Turbine compressor surge early warning method and system based on hybrid modeling and neural network

Abstract

The invention discloses a turbine compressor surge early warning method and system based on hybrid modeling and a neural network, and belongs to the technical field of compressor pipe network system surge early warning. According to the method, a hybrid model driven by target compressor pipe network system data is established on the basis of a mechanism model, high-precision analog simulation of the surge working condition is achieved, key equipment of a compressor system is subjected to physical and simulation application treatment through the hybrid modeling technology, a parameter transmission mode is optimized, and the system reliability is improved. High-precision simulation of the surge working condition of the complex compressor pipe network system can be achieved, and powerful support is provided for accurate recognition and early warning of the surge working condition of the compressor; according to the method, on the basis of experimental research equipment, the surge characteristics of the compressor are obtained through the modeling simulation technology. On the premise that the dynamic simulation precision is met, high cost of destructive test research is avoided, and advancement and practicability are achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of compressor pipe network system surge warning, and in particular relates to a turbine compressor surge warning method and system based on hybrid modeling and neural network. Background Art

[0002] Turbine compressors are important equipment in the industrial field and are widely used in the fields of chemical industry, metallurgy, aviation, etc. They increase the gas pressure through the interaction between the high-speed rotating impeller and the working gas, and at the same time accelerate the airflow to obtain kinetic energy, which is input into the downstream equipment after being pressurized in the diffuser. When the compressor pipe network system is working, surge phenomenon in which the compressor pressure fluctuates repeatedly is prone to occur under low flow conditions. The periodic strong vibrations that are usually accompanied by surge can cause devastating damage to the equipment and need to be avoided during operation. With the improvement of the automation degree of production equipment and the reduction of personnel, the timely prediction and early warning of surge conditions in the compressor system are particularly important for realizing intelligent production.

[0003] Due to the coupling of multiple components in the turbine compressor pipe network system, the cost of surge test is high. Using numerical modeling to simulate surge conditions and obtain accurate and efficient surge warning methods has always been a hot topic for scholars at home and abroad. However, the existing surge experiments are difficult to achieve surge prevention and control of complex compressor pipe network systems. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a turbine compressor surge warning method and system based on hybrid modeling and neural network to realize the identification of compressor system surge conditions, perform real-time surge risk warning, solve the surge prevention and control problems of complex compressor pipe network systems, and realize accurate identification and warning of compressor surge conditions based on compressor system simulation. In order to achieve the above object, the present invention adopts the following technical solutions: A turbine compressor surge early warning method based on hybrid modeling and neural network includes the following steps: S1, collecting the working parameters of the compressor pipe network system; S2, based on the operating parameters, predicting the compressor operating condition through the surge risk identification network; the compressor operating condition includes a surge operating condition and a normal operating condition, if it is a surge operating condition, execute S3, otherwise execute S1 and S2 cyclically; The process of obtaining the surge risk identification network is as follows: Construct a simulation model of the compressor pipe network system; adjust the opening of the upstream valve and the downstream valve of the compressor in the compressor pipe network system, simulate the surge condition and the normal condition through the simulation model of the compressor pipe network system, and obtain the working parameters of the surge condition and the normal condition; perform Fourier transformation on the signal of the working parameter to obtain the time-frequency spectrum; train the convolutional neural network through the time-frequency spectrum and the corresponding working conditions to obtain the surge risk identification network; S3, issue an early warning.

[0005] A further improvement of the present invention is: Preferably, in S1, the operating parameters include compressor temperature, compressor pressure and compressor flow.

[0006] Preferably, in S2, the signal of the operating parameter is subjected to Fourier transform to obtain a time-frequency spectrum, and the time-frequency spectrum is input into a surge risk identification network to predict the compressor operating condition.

[0007] Preferably, in S2, the oscillating part in the signal of the operating parameter is a surge condition.

[0008] Preferably, in S2, the surge condition and the normal condition are simulated respectively by the compressor pipe network system simulation model; a training data set corresponding to the surge condition and a data set corresponding to the normal condition are obtained respectively; and a convolutional neural network is trained together.

[0009] Preferably, after S3, the operating condition is adjusted by adjusting the opening of the upstream valve or the downstream valve of the compressor after the early warning.

[0010] A turbine compressor surge warning system based on hybrid modeling and neural network includes a software part and a hardware part. The software part is used for the above method, and the hardware part includes a compressor pipe network system, a data acquisition and conversion device, and a computer. A sensor is provided in the compressor pipe network system, and the sensor is electrically connected to the data acquisition and conversion device, and the data acquisition and conversion device is electrically connected to the computer.

[0011] Preferably, the software part includes: Collection module, collecting the working parameters of the compressor pipe network system; A prediction module predicts the compressor operating condition through a surge risk identification network based on the operating parameters; the compressor operating condition includes a surge condition and a normal condition. If it is a surge condition, the early warning module is executed, otherwise the acquisition module and the prediction module are executed cyclically; The process of obtaining the surge risk identification network is as follows: Construct a simulation model of the compressor pipe network system; adjust the opening of the upstream valve and the downstream valve of the compressor in the compressor pipe network system, simulate the surge condition and the normal condition through the simulation model of the compressor pipe network system, and obtain the working parameters of the surge condition and the normal condition; perform Fourier transformation on the signal of the working parameter to obtain the time-frequency spectrum; train the convolutional neural network through the time-frequency spectrum and the corresponding working conditions to obtain the surge risk identification network; Early warning module, for early warning.

[0012] Preferably, the compressor pipe network system includes a connected turbine and a compressor downstream pipe network, a compressor inlet regulating valve is arranged in front of the turbine, and a compressor outlet regulating valve is arranged after the compressor downstream pipe network.

[0013] Preferably, corresponding sensors are installed between the compressor inlet regulating valve and the turbine compressor, between the turbine compressor and the compressor downstream pipeline, between the compressor downstream pipeline and the compressor outlet regulating valve, the compressor downstream pipeline and the turbine compressor.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The present invention discloses a turbine compressor surge warning method based on hybrid modeling and neural network. The method establishes a hybrid model driven by data of a target compressor pipe network system on the basis of a mechanism model, realizes high-precision simulation of surge conditions, and processes key equipment of the compressor system by taking into account both physical and simulation applications through hybrid modeling technology, and optimizes the parameter transfer method, which can realize high-precision simulation of surge conditions of complex compressor pipe network systems, and provide strong support for accurate identification and early warning of compressor surge conditions; the method obtains compressor surge characteristics by using modeling and simulation technology on the basis of experimental research equipment. On the premise of meeting the dynamic simulation accuracy, the high cost of destructive experimental research is avoided, and it is advanced and practical.

[0015] Furthermore, the method extracts the surge condition characteristics of the target system and obtains a surge risk identification network based on a convolutional neural network. The risk identification network has been trained with a test data set and has high recognition accuracy, and can accurately warn the dynamic characteristics of the compressor pipe network system before the surge condition.

[0016] The present invention also discloses a turbine compressor surge warning system based on hybrid modeling and neural network, including 1) the experimental research device is divided into a hardware part and a software part; 2) the hardware part is a collection and control system composed of a pressure sensor, a temperature sensor, a flow sensor, a signal collection and processing device and a computer; 3) the software part is a dynamic simulation model of the target compressor pipe network system and a computer surge risk identification and warning; 4) the dynamic simulation model of the compressor pipe network system is a semi-mechanism model, and a semi-mechanism model with higher accuracy is obtained by deep learning of the test data on the basis of the compressor pipeline mechanism model. The semi-mechanism model is used to dynamically simulate the surge condition of the compressor pipe network system, and the simulation result is output; 5) the surge risk identification network obtained based on the convolutional neural network training is installed on the computer, and the real-time data obtained by the sensor and the signal transmission system is used to predict the surge risk of the target compressor system, and an alarm is issued for the surge condition. Compared with the prior art, the present invention is oriented to the surge warning field of the compressor pipe network system, and avoids the risk of surge test while ensuring the high reliability of the simulation result, with low cost, high safety, and both practicality and versatility. The compressor surge warning device of the present invention can warn of surge conditions while ensuring simulation accuracy, thereby greatly reducing the risk of the compressor entering a surge condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a technical block diagram of the present invention; Figure 2 This is the structural diagram of the compressor pipe network system test equipment; Figure 3 It is a dynamic simulation pressure change diagram under normal-surge conditions of the compressor pipe network system; Figure 4 It is a schematic diagram of the prediction results using the surge risk identification network; Among them, 1-compressor inlet regulating valve; 2-turbine compressor; 3-compressor downstream pipeline network; 4-compressor outlet regulating valve; 5-outlet regulating valve control signal; 6-A / D converter; 7-data transmission signal; 8-computer; 9-compressor test signal; 10-inlet regulating valve control signal. DETAILED DESCRIPTION

[0018] The present invention is further described in detail below in conjunction with the accompanying drawings: Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0019] The present invention discloses a turbine compressor surge early warning method based on hybrid modeling and neural network, the method comprising the following steps: S1, collect the working parameters of the compressor pipe network system.

[0020] The working parameters include the collected compressor temperature, compressor pressure and compressor flow rate. The collected working parameters can be used to determine in real time whether the compressor in the compressor pipe network system is surging.

[0021] S2, based on the operating parameters, predicting the compressor operating condition through the surge risk identification network; the compressor operating condition includes a surge operating condition and a normal operating condition, if it is a surge operating condition, execute S3, otherwise execute S1 and S2 cyclically; S3, issue an early warning.

[0022] In S2, the process of obtaining the surge risk identification network is as follows: S2.1, the mechanism model of the compressor system is obtained through the mass conservation, momentum conservation, and energy conservation equations, and the mechanism model is data-driven optimized through key test data to obtain the hybrid modeling of the compressor system and the dynamic simulation model. The construction of this simulation model refers to the published patent CN 115356096 A.

[0023] S2.2, based on the semi-mechanism dynamic simulation model of the compressor pipe network system, the compressor surge condition is simulated, and the compressor pipe network system is made to enter the surge condition and the normal condition by adjusting the opening of the upstream valve and the downstream valve of the compressor.

[0024] Output the compressor working parameters of the compressor pipe network system under surge conditions or normal conditions. The working parameters are the results of temperature, pressure or flow. After Fourier transforming the signal of any working parameter under surge and normal conditions, a time-frequency spectrum (time-frequency spectrum) is obtained. The time-frequency spectrum is associated with the working conditions as simulation data. Multiple simulation data constitute a data set. In particular, for surge conditions, the compressor surge characteristics in the signal fluctuation are extracted, which is the oscillating part in the simulation results, including amplitude frequency, etc.

[0025] S2.3, providing a data source for neural network training through simulation data sets of surge conditions and normal conditions of the compressor dynamic simulation system, and training the convolutional neural network using time-frequency graphs of normal conditions and surge conditions; obtaining a surge risk identification network; During the training process, if the recognition accuracy of the surge risk recognition network meets the standard, execute step 3, otherwise repeat steps 2.1-S2.3 to retrain the convolutional neural network; S3, through obtaining the compressor pipeline dynamic simulation system to dynamically simulate and predict a specific working condition, the simulation results are imported into the surge risk identification network with the required accuracy, the surge risk of the target compressor working condition is identified in real time, and the surge condition is alarmed.

[0026] Specifically, the input and output parameters of the target system are transmitted to the computer through sensors and signal transmission systems, the operating conditions of the system are diagnosed based on the surge recognition network, the diagnosis results are obtained in the computer software system, and real-time surge risk warning is achieved.

[0027] The turbine compressor surge early warning system based on hybrid modeling and neural network of the present invention gives full play to the advantages of experimental research and simulation research, adopts advanced modeling methods, obtains dynamic simulation of pressure, temperature and flow changes under typical working conditions of the compressor pipe network system, simulates and analyzes working conditions with experimental risks such as surge, and has reliable simulation results and a wide range of applications. It trains high-precision convolutional neural networks to obtain surge risk identification networks, extracts and identifies features of target compressor working conditions, and monitors the risk of the compressor system entering surge in real time, with high identification efficiency and reduced cost risks.

[0028] In some embodiments of the present invention, after S2, a test process is further performed after the surge risk identification network is obtained. The test process is: Step 1, see Figure 2 , establish a turbine compressor system turbine compressor surge warning system, obtain the outlet regulating valve control signal 5 and the data transmission compressor test signal 7 through the sensor, and realize data-driven optimization through the A / D converter 6.

[0029] The sensors and reading devices include compressor inlet and outlet pressure sensors, compressor outlet flow sensor, speed reading device, power reading device, opening reading devices at each valve, and atmospheric temperature and pressure sensors.

[0030] Step 2, using the opening of the compressor inlet regulating valve 1 and the compressor outlet regulating valve 4 as control parameters, actual simulation of surge conditions and normal conditions is implemented in the target compressor pipe network system to obtain a test data set; Step 3: For the data in the test data set, the pressure signal of the surge condition of the target compressor system is processed based on the short-time Fourier transform to generate a time-frequency spectrum, showing the changes of the pressure signal in the time and frequency dimensions.

[0031] Step 4: Input the time-frequency spectrum into the above-mentioned surge risk identification network, predict the surge condition, compare the prediction result with the actual simulation condition, and verify the accuracy and reliability of the surge risk identification network. If the accuracy is not met, re-verify.

[0032] According to compressor theory, surge is caused by excessive resistance downstream of the compressor and the compressor's ability to work unable to give the working fluid enough energy to leave the system. At this time, the downstream pipe network pressure is temporarily higher than the compressor outlet pressure, the flow rate decreases or even backflow occurs, which causes pressure fluctuations and the system enters a surge condition. In actual working conditions, when the valve opening downstream of the compressor decreases and the resistance of the pipe network downstream of the compressor increases, the system will enter a surge condition.

[0033] The second aspect of the present invention discloses a turbine compressor surge warning system based on hybrid modeling and neural network, see Figure 1 The device includes a hardware part and a software part. The hardware part is a computer data acquisition and control system, and the software part includes a dynamic simulation module and a computer surge warning module.

[0034] The dynamic simulation module is used to obtain the surge risk identification network. The specific process is as follows: construct a compressor pipe network system simulation model; adjust the compressor upstream valve opening and downstream valve opening in the compressor pipe network system, simulate the surge working condition and normal working condition through the compressor pipe network system simulation model, and obtain the surge working condition working parameters and normal working condition working parameters; perform Fourier transformation on the signal of the working parameter to obtain the time-frequency spectrum; train the convolutional neural network through the time-frequency spectrum and the corresponding working condition to obtain the surge risk identification network; The computer surge warning module includes an acquisition module, a prediction module and a warning module.

[0035] A collection module is used to collect the working parameters of the compressor pipe network system; A prediction module is used to predict the compressor operating condition based on the operating parameters through the surge risk identification network; the compressor operating condition includes a surge condition and a normal condition. If it is a surge condition, the early warning module is executed, otherwise the acquisition module and the prediction module are executed cyclically; The early warning module is used for early warning.

[0036] Figure 2 The structural diagram of the device described in the present invention is given, which is divided into a compressor pipe network system and a signal collection and processing device. The compressor pipe network system consists of a compressor inlet 1, a turbine compressor 2, a pipeline, and a compressor downstream regulating valve 3. The signal collection and processing device consists of temperature, pressure, flow sensors, speed, power, opening monitoring devices, signal collection and conversion equipment, and a computer, wherein each sensor is mainly distributed at the upstream inlet, downstream outlet, and valve of the compressor. Sensors are arranged on the pipeline, the regulating valve, or the turbine compressor 2.

[0037] The compressor pipe network system is composed of a compressor inlet regulating valve 1, a turbine compressor 2, a compressor downstream pipeline 3, and a compressor outlet regulating valve 4. Specifically, a compressor inlet regulating valve 1 is provided on the inlet pipeline of the turbine compressor 2, the outlet of the turbine compressor 2 is connected to the compressor downstream pipeline 3, and the compressor outlet regulating valve 4 is provided on the compressor downstream pipeline 3. A test point is provided between the compressor inlet regulating valve 1 and the centrifugal compressor 2, and a test point is provided on the inlet of the compressor downstream pipeline 3. Pressure sensors are respectively installed at the inlet and outlet of the compressor, a flow sensor, a speed monitoring device, and a power monitoring device are installed at the compressor outlet, and an opening reading device, an atmospheric environment setting temperature, and a pressure sensor are installed at each valve. Specifically, the flow sensor described in the present invention is a conical throttle and a pitot tube, and the temperature sensor is a thermometer.

[0038] The signal acquisition and processing device includes a register, a relay and an A / D converter 6. In the compressor performance test, the compressor operating point is moved by changing the opening of the compressor inlet regulating valve 1 and the compressor outlet regulating valve 4. In the compressor pipe network system, sensors are placed at the inlet and outlet of the valve and the compressor to obtain the test data of the system under different working conditions. The sensor at each test point transmits the compressor test signal 9 to the A / D converter 6, and finally transmits the test data to the software part in the computer 8. Various sensors are connected to the data acquisition software in the computer through the signal conversion device based on the ModBus communication protocol. When the compressor pipe network system is tested, the valve opening, compressor speed and power, pressure, temperature and flow data are collected by the sensor, input into the computer through the A / D converter, and collected by the computer software. Finally, the compressor valve system test data of pressure, temperature and flow changing with time is formed.

[0039] The research device of the present invention realizes valve control through software, specifically, inputs valve control instructions into the computer and software part 8, inputs them into the A / D converter 6 through the data transmission signal 7, and realizes the control of the regulating valve through the ModBus communication protocol and register, outputs communication instructions and relays in the computer software to realize the control of the regulating valve, inputs the outlet valve control signal 5 into the compressor outlet regulating valve 4, realizes the change of the action of the compressor outlet regulating valve 4, or inputs the inlet valve control signal 10 into the compressor inlet regulating valve 1, realizes the change of the action of the compressor outlet regulating valve 1, and changes the compressor working condition. Through the above process, the performance test of the compressor pipe network system under various working conditions can be carried out, and the test data of the pressure, temperature and flow at the compressor and the boundary can be collected and processed.

[0040] The present invention trains a convolutional neural network on the data obtained by simulating the mechanism model of the key components of the compressor pipe network system and the actual compressor pipe network system experimental data set to obtain a surge risk warning network and achieve high-precision surge prediction of the target compressor. By setting boundary conditions and initial conditions, the user can achieve real-time dynamic simulation of the target compressor pipe network system and output and store the simulation results. The simulation results include the dynamic changes of the upstream and downstream pressures, temperatures, and flows of the compressor over time. Through the real-time simulation of the compressor pipe network system, the dynamic changes of the corresponding operating pressure, temperature, and flow are obtained, and the characteristic analysis and behavior prediction of the surge condition are achieved.

[0041] The following is further described in conjunction with a specific embodiment: Step 1: Obtain the mechanism model of the compressor system through the mass conservation, momentum conservation, and energy conservation equations, and perform data-driven optimization on the mechanism model through key test data to obtain a simulation model of the compressor system. Step 2: Based on the semi-mechanism dynamic simulation model of the compressor pipe network system, the compressor surge condition is simulated, and the compressor pipe network system enters the surge condition by adjusting the opening of the upstream and downstream valves of the compressor. The dynamic simulation results of the compressor temperature, pressure, and flow rate when the compressor pipe network system enters the surge condition are output, and the compressor surge characteristics in the signal fluctuation are extracted, which is the oscillating part of the simulation result, including amplitude frequency, etc.

[0042] Step 3, the surge condition and normal condition simulation of the compressor dynamic simulation system are used to provide data sources for neural network training, and the time-frequency diagrams of normal and surge conditions are used to train the convolutional neural network. First, the corresponding time-frequency diagrams are generated by applying short-time Fourier transform (STFT) to the raw data of surge and normal conditions. Then, the trained convolutional neural network (CNN) is used to identify the time-frequency diagram. 16 2×1 convolution kernels are used in the convolution layer of the CNN network. By capturing the frequency characteristics of each condition, surge and normal conditions can be effectively distinguished. The recognition results are output in the form of labels corresponding to each condition category, with surge label 1 and normal condition label 2. 425 groups of surge data and 25 groups of normal condition data are used as training sets to train the CNN network, and 70 groups of surge data and 10 groups of normal condition data are used as test sets to verify the accuracy of the trained network. The accuracy of the training set and the test set is 100%, indicating that the quality of the trained network meets the requirements of compressor surge condition recognition.

[0043] Step 4, dynamically simulate and predict a specific working condition by obtaining the compressor pipe network dynamic simulation system, import the simulation results into the surge risk identification network with satisfactory accuracy, identify the surge risk of the target compressor working condition in real time, and alarm the surge condition.

[0044] In a dynamic simulation study, the valve opening was adjusted gradually, as follows: In the stage from 0 to 5 seconds, the pressure signal showed normal operation, but at 5 seconds, due to the change in valve action, the pressure increased and then remained stable, entering the stage of outputting high-pressure gas. At 8 seconds, due to improper valve operation, the compressor system entered a surge state, which lasted until 14 seconds. After 14 seconds, the control valve was adjusted to leave the fault condition and the system re-entered the normal operation stage.

[0045] Figure 3 , Figure 4 The simulation results of the pressure during surge of the dynamic simulation model of the compressor pipe network system and the prediction results of the surge risk identification network are given. When realizing the real-time abnormal operating condition monitoring of the compressor system, the amount of data read by the abnormal operating condition identification network each time is the amount of data contained in each group of data in the training set when training the network, that is, 10,000 data, which is equivalent to the amount of data within 1 second. Therefore, the network starts working from t=1s, and then reads data every 0.5 seconds to identify the operating status within 1 second before this moment, such as Figure 4 When all conditions are normal within 1 second, label 2 is output; when surge abnormality is included within 1 second, label 1 is output. By comparison Figure 3 and Figure 4 It is found that the compressor abnormal condition identification result is consistent with the actual situation. It can be seen that the turbine compressor surge warning system based on hybrid modeling and neural network proposed in the present invention is reliable.

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A turbine compressor surge early warning method based on hybrid modeling and neural network, characterized in that: The following steps are involved: S1, collecting the working parameters of the compressor pipe network system; S2, based on the operating parameters, the compressor operating condition is predicted through the surge risk identification network; The compressor operating condition includes a surge condition and a normal condition. If it is a surge condition, S3 is executed, otherwise S1 and S2 are executed cyclically; The process of obtaining the surge risk identification network is as follows: Construct a simulation model of the compressor pipe network system; adjust the opening of the upstream valve and the downstream valve of the compressor in the compressor pipe network system, simulate the surge condition and the normal condition through the simulation model of the compressor pipe network system, and obtain the working parameters of the surge condition and the normal condition; perform Fourier transformation on the signal of the working parameter to obtain the time-frequency spectrum; train the convolutional neural network through the time-frequency spectrum and the corresponding working conditions to obtain the surge risk identification network; S3, issue an early warning.

2. A turbine compressor surge early warning method based on hybrid modeling and neural network according to claim 1, characterized in that: In S1, the operating parameters include compressor temperature, compressor pressure and compressor flow.

3. The turbine compressor surge early warning method based on hybrid modeling and neural network according to claim 1 is characterized in that: In S2, the signal of the working parameter is transformed by Fourier to obtain a time-frequency spectrum, which is then input into a surge risk identification network to predict the compressor operating condition.

4. The turbine compressor surge early warning method based on hybrid modeling and neural network according to claim 1 is characterized in that: In S2, the oscillating part in the signal of the operating parameter is a surge condition.

5. The turbine compressor surge early warning method based on hybrid modeling and neural network according to claim 1 is characterized in that: In S2, the surge condition and the normal condition are simulated respectively through the compressor pipe network system simulation model; the training data set corresponding to the surge condition and the data set corresponding to the normal condition are obtained respectively; and the convolutional neural network is trained together.

6. The turbine compressor surge early warning method based on hybrid modeling and neural network according to claim 1, characterized in that: After S3, the operating condition is adjusted by adjusting the opening of the upstream valve or the downstream valve of the compressor after the early warning.

7. A turbine compressor surge warning system based on hybrid modeling and neural network, characterized in that: It includes a software part and a hardware part, the software part is used to implement the method described in claim 1, and the hardware part includes a compressor pipe network system, a data acquisition and conversion device, and a computer; a sensor is provided in the compressor pipe network system, the sensor is electrically connected to the data acquisition and conversion device, and the data acquisition and conversion device is electrically connected to the computer.

8. The turbine compressor surge warning system based on hybrid modeling and neural network according to claim 7 is characterized in that: The software part includes: Collection module, collecting the working parameters of the compressor pipe network system; A prediction module predicts the compressor operating condition through a surge risk identification network based on the operating parameters; the compressor operating condition includes a surge condition and a normal condition. If it is a surge condition, the early warning module is executed, otherwise the acquisition module and the prediction module are executed cyclically; The process of obtaining the surge risk identification network is as follows: Construct a simulation model of the compressor pipe network system; adjust the opening of the upstream valve and the downstream valve of the compressor in the compressor pipe network system, simulate the surge condition and the normal condition through the simulation model of the compressor pipe network system, and obtain the working parameters of the surge condition and the normal condition; perform Fourier transformation on the signal of the working parameter to obtain the time-frequency spectrum; train the convolutional neural network through the time-frequency spectrum and the corresponding working conditions to obtain the surge risk identification network; Early warning module, for early warning.

9. The turbine compressor surge warning system based on hybrid modeling and neural network according to claim 6, characterized in that: The compressor pipe network system comprises a connected turbine (2) and a compressor downstream pipe network (3), a compressor inlet regulating valve (1) is arranged in front of the turbine (2), and a compressor outlet regulating valve (4) is arranged behind the compressor downstream pipe network (3).

10. The turbine compressor surge warning system based on hybrid modeling and neural network according to claim 7, characterized in that: Corresponding sensors are installed between the compressor inlet regulating valve (1) and the turbine compressor (2), between the turbine compressor (2) and the compressor downstream pipeline (3), between the compressor downstream pipeline (3) and the compressor outlet regulating valve (4), and on the compressor downstream pipeline (3) and the turbine compressor (2).

Images