A method and system for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification

Abstract

The application discloses a kind of based on online position observation and equivalent disturbance identification out-of-control ship drift trajectory prediction correction method and system, belong to out-of-control ship drift trajectory online prediction and bridge ship collision risk early warning technical field, solve the existing prediction mode difficult to correct systematic and continuous deviation of dynamics model, prone to observation abnormal interference and appear excessive correction, difficult to stabilize the output complete ship collision risk parameter problem.The application obtains the online position observation sequence of out-of-control ship, selects observation window, and obtains model playback trajectory by initial dynamics model with window playback, calculates the residual sequence of observation trajectory and playback trajectory and decomposes to tangential, normal and bow shake / curvature channel, identifies equivalent disturbance and is injected into dynamics model after innovation truncation, amplitude limiting, reliability evaluation and prediction attenuation processing, recursively generates the future drift trajectory after correction, outputs bridge ship collision risk parameter in combination with bridge area geometric boundary, and is suitable for bridge area ship collision risk early warning scene.

Description

Technical Field

[0001] This invention belongs to the field of online prediction of drift trajectory of out-of-control ships and early warning of bridge collision risk, and in particular relates to a method and system for predicting and correcting the drift trajectory of out-of-control ships based on online position observation and equivalent disturbance identification. Background Technology

[0002] In bridge areas or restricted waters, after a vessel experiences main engine failure, loss of rudder effectiveness, or restricted maneuverability, it may drift towards bridge piers, collision avoidance facilities, navigation channel boundaries, or the shoreline under the combined effects of wind, current, inertia, and hydrodynamic response. Bridge area early warning systems need to continuously predict future drift trajectories based on online observations of the vessel's position and generate parameters such as impact point, impact speed, impact attitude, impact angle, minimum distance from the bridge, and risk level for engineering intervention.

[0003] Existing drift trajectory prediction methods typically include dynamic model extrapolation, state filtering estimation, geometric trajectory extrapolation, and data-driven prediction. Dynamic model extrapolation offers physical interpretability, but when actual wind disturbances, hydrodynamic parameters, initial states, or the local environment of the bridge area deviate from model assumptions, future trajectories are prone to systematic biases. State filtering methods primarily update the current state and are insufficient in representing persistent unmodeled disturbances or parameter biases in the future prediction segment. Geometric trajectory extrapolation makes limited use of dynamic mechanisms and struggles to stably output bridge collision risk parameters such as impact attitude and impact angle. Deep learning trajectory prediction usually relies on numerous similar scenario samples, which presents limitations in sample coverage, generalization ability, and interpretability in small-sample early warning scenarios of out-of-control vessels in bridge areas.

[0004] Therefore, it is necessary to propose a method for dynamic correction using model biases exposed in the observed location sequence. Summary of the Invention

[0005] In view of this, the present invention aims to propose a method and system for predicting and correcting the drift trajectory of out-of-control vessels based on online position observation and equivalent disturbance identification, so as to solve the problems that existing methods for predicting the drift trajectory of out-of-control vessels in bridge areas cannot effectively correct the systematic and continuous deviations of the dynamic model, are easily affected by abnormal observations and thus overcorrect, and are difficult to stably output complete bridge collision risk warning parameters.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification, comprising the following steps: Obtain the online position observation sequence of the out-of-control vessel. The online position observation sequence shall include at least the planar position coordinates of the out-of-control vessel in the bridge area or restricted waters. When the bow angle can be obtained, the online position observation sequence shall also include the bow angle. The body axis velocity and yaw rate shall be reconstructed from the position coordinates and / or the bow angle. Select an observation window from the online position observation sequence and align the observation time points, position coordinates, and optional heading angles within the observation window with time. Using the initial state of the observation window as the initial condition, the initial dynamic model is called to perform dynamic model playback within the same observation window to obtain the model playback trajectory corresponding to the observation time point. Calculate the residual sequence between the observation trajectory formed by the online location observation sequence within the observation window and the model playback trajectory; The residual sequence is projected or decomposed into the tangential channel, normal channel, and yaw / curvature channel to obtain the residual measurement values ​​for each channel; Based on the residual measurement values ​​of each channel, the equivalent disturbance is identified, and the correction amount corresponding to the equivalent disturbance is innovatively truncated, the correction amount is limited, and the limit ratio is statistically analyzed. The reliability scalar is calculated based on at least two of the following: residual fitting loss, observation window support range, correction amount limiting ratio, and observation window length. The correction amount is then subjected to prediction time or prediction distance attenuation based on the reliability scalar. Using the state at the end of the observation window as the initial state for prediction, the processed equivalent disturbance is injected into the dynamic model to recursively generate the corrected future drift trajectory. Based on the intersection relationship between the corrected future drift trajectory and the bridge piers, anti-collision facilities, navigation channel boundaries or pre-set danger zones, bridge collision risk parameters are output. The bridge collision risk parameters include at least three of the following: impact point, impact speed, impact attitude, impact angle, minimum distance from the bridge, and risk level.

[0007] Furthermore, the observation window can be selected in any of the following ways: a fixed time window, a fixed number of measurement points window, a sliding window, a distance window, or an adaptive window based on noise level and motion stability.

[0008] Furthermore, when only position coordinates are obtained, the direction of motion, drift velocity, acceleration trend, trajectory curvature, and yaw trend are estimated based on adjacent position points or locally fitted trajectories; when the heading angle is obtained, the heading angle is used to constrain or correct the reconstructed body axis velocity and yaw angular velocity.

[0009] Furthermore, the residual sequence includes one or more of the following: position residual, velocity residual obtained by differentiating the position residual, and yaw / curvature residual obtained by the heading angle or trajectory curvature. Projecting or decomposing the residual sequence into the tangential channel, normal channel, and yaw / curvature channel includes: establishing tangential unit vectors and normal unit vectors based on the local tangential direction of the observed trajectory or model playback trajectory; projecting the position residual into tangential residual and normal residual; and forming yaw / curvature channel residual measurements based on the variation trend of the heading angle difference, curvature difference, or normal residual with distance.

[0010] Furthermore, the residual sequence is decomposed into one or more of the following: initial state deviation, velocity scale deviation, lateral offset deviation, and curvature deviation.

[0011] Furthermore, the equivalent disturbance includes tangential equivalent disturbance, normal equivalent disturbance, and yaw / curvature equivalent disturbance. The normal equivalent disturbance and the yaw / curvature equivalent disturbance respectively contain steady-state components and transient components. The equivalent disturbance is injected into the dynamic model in the form of equivalent velocity correction, equivalent longitudinal drift force correction, equivalent lateral drift force correction, equivalent yaw moment correction, or a combination thereof. The steady-state component is used to characterize the system deviation that persists within the observation window, and the transient component is used to characterize the local disturbance or short-term deviation within the observation window. The transient component decays in the future prediction segment according to the prediction time or prediction distance.

[0012] Furthermore, the difference between the residual measurement value and the current equivalent disturbance estimate is used as the innovation quantity, and the innovation quantity is limited to the range determined by the standard deviation of observation noise, the dispersion of historical residuals, or the preset engineering threshold. The amplitude limiting ratio is statistically defined as the proportion of the number of measurement points within the statistical observation window that are limited by innovation or correction to the number of effective measurement points, and the proportion is used as the input for the calculation of the reliability scalar.

[0013] Furthermore, the prediction time or prediction distance decay is determined based on the prediction time, arc length distance, or straight-line distance from the prediction point to the end of the observation window, and a faster decay rate is applied to the transient components of the normal channel and the yaw / curvature channel than to the steady-state components.

[0014] Furthermore, the bridge-vehicle collision risk parameters determine whether there is a possibility of entering a dangerous area, crossing a navigation channel boundary, or contacting a bridge pier or collision protection facility by geometrically intersecting the corrected future drift trajectory with the outer contour of the bridge pier, the boundary of the collision protection facility, the boundary of the navigation channel, the radius of the danger zone, or the safe buffer distance. The risk level is mapped to multiple warning levels based on one or more of the following: the distance between the impact point or the nearest point and the geometric boundary of the bridge area, the impact speed, the impact angle, the impact posture, the minimum distance from the bridge, and the reliability scalar.

[0015] This invention also provides a system for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification. The correction system is used to execute the above method, and the system includes: The observation interface is used to receive radar, visual, AIS, BeiDou / GNSS or multi-source fusion positioning data and form an online position observation sequence that includes at least the planar position coordinates of the out-of-control vessel. The bridge area geometric parameter storage unit is used to store one or more of the following bridge area geometric parameters: pier coordinates, collision avoidance facility boundary, navigation channel boundary, danger zone radius, and safety buffer distance; The model playback module is used to perform dynamic model playback within the observation window and output the model playback trajectory, taking the starting state of the observation window as the initial condition. The residual decomposition module is used to calculate the residual sequence between the observed trajectory and the model playback trajectory, and decompose the residual sequence into the tangential channel, the normal channel, and the yaw / curvature channel; The equivalent disturbance identification module is used to identify equivalent disturbances based on the residual measurement values ​​of each channel, and perform innovative truncation, correction amount limiting, limiting ratio statistics, reliability scalar calculation, and prediction time or prediction distance attenuation processing. The correction prediction module is used to inject the processed equivalent disturbance into the dynamic model and generate the corrected future drift trajectory, with the state at the end of the observation window as the initial state for prediction. The risk output module is used to output corrected trajectory, uncorrected trajectory, reliability scalar, amplitude limit ratio, risk level, and bridge-vehicle collision risk parameters.

[0016] Compared with existing technologies, the advantages of this invention are as follows: Compared with simply extrapolating the initial dynamic model, this invention enables the systematic bias of the model to be identified within the same time window by replaying the dynamic model within the observation window and calculating the residual sequence. Compared with filtering methods that only update the current state, this invention decomposes the residuals within the observation window into tangential, normal, and yaw / curvature channels, and transforms them into equivalent perturbations that can be injected into the dynamic model, which is beneficial for correcting persistent biases in future prediction segments.

[0017] This invention reduces the impact of abnormal observation points or short-term noise on the equivalent disturbance identification results by innovatively truncating, limiting the correction amount, and statistically analyzing the limiting ratio. By adjusting the retention ratio of the equivalent disturbance in the future prediction segment through reliability scalar and prediction time or prediction distance attenuation, it reduces the impact of local biases in short windows on long-distance prediction, ensures the stability of the prediction correction process, and avoids over-correction caused by short-term observation anomalies.

[0018] This invention performs intersection determination between the corrected future drift trajectory and the geometric boundary of the bridge area, and can output bridge collision risk parameters such as impact point, impact speed, impact attitude, impact angle, minimum distance from the bridge and risk level, which facilitates bridge area early warning and engineering handling, and can provide complete decision support parameters for emergency response to bridge area ship collision risks. Attached Figure Description

[0019] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of a method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification, as described in this invention. Figure 2 This is a flowchart of the equivalent perturbation identification and correction prediction process described in this invention; Figure 3 This is a schematic diagram illustrating the observed growth and predicted convergence described in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other, and the described embodiments are only some embodiments of the present invention, not all embodiments.

[0021] See Figure 1-3 This embodiment describes a method for predicting and correcting the drift trajectory of an out-of-control vessel based on online position observation and equivalent disturbance identification. This method can be executed by a bridge area early warning platform, a shore-based monitoring system, a vessel traffic management system, or an independent computing device. The specific implementation process is as follows: Radar, vision equipment, AIS receivers, or BeiDou / GNSS receivers are deployed upstream of the bridge area. After an out-of-control vessel enters the observation range upstream of the bridge area, the system continuously receives its position coordinates, forming an online position observation sequence. The online position observation sequence includes at least the planar position coordinates of the out-of-control vessel in the bridge area or restricted waters; when the observation equipment can provide the heading angle, the online position observation sequence also includes the heading angle; when the heading angle cannot be provided, the system estimates the motion direction and yaw trend based on the tangential direction of the position trajectory and local curvature. The system standardizes the observation timestamps and can smooth the position sequence to reduce the impact of isolated noise points on velocity and curvature reconstruction. The reconstructed body axis velocity and yaw rate are used as state inputs or constraint information for model playback and future prediction. When only position coordinates are obtained, the motion direction, drift velocity, acceleration trend, trajectory curvature, and yaw trend are estimated based on adjacent position points or locally fitted trajectories; when the heading angle is obtained, it is used to constrain or correct the reconstructed body axis velocity and yaw rate results.

[0022] The online location observation sequence can be represented as: When the heading angle is obtained, the observation can be expanded to: When the heading angle is not obtained, the direction of motion can be estimated from adjacent position points or locally fitted trajectories: Given a heading angle In this case, the body axis velocity can be obtained from the global velocity projection: In the formula, This represents the planar position vector at the k-th observation time. and These are the vertical and horizontal coordinates in a unified planar coordinate system, respectively. This represents the k-th observation time point, and N represents the total number of measurement points in the observation sequence; This represents an observation vector that includes position and heading angle. The k-th observed heading angle; This represents the tangential direction angle of the trajectory obtained from adjacent position points or locally fitted trajectories, and atan2 is the arctangent function that determines the direction angle based on the longitudinal and lateral coordinate increments. This represents the magnitude of the global planar velocity in the k-th segment. = - Indicates the time interval between adjacent observations; and These represent the longitudinal velocity and the lateral velocity in the ship's coordinate system, respectively. Indicates yaw rate, This indicates the heading angle used for velocity projection.

[0023] The system selects an observation window from the online position observation sequence. The observation window is a continuous subsequence of observations within the online position observation sequence, used to perform model playback, residual calculation, and equivalent disturbance identification within the same time range or the same observation point range. The system also performs time alignment on the observation time points, position coordinates, and optional heading angles within the window. Depending on the sampling method, noise level, and motion stability of the observation data, the observation window can be selected using any of the following methods: fixed time window, fixed number of measurement points window, sliding window, distance window, or an adaptive window based on noise level and motion stability.

[0024] To avoid confusion between different window selection methods, let the set of measurement point indices for the j-th observation window be denoted as . Depending on the window construction method, It can be determined by any of the following methods.

[0025] When a fixed time window is used, the observation window is denoted as... Its definition is: When a fixed number of measuring points is used, the observation window is denoted as... Its definition is: in, The window duration. At the current observation endpoint, in the expression for a fixed number of measuring points within a window, M represents the number of measuring points in the window. Indicates the current observation terminal measurement point number. This represents the set of measurement point indices within the j-th observation window.

[0026] In the formula, k represents the measurement point number in the online location observation sequence. This represents the observation time corresponding to the k-th measuring point. Indicates the current observation end time. Indicates the current observation terminal measurement point number. M represents the window duration of a fixed time window, and M represents the number of measurement points included in the fixed measurement point window. This represents the set of measurement point indices selected in the j-th observation window based on time length. This represents the set of measurement point indices selected for the j-th observation window based on the number of measurement points. In practice, this can be selected according to the characteristics of the observation data. = or = .

[0027] When using a sliding window, as new observation points enter, the window's end point is updated from the previous moment to the current observation end point, and the window's starting point moves forward or is redefined accordingly. When using a distance window, the window can select a continuous range of measurement points whose cumulative trajectory arc length reaches a preset distance range. When using an adaptive window, the window length can be dynamically adjusted based on the observation noise level, measurement point missingness, velocity variation amplitude, or trajectory curvature stability. All of these different windowing methods are used to determine the effective set of observation points for model playback and residual identification.

[0028] The effective length and supported range of a window can be represented as: In the formula, This indicates the arc length or cumulative displacement length of the covered trajectory within the observation window. Indicates the starting point number of the observation window. This represents the support range coefficient of the observation window for the trajectory segment to be predicted. Indicates the length of the trajectory segment to be predicted or its engineering equivalent distance; The closer it is to 1, the more fully the observed trajectory length is relative to the predicted segment.

[0029] After selecting an observation window, the system uses the initial state of the observation window as the initial condition and calls the initial dynamic model to perform dynamic model playback within the same observation window, obtaining the model playback trajectory corresponding to the observation time point. The model playback trajectory is the result of the initial dynamic model reproducing the motion that has occurred within the same window, aligned with the observation time point, and may include one or more state variables such as position, heading angle, body axis velocity, and yaw rate. By performing model playback within the same observation window, the explanatory power of the initial dynamic model for the motion that has occurred can be compared, and this difference can be used as a basis for future prediction corrections.

[0030] The initial dynamic model, when replayed within the same observation window, can be represented as: In the formula, The state vector representing the initial dynamic model during the playback phase can include planar position, heading angle, body axis velocity, and yaw rate. This represents the initial dynamic model without injected equivalent disturbance; This indicates input from wind, airflow, or other external environments; it can also be the default environment input. Indicates the initial dynamic model parameters; Indicates the start time of the observation window. This indicates the starting state of the window reconstructed from the observation sequence; This represents the mapping matrix used to extract planar positions from the state vector. Representing the observation time Aligned model playback planar position.

[0031] After obtaining the model playback trajectory, the system calculates the residual sequence between the observed trajectory and the model playback trajectory. The residual sequence serves as the input for subsequent equivalent disturbance identification and is used to characterize the comprehensive model bias exposed within the observation window. The residual sequence includes one or more of the following: position residual, velocity residual obtained by differentiating the position residual, and yaw / curvature residual obtained by bow angle or trajectory curvature.

[0032] The position residual, velocity residual, and yaw / curvature residual can be expressed as follows: The residual fitting loss used to evaluate the playback bias within the same window can be defined as: In the formula, This represents the position residual of the k-th measuring point. This represents the velocity residual obtained by differentiating the residuals of adjacent positions. This represents the heading residual between the observed heading angle and the heading angle from the model playback. represents the curvature residual between the observed trajectory curvature and the model playback trajectory curvature; wrap represents the angle wrapping function that normalizes the angle difference to a preset angle range. and These represent the curvature of the observed trajectory and the curvature of the model playback trajectory, respectively. This represents the residual fitting loss. This represents the weight of the k-th measurement point. This represents the Euclidean norm of the position residual vector.

[0033] After obtaining the residual sequence, the system projects or decomposes it into tangential, normal, and yaw / curvature channels to obtain residual measurements for each channel. The tangential channel is established along the local motion direction or the tangential direction of the trajectory, used to characterize the velocity scale deviation along the motion direction; the normal channel is established perpendicular to the local motion direction or the tangential direction of the trajectory, used to characterize the lateral drift deviation; the yaw / curvature channel is formed by the variation trend of the bow angle difference, curvature difference, or normal residual with distance, used to characterize the yaw response or trajectory curvature trend. As needed, the residual sequence can be further decomposed into one or more of the following: initial state deviation, velocity scale deviation, lateral drift deviation, and curvature deviation.

[0034] Let the local tangential unit vector and the normal unit vector be respectively and ,but: The bow roll / curvature channel residual can be given by one of the following formulas: ∈{ , , } The residual can be decomposed into a deviation vector: In the formula, Let represent the unit vector along the tangential direction of the trajectory at the k-th measurement point. Indicates and The perpendicular normal unit vector; and These represent the projection values ​​of the position residual onto the tangential and normal channels, respectively. This represents the measured value of the bow roll / curvature channel residual, which can be obtained from the bow residual. Curvature residual or normal residual The rate of change of the arc length s along the trajectory is determined by one of three calculation methods; s represents the arc length coordinate measured along the observed trajectory or the model playback trajectory; This represents the bias vector obtained from the residual decomposition. , , and These represent the initial state deviation, velocity scale deviation, lateral offset deviation, and curvature deviation, respectively.

[0035] The system identifies equivalent disturbances based on the residual measurements of each channel. These equivalent disturbances are dynamic model correction inputs derived from the residual sequence and characterize the combined deviations caused by airflow errors, model parameter biases, initial state biases, and unmodeled disturbances. Equivalent disturbances include tangential, normal, and yaw / curvature equivalent disturbances. The normal and yaw / curvature equivalent disturbances each include steady-state and transient components. The steady-state components characterize persistent system biases within the observation window, while the transient components characterize local disturbances or short-term biases within the observation window. Equivalent disturbances can be injected into the dynamic model as equivalent velocity corrections, equivalent longitudinal drift force corrections, equivalent lateral drift force corrections, equivalent yaw moment corrections, or combinations thereof.

[0036] For any channel c∈{T,N,R}, the equivalent perturbation can be identified using a recursive form: In the formula, c represents the residual decomposition channel, and T, N and R represent the tangential channel, normal channel and yaw / curvature channel, respectively; This represents the estimated equivalent disturbance value of the c-th channel after the update at the k-th measurement point. This represents the estimated equivalent disturbance value of the c-th channel at the previous measurement point; This represents the residual measurement value of channel c. This represents the channel observation coefficient that maps the equivalent perturbation estimate to the residual measurement dimension; This represents the amount of innovation in channel c; This represents the innovation cutoff threshold for the c-th channel; This indicates the update gain for the c-th channel; and These represent the lower and upper limits of the allowable correction amount for the c-th channel, respectively.

[0037] The corrected input vector, composed of the equivalent perturbations of each channel, can be written as: When injected into the dynamic model in the form of equivalent flow velocity or equivalent force, it can be further written as In the formula, This represents the corrected input vector composed of the equivalent perturbations of each channel. , and These are the equivalent disturbances in the tangential, normal, and bow roll / curvature channels, respectively. This represents the equivalent flow rate correction. , and These represent the equivalent longitudinal drift force correction, equivalent lateral drift force correction, and equivalent yaw moment correction, respectively, when referred to in the dynamic model.

[0038] The system performs innovation truncation, correction amplitude limiting, amplitude limiting ratio statistics, and reliability evaluation on the correction amount corresponding to the equivalent disturbance. Innovation truncation is used to limit the abnormal innovation amount between the residual measurement value and the current equivalent disturbance estimate. Specifically, the difference between the residual measurement value and the current equivalent disturbance estimate is used as the innovation amount, and the innovation amount is limited to a range determined by the standard deviation of observation noise, the dispersion of historical residuals, or a preset engineering threshold. Correction amplitude limiting is used to constrain the amplitude of the dynamic correction input. Amplitude limiting ratio statistics are used to characterize the proportion of measurement points that are truncated or limited within the observation window. Specifically, the proportion of measurement points that are innovatively truncated or limited by correction amount within the observation window is counted to the total number of valid measurement points, and this proportion is used as the input for calculating the reliability scalar.

[0039] The limiting ratio can be defined as: in, The number of valid measurement points within the observation window that trigger innovative truncation or correction amplitude limiting. This represents the number of valid measurement points within the observation window.

[0040] The reliability scalar is calculated based on at least two of the following: residual fitting loss, observation window support range, correction threshold, and observation window length. It is used to adjust the retention rate of equivalent perturbations in future prediction segments. The reliability scalar ranges from 0 to 1; a higher reliability scalar increases the retention rate of equivalent perturbations in future prediction segments, while a lower reliability scalar decreases the retention rate of equivalent perturbations and accelerates prediction time or prediction distance decay.

[0041] The reliability scalar can be constructed in the following form: The prediction time or prediction distance decay factor can be expressed as: In the formula, This indicates the proportion of measurement points within the observation window that trigger innovation truncation or correction amplitude limiting. This indicates the number of valid measurement points that have been truncated or limited. This indicates the number of valid measurement points within the observation window; R represents a reliability scalar, with a value ranging from 0 to 1. , , , and This represents the preset weights or calibration coefficients used to construct the reliability scalar; This represents the attenuation factor of the c-th channel at the predicted distance s. This represents the attenuation characteristic length of the c-th channel; This represents the equivalent perturbation of the c-th channel of the dynamic model that is retained and injected at the predicted distance s.

[0042] After processing the equivalent disturbance, the system uses the state at the end of the observation window as the initial prediction state. The equivalent disturbance, after correction limiting, reliability evaluation, and prediction time or distance attenuation, is injected into the dynamic model to recursively generate the corrected future drift trajectory. The prediction time or distance attenuation is determined based on the prediction time, arc distance, or straight-line distance of the prediction point relative to the end of the observation window. For the transient components of the normal channel and the yaw / curvature channel, a faster attenuation than the steady-state components can be set to reduce the impact of local short-term deviations on long-distance prediction.

[0043] The modified predictive dynamics model can be expressed as: in, , and These represent the mapping relationships between the tangential, normal, and yaw / curvature channel equivalent disturbance injection dynamic models, respectively. This is the corrected future drift trajectory.

[0044] In the formula, This represents the dynamic state vector during the correction and prediction phase. Indicates the time at the end of the observation window. This indicates the end state of the window reconstructed from the online observation sequence; , and These represent the mapping relationships of the equivalent disturbance injection dynamics models for tangential, normal, and yaw / curvature channels, respectively; , and These represent the prediction corrections for the three channels after reliability evaluation and predicted distance attenuation, respectively. Indicates by The extracted corrected future drift trajectory.

[0045] After generating the corrected future drift trajectory, the system outputs bridge collision risk parameters based on the intersection relationship between the corrected future drift trajectory and the geometric boundaries of the bridge area. The geometric boundaries of the bridge area include one or more of the following: the outer contour of the bridge pier, the boundary of the anti-collision facility, the boundary of the navigation channel, the radius of the danger zone, and the safe buffer distance; the bridge collision risk parameters include one or more of the following: impact point, impact velocity, impact attitude, impact angle, minimum distance from the bridge, and risk level.

[0046] The bridge collision risk parameters can be determined by the minimum distance between the corrected future drift trajectory and the geometric boundary of the bridge area, and the intersection conditions. Let G_bridge be the geometric set of the outer contour of the bridge pier, the boundary of the collision protection facility, the boundary of the navigation channel, or the danger zone. Let P(t) be the position of the corrected future drift trajectory at the predicted time t, and V(t) be the velocity at the predicted time t. t belongs to the future prediction time interval [t_b, t_b+T_pred], where t_b is the end time of the observation window, and T_pred is the prediction duration.

[0047] The minimum distance between the predicted trajectory and the geometric boundary of the bridge area can be expressed as: The predicted time when this minimum distance is reached can be expressed as: In the formula, dist(P(t),G_bridge) represents the minimum geometric distance from the predicted trajectory point P(t) to the geometric set G_bridge of the bridge area; d_min represents the minimum distance between the corrected future drift trajectory and the geometric boundary of the bridge area within the predicted time interval; t* represents the prediction time at which this minimum distance is achieved.

[0048] When d_min is less than the preset safety buffer distance, or when the predicted trajectory P(t) intersects with G_bridge at a certain predicted time, a bridge-ship collision risk is determined, and the impact point, impact velocity, impact attitude, and impact angle at time t* are output. The impact point can be represented as P(t*), the impact velocity as V(t*), and the impact angle as: In the formula, n_bridge(P(t*)) represents the outward normal or collision normal of the bridge zone boundary at the impact point P(t*), and angle(V(t*),n_bridge(P(t*))) represents the angle between the predicted velocity direction and the normal of the bridge zone boundary.

[0049] Risk level can be expressed as: In the formula, Φ is the preset risk level mapping function, and R is the reliability scalar. The risk level can be mapped to multiple warning levels based on the minimum distance d_min, impact velocity V(t*), impact angle θ_imp, and reliability scalar R.

[0050] Combination Figure 3 The diagram showing the observation growth and prediction convergence illustrates the convergence characteristics of this method through the prediction results under different observation ratios. Figure 3 The dashed line represents the true / reference full trajectory, the thick solid line represents the observed first segment, the thin solid line represents the predicted second segment after correction, and the black dot represents the current observation end. As the observation ratio increases from 0.10 to 0.90, the trajectory information covered by the observation window gradually increases, and the constraints of the model playback residual on velocity scale bias, lateral offset bias, and curvature bias gradually strengthen, thus the equivalent perturbation identification results tend to stabilize.

[0051] The observation ratio in this example can be denoted as: in, Indicates the proportion of observations. The arc length of the observed trajectory. For the reference full trajectory arc length. For each observation scale The system uses the current observation endpoint as the initial prediction point, performs same-window model playback, residual decomposition, equivalent disturbance identification, and corrective prediction to obtain the predicted trajectory. .

[0052] Position error and heading error can be expressed as follows: The overall prediction accuracy score can be expressed in a normalized form: in, This indicates that when the observation ratio is q, the corrected predicted trajectory reaches the evaluation endpoint. The location, This indicates the position of the reference full trajectory at the same evaluation endpoint; and These represent the heading angles of the corrected predicted trajectory and the reference trajectory at the evaluation endpoint, respectively. Indicates the error in the endpoint position. The endpoint heading error is represented by TPAS(q); TPAS(q) represents the overall prediction accuracy score. and These represent the weights of the position error term and the heading error term, respectively, with π used to normalize the heading angle error.

[0053] The TPAS mentioned above is the trajectory prediction accuracy score defined in this embodiment, which is used to quantify the overall closeness of the corrected predicted trajectory to the reference trajectory under different observation ratios, and is not limited to a fixed index recognized in the field.

[0054] Combination Figure 3 As shown in the example, when the observation scale q=0.10, the TPAS is approximately 0.447, the position error is approximately 2633.1 mm, and the heading error is approximately 64.75 degrees; when q=0.30, the TPAS is approximately 0.824, the position error is approximately 1037.0 mm, and the heading error is approximately 21.19 degrees; when q=0.50, the TPAS is approximately 0.867, the position error is approximately 1192.2 mm, and the heading error is approximately 1. 3.39 degrees; when q=0.70, TPAS is approximately 0.940, position error is approximately 140.5 mm, and heading error is approximately 8.24 degrees; when q=0.80, TPAS is approximately 0.967, position error is approximately 102.9 mm, and heading error is approximately 4.50 degrees; when q=0.90, TPAS is approximately 0.977, position error is approximately 43.1 mm, and heading error is approximately 3.08 degrees.

[0055] As the above example demonstrates, with an increase in the observation ratio, the corrected predicted trajectory becomes generally closer to the true / reference full trajectory, TPAS improves overall, and position and heading errors decrease overall. However, due to the influence of observation noise, local variations in trajectory curvature, and residual decomposition channel coupling, individual errors at certain observation ratios may exhibit local fluctuations; for example, the position error at q=0.50 is higher than at q=0.30. Therefore, this example is used to illustrate the overall convergence and improved stability of the predicted trajectory after an increase in online observations, and does not imply that all error indices decrease strictly monotonically under all operating conditions.

[0056] The embodiments of the present invention are not limited to the above description. Various alternative embodiments can be adopted without departing from the core technology of the present invention. The observation window can be selected using a fixed time, a fixed number of measurement points, a sliding distance, or an adaptive method; residual decomposition can be based on the tangential direction of the observation trajectory or the tangential direction of the model playback trajectory; equivalent disturbance identification can be achieved using recursive updates, weighted least squares, constrained least squares, or piecewise fitting; prediction time or prediction distance attenuation can use exponential attenuation, linear attenuation, piecewise attenuation, or attenuation based on the distance to the risk area; risk levels can be mapped using level two, level three, level five, or other engineering early warning levels. None of the above alternative embodiments change the core technology of the present invention, namely, the combined relationship between the online position observation sequence, observation window, model playback trajectory, residual sequence, three-channel residual decomposition, equivalent disturbance identification, correction amplitude limiting, reliability scalar, prediction time or prediction distance attenuation, dynamic model correction prediction, and bridge-ship collision risk parameter output.

[0057] This embodiment also provides a system for predicting and correcting the drift trajectory of an out-of-control vessel based on online position observation and equivalent disturbance identification, used to execute the above method steps. The system includes an observation interface, a processor and memory, a bridge area geometric parameter storage unit, a model playback module, a residual decomposition module, an equivalent disturbance identification module, a correction prediction module, a risk output module, and a risk display or alarm output interface.

[0058] The observation interface is used to receive radar, visual, AIS, BeiDou / GNSS or multi-source fusion positioning data, and to form an online position observation sequence that includes at least the planar position coordinates of the out-of-control vessel; The bridge area geometric parameter storage unit is used to store one or more of the following bridge area geometric parameters: pier coordinates, collision avoidance facility boundary, navigation channel boundary, danger zone radius, and safety buffer distance; The model playback module is used to perform dynamic model playback within the observation window and output the model playback trajectory, taking the starting state of the observation window as the initial condition. The residual decomposition module is used to calculate the residual sequence between the observed trajectory and the model playback trajectory, and decompose the residual sequence into the tangential channel, normal channel and yaw / curvature channel; The equivalent disturbance identification module is used to identify equivalent disturbances based on the residual measurement values ​​of each channel, and to perform innovative truncation, correction amount limiting, limiting ratio statistics, reliability scalar calculation, and prediction of time or distance attenuation. The correction prediction module is used to inject the processed equivalent disturbance into the dynamic model and generate the corrected future drift trajectory, with the state at the end of the observation window as the initial state for prediction. The risk output module and the risk display or alarm output interface are used to output the corrected trajectory, uncorrected trajectory, reliability scalar, amplitude limiting ratio, risk level, and bridge-vehicle collision risk parameters. The risk display or alarm output interface simultaneously outputs the corrected trajectory, uncorrected trajectory, reliability scalar, amplitude limiting ratio, bridge area geometric boundary overlay display results, and risk level, for display, recording, or alarm triggering by the bridge area early warning platform.

[0059] The memory stores instructions that can be executed by the processor, and the processor executes the instructions to implement the steps of the above method.

[0060] The specific embodiments of the present invention disclosed above are merely illustrative of the invention. These embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification, characterized in that: Includes the following steps: Obtain the online position observation sequence of the out-of-control vessel. The online position observation sequence shall include at least the planar position coordinates of the out-of-control vessel in the bridge area or restricted waters. When the bow angle can be obtained, the online position observation sequence shall also include the bow angle. The body axis velocity and yaw rate shall be reconstructed from the position coordinates and / or the bow angle. Select an observation window from the online position observation sequence and align the observation time points, position coordinates, and optional heading angles within the observation window with time. Using the initial state of the observation window as the initial condition, the initial dynamic model is called to perform dynamic model playback within the same observation window to obtain the model playback trajectory corresponding to the observation time point. Calculate the residual sequence between the observation trajectory formed by the online location observation sequence within the observation window and the model playback trajectory; The residual sequence is projected or decomposed into the tangential channel, normal channel, and yaw / curvature channel to obtain the residual measurement values ​​for each channel; Based on the residual measurement values ​​of each channel, the equivalent disturbance is identified, and the correction amount corresponding to the equivalent disturbance is innovatively truncated, the correction amount is limited, and the limit ratio is statistically analyzed. The reliability scalar is calculated based on at least two of the following: residual fitting loss, observation window support range, correction amount limiting ratio, and observation window length. The correction amount is then subjected to prediction time or prediction distance attenuation based on the reliability scalar. Using the state at the end of the observation window as the initial state for prediction, the processed equivalent disturbance is injected into the dynamic model to recursively generate the corrected future drift trajectory. Based on the intersection relationship between the corrected future drift trajectory and the bridge piers, anti-collision facilities, navigation channel boundaries or pre-set danger zones, bridge collision risk parameters are output. The bridge collision risk parameters include at least three of the following: impact point, impact speed, impact attitude, impact angle, minimum distance from the bridge, and risk level.

2. The method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification according to claim 1, characterized in that: The observation window can be selected in any of the following ways: fixed time window, fixed number of measurement points window, sliding window, distance window, or adaptive window based on noise level and motion stability.

3. The method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification according to claim 1, characterized in that: When only position coordinates are obtained, the direction of motion, drift velocity, acceleration trend, trajectory curvature, and yaw trend are estimated based on adjacent position points or locally fitted trajectories; when the heading angle is obtained, the heading angle is used to constrain or correct the reconstructed body axis velocity and yaw angular velocity.

4. The method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification according to claim 1, characterized in that: The residual sequence includes one or more of the following: position residual, velocity residual obtained by differentiating the position residual, and yaw / curvature residual obtained by bow angle or trajectory curvature; Projecting or decomposing the residual sequence into the tangential, normal, and yaw / curvature channels includes: establishing tangential and normal unit vectors based on the local tangential direction of the observed trajectory or model playback trajectory; projecting the position residuals into tangential and normal residuals; and generating yaw / curvature channel residual measurements based on the variation trends of the yaw angle difference, curvature difference, or normal residuals with distance.

5. The method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification according to claim 4, characterized in that: The residual sequence is further decomposed into one or more of the following: initial state deviation, velocity scale deviation, lateral offset deviation, and curvature deviation.

6. The method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification according to claim 1, characterized in that: The equivalent disturbance includes tangential equivalent disturbance, normal equivalent disturbance, and yaw / curvature equivalent disturbance. The normal equivalent disturbance and the yaw / curvature equivalent disturbance contain steady-state components and transient components, respectively. The equivalent disturbance is injected into the dynamic model in the form of equivalent velocity correction, equivalent longitudinal drift force correction, equivalent lateral drift force correction, equivalent yaw moment correction, or a combination thereof. The steady-state component is used to characterize the system deviation that persists within the observation window, and the transient component is used to characterize the local disturbance or short-term deviation within the observation window. The transient component decays in the future prediction segment according to the prediction time or prediction distance.

7. The method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification according to claim 1, characterized in that: The difference between the residual measurement value and the current equivalent disturbance estimate is used as the innovation quantity, and the innovation quantity is limited to the range determined by the standard deviation of observation noise, the dispersion of historical residuals, or the preset engineering threshold. The amplitude limit ratio is statistically defined as the proportion of the number of measurement points within the observation window that are limited by innovation or correction to the number of effective measurement points, and the proportion is used as the input for the calculation of the reliability scalar.

8. The method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification according to claim 1, characterized in that: The prediction time or prediction distance decay is determined based on the prediction time, arc distance, or straight-line distance from the prediction point to the end of the observation window, and a faster decay rate is applied to the transient components of the normal channel and the yaw / curvature channel than to the steady-state components.

9. The method for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification according to claim 1, characterized in that: Bridge collision risk parameters are determined by geometric intersection of the corrected future drift trajectory with the outer contour of the bridge pier, the boundary of the collision protection facility, the boundary of the navigation channel, the radius of the danger zone, or the safe buffer distance. This determines whether there is a possibility of entering the danger zone, crossing the boundary of the navigation channel, or contacting the bridge pier and collision protection facility. The risk level is mapped to multiple warning levels based on one or more of the following: the distance between the impact point or the nearest point and the geometric boundary of the bridge area, the impact speed, the impact angle, the impact attitude, the minimum distance from the bridge, and the reliability scalar.

10. A system for predicting and correcting the drift trajectory of an out-of-control ship based on online position observation and equivalent disturbance identification, characterized in that: The correction system is used to perform the method according to any one of claims 1-9, the system comprising: The observation interface is used to receive radar, visual, AIS, BeiDou / GNSS or multi-source fusion positioning data and form an online position observation sequence that includes at least the planar position coordinates of the out-of-control vessel. The bridge area geometric parameter storage unit is used to store one or more of the following bridge area geometric parameters: pier coordinates, collision avoidance facility boundary, navigation channel boundary, danger zone radius, and safety buffer distance; The model playback module is used to perform dynamic model playback within the observation window and output the model playback trajectory, taking the starting state of the observation window as the initial condition. The residual decomposition module is used to calculate the residual sequence between the observed trajectory and the model playback trajectory, and decompose the residual sequence into the tangential channel, the normal channel, and the yaw / curvature channel; The equivalent disturbance identification module is used to identify equivalent disturbances based on the residual measurement values ​​of each channel, and perform innovative truncation, correction amount limiting, limiting ratio statistics, reliability scalar calculation, and prediction time or prediction distance attenuation processing. The correction prediction module is used to inject the processed equivalent disturbance into the dynamic model and generate the corrected future drift trajectory, with the state at the end of the observation window as the initial state for prediction. The risk output module is used to output corrected trajectory, uncorrected trajectory, reliability scalar, amplitude limit ratio, risk level, and bridge-vehicle collision risk parameters.

Images