CN-121979316-A - LNG power ship storage tank pressure data driving control method based on LSTM trend prediction

CN121979316ACN 121979316 ACN121979316 ACN 121979316ACN-121979316-A

Abstract

The invention discloses an LNG power ship storage tank pressure data driving control method based on LSTM trend prediction, and relates to the technical field of ship and ocean engineering automation control. Firstly, establishing a dynamic model of an executing mechanism of a BOG processing system of an LNG power ship, secondly, establishing an LSTM pressure time sequence prediction model, applying an LSTM advanced prediction value to an improved embedded tight format model-free self-adaptive control law, overcoming the time lag of the system, and finally realizing pressure control by dynamically adjusting generalized rotating speed instructions to cooperatively distribute the rotating speed of a compressor and the opening of a regulating valve. The invention solves the problems of severe pressure fluctuation of the storage tank, switching oscillation of the actuating mechanism and large overshoot of the actuating mechanism caused by model parameter mismatch and response lag of the LNG power ship under severe sea conditions, and effectively ensures the safety and reliability of pressure control of the LNG storage tank.

Inventors

YAO WENLONG
CHENG PENGFEI
YUE YAOBIN
FENG JIANLIANG
LIU YUCHUAN
TIAN SHUO

Assignees

青岛科技大学

Dates

Publication Date: 20260505
Application Date: 20260210

Claims (5)

1. An LNG power ship storage tank pressure data drive control method for LSTM trend prediction, the method comprising the steps of: S1, establishing a dynamic model of an executing mechanism of a BOG processing system of an LNG power ship; S2, constructing an LSTM pressure time sequence prediction model fused with sea condition disturbance characteristics; S3, designing a model-free self-adaptive controller embedded with prediction information, and calculating a generalized rotating speed control law by estimating pseudo partial derivative on line; and S4, establishing a dynamic cooperative allocation strategy and closed-loop feedback of the compressor and the regulating valve based on the generalized rotating speed.
2. The method for driving and controlling the pressure data of the tank of the LNG power ship according to claim 1, wherein in step S1, the establishing a dynamic model of the BOG processing system of the LNG power ship specifically includes: (1) Fitting and establishing the rotating speed of the compressor motor And exhaust gas mass flow rate Nonlinear function model between: ; In the formula, Is a fitting coefficient, and determines the minimum anti-surge rotating speed boundary of the compressor under the current working condition And maximum allowable rotation speed ; (2) Based on ISA standard flow equation, establishing the opening of the regulating valve And mass flow through the valve Mapping model between: ; In the formula, As a flow coefficient of the water, the water is mixed with water, As a function of the inherent flow characteristics of the valve, Is the coefficient of expansion of the gas, and is, For the inlet gas density to be the same, Is the pressure difference between the front and the back of the valve.
3. The method for driving and controlling the pressure data of the tank of the LNG power ship according to claim 1, wherein in step S2, the constructing the LSTM pressure time sequence prediction model that fuses the sea state disturbance characteristics specifically includes: (1) Constructing a composite input feature vector comprising pressure observations and first order differences To implicitly characterize sea state disturbances: ; In the formula, For the length of the sliding window, Representation of First order difference of time of day, introduction Is physically significant in that under steady sea conditions Approaching a constant, but under severe sea conditions The high-frequency oscillation distribution is presented, and the LSTM network can sense the current sea state disturbance level by utilizing the characteristics; (2) Will be Input LSTM network, utilizing forget gate Input door Cell status And an output door Extracting time sequence characteristics, wherein the calculation process is as follows: ; ; ; ; In the formula, The function is activated for Sigmoid, For the hyperbolic tangent activation function, The Hadamard product of the matrix is represented, And Respectively a weight matrix and a bias term; (3) Pure lag compensation and lead prediction: outputting future by training converged LSTM model Predicted value of step pressure The predicted value identifies a pressure jump trend in advance based on the "texture feature" of the historical pressure curve, wherein the predicted step size Is set to be greater than the pure lag time of the system control loop Thereby providing time advance for the action of the subsequent controller and effectively compensating the response delay of the physical system.
4. The method for driving and controlling the pressure data of the storage tank of the LNG power ship predicted by the LSTM trend according to claim 1, wherein the step S3 is designed to embed a model-free self-adaptive controller of the predicted information, and the generalized rotation speed control law is calculated by estimating the pseudo partial derivative on line specifically comprises the following steps: (1) Pressure variation of nonlinear system Linearization is expressed as: ; In the formula, Indicating the pressure change amount at the next time; A control increment representing the current time; Defined as the pseudo-partial derivative; (2) Introducing pseudo-partial derivatives Dynamic linearization of nonlinear systems, construction of which involves Is a function of the estimation criteria: ; (3) The alignment then functions partial derivative and utilizes projection algorithm to obtain iterative update formula of pseudo partial derivative: ; In the formula, As the weight factor of the weight factor, Is a step factor; (4) Constructing an objective function comprising future prediction bias and control input constraints: ; (5) Deriving final generalized rotation speed control instruction by combining linearization model : ; In the formula, The pressure is set for the target and, For the leading predicted value of the LSTM output, In order to predict the feedback gain factor, Is a step factor; Is a weight factor.
5. The method for controlling the driving of the tank pressure data of the LNG power ship according to claim 1, wherein the dynamic cooperative allocation strategy in step S4 specifically includes logic for determining and executing high-load and low-load conditions: (1) Compressor-dominant distribution strategy for high load zone: When (when) When the system is judged to be in a high-load working condition, the bypass reflux is forcibly closed at the moment, and an opening instruction of the regulating valve is set And directly mapping the generalized rotating speed command into a physical rotating speed command of the compressor Applying the maximum rotational speed at the same time Is limited by saturation: ; In the formula, Is the physical rotation speed of the compressor; is a generalized rotating speed instruction; is the minimum anti-surge rotational speed; Is the maximum rotation speed; (2) Collaborative compensation allocation strategy for low load area: When (when) When the system is in the low-load working condition, the physical rotating speed instruction of the compressor is forcedly locked to prevent the compressor from surging Simultaneously, calculating the excess mass flow between the actual exhaust gas flow of the compressor at the minimum rotation speed and the flow theoretically required by the system : ; Further, based on the regulating valve flow rate model in the step S1, the regulating valve opening command of the excessive mass flow rate is calculated by an inverse function method Setting the regulating valve to have equal percentage flow characteristic The opening instruction calculation formula is: ; (3) Global closed loop feedback mechanism: the calculation is carried out And (3) with Synchronously transmitting to an actuating mechanism, and collecting the pressure of the storage tank after the execution in real time And the first-order difference is used as a new state vector to be fed back to the LSTM network of the step S2 and the MFAC controller of the step S3, so that the rolling time domain closed-loop control of data driving is completed.

Description

LNG power ship storage tank pressure data driving control method based on LSTM trend prediction Technical Field The invention relates to the technical field of automatic control of ships and ocean engineering, in particular to a pressure data driving cooperative control method suitable for a BOG processing system of an LNG power ship under complex sea conditions. Background Liquefied Natural Gas (LNG) plays a critical role in global energy structure regulation and "two carbon" target implementation as a clean, efficient energy source. LNG powered vessels are driving the global shipping industry to a low carbon future "green heart" where safety and economy are of great concern. During LNG storage and ocean going, LNG in the storage tank continuously evaporates to produce boil-off gas (BOG) due to external heat intrusion and hull movement disturbance, resulting in an increase in pressure in the tank. If effective control is not performed, the too high pressure triggers the safety valve to discharge, so that huge economic loss is caused, the environment is polluted, and even the safety of the ship structure is threatened. Therefore, aiming at system time lag and disturbance under severe sea conditions, the construction of a data driving cooperative control strategy based on LSTM time sequence prediction has important engineering value, the feedforward compensation is carried out on pressure fluctuation by introducing a prediction mechanism, and the control authority of a compressor and a regulating valve is dynamically and cooperatively distributed so as to realize the control precision and the robustness of the pressure of the LNG storage tank. Although the existing BOG reliquefaction and combustion treatment system has been widely used for LNG power ships, the pressure control performance thereof still faces serious challenges when facing the deep open sea complex sea conditions. The conventional LNG storage tank pressure control is designed under static or stable sea conditions, and the LNG power ship is often subjected to severe sea condition interference such as stormy waves and currents in the course of voyage, the severe shaking of the ship body can induce the liquid cargo in the tank to generate severe shaking, the sudden change of the gas-liquid contact area leads to the explosive and nonlinear increase of the BOG generation rate, and if the control algorithm cannot make quick response, the tank pressure stability and even the safety valve jump are very easy to be difficult to maintain due to system time lag. Therefore, the pressure of the LNG storage tank is predictably and synergistically controlled under the complex sea condition, so that the BOG treatment system can be stably transited and pressure fluctuation can be restrained, and the method is a difficult problem to be solved in the current ship automation field. Pressure control of a ship BOG system is generally based on monitoring and adjusting thermodynamic processes, but because of the complex mechanism of gas-liquid phase in a storage tank, an accurate mathematical model is difficult to build. In early stage Shin proposed a compressor operation optimization strategy based on classical PID, balancing tank pressure by adjusting compressor load, but in a large hysteresis system, overshoot is very easy to occur under operating condition disturbance due to lack of predictive ability of PID to pressure trend. In order to solve the problem of model uncertainty, ghaemi builds a detailed dynamic thermodynamic model of the BOG reliquefaction system, designs a model-based predictive control scheme, and utilizes an equation to solve the predicted pressure evolution. However, the method is highly dependent on accurate physical parameters, and when the LNG power ship is subjected to sloshing to cause the abrupt change of the gas-liquid contact area, the mechanism model is extremely easy to mismatch, so that the control performance is obviously reduced. Aiming at dynamic process control, liu Yan designs a composite control strategy combining variable frequency regulation and cascade-branch control, which remarkably overcomes the defect of response lag of single variable frequency regulation when dealing with large load fluctuation, but lacks prospective prediction of multivariable coupling characteristics of an internal flow field of a compressor. In recent years, data driving methods are increasingly emerging. The university of Jiangsu ocean Zhang Huixia proposes a hybrid approach that combines CFD data and machine learning to predict in real time the temperature change in LNG tanks to optimize thermal management. However, most of the methods do not incorporate sea state information into a control closed loop, and the disturbance influence of the ship body sloshing on the BOG generation rate under severe sea conditions is ignored. Meanwhile, the traditional strategy lacks an advanced prediction mechanism for the pressure evolution trend, the problem of