CN-121978887-A - Temperature pressure PID dynamic response optimization method and system for liquid hydrogen storage tank

CN121978887ACN 121978887 ACN121978887 ACN 121978887ACN-121978887-A

Abstract

The invention discloses a temperature pressure PID dynamic response optimization method and system of a liquid hydrogen storage tank, comprising the steps of collecting key parameter historical data in the operation of the liquid hydrogen storage tank, preprocessing, constructing a multivariable dynamic response prediction model, training and verifying based on the key parameter historical data to obtain a trained optimal prediction model, simulating temperature pressure response processes under different PID control parameter combinations based on the optimal prediction model to obtain performance indexes corresponding to each PID control parameter combination, taking PID control parameters as actions of reinforcement learning agents, taking weighted values of the dynamic response performance indexes as rewards, simulating an environment by using the optimal prediction model, enabling the agents to iteratively execute the actions, obtain rewards, learn the optimal PID control parameter combination, deploying the optimal PID control parameter combination into a control system, collecting the operation data and the real-time performance indexes of the liquid hydrogen storage tank in real time, and performing response optimization based on the real-time performance indexes.

Inventors

ZHANG YANSONG
ZHAO LIANG
GAO YONGZHENG
ZHANG BIN

Assignees

北京科泰克科技有限责任公司

Dates

Publication Date: 20260505
Application Date: 20260108

Claims (12)

1. A method for optimizing the temperature-pressure PID dynamic response of a liquid hydrogen storage tank, the method comprising: Collecting key parameter historical data in the operation of a liquid hydrogen storage tank, and preprocessing the key parameter historical data, wherein the key parameter historical data comprises temperature, pressure, refrigerating system input quantity and air release valve opening; constructing a multivariable dynamic response prediction model based on a long-term and short-term memory network, and training and verifying based on the preprocessed key parameter historical data to obtain a trained optimal prediction model; based on the optimal prediction model, simulating temperature and pressure response processes under different PID control parameter combinations, obtaining performance indexes corresponding to each PID control parameter combination, and establishing a mapping data set of the PID control parameter combinations and the performance indexes; Taking PID control parameters as actions of reinforcement learning agents, taking dynamic response performance index weighted values as rewards, simulating an environment by using the optimal prediction model, enabling the agents to iteratively execute the actions, acquire the rewards, and learning optimal PID control parameter combinations; And the optimal PID control parameters are combined and deployed into a control system, operation data and real-time performance indexes of the liquid hydrogen storage tank are collected in real time, and response optimization is performed based on the real-time performance indexes.
2. The method of claim 1, wherein preprocessing the key parameter history data comprises: for each point in time in the key parameter history data Corresponding four-dimensional data points Performing missing value filling processing through the linear difference value, denoising and normalizing the data sequence of each dimension, In order to be able to determine the temperature, In the case of a pressure force, the pressure, For the input quantity of the refrigerating system, Is the opening degree of the air release valve.
3. The method of claim 1, wherein the multivariate dynamic response prediction model has an input layer dimension of 3 and an input vector of 3 The output layer is designed as a full connection layer, the dimension of the output layer is 2, and the output vector is Hidden layer dimension Initializing to super-parameters, defining hidden states And cell status Introducing splice vectors As a gating computation basis, an LSTM state update mechanism is set as follows: , , Wherein, the A temperature value representing a current time step; representing a pressure value; Representing a control input signal; for a single time step of temperature change in the future, The amount of pressure change for a single time step in the future; for the sigmoid activation function, Representing element-level multiplication, predicting variation , And For training parameters, all weight matrices are initialized with Xavier and bias terms are initialized to zero.
4. The method of claim 1, wherein simulating temperature-pressure response processes under different PID control parameter combinations based on the optimal predictive model, obtaining performance metrics corresponding to each PID control parameter combination, and establishing a mapping dataset of PID control parameter combinations and performance metrics, comprises: Determining parameter space with PID controller, including proportional gain Integral gain Differential gain To generate the total number of all combined points , wherein, 、 And Sampling points for each dimension; Controlling parameter combinations for each PID Executing closed loop system simulation, initializing the system state to zero, and applying the reference signal Determining an error signal for a unit step input Determining a controller output The method comprises the following steps: Calling the optimal prediction model, and inputting And the current system state, solving the model output by a numerical integration method I.e. temperature or pressure response curve, based on which the simulation process is continued until the system reaches steady state Calculating an overshoot OS; Wherein, the , Wherein, the Is a steady state value; In order to adjust the time period of the process, Satisfy the following requirements forall Searching a minimum time point through iteration; Calculating steady state errors as follows: , ; Recording all index values as data points To determine a mapping dataset of PID control parameter combinations and performance metrics.
5. The method of claim 1, wherein the method sets an agent's motion vector when defining the motion space in learning the optimal PID control parameter combination Representing PID controller parameters, designing an action space into a continuous range, fixing an action dimension into a three-dimensional vector, directly mapping the three-dimensional vector to PID structural parameters, setting a prediction model environment for simulating control system dynamics, and inputting the model into a current state in a discrete state space mode And actions Outputting the predicted next state And control response The bonus function is designed to be weighted based on dynamic response performance indexes, wherein the performance indexes comprise integral square error ISE and maximum overshoot And adjusting the time Rewarding (rewarding) Defined as a negative weighted sum, targeting a minimization indicator, the formula is: , Wherein, the 、、 Is the weight; , is a reference input and a predicted output Is a deviation of (2).
6. The method of claim 1, wherein performing response optimization based on the real-time performance metrics comprises: When the integral square error of the real-time performance index is larger than a preset threshold, judging that the performance is insufficient to trigger a subsequent optimization flow; And retraining a predictive model based on the newly acquired operation data, and re-optimizing PID control parameters based on the new predictive model until the integral square error is smaller than or equal to a preset threshold value, thereby achieving a performance target and completing optimization.
7. A system for optimizing the PID dynamic response of the temperature and pressure of a liquid hydrogen storage tank, said system comprising: The pretreatment unit is used for collecting key parameter historical data in the operation of the liquid hydrogen storage tank and preprocessing the key parameter historical data, wherein the key parameter historical data comprises temperature, pressure, refrigerating system input quantity and air release valve opening; The model training unit is used for constructing a multivariable dynamic response prediction model based on the long-term memory network, and training and verifying the multivariable dynamic response prediction model based on the preprocessed key parameter historical data so as to obtain a trained optimal prediction model; the mapping data set establishing unit is used for simulating temperature and pressure response processes under different PID control parameter combinations based on the optimal prediction model, acquiring performance indexes corresponding to each PID control parameter combination and establishing a mapping data set of the PID control parameter combinations and the performance indexes; The optimal parameter determining unit is used for taking PID control parameters as actions of the reinforcement learning agent, taking the weighted value of the dynamic response performance index as rewards, simulating an environment by using the optimal prediction model, enabling the agent to iteratively execute the actions, acquire the rewards and learn the optimal PID control parameter combination; and the optimizing unit is used for combining and deploying the optimal PID control parameters into a control system, collecting the operation data and the real-time performance index of the liquid hydrogen storage tank in real time, and performing response optimization based on the real-time performance index.
8. The system of claim 7, wherein the preprocessing unit preprocesses the key parameter history data, comprising: for each point in time in the key parameter history data Corresponding four-dimensional data points Performing missing value filling processing through the linear difference value, denoising and normalizing the data sequence of each dimension, In order to be able to determine the temperature, In the case of a pressure force, the pressure, For the input quantity of the refrigerating system, Is the opening degree of the air release valve.
9. The system of claim 7, wherein the multivariate dynamic response prediction model has an input layer dimension of 3 and an input vector of 3 The output layer is designed as a full connection layer, the dimension of the output layer is 2, and the output vector is Hidden layer dimension Initializing to super-parameters, defining hidden states And cell status Introducing splice vectors As a gating computation basis, an LSTM state update mechanism is set as follows: , , Wherein, the A temperature value representing a current time step; representing a pressure value; Representing a control input signal; for a single time step of temperature change in the future, The amount of pressure change for a single time step in the future; for the sigmoid activation function, Representing element-level multiplication, predicting variation , And For training parameters, all weight matrices are initialized with Xavier and bias terms are initialized to zero.
10. The system according to claim 7, wherein the mapping data set creating unit simulates a temperature pressure response process under different PID control parameter combinations based on the optimal prediction model, obtains a performance index corresponding to each PID control parameter combination, and creates a mapping data set of PID control parameter combinations and performance indexes, including: Determining parameter space with PID controller, including proportional gain Integral gain Differential gain To generate the total number of all combined points , wherein, 、 And Sampling points for each dimension; Controlling parameter combinations for each PID Executing closed loop system simulation, initializing the system state to zero, and applying the reference signal Determining an error signal for a unit step input Determining a controller output The method comprises the following steps: Calling the optimal prediction model, and inputting And the current system state, solving the model output through a numerical integration system I.e. temperature or pressure response curve, based on which the simulation process is continued until the system reaches steady state Calculating an overshoot OS; Wherein, the , Wherein, the Is a steady state value; In order to adjust the time period of the process, Satisfy the following requirements forall Searching a minimum time point through iteration; Calculating steady state errors as follows: , ; Recording all index values as data points To determine a mapping dataset of PID control parameter combinations and performance metrics.
11. The system according to claim 7, wherein the optimum parameter determining unit sets the motion vector of the agent when defining the motion space in learning the optimum PID control parameter combination Representing PID controller parameters, designing an action space into a continuous range, fixing an action dimension into a three-dimensional vector, directly mapping the three-dimensional vector to PID structural parameters, setting a prediction model environment for simulating control system dynamics, and inputting the model into a current state in a discrete state space mode And actions Outputting the predicted next state And control response The bonus function is designed to be weighted based on dynamic response performance indexes, wherein the performance indexes comprise integral square error ISE and maximum overshoot And adjusting the time Rewarding (rewarding) Defined as a negative weighted sum, targeting a minimization indicator, the formula is: , Wherein, the 、、 Is the weight; , is a reference input and a predicted output Is a deviation of (2).
12. The system of claim 7, wherein the optimization unit performs response optimization based on the real-time performance metrics, comprising: When the integral square error of the real-time performance index is larger than a preset threshold, judging that the performance is insufficient to trigger a subsequent optimization flow; And retraining a predictive model based on the newly acquired operation data, and re-optimizing PID control parameters based on the new predictive model until the integral square error is smaller than or equal to a preset threshold value, thereby achieving a performance target and completing optimization.

Description

Temperature pressure PID dynamic response optimization method and system for liquid hydrogen storage tank Technical Field The invention relates to the technical field of PID control parameter optimization, in particular to a temperature pressure PID dynamic response optimization method and system of a liquid hydrogen storage tank. Background The main problem to be solved by controlling the temperature and the pressure of the liquid hydrogen storage tank is to ensure the stability of the temperature and the pressure of the liquid hydrogen in the storage process, avoid safety accidents such as evaporation and leakage of the liquid hydrogen caused by overhigh temperature or overhigh pressure, reduce the energy consumption as much as possible and improve the efficiency and the economical efficiency of the liquid hydrogen storage. The core aim is to realize the accurate and stable control of the temperature and the pressure of the liquid hydrogen storage tank. In the field of temperature and pressure control of liquid hydrogen storage tanks, common related technical schemes and defects are as follows. The traditional PID control method mainly relies on experience to manually adjust PID parameters, is difficult to adapt to complex changes in the operation process of the liquid hydrogen storage tank, is difficult to dynamically adjust once the parameters are determined, and has poor control effect when the working conditions change. Although the fuzzy control method can process some uncertainties, the fuzzy control method lacks an accurate mathematical model, the establishment of control rules depends on expert experience, and high-precision control is difficult to realize. The neural network control method has strong nonlinear mapping capability, but has long training time, is easy to fall into a local optimal solution, and has high requirements on the quality and quantity of training data. The model predictive control method needs to establish an accurate mathematical model, and the operation process of the liquid hydrogen storage tank is complex, the difficulty of establishing the accurate model is large, the calculated amount is large, and the real-time performance is poor. Therefore, a temperature pressure PID dynamic response optimization method of the liquid hydrogen storage tank is needed, and the problems of low temperature pressure control precision, poor adaptability, high energy consumption and the like of the existing liquid hydrogen storage tank are solved. Disclosure of Invention The invention provides a temperature pressure PID dynamic response optimization method and a temperature pressure PID dynamic response optimization system for a liquid hydrogen storage tank, which aim to solve the problem of how to perform dynamic response optimization on PID control parameters of the temperature pressure of the liquid hydrogen storage tank, so that the temperature pressure control precision of the liquid hydrogen storage tank is high, the adaptability is good, and the energy consumption is low. In order to solve the above problems, according to an aspect of the present invention, there is provided a temperature pressure PID dynamic response optimization method of a liquid hydrogen storage tank, the method comprising: Collecting key parameter historical data in the operation of a liquid hydrogen storage tank, and preprocessing the key parameter historical data, wherein the key parameter historical data comprises temperature, pressure, refrigerating system input quantity and air release valve opening; constructing a multivariable dynamic response prediction model based on a long-term and short-term memory network, and training and verifying based on the preprocessed key parameter historical data to obtain a trained optimal prediction model; based on the optimal prediction model, simulating temperature and pressure response processes under different PID control parameter combinations, obtaining performance indexes corresponding to each PID control parameter combination, and establishing a mapping data set of the PID control parameter combinations and the performance indexes; Taking PID control parameters as actions of reinforcement learning agents, taking dynamic response performance index weighted values as rewards, simulating an environment by using the optimal prediction model, enabling the agents to iteratively execute the actions, acquire the rewards, and learning optimal PID control parameter combinations; And the optimal PID control parameters are combined and deployed into a control system, operation data and real-time performance indexes of the liquid hydrogen storage tank are collected in real time, and response optimization is performed based on the real-time performance indexes. Preferably, the preprocessing the key parameter historical data comprises: for each point in time in the key parameter history data Corresponding four-dimensional data pointsPerforming missing value filling processing thro