Self-adaptive adjustment ship shared condenser energy-saving system

Abstract

The invention relates to a self-adaptive adjusting ship common condenser energy-saving system which comprises at least one common condenser, and a fluid loop of the common condenser is connected to a main engine, an auxiliary generator and an air conditioner refrigerating system of a ship. The sensor array comprises a temperature sensor, a pressure sensor and a flow sensor which are arranged at each fluid passage inlet and outlet of the common condenser; the signal input end of the self-adaptive control unit is connected with the sensor array, and the control output end of the self-adaptive control unit is connected to the cooling fan frequency converter, the water pump frequency converter and the regulating valve of the common condenser. The method has the advantages that the self-adaptive control unit based on the reinforcement learning model is introduced, the control strategy can be dynamically adjusted according to real-time operation data and historical experience, the current state is responded, continuous evolution can be achieved through online fine adjustment, and therefore the energy efficiency is continuously improved within the full working condition range; and the limitation of strategy stiffness of a traditional control method is broken through.

Description

Technical Field

[0001] This invention relates to an adaptive adjustment energy-saving system for a shared marine condenser, belonging to the field of energy conservation. Background Technology

[0002] Improving ship operational energy efficiency is a core approach for the shipping industry to cope with increasingly stringent environmental regulations and operating cost pressures. On modern ships, multiple devices such as main engines, auxiliary generators, and air conditioning systems generate significant amounts of waste heat during operation, typically dissipated through their respective condensers or a shared condenser. While using a shared condenser simplifies the initial layout, it faces complex and fluctuating heat load sources. The load rates of various heat sources differ greatly under different operating conditions, such as navigation, operation, and berthing, resulting in a wide dynamic range of total heat load for the shared condenser, posing a severe challenge to real-time energy efficiency management of the system.

[0003] Currently, the control of ship condensing systems mostly relies on traditional PID controllers or on / off control based on fixed thresholds. These methods can only provide feedback adjustment for a few key parameters (such as condensing pressure or temperature), resulting in lag and an inability to perceive the overall energy efficiency of the system. They lack adaptability to variable environments (such as changes in water and air temperatures in different sea areas) and complex load coupling relationships. Specifically, existing technologies mainly suffer from the following drawbacks: First, the control strategy is rigid, failing to achieve dynamic optimization of energy consumption while ensuring heat dissipation requirements; second, the various actuators in the system (such as cooling fans and water pumps) often operate independently, lacking coordination, which easily leads to control conflicts and unnecessary energy waste; third, they rely entirely on real-time feedback, lacking the ability to predict load changes, making it difficult to cope with drastic changes in ship operating conditions, and hindering the achievement of optimal system stability and energy efficiency.

[0004] In summary, the fundamental technical bottleneck of existing shared condenser systems in ships lies in their insufficient perception and decision-making capabilities. They are unable to treat a multi-input, multi-output, strongly coupled nonlinear thermodynamic system as a whole for real-time energy efficiency assessment and collaborative optimization control. Although some solutions attempt to introduce simple energy-saving logic, they generally lack self-learning and adaptive mechanisms, making it impossible to continuously evolve and optimize throughout the ship's entire lifecycle.

[0005] Therefore, there is an urgent need in this field for an energy-saving system that can deeply integrate real-time data and possess intelligent decision-making and forward-looking adjustment capabilities, in order to break through the limitations of existing control modes and achieve a leapfrog improvement in ship energy utilization efficiency. This invention is proposed precisely to fundamentally solve this technical problem. Summary of the Invention

[0006] To overcome the shortcomings of existing technologies, this invention provides an adaptive adjustment energy-saving system for shared marine condensers. The technical solution of this invention is as follows:

[0007] An adaptive adjustment energy-saving system for a shared marine condenser includes:

[0008] At least one common condenser, the fluid circuit of which is connected to the ship’s main engine, auxiliary generator and air conditioning system, is used to receive and dissipate the combined heat load of several heat sources;

[0009] A sensor array, including temperature sensors, pressure sensors, and flow sensors installed at the inlet and outlet of each fluid passage of the common condenser, is used to collect the operating parameters of the system in real time.

[0010] An adaptive control unit, the signal input of which is connected to the sensor array, and the control output of which is connected to the cooling fan inverter, water pump inverter and regulating valve of the common condenser;

[0011] The adaptive control unit is configured to perform the following steps:

[0012] a. State perception and modeling: Based on the real-time data from the sensor array, calculate the total heat load rate and real-time energy efficiency ratio of the current system;

[0013] b. Strategy generation and optimization: The calculated total heat load rate and real-time energy efficiency ratio are input into a pre-trained reinforcement learning model. The reinforcement learning model uses minimizing the overall system energy consumption as the reward function and outputs the optimal set of control commands for the cooling fan, water pump and regulating valve.

[0014] c. Execution and Adaptation: Execute the optimal set of control commands and collect the system response again through the sensor array. Store the interactive experience data in the experience playback buffer for periodic online fine-tuning of the reinforcement learning model, thereby achieving continuous adaptive optimization of the system.

[0015] The high-temperature cooling circuit from the main engine and the medium-temperature refrigerant circuit from the air conditioning system exchange heat with the circulating working fluid of the common condenser through independent plate heat exchangers; the adaptive control unit independently adjusts the electric regulating valves on each heat exchange circuit through a decoupling control algorithm.

[0016] In step b, the reinforcement learning model is a deep deterministic policy gradient network based on the Actor-Critic framework. The Actor network serves as the policy function, with its input being a normalized system state vector and its output being a continuous action vector with the same dimension as the number of actuators. This action vector, after inverse normalization, is directly used as the frequency setpoint of the inverter and the opening command of the regulating valve. The Critic network serves as the value function, used to evaluate the expected reward of the Actor network's output action and guide the update direction of the Actor network.

[0017] In step b, the reward function R(t) is composed of the following multi-factor weighted sum in each control period t:

[0018] R(t) = -[α·P_total(t) + β·Σ|Δu_i(t)| + γ·|T_cond(t)T_cond_opt|]; where P_total(t) is the total power consumption of the cooling fan and water pump in this cycle; Σ|Δu_i(t)| is the sum of the absolute values ​​of the changes in control commands of all actuators in this cycle and the previous cycle, used to penalize action oscillations; |T_cond(t)T_cond_opt| is the absolute value of the difference between the current condensing temperature and the optimal condensing temperature dynamically calculated by the energy-saving model, used to ensure thermodynamic efficiency; α, β, γ are pre-set positive weighting coefficients used to balance the relationship between energy saving, equipment life and system stability.

[0019] The adaptive control unit also includes a multi-objective optimization decision submodule. This multi-objective optimization decision submodule uses a non-dominated sorting genetic algorithm with an elitist strategy to find the Pareto optimal solution set among the three objectives of total system power consumption, cumulative fatigue damage of key components, and condenser terminal difference stability. Based on the operational priority of the ship's current voyage stage, it automatically selects the best compromise solution from the solution set as the objective constraint for training the reinforcement learning model.

[0020] The online estimation method for the cumulative fatigue damage of the key components is as follows: based on Miner's linear cumulative damage theory, the shaft load spectrum derived from control commands and system pressure is analyzed in real time using the Rainflow counting method, and combined with the material's SN curve, the remaining service life of the cooling fan bearing and water pump mechanical seal is dynamically predicted.

[0021] The adaptive control unit also includes a feedforward prediction module. This feedforward prediction module uses a long short-term memory neural network model to perform time-series analysis on the planned speed and heading provided by the ship's integrated navigation system for the next 15-30 minutes, as well as the future ambient temperature and humidity change data received from satellite meteorological services. It predicts the future heat load change trend that the shared condenser will face and feeds this trend into the reinforcement learning model in advance as contextual information.

[0022] The advantages of this invention are:

[0023] 1. By introducing an adaptive control unit based on a reinforcement learning model, a complete closed loop of "perception-decision-execution-learning" is constructed. The system can dynamically adjust the control strategy based on real-time operating data and historical experience, not only responding to the current state but also continuously evolving through online fine-tuning, thereby continuously improving energy efficiency across the entire operating range and breaking through the limitations of rigid strategies in traditional control methods.

[0024] 2. It can directly output continuous and precise control commands to multiple actuators, achieving optimal coordinated control of cooling fans, water pumps, and regulating valves. The reward function simultaneously considers energy consumption, equipment operation smoothness, and thermodynamic efficiency, effectively avoiding control oscillations. While ensuring energy-saving effects, it significantly improves the stability of system operation and equipment lifespan.

[0025] 3. By integrating a multi-objective optimization decision-making submodule, the optimization objective is expanded from simply minimizing energy consumption to multiple dimensions such as fatigue life of key components and system stability. The system can automatically select the best trade-off based on the Pareto front, making energy-saving management strategies more intelligent and comprehensive, and meeting the diverse operational needs of ships at different voyage stages.

[0026] 4. A feedforward prediction module based on a long short-term memory neural network has been introduced, which can accurately predict future heat load trends. This enables the system to cope with drastic changes in ship operating conditions, achieving a leap from "passive response" to "proactive foresight," and greatly improving the regulation quality and robustness in dynamically changing environments.

[0027] 5. It deeply integrates thermodynamic models, fatigue damage models, etc., providing an effective solution for optimizing the control of complex ship energy systems. Attached Figure Description

[0028] Figure 1 This is a block diagram of the main structure of the present invention. Detailed Implementation

[0029] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer as a result. However, these embodiments are merely exemplary and do not constitute any limitation on the scope of the present invention. Those skilled in the art should understand that modifications or substitutions can be made to the details and form of the technical solutions of the present invention without departing from the spirit and scope of the present invention, but all such modifications and substitutions fall within the protection scope of the present invention.

[0030] See Figure 1 This invention relates to an adaptive adjustment energy-saving system for a shared marine condenser, comprising:

[0031] At least one common condenser 1, the fluid circuit of which is connected to the ship’s main engine, auxiliary generator and air conditioning system, for receiving and dissipating the combined heat load of several heat sources;

[0032] A sensor array 2 includes temperature sensors, pressure sensors, and flow sensors installed at the inlet and outlet of each fluid passage of the common condenser, for real-time acquisition of system operating parameters;

[0033] An adaptive control unit 3 is provided, the signal input terminal of which is connected to the sensor array, and the control output terminal is connected to the cooling fan inverter, water pump inverter and regulating valve of the common condenser.

[0034] The adaptive control unit is configured to perform the following steps:

[0035] a. State perception and modeling: Based on the real-time data from the sensor array, calculate the total heat load rate and real-time energy efficiency ratio of the current system;

[0036] b. Strategy generation and optimization: The calculated total heat load rate and real-time energy efficiency ratio are input into a pre-trained reinforcement learning model. The reinforcement learning model uses minimizing the overall system energy consumption as the reward function and outputs the optimal set of control commands for the cooling fan, water pump and regulating valve.

[0037] c. Execution and Adaptation: Execute the optimal set of control commands and collect the system response again through the sensor array. Store the interactive experience data in the experience playback buffer for periodic online fine-tuning of the reinforcement learning model, thereby achieving continuous adaptive optimization of the system.

[0038] By using a sensor array, adaptive control unit, closed-loop control of actuators, and online fine-tuning, an intelligent system is formed that can learn from actual operating results and continuously improve.

[0039] Reinforcement learning models can simultaneously process multiple input variables such as temperature, pressure, and flow rate, and collaboratively output a unified set of optimal control commands for cooling fans, water pumps, and regulating valves. This overall optimization capability of multivariable, strongly coupled systems solves the energy efficiency loss problem caused by the independent operation and lack of coordination of each actuator in traditional control.

[0040] It can adapt to the aging and scaling of ship equipment, as well as the differences in operation in different navigation areas and seasons, and always maintain the optimal or near-optimal operating state.

[0041] By seamlessly integrating the real-time sensing of the sensor array 2, the intelligent decision-making of the adaptive control unit 3, and the precise actions of the execution components, the need for manual intervention is reduced, and energy-saving measures can be implemented in a timely and accurate manner.

[0042] The high-temperature cooling circuit from the main engine and the medium-temperature refrigerant circuit from the air conditioning system exchange heat with the circulating working fluid of the common condenser through independent plate heat exchangers; the adaptive control unit independently adjusts the electric regulating valves on each heat exchange circuit through a decoupling control algorithm.

[0043] In step b, the reinforcement learning model is a deep deterministic policy gradient network based on the Actor-Critic framework. The Actor network serves as the policy function, with its input being a normalized system state vector and its output being a continuous action vector with the same dimension as the number of actuators. This action vector, after inverse normalization, is directly used as the frequency setpoint of the inverter and the opening command of the regulating valve. The Critic network serves as the value function, used to evaluate the expected reward of the Actor network's output action and guide the update direction of the Actor network.

[0044] This step can directly generate the precise frequency value of the frequency converter and the precise opening command of the regulating valve, realizing stepless fine adjustment of the actuator, thereby improving the smoothness of control and the accuracy of energy efficiency optimization.

[0045] The ship condenser system is a complex system with multiple parameters and strong coupling. The state vector has a high dimension. By combining the value function approximation of deep neural networks and the policy gradient method, it is possible to efficiently extract features from the high-dimensional, nonlinear system state and learn the optimal policy.

[0046] The action vectors output by the Actor network have the same dimension as the number of actuators. This means that the model is inherently designed to solve collaborative control problems like those of this invention, which require the simultaneous coordination of multiple components such as cooling fans, water pumps, and valves. It can automatically learn the inherent coupling relationships between these actuators and provide a globally optimal set of collaborative control instructions, rather than relying on isolated local optimizations.

[0047] In step b, the reward function R(t) is composed of the following multi-factor weighted sum in each control period t:

[0048] R(t) = -[α·P_total(t) + β·Σ|Δu_i(t)| + γ·|T_cond(t)T_cond_opt|]; where P_total(t) is the total power consumption of the cooling fan and water pump in this cycle; Σ|Δu_i(t)| is the sum of the absolute values ​​of the changes in control commands of all actuators in this cycle and the previous cycle, used to penalize action oscillations; |T_cond(t)T_cond_opt| is the absolute value of the difference between the current condensing temperature and the optimal condensing temperature dynamically calculated by the energy-saving model, used to ensure thermodynamic efficiency; α, β, γ are pre-set positive weighting coefficients used to balance the relationship between energy saving, equipment life and system stability.

[0049] By mathematically unifying the three crucial and often conflicting objectives of energy saving, equipment protection, and thermodynamic efficiency within a single framework and introducing a penalty term Σ|Δu_i(t)|, this function directly incorporates the smoothness of actuator (such as frequency converters and valves) as one of its optimization objectives. This effectively suppresses frequent and significant fluctuations in control commands, improves system stability, and significantly extends the service life of critical mechanical components.

[0050] The term |T_cond(t)T_cond_opt| in the function is crucial, as it constrains the system to achieve energy savings under conditions close to optimal thermodynamic efficiency. This avoids the model blindly reducing the condensing pressure in pursuit of extreme energy savings, which could lead to the condensing temperature deviating from a reasonable range and potentially causing long-term problems such as decreased heat exchange efficiency and increased equipment scaling. This ensures the scientific nature of the energy-saving strategy and the long-term reliable operation of the system.

[0051] The adaptive control unit also includes a multi-objective optimization decision submodule. This multi-objective optimization decision submodule uses a non-dominated sorting genetic algorithm with an elitist strategy to find the Pareto optimal solution set among the three objectives of total system power consumption, cumulative fatigue damage of key components, and condenser terminal difference stability. Based on the operational priority of the ship's current voyage stage, it automatically selects the best compromise solution from the solution set as the objective constraint for training the reinforcement learning model.

[0052] The online estimation method for the cumulative fatigue damage of the key components is as follows: based on Miner's linear cumulative damage theory, the shaft load spectrum derived from control commands and system pressure is analyzed in real time using the Rainflow counting method, and combined with the material's SN curve, the remaining service life of the cooling fan bearing and water pump mechanical seal is dynamically predicted.

[0053] By employing a non-dominated sorting genetic algorithm with an elitist strategy, this module can objectively find a series of non-dominated Pareto optimal solutions, revealing the optimal trade-off between different objectives and avoiding one-sided decisions that may result from single-objective optimization.

[0054] The system can automatically select the optimal solution from the Pareto solution set based on the operational priorities of the ship's current voyage stage (such as prioritizing energy conservation during ocean cruises and equipment reliability during port arrivals and departures). This makes the system's behavior strategy no longer fixed, but rather possesses a high degree of situational awareness and dynamic decision-making capabilities, enabling it to intelligently adapt to different task requirements and improve management precision and flexibility.

[0055] By quantifying the cumulative fatigue damage of key components as a long-term reliability indicator into an optimization target, the reinforcement learning model is guided to proactively consider the impact of control actions on the mechanical life of equipment when searching for energy-saving strategies. This reduces equipment losses from the source of control, achieving a balance between energy efficiency improvement and equipment life extension, and significantly reducing the operation and maintenance costs throughout the entire life cycle.

[0056] The adaptive control unit also includes a feedforward prediction module. This feedforward prediction module uses a long short-term memory neural network model to perform time-series analysis on the planned speed and heading provided by the ship's integrated navigation system for the next 15-30 minutes, as well as the future ambient temperature and humidity change data received from satellite meteorological services. It predicts the future heat load change trend that the shared condenser will face and feeds this trend into the reinforcement learning model in advance as contextual information.

[0057] By using LSTM neural networks to perform time-series analysis on future airspeed, environmental, and other data, the trend of heat load changes can be predicted 15-30 minutes in advance. This allows for pre-emptive adjustment of actuators before load shocks occur, completely transforming the traditional "passive response" mode into an "active mitigation" mode, thus improving control quality and system stability.

[0058] By incorporating the ship's navigation plan and real-time weather and other macroscopic task and environmental information as contextual inputs into the control model, the system can understand the impending changes in the ship's operating conditions, thereby formulating more forward-looking control strategies. This represents a leap from state-based control to intelligent decision-making based on the fusion of task and state.

[0059] This avoids abrupt actions by the actuators and drastic fluctuations in the system. This not only improves robustness but also avoids the additional energy waste caused by responding to sudden changes, thus maintaining excellent overall energy efficiency even in dynamic environments.

[0060] The adaptive adjustment energy-saving system for shared marine condensers described in this invention operates on the principle of a dynamic optimization process integrating real-time sensing, forward-looking prediction, intelligent decision-making, collaborative execution, and continuous self-learning. The entire process can be divided into the following steps:

[0061] Step 1: Comprehensive State Perception and Future Trend Prediction

[0062] 1. Real-time data acquisition: Sensor array 2, deployed at key nodes of the common condenser, starts working and continuously collects operating parameters such as temperature, pressure, and flow rate.

[0063] 2. State Calculation: The Adaptive Control Unit 3 receives this data and performs the "State Awareness and Modeling" step to calculate key state indicators such as the current total heat load rate and real-time energy efficiency ratio of the system.

[0064] 3. Forward-looking prediction: Simultaneously, the feedforward prediction module is activated. It uses an LSTM neural network to perform time-series analysis on data such as ship speed and ambient temperature obtained from ship navigation and weather services for the next 15-30 minutes, predicting the upcoming heat load change trend of the shared condenser.

[0065] Step 2: Multi-objective optimization and intelligent strategy generation

[0066] 4. Setting Optimization Objectives: The multi-objective optimization decision-making submodule begins operation. It searches for a Pareto optimal solution set among three objectives: total system power consumption, fatigue damage of key components, and condenser terminal temperature difference stability. Subsequently, based on the operational priorities of the ship's current voyage phase (e.g., energy saving for ocean voyages, equipment reliability for port entry and exit), it automatically selects an optimal compromise from the solution set as the objective for this round of control.

[0067] 5. Generate control commands: Input the real-time status indicators and predicted trends obtained in the first step, as well as the current target determined in the second step, into the pre-trained reinforcement learning model.

[0068] Based on this comprehensive information, the Actor network outputs a continuous set of optimal control commands with the same dimensions and number of actuators (i.e., the frequency of the cooling fan inverter, the frequency of the water pump inverter, and the opening degree of each regulating valve).

[0069] The Critic network evaluates the long-term value of the action and guides the update direction of the Actor network.

[0070] In this process, the reward function acts as a "guide", ensuring that the generated instructions reduce energy consumption (P_total), suppress device oscillations (Σ|Δu_i|), and maintain high thermodynamic efficiency (|T_condT_cond_opt|).

[0071] Step 3: Precise Collaborative Execution and System Response Acquisition

[0072] 6. Command Issuance and Execution: The adaptive control unit 3 issues the generated optimal set of control commands to each actuator (cooling fan inverter, water pump inverter, regulating valve). These actuators coordinate their actions according to the commands to achieve precise regulation of the condensing operation.

[0073] 7. Decoupling Control (Parallel): During the adjustment process, if there is a sudden change in the load of different heat sources (such as the main engine and the air conditioning system), the adaptive control unit will run the decoupling control algorithm in parallel and independently adjust the electric regulating valves on each heat exchange circuit to smooth out the heat load impact and ensure system stability.

[0074] Step 4: Experience Accumulation and Model Evolution (Self-Learning)

[0075] 8. Effect Evaluation and Experience Storage: After implementing the new control strategy, sensor array 2 collects the system's response data again. Adaptive control unit 3 stores this "state-action-new state-reward" sequence as a complete set of experience data in the experience playback buffer.

[0076] 9. Periodic Model Fine-Tuning: The system periodically or irregularly extracts data from the experience replay buffer to fine-tune the reinforcement learning model online. This enables the model to learn performance changes caused by equipment aging and scaling, as well as hidden patterns in new operating environments, thereby achieving continuous self-optimization and adaptation.

[0077] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An adaptive adjustment energy-saving system for a shared marine condenser, characterized in that, include: At least one common condenser, the fluid circuit of which is connected to the ship’s main engine, auxiliary generator and air conditioning system, is used to receive and dissipate the combined heat load of several heat sources; A sensor array, including temperature sensors, pressure sensors, and flow sensors installed at the inlet and outlet of each fluid passage of the common condenser, is used to collect the operating parameters of the system in real time. An adaptive control unit, the signal input of which is connected to the sensor array, and the control output of which is connected to the cooling fan inverter, water pump inverter and regulating valve of the common condenser; The adaptive control unit is configured to perform the following steps: a. State perception and modeling: Based on the real-time data from the sensor array, calculate the total heat load rate and real-time energy efficiency ratio of the current system; b. Strategy generation and optimization: The calculated total heat load rate and real-time energy efficiency ratio are input into a pre-trained reinforcement learning model. The reinforcement learning model uses minimizing the overall system energy consumption as the reward function and outputs the optimal set of control commands for the cooling fan, water pump and regulating valve. c. Execution and Adaptation: Execute the optimal set of control commands and collect the system response again through the sensor array. Store the interactive experience data in the experience playback buffer for periodic online fine-tuning of the reinforcement learning model, thereby achieving continuous adaptive optimization of the system.

2. The adaptive adjustment ship shared condenser energy-saving system according to claim 1, characterized in that, The high-temperature cooling circuit from the main engine and the medium-temperature refrigerant circuit from the air conditioning system exchange heat with the circulating working fluid of the common condenser through independent plate heat exchangers; the adaptive control unit independently adjusts the electric regulating valves on each heat exchange circuit through a decoupling control algorithm.

3. The adaptive adjustment ship shared condenser energy-saving system according to claim 1, characterized in that, In step b, the reinforcement learning model is a deep deterministic policy gradient network based on the Actor-Critic framework. The Actor network serves as the policy function, with its input being a normalized system state vector and its output being a continuous action vector with the same dimension as the number of actuators. This action vector, after inverse normalization, is directly used as the frequency setpoint of the inverter and the opening command of the regulating valve. The Critic network serves as the value function, used to evaluate the expected reward of the Actor network's output action and guide the update direction of the Actor network.

4. The adaptive adjustment ship shared condenser energy-saving system according to claim 3, characterized in that, In step b, the reward function R(t) is composed of the following multi-factor weighted sum in each control period t: R(t) = -[α·P_total(t) + β·Σ|Δu_i(t)| + γ·|T_cond(t)T_cond_opt|]; where P_total(t) is the total power consumption of the cooling fan and water pump in this cycle; Σ|Δu_i(t)| is the sum of the absolute values ​​of the changes in control commands of all actuators in this cycle and the previous cycle, used to penalize action oscillations; |T_cond(t)T_cond_opt| is the absolute value of the difference between the current condensing temperature and the optimal condensing temperature dynamically calculated by the energy-saving model, used to ensure thermodynamic efficiency; α, β, γ are pre-set positive weighting coefficients used to balance the relationship between energy saving, equipment life and system stability.

5. The adaptive adjustment ship shared condenser energy-saving system according to claim 1, characterized in that, The adaptive control unit also includes a multi-objective optimization decision submodule. This multi-objective optimization decision submodule uses a non-dominated sorting genetic algorithm with an elitist strategy to find the Pareto optimal solution set among the three objectives of total system power consumption, cumulative fatigue damage of key components, and condenser terminal difference stability. Based on the operational priority of the ship's current voyage stage, it automatically selects the best compromise solution from the solution set as the objective constraint for training the reinforcement learning model.

6. The adaptive adjustment ship shared condenser energy-saving system according to claim 5, characterized in that, The online estimation method for the cumulative fatigue damage of the key components is as follows: based on Miner's linear cumulative damage theory, the shaft load spectrum derived from control commands and system pressure is analyzed in real time using the Rainflow counting method, and combined with the material's SN curve, the remaining service life of the cooling fan bearing and water pump mechanical seal is dynamically predicted.

7. The adaptive adjustment ship shared condenser energy-saving system as described in claim 1, characterized in that, The adaptive control unit also includes a feedforward prediction module. This feedforward prediction module uses a long short-term memory neural network model to perform time-series analysis on the planned speed and heading provided by the ship's integrated navigation system for the next 15-30 minutes, as well as the future ambient temperature and humidity change data received from satellite meteorological services. It predicts the future heat load change trend that the shared condenser will face and feeds this trend into the reinforcement learning model in advance as contextual information.

Images