Predictive dynamic weight ship hybrid power system performance evaluation method and system

Abstract

The invention belongs to the field of ship hybrid power system control, and particularly discloses a ship hybrid power system performance evaluation method and system based on predictive dynamic weight. Comprising the following steps: constructing a multi-dimensional performance evaluation index system and ship perception data, and constructing a digital twin model mapped with a physical ship hybrid power system to deduce a system response state in a future preset time window, generating a future state vector including a future battery electric quantity trend, a future equipment thermal load state and a future energy consumption emission predicted value; building a predictive dynamic weight decision module based on LSTM and deep DDPG, and making a decision; and performing weighted summation calculation on the predictive dynamic weight vector and each index standardized value, and generating a targeted control optimization instruction according to the calculated value. According to the method, comprehensive performance evaluation and optimization support which covers the whole life cycle, has a prospective view and gives consideration to multi-target cooperation can be provided for the ship hybrid power system.

Description

Technical Field

[0001] This invention belongs to the field of marine hybrid power system control, and more specifically, relates to a predictive dynamic weighting method and system for evaluating the performance of marine hybrid power systems. Background Technology

[0002] With increasingly stringent emissions regulations from the International Maritime Organization (IMO) and the growing demand for "green shipping," marine hybrid power systems (integrating diesel engines, LNG, lithium batteries, fuel cells, etc.) have become a core development direction for the industry. However, the performance evaluation of such systems involves multiple interdependent dimensions, including economics, environmental protection, and technology, making the evaluation extremely complex.

[0003] Existing technical solutions, such as Chinese patent CN120509525A, while proposing energy optimization management methods, focus on generating real-time control commands and often employ fixed weights or simple feedback mechanisms based solely on the current state. These existing technologies suffer from the following main drawbacks: Evaluation lag: Adjusting weights based solely on "current" or "historical" data fails to predict future operating conditions (such as entering emission control zones or encountering severe sea conditions), resulting in slow system response to sudden changes in operating conditions and hindering "preventative" optimization. Rigid benchmarks: Data standardization typically uses fixed maximum / minimum values ​​as benchmarks, ignoring the objective fact of ship equipment aging over time (such as battery capacity degradation). This leads to distorted evaluation results for older ships, failing to reflect their relative optimal performance under current health conditions. Lack of global perspective: Limited training samples for single-ship agents and a lack of overall planning capabilities for future voyages make it difficult to achieve global optimization across the entire voyage.

[0004] Therefore, there is an urgent need for a performance evaluation method that can integrate future state predictions, has forward-looking decision-making capabilities, and adapts to changes throughout the entire life cycle of equipment. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a predictive dynamic weighting method and system for evaluating the performance of marine hybrid power systems. It constructs a digital twin model to perform real-time simulations of future navigation conditions and integrates the simulation results into a reinforcement learning decision-making mechanism based on LSTM-DDPG. This achieves a fundamental leap from "reactive evaluation based on the current situation" to "predictive evaluation based on the future." Simultaneously, by combining a dynamic standardization method that allows for drift with equipment health status, it effectively solves the problems of decision-making errors and evaluation distortions caused by evaluation lag and rigid benchmarks in existing technologies. This provides comprehensive performance evaluation and optimization support for marine hybrid power systems, covering the entire lifecycle, possessing a forward-looking perspective, and considering multi-objective collaboration.

[0006] To achieve the above objectives, according to one aspect of the present invention, a method for performance evaluation of a ship hybrid power system with predictive dynamic weights is proposed, comprising the following steps: Step 1: Construct a multi-dimensional performance evaluation index system and ship perception data. The evaluation index system includes economic efficiency, environmental protection, technicality and reliability. The ship perception data includes the current status data of the ship's hybrid power system and external environment data. The above data are then standardized. Step 2: Construct a digital twin model that maps to the physical ship hybrid power system. This model includes a ship motion dynamics model and an energy flow model. Based on the future environmental prediction data sequence obtained from weather forecasts and the future navigation mission sequence of the ship's navigation plan, the current state data is used as the initial boundary condition. The current state data, along with the future environmental prediction data sequence and the future navigation mission sequence, are input into the digital twin model to deduce the system response state within a future preset time window and generate a future state vector that includes the future battery charge trend, the future equipment heat load state, and the future energy consumption and emission prediction values. Step 3: Construct a predictive dynamic weight decision module based on LSTM and DDPG. The current state data and external environment data obtained in step S1 are fused with the future state vector generated in step 2 to construct an enhanced state vector. The time series features of the enhanced state vector are extracted using the LSTM network, and the decision is made through the Actor-Critic network architecture. Step 4: The predictive dynamic weight vector output in step S3 is weighted and summed with the standardized values ​​of each indicator. Based on the calculated value, targeted control optimization instructions are generated to complete the closed-loop management of the ship from assessment to optimization.

[0007] As a further preferred embodiment, in step one, the current status data includes the power battery health status, fuel cell performance degradation, engine real-time efficiency point, and current mission stage code obtained through the engine room monitoring system; the external environment data includes real-time wind speed, wave level, and water flow speed obtained through the ship's weather station and sensors, as well as real-time fuel price, liquefied natural gas price, and emission control zone status identifier of the current navigation area received through the satellite communication system.

[0008] As a further optimization, the real-time raw data of each performance indicator is obtained and dimensionless processing is performed using the dynamic range standardization method; for benefit-type indicators, their standardized values ​​are calculated; for cost-type indicators, their inverse standardized values ​​are calculated. In this way, all indicators are uniformly mapped to the numerical range of [0,1].

[0009] As a further preferred option, step two specifically includes the following steps: (21) Deploy a digital twin model corresponding one-to-one with the physical ship in a shore-based or high-performance edge computing unit. This model includes: The hull resistance model is based on the Holtrop-Mennen method or a CFD pre-computation database and calculates real-time resistance based on speed, draft, and wind and wave parameters. Propulsion system model: includes propeller open-water characteristic curves, main engine universal characteristic curves, motor efficiency map, and battery equivalent circuit model; Energy management strategy model: The twin operates with the same energy management logic as the real ship; (22) Read the navigation plan and wind, wave and current data on the route from the electronic chart system, and use the Runge-Kutta method to iteratively solve the digital twin model starting from the current measured state, so as to generate a future state vector containing the future battery power trend, the future equipment heat load state and the future energy consumption and emission prediction values.

[0010] As a further preferred embodiment, the derivation process of the digital twin model includes: Let the current time be The predicted step size is Construct future state vector Computational model: in, Represents the digital twin dynamics mapping function. The measured value of the system state at the current moment is collected. For future meteorological and environmental forecast data within a preset time window, Data for future voyage plans.

[0011] As a further preferred option, step three specifically includes the following steps: (31) The current state data and external environment data obtained in step one are fused with the future state vector generated in step two to construct an enhanced state vector, so that the decision module can simultaneously perceive the current state of the system and the future evolution trend. (32) The time series features of the enhanced state vector are extracted using the LSTM network, and the decision is made through the Actor-Critic network architecture. The Actor network outputs a four-dimensional weight vector that satisfies the normalization condition based on the input time series features, which correspond to the real-time weights of economy, environmental protection, technology and reliability respectively. The Critic network evaluates the expected cumulative reward value of the weight action in the current and future projection states, and continuously optimizes the weight generation strategy by maximizing the cumulative reward.

[0012] As a further preferred embodiment, the enhanced state vector The specific construction form is as follows: In the formula, It is a real-time state vector that includes current wind speed, wave height, oil price, mission stage, and equipment health. To predict the future state of the system, This represents the characteristics of future environmental change rates.

[0013] As a further preferred embodiment, during the training process of the predictive dynamic weight decision module, its reward function... A trend penalty term based on twin inference was introduced: in, To standardize operating costs, To standardize total emissions, For overall system efficiency; Basic weighting coefficients; For the prediction penalty function based on the results of digital twin inference, This is the reward coefficient.

[0014] As a further preferred option, the following also includes: When the evaluation results show that the current overall score is lower than the preset threshold, the system automatically identifies the key indicators that led to the low score: If the primary reason is an excessively low environmental score and the digital twin simulation indicates that the ship is about to enter an emission control zone, an automatic command will be generated to prioritize switching to fuel cell or lithium battery power mode. If the primary reason is an excessively low economic score and the ship is in open sea, it is recommended to switch to diesel engine direct drive mode and optimize to economical speed.

[0015] According to another aspect of the present invention, a predictive dynamic weighted performance evaluation system for a marine hybrid power system is also provided, comprising: The data acquisition and multi-dimensional sensing module is used to collect real-time data on the ship's power system status, mission data, market data, and environmental data through sensor networks, weather stations, and communication interfaces. The digital twin simulation module, connected to the data acquisition and multi-dimensional perception module, is used to store the digital twin model of the ship's hybrid power system, and, in conjunction with the received weather forecasts and navigation plans, to perform ultra-real-time simulation and simulation of the system's future operating state, and output the future state vector. The predictive dynamic weight decision module is connected to the data acquisition and multidimensional perception module and the digital twin inference module, respectively. It has a built-in intelligent agent based on the LSTM-DDPG algorithm to receive current state data and future state vectors. Through enhanced state space analysis and reinforcement learning strategies, it calculates and outputs the predictive dynamic weights of each evaluation dimension in real time. The data standardization processing module is used to receive raw performance index data and perform range standardization processing on the data using dynamic boundaries based on device health status. The weighted comprehensive evaluation module is used to receive standardized indicator data and predictive dynamic weights, calculate the comprehensive performance score, and display the score through the user interface or send optimization instructions to the ship control system.

[0016] In summary, compared with the prior art, the above-described technical solutions conceived by this invention mainly possess the following technical advantages: 1. This invention utilizes digital twin technology combined with weather forecasts and flight plans to simulate and predict future system states in advance (such as predicting impending battery depletion or emissions exceeding limits), and uses this "future state" as input to the intelligent agent. This enables the assessment system to "foresee" potential risks and guide the system to adjust its weighting strategy in advance (such as increasing environmental protection weights to reserve power), thereby avoiding a passive situation when responding to emergencies and significantly improving the system's environmental adaptability and safety.

[0017] 2. This invention proposes a dynamic benchmark standardization method based on State of Health (SOH). It utilizes a digital twin model to calculate the theoretical performance limits of equipment at its current aging level in real time, and dynamically adjusts the evaluation benchmark accordingly. This method effectively distinguishes between "performance degradation due to improper operation" and "physical degradation due to equipment aging," ensuring that the evaluation results remain objective and fair throughout the ship's entire lifecycle, providing a more accurate basis for operation and maintenance management decisions.

[0018] 3. This invention improves the reward function of reinforcement learning by introducing a future trend penalty term based on Siamese inference. This enables the evaluation system to consider not only immediate gains but also long-term comprehensive returns when weighing the weights of various indicators. The system can automatically suppress short-sighted behavior that sacrifices future safety or compliance for short-term cost savings, guiding the hybrid power system to achieve dynamic balance and globally optimal synergy across multiple objectives throughout the entire flight range. Attached Figure Description

[0019] Figure 1 This is a flowchart of a predictive dynamic weighting method for evaluating the performance of a ship hybrid power system, provided in an embodiment of the present invention. Figure 2 This is a flowchart of the performance evaluation method based on dynamic weights involved in the embodiments of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0021] like Figure 1 and Figure 2 As shown, this embodiment of the invention provides a method for performance evaluation of a ship hybrid power system with predictive dynamic weights, including the following steps: Step S1: Establish a multi-dimensional performance evaluation index system and data perception channel.

[0022] In this step, an evaluation index system is constructed that includes four dimensions: economy, environmental protection, technology, and reliability, and specific quantitative indicators for each dimension are defined. The ship's multi-source sensing system collects real-time data on the current status of the ship's hybrid power system and external environmental data. The current status data includes the state of health (SOH) of the power battery, the performance degradation of the fuel cell, the real-time efficiency point of the engine, and the current mission stage code obtained through the engine room monitoring system. The external environmental data includes real-time wind speed, wave level, and water current speed obtained through the ship's weather station and sensors, as well as real-time fuel price, liquefied natural gas price, and the emission control area (ECA) status indicator of the current navigation area received through the satellite communication system.

[0023] Traditional assessments often rely on offline data or single-dimensional dashboard data, leading to information silos. This step establishes a multi-dimensional performance evaluation index system and data perception channels to ensure that the assessment is "data-driven" rather than based on empirical assumptions. More specifically, this invention first constructs an index system encompassing four dimensions: economic efficiency, environmental friendliness, technicality, and reliability. Economic indicators (Total annual cost, investment payback period), environmental performance indicators (Carbon dioxide emission intensity, total nitrogen oxide emissions), technical indicators (System energy efficiency, power response characteristics) and reliability indicators (Mean Time Between Failures (MTBF) for critical equipment). Total annual cost includes real-time fuel / electricity consumption costs, maintenance costs based on operating hours, and amortized equipment depreciation. Dynamic payback period is the estimated time to recover system investment based on current fuel cost savings. System energy efficiency is the ratio of propulsion power to source input power (fuel calorific value + battery output). Power response time is the time required for the system to respond from idle to rated power, reflecting maneuverability. Mean Time Between Failures (MTBF) is the estimated remaining trouble-free operating time based on current equipment condition monitoring data (e.g., vibration, temperature).

[0024] The data sensing channel collects the following four types of variables in real time: environmental variables This includes real-time wind speed, wave intensity, and current speed obtained from the ship's onboard weather station. (Task variables) This includes the current mission phase code (berthing / arrival / departure / cruise) and ground speed. Market and policy variables. This includes real-time fuel prices and current marine emission limit levels. System state variables. This includes the health status of the power battery, the performance degradation of the fuel cell, the real-time operating point of the engine, and the score of the previous cycle, all obtained through the cabin monitoring system.

[0025] In this step, the system integrates multi-source data via Industrial Ethernet or CAN bus (NMEA 2000 protocol): Internal data is collected using the engine room monitoring and alarm system (AMS) and battery management system (BMS) to measure the state of charge (SOC), state of health (SOH), voltage, current, and temperature of the power battery; and to measure the engine speed, torque, fuel consumption rate, and exhaust temperature of the diesel engine / fuel cell. The data sampling frequency is set to 1Hz-10Hz.

[0026] External environment perception is achieved by acquiring relative wind speed and direction through a shipborne ultrasonic anemometer; by acquiring relative speed to water through a Doppler log (DVL); and by acquiring significant wave height and wave period through wave radar.

[0027] The latest energy market prices (fuel / LNG / shore power prices) and Emission Control Area (ECA) electronic fence data released by the Maritime Authority are periodically obtained (e.g., every 10 minutes) through the VSAT satellite communication system.

[0028] Step S2: Advance projection of future states based on digital twin models.

[0029] Hybrid power systems exhibit significant thermal and chemical inertia, and the navigation environment is time-varying. For example, if insufficient battery power is only discovered after the ship has entered an emission control area, the diesel engine must be started, leading to violations or hefty fines. Such impending risks cannot be identified solely based on the "current state." Therefore, this invention addresses the "hysteresis" of physical system responses by using a digital twin model to anticipate future states. The system utilizes the digital twin to "fast-forward" time in virtual space.

[0030] In this step, a digital twin model mapping to the physical ship hybrid power system is constructed. This model includes a ship motion dynamics model and an energy flow model. The system receives future environmental prediction data sequences from a weather forecast system and future navigation mission sequences generated by a ship navigation planning system. The current state data is used as the initial boundary conditions and input into the digital twin model along with the future environmental prediction data sequences and future navigation mission sequences. The digital twin model performs ultra-real-time simulation in virtual space to deduce the system response state within a future preset time window and generate a future state vector containing future battery charge trends, future equipment thermal load states, and future energy consumption and emission prediction values. More specifically, in this step, mathematical models corresponding one-to-one with the physical ship are deployed on shore or in high-performance edge computing units. This model includes: The hull resistance model is based on the Holtrop-Mennen method or a CFD pre-calculation database, and calculates real-time resistance based on speed, draft, and wind and wave parameters. The propulsion system model includes the propeller open-water characteristic curve, the main engine universal characteristic curve (Fuel Map), the motor efficiency map, and the battery equivalent circuit model (ECM). The Energy Management Strategy (EMS) model runs in a twin with energy management logic (such as rule-based logic or optimization algorithms) that is completely identical to that of the real ship.

[0031] The ultra-real-time simulation is performed based on the mathematical model established above. The specific process is as follows: First, set the simulation time window. (e.g., the next 30 minutes) and simulated step size (e.g., 1 minute).

[0032] Then, the system loads future boundary conditions and reads the navigation plan (planned speed and heading for the next 30 minutes) from the Electronic Chart System (ECDIS) and the gridded weather forecast (wind, wave and current data along the route for the next 30 minutes) provided by the weather service provider.

[0033] Next, state iteration calculations are performed, using the measured state at the current moment. Starting with the Runge-Kutta method, an iterative solution is performed: in, This represents the system state at the k-th simulation step. This represents the system state at the (k+1)th simulation step.

[0034] Finally, generate the future state vector. Finally, a future state vector containing time-series information is output, and the following key prediction values ​​are extracted: the battery SOC change curve in the next 30 minutes, the thermal load trend of whether the motor or engine temperature will exceed the limit, the ship will enter the ECA area, and the compliance prediction of the cumulative emissions at this time.

[0035] More specifically, in this step, the measured state of the system at the current time t is obtained. .

[0036] Setting boundary conditions: Future environment prediction sequence Future navigation mission sequence The above data is then input into the digital twin model. This model includes a ship resistance model, a propulsion model, and an energy management strategy model. The model uses... Using 1 minute as the step size, ultra-real-time simulation of the future. The operational status within a time window (e.g., the next 30 minutes). The extrapolation formula specifically includes: in, The system state vector for future moments is derived, with a focus on the predicted battery state of charge, predicted cumulative emissions, and predicted equipment thermal load. For the system dynamics mapping function of the digital twin model, The measured value of the system state at the current moment is collected. For future meteorological and environmental forecast data within a preset time window, Data for future voyage plans.

[0037] This step expands the time dimension of the evaluation from t to... This enables the system to identify potential energy crises or compliance risks, providing a physical basis for the advance adjustment of subsequent weights and achieving a qualitative change from "passive response" to "proactive prevention".

[0038] Step S3: Construct a predictive reinforcement learning dynamic weight decision mechanism.

[0039] In this step, a predictive dynamic weight decision-making module based on Long Short-Term Memory (LSTM) network and Deep Deterministic Policy Gradient (DDPG) is constructed. First, the state space is reconstructed by fusing the current state data and external environment data obtained in step S1 with the future state vector generated in step S2 to construct an enhanced state vector, enabling the decision-making module to simultaneously perceive the current state and future evolution trend of the system. Subsequently, the time series features of the enhanced state vector are extracted using the LSTM network, and the decision is made through an Actor-Critic network architecture. The Actor network outputs a four-dimensional weight vector that meets the normalization condition based on the input time series features, corresponding to the real-time weights of economy, environmental protection, technology, and reliability, respectively. The Critic network evaluates the expected cumulative reward value of the weight action in the current and future projected states, and continuously optimizes the weight generation strategy by maximizing the cumulative reward.

[0040] More specifically, this step utilizes artificial intelligence algorithms to solve the decision-making problem of "how to optimally allocate weights." This embodiment employs an improved LSTM-DDPG (Long Short-Term Memory Network-Deep Deterministic Policy Gradient) algorithm. First, the input (State) of the enhanced state space reconstruction agent is no longer merely the current sensor reading, but rather a fusion of panoramic information from the past, present, and future. The enhanced state vector is defined. , In the formula, The current real-time observation vector obtained from step one This refers to the real-time status at the current moment (oil price, wind speed, mission stage, etc.). This refers to the key future states projected in step two (such as the predicted final value of SOC). This represents the characteristics of future environmental change rates (such as the trend of "rapidly increasing wind and waves"). It is the historical operating pattern features extracted by the LSTM unit (such as the load volatility over the past hour).

[0041] The network architecture and training adopt an LSTM-DDPG architecture. The input layer receives data in time-series format. By extracting temporal features through LSTM layers, the inertia and delay issues in the navigation process are addressed. The Actor network outputs action vectors. That is, the current weight allocation: Constraints: In the above formula, These correspond to dynamic weight values ​​for economic efficiency, environmental friendliness, technicality, and reliability, respectively.

[0042] The input layer of the Actor network (policy network) receives... The hidden layer contains two fully connected layers (e.g., 128 and 64 neurons), using ReLU activation. The output layer contains four neurons, activated using the Softmax function to ensure the four weights of the output are evenly distributed. The sum is strictly 1, and each weight is between [0,1].

[0043] The Critic network (value network) is responsible for evaluating the "goodness" or "badness" of the weights. The input layer simultaneously receives... And the actions (weights) output by the Actor. The output layer outputs a scalar Q value, which represents the expected return of the weight allocation strategy in the long run.

[0044] Design a predictive reward function to guide the agent to make decisions that benefit long-term development. Defined as: in, To standardize operating costs, To standardize total emissions, For overall system efficiency; Basic weighting coefficients; This is a predictive penalty function based on the results of digital twin inference, for example, if the twin inference shows... If the amount of electricity in the system falls below the safe threshold (SOC < 20%) or emissions exceed the standard, the function will output a large negative value. This is the adjustment coefficient for the trend term.

[0045] In this invention, the standardized cost is based on the current moment. ,emission and efficiency Calculations. The goal is to achieve lower costs, lower emissions, and higher efficiency, with greater rewards (smaller negative values).

[0046] In this invention, future trend penalty It is a piecewise penalty function, if (Predicting future power depletion), then .like (If it is predicted that the vehicle will enter the emission zone but not use pure electric power), then .

[0047] Pre-training is performed using historical navigation data and a large number of simulated samples generated from twin models. During actual ship operation, the Actor network outputs the optimal weight vector at the current moment in milliseconds based on the real-time input augmented state. In this invention, LSTM is introduced to handle time-series dependencies (remembering history), and the future state of twin data is introduced to increase foresight (predicting the future). The DDPG algorithm is suitable for handling continuous action spaces (where weights are continuous numerical values).

[0048] Step S4: Dynamic benchmark standardization of performance index data.

[0049] In this step, real-time raw data of each performance index is obtained and dimensionless processing is performed using the dynamic range standardization method; for benefit-type indicators, their standardized values ​​are calculated; for cost-type indicators, their inverse standardized values ​​are calculated; during this process, the digital twin model in step S2 is used to assess the current aging degree of the equipment, and the maximum and minimum allowable values ​​in the standardization formula are dynamically adjusted to ensure that the standardization results reflect the relative performance level of the equipment at the current life cycle stage, and all indicators are uniformly mapped to the numerical range of [0,1]. The standardized calculation formula is as follows: For positive indicators (the larger the better): For contrarian indicators (the smaller the better) If the real-time value exceeds the dynamic boundary (such as sensor drift), it is truncated to the [0,1] interval.

[0050] To address the assessment distortion caused by equipment aging, a method based on equipment health status is adopted. The dynamic range standardization method for drift. Its calculation formula is shown below: In the formula: Let be the original measurement value of the i-th index at time t. This is the standardized dimensionless value. and It is not a fixed constant, but rather determined by the digital twin model based on the current device. Dynamic boundaries for real-time computation.

[0051] For benefit-related indicators (such as efficiency), use the above formula directly; for cost-related indicators (such as emissions), take... In one embodiment of the invention, for a new battery, It could be 1000kWh; for an aged battery with SOH=80%, the digital twin model will... The value is automatically corrected to 800 kWh. If the actual measured output is 790 kWh, the score is 0.79 under the old standard, while the score under this method is 790 / 800=0.98, which objectively reflects its excellent performance under the current conditions.

[0052] More specifically, in this invention, the aging sub-model in the digital twin model is used to calculate the current physical limits of the device in real time based on the currently monitored cumulative running time, charge and discharge cycle count, etc.

[0053] Step S5: Weighted comprehensive evaluation and closed-loop feedback control.

[0054] In this step, the predictive dynamic weight vector output in step S3 and the standardized values ​​of each index output in step S4 are weighted and summed to obtain the comprehensive performance score at the current moment. The system compares the score with the preset performance threshold. When the score is lower than the threshold or a significant decline in the score is predicted in the future, targeted control optimization instructions are generated based on the score results, including adjusting the power distribution ratio or suggesting a change in speed. The instructions are then fed back to the ship hybrid power control system to complete the closed-loop management from evaluation to optimization.

[0055] In this embodiment, the comprehensive scoring calculation model includes: The system updates the score every second and displays it on the human-computer interface (HMI) as a radar chart and trend graph. A score warning threshold is set. when If twin analysis predicts a significant drop in future scores, the system will trigger optimization suggestions.

[0056] Currently, diesel engines are being used, resulting in good economic performance and a high score. The twin model predicts that the system will enter the ECA phase in 30 minutes. If the current situation remains unchanged, the environmental score will plummet, causing the total score to fall below the threshold. The decision-making module should proactively increase the environmental weighting. The increased weighting of environmental protection in the overall scoring formula caused a sudden drop in the current score (diesel engine mode). The control system detected this score decrease and automatically searched for strategies to improve it—namely, proactively starting the diesel generator to charge the battery (although this increases current fuel consumption, it reserves energy for future electric navigation, aligning with the new weighting preferences). Under severe sea state warnings, the predicted wave height increases, thus enhancing the reliability weighting. The control system limits the high-load operation of individual units, starts the standby unit to connect to the grid, and increases the system redundancy. Although some economic efficiency is sacrificed, the overall high score is maintained.

[0057] This invention constructs a digital twin model to perform real-time simulations of future navigation conditions and integrates the simulation results into a reinforcement learning decision-making mechanism based on LSTM-DDPG. This achieves a fundamental leap in the evaluation system from "reactive evaluation based on the current situation" to "predictive evaluation based on the future." At the same time, by combining a dynamic standardization method that drifts with the health status of equipment, it effectively solves the problems of decision-making errors and evaluation distortions caused by evaluation lag and rigid benchmarks in existing technologies. It can provide comprehensive performance evaluation and optimization support for marine hybrid power systems that covers the entire life cycle, has a forward-looking vision, and takes into account multi-objective collaboration.

[0058] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A predictive dynamic weighting method for evaluating the performance of a ship hybrid power system, characterized in that, Includes the following steps: Step 1: Construct a multi-dimensional performance evaluation index system and ship perception data. The evaluation index system includes economic efficiency, environmental protection, technicality and reliability. The ship perception data includes the current status data of the ship's hybrid power system and external environment data. The above data are then standardized. Step 2: Construct a digital twin model that maps to the physical ship hybrid power system. This model includes a ship motion dynamics model and an energy flow model. Based on the future environmental prediction data sequence obtained from weather forecasts and the future navigation mission sequence of the ship's navigation plan, the current state data is used as the initial boundary condition. The current state data, along with the future environmental prediction data sequence and the future navigation mission sequence, are input into the digital twin model to deduce the system response state within a future preset time window and generate a future state vector that includes the future battery charge trend, the future equipment heat load state, and the future energy consumption and emission prediction values. Step 3: Construct a predictive dynamic weight decision module based on LSTM and DDPG, which integrates the current state data, external environment data and the future state vector generated in step 2 to construct an enhanced state vector. Use the LSTM network to extract the time series features of the enhanced state vector and make decisions through the Actor-Critic network architecture. Step four involves weighting and summing the output predictive dynamic weight vector with the standardized values ​​of each indicator, and generating targeted control optimization instructions based on the calculated values ​​to complete the closed-loop management of the ship from assessment to optimization.

2. The method for performance evaluation of a ship hybrid power system with predictive dynamic weights according to claim 1, characterized in that, In step one, the current status data includes the health status of the power battery, the performance degradation of the fuel cell, the real-time efficiency point of the engine, and the current mission stage code obtained through the engine room monitoring system; the external environment data includes real-time wind speed, wave level, and water flow speed obtained through the ship's weather station and sensors, as well as real-time fuel price, liquefied natural gas price, and emission control zone status identifier of the current navigation area received through the satellite communication system.

3. The method for performance evaluation of a ship hybrid power system with predictive dynamic weights according to claim 2, characterized in that, The real-time raw data of each performance indicator are obtained and dimensionless processing is performed using the dynamic range standardization method; for benefit-type indicators, their standardized values ​​are calculated. For cost-related indicators, calculate their inverse standardized values. In this way, all indicators are uniformly mapped to the numerical range of [0,1].

4. The method for performance evaluation of a ship hybrid power system with predictive dynamic weights according to claim 1, characterized in that, Step two specifically includes the following steps: (21) Deploy a digital twin model corresponding one-to-one with the physical ship in a shore-based or high-performance edge computing unit. This model includes: The hull resistance model is based on the Holtrop-Mennen method or a CFD pre-computation database and calculates real-time resistance based on speed, draft, and wind and wave parameters. Propulsion system model: includes propeller open-water characteristic curves, main engine universal characteristic curves, motor efficiency map, and battery equivalent circuit model; Energy management strategy model: The twin operates with the same energy management logic as the real ship; (22) Read the navigation plan and wind, wave and current data on the route from the electronic chart system, and use the Runge-Kutta method to iteratively solve the digital twin model starting from the current measured state, so as to generate a future state vector containing the future battery power trend, the future equipment heat load state and the future energy consumption and emission prediction values.

5. The method for performance evaluation of a ship hybrid power system with predictive dynamic weights according to claim 4, characterized in that, The derivation process of the digital twin model includes: Let the current time be The predicted step size is Construct future state vector Computational model: in, Represents the digital twin dynamics mapping function. The measured value of the system state at the current moment is collected. For future meteorological and environmental forecast data within a preset time window, Data for future voyage plans.

6. The method for performance evaluation of a ship hybrid power system with predictive dynamic weights according to claim 1, characterized in that, Step three specifically includes the following steps: (31) The current state data and external environment data obtained in step one are fused with the future state vector generated in step two to construct an enhanced state vector, so that the decision module can simultaneously perceive the current state of the system and the future evolution trend. (32) The time series features of the enhanced state vector are extracted using the LSTM network, and the decision is made through the Actor-Critic network architecture. The Actor network outputs a four-dimensional weight vector that satisfies the normalization condition based on the input time series features, which correspond to the real-time weights of economy, environmental protection, technology and reliability respectively. The Critic network evaluates the expected cumulative reward value of the weight action in the current and future projection states, and continuously optimizes the weight generation strategy by maximizing the cumulative reward.

7. The method for performance evaluation of a ship hybrid power system with predictive dynamic weights according to claim 6, characterized in that, The enhanced state vector The specific construction form is as follows: In the formula, It is a real-time state vector that includes current wind speed, wave height, oil price, mission stage, and equipment health. To predict the future state of the system, This represents the characteristics of future environmental change rates.

8. The method for performance evaluation of a ship hybrid power system with predictive dynamic weights according to claim 6, characterized in that, During the training process of the predictive dynamic weight decision module, its reward function A trend penalty term based on twin inference was introduced: in, To standardize operating costs, To standardize total emissions, For overall system efficiency; Basic weighting coefficients; For the prediction penalty function based on the results of digital twin inference, This is the reward coefficient.

9. A method for performance evaluation of a ship hybrid power system with predictive dynamic weights according to any one of claims 1-8, characterized in that, Also includes: When the evaluation results show that the current overall score is lower than the preset threshold, the system automatically identifies the key indicators that led to the low score: If the primary reason is an excessively low environmental score and the digital twin simulation indicates that the ship is about to enter an emission control zone, an automatic command will be generated to prioritize switching to fuel cell or lithium battery power mode. If the primary reason is an excessively low economic score and the ship is in open sea, it is recommended to switch to diesel engine direct drive mode and optimize to economical speed.

10. A predictive dynamic weighted performance evaluation system for marine hybrid power systems, characterized in that, include: The data acquisition and multi-dimensional sensing module is used to collect real-time data on the ship's power system status, mission data, market data, and environmental data through sensor networks, weather stations, and communication interfaces. The digital twin simulation module, connected to the data acquisition and multi-dimensional perception module, is used to store the digital twin model of the ship's hybrid power system, and, in conjunction with the received weather forecasts and navigation plans, to perform ultra-real-time simulation and simulation of the system's future operating state, and output the future state vector. The predictive dynamic weight decision module is connected to the data acquisition and multidimensional perception module and the digital twin inference module, respectively. It has a built-in intelligent agent based on the LSTM-DDPG algorithm to receive current state data and future state vectors. Through enhanced state space analysis and reinforcement learning strategies, it calculates and outputs the predictive dynamic weights of each evaluation dimension in real time. The data standardization processing module is used to receive raw performance index data and perform range standardization processing on the data using dynamic boundaries based on device health status. The weighted comprehensive evaluation module is used to receive standardized indicator data and predictive dynamic weights, calculate the comprehensive performance score, and display the score through the user interface or send optimization instructions to the ship control system.

Images