Flight route planning method and system based on deep learning large-scale extreme wind speed forecast

Abstract

The invention discloses a ship route planning method and system based on deep learning large-scale extreme wind speed forecast, and relates to the technical field of artificial intelligence and maritime navigation. An incremental principal component analysis method is introduced, a large-scale sea surface extreme wind speed forecasting model based on depth separable convolution is constructed, and potential physical sudden change disturbance information is decoded and reserved in a variable high-dimensional mapping space by the model through integration of an attention mechanism module and a residual learning algorithm; the attention mechanism module maps high-dimensional feature information in an extreme wind field time sequence to a multi-dimensional feature matrix of a coder-decoder. The method can perform refined dynamic forecasting on the large-scale sea surface extreme wind speed, accurately detect the extreme wind speed sea area and the boundary thereof, forecast and master the intermittent sudden change rule of the sea extreme wind speed in advance, help a ship driver to dynamically adjust the ship route in real time and avoid severe storm weather, and further ensure the navigation safety.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and maritime navigation technology, and in particular to a route planning method and system based on deep learning large-scale extreme wind speed forecasting. Background Art

[0002] Maritime transportation has been recognized as one of the indispensable modes of transportation for the development of the global logistics supply chain network. In recent years, with the rapid development of global trade and the rapid expansion of logistics networks, the world economic development has sharply increased the demand for reliable and efficient shipping. However, when it comes to the planning and design of transoceanic routes, maritime transport ships often face potential navigation risks. The instability and unpredictability of the meteorological environment will lead to great uncertainty in the marine environment. The intermittent changes and disturbances of the wind speed at sea will cause serious safety hazards to ship transportation, and then cause major accidents. For example, strong winds will make it difficult for ships to navigate, or even lose control, increasing the risk of collision, reefing or sinking. Extreme sea conditions will have a significant impact on the navigation safety of ships and the voyage decision-making process. Although ship operating companies can obtain reference route designs from meteorological navigation companies, real-time weather forecast information is becoming increasingly important for ships at sea. Accurate and timely onboard weather forecast data can support ship operators to design and determine the ship's path in advance. Ship drivers need to obtain reliable weather forecast information in real time so that they can take route adjustment measures such as deceleration and change of course in a timely manner, thereby further ensuring the safety of crew members and ships. Therefore, providing accurate and effective extreme weather forecasts is crucial to achieving efficient, safe and green ship path planning.

[0003] Traditional ocean and meteorological forecasts rely on numerical weather prediction models, which use collected initial conditions such as meteorological parameters and geographic boundaries and predict weather changes based on physical conservation equations. However, the inherent instability and randomness of the evolution of the Earth system make the prediction of the global weather system extremely challenging. In particular, with the increasing complexity and uncertainty of the Earth system caused by global climate change in recent decades, traditional numerical prediction models often fail to capture the complex spatiotemporal mutation disturbances in the Earth's evolutionary system. In addition, the computational cost of numerical methods based on physical models is very expensive, so these complex numerical models pose significant challenges in development and maintenance. The different spatiotemporal resolutions of meteorological elements also have a great impact on the prediction accuracy, such as grid and time interval resolution. For traditional numerical prediction models, increasing the grid resolution will lead to longer processing time and higher computational requirements. Most weather forecasts and meteorological observation systems can only provide sparse low-resolution data samples. In addition, due to the limitations of remote sensing observation technology, global large-scale observation systems often have missing wind speed observations. In addition, most of the current research methods on ship route design optimization use weather forecast information with very low spatiotemporal resolution. Weather forecast data with low spatiotemporal resolution has been proven to lead to large route design errors. Weather forecast models based on deep learning show great potential in revealing potential climate patterns from historical observations, and can provide high-resolution and accurate weather forecast data for ship route design by relying on powerful data mining and missing information completion capabilities, opening up new prospects for efficient and high-precision ship path planning. Researchers have been trying to explore different types of machine learning methods to obtain accurate forecasts of extreme wind speeds at sea. However, the complex nonlinear mutation characteristics of large-scale spatiotemporal weather systems pose a huge challenge to traditional linear regression theory and shallow machine learning techniques. Summary of the invention

[0004] In view of this, the present invention provides a ship route planning method and system based on deep learning extreme wind field forecast, introduces incremental principal component analysis, and constructs a large-scale sea surface extreme wind speed forecast model based on deep separable convolution; wherein the deep separable convolution module can improve the prediction efficiency and performance, while reducing the model's computational complexity and hardware memory requirements. Compared with traditional convolution, deep separable convolution greatly reduces the number of model parameters and computational requirements. This has strong practicality and feasibility for realizing shipboard weather forecasts when ships are sailing at sea. Among them, the incremental principal component analysis method can efficiently filter the evolution characteristics of extreme wind speeds and improve forecast accuracy by reducing data dimensions and redundant noise effects. In addition, the knowledge transferability and generalization performance of the forecast model are verified through extreme wind speed field forecasts in different geographical regions. Therefore, there is a gap in research technology in the design of heading planning based on high-resolution extreme wind speed forecasts. The present invention can make refined dynamic forecasts of large-scale two-dimensional sea surface extreme wind speeds, accurately detect extreme wind speed sea areas and their boundaries, forecast in advance and grasp the intermittent mutation laws of extreme wind speeds at sea, and help ship drivers to dynamically adjust ship routes in real time to avoid severe wind and wave weather, thereby ensuring navigation safety.

[0005] To this end, the present invention provides the following technical solutions:

[0006] On the one hand, the present invention discloses a route planning method based on deep learning large-scale extreme wind field forecast, comprising:

[0007] S1. Collect high-precision sea surface extreme wind speed field information in two different regions;

[0008] S2, normalize the wind field data, calculate the historical time series dependence and autocorrelation coefficient of the extreme wind speed two-dimensional field, obtain the weight factor of the wind field historical time series, and determine the tensor size of the model input data according to the weight factor and the time series dependence order of the wind field; the tensor size includes the time series order, length, width and depth, and the time series dependence is used to determine the first size of the tensor, the time series order;

[0009] S3. Introduce the incremental principal component analysis method to construct a large-scale sea surface extreme wind speed forecast model based on deep separable convolution; the incremental principal component analysis method performs dimension reduction and denoising preprocessing on the two-dimensional field input data of the model, retains the main components of the wind field element changes to reconstruct and calculate the effective wind field data; the model takes the time series dependency order of the reconstructed extreme wind field as input and outputs the extreme wind field forecast result; the model decodes the potential physical mutation disturbance information and retains it in the high-dimensional mapping space of the variable by integrating the attention mechanism module and the residual learning algorithm; the attention mechanism module maps the high-dimensional feature information in the extreme wind field time series to the multi-dimensional feature matrix of the encoder-decoder;

[0010] S4, training the model, and using the trained model to forecast large-scale offshore extreme wind speed field information at two different geographical locations;

[0011] S5. Detect areas with severe sea conditions based on extreme wind forecast information and dynamically adjust the planned route.

[0012] Furthermore, high-precision sea surface extreme wind speed field information is collected in two different regions, including:

[0013] The extreme wind speed field information in two different geographical locations was selected and collected, and the wind field forecast range in the selected area was summarized as the target variable for the potential spatial variation distribution of the extreme wind field.

[0014] Furthermore, the wind field data are normalized using min-max normalization.

[0015] Furthermore, the historical time series dependence and autocorrelation coefficient of the extreme wind speed two-dimensional field are calculated to obtain the weight factor of the historical time series of the wind field. The tensor size of the model input data is determined according to the weight factor and the order of the time series dependence of the wind field, including:

[0016] Establish the extreme wind field time series correlation function and partial autocorrelation analysis function;

[0017] Calculate the mean sequence of extreme wind field time series information, and use the correlation analysis function to calculate the correlation weight between the historical extreme wind field factors and the future target wind field factors;

[0018] The temporal dependence order of extreme wind field historical data is determined based on the correlation coefficient.

[0019] Furthermore, the Huber loss function is used as the prediction loss function of the model to calculate the error between the predicted value and the target value:

[0020]

[0021] Among them, Ψ is the forecast data of extreme wind speed field at sea, and O represents the reanalysis data of extreme wind speed field.

[0022] Furthermore, the model is trained, and the trained model is used to forecast large-scale offshore extreme wind speed field information at two different geographical locations, including:

[0023] Monte Carlo cross-validation method was used to divide the data set;

[0024] The normalized offshore wind speed dataset is divided into two parts: training set and test set;

[0025] Randomly divide the data set into training and validation, conduct model learning and training, and implement random sub-sampling validation;

[0026] The Huber loss error is used as the standard to calculate the prediction performance of the model, and the average prediction performance of the model is used as the final reference;

[0027] The trained model is used to forecast extreme wind fields in two different geographical regions.

[0028] Furthermore, based on the extreme wind forecast information, the severe sea conditions are detected and the planned route is dynamically adjusted, including:

[0029] Visualize the spatiotemporal map of extreme wind field forecasts and dynamically adjust the course based on the extreme wind speed areas detected in the forecast.

[0030] In another aspect, the present invention further discloses a route planning system based on deep learning large-scale extreme wind field forecast, comprising:

[0031] The data integration module selects the extreme wind speed field distribution information of two sea areas with different geographical locations;

[0032] A data processing module normalizes wind farm data, calculates the historical time series dependency and autocorrelation coefficient of the extreme wind speed two-dimensional field, obtains the weight factor of the wind farm historical time series, and determines the tensor size of the model input data according to the weight factor and the time series dependency order of the wind farm; the tensor size includes the time series order, length, width and depth, and the time series dependency is used to determine the first size of the tensor, the time series order;

[0033] The model building module introduces the incremental principal component analysis method to build a large-scale sea surface extreme wind speed forecast model based on deep separable convolution; the incremental principal component analysis method performs dimension reduction and denoising preprocessing on the two-dimensional field input data of the model, retains the main components of the wind field element changes to reconstruct and calculate the effective wind field data; the model takes the time series dependency order of the reconstructed extreme wind field as input and outputs the extreme wind field forecast result; the model decodes the potential physical disturbance information and retains it in the high-dimensional mapping space of the variable by integrating the attention mechanism module and the residual learning algorithm; the attention mechanism module maps the high-dimensional feature information in the extreme wind field time series to the multi-dimensional feature matrix of the encoder-decoder;

[0034] A wind field forecasting module, which trains the model and uses the trained model to forecast large-scale extreme offshore wind field information at two different geographical locations;

[0035] The route planning module dynamically adjusts the ship's planned route based on the high-resolution extreme wind field information predicted by the model to avoid severe weather and sea conditions.

[0036] Advantages and positive effects of the present invention:

[0037] Most of the existing route planning and design rely on low-resolution meteorological navigation data. In addition, due to the limitation of observation technology, the marine meteorological observation system cannot provide real-time and complete ocean navigation meteorological navigation data observation. The present invention establishes a route planning and design method system based on deep learning high-resolution large-scale extreme wind speed forecast, dynamically forecasts the high-resolution offshore extreme wind speed information distribution in different geographical locations, forecasts in advance and grasps the intermittent mutation law of extreme wind speed, facilitates ship drivers to accurately detect extreme wind speed sea areas and their boundaries, forecasts in advance and grasps the intermittent mutation law of extreme wind speed at sea, helps ship drivers to dynamically adjust ship routes in real time, avoid severe wind and wave weather, and thus ensure navigation safety.

[0038] In the forecast model of the present invention, the incremental principal component analysis method introduced can efficiently filter the evolution characteristics of extreme wind speed and improve the forecast accuracy by reducing the data dimension and redundant noise effect. A large-scale sea surface extreme wind speed forecast model based on deep separable convolution is constructed. The model decodes the potential physical disturbance information and retains it in the high-dimensional mapping space of the variable by integrating the attention mechanism module and the residual learning algorithm; the attention mechanism module maps the high-dimensional feature information in the extreme wind field time series to the multi-dimensional feature matrix of the encoder and decoder; wherein the deep separable convolution module can improve the prediction efficiency and performance, while reducing the calculation amount and hardware memory requirements of the model. Compared with traditional convolution, deep separable convolution greatly reduces the number of model parameters and calculation requirements. This is highly practical and feasible for realizing shipboard weather forecasts when ships are sailing at sea. In addition, the knowledge transferability and generalization performance of the forecast model are verified through the forecast of extreme wind speed fields in different geographical regions. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 This is a flow chart of a route planning model based on large-scale extreme wind speed forecast in an embodiment of the present invention;

[0041] Figure 2 This is a correlation analysis diagram of historical wind farm time series information in an embodiment of the present invention;

[0042] Figure 3 is an effective extreme wind field spatial eigenvector based on empirical orthogonal decomposition in an embodiment of the present invention;

[0043] Figure 4Route planning based on extreme wind speed forecast according to an embodiment of the present invention;

[0044] Figure 5 This is a route planning based on extreme wind speed forecasts at different geographical locations according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0046] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0047] like Figure 1 As shown, a route forecasting method based on deep learning large-scale extreme wind field forecasting in an embodiment of the present invention includes the following steps:

[0048] S1. Collect high-precision sea surface extreme wind speed field information in two different regions and analyze the relatively optimal historical time lag characteristics of wind field time series samples.

[0049] In the specific implementation, S1 specifically includes:

[0050] S11. Using the reanalysis dataset of the European Centre for Medium- and Long-Term Forecasts, the western Pacific Ocean and the North Atlantic Ocean were selected as two research areas with different geographical locations. The extreme wind field data in the corresponding sea areas were collected as the target variable for the potential spatial variation distribution of the wind speed field.

[0051] S12. Use the min-max normalization operation to normalize the wind speed field time series tensor data to further eliminate the calculation errors caused by different variable dimensions and variation ranges.

[0052]

[0053] where variable min is the minimum value of the sample variable, variable max Represents the maximum value of the sample variable.

[0054] S2. Use autocorrelation analysis to mine the time dependence of the overall wind field sequence lag, calculate the time dependence autocorrelation coefficient of the extreme wind field historical data, and determine the forecast model input. Among them, the model input refers to determining the time order (tensor size) of the wind field tensor. The tensor size includes the time order, length, width and depth. The time dependence is used to determine the first size of the tensor, the time order.

[0055] In specific implementation, S2 includes the following steps:

[0056] S21, establishing time series correlation function and partial autocorrelation function;

[0057] S22, calculating the mean sequence of extreme wind field time series information, and using the correlation analysis function to calculate the correlation weight between the historical extreme wind field factor and the future target wind field factor;

[0058] The calculation formula of the correlation coefficient between the extreme wind field historical time series is as follows:

[0059]

[0060] Among them, X, Y are the wind field time series at different times, σ X σ Y is the corresponding variance, μ Y μ X is the corresponding mean.

[0061] S23. Determine the time series dependence order of the extreme wind field historical data according to the correlation coefficient, and determine the forecast model input.

[0062] S3. Introduce the incremental principal component analysis method to construct a large-scale sea surface extreme wind speed forecast model based on deep separable convolution.

[0063] The model takes the highly positively correlated extreme wind field time series dependency order as input and outputs the extreme wind field forecast result. The model decodes the potential physical disturbance information and retains it in the high-dimensional mapping space of the variable by integrating the attention mechanism module and the residual learning algorithm; the attention mechanism module maps the high-dimensional feature information in the extreme wind field time series to the multi-dimensional feature matrix of the encoder-decoder; the deep separable convolution module can improve the prediction efficiency and performance, while reducing the model's computational workload and hardware memory requirements. Compared with traditional convolution, the deep separable convolution greatly reduces the number of model parameters and computational requirements. This is highly practical and feasible for realizing shipboard weather forecasting when ships are sailing at sea. Among them, the incremental principal component analysis method can efficiently filter the evolution characteristics of extreme wind speed and improve the forecast accuracy by reducing the data dimension and redundant noise effect. In addition, the knowledge transferability and generalization performance of the forecast model are verified through the forecast of extreme wind speed fields in different geographical regions.

[0064] In specific implementation, S3 includes the following steps:

[0065] S31. Introduce a method based on incremental principal component analysis to clean and reconstruct extreme wind field data. The incremental principal component analysis method can effectively filter the evolution characteristics of extreme wind speed and improve the forecast accuracy by reducing data dimensions and redundant noise effects.

[0066] The reconstruction calculation formula based on the empirical orthogonal decomposition of the wind field tensor is as follows:

[0067]

[0068] Among them, PC is the principal component variable, and EOF is the corresponding spatial eigenvector.

[0069] S32. Use the deep separable 2D CNN layer to build a deep encoder-decoder. The main task of the encoder is to gradually reduce the resolution of the input wind field tensor and extract the high-level semantic information of the wind field. The main task of the decoder is to gradually reconstruct the spatial cross-correlation high-dimensional information of the wind field tensor. The encoder-decoder uses deep separable convolution, in which the deep separable convolution decomposes the standard convolution into depthwise convolution and pointwise convolution, thereby significantly reducing the computational complexity while maintaining performance. Then, an encoding and decoding model based on a deep separable convolutional neural network is constructed. The 2D convolution kernel in the model is used to mine the coupling interaction characteristics between extreme wind field tensor factors. The decoder module exports high-dimensional semantic information, and the encoder part extracts high-dimensional feature maps of tensor sequences with low-order nonlinear information. The deep separable convolution block introduced in this method can significantly reduce the model calculation and memory requirements. Compared with traditional convolution, the number of parameters and calculations required is greatly reduced. This makes them more efficient and faster to train and run on devices with limited computing resources.

[0070] S33. Introduce the residual learning mechanism into the wind farm forecast network model. The residual learning mechanism controls the feature extraction and fidelity fusion of the model information flow, and further extracts the potential physical disturbance information and retains it in the original high-dimensional mapping space of each variable.

[0071] f(x)′=f(x)+x(6)

[0072] The output information flow of each residual module is the sum of the original information flow and the output information flow after convolution mapping.

[0073] S34. Further introduce the attention mechanism module into the encoding and decoding model.

[0074] The number of features of the time series tensor of the input multimodal meteorological factors of the encoder-decoder model is C, which is the number of two-dimensional fields of meteorological factors. H and W are the height and width of the model input tensor. Represents the input tensor of the model. The attention mechanism module uses dot-product to further calculate the three matrices of the input tensor: the key K, the query Q and the value V. and The corresponding mapping weight feature matrix value.

[0075]

[0076]

[0077]

[0078] The dimensions of the K and Q feature matrices should be consistent. and the jth key eigenvalue The similarity between them is determined by the regularization function Calculate and obtain. Due to the similarity calculation between different convolutional network layers, such as different positions and The similarity calculation is different.

[0079] In the calculation of attention feature weights, it is usually necessary to introduce a specific constraint factor ker(x, y) in the above regularization function: To ensure the non-negative characteristics of the attention calculation operation. According to the characteristic indicator function introduced by the constraint factor:

[0080]

[0081] Among them, when the numerator in the formula is expressed in vector form:

[0082]

[0083] The linear attention mechanism in high-dimensional feature extraction calculations can reduce the consumption of computing resources and improve the model calculation efficiency.

[0084] S34. Use the Huber loss function as the loss function of the prediction model to calculate the error between the predicted value and the target value:

[0085]

[0086] Among them, Ψ is the wind field forecast data created by the present invention, and O represents the reanalysis data of the offshore wind speed field, that is, the "observation data".

[0087] S4. The Monte Carlo cross-validation (Shuffle Split) method is used to encode and decode the model for model training, and the trained model is used to forecast large-scale extreme wind speed field information and accurately detect areas with severe sea conditions.

[0088] In specific implementation, the model training process is as follows:

[0089] S41. Use the Monte Carlo cross validation (Shuffle Split) method to divide the data set;

[0090] S42, dividing the normalized offshore wind speed data set into two parts: a training set and a test set;

[0091] S43. Assume there are 100 data samples, 20% of which are used as validation sets, 60% of which are used as training sets, and the remaining 20% ​​(100-(60+20)) will not be used by the model. The data set is randomly divided into training and validation sets to perform model learning and training.

[0092] S44. Repeat step S43 several times to perform random sub-sampling verification.

[0093] S45. Using the Huber loss error as a standard, calculate the prediction performance of the model in step S44, and take the average prediction performance of the model as the final reference.

[0094] S46. Use the trained model to forecast extreme wind fields in two different geographical regions.

[0095] S47. Visualize the spatiotemporal map of extreme wind field forecasts, and dynamically adjust the course according to the extreme wind speed sea areas detected in the forecast to avoid severe weather and sea conditions.

[0096] In the above embodiment, the present invention establishes a route planning method system based on deep learning large-scale extreme wind speed forecast. By introducing the incremental principal component analysis method, a large-scale sea surface extreme wind speed forecast model based on deep separable convolution is constructed. The model decodes the potential physical disturbance information and retains it in the high-dimensional mapping space of the variable by integrating the attention mechanism module and the residual learning algorithm; the attention mechanism module maps the high-dimensional feature information in the extreme wind field time series to the multi-dimensional feature matrix of the encoder and decoder; wherein the deep separable convolution module can improve the prediction efficiency and performance, while reducing the calculation amount and hardware memory requirements of the model. Compared with the traditional convolution, the deep separable convolution greatly reduces the number of model parameters and the calculation requirements. This is highly practical and feasible for realizing shipboard weather forecasts when ships are sailing at sea. Among them, the incremental principal component analysis method can efficiently filter the evolution characteristics of extreme wind speeds and improve the forecast accuracy by reducing the data dimension and redundant noise effects. In addition, the knowledge transferability and generalization performance of the forecast model are verified through the forecast of extreme wind speed fields in different geographical regions. The present invention can make a refined dynamic forecast of large-scale two-dimensional sea surface extreme wind speeds, accurately detect extreme wind speed sea areas and their boundaries, forecast in advance and grasp the intermittent mutation law of extreme wind speeds at sea, and help ship drivers to dynamically adjust ship routes in real time to avoid severe wind and wave weather, thereby ensuring navigation safety. Since it does not require complex physical and mathematical formula calculations and system simulation, the intelligent calculation forecast model breaks through the calculation bottleneck problem of traditional weather forecasting methods, and is therefore very valuable for obtaining real-time shipborne meteorological navigation data.

[0097] Corresponding to the route planning method based on deep learning large-scale extreme wind speed forecast in the above embodiment, this embodiment further provides a route planning system based on deep learning large-scale extreme wind speed forecast, including:

[0098] The data acquisition module selects the extreme wind speed field distribution information of two sea areas with different geographical locations;

[0099] A data processing module normalizes wind farm data, calculates the historical time series dependency and autocorrelation coefficient of the extreme wind speed two-dimensional field, obtains the weight factor of the wind farm historical time series, and determines the tensor size of the model input data according to the weight factor and the time series dependency order of the wind farm; the tensor size includes the time series order, length, width and depth, and the time series dependency is used to determine the first size of the tensor, the time series order;

[0100] The model building module first introduces the incremental principal component analysis method to build a large-scale sea surface extreme wind speed forecast model based on deep separable convolution. The model decodes the potential physical disturbance information and retains it in the high-dimensional mapping space of the variable by integrating the attention mechanism module and the residual learning algorithm; the attention mechanism module maps the high-dimensional feature information in the extreme wind field time series to the multi-dimensional feature matrix of the encoder-decoder; the deep separable convolution module can improve the prediction efficiency and performance, while reducing the model's computational workload and hardware memory requirements. Compared with traditional convolution, deep separable convolution greatly reduces the number of model parameters and computational requirements. This is highly practical and feasible for realizing shipboard weather forecasts when ships are sailing at sea. The incremental principal component analysis method can effectively filter the evolution characteristics of extreme wind speeds and improve forecast accuracy by reducing data dimensions and redundant noise effects. In addition, the knowledge transferability and generalization performance of the forecast model are verified through the forecast of extreme wind speed fields in different geographical regions. The present invention can make refined dynamic forecasts of large-scale two-dimensional sea surface extreme wind speeds, accurately detect extreme wind speed sea areas and their boundaries, forecast in advance and grasp the intermittent mutation laws of extreme wind speeds at sea, and help ship drivers to dynamically adjust ship routes in real time to avoid severe wind and wave weather, thereby ensuring navigation safety.

[0101] The wind field forecasting module trains the model and uses the trained model to forecast large-scale extreme offshore wind field information at two different geographical locations.

[0102] The route planning module dynamically adjusts the ship's planned route based on the high-resolution extreme wind field information predicted by the model to avoid severe weather and sea conditions.

[0103] As for the system in the embodiment of the present invention, since it corresponds to the method in the above embodiment, the description is relatively simple. For the relevant similarities, please refer to the description of the method part in the above embodiment, which will not be described in detail here.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A route planning method based on deep learning large-scale extreme wind field forecast, characterized in that: include: S1. Collect high-precision sea surface extreme wind speed field information in two different regions; S2, normalize the wind field data, calculate the historical time series dependence and autocorrelation coefficient of the extreme wind speed two-dimensional field, obtain the weight factor of the wind field historical time series, and determine the tensor size of the model input data according to the weight factor and the time series dependence order of the wind field; the tensor size includes the time series order, length, width and depth, and the time series dependence is used to determine the first size of the tensor, the time series order; S3. Introduce the incremental principal component analysis method to construct a large-scale sea surface extreme wind speed forecast model based on deep separable convolution; the incremental principal component analysis method performs dimension reduction and denoising preprocessing on the two-dimensional field input data of the model, retains the main components of the wind field element changes to reconstruct and calculate the effective wind field data; the model takes the time series dependency order of the reconstructed extreme wind field as input and outputs the extreme wind field forecast result; the model decodes the potential physical mutation disturbance information and retains it in the high-dimensional mapping space of the variable by integrating the attention mechanism module and the residual learning algorithm; The attention mechanism module maps the high-dimensional feature information in the extreme wind field time series to the multi-dimensional feature matrix of the encoder-decoder; S4, training the model, and using the trained model to forecast large-scale offshore extreme wind speed field information at two different geographical locations; S5. Detect areas with severe sea conditions based on extreme wind forecast information and dynamically adjust the planned route.

2. The route planning method based on deep learning large-scale extreme wind field forecast according to claim 1 is characterized in that: Collect high-precision extreme sea surface wind speed field information in two different regions, including: The extreme wind speed field information in two different geographical locations was selected and collected, and the wind field forecast range in the selected area was summarized as the target variable for the potential spatial variation distribution of the extreme wind field.

3. The route planning method based on deep learning large-scale extreme wind field forecast according to claim 1 is characterized in that: The wind field data is normalized using min-max normalization.

4. A route planning method based on deep learning large-scale extreme wind field forecast according to claim 2 or 3, characterized in that: Calculate the historical time series dependence and autocorrelation coefficient of the extreme wind speed two-dimensional field, obtain the weight factor of the wind field historical time series, and determine the tensor size of the model input data according to the weight factor and the time series dependence order of the wind field, including: Establish the extreme wind field time series correlation function and partial autocorrelation analysis function; Calculate the mean sequence of extreme wind field time series information, and use the correlation analysis function to calculate the correlation weight between the historical extreme wind field factors and the future target wind field factors; The temporal dependence order of extreme wind field historical data is determined based on the correlation coefficient.

5. The route planning method based on deep learning large-scale extreme wind field forecast according to claim 1 is characterized in that: The Huber loss function is used as the prediction loss function of the model to calculate the error between the predicted value and the target value: Among them, Ψ is the forecast data of extreme wind speed field at sea, and O represents the reanalysis data of extreme wind speed field.

6. The route planning method based on deep learning large-scale extreme wind field forecast according to claim 5 is characterized in that: The model is trained, and the trained model is used to forecast large-scale offshore extreme wind speed field information at two different geographical locations, including: Monte Carlo cross-validation method was used to divide the data set; The normalized offshore wind speed dataset is divided into two parts: training set and test set; Randomly divide the data set into training and validation, conduct model learning and training, and implement random subsampling validation; The Huber loss error is used as the standard to calculate the prediction performance of the model, and the average prediction performance of the model is used as the final reference; The trained model is used to forecast extreme wind fields in two different geographical regions.

7. The route planning method based on deep learning large-scale extreme wind field forecast according to claim 6 is characterized in that: According to the extreme wind forecast information, detect the severe sea conditions and dynamically adjust the planned route, including: Visualize the spatiotemporal map of extreme wind field forecasts and dynamically adjust the course based on the extreme wind speed areas detected in the forecast.

8. A route planning system based on deep learning large-scale extreme wind field forecast, characterized in that: include: The data integration module selects the extreme wind speed field distribution information of two sea areas with different geographical locations; A data processing module normalizes wind farm data, calculates the historical time series dependency and autocorrelation coefficient of the extreme wind speed two-dimensional field, obtains the weight factor of the wind farm historical time series, and determines the tensor size of the model input data according to the weight factor and the time series dependency order of the wind farm; the tensor size includes the time series order, length, width and depth, and the time series dependency is used to determine the first size of the tensor, the time series order; The model building module introduces the incremental principal component analysis method to build a large-scale sea surface extreme wind speed forecast model based on deep separable convolution; the incremental principal component analysis method performs dimension reduction and denoising preprocessing on the two-dimensional field input data of the model, retains the main components of the wind field element changes to reconstruct and calculate the effective wind field data; the model takes the time series dependency order of the reconstructed extreme wind field as input and outputs the extreme wind field forecast result; the model integrates the attention mechanism module and the residual learning algorithm to decode the potential physical disturbance information and retain it in the high-dimensional mapping space of the variable; The attention mechanism module maps the high-dimensional feature information in the extreme wind field time series to the multi-dimensional feature matrix of the encoder-decoder; A wind field forecasting module, which trains the model and uses the trained model to forecast large-scale extreme offshore wind field information at two different geographical locations; The route planning module dynamically adjusts the ship's planned route based on the high-resolution extreme wind field information predicted by the model to avoid severe weather and sea conditions.

Images