A capsizing prediction and warning method and system based on ship attitude perception

Abstract

The present invention discloses a capsizing prediction and warning method and system based on ship attitude perception, and the method includes data collection, data preprocessing, capsizing risk prediction, warning strategy generation and capsizing prediction and warning. The present invention relates to the field of ship capsizing prediction and warning technology, and specifically refers to a capsizing prediction and warning method and system based on ship attitude perception. The present scheme adopts a capsizing prediction and warning method generated by pre-classification prediction of capsizing risk combined with warning strategy. By first pre-classifying capsizing risk, a comprehensive evaluation of multiple factors and rapid risk judgment are obtained, and a comprehensive capsizing warning strategy is obtained by subsequent weighted prediction combined with multiple factors, thereby improving the overall reliability and data depth of the method; a causal-inspired graph neural network combined with dual-channel deep convolution feature extraction is used to pre-classify capsizing risk; and a bidirectional long short-term memory neural network combined with an integrated decision tree is used to generate a warning strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of ship capsizing prediction and early warning, and in particular to a capsizing prediction and early warning method and system based on ship attitude perception. Background Art

[0002] The capsizing prediction and warning method based on ship attitude perception is a technical system that monitors and analyzes the dynamic attitude data of the ship (such as roll, pitch and sway) in real time, and uses advanced algorithms and models (such as machine learning and neural networks) to predict whether the ship may capsize in the future and issue an early warning. Its main function is to issue an early warning signal in advance when the ship is in a shipping state, and provide timely decision-making support for ship operators, thereby effectively enhancing the safety of the ship, reducing the occurrence of capsizing accidents, and ensuring the safety of people and cargo on board.

[0003] However, among the existing capsizing prediction and warning methods, there is a traditional method that often adopts a simple binary classification prediction of a single model for the implementation of capsizing warning. However, in the actual navigation process, the factors affecting the capsizing of the ship are very complex. The correlation between the capsizing risk prediction and the capsizing warning strategy is often ignored in the traditional integrated method, and the simple capsizing probability prediction is difficult to meet the needs of the actual navigation process. Therefore, it is necessary to design a capsizing risk prediction and warning method that can consider the current posture characteristics of the ship and combine the external environment.

[0004] Among the existing capsizing risk prediction methods, there is a technical problem that the existing capsizing risk prediction basically focuses on the data source of the ship itself, but this results in the predicted risk probability being unable to fully reflect the environmental impact during the actual navigation process, and the detailed analysis and prediction of various factors are very low in interpretability, making it difficult to serve as an effective reference;

[0005] In the existing early warning strategy generation, there is a technical problem that the traditional method of generating early warning strategies is often divorced from the specific environmental factors and ship posture factors, and simply corresponds numerical values ​​and plans according to fixed rules. This inevitably reduces the reference value of the early warning strategy and fails to conduct a safety analysis of the ship's navigation process, thereby reducing the safety of the automatic system. Summary of the invention

[0006] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a capsizing prediction and warning method and system based on ship posture perception. In the existing capsizing prediction and warning methods, there is a problem that the traditional method often adopts a simple two-classification prediction of a single model for the implementation of capsizing warning. However, in the actual navigation process, the factors affecting the capsizing of the ship are very complex. The correlation between the capsizing risk prediction and the capsizing warning strategy is often ignored in the traditional integrated method, and the simple capsizing probability prediction is difficult to meet the needs of the actual navigation process. Therefore, it is necessary to design a capsizing risk prediction and warning system that can not only consider the current posture characteristics of the ship, but also combine the external environment. In order to solve the technical problems of the capsizing risk prediction method, this scheme creatively adopts a capsizing prediction and warning method generated by the pre-classification prediction of capsizing risk combined with the warning strategy. By first pre-classifying the capsizing risk, a comprehensive evaluation of multiple factors and a quick risk judgment are obtained, and then a comprehensive capsizing warning strategy is obtained by combining the weighted prediction of multiple factors, which improves the overall reliability and data depth of the method. In the existing capsizing risk prediction methods, there is a problem that the existing capsizing risk prediction basically focuses on the data source of the ship itself, but this results in the predicted risk probability being unable to fully reflect the environmental impact during the actual navigation process, and the detailed analysis and prediction of various factors are very low in interpretability, making it difficult to Taking the technical problem as an effective reference, this solution creatively adopts a causal-inspired graph neural network combined with dual-channel deep convolution feature extraction to pre-classify the capsizing risk. By setting up two channels of convolution feature extraction, the ship and external features are extracted respectively. After the feature fusion, the causal association analysis is realized through the causal-inspired graph neural network, thereby realizing a more detailed in-depth analysis of the causes of capsizing risk, which not only broadens the depth of capsizing analysis but also improves the reference and interpretability of the results, providing more effective support for subsequent steps; in the existing early warning strategy generation, there is a problem that the traditional method of generating early warning strategies is often divorced from the specific environmental factors and ship posture factors. This inevitably reduces the referenceability of the early warning strategy and fails to conduct a safety analysis of the ship's navigation process, thereby reducing the technical problem of the safety of the automatic system. This solution creatively uses a bidirectional long short-term memory neural network combined with an integrated decision tree to generate early warning strategies. By combining the weight value in the pre-classification result, the trigger comprehensive threshold is first judged to filter out the waste of computing resources. After the trigger threshold is comprehensively set, the strategy is matched and generated for the risk prediction results, which improves the direct availability and referenceability of the early warning strategy and improves the overall maintainability and executability of the method.

[0007] The technical solution adopted by the present invention is as follows: The present invention provides a capsizing prediction and early warning method based on ship attitude perception, the method comprising the following steps:

[0008] Step S1: data collection;

[0009] Step S2: data preprocessing;

[0010] Step S3: capsizing risk prediction;

[0011] Step S4: generating early warning strategies;

[0012] Step S5: capsize prediction and warning.

[0013] Furthermore, in step S1, the data collection is used to collect raw data of ship attitude, specifically, to obtain a raw data set of ship attitude perception through multimodal sensor data acquisition and external data collection; the raw data set of ship attitude perception specifically includes ship attitude data, ship environment data and ocean status data.

[0014] Further, in step S2, the data preprocessing is used to optimize the original data of the ship attitude, specifically to perform basic preprocessing and outlier processing operations on the original data set of ship attitude perception to obtain the optimized data set of ship attitude perception, including the following steps:

[0015] Step S21: outlier detection, specifically, using a method of combining manual data annotation with a standard support vector machine model to perform outlier detection on the original data set of ship attitude perception, identify and eliminate abnormal data that does not conform to the law of ship motion, and obtain an outlier elimination data set;

[0016] Step S22: data completion, specifically using a mean data filling method to fill in the missing data in the outlier elimination data set to obtain a data-filled data set;

[0017] Step S23: data calibration, specifically, using an artificial data detection method to perform data calibration on the padded data in the data-padded data set to obtain a calibration data set;

[0018] Step S24: spatiotemporal data association, specifically, performing time synchronization and association processing on the data in the calibration data set to obtain a time series optimized data set;

[0019] Step S25: data normalization, specifically, normalizing the data in the time series optimization data set by using a Z-score normalization method to obtain a normalized optimization data set;

[0020] Step S26: comprehensive preprocessing, specifically, performing comprehensive data preprocessing through the outlier detection, the data completion, the data calibration, the spatiotemporal data association and the data normalization to obtain a ship attitude perception optimization data set;

[0021] The ship attitude perception optimization data set specifically includes ship attitude optimization data, ship environment optimization data and ocean state optimization data.

[0022] Further, in step S3, the capsizing risk prediction is used for pre-classifying the capsizing risk, specifically, based on the ship attitude perception optimization data set, a causal heuristic graph neural network combined with dual-channel deep convolution feature extraction is used to pre-classify the capsizing risk to obtain the capsizing risk prediction data, specifically including the following steps:

[0023] Step S31: dual-channel deep convolutional network design, specifically, constructing a dual-input channel deep convolutional network, wherein the dual input channels include a posture motion channel and an environmental variable channel, and dual-channel convolution feature extraction is performed through dual-channel feature extraction and feature fusion;

[0024] The attitude motion channel is used to process the ship attitude optimization data;

[0025] The environmental variable channel is used to process the ship environment optimization data and the ocean state optimization data;

[0026] Step S32: dual-channel convolution feature extraction, specifically constructing a deep convolution layer, and performing dual-channel convolution feature extraction through the posture motion channel and the environmental variable channel, and applying feature fusion to obtain fused convolution features;

[0027] Step S33: feature dimensionality reduction processing, specifically constructing a maximum pooling layer and a fully connected layer, performing feature dimensionality reduction and feature mapping, and obtaining mapping feature vector data, which is used as an input data sample of the causal heuristic graph neural network;

[0028] Step S34: Causal relationship heuristic construction, specifically, constructing ship graph structure data according to the mapped feature vector data, and performing causal association deduction through a causal verification algorithm to obtain ship causal graph structure data, and defining a causal relationship matrix C according to the ship causal graph structure data, wherein each element in the causal relationship matrix C is used to represent the causal association strength in the ship causal graph structure data;

[0029] The ship graph structure data specifically includes node data and edge data, wherein the node data of the ship graph structure data is used to represent the posture characteristics of the ship, and the edge data of the ship graph structure data is used to represent the causal relationship in the time series;

[0030] Step S35: constructing a graph neural network, specifically, constructing a standard graph neural network based on the causal relationship matrix and the ship graph structure data to obtain graph convolution feature data;

[0031] Step S36: capsizing risk prediction, specifically, using the last layer of graph convolution feature data of the graph neural network as summary data features, and using a softmax function to perform capsizing risk pre-classification prediction to obtain capsizing risk prediction data;

[0032] The capsizing risk prediction data specifically includes capsizing risk probability reference data and attitude external correlation prediction reference data;

[0033] The capsizing risk probability reference data specifically include a capsizing probability quantified value and a capsizing pre-classification level;

[0034] The attitude external correlation prediction reference data is used to indicate the influence degree of the causal variable on the capsizing risk, specifically including the attitude influence weight and the external environment influence weight;

[0035] The attitude influence weight specifically includes the roll attitude weight , yaw attitude weight and pitch attitude weight ;

[0036] The external environment influence weight specifically includes the wind speed weight , Wave height weight and current velocity weights .

[0037] Further, in step S4, the warning strategy is generated to generate a warning strategy based on the result of the capsizing risk pre-classification, specifically, based on the ship attitude perception optimization data set and the capsizing risk prediction data, a bidirectional long short-term memory neural network combined with an integrated decision tree is used to generate a warning strategy to obtain capsizing warning strategy reference data, specifically including the following steps:

[0038] Step S41: constructing a bidirectional long short-term memory network, specifically constructing a standard bidirectional long short-term memory neural network, performing time series feature processing according to the ship attitude perception optimization data set, and obtaining a combined time series hidden state output;

[0039] Step S42: constructing an integrated decision tree, specifically, taking the merged temporal hidden state output as the input data of the integrated decision tree, adopting the standard extreme gradient boosting model as the basic prediction model, obtaining the contribution reference weight for early warning strategy generation through information gain calculation, and generating the contribution reference weight according to the early warning strategy, performing standard extreme gradient boosting model combination integration to obtain the integrated decision tree model, and generating the early warning strategy through the integrated decision tree model;

[0040] Step S43: combining models, specifically combining the bidirectional long short-term memory network and the integrated decision tree, and setting risk thresholds according to the capsizing risk prediction data, and calculating capsizing risk triggering comprehensive thresholds according to the risk threshold settings, and generating specific early warning strategies according to the capsizing risk triggering comprehensive thresholds to obtain early warning classification strategies;

[0041] Step S44: training the early warning strategy generation model, through the capsizing risk prediction, the construction of the bidirectional long short-term memory network and the construction of the integrated decision tree, the early warning strategy generation model training is performed to obtain the early warning strategy generation model Model SG ;

[0042] Step S45: generating an early warning strategy, specifically using the early warning strategy to generate a model Model SG , obtaining capsizing warning strategy reference data based on the ship attitude perception optimization data set;

[0043] The capsizing warning strategy reference data specifically includes a capsizing risk triggering comprehensive threshold, capsizing risk prediction data and a warning classification strategy.

[0044] Further, in step S5, the capsizing prediction and warning is used to perform a capsizing prediction and warning based on the capsizing risk pre-classification result and the warning strategy, specifically, the capsizing prediction and warning based on the ship posture perception is performed according to the capsizing warning strategy reference data to obtain the capsizing prediction and warning feedback data;

[0045] The capsizing prediction and warning feedback data specifically include a low-risk strategy, a medium-risk strategy and a high-risk strategy.

[0046] The present invention provides a capsizing prediction and warning system based on ship attitude perception, comprising a data collection module, a data preprocessing module, a capsizing risk prediction module, a warning strategy generation module and a capsizing prediction and warning module;

[0047] The data collection module is used for data collection, and obtains a raw data set of ship attitude perception through data collection, and sends the raw data set of ship attitude perception to the data preprocessing module;

[0048] The data preprocessing module is used for data preprocessing, and obtains a ship attitude perception optimization data set through data preprocessing, and sends the ship attitude perception optimization data set to the capsizing risk prediction module and the early warning strategy generation module;

[0049] The capsizing risk prediction module is used for capsizing risk prediction, obtains capsizing risk prediction data through capsizing risk prediction, and sends the capsizing risk prediction data to the early warning strategy generation module;

[0050] The warning strategy generation module is used for generating warning strategies, obtaining capsizing warning strategy reference data through the warning strategy generation, and sending the capsizing warning strategy reference data to the capsizing prediction and warning module;

[0051] The capsize prediction and warning module is used for capsize prediction and warning, and obtains capsize prediction and warning feedback data through capsize prediction and warning.

[0052] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0053] (1) In the existing capsizing prediction and warning methods, the traditional methods often use a simple binary classification prediction of a single model for the implementation of capsizing warning. However, in the actual navigation process, the factors affecting the capsizing of the ship are very complex. The correlation between the capsizing risk prediction and the capsizing warning strategy is often ignored in the traditional integrated method. The simple capsizing probability prediction is also difficult to meet the needs of the actual navigation process. Therefore, it is necessary to design a capsizing risk prediction and warning method that can not only consider the current posture characteristics of the ship but also combine the external environment. This scheme creatively adopts a capsizing prediction and warning method generated by the capsizing risk pre-classification prediction combined with the warning strategy. By first pre-classifying the capsizing risk, a comprehensive evaluation of multiple factors and a rapid risk judgment are obtained. Then, a comprehensive capsizing warning strategy is obtained by combining the weighted prediction of multiple factors, which improves the overall reliability and data depth of the method.

[0054] (2) In view of the technical problem that the existing capsizing risk prediction methods basically focus on the data source of the ship itself, the predicted risk probability cannot fully reflect the environmental impact during the actual navigation process, and the detailed analysis and prediction of various factors are very low in interpretability, making it difficult to serve as an effective reference, this solution creatively adopts a causal-inspired graph neural network combined with dual-channel deep convolution feature extraction to pre-classify the capsizing risk. By setting up two channels of convolution feature extraction, the ship and external features are extracted respectively, and after the feature fusion, the causal association analysis is realized through the causal-inspired graph neural network, thereby achieving a more detailed in-depth analysis of the causes of the capsizing risk, which not only broadens the depth of the capsizing analysis but also improves the reference and interpretability of the results, providing more effective support for subsequent steps;

[0055] (3) In view of the technical problem that in the generation of existing warning strategies, the traditional method of generating warning strategies is often divorced from the specific environmental factors and ship posture factors, and a simple correspondence between numerical values ​​and plans is performed according to fixed rules, which inevitably reduces the referenceability of the warning strategy and fails to conduct a safety analysis of the ship's navigation process, thereby reducing the safety of the automatic system. This scheme creatively adopts a bidirectional long short-term memory neural network combined with an integrated decision tree to generate warning strategies. By combining the weight value in the pre-classification result, a comprehensive trigger threshold judgment is first performed to filter out the waste of computing resources. After the trigger threshold is comprehensively set, the strategy is matched and generated according to the risk prediction results, which improves the direct availability and referenceability of the warning strategy and improves the overall maintainability and executability of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic flow chart of a capsizing prediction and early warning method based on ship attitude perception provided by the present invention;

[0057] Figure 2 A schematic diagram of a capsizing prediction and warning system based on ship attitude perception provided by the present invention;

[0058] Figure 3 This is a schematic diagram of the process of data preprocessing in step S2;

[0059] Figure 4 This is a schematic diagram of the process of capsizing risk prediction in step S3;

[0060] Figure 5 Schematic diagram of the process flow generated for the early warning strategy in step S4.

[0061] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0062] The technical solutions 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, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0063] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0064] Example 1, see Figure 1 The present invention provides a capsizing prediction and early warning method based on ship attitude perception, the method comprising the following steps:

[0065] Step S1: data collection;

[0066] Step S2: data preprocessing;

[0067] Step S3: capsizing risk prediction;

[0068] Step S4: generating early warning strategies;

[0069] Step S5: capsize prediction and warning.

[0070] By executing the above operations, in the existing capsizing prediction and warning methods, there is a problem that traditional methods often use a simple binary classification prediction of a single model for the implementation of capsizing warning. However, in the actual navigation process, the factors affecting the capsizing of the ship are very complex. The correlation between the capsizing risk prediction and the capsizing warning strategy is often ignored in the traditional integrated method, and the simple capsizing probability prediction is also difficult to meet the needs of the actual navigation process. Therefore, it is necessary to design a capsizing risk prediction and warning method that can consider the current posture characteristics of the ship and the external environment. The technical problem is that this scheme creatively adopts a capsizing prediction and warning method generated by the pre-classification prediction of the capsizing risk combined with the warning strategy. By first pre-classifying the capsizing risk, a comprehensive evaluation of multiple factors and a rapid risk judgment are obtained, and a comprehensive capsizing warning strategy is obtained through subsequent weighted prediction combining multiple factors, thereby improving the overall reliability and data depth of the method.

[0071] Example 2, see Figure 1 and Figure 2 In step S1, the data collection is used to collect raw data of ship attitude, specifically, to obtain a raw data set of ship attitude perception through multimodal sensor data collection and external data collection;

[0072] The multimodal sensor specifically includes an inertial measurement unit, an air pressure sensor and a positioning sensor;

[0073] The inertial measurement unit is used to detect the free angle motion of the ship, and specifically includes an accelerometer and a gyroscope;

[0074] The accelerometer is used to monitor the linear motion of the vessel and capture the rolling, acceleration and deceleration attitude changes;

[0075] The gyroscope is used to measure the angular velocity of the ship and determine the rotation and rolling attitude changes of the ship;

[0076] The free angle motion specifically includes pitch motion, roll motion, yaw motion and three-dimensional angular velocity motion data;

[0077] The air pressure sensor is used to assist in detecting changes in ocean air pressure and obtain wave height data in combination with ship height change information;

[0078] The positioning sensor is used to obtain position information and perform spatial analysis of motion trajectory;

[0079] The ship attitude perception raw data set specifically includes ship attitude data, ship environment data and ocean state data;

[0080] The ship attitude data specifically includes acceleration data, angular velocity data and attitude angle data;

[0081] The acceleration data is used to indicate the acceleration changes of the ship in the horizontal, vertical and depth directions. The specific data unit is meter per second squared (m / s 2 );

[0082] The angular velocity data is used to indicate the change in the rotation rate of the ship in the horizontal, vertical and depth directions, and the specific data unit is degree per second (° / s);

[0083] The ship environment data specifically includes air pressure data, positioning data and external environment data;

[0084] The air pressure data is used to indicate changes in ocean pressure and waves, and the specific data unit is Pascal;

[0085] The positioning data is used to indicate the geographical location of the vessel, specifically including longitude, latitude and altitude;

[0086] The external environment data specifically includes wind speed and wind direction data;

[0087] The ocean state data, specifically obtained through external data processing, includes wave height data and tide data.

[0088] Example 3, see Figure 1 , Figure 2 and Figure 3This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to optimize the original data of the ship attitude, specifically to perform basic preprocessing and outlier processing operations on the original data set of ship attitude perception to obtain the optimized data set of ship attitude perception, including the following steps:

[0089] Step S21: outlier detection, specifically, using a method of combining manual data annotation with a standard support vector machine model to perform outlier detection on the original data set of ship attitude perception, identify and eliminate abnormal data that does not conform to the law of ship motion, and obtain an outlier elimination data set;

[0090] Step S22: data completion, specifically using a mean data filling method to fill in the missing data in the outlier elimination data set to obtain a data-filled data set;

[0091] Step S23: data calibration, specifically, using an artificial data detection method to perform data calibration on the padded data in the data-padded data set to obtain a calibration data set;

[0092] Step S24: spatiotemporal data association, specifically, performing time synchronization and association processing on the data in the calibration data set to obtain a time series optimized data set;

[0093] Step S25: data normalization, specifically, normalizing the data in the time series optimization data set by using a Z-score normalization method to obtain a normalized optimization data set;

[0094] Step S26: comprehensive preprocessing, specifically, performing comprehensive data preprocessing through the outlier detection, the data completion, the data calibration, the spatiotemporal data association and the data normalization to obtain a ship attitude perception optimization data set;

[0095] The ship attitude perception optimization data set specifically includes ship attitude optimization data, ship environment optimization data and ocean state optimization data.

[0096] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the capsizing risk prediction is used to pre-classify the capsizing risk. Specifically, based on the ship attitude perception optimization data set, a causal heuristic graph neural network combined with dual-channel deep convolution feature extraction is used to pre-classify the capsizing risk to obtain the capsizing risk prediction data. Specifically, the following steps are included:

[0097] Step S31: dual-channel deep convolutional network design, specifically, constructing a dual-input channel deep convolutional network, wherein the dual input channels include a posture motion channel and an environmental variable channel, and dual-channel convolution feature extraction is performed through dual-channel feature extraction and feature fusion;

[0098] The attitude motion channel is used to process the ship attitude optimization data;

[0099] The environmental variable channel is used to process the ship environment optimization data and the ocean state optimization data;

[0100] Step S32: dual-channel convolution feature extraction, specifically constructing a deep convolution layer, and performing dual-channel convolution feature extraction through the posture motion channel and the environmental variable channel, and applying feature fusion to obtain fused convolution features;

[0101] The calculation formula for the dual-channel convolution feature extraction is:

[0102] ;

[0103] In the formula, is the characteristic data of the attitude motion channel, which is used to represent the characteristics of the ship attitude data extracted from the attitude motion channel. ReLU(·) is a nonlinear activation function. Z is the pose motion channel convolution weight, is the input data segment of the gesture motion channel, i is the time index, k is the data segment index, i:i+k-1 is the data segment from the time index i as the timing start point to the time index i+k-1 as the timing end point, b Z is the attitude motion channel bias term, is the characteristic data of the environmental variable channel, which is used to represent the characteristics of the ship environment and ocean state data extracted from the environmental variable channel. E is the environment variable channel convolution weight, is the environment variable input data segment, b E is the environmental variable channel bias term;

[0104] The calculation formula of the feature fusion is:

[0105] ;

[0106] In the formula, H i is the fused convolutional feature, where i is the time index, is the characteristic data of the attitude motion channel, is the environmental variable channel characteristic data, It is a feature fusion operation operator, which is specifically used to represent feature connection operations;

[0107] Step S33: feature dimensionality reduction processing, specifically constructing a maximum pooling layer and a fully connected layer, performing feature dimensionality reduction and feature mapping, and obtaining mapping feature vector data, which is used as an input data sample of the causal heuristic graph neural network;

[0108] Step S34: Causal relationship heuristic construction, specifically, constructing ship graph structure data according to the mapped feature vector data, and performing causal association deduction through a causal verification algorithm to obtain ship causal graph structure data, and defining a causal relationship matrix C according to the ship causal graph structure data, wherein each element in the causal relationship matrix C is used to represent the causal association strength in the ship causal graph structure data;

[0109] The causality test algorithm specifically refers to the application of the Granger causality test algorithm, and the calculation formula is:

[0110] ;

[0111] In the formula, v a is the ship causal graph structure data, v b is the ship cause graph structure data, the ship cause graph structure data and the ship cause graph structure data are used for causal verification, the causal verification specifically refers to calculating and judging the ship cause graph structure data v b Can the ship causal graph structure data v a has a significant impact, so as to make a causal relationship judgment, t is the time index, is a causal test constant term, used to represent the ship causal graph structure data v a The initial value of P is the ship causal graph structure data v a The total length of the autoregressive lag time is used to represent the predicted ship causal graph structure data v a The total amount of historical time required, p is the ship causal graph structure data v a The autoregressive lag time index of , is the ship causal graph structure data v a The autoregressive coefficient at the past tp time is used to reflect the ship causal graph structure data v a The impact of its own historical value on its current value, Q is the ship cause graph structure data v b The total length of the autoregressive lag time is used to represent the predicted ship genesis graph structure data v b The total amount of historical time required, q is the ship genesis graph structure data v b The autoregressive lag time index of , is the ship genesis graph structure data v b The autoregressive coefficient at the past time tp is used to represent the ship genesis graph structure data v bFor the ship causal graph structure data v a The influence of If it is significantly different from 0, it means that v b and v a There is a causal relationship between is the error term, which is used to represent random noise;

[0112] The ship graph structure data specifically includes node data and edge data, wherein the node data of the ship graph structure data is used to represent the posture characteristics of the ship, and the edge data of the ship graph structure data is used to represent the causal relationship in the time series;

[0113] Step S35: constructing a graph neural network, specifically, constructing a standard graph neural network based on the causal relationship matrix and the ship graph structure data to obtain graph convolution feature data, and the calculation formula is:

[0114] ;

[0115] In the formula, It is the first The l+1th layer graph convolution feature data of nodes, ReLU(·) is the nonlinear activation function, and N is the number of nodes. The set of neighboring nodes of Is a node The neighbor node index of is a causal relationship matrix, where the horizontal elements of the causal relationship matrix are used to represent the ship causal graph structure data, and the vertical elements of the causal relationship matrix are used to represent the ship cause graph structure data, W l is the neighbor node transfer weight, It is the first The node The l-th layer graph convolution feature data of neighboring nodes, is the initial weight of the node itself;

[0116] Step S36: capsizing risk prediction, specifically, using the last layer of graph convolution feature data of the graph neural network as summary data features, and using a softmax function to perform capsizing risk pre-classification prediction to obtain capsizing risk prediction data;

[0117] The capsizing risk prediction data specifically includes capsizing risk probability reference data and attitude external correlation prediction reference data;

[0118] The capsizing risk probability reference data specifically include a capsizing probability quantified value and a capsizing pre-classification level;

[0119] The capsizing probability quantized value specifically refers to the quantized probability representation of the capsizing probability, and the value range is [0,1];

[0120] The capsizing pre-classification level is specifically obtained through risk level classification;

[0121] Table 1 Schematic diagram of the quantitative probability range corresponding to the capsizing pre-classification level

[0122]

[0123] Table 1 is a schematic table of quantified probability ranges corresponding to capsizing pre-classification levels. As shown in the table, the capsizing pre-classification levels specifically include high risk, medium risk and low risk; the capsizing probability quantified range corresponding to the high risk is [0.75, 1], the capsizing probability quantified range corresponding to the medium risk is [0.35, 0.75), and the capsizing probability quantified range corresponding to the low risk is [0, 0.35);

[0124] The attitude external correlation prediction reference data is used to indicate the influence degree of the causal variable on the capsizing risk, specifically including the attitude influence weight and the external environment influence weight;

[0125] The attitude influence weight specifically includes the roll attitude weight , yaw attitude weight and pitch attitude weight ;

[0126] The external environment influence weight specifically includes the wind speed weight , Wave height weight and current velocity weights .

[0127] By performing the above operations, in view of the technical problem that in the existing capsizing risk prediction methods, the existing capsizing risk prediction basically focuses on the data source of the ship itself, but this leads to the predicted risk probability cannot fully reflect the environmental impact during the actual navigation process, and the detailed analysis and prediction of various factors are very low in interpretability, making it difficult to serve as an effective reference, this scheme creatively adopts a causal-inspired graph neural network combined with dual-channel deep convolutional feature extraction to pre-classify the capsizing risk, and extracts the ship and external features respectively by setting the convolutional feature extraction of two channels, and after the feature fusion, the causal correlation analysis is realized through the causal-inspired graph neural network, thereby achieving a more detailed in-depth analysis of the causes of the capsizing risk, which not only broadens the depth of the capsizing analysis but also improves the reference and interpretability of the results, providing more effective support for subsequent steps.

[0128] Example 5, see Figure 1 , Figure 2 and Figure 5This embodiment is based on the above embodiment. In step S4, the warning strategy is generated to generate a warning strategy based on the result of capsizing risk pre-classification. Specifically, based on the ship attitude perception optimization data set and the capsizing risk prediction data, a bidirectional long short-term memory neural network combined with an integrated decision tree is used to generate a warning strategy to obtain capsizing warning strategy reference data. Specifically, the following steps are included:

[0129] Step S41: constructing a bidirectional long short-term memory network, specifically constructing a standard bidirectional long short-term memory neural network, performing time series feature processing according to the ship attitude perception optimization data set, and obtaining a combined time series hidden state output;

[0130] The calculation formula for constructing the standard bidirectional long short-term memory neural network is:

[0131] ;

[0132] In the formula, is the forward hidden state output, is the backward hidden state output, t is the time index, LSTM(·) is the standard long short-term memory neural network function, X t is the reference data for the capsizing warning strategy, h t is the merged temporal hidden state output, [·;·] is the vector merge operator;

[0133] Step S42: constructing an integrated decision tree, specifically, taking the merged temporal hidden state output as the input data of the integrated decision tree, adopting the standard extreme gradient boosting model as the basic prediction model, obtaining the contribution reference weight for early warning strategy generation through information gain calculation, and generating the contribution reference weight according to the early warning strategy, performing standard extreme gradient boosting model combination integration to obtain the integrated decision tree model, and generating the early warning strategy through the integrated decision tree model;

[0134] The calculation formula for the information gain calculation is:

[0135] I(X)=H(Y)-H(Y|X);

[0136] Wherein, I(·) is the information gain calculation function, which is used to represent the overall influence of the input feature on the capsizing warning strategy and serves as the contribution reference weight for the warning strategy generation, X is the input feature, which is used to represent the combined temporal hidden state output, H(·) is the information entropy calculation function, Y is the target variable, which is used to represent the capsizing warning strategy, and H(Y|X) is the conditional entropy calculation function, which is used to represent the information entropy of the target variable Y for a given input feature X;

[0137] Step S43: combining models, specifically combining the bidirectional long short-term memory network and the integrated decision tree, and setting risk thresholds according to the capsizing risk prediction data, and calculating capsizing risk triggering comprehensive thresholds according to the risk threshold settings, and generating specific early warning strategies according to the capsizing risk triggering comprehensive thresholds to obtain early warning classification strategies;

[0138] The calculation formula for calculating the capsizing risk triggering comprehensive threshold is:

[0139] ;

[0140] Wherein, TS is the capsizing risk triggering comprehensive threshold, which is used to indicate the discrimination threshold for generating an early warning strategy. When the capsizing probability quantized value in the capsizing risk probability reference data is greater than the capsizing risk triggering comprehensive threshold, an early warning strategy is generated and a capsizing warning is performed;

[0141] is the roll attitude weight, is the roll angle at the current time t t Exceeding the set heel angle threshold The probability of , which is used to indicate the possibility of the ship tilting left or right to the risk angle;

[0142] is the yaw attitude weight, is the yaw angle yaw at the current time t t Exceeds the set yaw angle threshold The probability of , which is used to indicate the likelihood of the ship reaching the risk angle when turning;

[0143] is the pitch attitude weight, is the pitch angle at the current time t t Exceeding the set pitch angle threshold The probability of , which is used to indicate the possibility of the ship swaying forward and backward to reach the risk critical angle;

[0144] is the wind speed weight, is the wind speed at the current time t windspeed t Exceeding the set wind speed threshold The probability of , which is used to indicate the possibility that the external wind speed of the ship reaches the risk level;

[0145] is the wave height weight, is the wave height at the current time t t Exceeding the set wave height threshold The probability of the external wave height of the ship reaching the risk level;

[0146] is the ocean current velocity weight, Current time t ocean current speed oceanflow t Exceeding the set current speed threshold The probability of , which is used to indicate the possibility that the external current speed of the ship reaches the risk level;

[0147] The calculation formula for generating the early warning strategy is:

[0148] ;

[0149] Where Str(t) is the early warning classification strategy. The low-risk strategy specifically includes simple posture adjustment suggestions, environmental monitoring enhancement and safety reminders. The medium-risk strategy specifically includes immediate posture adjustment suggestions, active risk avoidance suggestions, intermediate safety alarms and equipment inspection suggestions. The high-risk strategy specifically includes emergency posture correction suggestions, immediate risk avoidance measures suggestions, ship-wide safety alarms, emergency equipment activation, crew evacuation preparations and external rescue request suggestions.

[0150] Step S44: training the early warning strategy generation model, through the capsizing risk prediction, the construction of the bidirectional long short-term memory network and the construction of the integrated decision tree, the early warning strategy generation model training is performed to obtain the early warning strategy generation model Model SG ;

[0151] Step S45: generating an early warning strategy, specifically using the early warning strategy to generate a model Model SG , obtaining capsizing warning strategy reference data based on the ship attitude perception optimization data set;

[0152] The capsizing warning strategy reference data specifically includes a capsizing risk triggering comprehensive threshold, capsizing risk prediction data and a warning classification strategy.

[0153] By executing the above operations, in view of the technical problem that in the generation of existing warning strategies, the traditional method of generating warning strategies is often separated from the specific environmental factors and ship posture factors, and the numerical values ​​and schemes are simply corresponded according to fixed rules, which inevitably reduces the referenceability of the warning strategy and fails to conduct a safety analysis of the ship's navigation process, thereby reducing the safety of the automatic system, this scheme creatively adopts a bidirectional long short-term memory neural network combined with an integrated decision tree to generate warning strategies. By combining the weight value in the pre-classification result, the comprehensive threshold judgment is first triggered to filter out the waste of computing resources, and after the trigger threshold is comprehensively set, the strategy is matched and generated according to the risk prediction results, which improves the direct availability and referenceability of the warning strategy and improves the overall maintainability and executability of the method.

[0154] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the capsizing prediction and warning is used to perform a capsizing prediction and warning based on the capsizing risk pre-classification result and the warning strategy, specifically, the capsizing prediction and warning based on the ship posture perception is performed according to the capsizing warning strategy reference data to obtain the capsizing prediction and warning feedback data;

[0155] The capsizing prediction and early warning feedback data specifically includes a low-risk strategy, a medium-risk strategy, and a high-risk strategy;

[0156] The low-risk strategies specifically include simple posture adjustment suggestions, enhanced environmental monitoring, and safety reminders;

[0157] The simple attitude adjustment suggestions specifically include slight adjustment of heading, slight adjustment of speed and suggestions for fine-tuning the current sailing attitude;

[0158] The aforementioned strengthening of environmental monitoring specifically refers to the recommendation to strengthen the monitoring of the current external environment;

[0159] The safety reminder refers to sending low-level safety notifications through the ship's communication system and reminding the crew to be alert;

[0160] The medium-risk strategies specifically include immediate posture adjustment suggestions, active risk avoidance suggestions, medium-level safety alerts, and equipment inspection suggestions;

[0161] The immediate attitude adjustment suggestion specifically includes immediately changing the navigation significantly, immediately reducing the speed and immediately changing the navigation route;

[0162] The active risk avoidance advice refers specifically to advising the crew to take actions to avoid dangerous areas;

[0163] The intermediate safety alarm refers to sending an intermediate-level alarm through the ship's communication system and requiring the crew to enter a risk preparedness state;

[0164] The equipment inspection recommendations specifically refer to inspecting key ship equipment and ensuring that the equipment is functioning properly;

[0165] The high-risk strategies specifically include emergency posture correction suggestions, immediate risk avoidance measures suggestions, ship-wide safety alarms, emergency equipment activation, crew evacuation preparations and external rescue request suggestions;

[0166] The emergency attitude correction suggestions specifically include forced heading change instructions, emergency stop operations and emergency reverse operations;

[0167] The immediate avoidance measures recommendation specifically refers to strongly recommending the crew to immediately change the sailing route and leave the dangerous environmental conditions as soon as possible;

[0168] The ship-wide safety alarm specifically refers to sending the highest level of safety alarm through the ship's communication system and requiring the crew to enter an emergency state;

[0169] The emergency equipment activation specifically refers to the activation of emergency equipment on board the ship;

[0170] The crew evacuation preparation specifically refers to sending evacuation preparation instructions through the ship's communication system;

[0171] The external rescue request suggestion specifically refers to sending a distress signal instruction to the outside through the ship communication system.

[0172] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and a capsizing prediction and warning system based on ship posture perception provided by the present invention includes a data collection module, a data preprocessing module, a capsizing risk prediction module, a warning strategy generation module and a capsizing prediction and warning module;

[0173] The data collection module is used for data collection, and obtains a raw data set of ship attitude perception through data collection, and sends the raw data set of ship attitude perception to the data preprocessing module;

[0174] The data preprocessing module is used for data preprocessing, and obtains a ship attitude perception optimization data set through data preprocessing, and sends the ship attitude perception optimization data set to the capsizing risk prediction module and the early warning strategy generation module;

[0175] The capsizing risk prediction module is used for capsizing risk prediction, obtains capsizing risk prediction data through capsizing risk prediction, and sends the capsizing risk prediction data to the early warning strategy generation module;

[0176] The warning strategy generation module is used for generating warning strategies, obtaining capsizing warning strategy reference data through the warning strategy generation, and sending the capsizing warning strategy reference data to the capsizing prediction and warning module;

[0177] The capsize prediction and warning module is used for capsize prediction and warning, and obtains capsize prediction and warning feedback data through capsize prediction and warning.

[0178] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0179] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0180] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A capsizing prediction and warning method based on ship attitude perception, characterized in that: The method comprises the following steps: Step S1: Data collection, obtaining the original data set of ship attitude perception, specifically including ship attitude data, ship environment data and ocean state data; Step S2: data preprocessing, specifically performing basic preprocessing and outlier processing operations on the original data set of ship attitude perception to obtain a ship attitude perception optimized data set, specifically including ship attitude optimized data, ship environment optimized data and ocean state optimized data; Step S3: capsizing risk prediction, specifically using a causal heuristic graph neural network combined with dual-channel deep convolution feature extraction to perform capsizing risk pre-classification to obtain capsizing risk prediction data, specifically including the following steps: Step S31: dual-channel deep convolutional network design, specifically, constructing a dual-input channel deep convolutional network, wherein the dual input channels include a posture motion channel and an environmental variable channel, and dual-channel convolution feature extraction is performed through dual-channel feature extraction and feature fusion; Step S32: dual-channel convolution feature extraction, specifically constructing a deep convolution layer, and performing dual-channel convolution feature extraction through the posture motion channel and the environmental variable channel, and applying feature fusion to obtain fused convolution features; Step S33: feature dimensionality reduction processing, specifically constructing a maximum pooling layer and a fully connected layer, performing feature dimensionality reduction and feature mapping, and obtaining mapping feature vector data, which is used as an input data sample of the causal heuristic graph neural network; Step S34: Causal relationship heuristic construction, specifically, constructing ship graph structure data according to the mapped feature vector data, and performing causal association deduction through a causal verification algorithm to obtain ship causal graph structure data, and defining a causal relationship matrix C according to the ship causal graph structure data, wherein each element in the causal relationship matrix C is used to represent the causal association strength in the ship causal graph structure data; The ship graph structure data specifically includes node data and edge data, wherein the node data of the ship graph structure data is used to represent the posture characteristics of the ship, and the edge data of the ship graph structure data is used to represent the causal relationship in the time series; Step S35: constructing a graph neural network, specifically, constructing a standard graph neural network based on the causal relationship matrix and the ship graph structure data to obtain graph convolution feature data; Step S36: capsizing risk prediction, specifically, using the last layer of graph convolution feature data of the graph neural network as summary data features and using a softmax function to perform capsizing risk pre-classification prediction to obtain capsizing risk prediction data; Step S4: generating a warning strategy, which is used to generate a warning strategy based on the result of the capsizing risk pre-classification, specifically, based on the ship attitude perception optimization data set and the capsizing risk prediction data, using a bidirectional long short-term memory neural network combined with an integrated decision tree to generate a warning strategy, and obtain capsizing warning strategy reference data, specifically including the following steps: Step S41: constructing a bidirectional long short-term memory network, specifically constructing a standard bidirectional long short-term memory neural network, performing time series feature processing according to the ship attitude perception optimization data set, and obtaining a combined time series hidden state output; Step S42: constructing an integrated decision tree, specifically, taking the merged temporal hidden state output as the input data of the integrated decision tree, adopting the standard extreme gradient boosting model as the basic prediction model, obtaining the contribution reference weight for early warning strategy generation through information gain calculation, and generating the contribution reference weight according to the early warning strategy, performing standard extreme gradient boosting model combination integration to obtain the integrated decision tree model, and generating the early warning strategy through the integrated decision tree model; Step S43: combining models, specifically combining the bidirectional long short-term memory network and the integrated decision tree, and setting risk thresholds according to the capsizing risk prediction data, and calculating capsizing risk triggering comprehensive thresholds according to the risk threshold settings, and generating specific early warning strategies according to the capsizing risk triggering comprehensive thresholds to obtain early warning classification strategies; Step S44: training the early warning strategy generation model, through the capsizing risk prediction, the construction of the bidirectional long short-term memory network and the construction of the integrated decision tree, the early warning strategy generation model training is performed to obtain the early warning strategy generation model Model SG ; Step S45: generating an early warning strategy, specifically using the early warning strategy to generate a model Model SG , obtaining capsizing warning strategy reference data based on the ship attitude perception optimization data set; Step S5: Capsizing prediction and warning, obtaining capsizing prediction and warning feedback data.

2. A capsizing prediction and early warning method based on ship attitude perception according to claim 1, characterized in that: In step S36, the capsizing risk prediction data specifically includes capsizing risk probability reference data and attitude external correlation prediction reference data; The capsizing risk probability reference data specifically include a capsizing probability quantified value and a capsizing pre-classification level; The attitude external correlation prediction reference data is used to indicate the influence degree of the causal variable on the capsizing risk, specifically including the attitude influence weight and the external environment influence weight; The attitude influence weights specifically include a roll attitude weight αroll, a yaw attitude weight αyaw and a pitch attitude weight αpitch; The external environment impact weights specifically include a wind speed weight αwindspeed, a wave height weight αheight, and an ocean current speed weight αoceanflow.

3. The method for capsizing prediction and early warning based on ship attitude perception according to claim 1, characterized in that: In step S45, the capsizing warning strategy reference data specifically includes a capsizing risk triggering comprehensive threshold, capsizing risk prediction data and a warning classification strategy.

4. The method for capsizing prediction and early warning based on ship attitude perception according to claim 1, characterized in that: In step S5, the capsizing prediction and warning is used to combine the capsizing risk pre-classification result and the warning strategy to perform a capsizing prediction and comprehensive warning, specifically, to perform a capsizing prediction and warning based on ship posture perception according to the capsizing warning strategy reference data to obtain capsizing prediction and warning feedback data; The capsizing prediction and warning feedback data specifically include a low-risk strategy, a medium-risk strategy and a high-risk strategy.

5. The method for capsizing prediction and early warning based on ship attitude perception according to claim 4 is characterized in that: In step S2, the data preprocessing includes the following steps: step S21: outlier detection; step S22: data completion; step S23: data calibration; step S24: spatiotemporal data association; step S25: data normalization; step S26: comprehensive preprocessing.

6. A capsizing prediction and warning system based on ship attitude perception, used to implement a capsizing prediction and warning method based on ship attitude perception as claimed in any one of claims 1 to 5, characterized in that: It includes a data collection module, a data preprocessing module, a capsizing risk prediction module, a warning strategy generation module and a capsizing prediction and warning module.

7. The capsizing prediction and warning system based on ship attitude perception according to claim 6 is characterized by: The data collection module is used for data collection, and obtains a raw data set of ship attitude perception through data collection, and sends the raw data set of ship attitude perception to the data preprocessing module; The data preprocessing module is used for data preprocessing, and obtains a ship attitude perception optimization data set through data preprocessing, and sends the ship attitude perception optimization data set to the capsizing risk prediction module and the early warning strategy generation module; The capsizing risk prediction module is used for capsizing risk prediction, obtains capsizing risk prediction data through capsizing risk prediction, and sends the capsizing risk prediction data to the early warning strategy generation module; The warning strategy generation module is used for generating warning strategies, obtaining capsizing warning strategy reference data through the warning strategy generation, and sending the capsizing warning strategy reference data to the capsizing prediction and warning module; The capsize prediction and warning module is used for capsize prediction and warning, and obtains capsize prediction and warning feedback data through capsize prediction and warning.

Images