Multi-stage compression dynamic correlation surge suppression system and method for compressed air energy storage power station

Abstract

The disclosure provides a compressed air energy storage power station multi-stage compression dynamic correlation surge suppression system and method, relating to the technical field of compressed air energy storage, comprising: acquiring real-time operation parameters of a multi-stage compressor system; based on the real-time operation parameters, estimating the internal state variables of the compressor using an extended Kalman filter algorithm; inputting the estimated internal state variables of the compressor into a surge dynamic boundary correlation model, calculating the stability margin and predicting the stability evolution trend in the future time; based on the stability margin calculation result, executing a control decision, calculating the optimal control action by an intelligent controller based on meta-reinforcement learning and a space-time graph neural network, and outputting a control instruction to an executing mechanism; the executing mechanism adjusts the compressor operation parameters to ensure that the system is away from the surge boundary; the disclosure forms a cooperative suppression closed loop from accurate perception, intelligent prediction to optimized decision and efficient execution, which can greatly reduce the probability and intensity of surge.

Description

Technical Field

[0001] This disclosure relates to the field of compressed air energy storage technology, specifically to a multi-stage compression dynamic correlation surge suppression system and method for compressed air energy storage power stations. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] With the accelerated pace of global energy transition, the installed capacity of renewable energy is expanding at an unprecedented rate. Compressed air energy storage systems, with their advantages of large-scale energy storage, long lifespan, and environmental friendliness, have become one of the key technologies supporting the stable operation of new power systems. Their operational stability directly affects not only the reliability and flexible control capabilities of the entire energy system but also the core link influencing the efficient absorption of renewable energy. The surge problem of multi-stage compressors, a technical challenge during the commissioning and operation of compressed air energy storage power stations, has become a bottleneck restricting the safe and efficient operation of the system.

[0004] Current single-stage surge control methods primarily focus on local adjustment of flow parameters within the same stage, neglecting the complex dynamic coupling effects between stages in multi-stage compressors. In actual operation, stages form a strongly correlated system through pressure wave transmission, flow redistribution, and temperature field superposition. Flow instability in one stage can trigger mismatches through inter-stage parameter disturbances, ultimately leading to a chain reaction of surge in the entire compressor. This single-point control mode struggles to achieve global vibration suppression, lacking a system-level approach that considers the dynamic correlation between stages in a multi-stage coordinated control system. Summary of the Invention

[0005] To address the aforementioned issues, this disclosure proposes a multi-stage compressor dynamic correlation surge suppression system and method for compressed air energy storage power stations. It designs a dynamic correlation model of the surge boundary of a multi-stage compressor, and achieves accurate identification of surge precursors and multi-stage linkage control by real-time mapping of inter-stage parameter coupling relationships. Furthermore, it integrates advanced predictive control algorithms and edge computing technology to construct a distributed collaborative control framework, enabling accurate prediction and multi-stage collaborative suppression of surge phenomena, thereby improving the overall operational stability and efficiency of the compressed air energy storage system.

[0006] According to some embodiments, the present disclosure adopts the following technical solutions:

[0007] A method for suppressing surge caused by dynamic correlation in multi-stage compression in compressed air energy storage power stations includes:

[0008] Obtain real-time operating parameters of a multi-stage compressor system;

[0009] Based on real-time operating parameters, the extended Kalman filter algorithm is used to estimate the internal state variables of the compressor.

[0010] The estimated internal state variables of the compressor are input into the surge dynamic boundary correlation model to calculate the stability margin;

[0011] Based on the stability margin calculation results, control decisions are made. The intelligent controller based on meta-reinforcement learning and spatiotemporal graph neural network calculates the optimal control action and outputs control commands to the actuator. The actuator adjusts the compressor operating parameters to ensure that the system stays away from the surge boundary.

[0012] The compressor thermodynamic model, surge dynamic model, and pipeline volume effect are coupled, and the local flow field is simulated in detail using three-dimensional CFD numerical simulation method to construct a multi-level coupled model. Based on the multi-level coupled model, the inter-level coupling effect is considered to obtain the surge dynamic boundary correlation model.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] A multi-stage compression dynamic correlation surge suppression system for compressed air energy storage power stations includes:

[0015] The parameter acquisition module is used to acquire real-time operating parameters of a multi-stage compressor system.

[0016] The state estimation module is used to estimate the internal state variables of the compressor based on real-time operating parameters using the extended Kalman filter algorithm.

[0017] The evolution prediction module is used to input the estimated internal state variables of the compressor into the surge dynamic boundary correlation model to calculate the stability margin;

[0018] The intelligent control module is used to execute control decisions based on the stability margin calculation results. The intelligent controller, based on meta-reinforcement learning and spatiotemporal graph neural network, calculates the optimal control action and outputs control commands to the actuator. The actuator adjusts the compressor operating parameters to ensure that the system stays away from the surge boundary.

[0019] The compressor thermodynamic model, surge dynamic model, and pipeline volume effect are coupled, and the local flow field is simulated in detail using three-dimensional CFD numerical simulation method to construct a multi-level coupled model. Based on the multi-level coupled model, the inter-level coupling effect is considered to obtain the surge dynamic boundary correlation model.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions:

[0021] A computer program product includes a computer program that, when executed by a processor, implements the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power stations.

[0022] According to some embodiments, the present disclosure adopts the following technical solutions:

[0023] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power stations.

[0024] According to some embodiments, the present disclosure adopts the following technical solutions:

[0025] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the multi-stage compression dynamic correlation surge suppression method for the compressed air energy storage power station.

[0026] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0027] This disclosure discloses a multi-stage compressor dynamic correlation surge suppression method for compressed air energy storage power plants. It designs a dynamic correlation model of the surge boundary of a multi-stage compressor, and achieves accurate identification of surge precursors and multi-stage linkage control by real-time mapping of inter-stage parameter coupling relationships. Furthermore, it integrates advanced predictive control algorithms and edge computing technology to construct a distributed collaborative control framework, thereby achieving accurate prediction and multi-stage collaborative suppression of surge phenomena, and thus improving the overall operational stability and efficiency of the compressed air energy storage system.

[0028] This disclosure presents a multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power plants. Addressing the problem that traditional surge models struggle to accurately describe the complex dynamic characteristics of multi-stage compressors, this method proposes a multi-stage coupled model architecture. This architecture couples the compressor thermodynamic model, the surge dynamic model, and the pipeline volume effect. Based on a one-dimensional unit stability prediction model of "stall-hysteresis-surge line / volume adjustment," the system utilizes three-dimensional CFD numerical simulation technology to refine the local flow field, thereby constructing a surge boundary prediction model that reflects the inter-stage coupling effect. This architecture not only accurately captures the mutual influence between compressor stages but also effectively reflects the disturbance effect of stabilization measures on the local flow field.

[0029] This disclosure presents a multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power plants. It innovatively proposes a surge boundary dynamic correlation model, which establishes a mathematical relationship between surge boundary prediction and system stability assessment by tracking the dynamic operating parameters of the multi-stage compressor in real time. This model is based on an improved form of the Greitzer model and fully considers the influence of inter-stage interactions and volume effects of the multi-stage compressor. The method also proposes a multi-stage coupling model, a surge boundary dynamic correlation mechanism, and a data exchange protocol during the execution of active surge control to achieve information sharing and collaborative working mode.

[0030] This disclosure presents a multi-stage compressor dynamic correlation surge suppression method for compressed air energy storage power plants. Based on establishing a dynamic correlation model of the surge boundary of a multi-stage compressor, it addresses the problem that traditional control strategies, such as PID control and nonlinear feedback control, often fail to achieve ideal control results when facing the complex dynamics of compressor systems, including strong nonlinearity, time-varying characteristics, and multi-stage coupling. A novel intelligent control method integrating meta-reinforcement learning and spatiotemporal graphical neural networks is proposed. By combining a multi-timescale prediction mechanism with an adaptive control strategy, it achieves feedforward suppression and dynamic optimization control of multi-stage compressor surge. Attached Figure Description

[0031] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0032] Figure 1 This is a flowchart illustrating the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power stations according to an embodiment of the present disclosure.

[0033] Figure 2 This is a diagram showing the effect of surge suppression on the operating point migration in the compressor characteristic diagram according to an embodiment of this disclosure. Detailed Implementation

[0034] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0035] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. 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 disclosure pertains.

[0036] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0037] Example 1

[0038] One embodiment of this disclosure provides a method for suppressing multi-stage compression dynamic correlation surge in a compressed air energy storage power station, the method steps including:

[0039] Step 1: Obtain the real-time operating parameters of the multi-stage compressor system;

[0040] Step 2: Based on real-time operating parameters, estimate the internal state variables of the compressor using the extended Kalman filter algorithm;

[0041] Step 3: Input the estimated internal state variables of the compressor into the surge dynamic boundary correlation model to calculate the stability margin;

[0042] Step 4: Based on the stability margin calculation results, execute control decisions. The intelligent controller based on meta-reinforcement learning and spatiotemporal graph neural network calculates the optimal control action and outputs control commands to the actuator. The actuator adjusts the compressor operating parameters to ensure that the system is far away from the surge boundary.

[0043] The compressor thermodynamic model, surge dynamic model, and pipeline volume effect are coupled, and the local flow field is simulated in detail using three-dimensional CFD numerical simulation method to construct a multi-level coupled model. Based on the multi-level coupled model, the inter-level coupling effect is considered to obtain the surge dynamic boundary correlation model.

[0044] As one embodiment, the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power plants disclosed herein proposes a multi-stage coupled model architecture that couples the compressor thermodynamic model, the surge dynamic model, and the pipeline volume effect. This model is based on one-dimensional unit stability prediction of "stall-hysteresis-surge line / volume adjustment," and utilizes three-dimensional CFD numerical simulation technology to refine the local flow field, thereby constructing a surge dynamic boundary correlation model that reflects the inter-stage coupling effect. This model can not only accurately capture the mutual influence between compressor stages but also effectively reflect the disturbance effect of stabilization measures on the local flow field. Furthermore, an intelligent control method integrating meta-reinforcement learning and spatiotemporal graph neural networks is proposed. By combining a multi-timescale prediction mechanism and an adaptive control strategy, feedforward suppression and dynamic optimization control of multi-stage compressor surge are achieved. The specific implementation process is as follows:

[0045] Step 1: Obtain the real-time operating parameters of the multi-stage compressor system;

[0046] Specifically, piezoelectric pressure sensors, thermocouple temperature sensors, vortex flow meters, and acceleration vibration sensors are arranged at the inlet and outlet of each stage of the compressor. Through the high-precision sensors arranged at the inlet and outlet of each stage of the compressor, pressure, temperature, flow and vibration data are collected in real time with a sampling frequency of up to 1kHz, covering all key points to ensure that the high-frequency characteristics of surge precursors can be captured. The collected data is transmitted to the central processing unit for processing through an optical fiber network.

[0047] Step 2: Based on real-time operating parameters, estimate the internal state variables of the compressor using the extended Kalman filter algorithm;

[0048] Using the extended Kalman filter algorithm, the internal state variables of the compressor are estimated based on sensor measurements. The estimated state variables include the mass flow rate, pressure ratio, and outlet temperature of each stage of the compressor, especially the mass flow rate and pressure fluctuations.

[0049] Step 3: Input the estimated internal state variables of the compressor into the surge dynamic boundary correlation model, calculate the stability margin, and predict the stability evolution trend in the future.

[0050] Specifically, the surge dynamic boundary correlation model couples the compressor thermodynamic model, the surge dynamic model, and the pipeline volume effect, taking into account the inter-stage matching characteristics of a multi-stage series compressor.

[0051] For those with N In a multi-stage compressor system, the outlet state of each stage becomes the inlet boundary condition of the next stage, establishing a coupling relationship described by the following set of equations:

[0052] (1)

[0053] (2)

[0054] (3)

[0055] (4)

[0056] (5)

[0057] in, For the first j Mass flow rate of the stage compressor, f mass,j () is the first j The mass flow characteristic function of a multistage compressor. For pressure ratio, For the first j The absolute pressure at the outlet of the stage compressor, For the first j The absolute pressure at the inlet of the stage compressor. For the first j The pressure ratio characteristic function of a multistage compressor. For rotational speed, For the inlet density, For insulation efficiency, Specific heat ratio, and These are the pressure and temperature losses between stages, respectively. For the first j The outlet gas temperature of the stage compressor, For the first j The inlet gas temperature of the stage compressor. For the first j +1 stage compressor inlet absolute pressure For the first j Level and First j Pressure loss between +1 levels, For the first j +1 stage compressor inlet gas temperature, For the first j Level and First j Temperature drop between +1 levels.

[0058] Furthermore, the coupled compressor thermodynamic model, surge dynamic model, and pipeline volume effect are described separately, as follows:

[0059] (1) Thermodynamic model of the compressor;

[0060] The compressor thermodynamic model considers the impact of interstage cooling on the gas state and compressor performance. Based on the principles of energy conservation and mass conservation, the first... j The dynamic process of a multi-stage heat exchanger can be described by the following equation:

[0061] (6)

[0062] (7)

[0063] in, For the first j The heat exchange capacity of the stage heat exchanger It is a time constant. For heat exchanger efficiency, This is the inlet temperature of the cryogenic medium.

[0064] (2) Surge dynamic model;

[0065] The surge boundary dynamic model establishes a mathematical relationship between surge boundary prediction and stability assessment by tracking the dynamic parameters of a multi-stage compressor in real time. This model is an improvement on the Greitzer model, taking into account the inter-stage interactions and volumetric effects of the multi-stage compressor.

[0066] For multi-stage compressor systems, the traditional Greitzer model is extended to:

[0067] (8)

[0068] (9)

[0069] (10)

[0070] in, and They represent the first j The dimensionless flow rate and pressure rise coefficient of the class and Indicates the equivalent circulation area and length. Indicates the aggregated volume. and These represent the rotor speed and the speed of sound, respectively. Indicates the flow resistance coefficient; For the first j The Greitzer stability parameter is a dimensionless number used to characterize the dynamic properties of a compression system and is a core criterion for predicting its instability mode (whether it tends towards surge or rotational stall).

[0071] The dynamic surge boundary is defined as the critical state in which the system loses stability, and is determined by solving the eigenvalues ​​of the Jacobian matrix of the dynamic system described above:

[0072] (11)

[0073] in, Let Jacobian matrix be the dynamical system shown in equations (8)-(9). This is the critical characteristic value.

[0074] Furthermore, the dynamic correlation surge suppression synergistic system considers the special characteristics of compressed air energy storage systems, namely the significant impact of storage tank pressure changes on compressor operating points. By introducing a storage tank thermodynamic model, the system can predict the impact of storage tank pressure changes on compressor stability:

[0075] (12)

[0076] (13)

[0077] in, For the gas quality in the gas storage facility, and These are the mass flow rates for compression and expansion, respectively. For the specific internal energy of the gas, and For the import / export enthalpy, and The heat exchange area and coefficient of the gas storage facility. and For gas temperature and ambient temperature.

[0078] (3) Pipeline volume effect;

[0079] The pipeline volume effect refers to the delay and buffering effect of the internal space of a pipeline system on gas flow. This effect affects the dynamic response characteristics and stability of a multi-stage compressor system. When the compressor outlet flow rate changes, the internal volume of the pipeline acts like a "gas container," storing or releasing gas, leading to asynchronous flow and pressure responses between the compressor outlet and downstream equipment (such as interstage coolers, the next stage compressor, or gas storage tanks). The pipeline volume effect is described by equations of mass conservation, momentum conservation, and energy conservation, and its gas flow can be represented by the following set of partial differential equations:

[0080] (14)

[0081] (15)

[0082] (16)

[0083] in, For gas density, For flow rate, For pressure, The coefficient of friction, For pipe diameter, For internal energy, Let be the specific enthalpy. For engineering applications, the above equations can usually be simplified by assuming that the gas temperature changes slowly within the pipe and focusing on the mass and momentum transfer processes. Using the lumped parameter method, the pipe system with distributed parameters can be simplified into a lumped volume model, and its dynamic equations can be simplified to:

[0084] (17)

[0085] (18)

[0086] in, For the volume of the pipe, and These are the mass flow rates of the inlet and outlet pipes, respectively. and These are the pipe length and cross-sectional area, respectively. Let be the inertia coefficient of the pipeline. The rate of change of mass flow rate over time. This refers to the absolute pressure at the pipe inlet. This refers to the absolute pressure at the pipe outlet. The flow resistance coefficient is... This refers to the pressure loss during pipeline flow.

[0087] Furthermore, for multi-stage compressors in compressed air energy storage systems, the traditional Greitzer model can be extended to include the inter-stage network volume effect, for the first stage... jThe dimensionless dynamic equation for the interstage piping system between the first-stage compressor and the (j+1)th-stage compressor can be expressed as:

[0088] a. Flow balance equation:

[0089] (19)

[0090] b. Pressure relationship equation:

[0091] (20)

[0092] in, For the dimensionless flow rate of the interstage pipeline, and The first Level and First Dimensionless flow rate of a multistage compressor For the Greitzer B parameters of interstage piping, For the first The dimensionless pressure rise of a multistage compressor For the dimensionless pressure of the interstage pipeline, is the flow resistance coefficient of the pipeline.

[0093] Furthermore, the Greitzer B parameter for interstage piping is defined as follows:

[0094] (twenty one)

[0095] in, For reference speed, For the speed of sound, For the volume of the interstage piping, and These represent the cross-sectional area and length of the interstage pipe, respectively.

[0096] The mechanism by which pipeline volume affects the surge boundary is that the pipeline volume effect alters the surge boundary of a multi-stage compressor system, requiring modifications to traditional surge boundary models to account for this effect. Based on the extended Greitzer model, the system stability incorporating the pipeline volume effect can be evaluated by analyzing the eigenvalues ​​of the system's Jacobian matrix. Considering the pipeline volume effect, the dynamic equations of the surge dynamic boundary correlation model for a multi-stage compressor system can be integrated as follows:

[0097] (twenty two)

[0098] The stability of the system is determined by the eigenvalues ​​of the linearized matrix of the nonlinear system at the equilibrium point:

[0099] (twenty three)

[0100] Here, F is a vector-valued function, which is essentially a mathematical description encapsulating all the nonlinear dynamic laws of the entire system. It maps the current state of the system to the rate of change of the state. Each component of F is a mathematical expression describing the corresponding state variable (e.g., ...). , How the time derivative of (etc.) depends on the current values ​​of all state variables.

[0101] The system is unstable when the real part of an eigenvalue is greater than zero; the dynamic surge boundary corresponds to the critical state where the real part of the eigenvalue crosses zero. The influence of the pipeline volume on the surge boundary is mainly reflected in two aspects:

[0102] 1) Phase lag effect: The interstage piping volume causes a delay in pressure transmission, resulting in inconsistent dynamic responses between the preceding and following compressor stages. This phase lag reduces the stability margin.

[0103] 2) Energy storage effect: The volume of the pipeline can store gas energy, which is released when the flow rate decreases, thus delaying the occurrence of surge. However, it will also absorb energy when the flow rate increases, thus delaying the recovery process.

[0104] Step 3: Input the estimated internal state variables of the compressor into the surge dynamic boundary correlation model to calculate the stability margin;

[0105] Specifically, firstly, through parameter sensitivity analysis, a quantitative relationship between the pipeline network volume and the system stability margin is established, defining the stability margin SM as the relative distance between the current operating point and the surge boundary:

[0106] (twenty four)

[0107] in, For the current dimensionless flow, The dimensionless flow rate at the surge boundary.

[0108] Empirical relationship between pipeline volume and stability margin:

[0109] (25)

[0110] in, This represents the stability margin when there is no pipeline volume effect. , , For system-specific parameters, For the Greitzer parameters of the pipeline network, This is the equivalent pipeline volume.

[0111] Step 4: Based on the stability margin calculation results, execute control decisions. The intelligent controller based on meta-reinforcement learning and spatiotemporal graph neural network calculates the optimal control action and outputs control commands to the actuator. The actuator adjusts the compressor operating parameters to ensure that the system is far away from the surge boundary.

[0112] Specifically, this disclosure proposes a reinforcement learning intelligent control architecture based on a meta-reinforcement learning controller, a spatiotemporal graph neural network predictor, and a hierarchical multi-objective optimizer. It fully integrates multi-timescale prediction and adaptive control strategies. Multi-timescale prediction can analyze surge conditions from different time dimensions, comprehensively capturing the development trend of surge. The adaptive control strategy dynamically adjusts control parameters according to actual operating conditions. This organic combination achieves feedforward suppression and dynamic optimization of surge in multi-stage compressors.

[0113] The reinforcement learning intelligent control architecture adopts a dual-timescale learning mechanism of internal fast parameter adjustment and external slow strategy optimization to solve the problem of insufficient adaptability of traditional reinforcement learning under changing conditions.

[0114] First, in meta-reinforcement learning, the surge control problem is modeled as a partially observable Markov decision process, where the tuples are defined as follows: ,in: It represents the state space, including parameters such as pressure, temperature, flow rate, and speed of each stage of the compressor; It indicates the action space, including control commands such as speed adjustment, guide vane angle change, and bypass valve opening; Indicates the state transition probability; The reward function is carefully designed to balance efficiency and stability. This represents the observation space, which, considering sensor limitations, is usually a subset of the state space; This represents the observation function, which describes the probability distribution of the observed values ​​under a given state.

[0115] The goal is to learn a strategy Maximize cumulative expected return:

[0116] (26)

[0117] in, For strategy parameters, Represents the trajectory. As a discount factor, For time range; The value is a scalar, representing the agent's position at time step. The instant reward value obtained.

[0118] This paper utilizes a meta-learning framework to acquire prior knowledge across tasks, enabling rapid adaptation to new working conditions. The method employs a model-agnostic meta-learning, with the following optimization objective:

[0119] (27)

[0120] in, Indicates the first One task, Indicates task distribution. Indicates in the task loss function on, This is the internal loop learning rate.

[0121] Furthermore, the spatiotemporal graph neural network predictor, composed of a graph convolutional network and a gated recurrent unit, can simultaneously capture the spatial topological relationships and temporal dynamic characteristics of a multi-stage compressor system. This models the multi-stage compressor system as a graph structure. , where nodes This indicates the compressors at each stage and their connecting components, with sides... This represents the connections between them. Each node In time The feature vector includes state variables such as pressure, temperature, and flow rate, denoted as . Graph convolution operations are used to capture spatial dependencies between nodes, and their basic form is:

[0122] (28)

[0123] in, Indicates the first Layer nodes The hidden state, Represents a node The neighborhood group, The normalization constant is and For learnable parameters, The activation function is used. To capture temporal dynamics, the graph convolutional network is combined with a gated recurrent unit to form a spatiotemporal graph convolutional block:

[0124] (29)

[0125] (30)

[0126] (31)

[0127] (32)

[0128] in, This represents the graph convolution operation. and These represent updating the door and resetting the door, respectively. This indicates element-wise multiplication.

[0129] Furthermore, the hierarchical multi-objective optimizer strategy addresses the trade-off between stability and efficiency in compressor control by decomposing the multi-objective optimization problem into a stability priority layer and an efficiency optimization layer, achieving a balance between the two objectives through dynamic weight adjustment.

[0130] The stability-first layer is the bottom-level control strategy. Its main objective is to prevent surge and ensure the safe operation of the system. The reward function for this layer is designed as follows:

[0131] (33)

[0132] in, This represents the stability margin of the j-th stage compressor. For critical stability margin, Indicates the rate of change of flow rate. and These are the weighting coefficients.

[0133] As one example, when the system detects that the stability margin is lower than the safety threshold, the stability priority layer will take over control and adopt a hierarchical control strategy:

[0134] Level 1 surge: Adjust the compressor speed and the opening of the auxiliary throttling element;

[0135] Secondary surge: Open the bypass valve to rapidly reduce the pressure ratio.

[0136] Furthermore, the efficiency optimization layer is a high-level control strategy that optimizes the compressor system efficiency while ensuring system stability. The reward function for this layer is:

[0137] (34)

[0138] in, Indicates the overall efficiency of the system. This represents the power consumption of the j-th stage compressor. Indicates the range of change in the control action. , and These are the weighting coefficients.

[0139] The efficiency optimization layer dynamically adjusts the weights of stability and efficiency based on the system state through an adaptive weight adjustment mechanism.

[0140] (35)

[0141] in, and These are the minimum and maximum weights. For the current stability margin, and To adjust the parameters, This is the sigmoid function.

[0142] As one embodiment, the reinforcement learning intelligent control method disclosed herein is implemented through a combination of offline pre-training and online fine-tuning, which not only ensures initial performance but also endows the system with the ability to continuously optimize. Its implementation process includes system initialization and pre-training, online learning and adaptation.

[0143] Furthermore, the system initialization and pre-training utilize historical operating data and simulation data to construct a training dataset, which covers various operating conditions and boundary situations. The pre-training process consists of three steps:

[0144] (1) Pre-training of the prediction model: Using supervised learning methods and historical data, a spatiotemporal graph neural network predictor is trained to accurately predict the dynamic behavior of a multi-stage compressor system. The loss function is defined as:

[0145] (36)

[0146] in, and Representing nodes respectively In time The predicted value and the actual value, Indicates model parameters, is the regularization coefficient.

[0147] (2) Meta-training of the controller: The meta-reinforcement learning controller is trained in a simulated environment using reinforcement learning algorithms, and the proximal policy optimization method is used to ensure training stability.

[0148] (37)

[0149] in, This is the estimated value of the dominance function. These are the trimming parameters.

[0150] (3) Value function modeling: Simultaneously train the value function network to provide accurate state value assessment for policy optimization:

[0151] (38)

[0152] in, For value network output, For target value, For about time steps The expected value.

[0153] Furthermore, after offline pre-training, the model enters the online learning and adaptation phase, continuously improving its performance through real-time data. The online learning strategy combines experience replay and model prediction guidance.

[0154] (1) The experience playback buffer pool stores the system's operating experience, and samples it according to priority for training;

[0155] (2) Model prediction guidance utilizes well-learned system models to generate simulation experience, accelerating the learning process;

[0156] (3) Safety constraints ensure that the exploration process will not cause surge and that the exploration is carried out in a controlled manner within the safety boundary.

[0157] When the system detects changes in operating conditions or performance degradation, it initiates an internal loop for rapid adjustment:

[0158] (39)

[0159] in, In response to the current task Adjusted parameters Internal learning rate This is the loss function for the current task.

[0160] Example 2

[0161] In one embodiment of this disclosure, taking the dynamic correlation surge suppression of multi-stage compressors in a 300 MW compressed air energy storage power station as an example, an implementation case of the dynamic correlation surge suppression method for multi-stage compressors in a compressed air energy storage power station disclosed in this disclosure is given. The specific process is as follows:

[0162] Step (1): Establish the overall system architecture;

[0163] A dynamic correlation surge suppression method framework is constructed, consisting of a multi-level coupled model architecture, a surge boundary dynamic correlation mechanism, and intelligent control execution. Through a high-precision sensor network, a real-time data processing unit, and a control actuator, coordinated monitoring and suppression of multi-stage compressors are achieved.

[0164] Furthermore, the architecture includes: a sensor layer, where pressure, temperature, flow, and vibration sensors are placed at the inlet and outlet of each compressor stage, with a sampling frequency of 1kHz, covering all key points; a data processing layer, employing an industrial computer cluster to run multi-level coupled models and state estimation algorithms, calculating stability margins in real time; a control layer, based on meta-reinforcement learning and spatiotemporal graph neural networks, which outputs control commands to the speed regulator, guide vane actuator, and bypass valve; and an execution layer, where high-speed actuators adjust compressor operating parameters to ensure the system stays away from surge boundaries. The overall workflow of this disclosure is: data acquisition → state estimation → boundary prediction → control decision → execution and verification, forming a closed-loop control.

[0165] Step (2): Construct a multi-level coupling model;

[0166] This embodiment targets a 300 MW multistage compressor system. The multistage coupled model architecture couples the compressor thermodynamic model, surge dynamic model, and pipeline volume effect to accurately describe interstage interactions and dynamic characteristics. This embodiment employs a 4-stage series compressor (j=1 to 4), with each stage driven by an electric motor and heat exchangers installed between stages for cooling.

[0167] Based on equations (1)-(5) of Example 1, the mass flow rate, pressure ratio, and outlet temperature of each stage compressor are calculated using the following set of equations:

[0168] (1) Mass flow rate: ;

[0169] (2) Pressure ratio: ;

[0170] (3) Outlet temperature: ;

[0171] (4) Inter-level transfer: ;

[0172] The parameters are based on a 300 MW-level design: (Specific heat ratio of air) For adiabatic efficiency (0.85-0.92), and Interstage loss (determined through CFD simulation, typical value) MPa, K). The model is updated using real-time sensor data, such as the first-stage inlet pressure. MPa, Level 4 outlet pressure MPa.

[0173] Furthermore, firstly, a thermodynamic model of the heat exchanger is established:

[0174] Interstage coolers have a significant impact on the gas state. Based on equations (6)-(7), the first stage cooler... j The dynamic process of a multi-stage heat exchanger is as follows:

[0175] (1) Outlet temperature: ;

[0176] (2) Dynamics of heat exchange: ;

[0177] in, (Time constant) (Heat exchanger efficiency) (Cooling water inlet temperature). This model uses real-time temperature data correction to ensure effective interstage cooling and prevent surge caused by excessive temperature.

[0178] Furthermore, a pipeline network volume effect model is established:

[0179] In a 300 MW system, the inter-stage pipe length can reach 50-100 m, and the volumetric effect cannot be ignored. Based on equations (14)-(18), the simplified pipe network model is as follows:

[0180] Conservation of mass: ;

[0181] Conservation of momentum: ;

[0182] Among them, the pipe volume Flow resistance coefficient This model is used to simulate the delay of interstage pressure fluctuations; for example, when the flow rate of the first-stage compressor changes, the inlet pressure response of the second stage exhibits a lag of 0.1–0.5 s. The multi-stage coupled model is validated through three-dimensional CFD numerical simulation. Local flow field analysis reveals the interstage eddies and pressure distribution, allowing for the correction of model parameters. The model output is used for real-time stability assessment and provides input for surge boundary prediction.

[0183] Furthermore, a dynamic correlation model of the surge boundary is constructed:

[0184] The surge boundary dynamic correlation mechanism is based on an improved Greitzer model. By tracking the dynamic parameters of the multi-stage compressor in real time, a mathematical relationship between surge boundary prediction and stability assessment is established based on equations (8)-(10). For a 300 MW system, the traditional Greitzer model is extended to equation:

[0185] The surge boundary dynamic correlation model is based on the improved Greitzer model. By tracking the dynamic parameters of the multi-stage compressor in real time, the mathematical relationship between surge boundary prediction and stability assessment is established based on equations (8)-(10). For a 300 MW system, the traditional Greitzer model is extended to equation:

[0186] Traffic dynamics: ;

[0187] Stress dynamics: ;

[0188] Greitzer parameters: ;

[0189] in, and For dimensionless flow rate and pressure rise coefficient, and For equivalent flow area and length (based on design values, e.g.) , ), To accumulate volume (approximately 20 cubic meters per level) ), Rotor speed (100) ), For the speed of sound (340 ), The flow resistance coefficient is (0.1). The dynamic surge boundary is determined by the eigenvalues ​​of the Jacobian matrix in equation (11):

[0190]

[0191] In actual implementation, the system calculates the feature value every 0.1 seconds. A warning is triggered at any time. The stability margin SM is calculated using the following equation:

[0192]

[0193] in, For the current dimensionless flow, The surge boundary flow rate is (fitted from historical data). The network volume effect is corrected for SM using the following equation:

[0194]

[0195] For a 300 MW system, empirical parameter values ​​are: , , , , , This mechanism ensures accurate predictions by dynamically adjusting the SM (Search Engine Controller) based on real-time data updates, such as changes in gas storage pressure.

[0196] Step (3): The intelligent controller based on meta-reinforcement learning and spatiotemporal graph neural network calculates the optimal control action and outputs control commands to the actuator; the actuator adjusts the compressor operating parameters to ensure that the system is far away from the surge boundary.

[0197] The intelligent control method consists of a meta-reinforcement learning controller, a spatiotemporal graph neural network predictor, and a hierarchical multi-objective optimizer, and is customized for a 300 MW system.

[0198] a. Meta-reinforcement learning controller: The surge control problem is modeled as a partially observable Markov decision process (POMDP), where the tuple is defined as... .

[0199] state space Parameters include pressure (0.1-10 MPa), temperature (20-200°C), flow rate (100-200 kg / s), and rotational speed (5000-10000 rpm).

[0200] Action space This includes speed adjustment (±100 rpm), guide vane angle change (±5 degrees), and bypass valve opening change (±10%).

[0201] reward function The design aims to balance efficiency and stability, based on equations (33) and (34). The objective is to achieve the optimal strategy. Maximize cumulative expected return Meta-learning employs Model-Agnostic Meta-Learning (MAML), with the optimization objective being the equation:

[0202]

[0203] In the formula, the task The loss function represents different operating conditions (such as startup and load changes). To control error, the internal loop learning rate is adjusted. External loop learning rate The controller adapts to the varied operating conditions of a 300MW system through offline pre-training and online fine-tuning.

[0204] b. Spatiotemporal Graph Neural Network Predictor: Models a multi-stage compressor system as a graph structure. , where nodes This represents a 4-stage compressor and its connecting components (10 nodes in total), with edges... This represents the pipe connection. Each node's feature vector includes state variables such as pressure, temperature, and flow rate. The predictor is composed of a graph convolutional network (GCN) and a gated recurrent unit (GRU), based on equations (28)-(32):

[0205] Graph convolution operations:

[0206] Spatiotemporal blocks: Capture temporal dynamics through update and reset gates. The predictor performs multi-timescale predictions: short-term predictions (1-5 seconds) are used for early surge warning, and long-term predictions (10-30 seconds) are used for stability assessment. The loss function is the equation:

[0207]

[0208] In the formula, The predictor is pre-trained on historical data and achieves an accuracy rate of over 95%.

[0209] c. Hierarchical multi-objective optimizer: Decomposes the control problem into a stability priority layer and an efficiency optimization layer.

[0210] Stability Priority Layer: The reward function is the equation:

[0211]

[0212] In the formula, For the first j Level stability margin, , , When SM < 5%, graded control is triggered: Level 1 surge, adjusting compressor speed and guide vane angle; Level 2 surge, opening the bypass valve (opening degree 50-100%), rapidly reducing the pressure ratio.

[0213] Efficiency optimization layer: The reward function is the equation:

[0214]

[0215] In the formula, For system efficiency (target >70%). For power consumption, , , The dynamic adjustment of weights is based on the equation:

[0216]

[0217] In the formula, For the sigmoid function, , The optimizer ensures maximum efficiency while maintaining stability.

[0218] As one embodiment, the dynamic correlation surge suppression model of this disclosure adopts a cooperative working mode, which uses closed-loop feedback control, and the specific process is as follows:

[0219] a. Data Acquisition. High-precision sensors, including piezoelectric pressure sensors (accuracy ±0.1%), thermocouple temperature sensors (accuracy ±0.5°C), vortex flow meters (accuracy ±0.2%), and acceleration vibration sensors, are deployed at the inlet and outlet of the 4-stage compressor. The sampling frequency is 1 kHz, and data is transmitted to the central processing unit via a fiber optic network, processing 10,000 data points per second to capture high-frequency characteristics of surge precursors (such as pressure fluctuation frequencies >100 Hz).

[0220] b. State Estimation. The Extended Kalman Filter (EKF) algorithm is used to estimate the compressor's internal state variables based on sensor measurements. The state vector includes mass flow rate, pressure fluctuation, and temperature for each stage. The EKF model is based on multi-stage coupled equations, and the covariance of process noise and observation noise is experimentally calibrated. For example, estimating the mass flow rate of stage 2. The error compared to the measured value is less than 1%.

[0221] c. Boundary Prediction. The estimated state variables are input into the dynamic boundary correlation model to calculate the distance (stability margin SM) between the current operating point and the surge boundary, and to predict the stability evolution trend over the next 30 seconds. The prediction model uses a spatiotemporal graph neural network to output the SM curve and warning level. When SM < 10%, the system issues a yellow warning; when SM < 5%, a red warning is issued.

[0222] d. Control Decision. When the stability margin falls below the threshold (SM < 10%), the intelligent controller is activated. The controller calculates the optimal control action based on meta-reinforcement learning and hierarchical multi-objective optimization. Control actions include: - Adjusting the compressor speed (via the frequency converter, variation range ±5%) - Adjusting the guide vane angle (range 0-90 degrees) - Adjusting the bypass valve opening (range 0-100%). The decision cycle is 0.1 seconds to ensure rapid response. For example, when a decrease in the third-stage SM is detected, the controller prioritizes adjusting the third-stage guide vane angle to increase flow and avoid surge.

[0223] e. Execution and Verification. Control commands are implemented through high-speed actuators, including electric actuators (response time <0.1s) and pneumatic bypass valves (response time <0.5s). The system continuously monitors the control effect and verifies it through real-time SM (Stable Motion) and efficiency indicators. If the SM does not improve after control, the system adjusts control parameters (e.g., increasing the bypass valve opening) to achieve adaptive optimization. Simultaneously, a gas storage thermodynamic model is used to predict the impact of pressure changes on stability, such as when the gas storage pressure... When the pressure increases from 14 MPa to 16 MPa, the compressor outlet pressure is adjusted accordingly to avoid the operating point from approaching the surge boundary.

[0224] like Figure 2The diagram shows the effect of surge suppression on the operating point migration in the compressor characteristic diagram. Without specific control measures, the operating point will move closer to the surge boundary and enter the surge zone when the operating conditions change. However, with dynamic correlation control, the system can push the operating point back to the safe area in time when it approaches the boundary, thereby achieving the effect of surge suppression.

[0225] The dynamic correlation surge suppression system and control method for multi-stage compressors in compressed air energy storage power stations disclosed herein offer the following advantages over existing technologies: By constructing a multi-stage coupled model that integrates thermodynamics, surge dynamics, and pipeline volume effects, and introducing a surge boundary dynamic correlation mechanism based on an improved Greitzer model, the system achieves accurate description of the complex dynamic characteristics of multi-stage compressors and real-time, high-precision prediction of surge boundaries. This fundamentally overcomes the shortcomings of traditional single models, such as poor adaptability and inaccurate prediction, and significantly improves the system's stability margin. Furthermore, it innovatively employs a fusion of meta-reinforcement learning and spatiotemporal graph neural networks. The intelligent control method of the network, combined with a hierarchical multi-objective optimization strategy, realizes feedforward suppression and dynamic optimization control of strongly nonlinear, time-varying coupled systems. This method has self-learning and rapid adaptation capabilities, and can effectively cope with various operating conditions. While ensuring stability, it also takes into account the system's operating efficiency, solving the problem of insufficient performance of traditional control methods such as PID under complex dynamics. The system forms a collaborative suppression closed loop from accurate perception and intelligent prediction to optimized decision-making and efficient execution, which can significantly reduce the probability and intensity of surge, and effectively ensure the safe, stable and efficient operation of the core equipment of large compressed air energy storage power stations.

[0226] Example 3

[0227] One embodiment of this disclosure provides a multi-stage compression dynamic correlation surge suppression system for a compressed air energy storage power station, comprising:

[0228] The parameter acquisition module is used to acquire real-time operating parameters of a multi-stage compressor system.

[0229] The state estimation module is used to estimate the internal state variables of the compressor based on real-time operating parameters using the extended Kalman filter algorithm.

[0230] The evolution prediction module is used to input the estimated internal state variables of the compressor into the surge dynamic boundary correlation model, calculate the stability margin, and predict the stability evolution trend in the future.

[0231] The intelligent control module is used to execute control decisions based on the stability margin calculation results. The intelligent controller, based on meta-reinforcement learning and spatiotemporal graph neural network, calculates the optimal control action and outputs control commands to the actuator. The actuator adjusts the compressor operating parameters to ensure that the system stays away from the surge boundary.

[0232] The compressor thermodynamic model, surge dynamic model, and pipeline volume effect are coupled, and the local flow field is simulated in detail using three-dimensional CFD numerical simulation method to construct a multi-level coupled model. Based on the multi-level coupled model, the inter-level coupling effect is considered to obtain the surge dynamic boundary correlation model.

[0233] Example 4

[0234] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power stations.

[0235] Example 5

[0236] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When these computer instructions are executed by a processor, they implement the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power stations.

[0237] Example 6

[0238] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the multi-stage compression dynamic correlation surge suppression method for the compressed air energy storage power station.

[0239] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0240] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0241] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. 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 this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for suppressing surge caused by multi-stage compression dynamic correlation in compressed air energy storage power stations, characterized in that, include: Obtain real-time operating parameters of a multi-stage compressor system; Based on real-time operating parameters, the extended Kalman filter algorithm is used to estimate the internal state variables of the compressor. The estimated internal state variables of the compressor are input into the surge dynamic boundary correlation model to calculate the stability margin; Based on the stability margin calculation results, control decisions are made. The intelligent controller based on meta-reinforcement learning and spatiotemporal graph neural network calculates the optimal control action and outputs control commands to the actuator. The actuator adjusts the compressor operating parameters to ensure that the system stays away from the surge boundary. The compressor thermodynamic model, surge dynamic model, and pipeline volume effect are coupled, and the local flow field is simulated in detail using three-dimensional CFD numerical simulation method to construct a multi-level coupled model. Based on the multi-level coupled model, the inter-stage coupling effect is considered to obtain the surge dynamic boundary correlation model. The multi-stage coupling model couples the compressor thermodynamic model, surge dynamic model, and pipeline volume effect. Considering the inter-stage matching characteristics of multi-stage series compressors, for a system with N-stage compressors, the outlet state of each stage is the inlet boundary condition of the next stage, thus constructing the coupling relationship. The compressor thermodynamic model considers the impact of interstage cooling on the gas state and compressor performance, and is constructed based on the principles of energy conservation and mass conservation. The surge dynamic model establishes a mathematical relationship between surge boundary prediction and stability assessment by tracking the dynamic parameters of a multi-stage compressor in real time. This model is based on an improvement of the Greitzer model and takes into account the inter-stage interaction and volume effect of the multi-stage compressor. The intelligent controller based on meta-reinforcement learning and spatiotemporal graph neural network integrates multi-timescale prediction and adaptive control strategies. Multi-timescale prediction analyzes surge from different time dimensions to comprehensively capture the development trend of surge. The adaptive control strategy dynamically adjusts control parameters according to the actual operating conditions. The combination of the two achieves feedforward suppression and dynamic optimization of surge in multi-stage compressors.

2. The method for suppressing multi-stage compression dynamic correlation surge in a compressed air energy storage power station as described in claim 1, characterized in that, The acquisition of real-time operating parameters of the multi-stage compressor system includes: High-precision sensors, including piezoelectric pressure sensors, thermocouple temperature sensors, vortex flow meters, and acceleration vibration sensors, are arranged at the inlet and outlet of the four-stage compressor. The collected data is transmitted to the central processing unit via an optical fiber network at a set sampling frequency to capture high-frequency operating parameters of surge precursors, including pressure, temperature, flow rate and vibration data.

3. The method for suppressing multi-stage compression dynamic correlation surge in a compressed air energy storage power station as described in claim 1, characterized in that, The estimation of the compressor's internal state variables based on real-time operating parameters using the extended Kalman filter algorithm includes: Based on the improved Greitzer model, a mathematical relationship between surge boundary prediction and stability assessment is established by tracking the real-time operating parameters of a multi-stage compressor. The Greitzer model is extended, and the extended Kalman filter algorithm is used to estimate the internal state variables of the compressor, which include mass flow rate, pressure fluctuation and temperature at each stage.

4. A multi-stage compression dynamic correlation surge suppression system for compressed air energy storage power stations, specifically implementing the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power stations as described in any one of claims 1-3, characterized in that, include: The parameter acquisition module is used to acquire real-time operating parameters of a multi-stage compressor system. The state estimation module is used to estimate the internal state variables of the compressor based on real-time operating parameters using the extended Kalman filter algorithm. The evolution prediction module is used to input the estimated internal state variables of the compressor into the surge dynamic boundary correlation model to calculate the stability margin; The intelligent control module is used to execute control decisions based on the stability margin calculation results. The intelligent controller, based on meta-reinforcement learning and spatiotemporal graph neural network, calculates the optimal control action and outputs control commands to the actuator. The actuator adjusts the compressor operating parameters to ensure that the system stays away from the surge boundary. The compressor thermodynamic model, surge dynamic model, and pipeline volume effect are coupled, and the local flow field is simulated in detail using three-dimensional CFD numerical simulation method to construct a multi-level coupled model. Based on the multi-level coupled model, the inter-level coupling effect is considered to obtain the surge dynamic boundary correlation model.

5. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power stations as described in any one of claims 1-3.

6. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power stations as described in any one of claims 1-3.

7. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the multi-stage compression dynamic correlation surge suppression method for compressed air energy storage power stations as described in any one of claims 1-3.

Images