Unmanned ship control system and method based on state prediction

Abstract

The invention discloses an unmanned ship control system and method based on state prediction, and relates to the technical field of control. The system comprises a flight state characteristic modeling module and the like, and the method comprises the steps of S1 to S5, calculating a predicted thrust course in combination with waves and shallow water effects, calculating a track offset matrix according to course correction and flow velocity, calculating a rudder angle and thrust demand to generate a prediction control sequence, executing feed-forward compensation and thrust injection in combination with a real-time rolling angle, and outputting a prediction control quantity; the method comprises the following steps: constructing a feature set by fusing a rolling angular velocity and a navigational speed change rate, improving prediction precision, correcting a thrust course through wave resistance and a shallow water effect, enhancing environmental immunity, generating a space-time migration matrix in combination with ocean current data, quantifying a track gradient, realizing high-precision prediction, processing rudder angle compensation and thrust increment in a time sequence cooperative manner, and improving the prediction precision. A standardized control sequence is generated, out-of-step is eliminated, feed-forward compensation and dynamic increment are combined for real-time monitoring, accurate regulation and control are achieved, delay is reduced, and real-time stability of track regression is ensured.

Description

Technical Field

[0001] The present invention relates to the field of control technology, and in particular to an unmanned ship control system and method based on state prediction. Background Art

[0002] Currently, the field of intelligent control technology encompasses a collection of technologies for dynamic behavior analysis and real-time decision-making in autonomous systems. Its core focus is on enabling adaptive system adjustments through multi-source data fusion and environmental perception. This area integrates automatic control theory, artificial intelligence algorithms, sensor networks, and actuator collaboration mechanisms, systematically covering the entire chain from data acquisition to decision-making output. It focuses on balancing system stability and response speed in complex environments, particularly in the area of ​​unmanned vehicles, which must cope with dynamic interference conditions and track and execute missions.

[0003] The state-prediction-based unmanned vessel control system is a technical solution that constructs a hull kinematic model and an environmental disturbance model, combining real-time sensor data to probabilistically predict the vessel's future motion state. For three technical issues: navigation path deviation correction, power unit parameter matching, and environmental disturbance compensation, time series analysis is used to predict the changing trends of the hull's position and attitude. Kalman filtering is used to integrate inertial navigation and satellite positioning data, and closed-loop adjustments to the propeller output power and rudder angle control are driven based on the predicted results. The specific implementation relies on multi-threaded embedded controllers and edge computing nodes to complete model calculations and command distribution.

[0004] In the existing technology, due to the insufficient adaptability of thrust and heading correction, the hull position prediction lacks the ability to model the spatiotemporal gradient, making it difficult to cope with the nonlinear offset caused by sudden changes in ocean currents. The control command generation ignores the timing coordination of the rudder angle and thrust increment, the actuator response out of step aggravates the path deviation, real-time control relies on closed-loop feedback, lacks feedforward compensation and dynamic incremental intervention, and the track correction lag affects mission efficiency.

[0005] Therefore, low prediction accuracy and poor stability have become technical problems that need to be solved urgently. Summary of the Invention

[0006] The present invention provides an unmanned ship control system and method based on state prediction, which solves the technical problems of low prediction accuracy and poor stability.

[0007] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0008] An unmanned ship control system based on state prediction includes a navigation characteristic modeling module, a prediction response generation module, a track deviation prediction module, a prediction collaborative optimization module and a navigation control execution module.

[0009] The flight state feature modeling module is used to obtain the roll angular velocity time series data output by the inertial measurement unit, collect the speed change rate of the Doppler log, perform sliding window integration on the roll angular velocity series to generate a displacement prediction curve, perform differential operation on the speed change rate to generate an acceleration prediction value, and fuse the displacement curve and acceleration value to construct a flight state prediction feature set;

[0010] A prediction response generation module is used to call the navigation state prediction feature set, input the wave model to calculate the theoretical thrust demand and heading angle correction of the propeller, superimpose the shallow water effect correction thrust prediction value, and generate a predicted thrust and heading set;

[0011] A track deviation prediction module is used to calculate the lateral and longitudinal deviations based on the heading correction value of the predicted thrust heading set and the flow velocity direction of the ocean current sensor, and to construct a time-space distribution matrix to generate a track prediction deviation matrix;

[0012] A prediction collaborative optimization module is used to calculate the rudder angle compensation and thrust increment requirements based on the track prediction offset matrix, perform phase alignment and amplitude processing to generate a prediction collaborative control sequence;

[0013] The flight state control execution module is used to call the rudder angle compensation value and thrust increment value of the predicted cooperative control sequence, execute the servo feedforward compensation and inject the thrust instruction, and output the predicted track control amount.

[0014] A further technical solution is that: the flight state prediction feature set includes a roll displacement curve and a linear acceleration prediction value, the predicted thrust heading set includes a fluid resistance correction thrust, a heading compensation angle and a bottom friction compensation thrust, the track prediction offset matrix includes a lateral offset distribution field, a longitudinal offset distribution field and a time-space offset gradient matrix, the predicted collaborative control sequence includes a rudder angle compensation timing quantity, a thrust increment timing quantity, a collaborative phase vector and a collaborative amplitude vector, and the predicted track control quantity includes a feedforward rudder angle quantity, a thrust increment pulse and a track regression drive quantity.

[0015] A further technical solution is that the flight state feature modeling module includes a data acquisition and preprocessing submodule, a dynamic parameter calculation submodule and a feature fusion modeling submodule. The data acquisition and preprocessing submodule is used to obtain the roll angular velocity time series data output by the inertial measurement unit, collect the speed change rate recorded by the Doppler odometer, input the roll angular velocity time series data into the sliding window integrator for time alignment, perform standardized noise reduction processing on the speed change rate, and generate a time series preprocessing set;

[0016] a dynamic parameter calculation submodule, configured to call the roll angular velocity data in the time series preprocessing set, perform integral accumulation operations based on a fixed time window length, generate a displacement increment sequence, synchronously perform differential operator processing on the speed change rate, and output acceleration prediction values ​​and displacement prediction curves;

[0017] The feature fusion modeling submodule is used to extract the longitudinal axis range parameters of the displacement prediction curve and the fluctuation parameters of the acceleration prediction value, weightedly superimpose the two types of parameters according to the timestamp, fuse the displacement and acceleration data according to the preset weight coefficient, and generate a flight state prediction feature set.

[0018] A further technical solution is that the predicted response generation module includes a theoretical parameter calculation submodule, an environmental compensation correction submodule and a feature integration output submodule. The theoretical parameter calculation submodule is used to call the displacement prediction curve in the navigation state prediction feature set, input the fluid resistance response function of the wave kinematic model, perform an integral operation on the ordinate value of the displacement prediction curve, calculate the thrust demand of the propeller in the future period, and simultaneously extract the lateral component of the fluid resistance response function to generate the theoretical thrust demand and the heading correction value.

[0019] an environmental compensation correction submodule, configured to call a bottom friction compensation coefficient of a shallow water effect model based on the amplitude range of the theoretical thrust requirement and the heading correction amount, perform a point-by-point multiplication operation on the bottom friction compensation coefficient and the theoretical thrust requirement, correct the longitudinal component of the thrust prediction value, and generate a corrected thrust prediction value;

[0020] The feature integration output submodule is used to extract the time series of the corrected thrust prediction value and the extreme value parameters of the heading correction amount, align the thrust prediction value and the heading correction amount according to the timestamp, perform a normalized superposition operation, and generate a predicted thrust and heading set.

[0021] A further technical solution is that the track deviation prediction module includes a data fusion processing submodule, an offset calculation submodule and a space-time matrix construction submodule, the data fusion processing submodule is used to call the heading angle theoretical correction value in the predicted thrust heading set, collect the flow velocity direction vector monitored in real time by the ocean current sensor, perform vector superposition operation on the heading angle theoretical correction value and the transverse component of the flow velocity direction vector, and generate a fusion correction vector;

[0022] an offset calculation submodule, configured to perform an integration operation on the time step of each future waypoint based on the lateral projection value and the longitudinal projection value of the fused correction vector, calculate the lateral displacement accumulation amount and the longitudinal displacement accumulation amount, and generate the lateral offset and the longitudinal offset;

[0023] The space-time matrix construction submodule is used to extract the time series of the lateral offset and the amplitude series of the longitudinal offset, arrange them into a two-dimensional grid structure according to the time dimension and space dimension, fill in the corresponding values ​​of the lateral and longitudinal offsets, and generate a track prediction offset matrix.

[0024] A further technical solution is that the prediction collaborative optimization module includes a control demand calculation submodule, a thrust demand calculation submodule and a collaborative sequence generation submodule, wherein the control demand calculation submodule is used to call the lateral offset in the track prediction offset matrix, perform a proportional amplification operation on the absolute value of the lateral offset based on a preset rudder efficiency coefficient, calculate the rudder angle adjustment amplitude per unit time, and generate a rudder angle compensation demand value;

[0025] a thrust demand calculation submodule, configured to extract the longitudinal offset from the track prediction offset matrix, perform an integral operation on the rate of change of the longitudinal offset according to the thrust response curve of the propeller, calculate the thrust increment required to maintain the ship's speed, and generate a thrust increment demand value;

[0026] The collaborative sequence generation submodule is used to align the time stamps of the time series of the rudder angle compensation demand value and the time series of the thrust increment demand value, perform normalization scaling on the amplitudes of the two types of demands, superimpose them into a composite signal according to the phase synchronization principle, and generate a predicted collaborative control sequence.

[0027] A further technical solution is that the flight state control execution module includes an execution parameter calling submodule, a feedforward compensation processing submodule and a control quantity output submodule, the execution parameter calling submodule is used to call the first step rudder angle compensation value in the predictive cooperative control sequence, collect the first step parameter of the thrust increment value, align the rudder angle compensation value and the thrust increment value according to the timestamp, and generate the rudder angle compensation initial value and the thrust increment initial value;

[0028] a feedforward compensation processing submodule, configured to obtain real-time roll angular velocity monitoring data based on the amplitude range of the initial rudder angle compensation value, perform a roll angular velocity inverse proportional correction operation on the initial rudder angle compensation value, and generate a compensated rudder angle value;

[0029] The control quantity output submodule is used to synchronize the correction result of the compensation rudder angle value with the execution timing of the initial value of the thrust increment, send a discrete instruction sequence of the thrust increment value to the thruster, and output the predicted track control quantity.

[0030] A method for controlling an unmanned vessel based on state prediction, used in the above-mentioned unmanned vessel control system based on state prediction, comprises the following steps:

[0031] S1: Obtain the roll angular velocity time series data output by the inertial measurement unit, collect the speed change rate of the Doppler log, input the roll angular velocity time series data into the sliding window integrator to generate a displacement prediction curve, perform a first-order differential operation on the speed change rate to generate an acceleration prediction value, and fuse the displacement prediction curve and the acceleration prediction value to construct a flight state prediction feature set;

[0032] S2: calling the displacement prediction curve in the navigation state prediction feature set, inputting the fluid resistance response function of the wave kinematics model to calculate the theoretical thrust demand and heading angle correction of the propeller, adding the bottom friction compensation coefficient of the shallow water effect model to correct the thrust prediction value, and generating a predicted thrust and heading set;

[0033] S3: Based on the theoretical correction of the heading angle in the predicted thrust heading concentration and the current velocity direction vector monitored in real time by the ocean current sensor, the time series of the lateral position offset and the longitudinal position offset are calculated to construct a track prediction offset matrix;

[0034] S4: Calculating the rudder angle compensation requirement based on the lateral offset in the trajectory prediction offset matrix, calculating the propeller thrust increment requirement based on the longitudinal offset, performing phase alignment and amplitude normalization processing on the two types of requirements in a time series, and generating a predicted cooperative control sequence;

[0035] S5: Call the first-step rudder angle compensation value and thrust increment value in the predicted cooperative control sequence, perform feedforward compensation action on the servo in combination with the real-time roll angular velocity monitoring data, inject thrust prediction increment instructions into the thruster, and output the predicted track control amount.

[0036] The beneficial effects of adopting the above technical solution are:

[0037] An unmanned ship control system based on state prediction includes a navigation feature modeling module, a prediction response generation module, a track offset prediction module, a prediction collaborative optimization module and a navigation control execution module. By fusing the roll angular velocity and the speed change rate to construct a multi-dimensional feature set, the spatiotemporal correlation accuracy of the hull motion state prediction is enhanced; the environmental anti-interference capability of the thrust and heading correction is improved through dynamic correction of wave resistance and shallow water effect; the spatiotemporal offset matrix is ​​generated by combining the real-time data of ocean currents, the track offset gradient is quantified, and high-precision dynamic prediction is achieved; the rudder angle compensation and the thrust increment are processed in a time-series collaborative manner to generate a standardized control sequence, eliminating the actuator response out-of-step; the feedforward compensation and dynamic increment instructions are combined with real-time monitoring data to accurately control the execution action, reduce control delay, and ensure the real-time and stability of the track regression.

[0038] A state prediction-based unmanned vessel control method includes steps S1 to S5, which include obtaining the roll angular velocity and speed change rate, calculating the predicted thrust heading based on wave and shallow water effects, calculating the track offset matrix based on the heading correction and current velocity, calculating the rudder angle and thrust demand to generate a predicted control sequence, performing feedforward compensation and thrust injection based on the real-time roll angle, and outputting the predicted control variable. Prediction accuracy is improved by integrating the roll angular velocity and speed change rate to construct a feature set. The thrust heading is corrected based on wave resistance and shallow water effects to enhance environmental immunity. A spatiotemporal offset matrix is ​​generated based on ocean current data, and the track gradient is quantified to achieve high-precision prediction. Rudder angle compensation and thrust increments are processed in a time-series coordinated manner to generate a standardized control sequence, eliminating step loss. Feedforward compensation and dynamic increments are combined with real-time monitoring to achieve precise control, reduce latency, and ensure real-time and stable track regression. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a principle block diagram of Example 1 of the present application;

[0040] Figure 2 This is the data flow diagram of the flight state feature modeling module in this application;

[0041] Figure 3 A data flow diagram for the predicted response generation module in this application;

[0042] Figure 4 This is the data flow diagram of the track deviation prediction module in this application;

[0043] Figure 5 This is the data flow diagram of the prediction collaborative optimization module in this application;

[0044] Figure 6 This is the data flow diagram of the flight control execution module in this application;

[0045] Figure 7 This is a flowchart of Example 2 of this application. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is in no way intended to limit the present application and its application or use. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0047] In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0048] Example 1:

[0049] like Figure 1 As shown, the present invention discloses an unmanned ship control system based on state prediction, including a navigation characteristic modeling module, a prediction response generation module, a track deviation prediction module, a prediction collaborative optimization module and a navigation control execution module, which are described in detail as follows.

[0050] The flight state feature modeling module is used to obtain the roll angular velocity time series data output by the inertial measurement unit, collect the speed change rate of the Doppler odometer, input the roll angular velocity time series data into the sliding window integrator to generate a displacement prediction curve, perform differential operation on the speed change rate to generate the acceleration prediction value, and fuse the displacement prediction curve and the acceleration prediction value to construct a flight state prediction feature set.

[0051] The prediction response generation module is used to call the displacement prediction curve in the navigation state prediction feature set, input the fluid resistance response function of the wave kinematic model, calculate the theoretical thrust demand and heading angle correction of the propeller in the future period, superimpose the bottom friction compensation coefficient of the shallow water effect model to correct the thrust prediction value, and generate a predicted thrust and heading set that includes environmental response.

[0052] The track deviation prediction module is used to calculate the lateral position offset and longitudinal position offset of future waypoints based on the theoretical correction of the heading angle of the predicted thrust heading concentration and the flow velocity direction vector monitored in real time by the ocean current sensor, construct a spatiotemporal distribution matrix of the offset changing with time, and generate a track prediction deviation matrix.

[0053] The predictive collaborative optimization module is used to calculate the rudder angle compensation requirement based on the lateral offset in the track prediction offset matrix, calculate the propeller thrust increment requirement based on the longitudinal offset, and perform phase alignment and amplitude normalization on the two types of requirements in time series to generate a predictive collaborative control sequence.

[0054] The navigation state control execution module is used to call the first-step rudder angle compensation value and thrust increment value in the predicted collaborative control sequence, perform feedforward compensation action on the servo in combination with real-time roll angular velocity monitoring data, inject thrust prediction increment instructions into the thruster, and output the predicted track control quantity that drives the ship's track to return to the predicted state.

[0055] The flight state prediction feature set includes the roll displacement curve and the linear acceleration prediction value. The predicted thrust heading set includes the fluid resistance correction thrust, the heading compensation angle, and the bottom friction compensation thrust. The track prediction offset matrix is ​​specifically the lateral offset distribution field, the longitudinal offset distribution field, and the time-space offset gradient matrix. The predicted collaborative control sequence includes the rudder angle compensation time series, the thrust increment time series, the collaborative phase vector, and the collaborative amplitude vector. The predicted track control quantity is specifically the feedforward rudder angle quantity, the thrust increment pulse, and the track regression drive quantity.

[0056] like Figure 2 As shown, the flight state feature modeling module includes a data acquisition and preprocessing submodule, a dynamic parameter calculation submodule and a feature fusion modeling submodule.

[0057] The data acquisition and preprocessing submodule is used to obtain the roll angular velocity time series data output by the inertial measurement unit, collect the speed change rate recorded by the Doppler odometer, input the roll angular velocity time series data into the sliding window integrator for time alignment, perform standardized noise reduction on the speed change rate, and generate a time series preprocessing set.

[0058] Obtain the roll angular velocity time series data output by the inertial measurement unit, set the sliding window length to 0.2 seconds, and based on the statistical results of the typical motion cycle experiment of the unmanned ship, use 10 sampling points as the window. For example, if the data sequence within the timestamp 0.0s-0.2s is [0.15, 0.12, 0.18] rad / s, perform the integral accumulation within the window: Eliminate the transmission delay between sensors by aligning the timestamps. For example, if the timestamp deviation between GPS positioning data and IMU data exceeds 10ms, linear interpolation alignment is used. For the speed change rate recorded by the Doppler log, if the original data is [0.32, 0.35, 0.28] m / s 2 ) Perform Z-score normalization: (mean μ = 0.316, standard deviation σ = 0.027), set ±2σ as the noise threshold, interval [0.262, 0.370], and replace the out-of-range data with the mean, such as 0.35m / s 2 Reserved, 0.28m / s 2 Replaced with 0.316m / s 2 The pre-processed data is stored in the experience storage buffer and shared with the environmental change detection module. When the standard deviation within the window continuously exceeds 0.05 rad / s, the environmental status flag is triggered.

[0059] The dynamic parameter calculation submodule is used to call the roll angular velocity data in the time series preprocessing set, perform integral accumulation operations based on a fixed time window length, generate a displacement increment sequence, synchronously perform differential operator processing on the speed change rate, and output the acceleration prediction value and displacement prediction curve.

[0060] Call the roll angular velocity data of the time series preprocessing set, use a fixed window of 0.2 seconds, corresponding to a 5Hz sampling rate, and perform integration operation to generate a displacement increment sequence: If ω=[0.15,0.12,0.18]rad / s, Δt=0.2s, then Δs=0.09m. Perform first-order difference calculation on the speed change rate to calculate the instantaneous acceleration: a t =(v t+1 -v t ) / Δt, such as v t =1.2m / s, v t+1 =1.15m / s, then a t =-0.25m / s 2 . Combined with the information gain calculation, when the absolute error between the acceleration prediction value and the actual observation value exceeds 0.1m / s 2 When the predicted value is -0.25m / s 2 , measured value -0.12m / s 2 , error 0.13m / s 2 , the window is marked as a key experience, the weighting coefficient of its displacement increment sequence is increased to 1.5 times, the default is 1.0, and it is input into the teacher network training framework as a priority sample.

[0061] The feature fusion modeling submodule is used to extract the longitudinal axis range parameters of the displacement prediction curve and the fluctuation parameters of the acceleration prediction value, perform weighted superposition of the two types of parameters according to the timestamp, fuse the displacement and acceleration data according to the preset weight coefficient, and generate a flight state prediction feature set.

[0062] Extract the vertical axis range parameter R of the displacement prediction curve s =max(Δs)-min(Δs), if Δs=[0.09,0.08,0.10]m in the window, then R s =0.02m, calculate the acceleration fluctuation parameters For example, a=[-0.25,-0.18,-0.22]m / s 2 , μ a =-0.217, σ a =0.028m / s 2 According to the adaptive weight adjustment strategy, the initial weight is set to α = 0.6 corresponding to displacement and β = 0.4 corresponding to acceleration. When the environmental change detection unit is triggered, such as three consecutive windows R s >0.15m, dynamically adjust the weights to α′=α·0.8, β′=β·1.2 to enhance the acceleration characteristic response. Fusion eigenvalue F=αR s +βσ a, such as F = 0.6 × 0.02 + 0.4 × 0.028 = 0.023, the input policy optimization unit is used as the state feature vector, and the cosine similarity matching is performed with the teacher network output in the soft target loss calculation, with a threshold of 0.85.

[0063] like Figure 3 As shown, the prediction response generation module includes a theoretical parameter calculation submodule, an environmental compensation correction submodule and a feature integration output submodule.

[0064] The theoretical parameter calculation submodule is used to call the displacement prediction curve in the navigation state prediction feature set, input the fluid resistance response function of the wave kinematic model, perform integral operation on the ordinate value of the displacement prediction curve, calculate the thrust demand of the propeller in the future period, and simultaneously extract the lateral component of the fluid resistance response function to generate the theoretical thrust demand and heading correction.

[0065] Call the displacement prediction curve in the navigation state prediction feature set. For example, when the time window t = 0.4s, the displacement sequence Δs = [0.09, 0.12, 0.15]m, and input the fluid resistance response function R(v) = 0.5·C of the wave kinematic model. d ·ρ·A·v 2 , where the resistance coefficient C d =0.8, based on the experimental calibration value of hull streamline, seawater density ρ = 1025 kg / m 3 , measured in real time by salinity sensor, hull cross-sectional area A=2.5m 2 , calculated based on the hull size. Perform an integral operation on the displacement curve with a window of Δt = 0.2 seconds to calculate the instantaneous velocity: Based on the kinetic equation T req =R(v)+m·a, where the hull mass m = 1200 kg, including the load, and the acceleration a = 0.15 m / s 2 , from the dynamic parameter calculation submodule, substitute: T req =0.5×0.8×1025×2.5×0.6 2 +1200×0.15=738+180=918N, and extract the lateral component of fluid resistance R at the same time lat = 0.3·R(v) = 0.3×738 = 221.4N, calculate the heading correction: T req and θ adj Input the teacher network training framework as the key experience mark, if |θ adj |>10° is marked as a high information gain sample and triggers the weight update of the soft target loss in the policy distillation.

[0066] The environmental compensation correction submodule is used to call the bottom friction compensation coefficient of the shallow water effect model based on the amplitude range of the theoretical thrust demand and the heading correction amount, perform point-by-point multiplication operation on the bottom friction compensation coefficient and the theoretical thrust demand, correct the longitudinal component of the thrust prediction value, and generate a corrected thrust prediction value.

[0067] Based on the theoretical thrust requirement T req =918N and θ adj =13.5°, and invoke the shallow water effect model in the Bayesian uncertainty quantification module. The bottom friction compensation coefficient is calculated as: The critical water depth d crit =3m, set according to the hull draft, real-time water depth d water =2.2m, measured by sonar, and substituted into: Corrected thrust prediction value: T corr =T req ·k fric =918×0.987≈906N, safety is verified by the safety assurance module: If T corr >0.8·T max , T max =1500N is the maximum thrust of the propeller, then press T safe =min(T corr ,0.9·T max ) derating, in this case 906N < 1200N, keep the original value, so the corrected T corr and θ adj Store in the normal buffer and trigger the information gain calculation. If |T corr -T req If |>50N, it is marked as key experience and prioritized for sampling in subsequent student network training.

[0068] The feature integration output submodule is used to extract the time series of the corrected thrust prediction value and the extreme value parameters of the heading correction value, align the thrust prediction value and the heading correction value according to the timestamp, perform normalization and superposition operations, and generate a predicted thrust and heading set.

[0069] Extract the time series of the corrected thrust prediction value, such as T corr =[906,892,915]N, calculate the range parameter: R T =max(T corr )-min(T corr )=915-892=23N, and extract the heading correction sequence θ at the same time adj = the extreme value θ of [13.5°, 12.8°, 14.2°] max = 14.2°. Perform normalized stacking on the timestamp aligned data: in σ θ =0.7°. Taking the first time point as an example: According to the adaptive weight adjustment strategy (dynamic threshold α=0.6): If F out >α, output to the strategy execution module to generate navigation actions, such as α=0.6, 0.17<0.6, no action is triggered, if F out When <α, the environment change detection is triggered to resample the data and update the weight distribution in the experience storage.

[0070] like Figure 4 As shown, the track deviation prediction module includes a data fusion processing submodule, an offset calculation submodule, and a space-time matrix construction submodule.

[0071] The data fusion processing submodule is used to call the theoretical correction value of the heading angle in the predicted thrust heading set, collect the flow velocity direction vector monitored in real time by the ocean current sensor, perform vector superposition operation on the theoretical correction value of the heading angle and the lateral component of the flow velocity direction vector, and generate a fusion correction vector.

[0072] Call the theoretical heading angle correction value θ of the predicted thrust heading concentration adj =13.5°, θ from the feature integration output submodule adj The sequence [13.5°, 12.8°, 14.2°] is combined with the ocean current probability distribution provided by the Bayesian uncertainty quantification module to collect the velocity direction vector monitored in real time by the ocean current sensor. The velocity v = 0.3 m / s, the measurement error is ±0.02 m / s, the direction angle φ = 15°, calibrated by an electronic compass, with an accuracy of ±0.5°. Execute the lateral component extraction: v lat =v·sin(φ-θ adj )=0.3·sin(15°-13.5°)=0.3·sin(1.5°)≈0.0078m / s, longitudinal component extraction: v lon =v·cos(φ-θ adj )=0.3·cos(1.5°)≈0.2998m / s, and the heading angle theoretical correction value θ adj Convert to thrust direction vector: Perform a vector superposition operation: Determine the environmental stability through the environmental change detection module: If |v lat If the value is greater than 0.01 m / s, the threshold is 10% of the ship's drift resistance of 0.1 m / s. In this case, the value is marked as critical experience and fed into the teacher network training. Otherwise, it is stored in the general buffer. In this example, 0.0078 m / s < 0.01 m / s, so it is marked as general experience.

[0073] The offset calculation submodule is used to perform an integration operation on the time step of each future waypoint based on the lateral projection value and the longitudinal projection value of the fused correction vector, calculate the lateral displacement accumulation and the longitudinal displacement accumulation, and generate the lateral offset and the longitudinal offset.

[0074] The lateral projection value F based on the fusion correction vector lat =211.3N and longitudinal projection value F lon =881.5N, combined with the hull mass m = 1200kg fed back by the strategy optimization module, including a load error of ±50kg, with a time step of Δt = 0.2s, corresponding to a control system sampling frequency of 5Hz, and performing discrete integration operations:

[0075] Lateral acceleration calculation:

[0076] Cumulative lateral displacement:

[0077] Longitudinal acceleration calculation:

[0078] Cumulative longitudinal displacement:

[0079] The importance of the offset is evaluated by the information gain calculation module: if Δd lat >0.02m, 1% of the hull width of 2m or Δd lon >0.1m, 1% of the ship length of 10m, is marked as a key experience; in this case, Δd lat =0.0175m and Δd lon =0.0735m are all within the limit and are stored in the common buffer for priority sampling probability p(s,a)∝(|Δd|+0.01) during student network training 0.5 Extraction.

[0080] The space-time matrix construction submodule is used to extract the time series of the lateral offset and the amplitude series of the longitudinal offset, arrange them into a two-dimensional grid structure according to the time dimension and space dimension, fill in the corresponding values ​​of the lateral and longitudinal offsets, and generate the track prediction offset matrix.

[0081] Extract the lateral offset sequence Δd lat =[0.0175,0.0152,0.0183]m and longitudinal offset Δd lon =[0.0735,0.0711,0.0752]m, and construct a two-dimensional grid at timestamp t=[0.0,0.2,0.4]s:

[0082] Horizontal offset grid: G lat(t=0.0,x)=0.0175m,G lat (t=0.2,x)=0.0152m,G lat (t=0.4,x)=0.0183m;

[0083] Vertical offset grid: G lon (t=0.0,y)=0.0735m,G lon (t=0.2,y)=0.0711m,G lon (t=0.4,y)=0.0752m;

[0084] Merge to generate the track prediction offset matrix M pred :

[0085]

[0086] The matrix is ​​input into the policy optimization unit and compared with the teacher network prediction value M tea =[0.016,0.070] to calculate the error: lateral error: |0.0175-0.016|=0.0015m, longitudinal error: |0.0735-0.070|=0.0035m, if the average error That is, the threshold, which triggers the student network training update. In this case, the limit is not exceeded and it is only recorded in the experience storage.

[0087] like Figure 5 As shown, the prediction collaborative optimization module includes a control demand calculation submodule, a thrust demand calculation submodule and a collaborative sequence generation submodule.

[0088] The control demand calculation submodule is used to call the lateral offset in the track prediction offset matrix, perform a proportional amplification operation on the absolute value of the lateral offset based on the preset rudder efficiency coefficient, calculate the rudder angle adjustment amplitude per unit time, and generate the rudder angle compensation demand value.

[0089] Call the lateral offset Δd in the track prediction offset matrix lat =[0.0175,0.0152,0.0183]m, M from the spatiotemporal matrix construction submodule pred Matrix, based on the preset rudder efficiency coefficient k rud =2.5deg / m, calibrated by ship model towing test, the lateral displacement Δd in the test lat =0.4m, rudder angle compensation of 1° is required, and reverse k rud =1 / 0.4=2.5, and the absolute value of the lateral offset at each time step is scaled up:

[0090] Time t = 0.0s: Δθ req =2.5×0.0175=0.04375°;

[0091] Time t = 0.2s: Δθ req =2.5×0.0152=0.038°;

[0092] Time t = 0.4s: Δθ req =2.5×0.0183=0.04575°.

[0093] Calculate the rudder angle adjustment rate ω rud =Δθ req / Δt (Δt=0.2s):

[0094] t=0.0s:ω rud =0.04375 / 0.2=0.21875deg / s;

[0095] t=0.2s:ω rud =0.038 / 0.2=0.19deg / s;

[0096] t=0.4s:ω rud =0.04575 / 0.2=0.22875deg / s.

[0097] Generate rudder angle compensation demand sequence ω rud =[0.21875,0.19,0.22875]deg / s, the servo execution capability is verified by the safety assurance module, the maximum servo speed is 1deg / s, if ω rud >0.2deg / s, the threshold is set to 20% of the maximum rate, triggering the Bayesian uncertainty quantification module to reassess the impact of ocean currents. In this case, 0.22875deg / s exceeds the limit, which is marked as a key experience and input into the teacher network training. The rest is stored in the normal buffer.

[0098] The thrust demand calculation submodule is used to extract the longitudinal offset in the track prediction offset matrix, perform an integral operation on the rate of change of the longitudinal offset according to the thrust response curve of the thruster, calculate the thrust increment required to maintain the speed, and generate the thrust increment demand value.

[0099] Extract the longitudinal offset Δd from the track prediction offset matrix lon =[0.0735,0.0711,0.0752]m, calculate the longitudinal rate of change at each time step

[0100]

[0101] If the thrust response curve T resp (v) = 150·v, perform integral operation

[0102] t=0.0s:ΔT req =150×0.3675×0.2=11.025N;

[0103] t=0.2s:ΔT req =150×0.3555×0.2=10.665N;

[0104] t=0.4s:ΔT req =150×0.376×0.2=11.28N.

[0105] Combined with the safety module thrust upper limit T max =1500N, if ΔT req >0.1·T max =150N, if we assume Then ΔT req =150×1×0.2=30N<150N, keep the original value and generate the thrust increment sequence ΔT req =[11.025,10.665,11.28]N, input experience storage buffer, when ΔT req When the force exceeds 15N for three consecutive times, the environmental change detection is triggered and the response curve is recalibrated.

[0106] The collaborative sequence generation submodule is used to align the time stamps of the time series of the rudder angle compensation demand value and the time series of the thrust increment demand value, perform normalization scaling on the amplitudes of the two types of demands, superimpose them into a composite signal according to the phase synchronization principle, and generate a predicted collaborative control sequence.

[0107] The rudder angle is compensated for the demand ω rud =[0.21875,0.19,0.22875]deg / s and thrust increment ΔT req = [11.025, 10.665, 11.28]N aligned by timestamp t = [0.0, 0.2, 0.4]s, perform normalized scaling:

[0108] Rudder angle normalization: The denominator 1deg / s is the maximum speed threshold of the servo;

[0109] Thrust normalization: The denominator 150N is 10% of the safety threshold;

[0110] Superposition is based on the phase synchronization principle, with weights α = 0.6 and β = 0.4. The steering efficiency priority based on feedback from the strategy optimization module is:

[0111] t=0.0s:S syn=0.6×0.21875+0.4×0.0735=0.16065;

[0112] t=0.2s:S syn =0.6×0.19+0.4×0.0711=0.14844;

[0113] t=0.4s:S syn =0.6×0.22875+0.4×0.0752=0.16817.

[0114] Generate predicted cooperative control sequence S syn =[0.16065,0.14844,0.16817], the input strategy execution module generates PWM control signals, such as S syn =0.16 corresponds to a duty cycle of 16%. At the same time, it is compared with the theoretical signal output by the teacher network. If the absolute error is greater than 0.05, such as the teacher output is 0.15 and the error is 0.01, the student network is triggered to update the weights online.

[0115] like Figure 6 As shown, the flight state control execution module includes an execution parameter calling submodule, a feedforward compensation processing submodule and a control quantity output submodule.

[0116] The execution parameter calling submodule is used to call the first-step rudder angle compensation value in the predicted cooperative control sequence, collect the first-step parameters of the thrust increment value, align the rudder angle compensation value and the thrust increment value according to the timestamp, and generate the rudder angle compensation initial value and the thrust increment initial value.

[0117] Call the first step rudder angle compensation value S in the predictive cooperative control sequence rud = 0.21875 deg / s, S from the collaborative sequence generation submodule syn =[0.16065,0.14844,0.16817] The first data is extracted, the time window Δt = 0.2s is synchronized with the control system clock, and the initial value of the rudder angle compensation θ is calculated. init =S rud Δt = 0.21875 × 0.2 = 0.04375°, and the first step parameter ΔT of the thrust demand calculation submodule is called at the same time req =11.025N, from ΔT req =[11.025,10.665,11.28]N, align to timestamp t=0.0s to generate the initial value of rudder angle compensation θ init =0.04375° and the initial value of thrust increment T init =11.025N, stored in the real-time control buffer, check the initial value range: rudder angle threshold θ safe =0.04°, set according to 4 times of the minimum resolution of the servo 0.01°, if θinit >θ safe In this case, 0.04375°>0.04° triggers the safety assurance module to verify the servo's response capability. Does the servo's maximum angular velocity of 1deg / s meet θ init / Δt=0.21875deg / s<1deg / s, after passing the verification, it is marked as a key instruction and input into the teacher network verification module, otherwise it is stored in the ordinary instruction queue.

[0118] The feedforward compensation processing submodule is used to obtain real-time roll angular velocity monitoring data based on the amplitude range of the initial rudder angle compensation value, perform roll angular velocity inverse proportional correction operation on the initial rudder angle compensation value, and generate a compensated rudder angle value.

[0119] Based on the initial value of rudder angle compensation θ init =0.04375°, preset safety threshold θ safe =0.05°, according to the maximum allowable instantaneous angle setting of the servo, call the real-time monitoring data of the roll angular velocity sensor ω roll =0.15deg / s, sample three times [0.14, 0.15, 0.16]deg / s and take the average value, and perform the inverse proportional correction operation: The correction factor k roll =0.2deg / s, calibrated by ship model rolling test: when ω roll =0.2deg / s) the rudder effect is completely ineffective and the compensation value is zero), and we can get:

[0120] Correction result verification: If θ comp <0, such as ω roll =0.25deg / s, θ comp =0.04375×(1-1.25)=-0.01094°, then θ is forced to be set comp =0°, generate compensation rudder angle value θ comp =0.01094°, input control quantity calibration unit, when compensation deviation |θ comp -θ init |>0.02°, triggering the Bayesian uncertainty quantification unit to re-evaluate the roll disturbance model. In this example, |0.01094-0.04375|=0.03281°>0.02°, then update k roll The coefficient, such as adjusting from 0.2 to 0.18.

[0121] The control quantity output submodule is used to synchronize the correction result of the compensation rudder angle value with the execution timing of the initial value of the thrust increment, send a discrete instruction sequence of the thrust increment value to the thruster, and output the predicted track control quantity.

[0122] The compensation rudder angle value θ comp =0.01094° and the initial value of thrust increment T init =11.025N According to the execution timing matching, timestamp t=0.0s, allowable deviation Δt err <0.005s, for θ comp Perform normalization: where θ max =0.05° is the rudder angle safety threshold, which comes from the maximum instantaneous servo angle limit and generates the servo PWM duty cycle command The thrust command remains at the original value T out =T init =11.025N, a discrete command sequence is sent to the thruster via the CAN bus protocol [D rud ,T out ], verification instruction sending time t send = 0.001s and the deviation Δt from the theoretical time t = 0.0s err =0.001s<0.005s, if it exceeds the limit, such as Δt err = 0.006s, the current instruction is discarded and the backup data in the real-time control buffer is called, such as the θ of the previous sequence. comp =0.015° and T init =10.5N.

[0123] Table 1: Control instruction output timing table

[0124]

[0125] As shown in Table 1, the normalized rudder angle and thrust command are arranged in a strict time sequence to ensure that the control signal is continuous and stable.

[0126] The advantages and disadvantages of the prior art and the present application are compared as follows.

[0127] Existing technologies rely on a single model and static data fusion, and fail to integrate dynamic environmental parameters such as wave resistance and shallow water effects, resulting in insufficient adaptability of thrust and heading corrections. The hull position prediction lacks the ability to model spatiotemporal gradients, making it difficult to cope with nonlinear offsets caused by sudden ocean current changes. Control command generation ignores the timing coordination of rudder angle and thrust increment, and the actuator response out of step aggravates path deviation. Real-time control relies on closed-loop feedback, lacks feedforward compensation and dynamic incremental intervention, and track correction lags affect mission efficiency.

[0128] This application constructs a feature set by integrating roll angular velocity and speed change rate to enhance the spatiotemporal correlation accuracy of hull motion state prediction; improves the environmental interference resistance of thrust and heading correction through dynamic correction of wave resistance and shallow water effect; generates a spatiotemporal offset matrix based on real-time ocean current data, quantifies the track offset gradient, and achieves high-precision dynamic prediction; generates a standardized control sequence through time-series collaborative processing of rudder angle compensation and thrust increment to eliminate actuator response loss; combines feedforward compensation and dynamic increment instructions with real-time monitoring data to accurately control execution actions, reduce control delays, and ensure the real-time and stability of track regression.

[0129] Example 2:

[0130] like Figure 7 As shown, the present invention discloses a control method for an unmanned ship based on state prediction, comprising the following steps:

[0131] S1: Obtain the roll angular velocity time series data output by the inertial measurement unit, collect the speed change rate of the Doppler odometer, input the roll angular velocity time series data into the sliding window integrator to generate a displacement prediction curve, perform a first-order differential operation on the speed change rate to generate an acceleration prediction value, and fuse the displacement prediction curve and the acceleration prediction value to construct a flight state prediction feature set.

[0132] S2: Call the displacement prediction curve in the navigation state prediction feature set, input the fluid resistance response function of the wave kinematic model to calculate the theoretical thrust demand of the propeller and the theoretical correction of the heading angle, superimpose the bottom friction compensation coefficient of the shallow water effect model to correct the thrust prediction value, and generate a predicted thrust and heading set.

[0133] S3: Based on the theoretical correction of the heading angle of the predicted thrust heading concentration and the flow velocity direction vector monitored in real time by the ocean current sensor, the time series of the lateral position offset and the longitudinal position offset are calculated to construct the track prediction offset matrix.

[0134] S4: Calculate the rudder angle compensation requirement based on the lateral offset in the track prediction offset matrix, and calculate the propeller thrust increment requirement based on the longitudinal offset. Perform phase alignment and amplitude normalization on the two types of requirements in time series to generate a predicted collaborative control sequence.

[0135] S5: Call the first-step rudder angle compensation value and thrust increment value in the predicted cooperative control sequence, perform feedforward compensation action on the servo in combination with the real-time roll angular velocity monitoring data, inject thrust prediction increment instructions into the thruster, and output the predicted track control amount.

Claims

1. An unmanned vessel control system based on state prediction, characterized by: It includes flight state feature modeling module, prediction response generation module, track deviation prediction module, prediction collaborative optimization module and flight state control execution module. The flight state feature modeling module is used to obtain the roll angular velocity time series data output by the inertial measurement unit, collect the speed change rate of the Doppler log, perform sliding window integration on the roll angular velocity series to generate a displacement prediction curve, perform differential operation on the speed change rate to generate an acceleration prediction value, and fuse the displacement curve and acceleration value to construct a flight state prediction feature set; A prediction response generation module is used to call the navigation state prediction feature set, input the wave model to calculate the theoretical thrust demand and heading angle correction of the propeller, superimpose the shallow water effect correction thrust prediction value, and generate a predicted thrust and heading set; A track deviation prediction module is used to calculate the lateral and longitudinal deviations based on the heading correction value of the predicted thrust heading set and the flow velocity direction of the ocean current sensor, and to construct a time-space distribution matrix to generate a track prediction deviation matrix; A prediction collaborative optimization module is used to calculate the rudder angle compensation and thrust increment requirements based on the track prediction offset matrix, perform phase alignment and amplitude processing to generate a prediction collaborative control sequence; The flight state control execution module is used to call the rudder angle compensation value and thrust increment value of the predicted cooperative control sequence, execute the servo feedforward compensation and inject the thrust instruction, and output the predicted track control amount.

2. The unmanned vessel control system based on state prediction according to claim 1, characterized in that: The flight state prediction feature set includes a roll displacement curve and a linear acceleration prediction value; the predicted thrust heading set includes a fluid resistance correction thrust, a heading compensation angle, and a bottom friction compensation thrust; the track prediction offset matrix includes a lateral offset distribution field, a longitudinal offset distribution field, and a time-space offset gradient matrix; the predicted collaborative control sequence includes a rudder angle compensation time series, a thrust increment time series, a collaborative phase vector, and a collaborative amplitude vector; and the predicted track control quantity includes a feedforward rudder angle quantity, a thrust increment pulse, and a track regression drive quantity.

3. The unmanned vessel control system based on state prediction according to claim 1, characterized in that: The flight state feature modeling module includes a data acquisition and preprocessing submodule, a dynamic parameter calculation submodule, and a feature fusion modeling submodule. The data acquisition and preprocessing submodule is used to obtain the roll angular velocity time series data output by the inertial measurement unit, collect the speed change rate recorded by the Doppler log, input the roll angular velocity time series data into the sliding window integrator for time alignment, perform standardized noise reduction processing on the speed change rate, and generate a time series preprocessing set; a dynamic parameter calculation submodule, configured to call the roll angular velocity data in the time series preprocessing set, perform integral accumulation operations based on a fixed time window length, generate a displacement increment sequence, synchronously perform differential operator processing on the speed change rate, and output acceleration prediction values ​​and displacement prediction curves; The feature fusion modeling submodule is used to extract the longitudinal axis range parameters of the displacement prediction curve and the fluctuation parameters of the acceleration prediction value, weightedly superimpose the two types of parameters according to the timestamp, fuse the displacement and acceleration data according to the preset weight coefficient, and generate a flight state prediction feature set.

4. The unmanned vessel control system based on state prediction according to claim 1, characterized in that: The predicted response generation module includes a theoretical parameter calculation submodule, an environmental compensation correction submodule, and a feature integration output submodule. The theoretical parameter calculation submodule is used to call the displacement prediction curve in the navigation state prediction feature set, input the fluid resistance response function of the wave kinematic model, perform an integral operation on the ordinate value of the displacement prediction curve, calculate the thrust demand of the propeller in the future period, and simultaneously extract the lateral component of the fluid resistance response function to generate the theoretical thrust demand and the heading correction value. an environmental compensation correction submodule, configured to call a bottom friction compensation coefficient of a shallow water effect model based on the amplitude range of the theoretical thrust requirement and the heading correction amount, perform a point-by-point multiplication operation on the bottom friction compensation coefficient and the theoretical thrust requirement, correct the longitudinal component of the thrust prediction value, and generate a corrected thrust prediction value; The feature integration output submodule is used to extract the time series of the corrected thrust prediction value and the extreme value parameters of the heading correction amount, align the thrust prediction value and the heading correction amount according to the timestamp, perform a normalized superposition operation, and generate a predicted thrust and heading set.

5. The unmanned vessel control system based on state prediction according to claim 1, characterized in that: The track deviation prediction module includes a data fusion processing submodule, an offset calculation submodule and a space-time matrix construction submodule. The data fusion processing submodule is used to call the heading angle theoretical correction value in the predicted thrust heading set, collect the velocity direction vector monitored in real time by the ocean current sensor, perform vector superposition operation on the heading angle theoretical correction value and the transverse component of the velocity direction vector, and generate a fusion correction vector. an offset calculation submodule, configured to perform an integration operation on the time step of each future waypoint based on the lateral projection value and the longitudinal projection value of the fused correction vector, calculate the lateral displacement accumulation amount and the longitudinal displacement accumulation amount, and generate the lateral offset and the longitudinal offset; The space-time matrix construction submodule is used to extract the time series of the lateral offset and the amplitude series of the longitudinal offset, arrange them into a two-dimensional grid structure according to the time dimension and space dimension, fill in the corresponding values ​​of the lateral and longitudinal offsets, and generate a track prediction offset matrix.

6. The unmanned vessel control system based on state prediction according to claim 1, characterized in that: The prediction collaborative optimization module includes a control demand calculation submodule, a thrust demand calculation submodule and a collaborative sequence generation submodule. The control demand calculation submodule is used to call the lateral offset in the track prediction offset matrix, perform a proportional amplification operation on the absolute value of the lateral offset based on a preset rudder efficiency coefficient, calculate the rudder angle adjustment amplitude per unit time, and generate a rudder angle compensation demand value; a thrust demand calculation submodule, configured to extract the longitudinal offset from the track prediction offset matrix, perform an integral operation on the rate of change of the longitudinal offset according to the thrust response curve of the propeller, calculate the thrust increment required to maintain the ship's speed, and generate a thrust increment demand value; The collaborative sequence generation submodule is used to align the time stamps of the time series of the rudder angle compensation demand value and the time series of the thrust increment demand value, perform normalization scaling on the amplitudes of the two types of demands, superimpose them into a composite signal according to the phase synchronization principle, and generate a predicted collaborative control sequence.

7. The unmanned vessel control system based on state prediction according to claim 1, characterized in that: The navigation state control execution module includes an execution parameter calling submodule, a feedforward compensation processing submodule and a control quantity output submodule. The execution parameter calling submodule is used to call the first-step rudder angle compensation value in the predicted cooperative control sequence, collect the first-step parameters of the thrust increment value, align the rudder angle compensation value and the thrust increment value according to the timestamp, and generate the rudder angle compensation initial value and the thrust increment initial value; a feedforward compensation processing submodule, configured to obtain real-time roll angular velocity monitoring data based on the amplitude range of the initial rudder angle compensation value, perform a roll angular velocity inverse proportional correction operation on the initial rudder angle compensation value, and generate a compensated rudder angle value; The control quantity output submodule is used to synchronize the correction result of the compensation rudder angle value with the execution timing of the initial value of the thrust increment, send a discrete instruction sequence of the thrust increment value to the thruster, and output the predicted track control quantity.

8. A method for controlling an unmanned vessel based on state prediction, used in the unmanned vessel control system based on state prediction according to any one of claims 1 to 7, characterized in that: The following steps are involved: S1: Obtain the roll angular velocity time series data output by the inertial measurement unit, collect the speed change rate of the Doppler log, input the roll angular velocity time series data into the sliding window integrator to generate a displacement prediction curve, perform a first-order differential operation on the speed change rate to generate an acceleration prediction value, and fuse the displacement prediction curve and the acceleration prediction value to construct a flight state prediction feature set; S2: calling the displacement prediction curve in the navigation state prediction feature set, inputting the fluid resistance response function of the wave kinematics model to calculate the theoretical thrust demand and heading angle correction of the propeller, adding the bottom friction compensation coefficient of the shallow water effect model to correct the thrust prediction value, and generating a predicted thrust and heading set; S3: Based on the theoretical correction of the heading angle in the predicted thrust heading concentration and the current velocity direction vector monitored in real time by the ocean current sensor, the time series of the lateral position offset and the longitudinal position offset are calculated to construct a track prediction offset matrix; S4: Calculating the rudder angle compensation requirement based on the lateral offset in the trajectory prediction offset matrix, calculating the propeller thrust increment requirement based on the longitudinal offset, performing phase alignment and amplitude normalization processing on the two types of requirements in a time series, and generating a predicted cooperative control sequence; S5: Call the first-step rudder angle compensation value and thrust increment value in the predicted cooperative control sequence, perform feedforward compensation action on the servo in combination with the real-time roll angular velocity monitoring data, inject thrust prediction increment instructions into the thruster, and output the predicted track control amount.

Images