stage.

[0112] The real-time acquired flow rate, rotational speed, pressure, temperature, and sound pressure signals are filtered and feature extracted to form a real-time input vector. .

[0113] b. Judgment calculation stage.

[0114] Real-time input vectors The trained hybrid regression model calculates and outputs the second and third surge thresholds. Finally, combined with the previously obtained first surge threshold, the predicted surge threshold is obtained. Specifically:

[0115] 1) By fusing multi-source data, experimental data, historical operational data, and simulation results are integrated, using a weighted average and Bayesian update method:

[0116]

[0117] In the formula, This indicates the predicted surge threshold. Indicates weight, This indicates the first surge threshold, the second surge threshold, or the third surge threshold. =1,2 or 3; This represents the first surge threshold value. This indicates the second surge threshold. This indicates the third surge threshold value; The weight representing the first surge threshold. The weight representing the second surge threshold. The weight representing the third surge threshold. It is determined based on data reliability and time relevance.

[0118] For real-time running data, the weights can decay over time:

[0119]

[0120] The time decay coefficient, To generate the i-th surge threshold The calculation time of the data used.

[0121] 2) Through the online model correction mechanism, during the long-term operation of the compressor, the system characteristics drift due to component wear or temperature changes. The predicted surge critical value is dynamically corrected through the adaptive learning module.

[0122]

[0123] In the formula, For loss function, For learning rate, express Model parameters at time 10:00 express The model parameters at each time step. This mechanism can maintain... The discrimination accuracy has remained stable over a long period.

[0124] c. Logic discrimination stage.

[0125] The discrimination parameters of the compressor under monitoring are calculated in real time. Based on the predicted surge critical value of the compressor under monitoring, and combined with the set empirical safety margin, the stability of the compressor under monitoring is judged.

[0126] If the discrimination parameters are currently calculated in real time satisfy:

[0127]

[0128] In the formula, With an empirical safety margin (typical value 0.05–0.1), the system determines that it is about to enter the surge zone and issues a warning signal. This indicates the predicted surge threshold value for the compressor to be monitored; This indicates the discrimination parameters for the compressor to be monitored.

[0129] d. Output stage.

[0130] The model output includes: real-time predictions Compressor stability class (safe, critical, surge) and recommended adjustment parameters (such as throttle opening or speed correction value).

[0131] In a real-time operating environment, this invention continuously acquires multi-point pressure, flow, temperature, and speed signals of the axial compressor through a high-speed sampling system. The signal processing and real-time discrimination algorithm implementation process is as follows:

[0132] (1) Signal synchronization and time calibration.

[0133] The signals from each measurement point are time-calibrated through a unified clock synchronization module to ensure that all signal sampling points correspond under the same time base. The system time synchronization error is controlled within... Within microseconds.

[0134] (2) Filtering algorithm design.

[0135] To remove mechanical vibration, electromagnetic interference, and high-frequency noise, this invention employs a composite filtering strategy:

[0136]

[0137] In the formula, Indicates wavelet denoising; This indicates a low-pass Butterworth filter. This represents the filtered signal; This represents the signal before filtering.

[0138] Low-pass cutoff frequency ,in The sampling frequency retains both low-frequency (0–100Hz) and mid-frequency (100–1000Hz) components of the signal to capture rotational stall and surge precursors.

[0139] (3) Characteristic signal decomposition.

[0140] Perform Empirical Mode Decomposition (EMD) or Variational Mode Decomposition (VMD) on the preprocessed signal to obtain several internal mode functions. .

[0141] Each mode corresponds to different physical characteristics: high frequency mode: tip vortex structure and local stall cell motion; mid frequency mode: rotational stall fluctuation; low frequency mode: overall system surge characteristics.

[0142] (4) Feature parameter extraction.

[0143] Calculate the root mean square value based on the filtered and decomposed signal. , main frequency Coherence coefficient (CC), instantaneous frequency, and envelope characteristics.

[0144] (5) Real-time calculation of discrimination criterion parameters.

[0145] In signal processing, the discrimination parameters are calculated in real time:

[0146]

[0147] This parameter is updated every millisecond, and... By making comparisons, an immediate stability assessment can be formed.

[0148] In addition, its first-order time derivative is calculated:

[0149]

[0150] in, Indicates a time interval.

[0151] like Continuous negative growth and , The preset threshold indicates that the system is being judged to be trending towards instability.

[0152] (6) Classification and discrimination logic.

[0153] a. Feature fusion input layer, which combines the extracted parameters mentioned above. Construct the input vector:

[0154] in, This indicates the signal-to-noise ratio.

[0155] b. Intelligently identify network structure.

[0156] This invention employs a Lightweight Neural Network (Light-MLP) model to achieve real-time classification. The network structure is as follows:

[0157]

[0158] in, Represents the eigenvector; This indicates that this fully connected layer has 16 neurons; Indicates the activation function; This indicates that the second fully connected layer contains 8 neurons; This indicates that the normalized exponential function has 3 output neurons.

[0159] The output category is: Stable region; Critical region; Surge zone.

[0160] c. Judgment and alarm logic.

[0161] Model output confidence , Indicates the first The confidence level of the output category.

[0162] And set the discrimination rules:

[0163] like The system immediately triggered a surge alarm; The output category is Confidence level of surge region.

[0164] like It issues a "near critical" warning; The output category is Confidence level of the critical region.

[0165] like The system continues to operate normally; The output category is Confidence level of the stable region.

[0166] (7) Adaptive threshold adjustment and feedback.

[0167] Since the noise characteristics of different compressor structures and operating conditions vary, this embodiment of the invention introduces an adaptive threshold feedback mechanism into the algorithm.

[0168] When the system detects feature drift, it automatically corrects the threshold. :

[0169]

[0170] In the formula, express The correction threshold at any given time; express The correction threshold at any given time; With a correction factor of (0.01–0.05), this mechanism can significantly improve the versatility and robustness between different models.

[0171] Finally, taking a certain compressed air energy storage power station as an example, a specific implementation of the present invention is given.

[0172] First, a controllable axial compressor-pipeline system experimental rig was constructed. The experimental system mainly includes:

[0173] a. Three-stage tandem axial compressor body. Each stage of the blades adopts an adjustable stator structure to adjust the pressure ratio and flow characteristics.

[0174] b. Adjustable throttle valve and gas storage chamber system. Gas storage volume. The throttle valve is adjustable within the range of 5–50 m³ and is used to control system load and flow rate changes.

[0175] c. Measurement and Data Acquisition System. Pressure sensors are installed at three points each: the blade tip outlet, the rotor rear section, the outlet pipe, and the gas storage cavity; speed sensor; mass flow meter; temperature measurement sensor; data acquisition card.

[0176] d. Signal Processing and Analysis System. Real-time signal preprocessing, using an embedded DSP for filtering and synchronization; host computer, employing an industrial computer to run data fusion, modeling, and discrimination algorithms; display and alarm interface, used to display the flow-pressure ratio curve in real time and Trends in change.

[0177] Secondly, experimental conditions and procedures were set, including different combinations of rotational speed and gas storage volume, to verify... The variation pattern is as follows: Gradually reduce the throttle valve opening at rated speed to decrease the compressor flow rate.

[0178] The steps are as follows: record the pressure fluctuation signals at the blade tip outlet and in the pipeline network in real time; calculate the dimensionless discriminant number. ; Use a discriminant algorithm to determine the current system state (stable / rotating stall / surge); repeat different , Combining and statistically analyzing the first occurrence of surge value.

[0179] Then, the experimental results were analyzed, mainly including:

[0180] (1) Airflow pulsation characteristics and spectrum analysis. Analyzing the outlet pressure fluctuation spectrum under the same operating conditions, when the dominant frequency appears around 600Hz, it belongs to the typical characteristics of rotating stall, and the waveform is a periodic small-amplitude sinusoidal shape; when the spectrum energy is concentrated in the low-frequency range of 20–30Hz, periodic large-amplitude fluctuations occur, indicating that the system has entered a surge state. As the value gradually increases, the proportion of low-frequency energy in the system rises significantly. When the spectral energy ratio is greater than the set value, it indicates that the flow field has transitioned from local instability to overall flow reversal.

[0181] (2) Analysis of operating point change trajectory. Analyze the compressor flow-pressure ratio characteristic curve, when... At that time, the operating point exhibited small-scale oscillations within the stall region, but eventually stabilized; when When slow-decreasing oscillations occur, the system remains stable; when At this point, the oscillation amplitude no longer decays, and the flow rate and pressure change in opposite directions periodically, resulting in a typical surge closed trajectory.

[0182] (3) Measured critical value The determination was made through repeated experiments and statistical analysis of the discrimination results. The numerical results are consistent across different rotational speeds and gas storage volumes, verifying the stability and repeatability of the method in this embodiment of the invention.

[0183] (4) Comparison between model prediction and actual measurement. The established regression model was used for prediction, and the prediction critical value was obtained. and The comparison of measured values ​​verifies that the machine learning model has good fitting and prediction performance.

[0184] Finally, historical operating data was verified. To verify the applicability of the algorithm in a real operating environment, historical operating data (e.g., three months) of data from an experimental axial compressor were analyzed. After feature extraction and discrimination, if the model's "critical zone" alarm was consistent with the verification, it indicates that the method in this embodiment is not only suitable for laboratory verification, but also for online stability monitoring and safety early warning of long-term operating systems.

[0185] This embodiment organically combines experimental physical modeling, historical operational data regression analysis, and intelligent algorithm discrimination, resulting in significant technological advancements. Compared to traditional methods that rely on empirical critical curves, pressure-flow ratio fitting, or single experiments, this invention achieves more accurate and reliable quantitative identification and predictable determination of the critical stability boundary of the compressor system.

[0186] The advantage of this invention lies in its integration of historical operating data, constructing a complete technical chain from rigorous theoretical derivation to efficient engineering implementation, thereby achieving higher physical consistency and model scalability. This method not only ensures the repeatability of experimental conclusions but also overcomes the limitations of traditional static thresholds, enabling online adaptive updates and real-time judgment based on the actual operating status of the system, thus improving early warning and protection capabilities under dynamic operating conditions.

[0187] The embodiments of the present invention have excellent robustness. Even when faced with complex situations such as measurement noise, data errors, external interference, and even partial data loss that are unavoidable in the field, its core algorithm can still maintain a low false judgment rate.

[0188] Therefore, this embodiment is a highly reliable solution that combines high-precision discrimination capability and early warning effectiveness, providing a solid technical guarantee for the safe, stable and efficient operation of axial compressors in key facilities such as compressed air energy storage power stations.

[0189] Example 2

[0190] This embodiment discloses a compressor surge discrimination critical value determination system based on multi-source data fusion.

[0191] A multi-source data fusion-based compressor surge discrimination threshold determination system includes:

[0192] The first critical value prediction module is configured to: build and run a compressor surge test system, identify the surge start point during the test, and calculate the first surge critical value based on the test data at the surge start point.

[0193] The model determination module is configured to: acquire multi-source historical data related to compressor surge, select feature variables to construct historical input vectors, fit a multivariate regression model using the historical input vectors to obtain a discriminant model; and simultaneously train a machine learning model using the historical input vectors to obtain a trained machine learning model.

[0194] The second critical value prediction module is configured to: construct a real-time input vector based on the real-time collected data of the compressor to be monitored, and input the real-time input vector into the discrimination model and the trained machine learning model to obtain the second surge critical value and the third surge critical value;

[0195] The fusion module is configured to fuse the first surge threshold, the second surge threshold, and the third surge threshold to obtain the predicted surge threshold of the compressor to be monitored.

[0196] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0197] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for determining the critical value of compressor surge detection based on multi-source data fusion, characterized in that, Includes the following steps: A compressor surge test system was built and run to identify the surge initiation point during the test. Based on the test data at the surge initiation point, the first surge critical value was calculated. Multi-source historical data related to compressor surge are obtained, feature variables are selected to construct historical input vectors, and a multivariate regression model is fitted using the historical input vectors to obtain a discriminant model; at the same time, a machine learning model is trained using the historical input vectors to obtain a trained machine learning model; A real-time input vector is constructed based on the real-time collected data of the compressor to be monitored. The real-time input vector is then input into the discrimination model and the trained machine learning model to obtain the second surge threshold and the third surge threshold. The first surge threshold, the second surge threshold, and the third surge threshold are fused together to obtain the predicted surge threshold of the compressor to be monitored. Multi-source historical data related to compressor surge, specifically including axial compressor bench test data, industrial field operation record data, and compressor instability boundary data under different geometric parameters; From multi-source historical data, various feature variables are selected to construct a historical input vector. , ;in, This indicates the tip tangential velocity of the compressor blades; Indicates the speed of sound; Indicates the volume of the gas storage tank; This indicates the cross-sectional area of ​​the compressor outlet pipe; Indicates the equivalent length of the outlet pipeline; Indicates the pressure ratio; Indicates the flow coefficient; Indicates adiabatic efficiency; Represents the Reynolds number; Represents the Strauhal number; The measured surge critical discrimination criterion value is used as the output variable. ; By fitting a multiple regression model using historical input vectors, a discriminant model is obtained, which specifically includes: The formula for constructing a multiple regression model is as follows: ； in, For constant terms; These are the regression coefficients; These are higher-order nonlinear function terms; For the first One feature parameter; For the first One feature parameter; This is the surge threshold value; Fitting historical input vectors using the least squares method With output variables To obtain basic statistical relationships and form a discriminant model; The machine learning model is trained using historical input vectors, resulting in a trained machine learning model. The training objective function is: ； in, and These represent the model weights and biases, respectively. Indicates the number of samples; Indicates the first One sample; Indicates the first The true surge threshold value for each sample; Indicates the first The surge threshold value output by the machine learning model for each sample; Represents the regularization coefficient; The predicted surge threshold of the compressor to be monitored is calculated as follows: ； in, This indicates the predicted surge threshold. Indicates weight, This indicates the first surge threshold, the second surge threshold, or the third surge threshold. =1,2 or 3; This represents the first surge threshold value. This indicates the second surge threshold. This indicates the third surge threshold value; The weight representing the first surge threshold. The weight representing the second surge threshold. This represents the weight of the third surge threshold.

2. The method for determining the critical value of compressor surge discrimination based on multi-source data fusion as described in claim 1, characterized in that, The process of setting up the compressor surge test system includes: A multi-stage adjustable stationary vane axial flow compressor was selected as the main test unit. A pipeline is installed between the multi-stage adjustable stationary vane axial flow compressor and the air receiver, with a throttle valve at the end of the pipeline. Multiple sensors are deployed: a pressure sensor is installed at the compressor rotor blade tip outlet, multiple pressure sampling points are arranged in a circumferential distribution at the compressor outlet section, a static pressure sensor is installed in the middle section of the pipeline, pressure and temperature sensors are installed inside the gas storage tank, and flow meters and thermometers are installed in the intake pipeline. or, Operating the compressor surge test system specifically includes: The throttle valve is controlled by adjusting its opening to change the compressor's back pressure and flow rate, thereby controlling the operating point state; the motor is used as a power source to control the compressor speed. During each test, the compressor was controlled to enter the near-unstable region, the unstable region, and the safe retreat phase in sequence from the stable operation phase. During this process, the pressure fluctuation at the rotor tip outlet, the average static pressure at the compressor outlet, the pressure fluctuation in the middle section of the pipeline, the pressure and temperature of the gas accumulator, and the compressor speed and flow rate were recorded. The experiment was repeated under different gas storage tank volumes, equivalent lengths of outlet pipes, and rotational speeds, and the obtained experimental data were used to construct a multidimensional database.

3. The method for determining the critical value of compressor surge discrimination based on multi-source data fusion as described in claim 1, characterized in that, The first surge threshold is calculated as follows: ； ； in, Indicates surge discrimination parameters, Indicates time; when The value is calculated at the surge initiation point. This is the first surge threshold value; This indicates the tip tangential velocity of the compressor blades; Indicates the speed of sound; Indicates the volume of the gas storage tank; This indicates the cross-sectional area of ​​the compressor outlet pipe; Indicates the equivalent length of the outlet pipeline; Indicates time; Indicates specific heat ratio; Represents the gas constant; This indicates the temperature of the gas storage tank.

4. The method for determining the critical value of compressor surge discrimination based on multi-source data fusion as described in claim 1, characterized in that, A real-time input vector is constructed based on the real-time acquired data of the compressor under monitoring, specifically including: The flow rate, speed, pressure, temperature, and sound pressure signals of the compressor under monitoring are collected in real time, and then filtered and feature extracted to form a real-time input vector. ; The real-time input vector Compared with historical input vectors They have the same types of characteristics.

5. The method for determining the critical value of compressor surge discrimination based on multi-source data fusion as described in claim 4, characterized in that, It also includes real-time calculation of the discrimination parameters of the compressor under monitoring, using the predicted surge critical value of the compressor under monitoring as a benchmark, and combining it with the set empirical safety margin to determine the stability of the compressor under monitoring; The judgment formula is expressed as follows: ； In the formula, For experience-based safety margin; This indicates the predicted surge threshold value for the compressor to be monitored; Indicates surge discrimination parameters, Indicates time.

6. A compressor surge discrimination critical value determination system based on multi-source data fusion, characterized in that, include: The first critical value prediction module is configured to: build and run a compressor surge test system, identify the surge start point during the test, and calculate the first surge critical value based on the test data at the surge start point. The model determination module is configured to: acquire multi-source historical data related to compressor surge, select feature variables to construct historical input vectors, fit a multivariate regression model using the historical input vectors to obtain a discriminant model; and simultaneously train a machine learning model using the historical input vectors to obtain a trained machine learning model. The second critical value prediction module is configured to: construct a real-time input vector based on the real-time collected data of the compressor to be monitored, and input the real-time input vector into the discrimination model and the trained machine learning model to obtain the second surge critical value and the third surge critical value; The fusion module is configured to fuse the first surge threshold, the second surge threshold, and the third surge threshold to obtain the predicted surge threshold of the compressor to be monitored. Multi-source historical data related to compressor surge, specifically including axial compressor bench test data, industrial field operation record data, and compressor instability boundary data under different geometric parameters; From multi-source historical data, various feature variables are selected to construct a historical input vector. , ;in, This indicates the tip tangential velocity of the compressor blades; Indicates the speed of sound; Indicates the volume of the gas storage tank; This indicates the cross-sectional area of ​​the compressor outlet pipe; Indicates the equivalent length of the outlet pipeline; Indicates the pressure ratio; Indicates the flow coefficient; Indicates adiabatic efficiency; Represents the Reynolds number; Represents the Strauhal number; The measured surge critical discrimination criterion value is used as the output variable. ; By fitting a multiple regression model using historical input vectors, a discriminant model is obtained, which specifically includes: The formula for constructing a multiple regression model is as follows: ； in, For constant terms; These are the regression coefficients; These are higher-order nonlinear function terms; For the first One feature parameter; For the first One feature parameter; This is the surge threshold value; Fitting historical input vectors using the least squares method With output variables To obtain basic statistical relationships and form a discriminant model; The machine learning model is trained using historical input vectors, resulting in a trained machine learning model. The training objective function is: ； in, and These represent the model weights and biases, respectively. Indicates the number of samples; Indicates the first One sample; Indicates the first The true surge threshold value for each sample; Indicates the first The surge threshold value output by the machine learning model for each sample; Represents the regularization coefficient; The predicted surge threshold of the compressor to be monitored is calculated as follows: ； in, This indicates the predicted surge threshold. Indicates weight, This indicates the first surge threshold, the second surge threshold, or the third surge threshold. =1,2 or 3; This represents the first surge threshold value. This indicates the second surge threshold. This indicates the third surge threshold value; The weight representing the first surge threshold. The weight representing the second surge threshold. This represents the weight of the third surge threshold.

Images