A Smart Control Optimization Method for Cryogenic Liquid Evaporators

Abstract

This invention discloses an intelligent control optimization method for cryogenic liquid evaporators. The method includes modeling a multi-valve control unit for the cryogenic liquid evaporator, designing an adaptive evaporator loss function, designing a centralized execution evaluation architecture for the multi-valve control unit, prioritizing experience replay training, and optimizing evaporator control. This invention belongs to the field of intelligent control, specifically referring to an intelligent control optimization method for cryogenic liquid evaporators. This scheme constructs precision control reward terms, energy consumption penalty terms, and safety penalty terms to balance control precision and system synergy. Based on the adaptive evaporator loss function design, conflict penalties are used to avoid contradictions in action logic, thereby improving the stability of control optimization. The degree of action conflict is incorporated into the experience priority calculation, prioritizing the learning of conflict scenarios. Importance weights and collaborative penalty designs are added to evaluate network loss, significantly reducing the probability of action conflict. Importance sampling is used to correct distribution offsets, thereby improving the reliability of control optimization.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to an intelligent control optimization method for cryogenic liquid evaporators. Background Technology

[0002] Low-temperature liquid evaporator control optimization methods are technical means to optimize valve opening and operating parameters by adjusting the actions of key components such as feed valves, heat source valves, and outlet valves to achieve goals such as stable outlet medium temperature, compliant system pressure, and low energy consumption. However, general evaporator control optimization methods suffer from poor control coordination, lack of stage adaptability, slow start-up convergence, easy steady-state fluctuations, and a disconnect between safety and energy consumption goals, resulting in poor control optimization effects. Furthermore, general evaporator control optimization methods suffer from insufficient control action stability, weak disturbance rejection capability, low experience learning efficiency, and difficulty in handling complex scenarios, leading to low reliability of control optimization. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an intelligent control optimization method for cryogenic liquid evaporators. Addressing the problems of poor control coordination, lack of stage adaptability, slow start-up convergence, easy steady-state fluctuations, and a disconnect between safety and energy consumption targets, leading to poor control optimization results, this solution models a multi-valve control unit, constructing separate precision control reward, energy consumption penalty, and safety penalty terms to balance control precision and system coordination. Based on an adaptive evaporator loss function design, it achieves rapid convergence during start-up, shortening the heating time, and reduces frequent valve operations to lower energy consumption and wear during steady-state operation. Simultaneously, it uses conflict penalty... To avoid logical contradictions in actions and thus improve the stability of control optimization, this solution addresses the problems of insufficient control action stability, weak anti-disturbance capability, low experience learning efficiency, and difficulty in handling complex scenarios, leading to low reliability of control optimization. This solution employs an independent execution network + shared evaluation network architecture, incorporating the degree of action conflict into the experience priority calculation and prioritizing the learning of conflict scenarios. Furthermore, it incorporates importance weights and collaborative penalty design to evaluate the network loss, rapidly forming a collaborative control strategy and significantly reducing the probability of action conflicts. Importance sampling is used to correct distribution bias, avoiding training bias caused by priority sampling, thereby improving the reliability of control optimization.

[0004] The technical solution adopted by this invention is as follows: This invention provides an intelligent control and optimization method for a cryogenic liquid evaporator, which includes the following steps:

[0005] Step S1: Modeling the multi-valve control unit of the cryogenic liquid evaporator;

[0006] Step S2: Adaptive evaporator loss function design;

[0007] Step S3: Design of a centralized execution evaluation architecture for multi-valve control units;

[0008] Step S4: Prioritize experience replay training;

[0009] Step S5: Evaporator control optimization.

[0010] Furthermore, in step S1, the modeling of the multi-valve control unit of the cryogenic liquid evaporator specifically includes:

[0011] Define a quintuple, including evaporator operating state S, valve control unit output action, state transition T, quantified control objective reward R, and discount factor γ; evaporator operating state is obtained by collecting cryogenic liquid feed flow rate data from sensors. Heat source temperature evaporator shell side pressure outlet gas temperature outlet gas pressure and ambient temperature The valve control unit outputs the valve opening, normalized to [-1, 1], which represents the valve opening of the feed valve control unit. Heat source valve control unit opening degree and outlet valve control unit opening degree The state transition, based on the evaporator heat balance equation, describes the dynamic change from action to state, and is expressed as: Where t is the sampling time; , and These are the time constant, feed cooling coefficient, and temperature regulation coefficient, respectively. This is the target value for the outlet temperature;

[0012] Define reward function Simultaneously, it incentivizes compliance with control standards and energy consumption reduction, while penalizing overheating and overpressure, thus constructing a precision control reward system. , represented as: Construct energy consumption penalty items , represented as: ; Construct security penalty items , represented as: ; ;in, and It is the temperature compliance bonus coefficient; and It is the pressure target achievement reward coefficient; It is the energy consumption coefficient; It is the over-temperature penalty coefficient. It exceeds the temperature threshold; It is the overpressure penalty coefficient. It is the overpressure threshold; This is the target value for export pressure.

[0013] Furthermore, in step S2, the adaptive evaporator loss function design defines an adaptive piecewise MSE loss. , represented as: Among them, segmented weights Dynamically adjusted according to the control phase, as shown below: ; This is the current Q value of valve control unit i; It is the target Q value; A is the combined action of the three valve control units; This is a penalty for action conflict, represented as: ; It is the conflict penalty coefficient.

[0014] Furthermore, in step S3, the centralized execution evaluation architecture design of the multi-valve control unit specifically includes the following:

[0015] Step S31: Architecture design; Equip the feed valve, heat source valve, and outlet valve with independent execution networks, with each network only outputting the opening command of the corresponding valve; Share one evaluation network, input the global operating status of the evaporator and the joint action of the three valves, evaluate the long-term control value of the current action combination, and feed the evaluation results back to the three execution networks;

[0016] Step S32: Execute network definition; each execution network is only responsible for calculating the opening degree of a single valve, with the input being the global state of the evaporator and the output being the normalized opening degree of the valve;

[0017] Step S33: Evaluate the network definition; centrally evaluate the value of joint actions within the network;

[0018] Step S34: Calculate the target Q value.

[0019] Furthermore, in step S4, the priority experience replay training specifically includes the following:

[0020] Step S41: Experience pool construction; Store the interactive experience of each step of the evaporator as a tuple;

[0021] Step S42: Evaluate the value of the experience; prioritize the experience. The combined Q-value error and the degree of conflict are expressed as: rank(·) is a sorting function; It is a conflict weight; This is the current Q value of valve control unit i; It is the target Q value;

[0022] Step S43: Importance sampling; introduce importance weights to correct distribution bias, expressed as: ; ; and These are the importance weights before and after the correction, respectively; D is the experience pool capacity. These are weighting coefficients;

[0023] Step S44: Update network parameters; incorporate importance weights and collaborative penalty design to evaluate network loss. , represented as: ;in, It is an expectation.

[0024] Furthermore, in step S5, the evaporator control optimization involves deploying a well-trained multi-valve control unit model to the evaporator DCS system, forming a closed loop of real-time acquisition, distributed execution, centralized evaluation, and online update.

[0025] The beneficial effects achieved by adopting the above solution are as follows:

[0026] (1) To address the problems of poor control coordination, lack of stage adaptability, slow start-up convergence, easy fluctuation in steady state, and disconnect between safety and energy consumption targets in general evaporator control optimization methods, which lead to poor control optimization results, this solution models a multi-valve control unit and constructs a precision control reward term, an energy consumption penalty term, and a safety penalty term to balance control precision and system coordination. Based on the design of an adaptive evaporator loss function, it achieves rapid convergence during start-up to shorten the heating time and reduces frequent valve actions to lower energy consumption and wear during steady state. At the same time, it avoids contradictions in action logic through conflict penalties, thereby improving the stability of control optimization.

[0027] (2) To address the problems of insufficient stability of control actions, weak anti-disturbance ability, low efficiency of experience learning, and difficulty in coping with complex scenarios in general evaporator control optimization methods, which leads to low reliability of control optimization, this solution designs an independent execution network + shared evaluation network architecture, incorporates the degree of action conflict into the calculation of experience priority, and prioritizes the learning of conflict scenarios; and adds importance weight and collaborative penalty to design the evaluation network loss, quickly forming a collaborative control strategy and significantly reducing the probability of action conflict; and corrects the distribution offset through importance sampling to avoid training bias caused by priority sampling; thereby improving the reliability of control optimization. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating an intelligent control optimization method for a cryogenic liquid evaporator provided by the present invention.

[0029] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0030] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0031] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0032] Example 1, see Figure 1 This invention provides an intelligent control optimization method for cryogenic liquid evaporators, which includes the following steps:

[0033] Step S1: Modeling the multi-valve control unit of the cryogenic liquid evaporator; the control problem of the cryogenic liquid evaporator is transformed into a decision process of a three-valve control unit, defining a quintuple containing state, action, transition, reward and discount factor, and decoupling it into 3 valve control units;

[0034] Step S2: Adaptive evaporator loss function design; design a loss function for the three-stage control characteristics of the evaporator, including piecewise dynamic weights and a collaborative penalty term for the valve control unit, and update the execution network parameters based on the loss gradient;

[0035] Step S3: Design of centralized execution evaluation architecture for multi-valve control units; including independent execution network, shared evaluation network, target network, and parameter definition;

[0036] Step S4: Prioritize experience replay training; this includes building an experience pool, evaluating the value of experiences, sampling importance, and updating network parameters.

[0037] Step S5: Evaporator control optimization; Evaporator control optimization is achieved based on a multi-valve control unit.

[0038] Example 2, see Figure 1This embodiment is based on the above embodiment. In step S1, the modeling of the multi-valve control unit of the cryogenic liquid evaporator is to transform the control problem into an interactive Markov decision process of three valve control units. Each valve control unit corresponds to a control variable. During training, a centralized evaluator is used to evaluate the joint action, and during execution, the action is performed in a decentralized manner to balance control accuracy and system synergy.

[0039] The specific operation is as follows: Define a quintuple, including the evaporator operating state S, the valve control unit output action, the state transition T, the reward R for quantifying the control objective, and the discount factor γ; the evaporator operating state is obtained by collecting the cryogenic liquid feed flow rate through sensors. Heat source temperature evaporator shell side pressure outlet gas temperature outlet gas pressure and ambient temperature The valve control unit outputs the valve opening, normalized to [-1, 1], which represents the valve opening of the feed valve control unit. Heat source valve control unit opening degree and outlet valve control unit opening degree The state transition, based on the evaporator heat balance equation, describes the dynamic change from action to state, and is expressed as: Where t is the sampling time; , and These are the time constant, feed cooling coefficient, and temperature regulation coefficient, respectively. This is the target value for the outlet temperature;

[0040] Define reward function Simultaneously, it incentivizes compliance with control standards and energy consumption reduction, while penalizing overheating and overpressure, thus constructing a precision control reward system. , represented as: Construct energy consumption penalty items , represented as: ; Construct security penalty items , represented as: ; ;in, and It is the temperature compliance bonus coefficient; and It is the pressure target achievement reward coefficient; It is the energy consumption coefficient, to avoid wasting energy by fully opening the heat source valve; It is the over-temperature penalty coefficient. It exceeds the temperature threshold; It is the overpressure penalty coefficient. It is the overpressure threshold; This is the target value for export pressure;

[0041] By decoupling the feed-heat source-outlet into three valve control units, the risks of overheating and overpressure are avoided due to single-variable control. The system also directly mitigates these risks by using safety penalty and reward mechanisms.

[0042] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the adaptive evaporator loss function design is designed for the three-stage control characteristics of the evaporator from start-up to transition to steady state. A piecewise weighted loss is introduced. The weight of the large error is increased in the start-up stage to accelerate convergence, and the weight is reduced in the steady state stage to suppress small fluctuations. At the same time, a cooperative penalty term for the valve control unit is introduced to resolve the action conflict that occurs when there are multiple valve control units.

[0043] The specific steps are as follows:

[0044] Adaptive piecewise MSE loss Defined as: Among them, segmented weights Dynamically adjusted according to the control phase, as shown below: The weights are doubled during the startup phase to accelerate the correction of large errors; the weights are halved during the steady-state phase to avoid overfitting to small perturbations. This is the current Q value of valve control unit i; It is the target Q value; A is the combined action of the three valve control units; It is a conflict penalty for actions, used to quantify the coordination of actions of the valve control unit, and is expressed as: ; It is a conflict penalty coefficient, ensuring that collaboration takes precedence over univariate optimization; the execution network parameters of the valve control unit are updated based on the loss gradient.

[0045] By performing the above operations, this solution addresses the problems of poor control coordination, lack of stage adaptability, slow start-up convergence, easy steady-state fluctuations, and the disconnect between safety and energy consumption targets, which lead to poor control optimization results in general evaporator control optimization methods. It models a multi-valve control unit, constructing separate precision control reward, energy consumption penalty, and safety penalty terms to balance control precision and system coordination. Based on an adaptive evaporator loss function design, it achieves rapid convergence during start-up to shorten heating time and reduces frequent valve actions to lower energy consumption and wear during steady-state operation. Simultaneously, it avoids logical inconsistencies in action through conflict penalties, thereby improving the stability of control optimization.

[0046] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the centralized execution evaluation architecture design of the multi-valve control unit specifically includes the following:

[0047] Step S31: Architecture design; Equip the feed valve, heat source valve, and outlet valve with independent execution networks. Each network only outputs the opening command of the corresponding valve and performs parallel calculations to meet the real-time control requirements of the evaporator; Share one evaluation network, input the global operating status of the evaporator and the joint action of the three valves, evaluate the long-term control value of the current action combination, and feed the evaluation results back to the three execution networks to guide the optimization of actions and avoid conflicts;

[0048] Step S32: Execution network definition; each execution network is only responsible for calculating the opening degree of a single valve, with the global state of the evaporator as input and the normalized opening degree of the valve as output, ensuring that the action can directly drive the actuator; the execution network of valve control unit i outputs the action. Represented as: The output is normalized to [-1, 1] by the tanh activation function, which directly corresponds to the valve opening. It is the strategy function of valve control unit i, which maps states to actions according to the rules. These are the corresponding network parameters; S(t) is the evaporator operating state at time t;

[0049] Step S33: Evaluate the network definition; the control effect of the evaporator requires global judgment, therefore the evaluation network needs to input the global state + combined actions, and output a quantified action value Q value. The higher the Q value, the better the current action combination can achieve multi-objective optimization of temperature target achievement + pressure stability + low energy consumption, thereby guiding the execution network to adjust the valve opening; the centralized evaluation network evaluates the value of combined actions, expressed as: MLP stands for Multilayer Perceptron, with an input layer of 6+3=9 dimensions, a hidden layer of 256 dimensions, and an output of a 1-dimensional Q-value. It evaluates network parameters, which are shared globally and updated based on joint error; A(t) is the joint action of the three valves at time t; ω and b are the weight and bias parameters of the network, respectively.

[0050] Step S34: Calculate the target Q-value. If the Q-value of the current evaluation network is directly used to guide network optimization, the Q-value will be unstable due to fluctuations in the current network parameters, leading to frequent changes in valve opening and exacerbating evaporator temperature and pressure fluctuations. Therefore, a target network is introduced to calculate a stable target Q-value as the benchmark for optimization. The target Q-value is expressed as: ; It is to evaluate the target network; These are the target network parameters, obtained through smooth parameter updates; the target execution network output. Represented as: ; , and It is the policy function of the target execution network of the three valve control units. , and This corresponds to the network parameters to be executed for the target.

[0051] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the priority experience playback training refers to high-value experiences during evaporator operation that have multi-valve control unit characteristics, including multivariable disturbance experiences (such as simultaneous fluctuations in feed flow and ambient temperature, requiring priority learning of collaborative disturbance rejection strategies); and action conflict experiences (such as valve control unit 1 opening and valve control unit 2 closing causing a sudden temperature drop, requiring priority learning of conflict avoidance). Therefore, a multi-valve control unit experience value assessment is introduced to ensure that high-cooperative-value experiences are prioritized for learning. Specifically, this includes the following:

[0052] Step S41: Experience pool construction; Store the interaction experience of each step of the evaporator as a tuple, represented as: ; It is the nth empirical tuple;

[0053] Step S42: Evaluate the value of the experience; prioritize the experience. The combined Q-value error and the degree of conflict are expressed as: rank(·) is a sorting function; It is a conflict weight; This is the current Q value of valve control unit i; It is the target Q value;

[0054] Step S43: Importance sampling; introduce importance weights to correct distribution bias, expressed as: ; ; and These are the importance weights before and after the correction, respectively; D is the experience pool capacity. These are weighting coefficients;

[0055] Step S44: Update network parameters; incorporate importance weights and collaborative penalty design to evaluate network loss. , represented as: ;in, It is the expectation; it executes network parameter updates based on loss gradients; it performs smooth parameter updates on the target network.

[0056] By performing the above operations, this solution addresses the problems of insufficient control action stability, weak anti-disturbance capability, low experience learning efficiency, and difficulty in coping with complex scenarios, which lead to low reliability of control optimization in general evaporator control optimization methods. It designs an independent execution network + shared evaluation network architecture, incorporating the degree of action conflict into the experience priority calculation and prioritizing the learning of conflict scenarios. Furthermore, it incorporates importance weights and collaborative penalties to design the evaluation network loss, rapidly forming a collaborative control strategy and significantly reducing the probability of action conflicts. Importance sampling corrects distribution offsets, avoiding training bias caused by priority sampling, thereby improving the reliability of control optimization.

[0057] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the evaporator control optimization involves deploying a well-trained multi-valve control unit model to the evaporator DCS system to form a closed loop of real-time acquisition, distributed execution, centralized evaluation, and online update: the sensor acquires the state S(t) every 1 second and sends it to the execution network of each valve control unit; the three valve control units output the valve opening in parallel to drive the actuator; the centralized evaluation network evaluates the control effect in real time and stores new experience in the experience pool; the model is fine-tuned using historical data every morning to avoid performance degradation caused by evaporator aging.

[0058] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0059] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0060] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An intelligent control optimization method for a cryogenic liquid evaporator, characterized in that: The method includes the following steps: Step S1: Modeling the multi-valve control unit of the cryogenic liquid evaporator; the control problem of the cryogenic liquid evaporator is transformed into a decision process of the multi-valve control unit, defining a quintuple containing state, action, transition, reward and discount factor, and decoupling it into 3 valve control units; Step S2: Adaptive evaporator loss function design; design a loss function for the three-stage control characteristics of the evaporator, including piecewise dynamic weights and a collaborative penalty term for the valve control unit, and update the execution network parameters based on the gradient of the adaptive evaporator loss function; Step S3: Design of centralized execution evaluation architecture for multi-valve control units; including independent execution network, shared evaluation network, target network, and parameter definition; Step S4: Prioritize experience replay training; this includes building an experience pool, evaluating the value of experiences, sampling importance, and updating network parameters. Step S5: Evaporator control optimization; Evaporator control optimization is achieved based on a multi-valve control unit; In step S1, the modeling of the multi-valve control unit of the cryogenic liquid evaporator specifically includes: Define a quintuple, including evaporator operating state S, valve control unit output action, state transition T, quantified control objective reward R, and discount factor γ; evaporator operating state is obtained by collecting cryogenic liquid feed flow rate data from sensors. Heat source temperature evaporator shell side pressure outlet gas temperature outlet gas pressure and ambient temperature The valve control unit outputs the valve opening, normalized to [-1, 1], which represents the valve opening of the feed valve control unit. Heat source valve control unit opening degree and outlet valve control unit opening degree The state transition, based on the evaporator heat balance equation, describes the dynamic change from action to state, and is expressed as: Where t is the sampling time; , and These are the time constant, feed cooling coefficient, and temperature regulation coefficient, respectively. This is the target value for the outlet temperature; Define a reward function that incentivizes both achieving control targets and reducing energy consumption, while penalizing over-temperature and over-pressure conditions, thus constructing a precision control reward program. , is represented as: Construct energy consumption penalty items , is represented as: ; Construct security penalty items , is represented as: ; ;in, and It is the temperature compliance bonus coefficient; and It is the pressure target achievement reward coefficient; It is the energy consumption coefficient; It is the over-temperature penalty coefficient. It exceeds the temperature threshold; It is the overpressure penalty coefficient. It is the overpressure threshold; This is the target value for export pressure; In step S2, the adaptive evaporator loss function design defines an adaptive piecewise MSE loss. , is represented as: Among them, segmented weights Dynamically adjusted according to the control phase, as shown below: ; This is the current Q value of valve control unit i; It is the target Q value; A is the combined action of the multi-valve control unit; This is a penalty for action conflict, represented as: ; It is the conflict penalty coefficient.

2. The intelligent low-temperature liquid evaporator control optimization method according to claim 1, characterized in that: In step S3, the centralized execution evaluation architecture design of the multi-valve control unit specifically includes the following: Step S31: Architecture design; Equip the feed valve, heat source valve and outlet valve with independent execution networks. Each independent execution network only outputs the opening command of the corresponding valve; Share one evaluation network, input the global operating status of the evaporator and the joint action of the three valves, evaluate the long-term control value of the current action combination, and feed the evaluation results back to the three execution networks; Step S32: Execute network definition; each execution network is only responsible for calculating the opening degree of a single valve, with the input being the global state of the evaporator and the output being the normalized opening degree of the valve; Step S33: Evaluate the network definition; evaluate the value of joint actions within the network. Step S34: Calculate the target Q value.

3. The intelligent low-temperature liquid evaporator control optimization method according to claim 2, characterized in that: In step S4, the priority experience replay training specifically includes the following: Step S41: Experience pool construction; Store the interactive experience of each step of the evaporator as a tuple; Step S42: Evaluate the value of the experience; prioritize the experience. The combined Q-value error and the degree of conflict are expressed as: rank(·) is the sorting function; It is a conflict weight; This is the current Q value of valve control unit i; It is the target Q value; Step S43: Importance sampling; introduce importance weights to correct distribution bias, expressed as: ; ; and These are the importance weights before and after the correction, respectively; D is the capacity of the experience pool; These are weighting coefficients; Step S44: Update network parameters; incorporate importance weights and collaborative penalty design to evaluate network loss. , is represented as: ;in, It is the expectation; gradient updates of network parameters are performed based on the adaptive evaporator loss function; smooth parameter updates are performed on the target network.

4. The intelligent low-temperature liquid evaporator control optimization method according to claim 3, characterized in that: In step S5, the evaporator control optimization involves deploying a well-trained multi-valve control unit model to the evaporator DCS system, forming a closed loop of real-time acquisition, distributed execution, centralized evaluation, and online updates.

Images