Energy-optimized zero-emission navigation control method and system for electric ships

Abstract

The application discloses a kind of energy optimization-based electric ship zero-emission navigation control method and system, it is related to edge computing technical field, including, according to the strategy instruction, the energy management intelligent agent obtains candidate power distribution scheme by dynamically adjusting energy priority, simultaneously, the energy loss of different candidate power distribution scheme is previewed by multi-objective optimization algorithm, and optimal power distribution scheme is generated;The optimal power distribution scheme is carried out deviation effectiveness check and obtains execution deviation, generates navigation control instruction based on the dynamic calibration of the execution deviation course trajectory and corrects load.This application solves the problem of insufficient collaborative use of multi-source heterogeneous data and single target optimization inefficiency through energy coordination model and collaborative energy optimization network.

Description

TECHNICAL FIELD

[0001] The application relates to an energy optimization-based zero-emission navigation control method and system for an electric ship, and belongs to the technical field of edge computing. BACKGROUND

[0002] Under the low-carbon transformation of the global shipping industry, as the core carrier for realizing zero-emission navigation, the advancement of energy management technology directly determines the energy efficiency level and operation economy of the ship. In recent years, with the large-scale application of clean energy such as lithium batteries and fuel cells, electric ships have shown the development characteristics of'multi-energy coupling, multi-device cooperation and multi-scene adaptation'. Traditional ship energy management relies on rule engines or static optimization algorithms based on experience, which can meet the basic control requirements, but there are problems such as low energy utilization efficiency and dynamic response lag in complex sea conditions. With the integrated development of intelligent navigation and the Internet of Things (IoT), data-driven energy management methods have gradually become a research hotspot.

[0003] The main shortcomings of the prior art are that, on the one hand, traditional methods do not sufficiently mine the spatio-temporal correlation characteristics of multi-source heterogeneous data (such as the state of an electric ship, cross-ship energy interaction and marine environment prediction), resulting in significant errors in global energy demand prediction; on the other hand, the offline optimization framework of the fixed strategy mode cannot respond to the dynamic changes of complex sea conditions and task loads in real time, which easily leads to energy supply and demand imbalance or abnormal wear and tear of equipment. SUMMARY

[0004] The application provides an energy optimization-based zero-emission navigation control method and system for an electric ship, which mainly aims to solve the problems of insufficient collaborative use of multi-source heterogeneous data and low efficiency of single-target optimization.

[0005] To achieve the above-mentioned purpose, the energy optimization-based zero-emission navigation control method for an electric ship provided by the application comprises,

[0006] Obtaining energy optimization collaborative data and preprocessing, extracting multi-source time sequence features based on the preprocessed energy optimization collaborative data;

[0007] Inputting the multi-source time sequence features into an energy collaboration model, performing local feature extraction by an energy feature screening layer, performing nonlinear correlation mining by a correlation mining layer, and generating a global energy demand prediction report;

[0008] Constructing a collaborative energy optimization network based on the multi-source time sequence features, initializing an energy management agent according to the collaborative energy optimization network, and iteratively optimizing the energy management agent according to the global energy demand prediction report to output a strategy instruction;

[0009] According to the strategy instruction, the energy management intelligent agent obtains a candidate power allocation scheme by dynamically adjusting an energy priority, and generates an optimal power allocation scheme by pre-playing energy loss of the different candidate power allocation schemes through a multi-objective optimization algorithm;

[0010] The optimal power allocation scheme is subjected to bias effectiveness verification to obtain an execution bias, and a flight control instruction is generated by dynamically calibrating a flight path trajectory and correcting a load based on the execution bias.

[0011] Optionally, the energy optimization collaborative data includes electric ship state data, cross-ship network data and environment prediction data.

[0012] The preprocessing includes data cleaning, format conversion, deduplication, normalization and outlier processing.

[0013] Optionally, the multi-source time series features are extracted by the following specific steps,

[0014] The electric ship state data is subjected to sliding window statistics and mutual information analysis to obtain power coupling features;

[0015] The cross-ship network data is subjected to dynamic time warping and fluctuation period extraction to obtain energy interaction features;

[0016] The environment prediction data is subjected to STL decomposition and time series standardization to obtain environment disturbance features;

[0017] The power coupling features, the energy interaction features and the environment disturbance features are integrated to generate multi-source time series features.

[0018] Optionally, the energy collaborative model is constructed by the following specific steps,

[0019] A time fusion Transformer and a graph attention network are called and initialized to build a time dependence modeling layer and an energy feature screening layer;

[0020] A multi-task learning loss function is used to perform joint training and parameter updating on the time dependence modeling layer and the energy feature screening layer to construct an energy collaborative model.

[0021] Optionally, the global energy demand prediction report is generated by the following specific steps,

[0022] The multi-source time series features are input into the energy collaborative model, the energy feature screening layer performs local physical feature extraction through wavelet packet decomposition to form a multi-scale time-frequency feature subset;

[0023] The association mining layer performs nonlinear association mining through transfer entropy to generate a causal association graph;

[0024] The multi-scale time-frequency feature subset and the causal correlation map are integrated through a feature splicing channel to generate a global energy demand prediction report.

[0025] Optionally, the specific steps for constructing the cooperative energy optimization network based on the multi-source temporal features are as follows:

[0026] The temporal correlation periods between the multi-source temporal features are identified, and the response intensity between different temporal correlation periods is obtained to generate a set of network response patterns.

[0027] Based on the network response pattern set, energy allocation priorities and adjustment strategies are defined, and an energy allocation rule base is generated.

[0028] By integrating the network response pattern set and energy allocation rule base, and defining node attributes, edge weights, and interaction protocols, a collaborative energy optimization network is generated.

[0029] Optionally, the output strategy instruction includes the following specific steps:

[0030] The node attributes, edge weights, and interaction protocols of the collaborative energy optimization network are mapped to the agent's state space, action space, and policy constraints to initialize the energy management agent;

[0031] The global energy demand prediction report is input into the initialized energy management agent, and multi-source information fusion and rule matching are performed by combining the network response pattern set and the energy allocation rule base to output policy instructions.

[0032] Optionally, the specific steps for generating the optimal power allocation scheme are as follows:

[0033] The energy management agent obtains a response pattern quantization table by analyzing the network response pattern, and dynamically adjusts the energy priority in combination with the energy allocation rule base to generate candidate power allocation schemes;

[0034] Based on the energy interaction characteristics and the environmental disturbance characteristics, the candidate power allocation scheme is simulated using a multi-objective optimization algorithm to form a loss load evaluation matrix;

[0035] Candidate power allocation schemes are filtered by rules based on the loss load assessment matrix and the energy allocation rule base, and weighted scoring is performed by combining the network response pattern set to generate the optimal power allocation scheme.

[0036] Optionally, the specific steps for generating navigation control commands are as follows:

[0037] Real-time data from the device is collected via an edge computing terminal and compared with the optimal power allocation scheme to obtain the execution deviation.

[0038] Based on the execution deviation, the flight path is dynamically calibrated using a path optimization algorithm. At the same time, the execution deviation is matched with the energy allocation rule base to generate an edge correction parameter table.

[0039] The edge correction parameter table is superimposed onto the optimal power allocation scheme, and the load is corrected based on the node attributes of the cooperative energy optimization network to generate navigation control commands.

[0040] To address the aforementioned problems, the present invention also provides an energy-optimized zero-emission navigation control system for electric ships, comprising,

[0041] The data acquisition module is used to acquire energy optimization collaborative data and perform preprocessing, and extract multi-source time-series features based on the preprocessed energy optimization collaborative data;

[0042] The demand forecasting module is used to input the multi-source time-series features into the energy collaborative model, the energy feature filtering layer performs local feature extraction, the correlation mining layer performs nonlinear correlation mining, and generates a global energy demand forecasting report.

[0043] The intelligent optimization module is used to construct a collaborative energy optimization network based on the multi-source time-series features, initialize the energy management agent according to the collaborative energy optimization network, iteratively optimize the energy management agent according to the global energy demand prediction report, and output strategy instructions.

[0044] The deviation calibration module is used to obtain candidate power allocation schemes by dynamically adjusting energy priority according to the strategy instructions, and at the same time, generate the optimal power allocation scheme by pre-simulating the energy loss of different candidate power allocation schemes through a multi-objective optimization algorithm.

[0045] The instruction generation module is used to verify the validity of the optimal power allocation scheme to obtain the execution deviation, dynamically calibrate the flight path and correct the load based on the execution deviation, and generate navigation control instructions.

[0046] Compared to the problems described in the background art, this invention enhances the accuracy and comprehensiveness of global energy demand prediction and improves the accuracy and effectiveness of optimization decisions by extracting local features and mining nonlinear correlations through an energy collaboration model; it also simultaneously constructs a collaborative energy optimization network to capture energy correlation features in dynamic scenarios such as sudden changes in sea state and task switching in real time, thereby improving the energy utilization efficiency of multi-ship collaborative navigation. Attached Figure Description

[0047] Figure 1 A flowchart illustrating an energy-optimized zero-emission navigation control method for electric ships according to an embodiment of the present invention;

[0048] Figure 2This is a schematic diagram of the overall logical framework of an energy-optimized zero-emission navigation control method for electric ships provided in an embodiment of the present invention.

[0049] Figure 3 This is a schematic diagram of a module for implementing the energy-optimized zero-emission navigation control system for electric ships, provided as an embodiment of the present invention.

[0050] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0051] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0052] This application provides an energy-optimized zero-emission navigation control method for electric ships. The executing entity of this energy-optimized zero-emission navigation control method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the energy-optimized zero-emission navigation control method for electric ships can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0053] Reference Figure 1 The diagram shown is a flowchart illustrating a zero-emission navigation control method for electric ships based on energy optimization, according to an embodiment of the present invention. In this embodiment, the zero-emission navigation control method for electric ships based on energy optimization includes:

[0054] S1. Acquire energy optimization collaborative data and preprocess it, and extract multi-source time-series features based on the preprocessed energy optimization collaborative data;

[0055] The energy optimization collaborative data refers to the multi-dimensional data that needs to be integrated to achieve zero-emission navigation control for electric ships, encompassing internal ship status, external collaborative resources, and environmental predictions. Specifically, this includes electric ship status data, cross-ship network data, and environmental prediction data. The electric ship status data refers to parameters reflecting the ship's energy operation status collected through the ship's sensor network, covering state of charge (SOC), propulsion power, remaining energy storage capacity, and fuel cell hydrogen reserves. For example, when an electric ship is sailing at 12 knots, sensors may collect electric ship status data with a SOC of 78%, propulsion power of 150kW, and remaining energy storage capacity of 200kWh. This electric ship status data directly reflects the electric ship's... The data includes real-time status of ship energy supply and consumption; the cross-ship network data refers to clean energy data that can be shared by neighboring ships, obtained through inter-ship communication networks, including photovoltaic / wind power output, sharing availability time, response latency, etc. For example, when sailing in formation, it may be detected that a neighboring ship currently has a shareable power of 50kW and a remaining sharing time of 2 hours. Cross-ship network data provides a resource basis for cross-ship energy coordination; the environmental prediction data refers to the predicted values ​​of environmental parameters for future periods obtained through marine meteorological stations and other means, covering wind speed, wave height, ocean current speed, and light intensity. For example, the predicted wind speed for the next 2 hours is 10m / s and the wave height is 1.5m. Environmental prediction data is used to assess the impact of environmental disturbances on energy consumption.

[0056] The preprocessing refers to data cleaning of the collected energy optimization and collaboration data, including electric vessel status data, cross-ship network data, and environmental prediction data, to remove incomplete, erroneous, or irrelevant data records. Format conversion is performed to ensure all data sources adhere to a unified format standard for subsequent processing. Then, deduplication is performed to eliminate duplicate data entries and ensure the uniqueness of the dataset. A normalization step maps data values ​​from different sources to the same scale, preventing certain features from unduly affecting the results due to excessively large differences in magnitude. Finally, outlier handling identifies and corrects data points that deviate from the normal range, ensuring the accuracy and reliability of the dataset.

[0057] This invention provides a foundation for data quality assurance in zero-emission navigation control of electric ships by acquiring and preprocessing energy optimization collaborative data. The preprocessing process eliminates sensor error interference, resolves the time misalignment problem of multi-source data, and unifies the scale of data with different dimensions, ensuring the authenticity, reliability, and consistency of the input data. This step directly determines the accuracy of subsequent energy demand forecasting and the reliability of zero-emission strategy formulation, and is a prerequisite for achieving precise energy management and zero-emission control.

[0058] Furthermore, the multi-source time series characteristics refer to the regular patterns that reflect the dynamic correlation between energy demand, supply and environmental disturbances, extracted from electric ship status data, cross-ship network data and environmental prediction data through time series analysis methods such as sliding window statistics, dynamic time warping and STL decomposition.

[0059] As an embodiment of the present invention, the preprocessed energy optimization collaborative data extraction of multi-source time-series features includes: performing sliding window statistics and mutual information analysis on electric ship state data to obtain power coupling features; performing dynamic time warping and fluctuation period extraction on cross-ship network data to obtain energy interaction features; performing STL decomposition and time-series standardization on environmental prediction data to obtain environmental disturbance features; and integrating power coupling features, energy interaction features, and environmental disturbance features to generate multi-source time-series features.

[0060] Among them, power coupling characteristics refer to the dynamic dependence between energy supply and energy demand over time in the state data of electric ships. For example, the correlation pattern of "for every 1% decrease in SOC, the propulsion power demand will increase by 5kW in the next 20 minutes" reflects the supply and demand coordination characteristics of electric ship energy. Energy interaction characteristics refer to the degree of time matching between the clean energy shared by neighboring ships and the energy demand of the ship itself in cross-ship network data. This quantifies the response delay, periodic fluctuations of the shared power of neighboring ships and the synchronicity of the ship's demand. For example, "the photovoltaic output of neighboring ships from 10:00 to 1:00 daily..." The correlation patterns of "peak power of 55kW at 4:00" and "power sharing delay of 5 minutes for neighboring ships under high wave conditions" reflect the spatiotemporal matching characteristics of cross-ship energy collaboration. The environmental disturbance characteristics refer to the nonlinear impact of changes in environmental parameters such as wind speed, wave height, and light intensity on energy consumption in environmental prediction data, revealing the causal relationship between environmental disturbances and energy demand fluctuations. For example, the correlation patterns of "propulsion efficiency decreases by 15% when wave height > 1.8m" and "photovoltaic output increases by 20% when wind speed > 8m / s" reflect the dynamic impact characteristics of environmental factors on zero-emission energy supply.

[0061] Furthermore, the acquisition of power coupling characteristics can be achieved through sliding window statistics and mutual information analysis of electric ship status data. In specific operation, a sliding window with a fixed time length is set, and statistical quantities such as the mean, variance, and covariance of the electric ship status data at different time lags within the sliding window are obtained to quantify the short-term fluctuation patterns of energy supply and demand. Subsequently, by comparing the mutual information values ​​under different lag times (such as 5 minutes, 10 minutes, 20 minutes, etc.), the lag time that maximizes the mutual information value is identified to complete the lag relationship analysis (such as the lag dependency of SOC change and propulsion power demand of about 20 minutes). Finally, the short-term fluctuation patterns of sliding window statistics and the lag relationship of mutual information analysis are integrated to extract power coupling characteristics, such as "there is a non-linear coupling relationship between SOC state and propulsion power demand, and the change of SOC will affect propulsion power demand after 20 minutes";

[0062] The acquisition of energy interaction characteristics enables dynamic time warping and fluctuation periodization of cross-ship network data. Specifically, the input data includes the clean energy output power sequence of neighboring ships (e.g., real-time photovoltaic / wind power output), the shared response delay sequence (the time it takes for neighboring ship power to go from command to actual output), and the energy demand sequence of the ship for the corresponding time period (e.g., propulsion power deficit). Using the ship's energy demand sequence as a reference sequence, the clean energy output power sequence of neighboring ships is aligned with the reference sequence, allowing for local stretching or compression of the time axis (e.g., a 5-minute delay in neighboring ship power output). A similarity score is obtained between the reference sequence and the clean energy output power sequence of neighboring ships, addressing the issue of cross-ship data... To address the time misalignment issue caused by communication delays or differences in device response, wavelet transform was employed to select a suitable mother wavelet, such as the Morlet wavelet, for the energy data. This involved multi-resolution decomposition of the clean energy output power sequence from neighboring vessels to identify the main periodic component (e.g., the daily peak from 10:00 to 14:00). A sliding window with a fixed time was used to obtain the autocorrelation coefficients at different lag times to verify periodic stability (e.g., the autocorrelation coefficient was still higher than the previous day's peak at a 24-hour lag). Finally, the periodic pattern of the neighboring vessel's clean energy output was extracted (e.g., "Daily photovoltaic output peaks from 12:00 to 13:00, periodic stability > 90%"). The alignment results and periodic patterns were then integrated to generate energy interaction characteristics (e.g., "The time matching score between the neighboring vessel's clean energy output and this vessel's energy demand is 0.85, indicating a daily peak period from 12:00 to 13:00, with the matching degree decreasing by 15% under high wave conditions").

[0063] The acquisition of environmental disturbance characteristics can be achieved by performing STL decomposition and time-series standardization on environmental prediction data. Specifically, the environmental prediction data is split into three parts using a seasonal-trend-residual decomposition algorithm: a trend term (reflecting long-term changes, such as a fitted straight line showing an annual average wind speed increase of 0.5 m / s), a seasonal cycle term (capturing periodic fluctuations, such as an annual cycle of wave height averaging 1.2 m per day in summer and 0.8 m per day in winter), and a random residual term (characterizing sudden disturbances, such as wind speed fluctuations of ±2 m / s caused by gusts). Then, time-series standardization is performed, using the Z-score method to obtain the mean and standard deviation of the environmental prediction data. This converts the environmental prediction data at each time point into a standard normal distribution, eliminating the influence of different dimensions such as wind speed and wave height, and unifying the data scale. Finally, the decomposition and standardization results are integrated to extract environmental disturbance characteristics, such as "long-term wind speed trend increases by 0.5 m / s every 10 years," "the standard deviation of the summer wave height cycle term is 1.5 m," and "propulsion efficiency decreases by 15% when random residuals > 2 m / s."

[0064] This invention extracts multi-source time-series features from preprocessed energy optimization and coordination data, revealing the dynamic coupling patterns between different data sources. These multi-source time-series features capture the temporal correlation patterns between energy demand and environmental supply during navigation, providing a crucial "dynamic matching basis" for energy coordination models. By modeling multi-source time-series features, future energy gaps can be predicted more accurately, the impact of environmental disturbances can be assessed, and energy allocation strategies can be dynamically optimized. This allows for the optimal balance between navigation efficiency and energy supply under zero-emission constraints, improving the adaptability and robustness of zero-emission control.

[0065] S2. Input the multi-source time-series features into the energy collaborative model. The energy feature screening layer performs local feature extraction, and the association mining layer performs nonlinear association mining to generate a global energy demand prediction report.

[0066] The energy coordination model refers to a collaborative analysis framework built based on multi-source time-series features. The input of the energy coordination model is the time-series correlation information of electric ship status data, cross-ship network data and environmental prediction data. By integrating the dynamic change patterns of the three types of data, a full-scenario analysis tool supporting energy demand prediction is formed, providing a foundation for subsequent feature screening and correlation mining.

[0067] As an embodiment of the present invention, the construction of the energy coordination model includes: calling and initializing the time fusion Transformer and graph attention network, and building a time-dependent modeling layer and an energy feature filtering layer;

[0068] By using a multi-task learning loss function, joint training and parameter updates are performed on the time-dependent modeling layer and the energy feature selection layer to construct an energy collaborative model.

[0069] The energy feature filtering layer refers to the sub-model in the energy coordination model used for local feature extraction. Its core function is to filter out key features strongly related to the mission (such as propulsion power-SOC coupling features during high-speed navigation) from redundant multi-source time-series features based on the core requirements of the current navigation phase (e.g., high power response during acceleration and low power maintenance during cruising). Simultaneously, it removes interfering features unrelated to the current mission (such as minor fluctuations in energy storage output during low-speed navigation). For example, during the acceleration phase, the filtering layer will primarily retain the "propulsion power-SOC lag of 20 minutes coupling feature" and the "neighboring ship clean energy sharing delay of 5 minutes," while ignoring irrelevant features such as "minor fluctuations in energy storage output of 0.5kW," thereby improving the computational efficiency and predictive accuracy of the energy coordination model.

[0070] The correlation mining layer refers to the sub-model in the energy coordination model used for nonlinear correlation mining. Its core function is to capture complex dependencies between multi-source time-series features through deep learning (such as the 20-minute lag between a 2-meter surge in wave height and a 10% decrease in energy storage output due to environmental disturbances). For example, the correlation mining layer can identify nonlinear causal chains such as "high waves (wave height > 1.8m) - neighboring vessel shared delay (> 5 minutes) - energy storage compensation demand (+20% propulsion power)," overcoming the limitations of traditional linear models in misjudging complex patterns. This generates a more accurate global energy demand forecast report, providing a reliable decision-making basis for energy scheduling in zero-emission navigation.

[0071] Furthermore, the construction of the time-dependent modeling layer and energy feature filtering layer can be achieved by calling and initializing the Time Fusion Transformer (a neural network architecture based on self-attention mechanism) and the Graph Attention Network. Specifically, in the PyTorch framework, the Time Fusion Transformer and the Graph Attention Network are called, and a time embedding layer and a multi-head self-attention mechanism are embedded in the Time Fusion Transformer, while a node feature encoding layer and an edge attention computation layer are embedded in the Graph Attention Network. First, the Time Fusion Transformer is initialized: the timestamp information and multi-source time-series features are fused through the time embedding layer, and the multi-head self-attention mechanism is used to capture the long-range dependency relationship at any time point in the multi-source time-series features (such as the lagged impact of historical propulsion power fluctuations on current energy storage demand). The time-dependent modeling layer is constructed, and the time-dependent representation vector is output to characterize the time evolution law of energy dynamic response. Simultaneously, the graph attention network is initialized: key variables of multi-source time-series features (such as SOC, shared power of neighboring ships, and wave height) are mapped into high-dimensional embedding vectors through the node feature encoding layer. The correlation strength (such as mutual information and hysteresis correlation) between high-dimensional embedding vectors is obtained using the edge attention computing layer, and an energy feature correlation graph is constructed. When building the energy feature screening layer, the attention coefficients between nodes are obtained through the forward propagation of the graph attention layer. Combined with the edge attention mechanism, strong correlation edges are selected according to the correlation strength, and the key feature subset after screening is output. Finally, the time dependency representation vector output by the time dependency modeling layer and the key feature subset output by the energy feature screening layer are fused through a fully connected layer to complete the construction of the time dependency modeling layer and the energy feature screening layer. This provides dual input support of "time pattern + key features" for subsequent nonlinear correlation mining, and enhances the dynamic modeling capability of the energy collaboration model for complex energy.

[0072] The energy synergy model is constructed by jointly training and updating the time-dependency modeling layer and the energy feature selection layer using a multi-task learning loss function. Specifically, preprocessed multi-source temporal features are simultaneously input into both the time-dependency modeling layer and the energy feature selection layer. Then, based on the outputs of these two layers, a joint loss (combining time-dependency modeling error and feature selection error) is obtained using the multi-task learning loss function. Backpropagation is then performed on the joint loss to generate gradient information for each layer. An adaptive optimizer is used to update the attention weights and temporal embedding parameters of the time-dependency modeling layer, as well as the node encoding weights and edge attention weights of the energy feature selection layer, based on the gradient information. Simultaneously, residual connections are used to fuse the time-dependency representation vector and key feature subsets across layers, adjusting network hyperparameters (such as the number of attention heads and feature embedding dimension) to optimize the energy synergy model's performance. Finally, the constructed energy synergy model is output.

[0073] Next, the energy collaboration model is trained. Further, the multi-source time-series features are divided into a sample set, a training set, and a validation set according to their data characteristics (e.g., a 7:2:1 ratio). Linear interpolation is used to augment the features of the sample set, and batch normalization is applied to standardize the data, forming enhanced standard samples. The Adam optimizer is used to dynamically adjust the learning rate of the enhanced standard samples, and early stopping is simultaneously applied for training monitoring to obtain intermediate parameters. On the validation set, the root mean square error loss function is used to quantize the loss of the intermediate parameters, obtaining the prediction error on the validation set. When the prediction error on the validation set exceeds the convergence threshold for 10 consecutive rounds, training terminates, and the trained energy collaboration model is output simultaneously.

[0074] It should be noted that the convergence threshold is defined based on the rate of change of the prediction error on the validation set, and its value ranges from 0.0001 to 0.01.

[0075] This invention, by constructing an energy coordination model, can integrate the dynamic correlation information of electric ship status data, cross-ship network data, and environmental prediction data, breaking the limitations of a single data source and providing multi-dimensional, full-scenario data support for energy demand forecasting.

[0076] As an embodiment of the present invention, the generation of the global energy demand prediction report includes: inputting multi-source time-series features into an energy collaborative model; the energy feature screening layer extracts local physical features through wavelet packet decomposition to form a multi-scale time-frequency feature subset; the association mining layer performs nonlinear association mining through transfer entropy to generate a causal association map; and the multi-scale time-frequency feature subset and the causal association map are integrated through a feature splicing channel to generate the global energy demand prediction report.

[0077] The multi-scale time-frequency feature subset refers to a set of local physical features extracted from multi-source time-series features using wavelet packet decomposition technology, covering different time resolutions and frequency ranges, such as "high-frequency fluctuations in shared power between neighboring ships within 15 minutes". The causal relationship graph refers to a visual relationship network generated by analyzing the multi-scale time-frequency feature subset using transfer entropy, with features as nodes and causal relationships as directed edges. The causal relationship graph identifies causal relationships and assigns weights (such as the causal edge weight of "wave height fluctuation → energy storage output decrease" of 0.85) by obtaining the information transmission direction and intensity between different features (such as "whether the high-frequency fluctuation of wave height at time t will lead to the low-frequency decrease of energy storage output at t+30 minutes"), intuitively revealing the causal dependency logic between each feature, such as "the delay in shared power between neighboring ships (at time t) will cause a gap in propulsion power at t+15 minutes", etc., providing causal basis for error correction and strategy optimization of the energy coordination model.

[0078] Furthermore, the generation of multi-scale time-frequency feature subsets can be achieved through local physical feature extraction via wavelet packet decomposition. Specifically, the decomposition level is set according to the temporal resolution and feature period of the multi-source time-series features, decomposing them layer by layer into sub-bands with different time windows and frequency components. Lower-level decomposition captures the long-term trend of the multi-source time-series features (e.g., hourly SOC decay), while higher-level decomposition refines them into short-term disturbances (e.g., minute-level wave height fluctuations), ultimately yielding a set of sub-bands covering multiple time resolutions and frequency ranges. Finally, sub-bands reflecting key dynamics are retained (e.g., short-term fluctuation sub-bands of shared power between neighboring vessels, mid-term periodic sub-bands of wave height, and high-frequency disturbance sub-bands of SOC), while redundant and noisy sub-bands (e.g., high-frequency noise sub-bands caused by sensor instantaneous errors) are removed, outputting the multi-scale time-frequency features.

[0079] The generation of the causal relationship graph can be achieved through nonlinear correlation mining using transfer entropy. Specifically, the multi-scale time-frequency features output from the energy feature filtering layer are used as input, and any two of these features are selected as the source variable and the target variable, respectively. The strength of the nonlinear causal influence between the source and target variables is quantified by obtaining the predictive power of the probability distribution of the source variable within its lag period on the probability distribution of the target variable (i.e., the transfer entropy value). The lag period is determined using the mutual information maximization method, and the transfer entropy value is obtained using kernel density estimation to accommodate nonlinear relationships. Subsequently, the transfer entropy values ​​are filtered based on a preset transfer entropy threshold, and the variable pairs corresponding to the filtered transfer entropy values ​​(e.g., "source variable → target variable") are identified as causal relationships. Finally, a causal relationship graph is constructed using the multi-scale time-frequency features as nodes and the causal relationships identified by the filtered transfer entropy values ​​as directed edges.

[0080] It should be noted that the transfer entropy threshold is based on the distribution definition of statistical tests, and the exemplary range of values ​​is (0.3, 1).

[0081] This invention extracts local features through an energy feature screening layer, which can accurately select key multi-source time series features that are strongly correlated with the current navigation stage from redundant multi-source time series features, avoiding interference from invalid information and improving the computational efficiency and prediction accuracy of the energy coordination model. Through nonlinear correlation mining through an association mining layer, it can capture the complex dependencies between multi-source time series features, overcome the limitations of traditional linear models in misjudging complex patterns, and improve the accuracy of energy demand prediction.

[0082] S3. Construct a collaborative energy optimization network based on the multi-source time-series features, initialize an energy management agent according to the collaborative energy optimization network, and iteratively optimize the energy management agent according to the global energy demand prediction report, and output policy instructions;

[0083] The collaborative energy optimization network refers to a collaborative analysis framework built upon multi-source time-series characteristics such as electric ship status data, cross-ship network data, and environmental prediction data. Its core function is to generate a collaborative optimization strategy combination of multiple energy variables (such as energy storage charging and discharging, neighbor ship power sharing, and fuel cell startup) under zero-emission constraints (such as carbon emission caps) and mission objectives (such as load priorities), rather than adjusting a single variable independently. For example, when high waves cause a decrease in energy storage output, the collaborative energy optimization network can simultaneously evaluate options for neighbor ship power sharing, fuel cell startup, and load reduction, selecting the optimal combination with the lowest overall cost (energy consumption + emissions). This promotes a shift in energy management from "passive adaptation" to "proactive decision-making," improving the global resource allocation efficiency during zero-emission navigation.

[0084] As an embodiment of the present invention, the construction of a collaborative energy optimization network based on multi-source temporal features includes: identifying the temporal correlation periods between multi-source temporal features, and simultaneously obtaining the response intensity between different temporally correlated periods to generate a network response pattern set; defining energy allocation priorities and adjustment strategies based on the network response pattern set to generate an energy allocation rule base; and integrating the network response pattern set and the energy allocation rule base to generate a collaborative energy optimization network by defining node attributes, edge weights, and interaction protocols.

[0085] The time-series correlated period refers to a time interval in which there is a clear correlation between multiple time-series features. It is identified through methods such as mutual information, correlation coefficient, and Granger causality test, reflecting the coordinated change characteristics of different multiple time-series features within a specific time period. For example, if the SOC decay rate during the morning peak navigation period (8:00-10:00) shows a strong correlation with propulsion power fluctuations and the frequency of shared power calls from neighboring ships, then the morning peak navigation period is identified as a time-series correlated period.

[0086] The network response pattern set refers to the behavioral patterns and rules of the collaborative energy optimization network in different associated time periods, summarized based on the response intensity (such as the change value of the correlation intensity between features, the lag time difference, etc.) during different time-series associated periods. For example, during the morning peak associated period, the network response pattern is "for every 5% decrease in SOC, the shared power of neighboring ships needs to be increased by 100kW to compensate for the energy storage gap"; the network response pattern is obtained through statistical analysis and other methods, and is used to guide the dynamic adjustment of subsequent energy allocation strategies.

[0087] The energy allocation priority and adjustment strategy refers to the priority order and dynamic adjustment rules set for different energy variables (such as energy storage charging and discharging, neighbor ship sharing and calling, and fuel cell startup) based on the network response pattern set. For example, during high correlation periods (such as morning peak), because the compensation effect of neighbor ship shared power on SOC decay is highly responsive, the energy allocation priority is "neighbor ship sharing and calling > fuel cell startup > energy storage discharging"; while during low correlation periods (such as off-peak), the autonomous adjustment capability of energy storage dominates, and the priority is adjusted to "energy storage discharging > fuel cell startup > neighbor ship sharing and calling".

[0088] The energy allocation rule base refers to a set of specific operational rules formed by integrating network response patterns with energy allocation priorities and adjustment strategies. These rules are stored in the form of "condition-action" pairs and are used to guide the actual energy scheduling of the collaborative energy optimization network. For example, the energy allocation rule base includes: "When it is during the morning peak associated period (8:00-10:00) and SOC < 40% and wave height > 1.5 meters, trigger the action 'share 200kW of energy with neighboring ships and discharge 100kW of energy storage'".

[0089] The node attributes, edge weights, and interaction protocols refer to the specific definitions of the topology and interaction logic of the collaborative energy optimization network, used to standardize the cooperation methods of various components in the network. Node attributes include basic information on multi-source time-series characteristics, such as time resolution, value range, and data type; edge weights represent the strength of the association between nodes (e.g., the edge weight of SOC sharing power with a neighboring ship is 0.82, reflecting a strong association); interaction protocols specify the direction of information transmission between nodes (e.g., SOC data is transmitted to the neighboring ship's sharing decision node in real time), update frequency, and conflict resolution rules (e.g., when energy storage discharge conflicts with the neighboring ship's sharing call, the neighboring ship's sharing is prioritized). By clarifying node attributes, edge weights, and interaction protocols, the collaborative energy optimization network forms a standardized cooperation framework, supporting the implementation of energy allocation rules.

[0090] Furthermore, the generated network response pattern set can be achieved by identifying the temporal correlation periods between multi-source temporal features and simultaneously obtaining the response intensity between different temporal correlation periods. Specifically, based on the multi-source temporal features, a sliding window of fixed time length (e.g., 2 hours) is set. The Pearson correlation coefficients of any two multi-source temporal features within the sliding window are obtained. When the absolute value of the Pearson correlation coefficient exceeds a preset coefficient threshold, the time interval corresponding to the Pearson correlation coefficient is marked as a "high correlation period," and when it is below the coefficient threshold, it is marked as a "low correlation period," thus completing the identification of temporal correlation periods. Subsequently, for each correlation period, the Pearson correlation coefficients of the multi-source temporal features at different delay durations (e.g., 10 minutes, 20 minutes, 30 minutes) are further obtained. The maximum value is selected as the response intensity for the current delay duration. Simultaneously, the mean and variance of all Pearson correlation coefficients within the window are statistically analyzed to quantify the persistence of the correlation and the stability of the temporal dependency. Finally, the analysis results of each sliding window (time range, associated feature pairs, associated time period level, maximum Pearson correlation coefficient, and mean / variance) are integrated into a structured set of network response patterns.

[0091] It should be noted that the coefficient threshold is based on the distribution test definition of the Pearson correlation coefficient, and an exemplary value range is (between 0.5 and 0.8).

[0092] The generated energy allocation rule base can define energy allocation priorities and adjustment strategies based on the network response pattern set. In specific operation, energy allocation priorities are defined based on response intensity and time-series correlation periods: feature pairs with high response intensity in high correlation periods (such as "SOC-neighbor ship shared power", maximum Pearson correlation coefficient 0.75) are listed as priority allocation objects because they have a direct impact on energy dispatch; feature pairs with medium response intensity in medium correlation periods (such as off-peak 14:00-16:00) (such as "wave height-energy storage output", maximum Pearson correlation coefficient 0.6) are listed as secondary priority allocation objects; feature pairs with low response intensity in low correlation periods (such as midday 12:00-14:00) (such as "photovoltaic output-propulsion power", maximum Pearson correlation coefficient 0.4) are listed as subsequent allocation objects. Simultaneously, the allocation order is adjusted based on time-period objectives. For example, during the morning peak, the goal is to "ensure continued operation," prioritizing the power sharing needs of neighboring vessels strongly correlated with SOC; during the off-peak, the goal is to "reduce costs," prioritizing the allocation of photovoltaic output weakly correlated with wave height. Next, specific triggering conditions (e.g., "SOC < 30%" and "Wave height > 1.5 meters" during the morning peak) and adjustment actions (e.g., "When SOC < 30% and it is during a highly correlated morning peak period, the power sharing of neighboring vessels increases by 200kW") are set for each correlated time period. Conflict resolution rules are also clarified (e.g., when there is a conflict between neighboring vessel sharing and fuel cell startup, the higher-priority neighboring vessel sharing is prioritized). Finally, the energy allocation priority, triggering conditions, and adjustment actions are integrated into a "condition-action" energy allocation rule base (e.g., "IF Time ∈ [8:00, 10:00] AND SOC < 30% THEN 200kW allocated to neighboring vessels + 100kW allocated to energy storage").

[0093] The generated collaborative energy optimization network can be achieved by integrating a set of network response patterns and an energy allocation rule base, and by defining node attributes, edge weights, and interaction protocols. Specifically, multi-source temporal feature variables are defined as network nodes, each including basic attributes such as time resolution, value range, and data type. Then, based on the response intensity in the set of network response patterns, the maximum Pearson correlation coefficient of associated feature pairs is mapped to the edge weight. For example, if the maximum Pearson correlation coefficient of "SOC-neighbor ship shared power" is 0.75 during the morning peak period, then the edge weight is set to 0.75. Next, a node interaction protocol is formulated: combining the "condition-action" rules in the energy allocation rule base, the information transmission direction and triggering logic between network nodes are defined. For example, the rule "IF time ∈ [8:00, 10:00] AND SOC < 30% THEN neighboring ship's shared power increases by 200kW" in the rule base can be transformed into an interaction protocol: when the state value of the SOC node (time window 8:00-10:00) is lower than 30%, it sends an adjustment instruction of "increase by 200kW" to the neighboring ship's shared power node; at the same time, a reverse feedback mechanism is clearly defined (such as after the neighboring ship's shared power node performs the adjustment, it needs to feed back the actual output value to the SOC node to verify the rule's effect). Finally, by integrating node attributes, edge weights, and interaction protocols, a collaborative energy optimization network is constructed.

[0094] This invention constructs a collaborative energy optimization network based on multi-source time-series characteristics, which can deeply integrate dispersed energy data (such as SOC, clean energy output of neighboring ships, and wave height) with zero-emission constraints (such as carbon emission limits and energy storage capacity limitations) to form a full-link collaborative optimization framework covering demand forecasting, resource allocation, and emission control, thereby improving the global allocation efficiency of resources during zero-emission navigation.

[0095] The energy management agent refers to a decision-making entity initialized according to the collaborative optimization logic of the collaborative energy optimization network. This agent inherits the network's dynamic analysis capabilities for multi-source temporal characteristics and its multi-objective optimization rules under zero-emission constraints. In actual operations, it can directly invoke collaborative strategies generated by the network, such as "high wave scenario - energy storage compensation - neighboring vessel sharing," achieving rapid transformation from model analysis (e.g., energy collaboration model) to actual operation. For example, during the initialization phase, the energy management agent can directly invoke the collaborative strategy generated by the network, eliminating the need for retraining and significantly shortening the cycle required for practical application. This also ensures the scientific validity and feasibility of the initial strategy, providing a reliable starting point for subsequent iterative optimization.

[0096] As an embodiment of the present invention, the initialization of the energy management agent according to the cooperative energy optimization network includes: mapping the node attributes, edge weights and interaction protocols of the cooperative energy optimization network into the agent's state space, action space and policy constraints, and initializing the energy management agent.

[0097] The initialization of the energy management agent can be achieved by mapping the node attributes, edge weights, and interaction protocols of the collaborative energy optimization network to the agent's state space, action space, and policy constraints. Specifically, based on the basic attributes of network nodes, the state space of the energy management agent is defined as a multi-dimensional set of environmental perception variables, with each dimension corresponding to a basic attribute of a network node. Then, with edge weights as the core, policy constraint rules for the agent are defined. Higher edge weights correspond to higher collaborative influence between network nodes. When making decisions, the energy management agent must prioritize the impact of state changes of high-weight network nodes on the target variable (e.g., when adjusting the shared power of neighboring ships, it needs to focus on the high-weight association with SOC to ensure their collaboration). Next, based on the triggering conditions and adjustment actions in the interaction protocol, the action space of the energy management agent is defined as a set of executable discrete operations (e.g., "increase shared power of neighboring ships," "start fuel cells," "reduce energy storage discharge," etc.). Each discrete operation is bound with triggering conditions and execution parameters (e.g., "when SOC < 30% and during the morning peak period, execute the action 'shared power of neighboring ships + 200kW'"). Ultimately, an intelligent energy management agent with environmental perception, dynamic decision-making, and collaborative adjustment capabilities will be formed, providing support for the real-time optimization of subsequent energy dispatch.

[0098] Furthermore, this invention iteratively optimizes the energy management agent through a global energy demand prediction report and outputs strategy instructions, enabling the reverse correction of the energy management agent's decision parameters using actual operating data.

[0099] The strategy instructions refer to the energy scheduling operation guidelines output after iterative optimization of the energy management agent based on the global energy demand forecast report. The generation process uses actual operating data to reverse-correct the agent's decision parameters, guiding the actual energy scheduling of electric ships. For example, when the global energy demand forecast report shows that insufficient shared power from neighboring ships leads to excessive emissions during a certain period, the energy management agent can adjust the weight parameters of "shared power from neighboring ships" through methods such as gradient descent, and automatically increase the sharing ratio when encountering similar scenarios in the future. This continuous "prediction-execution-correction" closed loop makes the output strategy instructions both accurate and robust, providing a dynamic and reliable operational basis for energy scheduling in zero-emission navigation.

[0100] As an embodiment of the present invention, the step of iteratively optimizing the energy management agent through the global energy demand prediction report and outputting policy instructions includes: inputting the global energy demand prediction report into the initialized energy management agent, combining the network response pattern set and the energy allocation rule base to perform multi-source information fusion and rule matching, and outputting policy instructions.

[0101] The strategy instructions can be obtained by inputting the global energy demand forecast report into the initialized energy management agent, and by combining the network response pattern set and the energy allocation rule base for multi-source information fusion and rule matching. For example, the demand data in the global energy demand forecast report is spatiotemporally aligned with the correlation feature pairs and response intensity in the network response pattern set. At the same time, the triggering conditions and adjustment actions of the corresponding time period in the energy allocation rule base are combined to form a fused decision context (e.g., "During the morning peak from 8:00 to 10:00, the SOC is expected to decrease by 5%, and the power shared by neighboring ships should be prioritized"). Then, based on the fused decision context, the "condition-action" rules in the energy allocation rule base are traversed to verify whether the current state meets the triggering conditions one by one. If the conditions are met, the corresponding action is extracted; if multiple rules are triggered simultaneously, the optimal action combination is selected based on the response intensity in the network response pattern set and the energy allocation priority in the energy allocation rule base. The matched action combination is converted into a specific energy scheduling instruction (e.g., "8:00-8:30, increase the shared power of neighboring ships by 200kW; 8:30-9:00, start the fuel cell to supplement 150kW"). The feasibility of the instruction is verified through the state space feedback mechanism of the agent (e.g., check whether the shared power of neighboring ships exceeds the available quota and whether the fuel cell start time is matched). Finally, an executable strategy instruction is generated to guide the real-time scheduling of electric ship energy.

[0102] This invention initializes the energy management agent based on the collaborative energy optimization network, which can transform the collaborative optimization logic of the collaborative energy optimization network into the initial decision-making ability of the energy management agent. The energy management agent not only inherits the collaborative energy optimization network's ability to dynamically analyze multi-source time-series characteristics (such as capturing the lag correlation between wave height fluctuations and energy storage output), but also incorporates multi-objective optimization rules under zero-emission constraints (such as prioritizing the use of clean energy and minimizing fossil fuel supplementation).

[0103] S4. According to the strategy instruction, the energy management agent obtains candidate power allocation schemes by dynamically adjusting energy priorities, and at the same time generates the optimal power allocation scheme by pre-simulating the energy loss of different candidate power allocation schemes through a multi-objective optimization algorithm.

[0104] The candidate power allocation scheme refers to the preliminary scheduling scheme generated by the energy management agent after dynamically adjusting the energy priority according to the strategy instructions. It includes different combinations of energy variables and covers the possibilities of energy allocation under different scenarios. For example, during the morning peak period of 8:00-10:00, if the strategy instructions require "prioritizing the protection of the voyage", the candidate power allocation scheme may include combinations such as "200kW of shared power with neighboring ships + 100kW of energy storage discharge" and "150kW of shared power with neighboring ships + 120kW of fuel cell start-up". The candidate power allocation scheme is generated by adjusting the energy priority and provides alternative options for subsequent optimization.

[0105] As an embodiment of the present invention, the step of obtaining candidate power allocation schemes by dynamically adjusting energy priorities according to policy instructions includes: the energy management agent obtaining a response pattern quantization table by parsing network response patterns, and dynamically adjusting energy priorities in conjunction with an energy allocation rule base to generate candidate power allocation schemes.

[0106] The aforementioned response pattern quantification table is a reference tool that quantifies and presents network response patterns in a structured table format. It includes time-series correlation periods, correlation feature pairs, response strength, correlation stability indicators, and dynamic behavior patterns, and is used for quick querying and guiding energy priority adjustments.

[0107] Furthermore, the candidate power allocation scheme can obtain a response pattern quantification table by analyzing the network response pattern through an energy management intelligent agent, and dynamically adjust the energy priority acquisition in combination with the energy allocation rule base. In specific operation, the structured content of the network response pattern set is analyzed, and the time range, related feature pairs, response intensity, correlation stability index and dynamic behavior pattern of the time-series correlation period are extracted. The data is then organized into a response pattern quantification table according to the fields of "time period-feature pair-maximum correlation coefficient-mean-variance-dynamic pattern", forming a collaborative pattern reference tool that can be quickly queried. Subsequently, by combining the "condition-action" rules in the energy allocation rule base, and matching the current time with the real-time status of the equipment (such as SOC=25%, wave height=1.3 meters) according to the matching rule conditions, the priority order of energy variables is dynamically adjusted. Finally, based on the adjusted priority order and combined with the global energy demand forecast report, multiple candidate power allocation schemes are generated. For example, the candidate power allocation schemes during the morning peak period may include "200kW shared power with neighboring ships + 100kW energy storage discharge" and "180kW shared power with neighboring ships + 120kW fuel cell start-up", covering the possibility of different resource combinations and providing alternative options for subsequent optimization.

[0108] This invention dynamically adjusts energy priorities and obtains candidate power allocation schemes based on strategy instructions. It can flexibly adjust the priority order of energy variables according to real-time needs, avoid resource mismatch caused by fixed priorities, ensure that candidate schemes cover the optimal resource combination under different scenarios, and improve the adaptability of energy allocation to dynamic needs.

[0109] The optimal power allocation scheme refers to the scheme with the lowest overall cost that meets the constraints after pre-simulating candidate power allocation schemes in terms of energy loss, emission compliance, and mission reliability using a multi-objective optimization algorithm. For example, during the morning peak period of 8:00-10:00, if the total loss of the candidate power allocation scheme "200kW shared power with neighboring ships + 100kW energy storage discharge" is 50kW·h and the carbon emission is 80kg, which is lower than other candidate power allocation schemes (such as "150kW shared power with neighboring ships + 120kW fuel cell" with a total loss of 65kW·h and carbon emission of 110kg), then this scheme is determined to be optimal. The optimal scheme provides a globally optimal energy scheduling strategy for zero-emission navigation by balancing multiple objectives (loss, emission, and reliability).

[0110] As an embodiment of the present invention, the step of generating the optimal power allocation scheme by pre-simulating the energy loss of different candidate power allocation schemes through a multi-objective optimization algorithm includes: simulating the candidate power allocation schemes based on energy interaction characteristics and environmental disturbance characteristics using a multi-objective optimization algorithm to form a loss load evaluation matrix; filtering the candidate power allocation schemes according to the loss load evaluation matrix and the energy allocation rule base, and performing weighted scoring in combination with the network response rule set to generate the optimal power allocation scheme.

[0111] The loss load evaluation matrix refers to a quantitative comparison matrix formed after simulating candidate power allocation schemes through a multi-objective optimization algorithm. It includes the multi-dimensional evaluation indicators of each scheme and is used to comprehensively evaluate the performance of the schemes.

[0112] Furthermore, the loss load assessment matrix can be simulated using a multi-objective optimization algorithm based on energy interaction characteristics and environmental disturbance characteristics. Specifically, the candidate power allocation scheme is taken as input, and the objective function is set with the goal of minimizing total energy loss, minimizing total carbon emissions, and maximizing the endurance guarantee rate. The remaining energy storage capacity, the upper limit of shared power between adjacent ships, the upper limit of equipment operating power, and environmental disturbance restrictions are used as constraints to simulate the operation process of the candidate power allocation scheme under different environmental disturbances. Finally, the loss load assessment matrix is ​​constructed based on the simulation results: with the candidate power allocation scheme as the row and the total energy loss, total carbon emissions, endurance guarantee rate, and load balance (the proportion of each energy load to the total demand) as the column, the quantitative values ​​of each candidate power allocation scheme under typical operating conditions such as normal operating conditions, high wave operating conditions, and low light conditions are filled in.

[0113] The optimal power allocation scheme can be generated by filtering candidate power allocation schemes using a loss load assessment matrix and an energy allocation rule base, and by weighting a score based on a network response pattern set. In practice, based on the loss load assessment matrix and the energy allocation rule base, candidate power allocation schemes that do not meet the constraints or have low priority (such as choosing high-loss neighboring ships to share power during off-peak hours) are excluded, forming a pool of effective schemes after preliminary screening. Subsequently, taking the candidate power allocation schemes in the effective scheme pool as objects, the correlation strength and dynamic behavior patterns extracted from the network response pattern set are used as benchmarks to allocate indicator weights. Among them, the higher the correlation strength of the indicator (such as the loss of power shared by neighboring ships), the greater the indicator weight. Finally, each indicator in the candidate power allocation scheme (including total energy loss, carbon emissions, and endurance guarantee rate) is weighted and summed with the corresponding indicator weight, and the candidate power allocation scheme with the largest value is taken as the optimal power allocation scheme.

[0114] Furthermore, this invention uses a multi-objective optimization algorithm to pre-simulate the energy loss of candidate power allocation schemes and generate the optimal power allocation scheme. It can simultaneously evaluate the energy loss, emission compliance, and mission reliability of multiple candidate power allocation schemes. Through Pareto front analysis, it selects the candidate power allocation scheme with the lowest overall cost and that meets mission constraints, avoiding the local optimum trap caused by single-objective optimization and providing a globally optimal power allocation strategy for zero-emission navigation.

[0115] S5. Verify the validity of the optimal power allocation scheme to obtain the execution deviation, dynamically calibrate the route trajectory and correct the load based on the execution deviation, and generate navigation control commands.

[0116] The execution deviation refers to the difference between the actual result and the expected value of the optimal power allocation scheme during actual execution due to environmental disturbances or equipment errors (such as equipment performance fluctuations and dynamic deviations in correlation strength). For example, specific deviation values ​​such as "the predicted shared power of neighboring ships is 200kW, but the actual output is only 160kW" or "the predicted energy storage discharge time is 2 hours, but the actual discharge time is only 1.5 hours".

[0117] As an embodiment of the present invention, the step of verifying the validity of the optimal power allocation scheme to obtain the execution deviation includes: collecting real-time data from the device through an edge computing terminal and comparing it with the optimal power allocation scheme to obtain the execution deviation.

[0118] Among them, edge computing terminals refer to dedicated computing devices (such as industrial-grade edge gateways, embedded controllers, or shipborne intelligent terminals) deployed at the edge of the energy devices of electric ships. They are responsible for collecting device data locally, performing local calculations and analysis, and can quickly complete data comparison and deviation acquisition without relying on the cloud. Real-time device data refers to the dynamic status information collected in real time by sensors of various energy-related devices on the ship during operation, specifically including: energy output data, energy storage status data, load demand data, and environmental correlation data.

[0119] Furthermore, the real-time data of the device can be collected through an edge computing terminal. In specific operation, the edge computing terminal establishes a connection with multi-source sensors (such as current sensors, voltage sensors, temperature sensors, wave height sensors, etc.) deployed on the energy equipment through wired (such as RS485, CAN bus) and wireless (such as LoRa, 5G / satellite communication) communication interfaces. The multi-source sensors collect real-time data of the device, including the discharge power and SOC value of the energy storage battery, the actual transmission value of the shared power of the neighboring vessel, the real-time load of the propulsion motor, etc., and transmit the digital signals to the edge computing terminal. The edge computing terminal performs analog-to-digital conversion, filtering and noise reduction, and protocol parsing on the digital signals to extract valid data (such as "the actual output of the shared power of the neighboring vessel is 180kW" and "the current SOC value is 65%)).

[0120] The execution deviation can be obtained by comparing the real-time data of the device with the optimal power allocation scheme. In specific operation, the edge computing terminal classifies the collected real-time data of the device according to the index types such as power, status, and load. Then, the expected value of the current optimal power allocation scheme (such as 200kW shared power of neighboring ships, 70% SOC, and 300kW energy storage discharge power) is aligned with the real-time data of the device by timestamp, and the index deviation is obtained for different types of indicators. Finally, the index deviations are sorted into a structured report and output to obtain the execution deviation.

[0121] This invention obtains execution deviations based on deviation validity verification, which can verify the degree of matching between the optimal power allocation scheme and the theoretical expectation in actual execution, identify execution deviations caused by environmental disturbances or energy coordination model errors, avoid energy shortages or efficiency losses caused by the difference between "theoretically optimal" and "actually feasible", and ensure the reliability of the implementation of the optimal power allocation scheme.

[0122] The navigation control commands refer to a set of specific operation commands generated based on the calibrated route trajectory and the corrected load requirements. These commands are used to guide each electric vessel to perform actions, such as "adjust propulsion power to 250kW", "switch to fuel cell supplement 100kW", and "limit energy storage discharge to 300kW·h", to ensure that the navigation process is consistent with the dynamically optimized optimal power allocation scheme.

[0123] As an embodiment of the present invention, the step of dynamically calibrating the flight path and correcting the load based on the execution deviation to generate navigation control commands includes: dynamically calibrating the flight path through a path optimization algorithm based on the execution deviation, and matching the execution deviation with an energy allocation rule base to generate an edge correction parameter table; superimposing the edge correction parameter table onto the optimal power allocation scheme, and correcting the load based on the node attributes of the collaborative energy optimization network to generate navigation control commands.

[0124] The edge correction parameter table refers to a set of specific parameters generated based on the matching results between the execution deviation and the energy allocation rule base. It is used to guide the calibration of the flight path and the load correction, including power adjustment amount, path offset amount, load correction coefficient, etc., and directly adapts to the impact of dynamic environmental disturbances on navigation control.

[0125] Furthermore, the generated edge correction parameter table can dynamically calibrate the route trajectory based on the execution deviation using a path optimization algorithm. Simultaneously, the execution deviation is matched with an energy allocation rule base. Specifically, the current execution deviation and the ship's real-time status are input into the path optimization algorithm. Using "minimizing energy consumption deviation + avoiding high-risk areas" as the objective function, the algorithm corrects the original route by adjusting trajectory parameters (such as heading and propulsion power allocation). For example, for high-risk areas in high-energy-consuming segments reflected in the execution deviation, the path optimization algorithm replans the route, reducing dependence on deviation-related energy variables (such as shared power of neighboring ships and energy storage output), generating a calibrated new route. Subsequently, the execution deviation is matched with the energy allocation rule base to identify triggered rule conditions (such as energy variables deviating from expectations), and corresponding correction parameters are obtained based on these conditions. For example, if the execution deviation shows that a certain energy variable (such as shared power of neighboring ships) does not meet expectations, the energy allocation rule base triggers the rule "adjust the weight of this variable when environmental disturbance exceeds a threshold." Simultaneously, combined with the node attributes of the collaborative energy optimization network, correction parameters related to network nodes (such as load allocation coefficients and interface transmission correction values) are supplemented. Finally, all correction parameters were organized into an edge correction parameter table according to the structure of "parameter type-correction value-applicable scenario-priority" to guide the precise execution of subsequent navigation control commands.

[0126] The navigation control commands can be generated by overlaying an edge correction parameter table onto the optimal power allocation scheme and correcting the load generation based on the node attributes of the collaborative energy optimization network. Specifically, correction parameters related to energy allocation and path adjustment are extracted from the edge correction parameter table. These parameters are then adaptively adjusted according to their priority and applicable scenarios in the optimal power allocation scheme. Subsequently, based on the node attributes of the collaborative energy optimization network, including real-time status information such as the health status of energy storage devices, the transmission efficiency of energy interfaces, and the real-time output capacity of renewable energy devices, the correction parameters are further adjusted. For example, if the health status of energy storage devices declines (manifested as reduced charging and discharging efficiency), the set value of the discharge power is reduced to avoid over-discharge; if the transmission efficiency of the shared interface of neighboring ships decreases (manifested as increased transmission loss), the demand value of the shared power of neighboring ships is increased to compensate for the loss, ensuring that the corrected load allocation matches the actual capacity of the nodes. Finally, the further adjusted correction parameters are combined with the adjusted related parameters (such as heading and propulsion power allocation) to transform them into specific navigation control commands.

[0127] This invention dynamically calibrates the flight path and corrects the load based on the execution deviation, and generates navigation control commands. It can flexibly adjust the navigation strategy according to the real-time deviation, thereby improving the navigation's adaptability to dynamic environments.

[0128] Specifically, for a more intuitive understanding of the execution logic and data flow relationships of the energy-optimized zero-emission navigation control system for electric ships, please refer to [link to relevant documentation]. Figure 2 A schematic diagram of the overall logical framework of an energy-optimized zero-emission navigation control method for electric ships. Figure 2 The system clearly presents a complete closed-loop chain from multi-source data acquisition to dynamic command execution: battery status monitoring sensors, motor speed sensors, environmental information acquisition and path planning, speed control strategies, etc., capture real-time power parameters (such as battery status and motor speed), navigation environment data (such as sea state information), and mission command information of the ship. This information is then aggregated to the energy optimization controller via a unified data transmission channel. Upon receiving the data, the controller combines the target path of the path planning with the speed adjustment requirements of the speed control strategy, comprehensively analyzes the battery status, motor operating parameters, and environmental variables, and generates optimized control commands such as propulsion power allocation and course adjustment. These optimized control commands act directly on the propulsion motor drive unit and course adjustment mechanism in the form of control signals, driving the ship to execute specific power output and course correction actions. Simultaneously, they transmit energy consumption management strategies to the energy consumption monitoring device, enabling real-time quantitative recording of the power system's energy consumption. It should be noted that the logical framework diagram of the energy-optimized zero-emission navigation control method for electric ships essentially represents the interaction logic of "perception-decision-execution-feedback". In actual operation, the millisecond-level synchronous transmission of multi-source data (such as the precise alignment of sensor signals and timing control commands), the dynamic strategy generation under complex sea conditions (such as real-time adjustment of power distribution during surge impacts), and the error compensation mechanism of the actuator (such as adaptive correction of heading deviations) are all achieved through the above closed-loop link for intelligent coordination. This logical framework diagram of the energy-optimized zero-emission navigation control method for electric ships only extracts the core data flow path and control logic, providing an intuitive reference for understanding the autonomous energy management operation mode of the energy-optimized zero-emission navigation control system for electric ships.

[0129] Compared to the problems described in the background art, this invention enhances the accuracy and comprehensiveness of global energy demand prediction and improves the accuracy and effectiveness of optimization decisions by extracting local features and mining nonlinear correlations through an energy collaboration model; it also simultaneously constructs a collaborative energy optimization network to capture energy correlation features in dynamic scenarios such as sudden changes in sea state and task switching in real time, thereby improving the energy utilization efficiency of multi-ship collaborative navigation.

[0130] like Figure 3 The diagram shown is a functional block diagram of the energy-optimized zero-emission navigation control system for electric ships based on the present invention.

[0131] The energy-optimized zero-emission navigation control system 200 for electric ships described in this invention can be installed in an electronic device. Depending on the functions implemented, the energy-optimized zero-emission navigation control system for electric ships may include a network construction module 201, a safety value calculation module 202, a logic construction module 203, an index calculation module 204, and a scheme formulation module 205. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0132] In this embodiment of the invention, the functions of each module / unit are as follows:

[0133] The data acquisition module 201 is used to acquire energy optimization collaborative data and preprocess it, and extract multi-source time-series features based on the preprocessed energy optimization collaborative data.

[0134] The demand forecasting module 202 is used to input the multi-source time-series features into the energy collaborative model, the energy feature screening layer performs local feature extraction, the association mining layer performs nonlinear association mining, and generate a global energy demand forecasting report.

[0135] The intelligent optimization module 203 is used to construct a collaborative energy optimization network based on the multi-source time-series features, initialize an energy management agent according to the collaborative energy optimization network, iteratively optimize the energy management agent according to the global energy demand prediction report, and output strategy instructions.

[0136] The deviation calibration module 204 is used to obtain candidate power allocation schemes by dynamically adjusting energy priority according to the strategy instructions, and at the same time generate the optimal power allocation scheme by pre-simulating the energy loss of different candidate power allocation schemes through a multi-objective optimization algorithm.

[0137] The instruction generation module 205 is used to verify the validity of the optimal power allocation scheme to obtain the execution deviation, dynamically calibrate the flight path and correct the load based on the execution deviation, and generate navigation control instructions.

[0138] In detail, the modules in the energy-optimized electric ship zero-emission navigation control system 200 described in this embodiment of the invention employ the same methods as described above during use. Figure 1 The method uses the same technical means as the energy-optimized zero-emission navigation control method for electric ships described in the article, and can produce the same technical effect, so it will not be repeated here.

[0139] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0140] Finally, it should be noted that deleting any one of the above embodiments does not affect the technical solutions of other embodiments. The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A zero-emission navigation control method for electric ships based on energy optimization, characterized in that, include, Energy optimization collaborative data is acquired and preprocessed, and multi-source time-series features are extracted based on the preprocessed energy optimization collaborative data. The energy optimization collaborative data includes electric ship status data, cross-ship network data, and environmental prediction data. The preprocessing includes data cleaning, format conversion, deduplication, normalization, and outlier handling; The specific steps for extracting multi-source temporal features are as follows: Sliding window statistics and mutual information analysis are performed on the electric ship status data to obtain power coupling characteristics; Dynamic time warping and fluctuation period extraction are performed on the cross-ship network data to obtain energy interaction characteristics; The environmental prediction data is decomposed using STL and standardized over time to obtain environmental disturbance characteristics. By integrating the power coupling characteristics, the energy interaction characteristics, and the environmental disturbance characteristics, multi-source time-series characteristics are generated; The multi-source time-series features are input into the energy collaborative model. The energy feature screening layer performs local feature extraction, and the association mining layer performs nonlinear association mining to generate a global energy demand prediction report. A collaborative energy optimization network is constructed based on the multi-source temporal features. An energy management agent is initialized according to the collaborative energy optimization network, and the energy management agent is iteratively optimized according to the global energy demand prediction report, outputting policy instructions. The specific steps are as follows. The node attributes, edge weights, and interaction protocols of the collaborative energy optimization network are mapped to the agent's state space, action space, and policy constraints to initialize the energy management agent; The global energy demand prediction report is input into the initialized energy management agent, and multi-source information fusion and rule matching are performed by combining the network response pattern set and the energy allocation rule base to output policy instructions. According to the strategy instructions, the energy management agent obtains candidate power allocation schemes by dynamically adjusting energy priorities, and generates the optimal power allocation scheme by pre-simulating the energy loss of different candidate power allocation schemes through a multi-objective optimization algorithm. The optimal power allocation scheme is validated for deviation to obtain the execution deviation. Based on the execution deviation, the flight path is dynamically calibrated and the load is corrected to generate navigation control commands.

2. The zero-emission navigation control method for electric ships based on energy optimization as described in claim 1, characterized in that, The specific construction steps of the energy coordination model are as follows. Call and initialize the temporal fusion Transformer and graph attention network, and build the temporal dependency modeling layer and energy feature filtering layer; By using a multi-task learning loss function, joint training and parameter updates are performed on the time-dependent modeling layer and the energy feature selection layer to construct an energy collaborative model.

3. The zero-emission navigation control method for electric ships based on energy optimization as described in claim 1, characterized in that, The specific steps for generating the global energy demand forecast report are as follows: The multi-source time-series features are input into the energy collaborative model, and the energy feature filtering layer extracts local physical features through wavelet packet decomposition to form a multi-scale time-frequency feature subset. The association mining layer performs nonlinear association mining through transfer entropy to generate a causal association map. The multi-scale time-frequency feature subset and the causal correlation map are integrated through a feature splicing channel to generate a global energy demand prediction report.

4. The zero-emission navigation control method for electric ships based on energy optimization as described in claim 1, characterized in that, The specific steps for constructing a collaborative energy optimization network based on the multi-source temporal features are as follows. The temporal correlation periods between the multi-source temporal features are identified, and the response intensity between different temporal correlation periods is obtained to generate a set of network response patterns. Based on the network response pattern set, energy allocation priorities and adjustment strategies are defined, and an energy allocation rule base is generated. By integrating the network response pattern set and energy allocation rule base, and defining node attributes, edge weights, and interaction protocols, a collaborative energy optimization network is generated.

5. The zero-emission navigation control method for electric ships based on energy optimization as described in claim 1, characterized in that, The specific steps for generating the optimal power allocation scheme are as follows: The energy management agent obtains a response pattern quantization table by analyzing the network response pattern, and dynamically adjusts the energy priority in combination with the energy allocation rule base to generate candidate power allocation schemes; Based on the energy interaction characteristics and the environmental disturbance characteristics, the candidate power allocation scheme is simulated using a multi-objective optimization algorithm to form a loss load evaluation matrix; Candidate power allocation schemes are filtered by rules based on the loss load assessment matrix and the energy allocation rule base, and weighted scoring is performed by combining the network response pattern set to generate the optimal power allocation scheme.

6. The zero-emission navigation control method for electric ships based on energy optimization as described in claim 1, characterized in that, The specific steps for generating navigation control commands are as follows: Real-time data from the device is collected via an edge computing terminal and compared with the optimal power allocation scheme to obtain the execution deviation. Based on the execution deviation, the flight path is dynamically calibrated using a path optimization algorithm. At the same time, the execution deviation is matched with the energy allocation rule base to generate an edge correction parameter table. The edge correction parameter table is superimposed onto the optimal power allocation scheme, and the load is corrected based on the node attributes of the cooperative energy optimization network to generate navigation control commands.

7. A zero-emission navigation control system for electric ships based on energy optimization, characterized in that, include, The data acquisition module is used to acquire energy optimization collaborative data and perform preprocessing, and extract multi-source time-series features based on the preprocessed energy optimization collaborative data; The demand forecasting module is used to input the multi-source time-series features into the energy collaborative model, the energy feature filtering layer performs local feature extraction, the correlation mining layer performs nonlinear correlation mining, and generates a global energy demand forecasting report. The intelligent optimization module is used to construct a collaborative energy optimization network based on the multi-source time-series features, initialize the energy management agent according to the collaborative energy optimization network, iteratively optimize the energy management agent according to the global energy demand prediction report, and output strategy instructions. The deviation calibration module is used to obtain candidate power allocation schemes by dynamically adjusting energy priority according to the strategy instructions, and at the same time, generate the optimal power allocation scheme by pre-simulating the energy loss of different candidate power allocation schemes through a multi-objective optimization algorithm. The instruction generation module is used to verify the validity of the optimal power allocation scheme to obtain the execution deviation, dynamically calibrate the flight path and correct the load based on the execution deviation, and generate navigation control instructions.

Images