A Method and System for Ship Navigation Route Planning Based on Wave State Analysis

Abstract

The ship navigation route planning method and system based on wave state analysis provided in this application combines deep learning networks and marine environmental monitoring information to provide accurate navigation route planning for ships, significantly reducing risks and energy consumption during navigation. The navigation route prediction results generated by this scheme are based not only on current ocean conditions but also on historically effective navigation routes, making the predictions more reliable. This method can avoid navigation risks caused by sudden changes in the marine environment and reduce energy consumption when selecting the optimal navigation route. Optimization of neural network weight configuration is achieved by using target and joint network tuning quality indicators, enabling the network to learn and improve itself for more accurate navigation route prediction. The resulting target navigation route planning decision network can formulate the optimal navigation plan in complex marine environments, improving ship navigation safety and economic efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a ship navigation route planning method and system based on sea wave state analysis. BACKGROUND

[0002] Ship navigation route planning is of great significance to ensure the safety of ship navigation and improve energy efficiency. However, traditional navigation route planning methods mainly rely on manual experience and fixed algorithms, which cannot adapt to complex and changing marine environments such as sea waves, currents and other factors in real time. Especially in severe marine environments, traditional navigation route planning methods may not provide the optimal navigation plan, thereby increasing the risk of navigation and energy consumption.

[0003] In recent years, artificial intelligence technology has been widely applied in various fields. Deep learning, as an important branch of artificial intelligence, can effectively process large amounts of complex data by simulating the working mechanism of the human brain, thereby extracting useful information. In the field of navigation route planning, sea wave state analysis technology based on deep learning is particularly attractive.

[0004] This technology uses deep learning networks to monitor and predict sea wave states in real time, taking sea wave states as an important reference factor for navigation route planning, which can help develop more reasonable and safe navigation plans. Deep learning networks can learn and improve themselves, and their prediction accuracy will continue to improve as the amount of training data increases.

[0005] In addition, sea wave state analysis technology based on deep learning can also combine other marine environmental factors such as sea water temperature and salinity, through linkage feature mining, to obtain more comprehensive marine environmental information, providing stronger decision support for navigation route planning.

[0006] However, there are still some problems in the practical application of this technology. Therefore, how to effectively apply this sea wave state analysis technology based on deep learning to navigation route planning is still a problem to be solved. SUMMARY

[0007] In order to improve the above problems, the present application provides a ship navigation route planning method and system based on sea wave state analysis.

[0008] In a first aspect, a ship navigation route planning method based on sea wave state analysis is provided, applied to a navigation route planning system, the method comprising:

[0009] The sea area environment monitoring information sample, the sea wave state trend data sample set corresponding to the sea area environment monitoring information sample, and the sailing route planning authentication result are obtained, the first deep learning network debugging sample corresponding to the sea area environment monitoring information sample, the labeled debugging sample, and the labeled deep learning network debugging sample are obtained; the labeled debugging sample is a sample labeled after the second deep learning network debugging sample is labeled, the labeled deep learning network debugging sample is a remaining sample except the labeled debugging sample in the second deep learning network debugging sample, and the first deep learning network debugging sample and the second deep learning network debugging sample are determined from the sea wave state trend data sample set and the sailing route planning authentication result;

[0010] The sea area environment monitoring information sample and the sea wave state trend data sample set are input into a to-be-debugged sailing route planning decision network to obtain a sailing route planning prediction result, and a target network debugging quality index is generated according to a difference between the sailing route planning prediction result and the sailing route planning authentication result;

[0011] The labeled deep learning network debugging sample and the first deep learning network debugging sample are input into the to-be-debugged sailing route planning decision network to obtain a labeled training result, and a joint network debugging quality index is generated according to a difference between the labeled debugging sample and the labeled training result;

[0012] The neural network configuration weight of the to-be-debugged sailing route planning decision network is optimized according to the target network debugging quality index and the joint network debugging quality index until a training standard is met, and a target sailing route planning decision network is obtained; the target sailing route planning decision network is used for sailing route planning processing according to sea area environment monitoring information.

[0013] In some possible technical solutions, the joint network debugging quality index includes a route updating debugging quality index, and the first deep learning network debugging sample, the labeled debugging sample, and the labeled deep learning network debugging sample corresponding to the sea area environment monitoring information sample are obtained, including:

[0014] The sea wave state trend data sample set is obtained as the first deep learning network debugging sample, and the sailing route planning authentication result is obtained as the second deep learning network debugging sample;

[0015] The sailing route planning authentication result is labeled to obtain a sailing route labeled debugging sample and an energy consumption labeled debugging sample, the sailing route labeled debugging sample is obtained as the labeled debugging sample, and the energy consumption labeled debugging sample is obtained as the labeled deep learning network debugging sample;

[0016] The labeled deep learning network debugging sample and the first deep learning network debugging sample are input into a to-be-debugged voyage route planning decision network to obtain a labeled training result, and a joint network debugging quality index is generated according to a difference between the labeled debugging sample and the labeled training result, including:

[0017] The sea wave state trend data sample set and the energy consumption labeled debugging sample are input into a to-be-debugged voyage route planning decision network to obtain a voyage labeled training result;

[0018] A route updating debugging quality index is generated according to a difference between the voyage labeled debugging sample and the voyage labeled training result.

[0019] In some possible technical solutions, the joint network debugging quality index includes a voyage risk prediction training error, and the first deep learning network debugging sample, the labeled debugging sample and the labeled deep learning network debugging sample corresponding to the sea area environment monitoring information sample are obtained, including:

[0020] The voyage route planning certification result is obtained as the first deep learning network debugging sample, and the sea wave state trend data sample set is obtained as the second deep learning network debugging sample;

[0021] The sea wave state trend data sample set is labeled to obtain a voyage risk labeled debugging sample and a risk trend debugging sample, the voyage risk labeled debugging sample is obtained as the labeled debugging sample, and the risk trend debugging sample is obtained as the labeled deep learning network debugging sample;

[0022] The labeled deep learning network debugging sample and the first deep learning network debugging sample are input into a to-be-debugged voyage route planning decision network to obtain a labeled training result, and a joint network debugging quality index is generated according to a difference between the labeled debugging sample and the labeled training result, including:

[0023] The voyage route planning certification result and the risk trend debugging sample are input into a to-be-debugged voyage route planning decision network to obtain a voyage risk labeled training result;

[0024] A voyage risk prediction training error is generated according to a difference between the voyage risk labeled debugging sample and the voyage risk labeled training result.

[0025] In some possible technical solutions, the sea area environment monitoring information sample and the sea wave state trend data sample set are input into a to-be-debugged voyage route planning decision network to obtain a voyage route planning prediction result, including:

[0026] The marine environment monitoring information sample, the wave state trend data sample set, and the navigation route planning certification results are entered into the navigation route planning decision network to be debugged.

[0027] Knowledge feature mining is performed on the marine environment monitoring information samples and the wave state trend data sample sets respectively to obtain marine environment monitoring knowledge vector samples and wave state trend knowledge vector samples;

[0028] By aggregating the marine environment monitoring knowledge vector examples and the wave state trend knowledge vector examples, a cross-knowledge vector for route planning is obtained.

[0029] Based on the certification results of the navigation route planning, the cross-knowledge vector of the route planning is predicted to obtain the navigation route planning prediction result.

[0030] In some possible technical solutions, the step of performing knowledge feature mining on the marine environment monitoring information samples and the wave state trend data sample sets respectively to obtain marine environment monitoring knowledge vector samples and wave state trend knowledge vector samples includes:

[0031] The marine environment monitoring information samples are subjected to attribute extraction to obtain the original marine environment attribute vector. The original marine environment attribute vector is then subjected to knowledge mapping to obtain the marine environment monitoring knowledge vector samples.

[0032] State event identification is performed on the wave state trend data samples in the wave state trend data sample set to obtain at least one state event inference vector corresponding to a target state event.

[0033] Based on the state event deduction vectors corresponding to each state event in the wave state trend data sample set, the original wave state trend description corresponding to the wave state trend data sample set is obtained.

[0034] Knowledge mapping is performed on the original wave state trend description to obtain a sample of the wave state trend knowledge vector.

[0035] In some possible technical solutions, obtaining the original wave state trend description corresponding to the wave state trend data sample set based on the inference vectors of each state event corresponding to each wave state trend data sample set includes:

[0036] The trend keywords, state event keywords, and distribution features corresponding to each wave state trend data sample are extracted to obtain the trend keyword vector, state event keyword vector, and relative distribution variable corresponding to each distribution feature.

[0037] Based on the state event keyword vector, relative distribution variables, state event inference vector, and trend keyword vector corresponding to the wave state trend data sample, the original stage state trend description of the target state event is obtained.

[0038] The original wave state trend description is obtained based on the original stage state trend description corresponding to each target state event in each wave state trend data sample.

[0039] In some possible technical solutions, the current input information is defined as the original marine environment attribute vector or the original wave state trend description. Knowledge mapping is then performed on the current input information to obtain the corresponding current sample information, including:

[0040] Perform a feature focusing operation on the current input information to obtain the currently focused input information;

[0041] By aggregating the current input information and the currently focused input information, initial linkage input information is obtained;

[0042] Latent space vector mining is performed on the initial linkage input information to obtain the current latent space vector;

[0043] By aggregating the current latent space vector and the initial linkage input information, the target linkage input information is obtained;

[0044] The current sample information is obtained based on the target linkage input information.

[0045] In some possible technical solutions, the aggregation of the marine environment monitoring knowledge vector examples and the wave state trend knowledge vector examples to obtain the route planning cross-knowledge vector includes:

[0046] Perform heterogeneous feature focusing operation on the marine environment monitoring knowledge vector sample and the wave state trend knowledge vector sample to obtain linked focused input information;

[0047] Based on the marine environment monitoring knowledge vector example and the linked focused input information, the marine state element optimization vector is obtained;

[0048] Based on the optimized vector of the sea area state elements, knowledge filtering is performed on the linked focused input information to obtain the route-related knowledge vector;

[0049] The knowledge vectors involved in the route and the marine environmental monitoring knowledge vectors are combined and processed to obtain the cross-knowledge vectors for route planning.

[0050] In some possible technical solutions, obtaining the optimized vector of marine state elements based on the marine environment monitoring knowledge vector example and the linked focused input information includes:

[0051] By combining the marine environment monitoring knowledge vector sample and the linked focused input information, a first combined knowledge vector is obtained;

[0052] Based on the weights configured in the first neural network, latent space vector mining is performed on the first combined knowledge vector to obtain the first latent space vector.

[0053] Vector projection is performed on the first latent space vector to obtain the optimized vector of the sea area state elements.

[0054] In some possible technical solutions, the navigation route planning certification results include multiple prior ship navigation route diagrams with a sequential order;

[0055] Based on the certification results of the navigation route planning, the cross-knowledge vector of the route planning is predicted to obtain the navigation route planning prediction result, including:

[0056] The target planning area is determined from each route area corresponding to the navigation route planning certification results;

[0057] From the navigation route planning and certification results, the prior ship navigation route map before the target planning area is obtained as the reference ship navigation route map. Knowledge feature mining is performed on the reference ship navigation route map to obtain the spatiotemporal vector of the reference ship navigation route.

[0058] Based on the cross-knowledge vector of the route planning and the spatiotemporal vector of the reference ship navigation route, the spatiotemporal vector of the ship navigation route to be processed is obtained;

[0059] Predict the spatiotemporal vector of the ship navigation route to be processed to obtain the ship navigation route map corresponding to the target planning area;

[0060] The next route area is obtained as the target planning area. Then, the process jumps to the step of obtaining the prior ship navigation route map before the target planning area from the navigation route planning certification result as the reference ship navigation route map. This process continues until the termination condition is triggered, resulting in multiple ship navigation route maps to be processed.

[0061] The navigation route planning and prediction results are obtained based on the navigation route maps of each vessel to be processed.

[0062] In some possible technical solutions, obtaining the spatiotemporal vector of the ship's navigation route to be processed based on the cross-knowledge vector of the route planning and the spatiotemporal vector of the reference ship navigation route includes:

[0063] A feature focusing operation is performed on the spatiotemporal vector of the reference ship's navigation route to obtain initial focused input information. Based on the initial focused input information and the spatiotemporal vector of the reference ship's navigation route, the initial spatiotemporal vector of the ship's navigation route is obtained.

[0064] The initial ship navigation route spatiotemporal vector and the route planning cross-knowledge vector are linked and feature-focused to obtain interactive focused input information. Based on the interactive focused input information and the initial ship navigation route spatiotemporal vector, the transition ship navigation route spatiotemporal vector is obtained.

[0065] Latent space vector mining is performed on the spatiotemporal vector of the transition vessel's navigation route to obtain the spatiotemporal vector of the target vessel's navigation route. Based on the spatiotemporal vectors of the transition vessel's navigation route and the target vessel's navigation route, the spatiotemporal vector of the vessel to be processed is obtained.

[0066] In some possible technical solutions, the learning example pool corresponding to the navigation route planning decision network to be debugged includes deep learning network debugging samples corresponding to multiple marine environmental monitoring information samples. The deep learning network debugging samples include marine environmental monitoring information samples and corresponding wave state trend data sample sets, navigation route planning certification results, first deep learning network debugging samples, labeled debugging samples, labeled deep learning network debugging samples and prior navigation energy consumption tolerance values. The learning example pool includes at least one prior navigation energy consumption tolerance value.

[0067] The process involves inputting the marine environment monitoring information samples and the wave state trend data sample set into the navigation route planning decision network to be debugged, obtaining navigation route planning prediction results, and generating target network debugging quality indicators based on the differences between the navigation route planning prediction results and the navigation route planning certification results, including:

[0068] The marine environment monitoring information samples, the corresponding wave state trend data sample set, and the prior navigation energy consumption tolerance value in the learning example pool are entered into the navigation route planning decision network to be debugged, so as to obtain the navigation route planning prediction result matched with the prior navigation energy consumption tolerance value corresponding to the marine environment monitoring information sample.

[0069] Based on the difference between the navigation route planning prediction results and the navigation route planning certification results corresponding to the same marine environmental monitoring information sample, a target local network debugging quality index is generated. Based on the target local network debugging quality index corresponding to each marine environmental monitoring information sample, the target network debugging quality index is obtained.

[0070] The step of inputting the labeled deep learning network debugging samples and the first deep learning network debugging samples into the navigation route planning decision network to be debugged, obtaining labeled training results, and generating a joint network debugging quality index based on the difference between the labeled debugging samples and the labeled training results includes:

[0071] The first deep learning network debugging sample, the labeled deep learning network debugging sample, and the prior navigation energy consumption tolerance value corresponding to the marine environment monitoring information sample in the learning example pool are entered into the navigation route planning decision network to be debugged, so as to obtain the labeled training result paired with the prior navigation energy consumption tolerance value corresponding to the marine environment monitoring information sample.

[0072] A local joint network debugging quality index is generated based on the difference between the labeled debugging samples and the labeled training results corresponding to the environmental monitoring information samples of the same sea area. A joint network debugging quality index is obtained based on the local joint network debugging quality index corresponding to each sea area environmental monitoring information sample.

[0073] In some possible technical solutions, the navigation route planning decision network to be debugged includes a first feature mining branch, a second feature mining branch, and a route planning prediction branch;

[0074] The process involves inputting the marine environment monitoring information samples and the wave state trend data sample set into the navigation route planning decision network to be debugged, obtaining navigation route planning prediction results, and inputting the labeled deep learning network debugging samples and the first deep learning network debugging samples into the navigation route planning decision network to be debugged, obtaining labeled training results, including:

[0075] The marine environment monitoring information sample is entered into the first feature mining branch, the wave state trend data sample set is entered into the second feature mining branch, the linkage feature mining information is obtained based on the output of the first feature mining branch and the second feature mining branch, and the linkage feature mining information and the navigation route planning certification result are entered into the route planning prediction branch to obtain the navigation route planning prediction result.

[0076] The labeled deep learning network debugging samples and the first deep learning network debugging samples are entered into the second feature mining branch to obtain the labeled training results.

[0077] In a second aspect, a navigation route planning system is provided, comprising a processor and a memory that communicate with each other, the processor being configured to retrieve a computer program from the memory and to implement the method described in the first aspect by running the computer program.

[0078] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon, the computer program implementing the method described in the first aspect when it is run.

[0079] The ship navigation route planning method and system based on wave state analysis provided in this application can accurately plan ship navigation routes, significantly reducing risks and energy consumption during navigation. It uses a deep learning network to incorporate factors such as marine environmental monitoring information and wave state trend data to generate high-quality navigation route predictions. These predictions are based not only on current ocean conditions but also on historical navigation route planning verification results, meaning that effective navigation routes have been actually implemented and verified under similar conditions.

[0080] This approach, which combines current marine conditions with historical experience, makes predicted navigation routes more reasonable and reliable, avoiding navigational risks caused by sudden adverse marine conditions such as waves and currents. Furthermore, it can reduce energy consumption by selecting optimal navigation routes, such as avoiding headwinds or counter-currents, thereby improving fuel efficiency.

[0081] This technical solution further utilizes target network tuning quality metrics and joint network tuning quality metrics to optimize the weight configuration of the neural network. This optimization method enables the neural network to learn and improve itself, thereby more accurately predicting navigation routes. Through repeated training and optimization, the network's predictive ability can be continuously improved until it meets the preset training achievement requirements.

[0082] Finally, the resulting target navigation route planning decision network has high practical value. It can not only provide useful references for actual navigation, but also serve as an important component of the ship navigation route planning system, helping ships to formulate optimal navigation plans in various complex marine environments.

[0083] In summary, this technical solution introduces deep learning technology into the field of navigation route planning, effectively solving the problems of risk and energy consumption during navigation, and improving the safety and economic benefits of ship navigation. Attached Figure Description

[0084] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0085] Figure 1 A flowchart illustrating a ship navigation route planning method based on ocean wave state analysis, provided as an embodiment of this application. Detailed Implementation

[0086] To better understand the above technical solutions, the technical solutions of this application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this application and the specific features in the embodiments are detailed descriptions of the technical solutions of this application, rather than limitations on the technical solutions of this application. In the absence of conflict, the embodiments of this application and the technical features in the embodiments can be combined with each other.

[0087] Figure 1 A method for ship navigation route planning based on ocean wave state analysis is presented and applied to a navigation route planning system. The method includes the following steps 110-140.

[0088] Step 110: Obtain marine environmental monitoring information samples and the corresponding wave state trend data sample set and navigation route planning certification results. Obtain the first deep learning network debugging sample, the labeled debugging sample and the labeled deep learning network debugging sample corresponding to the marine environmental monitoring information samples.

[0089] The labeled debugging sample is a sample that has been labeled after the second deep learning network debugging sample has been labeled. The labeled deep learning network debugging sample is the remaining sample in the second deep learning network debugging sample excluding the labeled debugging sample. The first deep learning network debugging sample and the second deep learning network debugging sample are determined from the wave state trend data sample set and the navigation route planning certification result.

[0090] Step 120: Input the marine environment monitoring information sample set and the wave state trend data sample set into the navigation route planning decision network to be debugged, obtain the navigation route planning prediction result, and generate the target network debugging quality index based on the difference between the navigation route planning prediction result and the navigation route planning certification result.

[0091] Step 130: Input the labeled deep learning network debugging samples and the first deep learning network debugging samples into the navigation route planning decision network to be debugged, obtain the labeled training results, and generate a joint network debugging quality index based on the difference between the labeled debugging samples and the labeled training results.

[0092] Step 140: Optimize the neural network configuration weights of the navigation route planning decision network to be debugged based on the target network debugging quality index and the joint network debugging quality index until the training requirements are met, and obtain the target navigation route planning decision network.

[0093] The target navigation route planning decision network is used to perform navigation route planning based on marine environmental monitoring information.

[0094] In this embodiment of the application, the technical terms involved in step 110 include marine environmental monitoring information samples, wave state trend data sample set, navigation route planning certification results, first deep learning network debugging samples, labeled debugging samples, and labeled deep learning network debugging samples.

[0095] Marine environmental monitoring information samples refer to data samples collected from the marine environment, such as wind speed, wind direction, wave height, ocean current direction, and temperature.

[0096] The wave state trend data sample set is based on past marine environmental monitoring information to predict the trend of wave state over a future period of time.

[0097] The route planning certification results are route planning results that have been verified by experts or systems, and can be used as a standard reference for training deep learning networks.

[0098] The first deep learning network debugging samples are sample data used to train the first deep learning network, including marine environmental monitoring information samples and corresponding wave state trend data sample sets.

[0099] Labeled debug samples are second-generation debug samples for deep learning networks after being labeled. They can contain specific attributes or labels to assist deep learning networks in training.

[0100] Labeled deep learning network debug samples are a second type of deep learning network debug samples besides labeled debug samples, and they are also used for training deep learning networks.

[0101] In a sample application scenario, marine environmental monitoring information samples can be obtained from various sources such as satellites and buoys, including hourly wind speed, wind direction, and wave height data for a certain area in the North Atlantic. Then, historical data and relevant models can be used to predict the wave state trend over a future period (e.g., 24 hours), and this data can be used as a sample set of wave state trend data. Simultaneously, expert-developed navigation route planning certification results—that is, optimal navigation paths based on current and expected sea conditions—are available. These paths have been validated and are considered the best choice under given conditions. Next, the marine environmental monitoring information samples and the corresponding wave state trend data sample set are used as the first deep learning network debugging samples. Additionally, a second deep learning network debugging sample set is collected, some of which are specially labeled, such as those indicating specific weather or ocean current conditions, serving as labeled debugging samples; the remainder are used as labeled deep learning network debugging samples. These steps provide the necessary data and resources for training and optimizing the deep learning network, aiming to find the optimal navigation route planning scheme through machine learning techniques.

[0102] In this embodiment of the application, the technical terms involved in step 120 include the navigation route planning decision network to be debugged, the navigation route planning prediction result, and the target network debugging quality index.

[0103] The navigation route planning decision network to be debugged is a deep learning network that has not yet been fully trained and optimized. Its goal is to predict the optimal navigation route based on the input marine environmental monitoring information samples and wave state trend data sample sets.

[0104] The navigation route planning prediction result is the predicted value output by the navigation route planning decision network to be debugged, that is, the optimal navigation route predicted by the network under the given marine environmental monitoring information and wave state trend.

[0105] The target network debugging quality index (target network debugging loss) is a metric for measuring the performance of the navigation route planning decision network to be debugged. It is usually obtained by calculating the difference between the navigation route predicted by the network and the actual certification result. This difference is called "loss," and the smaller the loss, the higher the accuracy of the network prediction.

[0106] For example, sample data on marine environmental monitoring and corresponding wave state trend data for a certain region of the North Atlantic have already been collected. Next, this data will be input into the navigation route planning decision network to be tested.

[0107] After a series of complex calculations and processing, the network finally outputs a prediction result, which is the safest and most efficient navigation route under a given marine environment and wave state trend. This is the so-called "navigation route planning prediction result".

[0108] Then, this prediction is compared with the certified route planning results verified by experts or the system, and the difference between the two is calculated. This difference is used to generate a target network debugging quality metric, commonly known as the "loss". For example, if the predicted route differs greatly from the actual certified route, the loss will be high; conversely, if the two are very close, the loss will be low.

[0109] In this way, the navigation route planning decision network can be continuously optimized and adjusted, enabling it to more accurately predict the optimal navigation route in the future.

[0110] In this embodiment of the application, the technical terms involved in step 130 include labeled training results and joint network debugging quality indicators.

[0111] The labeled training results refer to the prediction results obtained after inputting labeled deep learning network debugging samples and first deep learning network debugging samples into the navigation route planning decision network to be debugged. These results will be processed or compared differently depending on the label of the samples.

[0112] The joint network debugging quality index is a metric for measuring the performance of a navigation route planning decision network under debugging. It is primarily obtained by calculating the difference between labeled debugging samples and labeled training results. This difference, also known as the "reconstruction loss," reflects the network's accuracy in processing data with specific labels.

[0113] For example, we have already obtained the first deep learning network debugging sample and the labeled deep learning network debugging sample corresponding to the marine environmental monitoring information sample of a certain area in the North Atlantic.

[0114] Next, this data is input into the navigation route planning decision network to be debugged. After a series of complex calculations and processing, the network finally outputs a prediction result, which is the safest and most efficient navigation route under the given ocean environment and wave state trends. This is called "labeled training result".

[0115] The prediction result is then compared with the marked debugging samples, and the difference between the two is calculated. This difference is used to generate a joint network debugging quality metric, commonly known as the "reconstruction loss". For example, if the predicted route differs greatly from the actual marked route, the reconstruction loss will be high; conversely, if the two are very close, the reconstruction loss will be low.

[0116] In this way, the navigation route planning decision network can be continuously optimized and adjusted, enabling it to more accurately predict the optimal navigation route in the future, especially for situations with specific markers.

[0117] In this embodiment of the application, the technical terms involved in step 140 include neural network configuration weights and target navigation route planning decision network.

[0118] Neural network weight configuration: In a neural network, each connection (i.e., the connection between neurons) has a weight, which determines the influence of each input signal on the output. By optimizing these weights, the performance of the network can be improved.

[0119] The target navigation route planning decision network is an optimized navigation route planning decision network that meets specific training objectives. It can perform more accurate navigation route planning based on marine environmental monitoring information.

[0120] In the previous step (step 130), the target network debugging quality index and the joint network debugging quality index were obtained. These two indexes reflect the current network performance and accuracy.

[0121] Now, this information will be used to optimize the neural network configuration weights of the decision network for the navigation route planning task. Specifically, if the target network's or joint network's debugging quality index is too high (i.e., the predicted results differ significantly from the actual results), the weights in the neural network will be adjusted accordingly to make the network's predictions closer to the actual results. This process may require multiple iterations and optimizations. This optimization process will continue until the network's performance reaches the preset training qualification requirements. Once these requirements are met, the target navigation route planning decision network is obtained.

[0122] This target network will be used to process more accurate navigation routes based on marine environmental monitoring information. For example, it may determine the safest and fastest navigation route for ships based on information such as waves, wind direction, and ocean currents.

[0123] The following section uses a cargo ship navigation route planning example to explain in detail steps 110-140 above.

[0124] First, sample marine environmental monitoring information from a certain region of the North Atlantic was acquired, including wave height, wind speed, and wind direction, along with corresponding wave trend data samples, such as the potential increase in wave height over the next few days. Simultaneously, the actual navigation routes chosen by similar vessels under identical or similar conditions in the past were collected as navigation route planning verification results. Next, based on this data, first and second deep learning network debugging samples were generated. The labeled debugging samples are those that have been marked, for example, those indicating particularly difficult navigation conditions; while the labeled deep learning network debugging samples are the remaining samples in the second deep learning network debugging samples excluding the labeled debugging samples.

[0125] Then, sample sets of marine environmental monitoring information and wave state trend data are input into the navigation route planning decision network to be debugged. This network predicts a navigation route through complex calculations and processing. This prediction result is compared with previously obtained navigation route planning certification results to generate a target network debugging quality index, which reflects the gap between the predicted and actual results.

[0126] The labeled deep learning network debugging samples and the first deep learning network debugging samples are then input again into the navigation route planning decision network to be debugged, resulting in a new navigation route prediction. Finally, based on the difference between this prediction and the labeled debugging samples, a joint network debugging quality index is generated.

[0127] Finally, based on the target network debugging quality index and the joint network debugging quality index, the weight configuration of the neural network is adjusted and optimized to make the predicted navigation route closer to the actual selected navigation route. This process may require multiple iterations until the preset training achievement requirements are met, that is, both the target network debugging quality index and the joint network debugging quality index are within acceptable ranges. At this point, the target navigation route planning decision network is obtained, which can predict the optimal navigation route based on marine environmental monitoring information.

[0128] The implementation of the technical solution in this application enables precise planning of ship navigation routes, significantly reducing risks and energy consumption during navigation. It utilizes a deep learning network to incorporate factors such as marine environmental monitoring information and wave trend data to generate high-quality navigation route predictions. These predictions are based not only on current ocean conditions but also on historical navigation route planning and validation results—that is, navigation routes that have been actually implemented and validated under similar conditions.

[0129] This approach, which combines current marine conditions with historical experience, makes predicted navigation routes more reasonable and reliable, avoiding navigational risks caused by sudden adverse marine conditions such as waves and currents. Furthermore, it can reduce energy consumption by selecting optimal navigation routes, such as avoiding headwinds or counter-currents, thereby improving fuel efficiency.

[0130] This technical solution further utilizes target network tuning quality metrics and joint network tuning quality metrics to optimize the weight configuration of the neural network. This optimization method enables the neural network to learn and improve itself, thereby more accurately predicting navigation routes. Through repeated training and optimization, the network's predictive ability can be continuously improved until it meets the preset training achievement requirements.

[0131] Finally, the resulting target navigation route planning decision network has high practical value. It can not only provide useful references for actual navigation, but also serve as an important component of the ship navigation route planning system, helping ships to formulate optimal navigation plans in various complex marine environments.

[0132] In summary, this technical solution introduces deep learning technology into the field of navigation route planning, effectively solving the problems of risk and energy consumption during navigation, and improving the safety and economic benefits of ship navigation.

[0133] In some possible embodiments, the joint network debugging quality index includes the line update debugging quality index. Then, in step 110, obtaining the first deep learning network debugging sample, the labeled debugging sample, and the labeled deep learning network debugging sample corresponding to the marine environment monitoring information sample includes: obtaining the wave state trend data sample set as the first deep learning network debugging sample, obtaining the navigation route planning certification result as the second deep learning network debugging sample; performing labeling processing on the navigation route planning certification result to obtain the route labeled debugging sample and the energy consumption labeled debugging sample, obtaining the route labeled debugging sample as the labeled debugging sample, and obtaining the energy consumption labeled debugging sample as the labeled deep learning network debugging sample.

[0134] Step 130, which involves inputting the labeled deep learning network debugging samples and the first deep learning network debugging samples into the navigation route planning decision network to be debugged, to obtain the labeled training results, and generating a joint network debugging quality index based on the difference between the labeled debugging samples and the labeled training results, includes: inputting the wave state trend data sample set and the energy consumption labeled debugging samples into the navigation route planning decision network to be debugged, to obtain the route label training results; and generating a route update debugging quality index based on the difference between the route label debugging samples and the route label training results.

[0135] Taking the navigation route planning of a large cargo ship in the North Pacific as an example, the following steps illustrate the process: First, in step 110, a sample set of wave state trend data for a certain region of the North Pacific is acquired, such as information on wave height, wind direction, and wind speed for the next few days. This data is used as the first deep learning network debugging sample. Simultaneously, the effective navigation routes actually selected and verified by similar cargo ships under similar conditions in the past are collected, i.e., the navigation route planning certification results, and these are used as the second deep learning network debugging sample.

[0136] Next, the route planning and certification results are labeled, for example, identifying which routes require more fuel in actual execution (energy consumption labeling debugging samples) and which routes face more navigation difficulties in actual execution (route labeling debugging samples). Based on this, the route labeling debugging samples are used as labeling debugging samples, and the energy consumption labeling debugging samples are used as labeling deep learning network debugging samples.

[0137] Then, in step 130, the wave state trend data sample set and energy consumption marker debugging samples are input into the navigation route planning decision network to be debugged. After network calculation and processing, the route marker training results are obtained. Then, based on the difference between this training result and the route marker debugging samples, a route update debugging quality index is generated.

[0138] In this process, the route update and debugging quality indicators are mainly used to evaluate the difference between the navigation routes predicted by the network and the actual, executed, and verified effective navigation routes. This approach allows for a better understanding of the network's performance and enables optimization of the network configuration to predict navigation routes more accurately.

[0139] This design combines various information sources, including wave state trend data, navigation route planning certification results, route marker adjustment samples, and energy consumption marker adjustment samples, to generate high-quality navigation route predictions through a deep learning network. This method not only improves the accuracy of navigation route planning but also provides ships with more navigation options, helping to reduce risks and energy consumption during navigation.

[0140] In other examples, the joint network debugging quality index includes the navigation risk prediction training error. Therefore, step 110, obtaining the first deep learning network debugging sample, labeled debugging sample, and labeled deep learning network debugging sample corresponding to the marine environment monitoring information sample, includes: obtaining the navigation route planning certification result as the first deep learning network debugging sample, and obtaining the wave state trend data sample set as the second deep learning network debugging sample; performing labeling processing on the wave state trend data sample set to obtain navigation risk labeled debugging samples and risk trend debugging samples; obtaining the navigation risk labeled debugging sample as the labeled debugging sample, and obtaining the risk trend debugging sample as the labeled deep learning network debugging sample.

[0141] Step 130, which involves inputting the labeled deep learning network debugging samples and the first deep learning network debugging samples into the navigation route planning decision network to be debugged, to obtain labeled training results, and generating a joint network debugging quality index based on the difference between the labeled debugging samples and the labeled training results, includes: inputting the navigation route planning certification results and the risk trend debugging samples into the navigation route planning decision network to be debugged, to obtain navigation risk record training results; and generating a navigation risk prediction training error based on the difference between the navigation risk labeled debugging samples and the navigation risk record training results.

[0142] Taking the navigation route planning of a medium-sized fishing vessel in the Pacific Ocean as an example, the following steps illustrate the process: First, in step 110, the certification results of navigation route planning in a certain area of ​​the Pacific Ocean over a past period are obtained, such as which routes have been actually executed and verified as effective. These are used as the first deep learning network debugging samples. Simultaneously, a sample set of wave state trend data for the next few days is collected, including expected wave height, wind direction, and wind speed. This information is used as the second deep learning network debugging samples.

[0143] Next, the sample set of wave state trend data is labeled, for example, each sample is labeled as "low risk," "medium risk," or "high risk" according to the wind and wave conditions, generating navigation risk labeling debugging samples. At the same time, risk trend debugging samples are generated based on the risk trends analyzed from historical wind and wave data (such as stormy seasons or calm seasons) to predict possible future changes in risk levels.

[0144] Then, in step 130, the navigation route planning certification results and risk trend debugging samples are input into the navigation route planning decision network to be debugged. After processing the input data, the network obtains the navigation risk training results. Then, based on the difference between this training result and the navigation risk labeled debugging samples, the navigation risk prediction training error is generated.

[0145] In this process, the navigation risk prediction training error is mainly used to evaluate the difference between the navigation risk level predicted by the network and the actual labeled risk level. This allows for a better understanding of the network's performance and enables optimization of the network configuration to more accurately predict navigation risks.

[0146] This design combines various information sources, including navigation route planning and certification results, wave state trend data, navigation risk marker adjustment samples, and risk trend adjustment samples, to generate high-quality navigation risk predictions through a deep learning network. This method not only improves the accuracy of navigation risk predictions but also provides ships with a more scientific basis for navigation decisions, helping to reduce safety risks during navigation.

[0147] In some preferred embodiments, step 120, which involves inputting the marine environment monitoring information samples and the wave state trend data sample set into the navigation route planning decision network to be debugged, to obtain the navigation route planning prediction result, includes: inputting the marine environment monitoring information samples, the wave state trend data sample set, and the navigation route planning authentication result into the navigation route planning decision network to be debugged; performing knowledge feature mining on the marine environment monitoring information samples and the wave state trend data sample set to obtain marine environment monitoring knowledge vector samples and wave state trend knowledge vector samples; aggregating the marine environment monitoring knowledge vector samples and the wave state trend knowledge vector samples to obtain a route planning cross-knowledge vector; and predicting the route planning cross-knowledge vector based on the navigation route planning authentication result to obtain the navigation route planning prediction result.

[0148] Taking the navigation route planning of a commercial transport ship in the Atlantic Ocean as an example, the process is as follows: First, in step 120, sample marine environmental monitoring information for a certain area of ​​the Atlantic Ocean is collected, such as seawater temperature, salinity, and current speed. Simultaneously, sample data on wave trends for the next few days are also acquired, including predicted wave height, wind direction, and wind speed. Additionally, the navigation route planning certification results for this area over a past period are obtained, i.e., which routes have been actually executed and verified as effective. All three parts of this information are entered into the navigation route planning decision network to be debugged.

[0149] Next, knowledge feature mining is performed on the sample sets of marine environmental monitoring information and wave state trend data. For example, using statistical methods and deep learning techniques, important features related to navigation route planning are extracted, such as the intensity and frequency of waves under certain seawater temperature and salinity conditions. Through this process, marine environmental monitoring knowledge vector samples and wave state trend knowledge vector samples are obtained.

[0150] Then, the marine environment monitoring knowledge vector examples and the wave state trend knowledge vector examples are aggregated to obtain the route planning cross-knowledge vector. This cross-knowledge vector integrates information on the marine environment and wave state, providing more comprehensive data support for navigation route planning.

[0151] Finally, based on the navigation route planning certification results, the cross-knowledge vectors of route planning are predicted. For example, using neural networks or other machine learning algorithms, based on past navigation route planning practices, the optimal route is predicted for a given sea environment and wave conditions. Thus, the navigation route planning prediction results are obtained.

[0152] In this way, by comprehensively utilizing marine environmental monitoring information, wave state trend data, and navigation route planning certification results, deep knowledge features are extracted and cross-knowledge vectors for route planning are generated, thereby predicting the optimal navigation route plan. This method not only improves the accuracy and effectiveness of navigation route planning but also helps reduce the risks and energy consumption of ship navigation, thereby increasing navigation efficiency.

[0153] In some other preferred embodiments, the step of performing knowledge feature mining on the marine environment monitoring information samples and the wave state trend data sample set to obtain marine environment monitoring knowledge vector samples and wave state trend knowledge vector samples includes: extracting attributes from the marine environment monitoring information samples to obtain original marine environment attribute vectors; performing knowledge mapping on the original marine environment attribute vectors to obtain marine environment monitoring knowledge vector samples; identifying state events from the wave state trend data samples in the wave state trend data sample set to obtain state event inference vectors corresponding to at least one target state event; obtaining the original wave state trend description corresponding to the wave state trend data sample set based on the state event inference vectors corresponding to each wave state trend data sample in the wave state trend data sample set; and performing knowledge mapping on the original wave state trend description to obtain wave state trend knowledge vector samples.

[0154] Taking a research vessel sailing in the Arctic Ocean as an example, the process is as follows: First, in step 120, sample marine environmental monitoring information of a certain region of the Arctic Ocean is acquired, such as seawater temperature, salinity, and climate conditions. Then, attributes are extracted from these marine environmental monitoring information samples, such as converting continuous temperature data into temperature ranges, resulting in raw marine environmental attribute vectors. Next, through knowledge mapping, these raw marine environmental attribute vectors are converted into marine environmental monitoring knowledge vector samples with semantic information, such as "extremely cold region" and "low salinity."

[0155] Simultaneously, a sample set of wave state trend data for the next few days was collected, including information such as predicted wave height, wind direction, and wind speed. State event identification was performed on these wave state trend data samples, such as identifying possible extreme weather events (e.g., storms), obtaining a state event projection vector corresponding to at least one target state event. Then, based on the state event projection vectors corresponding to all wave state trend data samples, an original wave state trend description was generated, such as "There may be two storm events within the next three days." Next, knowledge mapping was performed on these original wave state trend descriptions to obtain wave state trend knowledge vector samples with semantic information, such as "Frequent storms recently."

[0156] Thus, by extracting attributes and identifying state events from sample sets of marine environmental monitoring information and wave state trend data, and further performing knowledge mapping, we obtained semantically rich marine environmental monitoring knowledge vector samples and wave state trend knowledge vector samples. These knowledge vectors can better reflect the actual situation of the marine environment and wave state, and provide more accurate and comprehensive information support for navigation route planning, thereby effectively improving the accuracy and safety of navigation route planning.

[0157] In another preferred embodiment, obtaining the original wave state trend description corresponding to the wave state trend data sample set based on the state event inference vectors corresponding to each wave state trend data sample set includes: extracting attributes from the trend keywords corresponding to each wave state trend data sample set, the state event keywords corresponding to each target state event, and the distribution features corresponding to each target state event, to obtain the trend keyword vectors corresponding to each wave state trend data sample set, the state event keyword vectors corresponding to each target state event, and the relative distribution variables corresponding to each distribution feature; obtaining the original stage state trend description corresponding to the target state event based on the state event keyword vector, relative distribution variables, state event inference vector, and the trend keyword vector corresponding to the wave state trend data sample set; and obtaining the original wave state trend description based on the original stage state trend descriptions corresponding to each target state event in each wave state trend data sample set.

[0158] Let's take a yacht sailing in the Mediterranean Sea as an example: First, in step 120, a sample set of wave state trend data for the next few days is acquired, including information such as expected wave height, wind direction, and wind speed. Then, state events are identified for each wave state trend data sample, such as possible weather conditions like storms or clear skies; these are identified as target state events.

[0159] Next, the trend keywords (e.g., "wind gradually increases") corresponding to each wave state trend data sample, the state event keywords (e.g., "storm") corresponding to each target state event, and the distribution characteristics (e.g., the probability of a storm occurring) corresponding to each target state event are extracted to obtain the trend keyword vector, the state event keyword vector, and the relative distribution variable corresponding to each distribution characteristic for each wave state trend data sample.

[0160] Then, based on the state event keyword vector, relative distribution variables, state event inference vector, and trend keyword vector corresponding to the same target state event, the original stage state trend description corresponding to the target state event is obtained. For example, for the target state event "storm", a description such as "as the wind force gradually increases, the probability of a storm occurring also increases" might be obtained.

[0161] Finally, based on the original stage state trend descriptions corresponding to each target state event in each wave state trend data sample, the original wave state trend description corresponding to the entire wave state trend data sample set is obtained. For example, all descriptions of states such as "storm" and "sunny" are integrated to form a complete wave state trend description.

[0162] Thus, by deeply mining and refining the attributes of the sample data on ocean wave state trends, rich and meaningful original descriptions of ocean wave state trends are generated, which are of significant reference value for navigation route planning. This not only more accurately reflects the changing trends of ocean wave states but also identifies important state events that may affect navigation safety, thereby helping ships make more scientific and safer navigation decisions.

[0163] In some optional examples, if the current input information is the original marine environment attribute vector or the original wave state trend description, then knowledge mapping is performed on the current input information to obtain the corresponding current sample information, including: performing feature focusing operation on the current input information to obtain current focused input information; aggregating the current input information and the current focused input information to obtain initial linkage input information; performing latent space vector mining on the initial linkage input information to obtain current latent space vector; aggregating the current latent space vector and the initial linkage input information to obtain target linkage input information; and obtaining the current sample information based on the target linkage input information.

[0164] Taking a cargo ship sailing in the Indian Ocean as an example, the following steps illustrate the process: First, in step 120, sample data of marine environmental monitoring information and wave state trend data for a certain area of ​​the Indian Ocean are acquired. This information is then processed to obtain the original marine environmental attribute vector and the original wave state trend description.

[0165] When the current input information is the original marine environmental attribute vector, a feature focusing operation is first performed on it, for example, by selecting key features such as seawater temperature and salinity to obtain the current focused input information. Then, the current input information (i.e., the original marine environmental attribute vector) and the current focused input information are aggregated to obtain the initial linked input information.

[0166] Next, latent space vector mining is performed on the initial linkage input information. This process can use deep learning methods, such as autoencoders, to obtain the current latent space vector, which contains deeper-level marine environmental feature information.

[0167] Then, the current latent space vector and the initial linkage input information are aggregated to obtain the target linkage input information, which simultaneously contains the original features, focused features, and deep features.

[0168] Finally, based on the target linkage input information, the current sample information is obtained. This information will be used to input the navigation route planning decision network to perform navigation route planning prediction.

[0169] It is evident that by deeply processing and mining the original marine environmental attribute vectors or original wave state trend descriptions, richer and deeper information is obtained. This not only more accurately reflects the actual situation of the marine environment and wave state, but also provides more comprehensive data support for navigation route planning, thereby improving the accuracy and safety of navigation route planning.

[0170] In some alternative embodiments, the step of aggregating the marine environment monitoring knowledge vector samples and the wave state trend knowledge vector samples to obtain the route planning cross-knowledge vector includes: performing heterogeneous feature focusing operations on the marine environment monitoring knowledge vector samples and the wave state trend knowledge vector samples to obtain linked focused input information; obtaining marine state element optimization vectors based on the marine environment monitoring knowledge vector samples and the linked focused input information; performing knowledge filtering on the linked focused input information based on the marine state element optimization vectors to obtain route-related knowledge vectors; and combining the route-related knowledge vectors and the marine environment monitoring knowledge vector samples to obtain the route planning cross-knowledge vector.

[0171] Taking a research vessel sailing in the Atlantic Ocean as an example, the process is as follows: First, in step 120, marine environmental monitoring knowledge vector samples (e.g., "high temperature," "low salinity") and wave state trend knowledge vector samples (e.g., "increasing wind force," "high wave height") for a certain area of ​​the Atlantic Ocean are acquired. Then, heterogeneous feature focusing operations are performed on these two different sources of information, such as selecting features closely related to navigation safety and comfort, to obtain linked focused input information.

[0172] Next, based on the marine environment monitoring knowledge vector examples and the linked focused input information, an optimization algorithm (such as linear regression, support vector machine, etc.) is used to obtain the optimized vector of marine state elements. This vector may contain the core influencing factors of the marine environment, such as "temperature" and "salinity".

[0173] Then, based on the optimized vector of sea area state elements, knowledge filtering is performed on the linked focused input information, retaining only the knowledge features related to the optimized vector to obtain the route-related knowledge vector. For example, if "temperature" and "salinity" are considered core influencing factors in the optimized vector, then the route-related knowledge vector may include information such as "high temperature" and "low salinity".

[0174] Finally, the knowledge vectors related to the route and the marine environment monitoring are combined to obtain the cross-knowledge vector for route planning. This vector integrates key features of the marine environment and wave conditions, providing a comprehensive and accurate reference for subsequent navigation route planning.

[0175] It is evident that by deeply mining, optimizing, and combining the knowledge vector samples from marine environmental monitoring and wave state trends, a more comprehensive and accurate cross-knowledge vector for route planning is obtained. This method retains important features from the original information while incorporating key features that significantly impact navigation, thus providing more effective support for route planning and improving navigation safety and efficiency.

[0176] In one possible design approach, obtaining the optimized vector of marine state elements based on the marine environment monitoring knowledge vector examples and the linked focused input information includes: combining the marine environment monitoring knowledge vector examples and the linked focused input information to obtain a first combined knowledge vector; performing latent space vector mining on the first combined knowledge vector based on the weights configured in the first neural network to obtain a first latent space vector; and performing vector projection on the first latent space vector to obtain the optimized vector of marine state elements.

[0177] Taking a research vessel conducting scientific research in the South China Sea as an example, the following explanation is provided: First, in step 120, marine environmental monitoring knowledge vector samples (e.g., "high temperature," "low salinity") and wave state trend knowledge vector samples (e.g., "wind force gradually increases," "waves gradually rise") for a certain area of ​​the South China Sea are acquired. Then, heterogeneous feature focusing operations are performed on these two types of information from different sources, such as selecting features closely related to scientific research and equipment safety, to obtain linked focused input information.

[0178] Next, the marine environmental monitoring knowledge vector sample and the linked focused input information are combined to obtain the first combined knowledge vector. This vector contains the original marine environmental attribute information and the key feature information after focusing processing.

[0179] Then, based on the weights configured in the first neural network, latent space vector mining is performed on the first combined knowledge vector. This process can use deep learning methods, such as autoencoders, to obtain the first latent space vector. This vector contains deeper information about the marine environment and wave state features.

[0180] Finally, vector projection is performed on the first latent space vector, for example, through dimensionality reduction methods such as principal component analysis (PCA), to obtain an optimized vector of marine state elements. This vector retains the most important feature information while reducing the dimensionality and complexity of the data.

[0181] In this way, through in-depth processing, mining, and optimization of marine environmental monitoring knowledge vector samples and linked focused input information, a more refined and useful optimized vector of marine state elements is obtained. This method can not only better reflect the actual situation of the marine environment and wave conditions, but also provide more accurate data support for scientific research and equipment safety, thereby improving the accuracy of research and the safety of equipment.

[0182] In some examples, the navigation route planning certification results include multiple prior ship navigation route maps with a sequential order. The step of predicting the navigation route planning cross-knowledge vector based on the navigation route planning certification results to obtain the navigation route planning prediction result includes: determining a target planning area from each route area corresponding to the navigation route planning certification results; obtaining a prior ship navigation route map preceding the target planning area from the navigation route planning certification results as a reference ship navigation route map, performing knowledge feature mining on the reference ship navigation route map to obtain a reference ship navigation route spatiotemporal vector; obtaining a ship navigation route spatiotemporal vector to be processed based on the navigation route planning cross-knowledge vector and the reference ship navigation route spatiotemporal vector; predicting the ship navigation route spatiotemporal vector to be processed to obtain a ship navigation route map to be processed corresponding to the target planning area; obtaining the next route area as the target planning area, and then proceeding to the step of obtaining a prior ship navigation route map preceding the target planning area from the navigation route planning certification results as a reference ship navigation route map, until a termination condition is triggered, resulting in multiple ship navigation route maps to be processed; and obtaining the navigation route planning prediction result based on each ship navigation route map to be processed.

[0183] Let's take a cargo ship currently on a long-haul voyage in the North Atlantic as an example: First, in step 120, a set of prior ship navigation route maps with a chronological order are obtained. These maps are the results of historical navigation route planning verification for similar routes. Furthermore, the route planning cross-knowledge vector has already been obtained through the preceding steps.

[0184] Next, the first target planning area is determined from the various route regions corresponding to this set of navigation route planning certification results, such as the northeastern region of the North Atlantic. Then, prior vessel navigation route maps that have existed before this target planning area are obtained from the navigation route planning certification results as reference vessel navigation route maps. For example, navigation route maps of vessels that have previously navigated the same area in this season and whose navigation conditions are good are selected.

[0185] Then, knowledge feature mining is performed on the reference ship's navigation route map to obtain the spatiotemporal vector of the reference ship's navigation route. This vector may include factors such as navigation time, speed, weather conditions, and sea conditions.

[0186] Next, based on the cross-knowledge vector of route planning and the spatiotemporal vector of the reference ship's navigation route, the spatiotemporal vector of the ship's navigation route to be processed is obtained. This vector contains the actual situation and challenges that the cargo ship may face.

[0187] Then, the spatiotemporal vector of the ship's navigation route to be processed is predicted to obtain the navigation route map of the target planning area. This map provides a preliminary navigation route plan for cargo ships.

[0188] Next, the next route area is obtained as the new target planning area, and the above process is repeated until the termination condition is triggered, such as when all route areas have been planned and multiple ship navigation route maps to be processed are obtained.

[0189] Finally, based on the navigation route diagrams of each vessel to be processed, navigation route planning and prediction results were obtained. These results can guide cargo ships to complete long-distance voyages safely and effectively.

[0190] Thus, by incorporating historical navigation route maps and combining them with advanced knowledge feature mining and prediction methods, practically valuable navigation route planning prediction results are generated. This not only improves navigation safety and efficiency but also provides more experience and reference for future navigation route planning.

[0191] Under some possible design approaches, obtaining the spatiotemporal vector of the ship's navigation route to be processed based on the cross-knowledge vector of the route planning and the spatiotemporal vector of the reference ship's navigation route includes: performing a feature focusing operation on the spatiotemporal vector of the reference ship's navigation route to obtain initial focused input information; obtaining the initial spatiotemporal vector of the ship's navigation route based on the initial focused input information and the spatiotemporal vector of the reference ship's navigation route; performing a linked feature focusing operation on the initial spatiotemporal vector of the ship's navigation route and the cross-knowledge vector of the route planning to obtain interactive focused input information; obtaining the transitional spatiotemporal vector of the ship's navigation route based on the interactive focused input information and the initial spatiotemporal vector of the ship's navigation route; performing latent space vector mining on the transitional spatiotemporal vector of the ship's navigation route to obtain the spatiotemporal vector of the target ship's navigation route; and obtaining the spatiotemporal vector of the ship's navigation route to be processed based on the spatiotemporal vector of the transitional ship's navigation route and the spatiotemporal vector of the target ship's navigation route.

[0192] Let's take a cruise ship currently on a long-distance voyage in the Mediterranean as an example: First, in step 120, the spatiotemporal vector of the reference ship's route is obtained. This vector may contain information such as the sailing time, speed, weather conditions, and sea conditions of similar routes in the past. Furthermore, the route planning cross-knowledge vector has already been obtained through the preceding steps.

[0193] Next, a feature focusing operation is performed on the spatiotemporal vector of the reference ship's navigation route. For example, navigation time and wave conditions, two features that have a significant impact on navigation, are selected to obtain initial focused input information. Then, based on the initial focused input information and the spatiotemporal vector of the reference ship's navigation route, an initial spatiotemporal vector of the ship's navigation route is obtained. This vector contains more focused feature information.

[0194] Then, the initial ship navigation route spatiotemporal vector and the route planning cross-knowledge vector are linked and feature-focused to obtain interactive focused input information. For example, based on the navigation time and wave conditions in the initial ship navigation route spatiotemporal vector, combined with the environmental influencing factors in the route planning cross-knowledge vector, feature-focusing is performed. Next, based on the interactive focused input information and the initial ship navigation route spatiotemporal vector, the transition ship navigation route spatiotemporal vector is obtained.

[0195] Next, latent space vector mining is performed on the spatiotemporal vectors of the transitional vessel's navigation route to obtain the spatiotemporal vectors of the target vessel's navigation route. This process can utilize deep learning methods, such as autoencoders, to extract deeper-level feature information.

[0196] Finally, based on the spatiotemporal vectors of the transition vessel's route and the target vessel's route, the spatiotemporal vector of the vessel's route to be processed is obtained. This vector will be used for subsequent route prediction.

[0197] In this way, by performing multi-level feature focusing, linkage, and in-depth mining on the spatiotemporal vector of the reference ship's navigation route, a more refined and useful spatiotemporal vector of the ship's navigation route to be processed is obtained. This method can not only better reflect the actual situation and challenges that cruise ships may face, but also provide more accurate data support for navigation route planning, thereby improving the safety and efficiency of navigation.

[0198] In some exemplary embodiments, the learning example pool corresponding to the navigation route planning decision network to be debugged includes deep learning network debugging samples corresponding to multiple marine environmental monitoring information samples. The deep learning network debugging samples include marine environmental monitoring information samples and corresponding wave state trend data sample sets, navigation route planning certification results, first deep learning network debugging samples, labeled debugging samples, labeled deep learning network debugging samples, and prior navigation energy consumption tolerance values. The learning example pool includes at least one prior navigation energy consumption tolerance value. The step of inputting the marine environment monitoring information samples and the wave state trend data sample set into the navigation route planning decision network to be debugged, obtaining the navigation route planning prediction result, and generating the target network debugging quality index based on the difference between the navigation route planning prediction result and the navigation route planning certification result includes: inputting the marine environment monitoring information samples and the corresponding wave state trend data sample set and the prior navigation energy consumption tolerance value from the learning example pool into the navigation route planning decision network to be debugged, obtaining the navigation route planning prediction result paired with the prior navigation energy consumption tolerance value corresponding to the marine environment monitoring information sample; generating the target local network debugging quality index based on the difference between the navigation route planning prediction result and the navigation route planning certification result corresponding to the same marine environment monitoring information sample; and obtaining the target network debugging quality index based on the target local network debugging quality index corresponding to each marine environment monitoring information sample.

[0199] The step of inputting the labeled deep learning network debugging samples and the first deep learning network debugging samples into the navigation route planning decision network to be debugged, obtaining labeled training results, and generating a joint network debugging quality index based on the difference between the labeled debugging samples and the labeled training results includes: inputting the first deep learning network debugging samples, labeled deep learning network debugging samples, and prior navigation energy consumption tolerance values ​​corresponding to the marine environmental monitoring information samples in the learning example pool into the navigation route planning decision network to be debugged, obtaining labeled training results paired with the prior navigation energy consumption tolerance values ​​corresponding to the marine environmental monitoring information samples; generating a local joint network debugging quality index based on the difference between the labeled debugging samples and the labeled training results corresponding to the same marine environmental monitoring information samples; and obtaining a joint network debugging quality index based on the local joint network debugging quality indexes corresponding to each marine environmental monitoring information sample.

[0200] Taking a cargo ship on a long-haul voyage in the North Pacific as an example, the following steps are illustrated: First, in step 120, marine environmental monitoring information samples (such as seawater temperature and salinity), corresponding wave state trend data sample sets, navigation route planning certification results, first deep learning network debugging samples, labeled debugging samples, labeled deep learning network debugging samples, and prior navigation energy consumption tolerance values ​​are acquired. These are all stored in the learning example pool corresponding to the navigation route planning decision network to be debugged.

[0201] Next, the marine environmental monitoring information samples, the corresponding wave state trend data sample set, and the prior navigation energy consumption tolerance value from the learning example pool are entered into the navigation route planning decision network to be debugged, so as to obtain the navigation route planning prediction results paired with the prior navigation energy consumption tolerance value corresponding to the marine environmental monitoring information sample.

[0202] Then, based on the difference between the navigation route planning prediction results and the navigation route planning verification results corresponding to the environmental monitoring information samples of the same sea area, a target local network debugging quality index is generated. For example, the mean squared error (MSE) or cross entropy loss can be used to measure this difference. Then, based on the target local network debugging quality index corresponding to each marine environmental monitoring information sample, the target network debugging quality index is obtained.

[0203] Similarly, the first deep learning network debugging sample, the labeled deep learning network debugging sample, and the prior navigation energy consumption tolerance value corresponding to the marine environmental monitoring information sample in the learning example pool are entered into the navigation route planning decision network to be debugged, so as to obtain the labeled training result paired with the prior navigation energy consumption tolerance value corresponding to the marine environmental monitoring information sample.

[0204] Next, based on the differences between the labeled debugging samples and the labeled training results corresponding to the environmental monitoring information examples in the same sea area, a local joint network debugging quality index is generated. Then, based on the local joint network debugging quality indices corresponding to the environmental monitoring information examples in each sea area, a joint network debugging quality index is obtained.

[0205] In this way, by debugging deep learning networks on marine environmental monitoring information samples and related data, the performance and accuracy of navigation route planning decision networks can be effectively improved. This not only helps improve navigation safety and efficiency but also helps save energy, further promoting the development of the maritime industry.

[0206] In other examples, the navigation route planning decision network to be debugged includes a first feature mining branch, a second feature mining branch, and a route planning prediction branch. The step of inputting the marine environment monitoring information samples and the wave state trend data sample set into the navigation route planning decision network to obtain the navigation route planning prediction result, and inputting the labeled deep learning network debugging samples and the first deep learning network debugging samples into the navigation route planning decision network to obtain the labeled training result, includes: inputting the marine environment monitoring information samples into the first feature mining branch, inputting the wave state trend data sample set into the second feature mining branch, obtaining linked feature mining information based on the outputs of the first and second feature mining branches, inputting the linked feature mining information and the navigation route planning authentication result into the route planning prediction branch to obtain the navigation route planning prediction result; and inputting the labeled deep learning network debugging samples and the first deep learning network debugging samples into the second feature mining branch to obtain the labeled training result.

[0207] Taking an oil tanker on a long-distance voyage in the Indian Ocean as an example, the following steps are illustrated: First, in step 120, marine environmental monitoring information samples (such as seawater temperature and salinity), corresponding wave state trend data sample sets, labeled deep learning network debugging samples, and first deep learning network debugging samples are acquired. The navigation route planning decision network to be debugged includes a first feature mining branch, a second feature mining branch, and a route planning prediction branch.

[0208] Next, samples of marine environmental monitoring information are entered into the first feature mining branch, which may be a neural network module specifically designed for feature extraction of marine environmental information. Simultaneously, a set of sample data on ocean wave trends is entered into the second feature mining branch, which may also be a neural network module specifically designed for feature extraction of ocean wave trends.

[0209] Then, based on the outputs of the first and second feature mining branches, the linked feature mining information is obtained. This information combines key features of marine environmental information and wave state trend information.

[0210] Next, the linked feature mining information and the navigation route planning certification results are entered into the route planning prediction branch to obtain the navigation route planning prediction result. This result is a prediction of the tanker's navigation route based on the current marine environment and wave state trends, as well as historical navigation route planning experience.

[0211] Simultaneously, the labeled deep learning network debugging samples and the first deep learning network debugging samples are input into the second feature mining branch to obtain the labeled training results. This process trains the model to better understand and identify the features of the wave state trend data.

[0212] This design, by utilizing multiple feature mining branches to process different types of input data in parallel and employing a route planning prediction branch to generate the final navigation route planning prediction result, effectively improves the performance and accuracy of the navigation route planning decision network. This not only helps improve navigation safety and efficiency but also helps save energy, further promoting the development of the maritime industry.

[0213] Based on the above, a navigation route planning system is provided, including a processor and a memory that communicate with each other. The processor is used to retrieve a computer program from the memory and implement the above-described method by running the computer program.

[0214] Based on the above, a computer-readable storage medium is provided, on which a computer program is stored, the computer program implementing the above method when running.

[0215] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A ship route planning method based on sea state analysis, characterized by, The method is applied to a voyage route planning system, and comprises the following steps: Obtaining a sea area environment monitoring information sample, a sea wave state trend data sample set corresponding to the sea area environment monitoring information sample, and a voyage route planning authentication result, obtaining a first deep learning network debugging sample corresponding to the sea area environment monitoring information sample, a labeled debugging sample, and a labeled deep learning network debugging sample; the labeled debugging sample is a sample labeled after a second deep learning network debugging sample is labeled, the labeled deep learning network debugging sample is a remaining sample except the labeled debugging sample in the second deep learning network debugging sample, and the first deep learning network debugging sample and the second deep learning network debugging sample are determined from the sea wave state trend data sample set and the voyage route planning authentication result; Inputting the sea area environment monitoring information sample and the sea wave state trend data sample set into a voyage route planning decision network to be debugged to obtain a voyage route planning prediction result, and generating a target network debugging quality index according to a difference between the voyage route planning prediction result and the voyage route planning authentication result; Inputting the labeled deep learning network debugging sample and the first deep learning network debugging sample into the voyage route planning decision network to be debugged to obtain a labeled training result, and generating a joint network debugging quality index according to a difference between the labeled debugging sample and the labeled training result; Optimizing a neural network configuration weight of the voyage route planning decision network to be debugged according to the target network debugging quality index and the joint network debugging quality index until a training standard is met, and obtaining a target voyage route planning decision network; the target voyage route planning decision network is used for voyage route planning processing according to sea area environment monitoring information; The inputting the sea area environment monitoring information sample and the sea wave state trend data sample set into the voyage route planning decision network to be debugged to obtain the voyage route planning prediction result comprises the following steps: Inputting the sea area environment monitoring information sample, the sea wave state trend data sample set, and the voyage route planning authentication result into the voyage route planning decision network to be debugged; Respectively performing knowledge feature mining on the sea area environment monitoring information sample and the sea wave state trend data sample set to obtain a sea area environment monitoring knowledge vector sample and a sea wave state trend knowledge vector sample; Aggregating the sea area environment monitoring knowledge vector sample and the sea wave state trend knowledge vector sample to obtain a route planning cross-knowledge vector; According to the voyage route planning authentication result, the route planning cross-knowledge vector is predicted to obtain a voyage route planning prediction result; The learning example pool corresponding to the to-be-debugged navigation route planning decision network includes a plurality of sea area environment monitoring information samples respectively corresponding to deep learning network debugging samples, the deep learning network debugging sample includes a sea area environment monitoring information sample, a corresponding sea wave state trend data sample set, a navigation route planning authentication result, a first deep learning network debugging sample, a labeled debugging sample, a labeled deep learning network debugging sample, and a prior navigation energy consumption tolerance value, the learning example pool includes at least one prior navigation energy consumption tolerance value; The sea area environment monitoring information sample and the sea wave state trend data sample set are input into the to-be-debugged navigation route planning decision network to obtain a navigation route planning prediction result, and a target network debugging quality index is generated according to the difference between the navigation route planning prediction result and the navigation route planning authentication result, including: The sea area environment monitoring information sample and the corresponding sea wave state trend data sample set, the prior navigation energy consumption tolerance value in the learning example pool are input into the to-be-debugged navigation route planning decision network to obtain a navigation route planning prediction result paired with the prior navigation energy consumption tolerance value corresponding to the sea area environment monitoring information sample; A target local network debugging quality index is generated based on the difference between the navigation route planning prediction result and the navigation route planning authentication result corresponding to the same sea area environment monitoring information sample, and a target network debugging quality index is obtained based on the target local network debugging quality index corresponding to each sea area environment monitoring information sample; The labeled deep learning network debugging sample and the first deep learning network debugging sample are input into the to-be-debugged navigation route planning decision network to obtain a labeled training result, and a joint network debugging quality index is generated according to the difference between the labeled debugging sample and the labeled training result, including: The first deep learning network debugging sample, the labeled deep learning network debugging sample, and the prior navigation energy consumption tolerance value corresponding to the sea area environment monitoring information sample in the learning example pool are input into the to-be-debugged navigation route planning decision network to obtain a labeled training result paired with the prior navigation energy consumption tolerance value corresponding to the sea area environment monitoring information sample; A local joint network debugging quality index is generated based on the difference between the labeled debugging sample and the labeled training result corresponding to the same sea area environment monitoring information sample, and a joint network debugging quality index is obtained based on the local joint network debugging quality index corresponding to each sea area environment monitoring information sample.

2. The method of claim 1, wherein, The joint network debugging quality index includes a route update debugging quality index, and the first deep learning network debugging sample, the labeled debugging sample, and the labeled deep learning network debugging sample corresponding to the sea area environment monitoring information sample are obtained, including: The sea wave state trend data sample set is obtained as the first deep learning network debugging sample, and the navigation route planning authentication result is obtained as the second deep learning network debugging sample; The navigation route planning authentication result is labeled to obtain a route labeled debugging sample and an energy consumption labeled debugging sample, the route labeled debugging sample is obtained as the labeled debugging sample, and the energy consumption labeled debugging sample is obtained as the labeled deep learning network debugging sample; The marked deep learning network debugging sample and the first deep learning network debugging sample are input into a to-be-debugged navigation route planning decision network, a marked training result is obtained, a joint network debugging quality index is generated according to a difference between the marked debugging sample and the marked training result, and the joint network debugging quality index comprises: The sea wave state trend data sample set and the energy consumption marked debugging sample are input into a to-be-debugged navigation route planning decision network, a route marked training result is obtained, and a route update debugging quality index is generated according to a difference between the route marked debugging sample and the route marked training result. The joint network debugging quality index comprises a navigation risk prediction training error, the first deep learning network debugging sample, the marked debugging sample and the marked deep learning network debugging sample corresponding to the sea area environment monitoring information sample are obtained, and the first deep learning network debugging sample comprises:

3. The method of claim 1, wherein, The navigation route planning authentication result is obtained as the first deep learning network debugging sample, and the sea wave state trend data sample set is obtained as the second deep learning network debugging sample; The sea wave state trend data sample set is marked to obtain a navigation risk marked debugging sample and a risk trend debugging sample, the navigation risk marked debugging sample is obtained as the marked debugging sample, and the risk trend debugging sample is obtained as the marked deep learning network debugging sample; The marked deep learning network debugging sample and the first deep learning network debugging sample are input into a to-be-debugged navigation route planning decision network, a marked training result is obtained, a joint network debugging quality index is generated according to a difference between the marked debugging sample and the marked training result, and the joint network debugging quality index comprises: The navigation route planning authentication result and the risk trend debugging sample are input into a to-be-debugged navigation route planning decision network, a navigation risk training result is obtained, and a navigation risk prediction training error is generated according to a difference between the navigation risk marked debugging sample and the navigation risk training result. The sea area environment monitoring information sample and the sea wave state trend data sample set are respectively subjected to knowledge feature mining to obtain a sea area environment monitoring knowledge vector sample and a sea wave state trend knowledge vector sample, and the sea area environment monitoring knowledge vector sample comprises: An attribute of the sea area environment monitoring information sample is refined to obtain an original sea area environment attribute vector, and the original sea area environment attribute vector is subjected to knowledge mapping to obtain the sea area environment monitoring knowledge vector sample; 4. The method of claim 1, wherein, A state event of a sea wave state trend data sample in the sea wave state trend data sample set is identified to obtain a state event deduction vector corresponding to at least one target state event; Original sea wave state trend descriptions corresponding to the sea wave state trend data sample set are obtained according to the state event deduction vectors corresponding to each sea wave state trend data sample in the sea wave state trend data sample set; The original sea wave state trend descriptions are subjected to knowledge mapping to obtain the sea wave state trend knowledge vector sample; and The navigation route planning authentication result is obtained as the first deep learning network debugging sample, and the sea wave state trend data sample set is obtained as the second deep learning network debugging sample; The sea wave state trend data sample set is marked to obtain a navigation risk marked debugging sample and a risk trend debugging sample, the navigation risk marked debugging sample is obtained as the marked debugging sample, and the risk trend debugging sample is obtained as the marked deep learning network debugging sample; The marked deep learning network debugging sample and the first deep learning network debugging sample are input into a to-be-debugged navigation route planning decision network, a marked training result is obtained, a joint network debugging quality index is generated according to a difference between the marked debugging sample and the marked training result, and the joint network debugging quality index comprises: The navigation route planning authentication result and the risk trend debugging sample are input into a to-be-debugged navigation route planning decision network, a navigation risk training result is obtained, and a navigation risk prediction training error is generated according to a difference between the navigation risk marked debugging sample and the navigation risk training result. The sea area environment monitoring information sample and the sea wave state trend data sample set are respectively subjected to knowledge feature mining to obtain a sea area environment monitoring knowledge vector sample and a sea wave state trend knowledge vector sample, and the sea area environment monitoring knowledge vector sample comprises: An attribute of the sea area environment monitoring information sample is refined to obtain an original sea area environment attribute vector, and the original sea area environment attribute vector is subjected to knowledge mapping to obtain the sea area environment monitoring knowledge vector sample; A state event of a sea wave state trend data sample in the sea wave state trend data sample set is identified to obtain a state event deduction vector corresponding to at least one target state event; Original sea wave state trend descriptions corresponding to the sea wave state trend data sample set are obtained according to the state event deduction vectors corresponding to each sea wave state trend data sample in the sea wave state trend data sample set; The original sea wave state trend descriptions are subjected to knowledge mapping to obtain the sea wave state trend knowledge vector sample; and The original sea wave state trend description corresponding to the set of sea wave state trend data samples is obtained according to each state event corresponding to each sea wave state trend data sample in the set of sea wave state trend data samples, and includes: attribute extraction is respectively performed on trend keywords corresponding to each sea wave state trend data sample, state event keywords corresponding to each target state event, and distribution characteristics corresponding to each target state event, to obtain a trend keyword vector corresponding to each sea wave state trend data sample, a state event keyword vector corresponding to each target state event, and a relative distribution variable corresponding to each distribution characteristic; the original stage state trend description corresponding to a target state event is obtained based on the state event keyword vector, the relative distribution variable, the state event deduction vector corresponding to the same target state event, and the trend keyword vector corresponding to the sea wave state trend data sample; and the original sea wave state trend description is obtained based on the original stage state trend description corresponding to each target state event in each sea wave state trend data sample. The current input information is defined as the original sea area environment attribute vector or the original sea wave state trend description, knowledge mapping is performed on the current input information to obtain corresponding current sample information, which includes: a feature focusing operation is performed on the current input information to obtain current focused input information; the current input information and the current focused input information are aggregated to obtain initial linkage input information; hidden space vector mining is performed on the initial linkage input information to obtain a current hidden space vector; and the current hidden space vector and the initial linkage input information are aggregated to obtain target linkage input information. The current sample information is obtained according to the target linkage input information.

5. The method of claim 1, wherein, The route planning cross-knowledge vector is obtained by aggregating the sea area environment monitoring knowledge vector sample and the sea wave state trend knowledge vector sample, which includes: A heterogeneous feature focusing operation is performed on the sea area environment monitoring knowledge vector sample and the sea wave state trend knowledge vector sample to obtain linkage focused input information; A sea area state element optimization vector is obtained according to the sea area environment monitoring knowledge vector sample and the linkage focused input information; Knowledge screening is performed on the linkage focused input information according to the sea area state element optimization vector to obtain a flight route involvement knowledge vector; The flight route involvement knowledge vector and the sea area environment monitoring knowledge vector sample are combined to obtain the route planning cross-knowledge vector. The sea area state element optimization vector is obtained according to the sea area environment monitoring knowledge vector sample and the linkage focused input information, which includes: The sea area environment monitoring knowledge vector sample and the linkage focused input information are combined to obtain a first combined knowledge vector; Hidden space vector mining is performed on the first combined knowledge vector based on a first neural network configuration weight to obtain a first hidden space vector; The first hidden space vector is projected to obtain the sea area state element optimization vector.

6. The method of claim 1, wherein, The navigation route planning certification result includes a plurality of prior ship navigation route maps with a sequence; The navigation route planning certification result includes a plurality of prior ship navigation route maps with a sequence; A target planning area is determined from each route area corresponding to the navigation route planning certification result; A prior ship navigation route map before the target planning area is obtained from the navigation route planning certification result as a reference ship navigation route map, and knowledge features of the reference ship navigation route map are mined to obtain a reference ship navigation route space-time vector; The line planning cross-knowledge vector and the reference ship navigation route space-time vector are used to obtain a to-be-processed ship navigation route space-time vector; The to-be-processed ship navigation route space-time vector is predicted to obtain a to-be-processed ship navigation route map corresponding to the target planning area; The next route area is obtained as the target planning area, and the step of obtaining a prior ship navigation route map before the target planning area from the navigation route planning certification result as a reference ship navigation route map is executed until a termination condition is triggered, and a plurality of to-be-processed ship navigation route maps are obtained; The navigation route planning prediction result is obtained based on each to-be-processed ship navigation route map; The line planning cross-knowledge vector and the reference ship navigation route space-time vector are used to obtain a to-be-processed ship navigation route space-time vector; The reference ship navigation route space-time vector is subjected to a feature focusing operation to obtain initial focalized input information, and an initial ship navigation route space-time vector is obtained based on the initial focalized input information and the reference ship navigation route space-time vector; The initial ship navigation route space-time vector and the line planning cross-knowledge vector are subjected to a linkage feature focusing operation to obtain interactive focalized input information, and a transition ship navigation route space-time vector is obtained based on the interactive focalized input information and the initial ship navigation route space-time vector; The transition ship navigation route space-time vector is subjected to hidden space vector mining to obtain a target ship navigation route space-time vector, and the to-be-processed ship navigation route space-time vector is obtained based on the transition ship navigation route space-time vector and the target ship navigation route space-time vector.

7. The method of claim 1, wherein, The to-be-debugged navigation route planning decision network includes a first feature mining branch, a second feature mining branch, and a route planning prediction branch; The sea area environment monitoring information sample and the sea wave state trend data sample set are input into the to-be-debugged navigation route planning decision network to obtain a navigation route planning prediction result, and the labeled deep learning network debugging sample and the first deep learning network debugging sample are input into the to-be-debugged navigation route planning decision network to obtain a labeled training result, including: The sea area environment monitoring information sample is input into the first feature mining branch, the sea wave state trend data sample set is input into the second feature mining branch, linkage feature mining information is obtained according to outputs of the first feature mining branch and the second feature mining branch, the linkage feature mining information and the navigation route planning authentication result are input into a route planning prediction branch, and the navigation route planning prediction result is obtained; The labeled deep learning network debugging sample and the first deep learning network debugging sample are input into the second feature mining branch, and the labeled training result is obtained.

8. A voyage planning system characterized by, The navigation route planning system includes processors and memories in communication with each other, the processors are used to call computer programs from the memories, and the computer programs are used to realize the method in any one of claims 1-7.

Images