CN-121765327-B - Coupling motion prediction method, system, medium and product of deep sea mining lifting system based on physical operators and multiple codes

CN121765327BCN 121765327 BCN121765327 BCN 121765327BCN-121765327-B

Abstract

The invention discloses a coupling motion prediction method, a coupling motion prediction system, a coupling motion prediction medium and a coupling motion prediction product of a deep sea mining lifting system based on physical operators and multiple codes. Firstly, a system high-fidelity dynamics model is established, and multi-working-condition motion response data are obtained as a training set. And introducing a trainable second-order dynamics physical operator into the transducer network, coding the environmental working condition and the continuous time as characteristics, inputting the characteristics into an encoder, and outputting a key section bending moment and an equivalent excitation prediction sequence. By calculating the physical consistency residual error and the data residual error, a multi-domain loss function comprising a time domain error, a frequency domain error, a dynamic residual error and a space smoothness constraint is designed. And combining the optimized model parameters and the physical operators to obtain a trained model, and outputting a prediction result by inputting a target working condition. The method combines prediction precision and dynamic characteristic depiction, improves prediction stability and physical interpretability, and is suitable for the scenes of deep sea mining system design verification, operation monitoring, risk early warning and the like.

Inventors

Guo Yingnuo
KANG JICHUAN
KANG ZHUANG
SUN YU

Assignees

哈尔滨工程大学三亚南海创新发展基地
哈尔滨工程大学

Dates

Publication Date: 20260508
Application Date: 20260302

Claims (7)

1. The coupled motion prediction method of the deep sea mining lifting system based on the physical operators and the multiple codes is characterized by comprising the following steps of: Step 1, establishing a dynamic model of a lifting system, acquiring coupling motion response time sequence data under multiple working conditions, and constructing a training set, wherein the lifting system comprises a water surface support ship, a lifting vertical pipe, a booster pump, a relay station, a flexible hose and a mining vehicle, the dynamic model is modeled by adopting a centralized mass method, hydrodynamic load is calculated by adopting a Morison formula, and environmental conditions comprise wave, ocean current and wind load; Step 2, introducing a trainable second-order dynamics physical operator at the output end of the converter sequence prediction network to construct a prediction model, inputting environmental working condition and continuous time position codes as characteristics to an encoder of the converter sequence prediction network, and outputting a prediction sequence of key monitoring section bending moment and equivalent excitation; equivalent second order dynamics equation of the trainable second order dynamics physical operator: Wherein, the 、 And Respectively an equivalent mass matrix, a damping matrix and a rigidity matrix to be learned, And Respectively predicting first-order and second-order time derivatives of bending moment; Is a bending moment prediction result; quality matrix assurance using Cholesky decomposition parameterization And stiffness matrix Symmetry positive qualitative of (c): Wherein, the And In order for the lower triangular matrix to be trainable, For the regularization constant, Is that A rank identity matrix; Damping matrix The Rayleigh damping form is adopted: Wherein, the The Rayleigh damping coefficient is the Rayleigh damping coefficient, so that the non-negativity is ensured; Step 3, calculating a physical consistency residual error of the prediction sequence based on the second order dynamics physical operator; Physical consistency residual of the predicted sequence In order to achieve this, the first and second, Wherein, the Is the first The number of discrete moments in time is, To approximate the velocity term with the center difference, To approximate the acceleration term with a center difference, Is the predicted equivalent stimulus; Step 4, constructing a multi-domain loss function comprising a time domain prediction error, a frequency domain characteristic error, a dynamic residual error and a space smoothness constraint; step 5, utilizing a training set to perform joint optimization on parameters of an encoder in the predicted model and a trainable second-order dynamics physical operator by taking a minimum multi-domain loss function as a target, so as to obtain a trained predicted model; And 6, inputting target working condition parameters into the trained prediction model, and outputting a coupling motion prediction result of the lifting system.
2. The method for predicting coupled motion of deep sea mining lifting system based on physical operators and multiple codes as recited in claim 1, wherein said ambient condition codes include a sense wave height Spectral peak period Ship speed Direction angles of waves, winds and sea waves Carrying out Min-Max normalization and projecting to obtain conditional feature vectors in time steps ; The continuous time position coding adopts a method based on sine function family to generate time characteristic vectors , The conditional feature vector is added to the temporal encoding as input to the transducer encoder: Stacking the input vectors of all time steps in time sequence to form an input sequence 。
3. The method for predicting the coupling motion of the deep sea mining lifting system based on the physical operators and the multi-codes according to claim 2, wherein the transducer encoder adopts a sequence coding structure based on a multi-head self-attention mechanism, For the first Layer input sequence Projected as a query Key and key Sum value Calculating a scaled dot product attention: After residual connection and layer normalization, the multi-head attention output is processed through a position feedforward network: Warp yarn After layer coding, a bending moment prediction result is obtained through a linear output head And equivalent excitation prediction ; Wherein, the Is the dimension of each of the attention heads, For the input characteristics of the FFN, And As a matrix of weights that can be learned, And Is a learnable bias vector.
4. The method for predicting coupled motion of a deep sea mining lifting system based on physical operators and multiple encodings as set forth in claim 1, wherein said multiple domain loss function is, Wherein, the Fitting the loss for the time domain data, and weighting Huber loss by adopting a channel; loss for physical residual error; is a space smoothing loss based on an Euler-Bernoulli beam theory; is the loss of consistency of the frequency domain amplitude; Amplitude statistics loss; For an equivalent excitation regularization loss, Is a weight coefficient; the weight coefficient is dynamically adjusted by adopting a linear annealing strategy: Wherein, the For the current training round of time, For the annealing cycle time it is desirable that, And The initial and final weights, respectively.
5. A computer system comprising a memory, a processor and a computer program stored on the memory, characterized in that the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 4.
6. A computer-readable storage medium, on which a computer program/instruction is stored, characterized in that the computer program/instruction, when executed by a processor, implements the steps of the method according to any one of claims 1 to 4.
7. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, realizes the steps of the method according to any one of claims 1 to 4.

Description

Coupling motion prediction method, system, medium and product of deep sea mining lifting system based on physical operators and multiple codes Technical Field The invention belongs to the technical field of deep sea mining equipment dynamics modeling and intelligent prediction, and particularly relates to a method, a system, a medium and a product for predicting coupling motion of a deep sea mining lifting system based on a physical operator and multiple codes. Background Deep sea mining lifting systems are typically comprised of a number of subsystems such as surface support vessels, ultra-long lifting risers, relay stations, flexible hoses, mining vehicles, and the like that exhibit significant multi-scale, strong coupling and non-stationary dynamics under the combined action of complex marine environmental loads and operational disturbances. Under the random environmental loads such as wind, waves, currents and the like and the internal slurry conveying action, the system is easy to generate obvious longitudinal, transverse and bending coupling motion responses, wherein the bending moment response of the lifting vertical pipe along the line is an important index for structural safety evaluation, and fatigue or buckling failure can be caused by excessive local bending. The existing high-fidelity numerical simulation method based on finite element or multi-body dynamics can describe the dynamic behavior of the system more accurately, but the modeling process is complex, the calculation cost is high, and the requirements of quick prediction and on-line monitoring are difficult to meet. In recent years, the emerging deep learning prediction method has advantages in terms of calculation efficiency by learning historical response data, but most methods do not explicitly introduce system dynamics physical constraint, and the problems of predicted result violation of physical rules, insufficient generalization capability, strong dependence on training data and the like easily occur. Particularly, in key positions such as ship-riser interfaces, the response is dominated by six-degree-of-freedom ship motion and presents high non-stationarity, so that the traditional neural network method is difficult to achieve acceptable prediction accuracy. Therefore, there is a need for a coupled motion prediction method that can integrate deep learning modeling capability with system dynamics physical constraints to improve physical consistency and engineering reliability of a prediction result while ensuring prediction accuracy. Disclosure of Invention The invention aims to provide a coupling motion prediction method, a coupling motion prediction system, a coupling motion prediction medium and a coupling motion prediction product for a deep sea mining lifting system based on physical operators and multiple codes. The aim of the invention is realized by the following technical scheme: a coupling motion prediction method of a deep sea mining lifting system based on a physical operator and multiple codes comprises the following steps: step 1, establishing a dynamic model of a lifting system, acquiring coupling motion response time sequence data under multiple working conditions, and constructing a training set; Step 2, introducing a trainable second-order dynamics physical operator into an output section of the converter sequence prediction network to construct a prediction model, inputting environmental working condition and continuous time position codes as characteristics to an encoder of the converter sequence prediction network, and outputting a prediction sequence of key monitoring section bending moment and equivalent excitation; Step 3, calculating a physical consistency residual error of the prediction sequence based on the second order dynamics physical operator; Step 4, constructing a multi-domain loss function comprising a time domain prediction error, a frequency domain characteristic error, a dynamic residual error and a space smoothness constraint; step 5, utilizing a training set to perform joint optimization on parameters of an encoder in the predicted model and a trainable second-order dynamics physical operator by taking a minimum multi-domain loss function as a target, so as to obtain a trained predicted model; And 6, inputting target working condition parameters into the trained prediction model, and outputting a coupling motion prediction result of the lifting system. Further, the lifting system comprises a water surface supporting ship, a lifting vertical pipe, a booster pump, a relay station, a flexible hose and a mining vehicle, wherein the dynamic model is modeled by adopting a centralized mass method, the hydrodynamic load is calculated by adopting a Morison formula, and the environmental conditions comprise wave, ocean current and wind load. Further, the ambient condition encoding includes encoding sense wave heightsSpectral peak periodShip speedDirection angles of waves, winds and sea wavesCarrying o