Construction status prediction method and system of trailing suction hopper vessel based on physical data dual drive

Abstract

The invention discloses a method and system for predicting the construction state of a trailing hopper vessel based on physical data dual drive, and specifically relates to the field of dredging construction state prediction, and is used to solve the problem of insufficient monitoring and prediction of the construction state of a trailing hopper vessel under high sea conditions. The invention realizes high-precision attitude prediction combining physical constraints with data drive by constructing a coupled dynamic model of a hull-rake arm-rake head-wave compensator-hoisting system, and combining an LSTM neural network deep learning method. Based on a time series anomaly diffusion index and an environment-physical model mismatch index, a risk assessment system for abnormal energy accumulation, diffusion capacity and environmental over-limit conditions is established, and the accuracy of safety warning is improved. A genetic algorithm is used to dynamically optimize construction control parameters, so that the attitude adjustment of the rake head and the rake arm has adaptability, and a closed-loop iteration of prediction, warning, optimization and control is realized, and the stability of construction under complex sea conditions is improved.

Description

Technical Field

[0001] The present invention relates to the field of dredging construction state prediction, and more specifically, to a construction state prediction method and system of a trailing suction hopper vessel based on physical data dual drive. Background Art

[0002] The hydrological and meteorological conditions in the dredging construction area have become a key factor restricting construction efficiency and quality. The dredger is the core equipment in the dredging project, and its construction efficiency and safety performance directly affect the quality and progress of the project. Under high sea conditions, the posture of the dredger head and dredger arm is significantly affected by natural factors such as wind, waves, and surges, which may cause equipment damage, reduced construction efficiency, and even safety accidents. Traditional posture control and early warning methods mainly rely on manual experience and simple sensor data, which are difficult to cope with complex and changeable high sea conditions.

[0003] At present, the state monitoring technology of trailing suction hopper dredger has been widely used, but most of the existing technologies only focus on real-time monitoring functions, and lack the ability to predict and warn potential risks in advance. In addition, the existing technologies rarely involve theoretical research on trailing suction hopper dredgers, and lack simulation analysis support based on physical models. The physical simulation model can quickly and accurately predict the movement posture of the dredger head and dredger arm to a certain extent, but its prediction accuracy is still insufficient and it is difficult to meet the requirements of high-precision construction, and it is urgently needed to be further optimized and improved.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a construction status prediction method and system of a trailing suction dredge based on physical data dual drive, by constructing a coupled dynamic model of the hull-trailer arm-trailer head-wave compensator-hoisting system, and combining the LSTM neural network deep learning method, a high-precision attitude prediction combining physical constraints and data drive is realized; based on the time series anomaly diffusion index and the environment-physical model mismatch index, a risk assessment system for abnormal energy accumulation, diffusion capacity and environmental over-limit conditions is established, thereby improving the accuracy of safety warnings; a genetic algorithm is used to dynamically optimize the construction control parameters, so that the attitude adjustment of the trailing head and the trailing arm has adaptability, realizing a closed-loop iteration of prediction, warning, optimization and control, and improving the stability of construction under complex sea conditions, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive includes the following steps:

[0008] S1: Use multi-source sensors to collect construction-related data in real time and perform cleaning and standardization to build a training data set;

[0009] S2: Establish a coupled dynamic model of the hull-raker arm-raker head-wave compensator-hoisting system, and combine numerical simulation to simulate the dynamic response of the equipment under high sea conditions;

[0010] S3: Integrate historical data and numerical simulation results to build a physically constrained LSTM model, optimize the loss function through the dynamic equation and complete training verification;

[0011] S4: Integrate 3D modeling and simulation platform to optimize the attitude prediction model, and combine physical simulation and data-driven model to output prediction results;

[0012] S5: By analyzing the abnormal energy accumulation and diffusion capabilities, combining the deviation between the current environmental parameters and the safety threshold preset by the physical model, and associating the historical failure probability assessment to present the risk level, the input parameters are finally inferred through the LSTM model to generate the safety posture optimization conditions of the rake head and rake arm;

[0013] S6: Dynamically optimize construction control parameters through genetic algorithms, perform posture adjustments and feed back optimization results to the model to achieve closed-loop iteration.

[0014] In a preferred embodiment, step S2 includes the following contents:

[0015] Step S201, by establishing the hull kinematic equation of the trailing suction hopper, considering the motion response of the hull in multiple degrees of freedom, and calculating the influence of the hull mass, additional mass and moment of inertia on the motion;

[0016] Step S202: simulating the dynamic response of the rake head and the rake arm by establishing kinematic equations of the rake head and the rake arm, calculating the displacements in three directions and the effects of soil, fluid and water flow resistance;

[0017] Step S203, by establishing a physical model of the wave compensator, calculating its damping and elastic coefficients, and simulating its hysteresis effect, reflecting the dynamic influence of the wave compensator on the posture of the rake head and rake arm;

[0018] Step S204, input environmental parameters and construction parameters, simulate the dynamic response of the hull and equipment under high sea conditions through finite element analysis and multi-body dynamics simulation methods, and predict the real-time posture changes of the rake head and rake arm.

[0019] In a preferred embodiment, step S3 includes the following contents:

[0020] Step S301, collect historically collected ship speed, ship attitude, and environmental parameter data, perform preprocessing and extract key features, and construct a unified training data set;

[0021] Step S302: constructing an LSTM neural network model to predict the posture changes of the rake head and rake arm through time series data;

[0022] Step S303: taking the coupled dynamics model of the rake head and the rake arm as a constraint condition and incorporating it into the loss function, the total loss function being a weighted sum of the prediction error term and the physical constraint term;

[0023] Step S304: Adopting an adaptive optimization algorithm to optimize the hyperparameters of the LSTM model to improve the convergence speed and prediction accuracy of the network;

[0024] Step S305: Use numerical simulation data to verify the prediction accuracy of the LSTM model and evaluate its performance in practical applications.

[0025] In a preferred embodiment, in step S5, based on the real-time collected data, big data analysis technology is used to identify potential risks and assess their severity;

[0026] The risk is comprehensively assessed through the time series anomaly diffusion index and the environment-physical model mismatch index, and the risk coefficient is calculated by weighted summation of the time series anomaly diffusion index and the environment-physical model mismatch index.

[0027] In a preferred embodiment, the logic for obtaining the time series anomaly diffusion index is as follows:

[0028] Abnormal energy accumulation in time: ,in, is the abnormal pressure deviation at a point in time, : Environmental dynamic damping coefficient, is the horizontal velocity of the wave, is the ocean current velocity; is the reference flow rate; the characteristic value is used to measure the intensity and stability of abnormal diffusion and evaluate the diffusion capacity: ,in, is the covariance matrix, where is the maximum eigenvalue of the covariance matrix, is the minimum eigenvalue of the covariance matrix. Finally, the time series anomaly diffusion index is obtained by comprehensive calculation of abnormal energy accumulation and diffusion capacity. .

[0029] In a preferred embodiment, the logic for obtaining the environment-physical model mismatch index is as follows:

[0030] Boundary conflict detection calculates the deviation between the current environmental parameters and the preset safety threshold of the physical model, and accumulates it if it exceeds the limit: ,in, and Respectively represent the current environmental parameters and the safety thresholds preset by the model, Represents the total number of environmental parameters, Indicates the index of the specific environment parameter currently being processed. It is an indicator function. When the environmental parameter exceeds the safety threshold, the square of its relative deviation is calculated, otherwise it is not taken into account. By calculating the failure frequency in similar historical scenarios, the failure probability in the current environment is directly estimated: ,in, Indicates the history of the work Boundary conflict value at the time of operation, is the total sample size of the historical data set; environment-physical model mismatch index: .

[0031] In a preferred embodiment, the warning determination rules are as follows:

[0032] When the risk coefficient is greater than or equal to the second critical value of risk, a first-level warning signal is generated and construction is immediately interrupted; when the risk coefficient is greater than or equal to the first critical value of risk and less than the second critical value of risk, a second-level warning signal is generated and the ship speed and wave compensation parameters are automatically adjusted; when the risk coefficient is less than the first critical value of risk, a third-level warning signal is generated, mainly to record abnormalities and make subsequent optimization adjustments.

[0033] In a preferred embodiment, after obtaining the secondary warning signal, the trained LSTM neural network model is used to infer the input parameters, and the rake head and rake arm postures that meet the safety conditions are generated through an optimization algorithm.

[0034] In a preferred embodiment, step S6 includes the following contents:

[0035] Step S601, taking the safe posture of the rake head and rake arm as the goal, using genetic algorithm to optimize input parameters, randomly generating a group of input parameters as the initial population, defining a fitness function, measuring whether the rake head and rake arm posture corresponding to each group of input parameters meets the safety conditions, generating a new generation of population through selection, crossover and mutation operations, gradually optimizing the input parameters, repeating this process for iterative optimization, until a combination of input parameters that meets the safety conditions is found;

[0036] Step S602, outputting the optimized input parameters to the central control system;

[0037] Step S603, generating specific response suggestions based on the optimization results to ensure the safety of personnel and equipment;

[0038] Step S604, automatically inputting the optimized input parameters into the ship control system, adjusting the ship speed, ship attitude and mud pump speed in real time, and ensuring that the rake head and rake arm return to a safe attitude;

[0039] Step S605, feeding back the optimized construction parameters and actual effects to the LSTM neural network model for online learning and continuous optimization of the model.

[0040] The construction status prediction system of the trailing suction hopper vessel based on physical data dual drive includes: data tracing module, simulation mapping module, intelligent driving module, risk anchoring module and adaptive control module;

[0041] Data tracing module: collects multi-source sensor data, performs data cleaning and standardization, builds training data sets, and passes the training data sets to the simulation mapping module for physical modeling;

[0042] Simulation mapping module: establishes the coupled dynamic model of the hull-raker arm-raker head-wave compensator-hoisting system, combines numerical simulation to simulate the dynamic response of the equipment under high sea conditions, and outputs physical modeling data to provide prediction support for the intelligent drive module;

[0043] Intelligent driving module: integrates historical data and numerical simulation results, builds a physically constrained LSTM model, optimizes the loss function through the dynamic equation, optimizes the posture prediction model on the 3D modeling simulation platform, outputs the predicted posture data of the rake head and rake arm, and delivers it to the risk anchoring module for safety assessment;

[0044] Risk anchoring module: Analyzes abnormal energy accumulation and diffusion capabilities, combines the current environmental parameters with the safety threshold deviation of the physical model, associates the historical failure probability to assess the risk level, and reversely infers the input parameters through the LSTM model to generate optimization conditions, evaluate abnormal conditions and optimize input parameters for the adaptive control module to adjust the posture;

[0045] Adaptive control module: Dynamically optimize construction control parameters through genetic algorithms, adjust the rake head and rake arm postures, and feed back the optimization results to the model to achieve closed-loop iteration, output the optimized control parameters, and feed back the adjusted construction data to the data traceability module to continuously improve the model accuracy.

[0046] The technical effects and advantages of the method and system for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive of the present invention are as follows:

[0047] Aiming at the inadequacy of monitoring and predicting the construction status of a dredge suction dredge under high sea conditions, the present invention proposes a construction status prediction method and system based on physical data dual drive, which overcomes the problem of insufficient prediction accuracy of the traditional single physical model or data-driven model. By constructing a coupled dynamic model of the hull-dredge arm-dredge head-wave compensator-hoisting system and combining the LSTM neural network deep learning method, a high-precision attitude prediction combining physical constraints and data drive is realized. Based on the time series anomaly diffusion index and the environment-physical model mismatch index, a risk assessment system for abnormal energy accumulation, diffusion capacity and environmental over-limit conditions is established, which improves the accuracy of safety warning. The construction control parameters are dynamically optimized by using a genetic algorithm, so that the attitude adjustment of the dredge head and the dredge arm is adaptable, and the closed-loop iteration of prediction, warning, optimization and control is realized, which improves the stability of construction under complex sea conditions. Compared with the prior art, the present invention not only improves the accuracy of attitude prediction and enhances the risk warning capability, but also optimizes the intelligent regulation of the construction process, and improves the operation safety and efficiency of the dredge suction dredge in high dynamic waters. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the process of the present invention.

[0049] Figure 2 It is a flow chart of coupled dynamic modeling of the hull-raker arm-raker head-wave compensator-hanging system of the present invention.

[0050] Figure 3 A flow chart is constructed for the physical data dual-driven model of the present invention.

[0051] Figure 4 It is a flow chart of the rake head and rake arm posture prediction model of the present invention.

[0052] Figure 5 It is a schematic diagram of the structure of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] Embodiment 1: Figure 1 The present invention provides a method for predicting the construction state of a trailing suction hopper vessel based on physical data dual drive, comprising:

[0055] S1: Use multi-source sensors to collect construction-related data in real time and perform cleaning and standardization to build a training data set;

[0056] S2: Establish a coupled dynamic model of the hull-raker arm-raker head-wave compensator-hoisting system, and combine numerical simulation to simulate the dynamic response of the equipment under high sea conditions;

[0057] S3: Integrate historical data and numerical simulation results to build a physically constrained LSTM model, optimize the loss function through the dynamic equation and complete training verification;

[0058] S4: Integrate 3D modeling and simulation platform to optimize the attitude prediction model, and combine physical simulation and data-driven model to output prediction results;

[0059] S5: By analyzing the abnormal energy accumulation and diffusion capabilities, combining the deviation between the current environmental parameters and the safety threshold preset by the physical model, and associating the historical failure probability assessment to present the risk level, the input parameters are finally inferred through the LSTM model to generate the safety posture optimization conditions of the rake head and rake arm;

[0060] S6: Dynamically optimize construction control parameters through genetic algorithms, perform posture adjustments and feed back optimization results to the model to achieve closed-loop iteration.

[0061] Step S1, data collection and preprocessing. Using various sensors on the tractor suction hopper (such as attitude sensor, pressure sensor, GPS, etc.), real-time collection of construction data such as the position, pressure, angle, ship attitude, and sea condition data (such as wave height, surge speed) of the tractor head. Clean and standardize the collected data to build a high-quality training data set.

[0062] Step S2, physical model construction and numerical simulation. Figure 2 The coupled dynamic model of the hull-raker arm-raker head-wave compensator-hanging system of the trailing suction hopper is established as shown in the figure, and the influence of external environmental factors such as wind, waves, and surge on the system can be considered. The specific process is as follows:

[0063] Step S201, hull model construction. The mathematical model of the trailing suction hopper dredger establishes the kinematic equation of the hull based on the maneuverability principle. The coordinate system is an attached coordinate system that moves with the hull.

[0064]

[0065] Where:

[0066] : The mass of the hull (kg), excluding the influence of the surrounding fluid.

[0067] , , : Additional mass in the x, y, and z directions (kg), due to the additional inertial effect of the surrounding water when the hull moves, it is related to the hull shape and fluid density.

[0068] , , : The linear displacement of the center of gravity of the ship in the ship coordinate system (m), corresponding to:

[0069] x: longitudinal (surge, forward and backward movement),

[0070] y: lateral (sway, left and right movement),

[0071] z: vertical (heave, up and down movement).

[0072] , , : Euler angle of the hull (rad), describing the rotation attitude:

[0073] : Roll angle (rotation around the x-axis),

[0074] : pitch angle (rotation around the y axis),

[0075] : Deflection angle (yaw, rotation around the z-axis).

[0076] , , : corresponding angular velocity (rad / s).

[0077] , , : linear acceleration (m / s²).

[0078] , , : angular acceleration (rad / s²).

[0079] , , : The moment of inertia of the hull around the x, y, and z axes (kg·m²), reflecting the resistance of the hull mass distribution to rotation.

[0080] , , : Additional moment of inertia (kg·m²), the additional rotational inertia due to the fluid rotating with the hull.

[0081] , , : Hydrodynamic force (N), generated by the movement of the hull in the water.

[0082] , , : Propulsion force (N), generated by the propulsion system (such as propeller).

[0083] , , : Rudder force (N), generated by the rudder to control steering.

[0084] , , : Wind force (N), generated by wind acting on the hull.

[0085] , , : Wave force (N), generated by wave action.

[0086] , , : Control force (N), generated by control devices (such as thrusters).

[0087] , , : The influence of dragline traction on the hull (N).

[0088] , , : Hydrodynamic torque (N·m), corresponding to roll, pitch, and yaw directions.

[0089] , , : Wind moment (N·m).

[0090] , , : Wave moment (N·m).

[0091] : Yawing moment generated by the rudder (N·m).

[0092] , : Torque generated by the traction force of the rake head (unit: N·m).

[0093] The physical model of the rake arm and rake head of the trailing suction hopper vessel can be constructed according to the target parameters and mathematical model such as the size of the rake arm and rake head, the center of gravity position of each part, the moment of inertia, etc.

[0094] Step S202, the model of the rake head and rake arm is constructed. Similarly, the kinematic equations of the rake head and rake arm can be established. The coordinate system is an attached coordinate system that moves with the hull.

[0095]

[0096] In the formula, is the total mass of the rake head and rake arm; , , are the displacements of the rake head in three directions respectively; , , They are soil resistance in three directions; , They are the soil resistance in three directions respectively; , are the water flow resistance in two directions respectively; It is the reaction force of high pressure water on the rake head; It is the force generated by the vacuum pressure difference of the rake head; , are the damping and elastic coefficients of the heave compensator.

[0097] The physical model of the rake arm and rake head of the trailing suction hopper vessel can be constructed according to the target parameters and mathematical model such as the size of the rake arm and rake head, the center of gravity position of each part, the moment of inertia, etc.

[0098] Step S203, wave compensator model construction. Wave compensator is an important equipment that affects the posture of rake head and rake arm. Its damping and elastic coefficients can be expressed as:

[0099]

[0100] In the formula, is the surface area of ​​the wave compensator piston; is the catheter diameter; The catheter is long; is the dynamic viscosity; the elastic coefficient consists of two parts, namely the system elastic coefficient Elastic coefficient in the direction of wire rope extension :

[0101]

[0102] In the equation, the air pressure in equilibrium is , the volume is , is an exponent, which is an integer. Its function is to affect the calculation result of the system elastic coefficient by adjusting the proportional relationship and power behavior. is the elastic modulus of the wire rope; is the cross-sectional area of ​​the wire rope; is the length of the wire rope; It is the elongation of the heave compensator rope, that is, the relative displacement between the hull and the drag head.

[0103] However, in actual operation, the heave compensator has a lag effect. Therefore, the lag in the rope tension transmission is used in the model to simulate this phenomenon. The lag time is Proportional to the length of the wire rope:

[0104]

[0105] In the formula, is the hysteresis coefficient.

[0106] Step S204, simulating the motion posture data of the rake head and rake arm. Input environmental parameters such as wind, wave and current and specified construction parameters such as wave compensator pressure and excavation depth, and simulate the dynamic response of the hull and equipment under high sea conditions through numerical simulation methods such as finite element analysis and multi-body dynamics simulation to predict the real-time posture of the rake head and rake arm.

[0107] Step S3, construct the physical data dual-driven model. The model construction flow chart is as follows: Figure 3 As shown, the steps of model construction are as follows:

[0108] Step S301, data preparation and preprocessing.

[0109] A unified training data set is constructed using historically collected data on ship speed, ship motion posture (such as roll, pitch, bow pitch), environmental parameters (such as wave height, surge velocity, wind speed), rake head pressure, rake arm angle, etc. as the training set, and data generated by numerical simulation as the validation set.

[0110] After preprocessing (such as denoising and standardization), key features are extracted, such as time-frequency domain features, statistical features, and physical features (such as mud pump speed, mud pump power, drag head digging depth, etc.).

[0111] Step S302, neural network model construction.

[0112] LSTM (Long Short-Term Memory Network) is a recurrent neural network (RNN) variant suitable for time series prediction, which can effectively capture long-term dependencies in time series data. In the present invention, LSTM neural network is used to predict the attitude changes of rake head and rake arm in high sea conditions. The following is the specific architecture design of the neural network:

[0113] The time step of the input layer is set according to actual needs. Usually, the data of the past few seconds or minutes are selected as input. For example, if the data of the past 60 seconds is selected and the data is collected once per second, the input dimension is (60, n1), where n1 is the number of input parameters (such as ship speed, attitude, environmental parameters, etc.).

[0114] The LSTM network contains multiple hidden layers, each containing several LSTM units. The number of hidden layers and units can be selected based on data complexity and computing resources. For example:

[0115] The first LSTM layer: contains 128 LSTM units, which are used to preliminarily extract time series features.

[0116] The second LSTM layer contains 64 LSTM units, which are used to further extract high-order features.

[0117] Dropout layer: Add a Dropout layer (such as dropoutrate=0.2) after each LSTM layer to prevent overfitting.

[0118] The tanh activation function is used inside the LSTM unit to control the flow and memory of information.

[0119] The output layer predicts the attitude parameters of the rake head and rake arm, including the horizontal position (x, y coordinates), vertical position (z coordinates), pitch and yaw of the rake head, and the angle and position of the rake arm. The dimension of the output layer is (m, k), where m is the predicted time step (such as the next 10 seconds) and k is the number of output parameters (such as rake head position, angle, etc.).

[0120] Step S303, performing physical law fusion.

[0121] The loss function is used to measure the error between the model prediction result and the true value. In the present invention, the loss function not only considers the prediction error, but also integrates the coupled dynamic model of the rake head and the rake arm to enhance the physical consistency of the model.

[0122] Use mean squared error (MSE) as the prediction error term to calculate the difference between the predicted value and the true value:

[0123]

[0124] in, is the true value, is the predicted value, is the sample size.

[0125] The coupled dynamics model of the rake head and rake arm is used as a constraint condition and incorporated into the loss function. For example, the forces on the rake head and rake arm are calculated according to the dynamics model and compared with the predicted results:

[0126]

[0127] in, Actual observed values ​​of the forces or motion states of the rake head and rake arms; It is the force or motion state of the rake head and rake arm calculated according to the coupled dynamics model.

[0128] The total loss function is the weighted sum of the prediction error term and the physical constraint term:

[0129]

[0130] in, and is the weight coefficient used to balance the impact of prediction error and physical constraints.

[0131] Step S304: perform model training and optimization.

[0132] Adaptive optimization algorithms (such as Adam) are used to optimize the weight parameters of the neural network to improve the convergence speed and prediction accuracy of the model.

[0133] Optimize the hyperparameters of the LSTM network (including the number of hidden layers, number of LSTM neurons, learning rate, etc.) through grid search or Bayesian optimization methods.

[0134] The preprocessed training set data is input into the LSTM network and the prediction value is calculated by forward propagation. The total loss function is calculated and the network weights are updated by back propagation. The above process is repeated until the model converges or the maximum number of training rounds (epochs) is reached.

[0135] Step S305: perform model verification and testing.

[0136] The data generated by numerical simulation is used as a validation set to evaluate the prediction accuracy and generalization ability of the model.

[0137] The actual collected high sea state data is used as the test set to verify the performance of the model in practical applications.

[0138] The mean square error (MSE), mean absolute error (MAE) and coefficient of determination (R²) were used as evaluation indicators to assess the prediction accuracy of the model.

[0139] Step S306: deploy and apply the model.

[0140] The trained LSTM model is deployed to the central control system of the trailing suction hopper to predict the posture changes of the drag head and the drag arm in real time.

[0141] The forecast results and construction status are displayed in real time through electronic chart displays or AR augmented reality devices to provide decision support for the captain.

[0142] Step S4: constructing a model for predicting the attitude of the rake head and rake arms. Figure 4 As shown in the figure, the 3D modeling technology is combined to build a 3D model of the rake head and rake arm, and the construction scenes under different sea conditions are simulated through simulation platforms such as Unity to optimize the accuracy of the prediction model. The physical data dual-drive model is combined with numerical simulation to further improve the accuracy of attitude prediction.

[0143] The combination of the physical data dual-driven model and simulation data can be optimized and adjusted in the following ways: Assuming that the suction operation is carried out under high sea conditions, the simulation model first simulates the attitude changes of the suction head and the suction arm based on historical sea condition data (such as waves, flow rate, etc.). For example, the simulation platform predicts that the pitch angle of the suction head changes by 15 degrees under a specific wave height and wind speed. However, in the actual construction process, the real-time sensor data shows that the current wave height is higher than the simulation prediction value, resulting in the actual pitch angle of the suction head being 17 degrees. At this time, the physical data dual-driven model inputs the real-time collected sensor data (such as wave height, ship attitude) into the model, and corrects the prediction results of the simulation model through physical constraints (such as fluid dynamics principles). The model will adjust the original simulation results according to the actual sea conditions, for example, by increasing the environmental damping coefficient to reflect the impact of the current high sea conditions, so that the predicted suction head attitude (pitch angle) is more accurate, avoiding the errors that may be caused by relying solely on simulation or real-time data. In this way, the combination of simulation data and real-time physical data makes the final attitude prediction result more consistent with the actual construction situation, improving the prediction accuracy and adaptability of the system.

[0144] Step S5, Operational Safety Risk Assessment and Warning. Based on the real-time collected data, use big data analysis technology to identify potential risks (such as rake head deviation, hydraulic system overload, etc.), and evaluate their severity and probability of occurrence. Design an intelligent warning system to automatically issue an alarm when abnormal data is detected (such as a sudden increase in rake head pressure, excessive posture deviation).

[0145] When conducting a comprehensive risk assessment before a safety warning is triggered, it is first necessary to identify potential abnormalities and, through a series of refined calculations and analyses, determine whether the abnormality is caused by the dynamic changes and diffusion of internal equipment or by the external environment exceeding the safety boundary preset by the model. This process not only relies on physical modeling, but also combines statistical methods to ensure the objectivity and accuracy of the assessment results. The following is the specific assessment process and calculation method.

[0146] Step S501, comprehensively assessing risk through the time series anomaly diffusion index and the environment-physical model mismatch index:

[0147] Temporal Anomaly Diffusion Index (TADI): It is used to evaluate the accumulation and diffusion capacity of abnormal energy and reflect whether the anomaly may trigger a chain reaction.

[0148] Environmental-Physical Model Mismatch Index (EPMI): used to detect whether the current environmental parameters exceed the safety boundary of the physical model and whether there is a trend of risk accumulation.

[0149] Step S502, the logic for obtaining the time series anomaly diffusion index is as follows:

[0150] Abnormal energy accumulation in time:

[0151]

[0152] in, is the abnormal pressure deviation at a time point (such as the percentage deviation between the actual value measured by the pressure sensor and the expected value), : Environmental dynamic damping coefficient, is the horizontal velocity of the wave, is the ocean current velocity; is the reference flow rate (50% of the historical maximum flow rate), which determines the dissipation rate of abnormal energy in a high flow rate environment.

[0153] The abnormal energy is accumulated over time to reflect the continuous impact of the abnormality. As time goes by, the impact of the abnormality gradually decays, and this decay is modeled by an exponential function. The dynamic characteristics of the environment are modeled by the environmental dynamic damping coefficient. The greater the flow rate, the stronger the attenuation effect, which conforms to the actual situation of fluid dynamics and highlights the impact of the fluid environment on the diffusion of abnormalities.

[0154] The intensity and stability of abnormal diffusion are measured by characteristic values, and the diffusion capacity is evaluated:

[0155]

[0156] in, is the covariance matrix, which contains different physical quantities (such as pressure deviation 、Hull roll angle deviation and hydraulic system load deviation covariance and variance between them).

[0157]

[0158] in, is the maximum eigenvalue of the covariance matrix, indicating the dominant direction of abnormal diffusion, is the minimum eigenvalue of the covariance matrix, indicating the stability of the system; Var represents variance and Cov represents covariance.

[0159] The intensity and stability of anomaly diffusion are measured by eigenvalues. The maximum eigenvalue represents the dominant direction of the anomaly in the system, while the minimum eigenvalue reflects the stability of the system. Using the square root of the minimum eigenvalue as the denominator avoids the introduction of deviations due to differences in system dimensions, thereby ensuring that the calculation of diffusion capacity is not affected by differences in different physical quantity units.

[0160] Finally, the time series anomaly diffusion index is obtained through the comprehensive calculation of abnormal energy accumulation and diffusion capacity. , for example, is calculated as follows:

[0161]

[0162] The time series anomaly diffusion index is the accumulation of abnormal energy Diffusion capacity The product of is finally calculated by adding 1 to the diffusion capacity and taking the logarithm to avoid numerical overflow when the diffusion capacity is close to zero.

[0163] When the abnormal energy accumulation is high and the abnormal diffusion ability is strong, the time series abnormal diffusion index will increase significantly, which means that the abnormality may trigger a chain reaction. Therefore, the time series abnormal diffusion index is used as a comprehensive indicator to evaluate the diffusion potential of the abnormality and help determine whether there is a possibility of risk expansion.

[0164] Step S503, the logic for obtaining the environment-physical model mismatch index is as follows:

[0165] Boundary conflict detection calculates the deviation between the current environmental parameters and the preset safety threshold of the physical model, and accumulates it if it exceeds the limit:

[0166]

[0167] in, and Respectively represent the current environmental parameters and the safety thresholds preset by the model, Represents the total number of environmental parameters, Indicates the index of the specific environment parameter currently being processed. It is an indicator function. When the environmental parameter exceeds the safety threshold, the square of its relative deviation is calculated, otherwise it is not taken into account.

[0168] Deviation calculation is performed only when the environmental parameter exceeds the safety threshold, and the risk of non-compliance with the safety condition is aggravated by the square of the deviation. The safety threshold is used as the denominator to ensure that the unit difference of different environmental parameters does not affect the uniformity of the results.

[0169] By calculating the failure frequency in similar historical scenarios, the failure probability in the current environment can be directly estimated:

[0170]

[0171] in, Indicates the history of the work Boundary conflict value at the time of operation, is the total sample size of the historical data set.

[0172] If the current environment does not exceed the model's safety threshold (i.e. ), the environment-physical model mismatch index is zero, indicating that the environment is risk-free; if there are excessive parameters ( ), the environment-physical model mismatch index takes the failure probability of similar historical scenarios.

[0173] Environment-Physical Model Mismatch Index:

[0174]

[0175] Finally, the risk factor The calculation and judgment process is as follows:

[0176]

[0177] in, , It is a preset proportional coefficient used to weight the contribution of the time series anomaly diffusion index and the environment-physical model mismatch index. The specific value can be adjusted according to the actual scenario and needs.

[0178] Step S504, the warning determination rules are as follows:

[0179] When the risk factor is greater than or equal to the second critical value of the risk, a first-level warning signal is generated, construction is immediately interrupted, the rake head is recovered, and the emergency stabilization system is activated.

[0180] When the risk factor is greater than or equal to the first critical value of risk and less than the second critical value of risk, a secondary warning signal is generated, and the ship speed and wave compensation parameters are automatically adjusted, and manual confirmation is required on site.

[0181] When the risk factor is less than the first critical value of the risk, a third-level warning signal is generated, which mainly records the abnormality and makes subsequent optimization adjustments. At this time, the environment and equipment operation are still within a relatively safe range and are only saved as warning information to facilitate future optimization of paths or operations.

[0182] When the secondary warning signal is obtained, the input parameters are inferred using the trained LSTM neural network model, and the rake head and rake arm postures that meet safety conditions are generated through optimization algorithms (such as genetic algorithms). The specific steps are as follows:

[0183] Input abnormal data. Input the detected abnormal data (such as rake head pressure and posture deviation value) into the LSTM neural network model.

[0184] Inversely infer input parameters. Through the back propagation algorithm, calculate the input parameters that lead to abnormal output (such as ship speed, ship attitude, environmental parameters, etc.).

[0185] Identify key parameters. Analyze the inverse results and identify the input parameters that have the greatest impact on abnormal output (such as wave height, surge velocity, etc.).

[0186] Step 6: Conduct emergency response and real-time control feedback. When the secondary warning signal is obtained, timely generate response suggestions, optimize construction parameters through reinforcement learning algorithm, dynamically adjust the rake head posture and ship position, and improve construction efficiency and safety. The specific steps are as follows:

[0187] Step S601, parameter optimization:

[0188] Aiming at the safe posture of the drag head and the drag arm, a genetic algorithm is used to optimize the input parameters (such as ship speed, ship posture, mud pump speed, etc.).

[0189] A set of input parameters (such as ship speed, ship attitude, mud pump speed, etc.) is randomly generated as the initial population.

[0190] Define a fitness function to measure whether the rake head and rake arm postures corresponding to each set of input parameters meet the safety conditions. For example:

[0191]

[0192] in, is the predicted rake head and rake arm posture, It is the safe posture range.

[0193] Through selection, crossover and mutation operations, a new generation of population is generated and the input parameters are gradually optimized.

[0194] The above process is repeated for iterative optimization until an input parameter combination that meets the safety conditions is found.

[0195] Step S602, outputting the optimization results: outputting the optimized input parameters (such as ship speed, ship attitude, mud pump speed, etc.) to the central control system.

[0196] Step S603, generating response suggestions: generating specific response suggestions based on the optimization results to ensure the safety of personnel and equipment, for example:

[0197] Adjust boat speed: Reduce boat speed to reduce rake head pressure.

[0198] Adjust the ship's attitude: restore the safe attitude of the rake head and rake arm by adjusting the roll and pitch angles of the ship.

[0199] Adjust the mud pump speed: reduce the mud pump speed to reduce the force on the rake head.

[0200] Stop construction: When the optimization results cannot meet safety conditions, it is recommended to stop construction immediately and evacuate the dangerous area.

[0201] Step S604, automatic control or manual intervention: the optimized input parameters are automatically input into the ship control system, and the ship speed, ship attitude and mud pump speed are adjusted in real time to ensure that the rake head and rake arm are restored to a safe attitude. On the basis of automatic control, the operator can make manual adjustments according to the actual situation to further improve the construction safety.

[0202] Step S605, automatic feedback: feeding back the optimized construction parameters and actual effects to the LSTM neural network model for online learning and continuous optimization of the model.

[0203] Through the above design, the LSTM neural network can accurately predict the posture changes of the rake head and rake arm, and combine with physical constraints to improve the prediction accuracy and physical consistency of the model, providing reliable technical support for dredging construction under high sea conditions.

[0204] Embodiment 2: Figure 5 The invention provides a construction state prediction system of a trailing suction hopper vessel based on physical data dual drive, comprising: a data tracing module, a simulation mapping module, an intelligent driving module, a risk anchoring module and an adaptive control module;

[0205] Data tracing module: collects multi-source sensor data, performs data cleaning and standardization, builds training data sets, and passes the training data sets to the simulation mapping module for physical modeling;

[0206] Simulation mapping module: establishes the coupled dynamic model of the hull-raker arm-raker head-wave compensator-hoisting system, combines numerical simulation to simulate the dynamic response of the equipment under high sea conditions, and outputs physical modeling data to provide prediction support for the intelligent drive module;

[0207] Intelligent driving module: integrates historical data and numerical simulation results, builds a physically constrained LSTM model, optimizes the loss function through the dynamic equation, optimizes the posture prediction model on the 3D modeling simulation platform, outputs the predicted posture data of the rake head and rake arm, and delivers it to the risk anchoring module for safety assessment;

[0208] Risk anchoring module: Analyzes abnormal energy accumulation and diffusion capabilities, combines the current environmental parameters with the safety threshold deviation of the physical model, associates the historical failure probability to assess the risk level, and reversely infers the input parameters through the LSTM model to generate optimization conditions, evaluate abnormal conditions and optimize input parameters for the adaptive control module to adjust the posture;

[0209] Adaptive control module: Dynamically optimize construction control parameters through genetic algorithms, adjust the rake head and rake arm postures, and feed back the optimization results to the model to achieve closed-loop iteration, output the optimized control parameters, and feed back the adjusted construction data to the data traceability module to continuously improve the model accuracy.

[0210] The above-mentioned time series anomaly diffusion index, environment-physical model mismatch index and risk factor calculation formula adopts dimensionless numerical value for calculation. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0211] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0212] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0213] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive, characterized in that: Includes steps: S1: Use multi-source sensors to collect construction-related data in real time and perform cleaning and standardization to build a training data set; S2: Establish a coupled dynamic model of the hull-raker arm-raker head-wave compensator-hoisting system, and combine numerical simulation to simulate the dynamic response of the equipment under high sea conditions; S3: Integrate historical data and numerical simulation results to build a physically constrained LSTM model, optimize the loss function through the dynamic equation and complete training verification; S4: Integrate 3D modeling and simulation platform to optimize the attitude prediction model, and combine physical simulation and data-driven model to output prediction results; S5: By analyzing the abnormal energy accumulation and diffusion capabilities, combining the deviation between the current environmental parameters and the safety threshold preset by the physical model, and associating the historical failure probability assessment to present the risk level, the input parameters are finally inferred through the LSTM model to generate the safety posture optimization conditions of the rake head and rake arm; S6: Dynamically optimize construction control parameters through genetic algorithms, perform posture adjustments and feed back optimization results to the model to achieve closed-loop iteration.

2. The method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive according to claim 1 is characterized in that: Step S2 includes the following contents: Step S201, by establishing the hull kinematic equation of the trailing suction hopper, considering the motion response of the hull in multiple degrees of freedom, and calculating the influence of the hull mass, additional mass and moment of inertia on the motion; Step S202: simulating the dynamic response of the rake head and the rake arm by establishing kinematic equations of the rake head and the rake arm, calculating the displacements in three directions and the effects of soil, fluid and water flow resistance; Step S203, by establishing a physical model of the wave compensator, calculating its damping and elastic coefficients, and simulating its hysteresis effect, reflecting the dynamic influence of the wave compensator on the posture of the rake head and rake arm; Step S204, input environmental parameters and construction parameters, simulate the dynamic response of the hull and equipment under high sea conditions through finite element analysis and multi-body dynamics simulation methods, and predict the real-time posture changes of the rake head and rake arm.

3. The method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive according to claim 2 is characterized in that: Step S3 includes the following contents: Step S301, collect historically collected ship speed, ship attitude, and environmental parameter data, perform preprocessing and extract key features, and construct a unified training data set; Step S302: constructing an LSTM neural network model to predict the posture changes of the rake head and rake arm through time series data; Step S303: taking the coupled dynamics model of the rake head and the rake arm as a constraint condition and incorporating it into the loss function, the total loss function being a weighted sum of the prediction error term and the physical constraint term; Step S304: Adopting an adaptive optimization algorithm to optimize the hyperparameters of the LSTM model to improve the convergence speed and prediction accuracy of the network; Step S305: Use numerical simulation data to verify the prediction accuracy of the LSTM model and evaluate its performance in practical applications.

4. The method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive according to claim 3 is characterized by: Step S5, based on the real-time collected data, using big data analysis technology to identify potential risks and assess their severity; The risk is comprehensively assessed through the time series anomaly diffusion index and the environment-physical model mismatch index. The risk coefficient is calculated by weighted summation of the time series anomaly diffusion index and the environment-physical model mismatch index. The calculation formulas for the time series anomaly diffusion index, the environment-physical model mismatch index and the risk coefficient adopt the dimensionless numerical value for calculation.

5. The method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive according to claim 4 is characterized in that: The logic for obtaining the time series anomaly diffusion index is as follows: Abnormal energy accumulation in time: ,in, is the abnormal pressure deviation at a point in time, : Environmental dynamic damping coefficient, is the horizontal velocity of the wave, is the ocean current velocity; is the reference flow rate; the characteristic value is used to measure the intensity and stability of abnormal diffusion and evaluate the diffusion capacity: ,in, is the covariance matrix, where is the maximum eigenvalue of the covariance matrix, is the minimum eigenvalue of the covariance matrix. Finally, the time series anomaly diffusion index is obtained by comprehensive calculation of anomaly energy accumulation and diffusion capacity. .

6. The method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive according to claim 4 is characterized in that: The logic for obtaining the environment-physical model mismatch index is as follows: Boundary conflict detection calculates the deviation between the current environmental parameters and the preset safety threshold of the physical model, and accumulates it if it exceeds the limit: ,in, and Respectively represent the current environmental parameters and the safety thresholds preset by the model, Represents the total number of environmental parameters, Indicates the index of the specific environment parameter currently being processed. It is an indicator function. When the environmental parameter exceeds the safety threshold, the square of its relative deviation is calculated, otherwise it is not taken into account. By calculating the failure frequency in similar historical scenarios, the failure probability in the current environment is directly estimated: ,in, Indicates the history of the work Boundary conflict value at the time of operation, is the total sample size of the historical data set; environment-physical model mismatch index: .

7. The method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive according to claim 4 is characterized in that: The warning judgment rules are as follows: When the risk coefficient is greater than or equal to the second critical value of risk, a first-level warning signal is generated and construction is immediately interrupted; when the risk coefficient is greater than or equal to the first critical value of risk and less than the second critical value of risk, a second-level warning signal is generated and the ship speed and wave compensation parameters are automatically adjusted; when the risk coefficient is less than the first critical value of risk, a third-level warning signal is generated, the abnormality is recorded and subsequent optimization and adjustment are carried out.

8. The method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive according to claim 7 is characterized by: When the second-level warning signal is obtained, the trained LSTM neural network model is used to reverse the input parameters, and the optimization algorithm is used to generate the rake head and rake arm postures that meet safety conditions.

9. The method for predicting the construction status of a trailing suction hopper vessel based on physical data dual drive according to claim 8, characterized in that: Step S6 includes the following contents: Step S601, taking the safe posture of the rake head and rake arm as the goal, using genetic algorithm to optimize input parameters, randomly generating a group of input parameters as the initial population, defining a fitness function, measuring whether the rake head and rake arm posture corresponding to each group of input parameters meets the safety conditions, generating a new generation of population through selection, crossover and mutation operations, gradually optimizing the input parameters, repeating this process for iterative optimization, until a combination of input parameters that meets the safety conditions is found; Step S602, outputting the optimized input parameters to the central control system; Step S603, generating specific response suggestions based on the optimization results to ensure the safety of personnel and equipment; Step S604, automatically inputting the optimized input parameters into the ship control system, adjusting the ship speed, ship attitude and mud pump speed in real time, and ensuring that the rake head and rake arm return to a safe attitude; Step S605, feeding back the optimized construction parameters and actual effects to the LSTM neural network model for online learning and continuous optimization of the model.

10. A system for predicting the construction status of a trailing hopper vessel based on dual-drive of physical data, used to implement the method for predicting the construction status of a trailing hopper vessel based on dual-drive of physical data as claimed in any one of claims 1 to 9, characterized in that: include: Data tracing module, simulation mapping module, intelligent driving module, risk anchoring module and adaptive control module; Data tracing module: collects multi-source sensor data, performs data cleaning and standardization, builds training data sets, and passes the training data sets to the simulation mapping module for physical modeling; Simulation mapping module: establishes the coupled dynamic model of the hull-raker arm-raker head-wave compensator-hoisting system, combines numerical simulation to simulate the dynamic response of the equipment under high sea conditions, and outputs physical modeling data to provide prediction support for the intelligent drive module; Intelligent driving module: integrates historical data and numerical simulation results, builds a physically constrained LSTM model, optimizes the loss function through the dynamic equation, optimizes the posture prediction model on the 3D modeling simulation platform, outputs the predicted posture data of the rake head and rake arm, and delivers it to the risk anchoring module for safety assessment; Risk anchoring module: Analyzes abnormal energy accumulation and diffusion capabilities, combines the current environmental parameters with the safety threshold deviation of the physical model, associates the historical failure probability to assess the risk level, and reversely infers the input parameters through the LSTM model to generate optimization conditions, evaluate abnormal conditions and optimize input parameters for the adaptive control module to adjust the posture; Adaptive control module: Dynamically optimize construction control parameters through genetic algorithms, adjust the rake head and rake arm postures, and feed back the optimization results to the model to achieve closed-loop iteration, output optimized control parameters, and feed back the adjusted construction data to the data traceability module to continuously improve model accuracy.

Images