Ship narrow water area navigation hidden danger identification method and system, storage medium and electronic equipment

Abstract

This invention discloses a method, system, storage medium, and electronic device for identifying navigation hazards in narrow waters. The method includes the following steps: acquiring sensor data when a ship enters narrow waters and performing data preprocessing; calculating sensor data confidence based on a pre-trained first prediction model; calculating a data inconsistency index based on a pre-trained second prediction model; calculating an obstacle presence index based on a pre-trained third prediction model; calculating a route safety index based on a pre-trained fourth prediction model; calculating a navigation hazard index based on the sensor data confidence, data inconsistency index, and / or obstacle presence index and / or route safety index; and identifying the current navigation hazard in the narrow waters based on the navigation hazard index. This invention solves the problem that existing methods for identifying navigation safety hazards in narrow waters are prone to cognitive errors due to unreliable, inconsistent, or incomplete data.

Description

Technical Field

[0001] This invention belongs to the field of intelligent ship technology, and in particular relates to a method, system, storage medium and electronic device for identifying potential hazards in narrow waterways. Background Technology

[0002] Narrow waterways (such as canals, fjords, port channels, and inland waterways) are among the highest-risk navigation areas. Here, in these confined spaces, the operational space for vessels is extremely limited, and the margin for error is very low. Navigation decisions heavily rely on real-time, accurate navigation data (radar, AIS, ECDIS, etc.). If this data is delayed, deviates, interfered with, or misread, a vessel may veer off course within seconds, leading to grounding, collisions with shorelines, collisions with other vessels, or damage to critical infrastructure such as bridges and pipelines.

[0003] Currently, the navigation safety technologies commonly used in the industry mainly include: (1) Independent sensor monitoring and alarm, such as CPA / TCPA collision hazard alarm and yaw alarm. (2) Basic data fusion display: AIS targets, radar-tracked targets and channel boundaries are overlaid on ECDIS or Integrated Bridge System (IBS). (3) Rule-based risk assessment: Based on the International Code for Preventing Collisions at Sea (COLREGs) and navigation experience, fixed safety distances, speed limits and other rules are formulated to make risk judgments. However, in complex narrow waterway environments, the reliability, consistency and integrity of sensor data face severe challenges, and systemic data misleading risks are very easy to occur, while existing technologies lack effective identification and mitigation mechanisms for this.

[0004] Current improvement solutions, such as multi-target tracking based on Kalman filtering, simple AIS-radar target association, and computer vision-based assisted lookout, have partially improved the above problems, but have the following fundamental limitations: (1) They mostly alarm after data anomalies or risks occur, lacking predictive identification of the risk of data misrepresentation itself. (2) They usually optimize for a single sensor or a specific risk (such as collision), without constructing a progressive evaluation framework from the systematic perspective of multi-source information fusion. (3) They rely on preset rules and fixed parameters, and cannot adapt to the changing geographical, traffic, and environmental conditions of narrow waterways.

[0005] To address the problem of cognitive errors caused by unreliable, inconsistent, and incomplete data when ships navigate in narrow waters, which makes it difficult to effectively identify navigation safety hazards, a method, system, storage medium, and electronic equipment for identifying navigation hazards in narrow waters are proposed. Summary of the Invention

[0006] This invention proposes a method, system, storage medium, and electronic device for identifying potential safety hazards when ships navigate in narrow waters, in order to at least solve the problem that existing ships have difficulty effectively identifying navigation safety hazards due to cognitive errors caused by unreliable, inconsistent, or incomplete data when navigating in narrow waters.

[0007] According to an embodiment of the present invention, a method for identifying potential navigation hazards in narrow waters is provided, comprising:

[0008] The system acquires sensor data when a vessel navigates through narrow waterways and performs data preprocessing; the sensor data includes radar data, GPS data, AIS data, vessel maneuvering data, and image data.

[0009] The confidence level of sensor data is calculated based on the pre-trained first prediction model; the first prediction model uses the changes in GPS data and / or AIS data and / or image data corresponding to different ship maneuvering data as input features;

[0010] The data inconsistency index is calculated based on the pre-trained second prediction model; the second prediction model uses the spatiotemporal deviation and / or data deviation of different sensor data of the ship as input features.

[0011] The presence index of the obstacle target is calculated based on the pre-trained third prediction model; the third prediction model uses historical radar data and / or the matching degree between radar data and AIS data and / or the dynamic changes in the image when the ship obstacle target appears as input features;

[0012] The route safety index is calculated based on the pre-trained fourth prediction model; the fourth prediction model uses waterway data and / or ship trajectory data and / or ship draft data in narrow waterways as input features;

[0013] The navigation hazard index is calculated based on the confidence level of sensor data, the data contradiction index, and / or the obstacle target existence index and / or the route safety index.

[0014] When the navigation hazard index exceeds the preset navigation hazard threshold, it is determined that there is a hazard in the current narrow waterway navigation.

[0015] In a preferred embodiment, the data preprocessing includes the following steps:

[0016] All collected sensor data are synchronized to a unified spatiotemporal reference and spatiotemporally aligned.

[0017] Calculate the data source quality assessment value based on GPS signal quality and / or radar performance status and / or AIS reception status;

[0018] Some sensor data is filtered out based on the data source quality assessment value at each time point.

[0019] In a preferred embodiment, calculating the confidence level of the sensor data based on the pre-trained first prediction model includes:

[0020] The GPS matching value is calculated based on the degree of matching between different ship maneuvering data and the corresponding changes in GPS data.

[0021] AIS matching values ​​are calculated based on the degree of matching between different ship maneuvering data and AIS data variations.

[0022] The image matching value is calculated based on the degree of matching between different ship maneuvering data and image data changes.

[0023] The first prediction model is constructed based on the positive correlation between GPS matching values ​​and / or AIS matching values ​​and / or image matching values ​​and sensor data confidence, and the sensor data confidence is calculated accordingly.

[0024] In a preferred embodiment, calculating the data inconsistency index based on the pre-trained second prediction model includes:

[0025] The spatiotemporal deviation value is calculated based on the degree of time deviation and / or position indication deviation of data from different sensors on the ship.

[0026] The data deviation value is calculated based on the degree of data deviation in the correlation of data from different sensors on the ship.

[0027] A second prediction model is constructed based on the positive correlation between the spatiotemporal deviation value and / or the data deviation value and the data inconsistency index, and the data inconsistency index is calculated accordingly.

[0028] In a preferred embodiment, calculating the obstacle target presence index based on the pre-trained third prediction model includes:

[0029] The continuity indication value of the obstacle target is calculated based on the continuity of historical radar data when the ship obstacle target appears;

[0030] The non-visual presence indication value of the obstacle target is calculated based on the degree of matching between radar data and AIS data when the ship obstacle target appears.

[0031] The visual presence indication value of the obstacle target is calculated based on the dynamic changes in the background of the image when the ship obstacle target appears.

[0032] A third prediction model is constructed based on the positive correlation between the obstacle target continuity indicator value and / or the obstacle target non-visible presence indicator value and / or the obstacle target visible presence indicator value and the obstacle target presence index, and the obstacle target presence index is calculated accordingly.

[0033] In a preferred embodiment, calculating the route safety index based on the pre-trained fourth prediction model includes:

[0034] Calculate the safety boundary of the narrow waterway based on the width and / or depth of the narrow waterway;

[0035] The ship trajectory safety indication value is calculated based on the distance and / or the trend of distance change between the ship trajectory data and the safety boundary of the narrow waterway;

[0036] The ship's draft safety indication value is calculated based on the distance and / or the trend of distance change between the ship's draft data and the safety boundary of the narrow waterway.

[0037] A fourth prediction model is constructed based on the positive correlation between the ship trajectory safety indication value and / or the ship draft safety indication value and the route safety index, and the route safety index is calculated accordingly.

[0038] In a preferred embodiment, the step of calculating the navigation hazard index based on the confidence level and data contradiction index of sensor data and / or the obstacle target presence index and / or the route safety index includes:

[0039] Calculate the index correction coefficient based on the degree of influence of the confidence level of sensor data on the data contradiction index and / or the obstacle target existence index and / or the route safety index;

[0040] The navigation anomaly index is calculated based on the degree of impact of data contradiction index and / or obstacle target existence index and / or route safety index on ship navigation safety.

[0041] The navigation hazard index is obtained by correcting the navigation anomaly index based on the index correction coefficient.

[0042] According to another embodiment of the present invention, a system for identifying potential navigation hazards in narrow waters is provided, comprising:

[0043] The ship data acquisition and preprocessing module is used to acquire sensor data and perform data preprocessing when a ship enters narrow waters.

[0044] The first prediction module is used to calculate the confidence level of sensor data based on a pre-trained first prediction model; the first prediction model uses the changes in GPS data and / or AIS data and / or image data corresponding to different ship maneuvering data as input features;

[0045] The second prediction module is used to calculate the data inconsistency index based on the pre-trained second prediction model; the second prediction model uses the spatiotemporal deviation and / or data deviation and / or data source quality of different sensor data of the ship as input features.

[0046] The third prediction module is used to calculate the presence index of the obstacle target based on the pre-trained third prediction model. The third prediction model uses historical radar data and / or the matching degree between radar data and AIS data and / or the dynamic changes in the image when the ship obstacle target appears as input features.

[0047] The fourth prediction module is used to calculate the route safety index based on the pre-trained fourth prediction model; the fourth prediction model uses waterway data and / or ship trajectory data and / or ship draft data in narrow waterways as input features.

[0048] Navigation Hazard Identification Module: This module is used to call the outputs of the first prediction module, the second prediction module, and / or the third prediction module and / or the fourth prediction module to calculate the navigation hazard index. When the navigation hazard index is greater than the preset navigation hazard threshold, it is determined that there is a hazard in the current narrow waterway.

[0049] According to another embodiment of the present invention, a computer-readable storage medium is provided that stores a computer program for electronic data exchange, wherein the computer program causes a computer to execute the above-described method for identifying potential navigation hazards in narrow waters.

[0050] According to another embodiment of the present invention, an electronic device is provided, comprising:

[0051] At least one processor;

[0052] and a memory communicatively connected to the at least one processor;

[0053] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the above-described method for identifying potential navigation hazards in narrow waters.

[0054] The advantages of the method, system, storage medium, and electronic device for identifying potential navigation hazards in narrow waters by ships according to the present invention are as follows:

[0055] (1) The present invention constructs a multi-layered progressive assessment framework of “data confidence - contradiction - target existence - route safety - hidden danger fusion”. Compared with the traditional isolated and single risk alarm scheme, it can effectively realize the identification of hidden dangers in the whole chain from raw data to navigation decision, thereby avoiding the risk of data misleading and comprehensively improving the structure and integrity of safety assessment.

[0056] (2) The present invention actively and quantitatively assesses the credibility of the data itself and the contradiction of multiple sources through the first and second prediction models. Compared with the traditional scheme that only alarms after obvious data anomalies or physical risks are triggered, it can effectively detect potential data misleading risks such as sensor performance degradation, signal interference and data conflict in advance, thereby achieving early warning of hidden dangers and giving drivers valuable extra time for judgment and handling.

[0057] (3) This invention uses a third prediction model to fuse multimodal information such as radar historical trajectory, AIS matching degree and image dynamic changes to make a comprehensive judgment on the existence of targets. Compared with traditional schemes that rely on a single sensor (especially AIS) or simple data association, it can effectively improve the robustness and accuracy of perception in adverse weather, clutter interference and when facing non-cooperative targets (such as small boats without AIS), thereby significantly reducing the false alarm rate and enhancing the reliability of situational awareness in complex environments.

[0058] (4) This invention integrates the ship's real-time draft, maneuver trajectory and static channel data and dynamic safety boundary through the fourth prediction model to conduct route safety assessment. Compared with the traditional scheme based on static electronic chart data and general safety margin, it can effectively realize personalized and dynamic calculation of route safety, thereby more accurately warning of risks such as grounding and collision, and making the safety assessment results more consistent with the ship's instantaneous state and the real navigation environment.

[0059] (5) By introducing an index correction coefficient, this invention quantitatively couples the data confidence to the upper-level risk assessment and finally outputs a comprehensive navigation hazard index. Compared with traditional scattered and isolated multiple sound and light alarm schemes, it can effectively provide drivers with an intuitive and quantitative overall risk level, thereby reducing their cognitive load in tense environments and providing clear decision support for them to take different levels of coping measures.

[0060] (6) This invention takes a pre-trained prediction model that can be optimized and iterated as its core. Compared with the traditional scheme that relies on fixed rules and thresholds, it can effectively learn from historical data (including accident scenarios) to adapt to the characteristics of different waters, ship types and traffic modes, so that the system has the ability to adapt and evolve, laying the technical foundation for moving towards a higher level of intelligent navigation. Attached Figure Description

[0061] Figure 1 This is a flowchart of a method for identifying potential navigation hazards in narrow waters according to an embodiment of the present invention;

[0062] Figure 2 This is a flowchart of step S01 in an embodiment of the present invention;

[0063] Figure 3 This is a flowchart of step S02 in an embodiment of the present invention;

[0064] Figure 4 This is a flowchart of step S03 in an embodiment of the present invention;

[0065] Figure 5 This is a flowchart of step S04 in an embodiment of the present invention;

[0066] Figure 6 This is a flowchart of step S05 in an embodiment of the present invention;

[0067] Figure 7 This is a flowchart of step S06 in an embodiment of the present invention;

[0068] Figure 8 This is an architecture diagram of a ship navigation hazard identification system in narrow waters according to an embodiment of the present invention;

[0069] Figure 9 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0070] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0071] According to an embodiment of the present invention, a method for identifying potential navigation hazards in narrow waters is provided, the flowchart of which is shown below. Figure 1 As shown, it includes:

[0072] Step S01: Acquire sensor data when the ship enters narrow waters and perform data preprocessing; the sensor data includes radar data, GPS data, AIS data, ship maneuvering data, and image data;

[0073] Step S02: Calculate the confidence level of sensor data based on the pre-trained first prediction model; the first prediction model uses the changes in GPS data and / or AIS data and / or image data corresponding to different ship maneuvering data as input features;

[0074] Step S03: Calculate the data contradiction index based on the pre-trained second prediction model; the second prediction model uses the spatiotemporal deviation and / or data deviation of different sensor data of the ship as input features;

[0075] Step S04: Calculate the obstacle target existence index based on the pre-trained third prediction model; the third prediction model uses historical radar data and / or the matching degree between radar data and AIS data and / or the dynamic changes in the image when the ship obstacle target appears as input features.

[0076] Step S05: Calculate the route safety index based on the pre-trained fourth prediction model; the fourth prediction model uses waterway data and / or ship trajectory data and / or ship draft data in narrow waterways as input features;

[0077] Step S06: Calculate the navigation hazard index based on the sensor data confidence level and data contradiction index and / or obstacle target existence index and / or route safety index;

[0078] Step S07: When the navigation hazard index is greater than the preset navigation hazard threshold, it is determined that there is a hazard in the current narrow waterway navigation.

[0079] In a preferred embodiment, the data preprocessing in step S01 is illustrated in the flowchart below. Figure 2 As shown, the steps include:

[0080] Step S011: Synchronize all collected sensor data to a unified spatiotemporal reference and perform spatiotemporal alignment;

[0081] Step S012: Calculate the data source quality assessment value based on the GPS signal quality and / or radar performance status and / or AIS reception status;

[0082] Step S013: Filter out some sensor data based on the data source quality assessment value at each time point.

[0083] In this embodiment, due to the different data acquisition frequencies (e.g., radar: 2-10 seconds / frame, AIS: 2-10 seconds / sentence, GPS: 1-10Hz) and internal clock deviations of sensors such as radar, GPS, and AIS, direct fusion will lead to severe spatiotemporal misalignment. Step S011 aims to establish a unified spatiotemporal reference and interpolate or resample all data streams to ensure consistent state descriptions at the same discrete time points. Specifically, each radar point cloud frame, each AIS message, and each GPS NMEA sentence is stamped with a high-precision reception timestamp (t). recv ); and the timestamps t of all sensor receptions recv The system clock is converted to a unified system clock (e.g., server time synchronized with GPS PPS signals); all sensor location data (e.g., radar polar coordinates, AIS geographic coordinates) are converted to a fixed Cartesian coordinate system centered on the ship (e.g., NE coordinate system); a fixed time interval (e.g., 1 second) is set, and for each aligned time point, an interpolation method is used to obtain the estimated value of each sensor at that moment. The interpolation method can be selected according to the data characteristics, such as linear interpolation or spherical linear interpolation for latitude and longitude.

[0084] In step S012, a dynamic, quantified data source quality assessment value Q is calculated for each time moment and for each type of sensor data. t (The range is usually [0,1]), where 1 represents the best quality and 0 represents unreliable quality, and t represents the sensor data at a certain moment. This value combines the sensor's own health status and external environmental interference.

[0085] The calculation of the data source quality assessment value based on GPS signal quality and / or radar performance status and / or AIS reception status is based on the positive correlation between the GPS signal quality assessment value, / or radar performance status assessment value, and / or AIS reception status assessment value and the data source quality assessment value. The GPS signal quality assessment value is represented by the variable Q. gps (t) represents the radar performance status assessment value using the variable Q. radar (t) represents the AIS receive status assessment value using the variable Q. ais (t) represents the data source quality assessment value Q. t =g1·Q gps (t) g2 +g3·Q radar (t) g4 +g5·Q ais (t) g6 +g7, or Q t =g8·Q gps (t) g9 ·Q ais (t) g10 ·Q ais (t) g11 +g12, where g1~g12 are the computational coefficients obtained through pre-training.

[0086] GPS signal quality assessment value Q gps (t) is based on the number of visible satellites (N) sat Position Precision Factor (PDOP), Mean Signal-to-Noise Ratio (SNR) avg The receiver autonomous integrity (RAIM) flag is calculated as shown in equation (1).

[0087] (1)

[0088] Where w1, w2, w3, and w4 are weights and w1+w2+w3+w4=1, the sigmoid function is used to smooth the satellite number threshold N. thresh PDOP scale For the scale parameter, I raim RAIM availability indicator (value is 0 or 1).

[0089] Radar performance status assessment value Q radar(t) is calculated based on the deviation of the sea clutter suppression level (STC), rain and snow clutter suppression level (FTC), receiver gain (Gain), antenna rotation speed (RPM) from their rated values ​​and the detection of sudden changes in the number of historical echoes, as shown in Equation (2).

[0090] (2)

[0091] Among them, f clutter It is a function that decreases as the level of inhibition increases, f gain and f ant It is a function that measures whether the parameter is within the normal range, I outlier α is a preset penalty coefficient used to indicate outliers detected based on historical echo number sequences.

[0092] AIS Receive Status Assessment Value Q ais (t) is based on the received signal strength (RSSI), message bit error rate (BER), and the stability of the message reception interval (Δt). std The result is obtained by calculation, as shown in equation (3).

[0093] (3)

[0094] Where β is a preset stability penalty coefficient.

[0095] In step S013, a minimum usable quality threshold is preset for each type of sensor. For any time t, if the data source quality assessment value Q... t If the quality is less than the minimum available quality threshold, the original observations of this sensor are temporarily removed from the fusion pool at that moment.

[0096] In another implementation, a critical usable quality threshold is pre-set. If the quality assessment value of a sensor data falls between the critical usable quality threshold and the lowest usable quality threshold, the feature weight corresponding to that data is reduced in subsequent models (such as the first and second prediction models). When a sensor data is discarded due to low quality, the system uses its most recent valid observation value, combined with a ship motion model (such as a constant speed model CV or a constant rotational speed model CTRV), to perform a short-term state prediction as a temporary replacement value for that sensor data to maintain the continuity of the data stream, but it will mark its prediction source and the lower quality assessment value.

[0097] In a preferred embodiment, step S02, calculating the confidence level of the sensor data based on the pre-trained first prediction model, is illustrated in the flowchart below. Figure 3 As shown, it includes:

[0098] Step S021: Calculate the GPS matching value based on the degree of matching between different ship maneuvering data and the corresponding GPS data changes;

[0099] Step S022: Calculate the AIS matching value based on the degree of matching between different ship maneuvering data and AIS data changes;

[0100] Step S023: Calculate the image matching value based on the degree of matching between different ship maneuvering data and image data changes;

[0101] Step S024: Construct a first prediction model based on the positive correlation between GPS matching value and / or AIS matching value and / or image matching value and sensor data confidence level, and calculate the sensor data confidence level accordingly.

[0102] In this embodiment, different ship maneuvering data include rudder angle (δ), main engine speed (RPM), heading, and other data.

[0103] In step S021, calculating the GPS matching value based on the degree of matching between different ship maneuvering data and the corresponding GPS data changes includes:

[0104] (1) Based on the current and historical rudder angle (δ), main engine speed (RPM) and other maneuvering data, combined with the ship hydrodynamic model, generate the ship's expected motion vector V at instant t. pred (t); The ship hydrodynamic model is the Nomoto model or an empirical model;

[0105] (2) Extract the actual observed motion vector V at instant t from the GPS sequence data by time difference and filtering. obs (t);

[0106] (3) Calculate the difference D between the expected and observed vectors based on the weighted Euclidean distance between the expected and observed vectors. gps (t), and map it to the GPS matching value M. gps (t), as shown in equation (4).

[0107] (4)

[0108] Among them, the mapping function Mapping_Function makes M gps (t)∈[0,1], and the function is similar to the difference D gps (t) shows a negative correlation.

[0109] In step S022, calculating the AIS matching value based on the degree of matching between different ship maneuvering data and AIS data changes includes:

[0110] (1) The theoretically reasonable relationship and actual deviation between the key parameters inside the AIS message are calculated based on the principles of nautical physics, and denoted as internal deviation E. internal (t). For example, calculating the difference between COG and HDG under conditions without strong current (COG). ais (t)-HDG ais (t)) and perform a difference operation with the theoretical drift angle Theoretical_Drift_Angle(δ(t)) under steering conditions to obtain E. internal (t);

[0111] (2) Based on the dynamic information (SOG) reported by AIS ais COG ais ) and GPS reference (SOG) gps COG gps The difference operation calculates the external deviation E. external (t);

[0112] (3) Based on the internal deviation E internal (t) and external deviation E external (t) Calculate the AIS matching value M ais (t), as shown in equation (5).

[0113] (5)

[0114] Where γ1 and γ2 are preset adjustment coefficients, such that M ais (t) ∈ [0,1].

[0115] In step S023, calculating the image matching value based on the degree of matching between different ship maneuvering data and image data variations includes:

[0116] (1) Extract the visual feature change ΔF representing the ship's motion trend from the image sequence within the time window [t-Δt, t] using optical flow, feature tracking, or deep learning models. vision (t);

[0117] (2) Generate the expected ship motion trend T based on the manipulation data (mainly rudder angle δ) in the same time window. pred (t) (e.g., the expected direction or magnitude of image feature changes);

[0118] (3) Calculate the change in visual features ΔF vision (t) and the expected ship motion trend T pred The cosine similarity of (t) is denoted as S. img (t), which is normalized to the image matching value M img M_img(t) takes values ​​in the interval [0,1].

[0119] In step S024, the step of constructing a first prediction model based on the positive correlation between GPS matching values ​​and / or AIS matching values ​​and / or image matching values ​​and sensor data confidence, and calculating the sensor data confidence based on this model, includes:

[0120] (1) Construct a feature vector Input using GPS matching values ​​and / or AIS matching values ​​and / or image matching values ​​as joint input features. Features (t);

[0121] (2) Calculate the sensor data confidence preset model C(t) = u1·M based on the positive correlation between GPS matching value and / or AIS matching value and / or image matching value and sensor data confidence. gps (t) u2 +u3·M ais (t) u4 +u5·M img (t) u6 +u7, or C(t) = u8·M gps (t) u9 ·M ais (t) u10 ·M img (t) u11 +u12, where u1~u12 are the computational coefficients obtained through pre-training.

[0122] (3) The constructed feature vector Input Features The sensor data confidence score C(t) is input into the pre-trained first prediction model (e.g., a fully connected neural network or a gradient boosting decision tree model) and the sensor data confidence score C(t) is obtained. C(t)∈[0, 1], and the higher the value, the higher the confidence score of the sensor data at the current time.

[0123] In a preferred embodiment, step S03, calculating the data inconsistency index based on the pre-trained second prediction model, is illustrated in the flowchart below. Figure 4 As shown, it includes:

[0124] Step S031: Calculate the spatiotemporal deviation value based on the degree of time deviation and / or position indication deviation of different sensor data of the ship;

[0125] Step S032: Calculate the data deviation value based on the degree of correlation data deviation of different sensor data on the ship;

[0126] Step S033: Construct a second prediction model based on the positive correlation between the spatiotemporal deviation value and / or the data deviation value and the data contradiction index, and calculate the data contradiction index accordingly.

[0127] In this embodiment, step S031, calculating the spatiotemporal deviation value based on the degree of time deviation and / or position indication deviation of different sensor data of the ship, includes:

[0128] (1) Calculate the time delay for the same target (such as the ship itself) to reach the same state (such as the maximum turning rate point) in different sensor data streams, and obtain the time deviation value Δt. ij Where i and j represent different sensor pairs;

[0129] (2) Under a unified coordinate system, calculate the Euclidean distance between the positioning results of different sensors for the same target (mainly other ships) at the same time t, and obtain the position indication deviation value d. ij (t);

[0130] (3) Based on the time deviation value Δt ij and / or position indication deviation value d ij (t) Calculate the spatiotemporal deviation value E st (t), as shown in equation (6).

[0131] (6)

[0132] Where ε1 and ε2 are computational coefficients obtained through pre-training.

[0133] In step S032, calculating the data deviation value based on the degree of correlation data deviation of different sensor data of the ship includes:

[0134] (1) Extract physically related paired parameters. For example: SOG from the ship's AIS report. ais SOG calculated with GPS gps COG reported by the ship's AIS ais COG of the ship calculated based on multi-frame radar echoes radar The speed v of the target tracked by the other ship's radar radar SOG with AIS report ais The heading θ tracked by the radar of another ship radar COG reported by AIS ais .

[0135] (2) Calculate the relative deviation value for each pair of correlation parameters, and record it as the correlation data deviation value e. k (t)=|P i (t)-P j (t)| / (Max(P i (t), P j (t)));

[0136] (3) Calculate the data deviation value E based on the degree of deviation of the associated data. data(t), as shown in equation (7).

[0137] (7)

[0138] Among them, w k The weight is the pre-defined weight of the deviation of the k-th group of related parameters.

[0139] In step S033, the step of constructing a second prediction model based on the positive correlation between the spatiotemporal deviation value and / or the data deviation value and the data inconsistency index, and calculating the data inconsistency index accordingly, includes:

[0140] (1) Construct a feature vector X by using the spatiotemporal deviation value and the data deviation value as joint input features. conflict (t);

[0141] (2) Calculate the pre-set model I for the data contradiction index based on the positive correlation between the spatiotemporal deviation value and / or the data deviation value and the data contradiction index. conflict (t)=u13·E st (t) u14 +u15·E data (t) u16 +u17, or I conflict (t)=u18·E st (t) u19 ·E data (t) u20 +u21, u13~u21 are the computational coefficients obtained through pre-training.

[0142] (3) Construct the feature vector X conflict (t) and the pre-defined model of the data contradiction index are input into a pre-trained second prediction model (e.g., using historical data (event segments marked with known sensor conflicts, spoofing, or malfunctions) for supervised learning to train a regression model (such as a neural network or support vector regression)) to obtain the data contradiction index I. conflict (t), I conflict (t)∈[0, 1], the higher the value, the greater the contradiction between the current multi-source data, the worse the internal consistency of the entire situation awareness system, and the higher the risk of systematic misleading.

[0143] In a preferred embodiment, step S04 involves calculating the obstacle target presence index based on the pre-trained third prediction model, as shown in the flowchart below. Figure 5 As shown, it includes:

[0144] Step S041: Calculate the obstacle target continuity indication value based on the continuity of historical radar data when the ship obstacle target appears;

[0145] Step S042: Calculate the non-visual presence indication value of the obstacle target based on the degree of matching between radar data and AIS data when the ship obstacle target appears;

[0146] Step S043: Calculate the visual presence indication value of the obstacle target based on the dynamic changes in the background in the image when the ship obstacle target appears;

[0147] Step S044: Construct a third prediction model based on the positive correlation between the obstacle target continuity indicator value and / or the obstacle target non-visible presence indicator value and / or the obstacle target visible presence indicator value and the obstacle target presence index, and calculate the obstacle target presence index accordingly.

[0148] In this embodiment, step S041, calculating the obstacle target continuity indication value based on the continuity of radar historical data when the ship obstacle target appears, includes:

[0149] (1) The position sequence (x) of the ship obstacle target within the time window [t-Δt, t] i , y i Perform least-squares linear fitting and calculate the average residual E of the trajectory. fit =mean(|y i -(a·x i +b)|), where a and b are the fitting coefficients;

[0150] (2) Calculate the standard deviation σ of the instantaneous velocity v(t) and heading θ(t) of the ship obstacle target. v and σ θ ;

[0151] (3) Based on the average residual E of the trajectory fit The obstacle-target continuity indication value P is calculated using the standard deviation of the target's instantaneous velocity v(t) and heading θ(t). cont As shown in equation (8).

[0152] (8)

[0153] Where λ1, λ2, and λ3 are preset attenuation coefficients, and P cont ∈(0,1], the higher the value, the more continuous and reasonable the target trajectory is.

[0154] In step S042, calculating the non-visual presence indication value of the obstacle target based on the degree of matching between radar data and AIS data when the ship obstacle target appears includes:

[0155] (1) For radar target T radar Searching for AIS target T within the spatiotemporal correlation threshold ais If a match exists, calculate the consistency score S for position and velocity.match As shown in equation (9).

[0156] (9)

[0157] Where dist represents the distance to the target, and f() is a pre-trained function that is negatively correlated with the difference between position distance and velocity.

[0158] (2) Based on the consistency score S of position and velocity match Calculate the non-visible presence indicator value P of the obstacle target nonvis As shown in equation (10).

[0159] (10)

[0160] Among them, W region The weight is assigned to a region (based on the strength of AIS coverage; a region with weak AIS coverage will have a higher weight).

[0161] In step S043, calculating the visual presence indication value of the obstacle target based on the dynamic changes in the background of the image when the ship obstacle target appears includes:

[0162] (1) Calculate the expected projection area R of the target in the image based on the radar target position and camera calibration parameters. proj ;

[0163] (2) In R pro Within the region, the average optical flow amplitude M within the region is calculated using the optical flow method. flow ;

[0164] (3) Use an object detection model (such as YOLO) to detect whether there are categories such as "ships" or "floating objects" in the area. det .

[0165] (4) Calculate the visual presence indicator value P of the obstacle target. vis As shown in equation (11).

[0166] (11)

[0167] Where η is the preset weight coefficient, and Normalize() is the normalization function.

[0168] In step S044, the step of constructing a third prediction model based on the positive correlation between the obstacle target continuity indicator value and / or the obstacle target non-visible presence indicator value and / or the obstacle target visible presence indicator value and the obstacle target presence index, and calculating the obstacle target presence index accordingly, includes:

[0169] (1) The obstacle target continuity indicator value Pcont The indicator value P for the non-visible presence of obstacles / targets nonvis Obstacle target visibility indicator value P vis The feature vector X is constructed using the joint input features. obstacle (t);

[0170] (2) Calculate the pre-set model I of obstacle target existence index based on the positive correlation between the obstacle target continuity indicator value and / or the obstacle target non-visible existence indicator value and / or the obstacle target visible existence indicator value and the obstacle target existence index. obstacle (t)=u22·P cont u23 +u24·P nonvis u25 +u26·P vis u27 +u28, or I obstacle (t)=u29·P cont u30 ·P nonvis u31 ·P vis u32 +u33, u22~u33 are the computational coefficients obtained through pre-training.

[0171] (3) Construct the feature vector X obstacle (t) and the obstacle target existence index preset model are input into a pre-trained third prediction model (e.g., using historical radar data labeled with real targets and false alarms to train a classification model (such as random forest or neural network) and outputting the existence probability), to obtain the obstacle target existence index I. obstacle (t), I obstacle (t)∈[0, 1], the higher the value, the higher the confidence that the target actually exists, and it should be taken into account for navigation risk.

[0172] In a preferred embodiment, step S05 involves calculating the route safety index based on the pre-trained fourth prediction model, as shown in the flowchart below. Figure 6 As shown, it includes:

[0173] Step S051: Calculate the safety boundary of the narrow waterway based on the channel width and / or channel depth;

[0174] Step S052: Calculate the ship trajectory safety indication value based on the distance and / or distance change trend between the ship trajectory data and the safety boundary of the narrow waterway;

[0175] Step S053: Calculate the ship's draft safety indication value based on the distance and / or the trend of distance change between the ship's draft data and the safety boundary of the narrow waterway;

[0176] Step S054: Construct a fourth prediction model based on the positive correlation between the ship trajectory safety indication value and / or the ship draft safety indication value and the route safety index, and calculate the route safety index accordingly.

[0177] In this embodiment, calculating the safety boundary of the narrow waterway based on the channel width and / or channel depth includes:

[0178] (1) Extract the channel centerline L from the electronic chart (ENC) center Official markings of the left and right boundaries of the waterway B left B right and contour data;

[0179] (2) Based on the left and right boundaries, contract inward by a safety margin M. h (Based on parameters such as ship beam, speed, visibility, wind and current pressure, and dynamically adjusted according to an empirical model), the horizontal safety boundary SB of the narrow waterway is obtained. horizontal (t)={B left -M h B right +M h};

[0180] (3) Calculate the vertical safety boundary SB of the narrow waterway based on the ship's draft, chart depth, and real-time tide height. vertical (t)=Charted Depth +Tide(t)-Ship Draft (t)-U k Charted Depth The chart shows the water depth, Tide(t) represents the real-time tide height, and Ship... Draft (t) represents the ship's real-time draft, U k This indicates the safe margin of water depth (dynamically adjusted based on parameters such as bottom sediment, speed, and ship maneuverability according to an empirical model).

[0181] In step S052, calculating the ship trajectory safety indication value based on the distance and / or distance change trend between the ship trajectory data and the safety boundary of the narrow waterway includes:

[0182] (1) Calculate the distance from the ship's current position P(t) to the nearest horizontal safety boundary SB. horizontal The shortest distance d of (t) min (t)=min_distance(P(t),SB horizontal (t));

[0183] (2) Based on the current heading COG and speed SOG, predict the position P after a future time Δt (e.g., 30 seconds). pred(t+Δt), and calculate the position from the nearest horizontal safety boundary SB after time Δt. horizontal The shortest distance d of (t) pred (t+Δt);

[0184] (3) Based on the shortest distance d from the current position to the nearest horizontal safety boundary min (t) and the shortest distance d from the position to the nearest horizontal safety boundary after time Δt. pred (t+Δt) Calculate the ship trajectory safety indication value S traj (t), as shown in equation (12).

[0185] (12)

[0186] Where τ is a preset trajectory trend prediction weight factor (0 < τ ≤ 1), Sig() is the Sigmoid function, used to map the distance to a safety score of [0, 1] (the greater the distance, the higher the score), and D safe It is a preset baseline safety distance threshold.

[0187] In step S053, calculating the ship's draft safety indication value based on the distance and / or the trend of distance change between the ship's draft data and the safety boundary of the narrow waterway includes:

[0188] (1) Select the positions x(t) of multiple key points (such as bow, midship, and stern) along the length of the ship, and calculate the distance from each point to the vertical safety boundary SB. vertical The minimum vertical distance U of (t) min (t)=min vertical (x(t), SB) vertical (t));

[0189] (2) Based on the ship's draft, chart depth, and real-time tide height, predict the location x of the key point after a future time Δt (e.g., 30 seconds). pred (t+Δt), based on which the time Δt is calculated and the nearest vertical safety boundary SB is obtained. vertical The shortest distance U of (t) pred (t+Δt);

[0190] (3) According to U min (t) and U pred Calculate the ship's draft safety indication value S using (t+Δt). draft (t), as shown in equation (12).

[0191] (12)

[0192] Where ρ is a preset draft trend prediction weight factor (0 < ρ ≤ 1), Sig() is the Sigmoid function used to map the distance to a security score of [0, 1] (the greater the distance, the higher the score), and U safe It is a preset baseline excess water depth threshold.

[0193] In step S054, the step of constructing a fourth prediction model based on the positive correlation between the ship trajectory safety indication value and / or the ship draft safety indication value and the route safety index, and calculating the route safety index accordingly, includes:

[0194] (1) Set the ship trajectory safety indication value S traj (t) and ship draft safety indication value S draft (t) is used as joint input features to construct feature vector X route (t);

[0195] (2) Calculate the pre-set model I of the route safety index based on the positive correlation between the ship trajectory safety indication value and / or the ship draft safety indication value and the route safety index. route (t)=u34·S traj (t) u35 +u36·S draft (t) u37 +u38, or I conflict (t)=u39·S traj (t) u40 ·S draft (t) u41 +u42, u34~42 are the computational coefficients obtained through pre-training.

[0196] (3) Construct the feature vector X route (t) and the preset model of the route safety index are input into a pre-trained fourth prediction model (e.g., using historical flight data (including accident / near-miss event labels) to train a regression model (such as XGBoost or a neural network) to obtain the route safety index I. route (t), I route (t)∈[0, 1], the higher the value, the safer the route, and the lower the value, the higher the risk of grounding or collision.

[0197] In a preferred embodiment, step S06 involves calculating a navigation hazard index based on the confidence level of the sensor data, the data contradiction index, and / or the obstacle / target presence index and / or the route safety index. The flowchart is shown below. Figure 7 As shown, it includes:

[0198] Step S061: Calculate the index correction coefficient based on the degree of influence of the confidence level of sensor data on the data contradiction index and / or obstacle target existence index and / or route safety index;

[0199] Step S062: Calculate the navigation anomaly index based on the degree of impact of data contradiction index and / or obstacle target existence index and / or route safety index on ship navigation safety;

[0200] Step S063: Correct the navigation anomaly index according to the index correction coefficient to obtain the navigation hazard index.

[0201] In this embodiment, the calculation of the index correction coefficient based on the impact of the confidence level of sensor data on the data contradiction index and / or the obstacle target existence index and / or the route safety index includes:

[0202] (1) The less reliable the data, the greater the risk of data contradiction itself. Therefore, the confidence level of sensor data is negatively correlated with the data contradiction index. Based on this, the correction coefficient α of the data contradiction index is calculated. conflict In one implementation, a negative correlation function is used to calculate the correction coefficient α for the data inconsistency index. conflict As shown in equation (13).

[0203] (13)

[0204] Where o1 represents the preset sensitivity coefficient (o1>0), and C(t) is the confidence level of the sensor data calculated in step S02;

[0205] (2) When the data is unreliable, radar / visual perception data may be distorted, leading to unreliable target existence judgment. Therefore, the confidence level of sensor data is positively correlated with the obstacle target existence index. Based on this, the obstacle target existence index correction coefficient α is calculated. obstacle In one implementation, the obstacle target existence index correction coefficient α obstacle The calculation method is shown in equation (14).

[0206] (14)

[0207] Where μ1 is a preset obstacle target existence index adjustment coefficient.

[0208] (3) Route safety assessment heavily relies on accurate positioning (GPS) and airway data. Low data confidence levels render the safety assessment unreliable; therefore, the confidence level of sensor data is positively correlated with route safety indicators. Based on this, the route safety indicator correction coefficient α is calculated. route In one implementation, the route safety index correction factor α routeThe calculation method is shown in equation (15).

[0209] (15)

[0210] Where μ2 is the preset route safety index adjustment coefficient.

[0211] In step S062, the calculation of the navigation anomaly index based on the impact of data contradiction index and / or obstacle target existence index and / or route safety index on ship navigation safety is a comprehensive calculation of three basic risk indicators (data contradiction index I). conflict (t), Obstacle Target Existence Indicator I obstacle (t), Route safety index I route (t) Based on the positive correlation between data contradiction indicators and navigation anomaly indicators, and the negative correlation between obstacle / target presence indicators and route safety indicators and navigation anomaly indicators, navigation anomaly index I is calculated. anomaly (t). In one embodiment, navigation anomaly index I anomaly (t) is shown in equation (16).

[0212] (16)

[0213] Among them, r1, r2, and r3 are preset weight coefficients (preset according to the characteristics of the water area).

[0214] In step S063, the navigation anomaly index is corrected according to the index correction coefficient to obtain the navigation hazard index I. hazard (t), as shown in equation (17).

[0215] (17)

[0216] Among them, z1, z2, and z3 are preset correction weight coefficients (preset according to different usage scenarios of the ship).

[0217] In another embodiment, I hazard (t) An additional risk arising from system uncertainty also needs to be considered, namely, a non-reliability penalty. The non-reliability penalty is calculated based on the average correction coefficient. The navigation hazard index I with the non-reliability penalty added is... hazard (t), as in equation (18).

[0218] (18)

[0219] in, This represents the average correction factor. .

[0220] When navigation hazard index I hazard(t) is greater than the preset navigation hazard threshold I TH If so, it is determined that there is a potential hazard in the current narrow waterway.

[0221] According to another embodiment of the present invention, a system for identifying potential navigation hazards in narrow waters is provided, the system framework diagram of which is shown below. Figure 8 As shown, it includes:

[0222] The ship data acquisition and preprocessing module is used to acquire sensor data and perform data preprocessing when a ship enters narrow waters.

[0223] The first prediction module is used to calculate the confidence level of sensor data based on a pre-trained first prediction model; the first prediction model uses the changes in GPS data and / or AIS data and / or image data corresponding to different ship maneuvering data as input features;

[0224] The second prediction module is used to calculate the data inconsistency index based on the pre-trained second prediction model; the second prediction model uses the spatiotemporal deviation and / or data deviation and / or data source quality of different sensor data of the ship as input features.

[0225] The third prediction module is used to calculate the presence index of the obstacle target based on the pre-trained third prediction model. The third prediction model uses historical radar data and / or the matching degree between radar data and AIS data and / or the dynamic changes in the image when the ship obstacle target appears as input features.

[0226] The fourth prediction module is used to calculate the route safety index based on the pre-trained fourth prediction model; the fourth prediction model uses waterway data and / or ship trajectory data and / or ship draft data in narrow waterways as input features.

[0227] Navigation Hazard Identification Module: This module is used to call the outputs of the first prediction module, the second prediction module, and / or the third prediction module and / or the fourth prediction module to calculate the navigation hazard index. When the navigation hazard index is greater than the preset navigation hazard threshold, it is determined that there is a hazard in the current narrow waterway.

[0228] In this embodiment, each module of the system executes the above-mentioned method for identifying potential navigation hazards in narrow waters, which will not be described in detail here.

[0229] According to another embodiment of the present invention, a computer-readable storage medium is provided that stores a computer program for electronic data exchange, wherein the computer program causes a computer to execute the above-described method for identifying potential navigation hazards in narrow waters.

[0230] According to another embodiment of the present invention, an electronic device is provided, the structural schematic diagram of which is shown below. Figure 9 As shown, it includes:

[0231] At least one processor;

[0232] and a memory communicatively connected to the at least one processor;

[0233] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the above-described method for identifying potential navigation hazards in narrow waters.

[0234] Of course, those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention and are not intended to limit the present invention. Any changes or modifications to the above embodiments that are within the scope of the present invention will fall within the protection scope of the present invention.

Claims

1. A method for identifying potential navigation hazards in narrow waters, characterized in that, include: The system acquires sensor data when a vessel navigates through narrow waterways and performs data preprocessing; the sensor data includes radar data, GPS data, AIS data, vessel maneuvering data, and image data. The confidence level of sensor data is calculated based on the pre-trained first prediction model; the first prediction model uses the changes in GPS data and / or AIS data and / or image data corresponding to different ship maneuvering data as input features; The data inconsistency index is calculated based on the pre-trained second prediction model; the second prediction model uses the spatiotemporal deviation and / or data deviation of different sensor data of the ship as input features. The presence index of the obstacle target is calculated based on the pre-trained third prediction model; the third prediction model uses historical radar data and / or the matching degree between radar data and AIS data and / or the dynamic changes in the image when the ship obstacle target appears as input features; The route safety index is calculated based on the pre-trained fourth prediction model; the fourth prediction model uses waterway data and / or ship trajectory data and / or ship draft data in narrow waterways as input features; The navigation hazard index is calculated based on the confidence level of sensor data, the data contradiction index, and / or the obstacle target existence index and / or the route safety index. When the navigation hazard index exceeds the preset navigation hazard threshold, it is determined that there is a hazard in the current narrow waterway navigation.

2. The method for identifying potential navigation hazards in narrow waters according to claim 1, characterized in that, The data preprocessing includes the following steps: All collected sensor data are synchronized to a unified spatiotemporal reference and spatiotemporally aligned. Calculate the data source quality assessment value based on GPS signal quality and / or radar performance status and / or AIS reception status; Some sensor data is filtered out based on the data source quality assessment value at each time point.

3. The method for identifying potential navigation hazards in narrow waters according to claim 1, characterized in that, The step of calculating the confidence level of sensor data based on the pre-trained first prediction model includes: The GPS matching value is calculated based on the degree of matching between different ship maneuvering data and the corresponding changes in GPS data. AIS matching values ​​are calculated based on the degree of matching between different ship maneuvering data and AIS data variations. The image matching value is calculated based on the degree of matching between different ship maneuvering data and image data changes. The first prediction model is constructed based on the positive correlation between GPS matching values ​​and / or AIS matching values ​​and / or image matching values ​​and sensor data confidence, and the sensor data confidence is calculated accordingly.

4. The method for identifying potential navigation hazards in narrow waters according to claim 1, characterized in that, The calculation of the data inconsistency index based on the pre-trained second prediction model includes: The spatiotemporal deviation value is calculated based on the degree of time deviation and / or position indication deviation of data from different sensors on the ship. The data deviation value is calculated based on the degree of data deviation in the correlation of data from different sensors on the ship. A second prediction model is constructed based on the positive correlation between the spatiotemporal deviation value and / or the data deviation value and the data inconsistency index, and the data inconsistency index is calculated accordingly.

5. The method for identifying potential navigation hazards in narrow waters according to claim 1, characterized in that, The calculation of the obstacle target presence index based on the pre-trained third prediction model includes: The continuity indication value of the obstacle target is calculated based on the continuity of historical radar data when the ship obstacle target appears; The non-visual presence indication value of the obstacle target is calculated based on the degree of matching between radar data and AIS data when the ship obstacle target appears. The visual presence indication value of the obstacle target is calculated based on the dynamic changes in the background of the image when the ship obstacle target appears. A third prediction model is constructed based on the positive correlation between the obstacle target continuity indicator value and / or the obstacle target non-visible presence indicator value and / or the obstacle target visible presence indicator value and the obstacle target presence index, and the obstacle target presence index is calculated accordingly.

6. The method for identifying potential navigation hazards in narrow waters according to claim 1, characterized in that, The calculation of route safety indicators based on the pre-trained fourth prediction model includes: Calculate the safety boundary of the narrow waterway based on the width and / or depth of the narrow waterway; The ship trajectory safety indication value is calculated based on the distance and / or the trend of distance change between the ship trajectory data and the safety boundary of the narrow waterway; The ship's draft safety indication value is calculated based on the distance and / or the trend of distance change between the ship's draft data and the safety boundary of the narrow waterway. A fourth prediction model is constructed based on the positive correlation between the ship trajectory safety indication value and / or the ship draft safety indication value and the route safety index, and the route safety index is calculated accordingly.

7. The method for identifying potential navigation hazards in narrow waters by ships according to claim 1, characterized in that, The calculation of navigational hazard indicators based on the confidence level and data contradiction index of sensor data and / or the presence index of obstacle targets and / or the route safety index includes: Calculate the index correction coefficient based on the degree of influence of the confidence level of sensor data on the data contradiction index and / or the obstacle target existence index and / or the route safety index; The navigation anomaly index is calculated based on the degree of impact of data contradiction index and / or obstacle target existence index and / or route safety index on ship navigation safety. The navigation hazard index is obtained by correcting the navigation anomaly index based on the index correction coefficient.

8. A system for identifying potential navigation hazards in narrow waters, characterized in that, include: The ship data acquisition and preprocessing module is used to acquire sensor data and perform data preprocessing when a ship enters narrow waters. The first prediction module is used to calculate the confidence level of sensor data based on a pre-trained first prediction model; the first prediction model uses the changes in GPS data and / or AIS data and / or image data corresponding to different ship maneuvering data as input features; The second prediction module is used to calculate the data inconsistency index based on the pre-trained second prediction model; the second prediction model uses the spatiotemporal deviation and / or data deviation and / or data source quality of different sensor data of the ship as input features. The third prediction module is used to calculate the presence index of the obstacle target based on the pre-trained third prediction model. The third prediction model uses historical radar data and / or the matching degree between radar data and AIS data and / or the dynamic changes in the image when the ship obstacle target appears as input features. The fourth prediction module is used to calculate the route safety index based on the pre-trained fourth prediction model; the fourth prediction model uses waterway data and / or ship trajectory data and / or ship draft data in narrow waterways as input features. Navigation Hazard Identification Module: This module is used to call the outputs of the first prediction module, the second prediction module, and / or the third prediction module and / or the fourth prediction module to calculate the navigation hazard index. When the navigation hazard index is greater than the preset navigation hazard threshold, it is determined that there is a hazard in the current narrow waterway.

9. A computer-readable storage medium storing a computer program for electronic data interchange, wherein, The computer program causes the computer to perform the method as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1-7.

Images