A six-degree-of-freedom underwater unmanned vehicle online model identification method

Abstract

The application provides a kind of six degrees of freedom underwater unmanned vehicle online model identification method, belongs to the technical field of system modeling of underwater unmanned vehicle.The application adopts multi-head self-attention mechanism based on steering control data stream to the time series and state variables of vehicle respectively, constructs the model identification strategy of underwater unmanned vehicle real state data missing, and based on this, an end-to-end underwater unmanned vehicle online model identification deep learning method with encoder-decoder structure is proposed.The method of the application can estimate or reconstruct the missing state data, and can establish an accurate six degrees of freedom underwater unmanned vehicle model in real time with small computational cost, so as to ensure that the vehicle can work stably and reliably in the marine environment.

Description

Technical Field

[0001] This invention relates to the field of system modeling technology for underwater unmanned vehicles, and more particularly to an online model identification method for a six-degree-of-freedom underwater unmanned vehicle. Background Technology

[0002] As crucial equipment for ocean exploration, underwater unmanned vehicles (UAVs) have attracted significant attention. Characterized by their small size, light weight, and high maneuverability, UAVs can navigate autonomously in complex and ever-changing marine environments, playing a vital role in ocean exploration, seabed search and rescue, and marine resource development. However, due to the highly nonlinear and strongly coupled hydrodynamic characteristics of UAVs, some of their dynamic parameters change with variations in the marine environment and load.

[0003] Accurate dynamic models are crucial for underwater unmanned vehicles (UAVs), forming the foundation for their navigation, guidance, and control system design. They not only help design teams reduce testing costs and identify and resolve potential problems before actual deployment, but also optimize vehicle performance, improving maneuverability and mission efficiency. Furthermore, models identified during navigation aid in fault diagnosis, ensuring stable and reliable operation in complex and variable underwater environments, ultimately achieving higher energy efficiency and longer operational time.

[0004] Model identification can identify underwater unmanned vehicle models online or offline at a relatively low cost. Classic model identification methods include maximum likelihood estimation and model reference adaptation. However, these methods suffer from excessive dependence on mathematical models and initial values. Numerical simulation methods mostly use fluid dynamics software, but the computational load is large, and the results are easily affected by the quality of the model mesh and boundary conditions. Model testing methods, while providing high accuracy in obtaining hydrodynamic parameters, are costly, time-consuming, and inevitably affected by sensor and operational errors. Summary of the Invention

[0005] To address the technical problems of existing underwater unmanned vehicle (UAV) model identification algorithms, this invention provides an online model identification method for six-DOF underwater unmanned vehicles. This invention offers an end-to-end deep learning algorithm for online model identification of underwater unmanned vehicles with an encoder-decoder structure. For time series data and vehicle state variables, a multi-head self-attention mechanism based on manipulation and control data flow is designed to construct a model identification framework for situations where real-world UAV state data is missing. This algorithm can establish accurate six-DOF underwater unmanned vehicle models in real time with relatively low computational cost.

[0006] The technical means employed in this invention are as follows: A method for online model identification of a six-degree-of-freedom underwater unmanned vehicle includes the following steps: Step 1: Derive the dynamic model of the six-degree-of-freedom underwater unmanned vehicle, use the collected navigation data as training data, input it into the online model identification method of the six-degree-of-freedom underwater unmanned vehicle, and identify the six-degree-of-freedom underwater unmanned vehicle model; Step 2: Design a multi-head self-attention mechanism based on the manipulation control data flow between the time series and the state variables of the six-degree-of-freedom underwater unmanned vehicle to mine the implicit information of time series correlation, interdependence and nonlinear relationship between state variables, and environmental influence in the data. Step 3: Construct a strategy to identify missing real-state data; estimate or reconstruct the missing state data; Step 4: Construct an end-to-end encoder-decoder structure. The output of the decoder is the current state of the six-DOF underwater unmanned vehicle, thereby obtaining the identification result. The end-to-end encoder-decoder structure includes an encoder and a decoder.

[0007] Furthermore, in step 1, according to the Newton-Euler equations of motion for a rigid body in a fluid, the six-degree-of-freedom dynamic model in the motion coordinate system is as follows: ; ; in, This represents the transformation matrix between the fixed coordinate system and the moving coordinate system. This represents the resultant force and torque of the thrust of the underwater unmanned vehicle and the disturbances it experiences in the water. This represents the linear and angular velocity vectors of the underwater unmanned vehicle in the motion coordinate system. This represents the mass and inertia matrix of the six-degree-of-freedom underwater unmanned vehicle. Represent the Coriolis and centripetal force matrices; and we have: ; in, Represents the rigid body's mass and inertia matrix. Represents the hydrodynamic added mass matrix; ; ; in, Indicates the mass of an underwater unmanned vehicle. This represents the inertia term of the six-degree-of-freedom underwater unmanned vehicle; X , Y , Z and K , M , NThese represent the components of force and torque on each axis of the aircraft, respectively, and the center of gravity is represented as... The buoyancy center is represented as .

[0008] Furthermore, the Coriolis and centripetal force matrix Includes: rigid body centripetal force matrix and the added mass inertia matrix The resulting Coriolis-like force matrix ,Right now: ; in,

[0009] ; Other: ; ; ; ; ; ； in, Let represent the fluid damping matrix, and we have: ; ; ; in, This represents the restoring force vector, or torque vector, generated by the gravity and buoyancy of the six-degree-of-freedom underwater unmanned vehicle. and Let represent the gravity and buoyancy of the six-degree-of-freedom underwater unmanned vehicle, respectively, and let the center of gravity be denoted as . Buoyancy is expressed as .

[0010] Furthermore, the multi-head self-attention mechanism generates a query, a key, and a value from the input sequence through a linear transformation; The weights of the Value vector are calculated based on the dot product of multiple Queries and multiple Keys; an attention score is calculated by performing a dot product operation between each query vector and its corresponding Key vector; the attention score is normalized using a softmax function; finally, the normalized attention weights are used to perform a weighted summation of the Value vectors; the calculation formula for the multi-head self-attention mechanism is as follows: ; To avoid gradient vanishing or exploding during training, the dot product result is divided by a scaling factor. .

[0011] Furthermore, in step 3, a strategy for identifying missing real state data is constructed; the missing state data is estimated or reconstructed. The missing data in the actual state of the aircraft is represented as follows: The aircraft state prediction data is represented as The result after the identification strategy reorganizes the data is represented as follows: Suppose that at some point, some of the vehicle's state data is lost or inaccurate due to sensor malfunction and noise interference. The vehicle's state data will be reconstructed as follows: Then it enters the next model identification cycle until the six-degree-of-freedom underwater unmanned vehicle stops operating.

[0012] Furthermore, the encoder includes: Each layer consists of a multi-head self-attention module and a fully connected feedforward network; both the multi-head self-attention module and the fully connected feedforward network contain residual connections and are normalized. The encoder's input is the previous sequence ordered in chronological order. k The vehicle status data at each point in time includes: the vehicle's speed. angular velocity of the aircraft propeller speed of the aircraft and rudder angle ; By adding To promote residual connectivity, where Indicates a function implemented by a sub-layer; The output of any encoder layer is represented as: ; The output of each encoder, along with the output of the initial embedding layer, enters the decoder layer.

[0013] Furthermore, the decoder includes The same layer; The decoder has the same multi-head self-attention module and feedforward network as the encoder; The decoder applies a multi-head self-attention mechanism to the encoder's output, constructing a third sub-layer of the decoder; Both the multi-head self-attention module and the fully connected feedforward network contain residual connections and are normalized. The decoder generates There are outputs, and the output of each layer is represented as follows: ; The decoder first transforms the output of each layer through linear projection, converting the feature representation generated by each layer into a prediction vector; by weighting and summing these individual prediction vectors, the final prediction result of the model is generated.

[0014] Compared with the prior art, the present invention has the following advantages: 1. The deep learning method for online model recognition of underwater unmanned vehicles provided by this invention achieves the elimination of complex feature extraction and preprocessing steps without human design through an end-to-end encoder-decoder structure.

[0015] 2. The multi-head self-attention mechanism considering the manipulation and control data flow provided by this invention, by independently applying the multi-head self-attention mechanism between time series and aircraft state variables, mines implicit information such as time series correlation, interdependence between state variables, nonlinear relationships and environmental influences in the data, and achieves higher identification efficiency and accuracy.

[0016] 3. The model identification framework for missing real-state data provided by this invention enables model identification of the vehicle even when sensor data is missing or inaccurate by estimating or reconstructing the missing state data, thereby ensuring that the vehicle can work stably and reliably in the marine environment.

[0017] Therefore, the technical solution of the present invention solves the problems of insufficient modeling, inaccurate identification, lack of real state data leading to unstable operation, need for data preprocessing, and need for different algorithm structures to be designed for different dimensions of the vehicle in the prior art. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This invention provides an online model identification framework for underwater unmanned vehicles.

[0020] Figure 2 This is a schematic diagram of the self-attention mechanism of the present invention.

[0021] Figure 3 This is the identification strategy for missing true state data in this invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] like Figure 1-3 As shown, this invention provides an online model identification method for a six-degree-of-freedom underwater unmanned vehicle, comprising the following steps: Step 1: Derive the dynamic model of the six-degree-of-freedom underwater unmanned vehicle, use the collected navigation data as training data, input it into the online model identification method of the six-degree-of-freedom underwater unmanned vehicle, and identify the six-degree-of-freedom underwater unmanned vehicle model; Step 2: Design a multi-head self-attention mechanism based on the manipulation control data flow for both the time series and the state variables of the six-degree-of-freedom underwater unmanned vehicle, respectively, to mine the implicit information of time series correlation, interdependence and nonlinear relationship between state variables, and environmental influence in the data.

[0025] Step 3: Construct a strategy to identify missing real-state data (real-state data of the aircraft transmitted back by sensors during the aircraft's flight); estimate or reconstruct the missing state data; Step 4: Construct an end-to-end encoder-decoder structure. The output of the decoder is the current state of the six-DOF underwater unmanned vehicle, thereby obtaining the identification result. The end-to-end encoder-decoder structure includes an encoder and a decoder.

[0026] The encoder's output is represented as the output of each layer. Finally, the output of each encoder, along with the output of the initial embedding layer, enters the decoder layer.

[0027] The output of the decoder layer is represented as .

[0028] As a preferred embodiment, in this application, in step 1, according to the Newton-Euler equations of motion for a rigid body in a fluid, the six-degree-of-freedom dynamic model in the motion coordinate system is: ; ; in, This represents the transformation matrix between the fixed coordinate system and the moving coordinate system. This represents the resultant force and torque of the thrust of the underwater unmanned vehicle and the disturbances it experiences in the water. This represents the linear and angular velocity vectors of the underwater unmanned vehicle in the motion coordinate system. This represents the mass and inertia matrix of the six-degree-of-freedom underwater unmanned vehicle. Represent the Coriolis and centripetal force matrices; and we have: ; in, Represents the rigid body's mass and inertia matrix. Represents the hydrodynamic added mass matrix; ; ; in, Indicates the mass of an underwater unmanned vehicle. This represents the inertia term of the six-degree-of-freedom underwater unmanned vehicle; X , Y , Z and K , M , N These represent the components of force and torque on each axis of the aircraft, respectively, and the center of gravity is represented as... The buoyancy center is represented as .

[0029] The Coriolis and centripetal force matrix Includes: rigid body centripetal force matrix and the added mass inertia matrix The resulting Coriolis-like force matrix ,Right now: ; in,

[0030] ; Other: ; ; ; ; ; ； in, Let represent the fluid damping matrix, and we have: ; ; ; in, This represents the restoring force vector, or torque vector, generated by the gravity and buoyancy of the six-degree-of-freedom underwater unmanned vehicle. and Let represent the gravity and buoyancy of the six-degree-of-freedom underwater unmanned vehicle, respectively, and let the center of gravity be denoted as . Buoyancy is expressed as .

[0031] As a preferred implementation method, in the online model identification process of underwater unmanned vehicles, independently applying a multi-head self-attention mechanism between the time series and the vehicle's state variables can significantly extract implicit information from the data, thereby improving the model's identification accuracy and robustness. Figure 2 As shown.

[0032] Sensor data collected chronologically constitutes the time series of the aircraft's state. Since the aircraft's state information at a single point in time is limited, employing a multi-head self-attention mechanism to simultaneously focus on data from different points in time allows the model to better capture long-term dependencies and potential patterns in the time series. For example, when dealing with continuously changing state variables such as velocity and attitude angles, the multi-head self-attention mechanism can identify periodic changes, abrupt changes, and other complex temporal dependency structures.

[0033] In addition to the vehicle's state time series, this invention applies a multi-head self-attention mechanism to the various state variables of the vehicle. The forward velocity is one of the vehicle's state variables. lateral velocity Vertical velocity Roll rate Pitch angular velocity angular velocity of heading propeller speed and rudder angle There are complex nonlinear relationships between them. Multi-head self-attention mechanisms can extract implicit information between variables from multiple state variables in parallel. For example, they can simultaneously focus on the relationship between forward velocity and lateral velocity, as well as the coupling effect between roll rate and pitch rate, thus providing a more comprehensive state estimate.

[0034] The multi-head self-attention mechanism generates queries, keys, and values ​​from the input sequence through a linear transformation; it calculates the weights of values ​​based on the dot product of multiple queries and multiple keys; it calculates an attention score by performing a dot product operation between each query vector and its corresponding key vector; it normalizes the attention scores using a softmax function; and finally, it performs a weighted summation of the value vectors using the normalized attention weights. The calculation formula for the multi-head self-attention mechanism is as follows: ; During the calculation, the dot product result increases with the vector dimension. To avoid the vanishing or exploding gradient problem during training, the dot product result is divided by a scaling factor. .

[0035] Multi-head self-attention mechanisms enable the model to simultaneously capture different locations in the input data and extract multiple features, facilitating the identification of various complex interdependencies among the aircraft's state variables. Even if some "heads" focus on specific details of the input, others can still broadly explore other parts of the input, greatly enriching the model's observation window and processing capabilities. Each "head" maps the raw input data into different representation subspaces by independently linearly projecting the input query, key, and value. In these subspaces, each "head" applies a scaling dot product attention mechanism and has a corresponding weight matrix. Finally, the outputs of all "heads" are connected through an additional linear transformation layer to compute the final result.

[0036] Preferably, in step 3, underwater unmanned vehicles (UAVs) often face harsh marine environments and complex underwater dynamics when performing deep-sea exploration, environmental monitoring, and search and rescue missions, leading to frequent gaps or inaccuracies in their sensor data. Based on this, the present invention designs a strategy for identifying missing real-state data, which can estimate or reconstruct missing state data during the real-time operation of the vehicle, thereby ensuring that the vehicle's control system can operate accurately and reliably. Figure 3 As shown, a strategy for identifying missing real-world state data is constructed; the missing state data is estimated or reconstructed. The missing data in the actual state of the aircraft is represented as follows: The aircraft state prediction data is represented as The result after the identification strategy reorganizes the data is represented as follows: Suppose that at some point, some of the vehicle's state data is lost or inaccurate due to sensor malfunction and noise interference. The vehicle's state data will be reconstructed as follows: Then, the system enters the next model identification cycle until the six-DOF underwater drone stops operating. Through a multi-head self-attention mechanism, the online model identification algorithm can focus on and extract valuable information from previous moments in the time series, filling in missing data. Simultaneously, the system can also mine potential correlations between various state variables, comprehensively utilizing other state data for accurate estimation and prediction.

[0037] In one preferred embodiment, the encoder includes: The same layer; (in this application) =3) Each layer contains: a multi-head self-attention module and a fully connected feedforward network; both the multi-head self-attention module and the fully connected feedforward network contain residual connections and are normalized; The encoder's input is the previous data ordered in chronological order. k The vehicle status data at each point in time includes: the vehicle's speed. angular velocity of the aircraft propeller speed of the aircraft and rudder angle All sub-layers within the model (including the embedding layer) generate dimensions of . The output.

[0038] By adding To promote residual connectivity, where Indicates a function implemented by a sub-layer; The output of any encoder layer is represented as: ; The output of each encoder, along with the output of the initial embedding layer, enters the decoder layer.

[0039] The decoder includes The decoder has the same multi-head self-attention module and feedforward network as the encoder; the decoder applies a multi-head self-attention mechanism to the encoder output, constructing the third sub-layer of the decoder; both the multi-head self-attention module and the fully connected feedforward network contain residual connections and are normalized; the decoder generates... There are outputs, and the output of each layer is represented as follows: The decoder first transforms the output of each layer through linear projection, converting the feature representation generated by each layer into a prediction vector; by weighting and summing these individual prediction vectors, the final prediction result of the model is generated.

[0040] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the above embodiments of the present invention, the descriptions of each embodiment have their own emphasis; parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. It should be understood that the disclosed technical content in the several embodiments provided in this application can be implemented in other ways.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for online model identification of a six-degree-of-freedom underwater unmanned vehicle, characterized in that, Includes the following steps: Step 1: Derive the dynamic model of the six-degree-of-freedom underwater unmanned vehicle (UVU). Use the collected navigation data as training data and input it into the online model identification method for the six-degree-of-freedom UVU to identify the UVU model. In Step 1, based on the Newton-Euler equations of motion for rigid bodies in a fluid, the six-degree-of-freedom dynamic model in the motion coordinate system is: ； ； in, This represents the transformation matrix between the fixed coordinate system and the moving coordinate system. This represents the resultant force and torque of the thrust of the underwater unmanned vehicle and the disturbances it experiences in the water. This represents the linear and angular velocity vectors of the underwater unmanned vehicle in the motion coordinate system. This represents the mass and inertia matrix of the six-degree-of-freedom underwater unmanned vehicle. Represents the Coriolis and centripetal force matrices; Represents the fluid damping matrix; This represents the restoring force vector, i.e., the torque vector, generated by the gravity and buoyancy of the six-degree-of-freedom underwater unmanned vehicle; and we have: ； in, Represents the rigid body's mass and inertia matrix. Represents the hydrodynamic added mass matrix; ； ； in, Indicates the mass of an underwater unmanned vehicle. This represents the inertia term of the six-degree-of-freedom underwater unmanned vehicle. X , Y , Z and K , M , N These represent the components of force and torque on each axis of the aircraft, respectively, and the center of gravity is represented as... The buoyancy center is represented as ; Step 2: Between the time series and the state variables of the six-degree-of-freedom underwater unmanned vehicle, construct a multi-head self-attention mechanism based on the manipulation control data flow for both the time series and the vehicle's state variables, and mine the implicit information of time series correlation, interdependence and nonlinear relationship between state variables, and environmental influence in the data; Step 3: Construct a strategy to identify missing real-state data; estimate or reconstruct the missing state data; Step 4: Construct an end-to-end encoder-decoder structure. The output of the decoder is the current state of the six-DOF underwater unmanned vehicle, thereby obtaining the identification result. The end-to-end encoder-decoder structure includes an encoder and a decoder.

2. The online model identification method for a six-degree-of-freedom underwater unmanned vehicle according to claim 1, characterized in that, The Coriolis and centripetal force matrix Includes: rigid body centripetal force matrix and the added mass inertia matrix The resulting Coriolis force matrix ,Right now: ； in, ； Other: ； ； ； ； ； ； in, Let represent the fluid damping matrix, and we have: ； ； ； in, This represents the restoring force vector, or torque vector, generated by the gravity and buoyancy of the six-degree-of-freedom underwater unmanned vehicle. and Let represent the gravity and buoyancy of the six-degree-of-freedom underwater unmanned vehicle, respectively, and let the center of gravity be denoted as . Buoyancy is expressed as .

3. The online model identification method for a six-degree-of-freedom underwater unmanned vehicle according to claim 1, characterized in that, The multi-head self-attention mechanism generates a query, a key, and a value from the input sequence through a linear transformation. The weights of the Value vector are calculated based on the dot product of multiple Queries and multiple Keys; an attention score is calculated by performing a dot product operation between each query vector and its corresponding Key vector; the attention score is normalized using a softmax function; finally, the normalized attention weights are used to perform a weighted summation of the Value vectors; the calculation formula for the multi-head self-attention mechanism is as follows: ； To avoid gradient vanishing or exploding during training, the dot product result is divided by a scaling factor. .

4. The online model identification method for a six-degree-of-freedom underwater unmanned vehicle according to claim 1, characterized in that, In step 3, a strategy for identifying missing real state data is constructed; the missing state data is estimated or reconstructed. The missing data in the actual state of the aircraft is represented as follows: The aircraft state prediction data is represented as The result after the identification strategy reorganizes the data is represented as follows: Suppose that at some point, some of the vehicle's state data is lost or inaccurate due to sensor malfunction and noise interference. The vehicle's state data will be reconstructed as follows: Then it enters the next model identification cycle until the six-degree-of-freedom underwater unmanned vehicle stops operating.

5. The online model identification method for a six-degree-of-freedom underwater unmanned vehicle according to claim 1, characterized in that, The encoder includes: Each layer consists of a multi-head self-attention module and a fully connected feedforward network; both the multi-head self-attention module and the fully connected feedforward network contain residual connections and are normalized. The encoder's input is the previous data ordered in chronological order. k The vehicle status data at each point in time includes: the vehicle's speed. angular velocity of the aircraft propeller speed of the aircraft and rudder angle ; By adding To promote residual connectivity, where Indicates a function implemented by a sub-layer; The output of any encoder layer is represented as: ； The output of each encoder, along with the output of the initial embedding layer, enters the decoder layer.

6. The online model identification method for a six-degree-of-freedom underwater unmanned vehicle according to claim 5, characterized in that, The decoder includes The same layer; The decoder has the same multi-head self-attention module and feedforward network as the encoder; The decoder applies a multi-head self-attention mechanism to the encoder's output, constructing a third sub-layer of the decoder; Both the multi-head self-attention module and the fully connected feedforward network contain residual connections and are normalized. The decoder generates There are outputs, and the output of each layer is represented as follows: ; The decoder first transforms the output of each layer through linear projection, converting the feature representation generated by each layer into a prediction vector; by weighting and summing these individual prediction vectors, the final prediction result of the model is generated.

Images