An ocean dynamic feature observation method and system based on information path planning

Abstract

The application provides a kind of marine dynamic feature observation method and system based on information path planning, the method is according to target observation area initialization autonomous underwater vehicle initial position and initial target waypoint, start real-time observation to target feature, when autonomous underwater vehicle runs to initial target waypoint, according to the current real-time observation data collected update the marine dynamic feature model constructed in advance, and after obtaining target feature prediction result according to the updated marine dynamic feature model, whether the target feature observation is sufficient is judged according to the current step length information amount calculated, if the target feature observation is sufficient, then end observation task, otherwise, the next target waypoint is obtained through local information path planning and observation is continued.The application can effectively improve the accuracy and efficiency of marine dynamic feature observation in complex environment while ensuring the autonomy of autonomous underwater vehicle.

Description

Technical Field

[0001] This invention relates to the field of marine observation technology, and in particular to a method, system, computer equipment, and storage medium for observing marine dynamic characteristics based on information path planning. Background Technology

[0002] Ocean dynamics are characterized by complex spatiotemporal variations. Traditional observation methods, such as underway surveys and buoys, are limited by spatiotemporal resolution, making it difficult to reconstruct the field. Autonomous Underwater Vehicles (AUVs), with their high flexibility and autonomy, can carry various observation equipment and achieve autonomous navigation over a large area, giving them an advantage in observing ocean dynamics.

[0003] Most existing intelligent AUV deployment solutions employ a combination of Gaussian process model simulation of observation characteristics and path planning. However, traditional path planning methods pre-determine actions and paths before the task begins, such as the path of a lawnmower on a horizontal plane. This observation method does not fully leverage the autonomous operation advantages of the mobile platform, and due to the limitations of AUV mobility and energy consumption, it is even more difficult to apply to the observation of dynamic ocean features in complex environments. Furthermore, the method of using Gaussian process modeling and the correlation of given data points to estimate the values ​​of other unknown points of interest for simulating spatiotemporally changing ocean dynamic features and path planning suffers from limitations due to the large datasets, resulting in low algorithm efficiency and making it unsuitable for complex path planning scenarios. In short, the current level of intelligent application of AUV mobile platforms cannot guarantee the accuracy and efficiency required for observing dynamic ocean features and urgently needs improvement. Summary of the Invention

[0004] The purpose of this invention is to provide a method for observing marine dynamic features based on information path planning. By combining information path planning with a sparse online Gaussian process model, it enables online construction of dynamic feature models in complex marine environments and real-time information path planning for autonomous underwater vehicles (AUVs). This addresses the shortcomings of existing intelligent AUV usage schemes in marine dynamic feature observation scenarios. Information path planning not only fully leverages the autonomy of AUVs, achieving the most efficient path planning for acquiring marine dynamic feature observations within the constraints of AUV mobility and energy consumption, but also ensures high solution efficiency through a sparse online Gaussian process, effectively improving the accuracy of marine dynamic feature observations and providing a reliable guarantee for obtaining high-quality marine dynamic feature observation data.

[0005] To achieve the above objectives, it is necessary to provide a method and system for observing marine dynamic characteristics based on information path planning, addressing the aforementioned technical problems.

[0006] In a first aspect, embodiments of the present invention provide a method for observing dynamic features of the ocean based on information path planning, the method comprising the following steps:

[0007] Based on the target observation area, initialize the initial position and initial target waypoint of the autonomous underwater vehicle, and begin real-time observation of target features;

[0008] When the autonomous underwater vehicle reaches the initial target waypoint, it collects current real-time observation data; the current real-time observation data includes the vehicle's spatial position and corresponding target feature observation data.

[0009] Based on the current real-time observation data, the pre-constructed ocean dynamic feature model is updated, and the target feature prediction result is obtained based on the updated ocean dynamic feature model.

[0010] Based on the target feature prediction results, calculate the current step size information content, and based on the current step size information content, determine whether the target feature observation is sufficient;

[0011] If sufficient target features are observed, the observation mission ends; otherwise, the next target waypoint is obtained through local information path planning, and observation continues.

[0012] Furthermore, the step of initializing the initial position and initial target waypoint of the autonomous underwater vehicle based on the target observation area includes:

[0013] The target observation area is divided into equally spaced grids according to a preset size to obtain the corresponding region grid.

[0014] The origin of the region grid is initialized to the initial position of the autonomous underwater vehicle;

[0015] The initial target waypoint is obtained based on the initial position, preset running step size, and heading angle; the initial target waypoint is represented as:

[0016] X next-ini =(Rcosψ,Rsinψ)

[0017] Among them, X next-ini R represents the initial target waypoint; R represents the preset running step size; ψ represents the heading angle corresponding to the initial position.

[0018] Furthermore, the ocean dynamics model is a sparse online Gaussian process model;

[0019] The step of initializing the initial position and initial target waypoint of the autonomous underwater vehicle based on the target observation area further includes:

[0020] The mean function and covariance function of the sparse online Gaussian process model are initialized to construct the ocean dynamic characteristic model, and the initial observation data corresponding to the initial position are collected.

[0021] Furthermore, the step of updating the pre-built ocean dynamic characteristic model based on the current real-time observation data includes:

[0022] The space position of the vehicle in the current real-time observation data and the target feature observation data are used as the input and output of the ocean dynamic feature model, respectively, to update the parameters of the ocean dynamic feature model.

[0023] Furthermore, before the step of updating the parameters of the ocean dynamic characteristic model, the method further includes:

[0024] Based on the space position of the vehicle in the current real-time observation data, calculate the position covariance and Gram matrix, and based on the position covariance, the Gram matrix and the target feature observation data, calculate the deviation coefficient corresponding to the current real-time observation data;

[0025] Based on the deviation coefficient and the preset tolerance parameter, determine whether the current real-time observation data is important to the ocean dynamic characteristic model;

[0026] If the current real-time observation data is important, then the current real-time observation data is added to the basic vector set, and the parameters of the ocean dynamic feature model are updated according to the updated basic vector set;

[0027] If the current real-time observation data is not important, there is no need to add the current real-time observation data to the basic vector set. Instead, the parameters of the ocean dynamic feature model can be updated directly by performing a renormalization calculation based on feature space decomposition projection using the existing basic vector set and the current real-time observation data.

[0028] Furthermore, the target feature prediction result includes the mean and covariance matrix of the target features;

[0029] The steps of calculating the current step size information based on the target feature prediction result, and determining whether the target feature observation is sufficient based on the current step size information include:

[0030] The target feature prediction result and the path corresponding to the current step size are input into a preset path information content objective function to obtain the corresponding current step size information content; the path information content objective function is expressed as:

[0031]

[0032] Wherein, P represents all possible paths within the target observation area; Indicates the currently planned path; x represents the spatial location; This represents the absolute value of the gradient of the target feature observation data; This represents the covariance of a normalized Gaussian process. α1 represents the product of the absolute value of the gradient of the target feature observation data and the normalized Gaussian process covariance; α2 and α3 represent the weight coefficients of the absolute value of the gradient of the target feature observation data, the normalized Gaussian process covariance, and the product of the absolute value of the gradient of the target feature observation data and the normalized Gaussian process covariance, respectively. This indicates the amount of information within the currently planned path;

[0033] If the current step size information is less than the preset information threshold, the target feature observation is determined to be sufficient; otherwise, the target feature observation is determined to be insufficient.

[0034] Furthermore, the step of obtaining the next target waypoint through local information path planning includes:

[0035] Based on the current spatial position of the vehicle and the preset operating step size, a set of candidate target waypoints is obtained; the distance between each candidate target waypoint in the set and the current spatial position of the vehicle is the preset operating step size.

[0036] Based on the path information objective function, the candidate class Dobbins path information corresponding to the current space position of the vehicle and each candidate target waypoint is obtained respectively.

[0037] The candidate target waypoint corresponding to the largest candidate type Dobins path information is taken as the next target waypoint.

[0038] Secondly, embodiments of the present invention provide a marine dynamic characteristic observation system based on information path planning, the system comprising:

[0039] The initialization module is used to initialize the initial position and initial target waypoint of the autonomous underwater vehicle according to the target observation area, and to start real-time observation of target features;

[0040] The data acquisition module is used to collect real-time observation data when the autonomous underwater vehicle reaches the target waypoint; the real-time observation data includes the vehicle's spatial position and the corresponding target feature observation data;

[0041] The feature prediction module is used to update the pre-constructed ocean dynamic feature model based on the current real-time observation data, and to obtain the target feature prediction result based on the updated ocean dynamic feature model.

[0042] The feature diagnosis module is used to calculate the current step size information based on the target feature prediction result, and to determine whether the target feature observation is sufficient based on the current step size information.

[0043] The path planning module is used to end the observation task when sufficient target features are observed, and to obtain the next target waypoint through local information path planning when insufficient features are observed, and to continue observation.

[0044] Thirdly, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0045] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0046] The present application provides a method, system, computer device, and storage medium for observing marine dynamic features based on information path planning. After initializing the initial position and initial target waypoint of an autonomous underwater vehicle (AUV) according to the target observation area, the method begins real-time observation of target features. When the AUV reaches the initial target waypoint, it collects current real-time observation data including the vehicle's spatial position and corresponding target feature observation data. The pre-constructed marine dynamic feature model is updated based on the current real-time observation data, and target feature prediction results are obtained based on the updated sparse online Gaussian process model. The current step size information is calculated based on the target feature prediction results, and the sufficiency of target feature observation is determined based on the current step size information. If the target feature observation is sufficient, the observation task ends; otherwise, the next target waypoint is obtained through local information path planning, and observation continues. Compared with existing technologies, this ocean dynamic feature observation method based on information path planning not only fully leverages the autonomy of AUVs through information path planning to achieve the most favorable path planning for obtaining ocean dynamic feature observations under the constraints of AUV mobility and energy consumption, but also ensures high solution efficiency through sparse online Gaussian processes and effectively improves the accuracy of ocean dynamic feature observations, providing a reliable guarantee for obtaining high-quality ocean dynamic feature observation data. Attached Figure Description

[0047] Figure 1 This is a schematic diagram illustrating the application scenario of the marine dynamic feature observation method based on information path planning in this embodiment of the invention;

[0048] Figure 2 This is a schematic diagram of the process framework for ocean dynamic feature observation based on information path planning in an embodiment of the present invention;

[0049] Figure 3 This is a flowchart illustrating the ocean dynamics observation method based on information path planning in an embodiment of the present invention.

[0050] Figure 4 This is another detailed flowchart of the marine dynamic feature observation method based on information path planning in this embodiment of the invention;

[0051] Figure 5 This is a schematic diagram of the actual underwater ice thickness in the ice thickness observation scenario of this invention embodiment;

[0052] Figure 6 This is a schematic diagram of the actual observation path for the ice thickness observation scenario in this embodiment of the invention;

[0053] Figure 7 This is a schematic diagram of the prediction results and observation path of the temperature field observation scenario in an embodiment of the present invention;

[0054] Figure 8 This is a schematic diagram of the marine dynamic characteristic observation system based on information path planning in an embodiment of the present invention;

[0055] Figure 9 This is an internal structural diagram of the computer device in an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and beneficial effects of this application clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the embodiments described below are only part of the embodiments of the present invention and are used to illustrate the present invention, but are not intended to limit the scope of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0057] The ocean dynamic feature observation method based on information path planning provided by this invention can be understood as a method that combines information path planning with a sparse online Gaussian process model, and follows a cyclical observation concept of "planning-guided sampling - sampling-model construction - model-updating planning" to effectively observe the ocean dynamic features of the target of interest. This method can be applied to applications such as... Figure 1 The terminal or server shown. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The server can be implemented using the features provided by this invention, depending on the actual application requirements. Figure 2The information path planning-based ocean dynamic feature observation method architecture shown herein enables efficient and accurate ocean dynamic feature observation, and the obtained ocean dynamic feature observation data is used for subsequent research on the server or transmitted to the terminal for terminal users to view and analyze; the following embodiments will provide a detailed description of the information path planning-based ocean dynamic feature observation method of the present invention.

[0058] In one embodiment, such as Figure 3 As shown, an ocean dynamics observation method based on information path planning is provided, including the following steps:

[0059] S11. Based on the target observation area, initialize the initial position and initial target waypoint of the autonomous underwater vehicle (AUV) and begin real-time observation of target features; wherein, the target observation area can be understood as the required observation area determined according to actual application needs (real observation scenario or simulated observation scenario); correspondingly, the initial position and initial target waypoint of the AUV can be understood as the position to be started and the next target position to be reached by the AUV mapped onto the target observation area in the actual application scenario, which can be determined according to the position, size, shape and other characteristics of the target observation area; specifically, the step of initializing the initial position and initial target waypoint of the AUV based on the target observation area includes:

[0060] The target observation area is divided into equally spaced grids according to a preset size to obtain the corresponding area grid; wherein, the preset size can be understood as the latitude and longitude size, and the specific size can be set according to the actual application requirements, such as dividing the target observation area into 100*100 data grids at equal intervals along the longitude and latitude directions, etc., without specific limitation here;

[0061] The origin of the region grid is initialized as the initial position of the autonomous underwater vehicle (AUV); where the initial position of the AUV is the position of the AUV at the start time t=1, which is the origin X. t＝1 = (0,0);

[0062] The initial target waypoint is obtained based on the initial position, preset running step size, and heading angle; the initial target waypoint is represented as:

[0063] X next-ini =(Rcosψ,Rsinψ)

[0064] Among them, X next-ini R represents the initial target waypoint; R represents the preset running step size; ψ represents the heading angle corresponding to the initial position. It should be noted that the preset running step size and the heading angle corresponding to the initial position can be set according to application requirements, and no specific limitation is made here.

[0065] In practical applications, the next target waypoint corresponding to the initial position (initial target waypoint) has already been determined by the above method and does not need to be obtained through information path planning. However, considering that the ocean dynamic features within a certain observation area have a certain spatial correlation, in order to facilitate the accumulation of samples for training the ocean dynamic feature model and thus ensure the effectiveness of subsequent model parameter updates, this embodiment preferably initializes and constructs the ocean dynamic feature model after determining the initial position by utilizing prior knowledge of the target observation area. At the same time, considering that although Gaussian processes can estimate the values ​​of other unknown points of interest by modeling and using the correlation of given data points to simulate the spatiotemporally changing ocean dynamic features, they have the limitation of excessively large datasets, making it difficult to guarantee computational efficiency. Therefore, this embodiment preferably uses a sparse online Gaussian process model as the ocean dynamic feature model for simulating and constructing ocean dynamic features and corresponding AUV real-time information path planning. It should be noted that the sparse online Gaussian process model can be understood as a sparse representation of the Gaussian process model. It can completely characterize the sequence structure of the relevant subsample data of the Gaussian process model, overcome the limitation of excessively large Gaussian process datasets, and can achieve fast computation, thereby ensuring the efficiency of simulating and constructing ocean dynamic features.

[0066] For sparse online Gaussian processes in practical applications, let the input be denoted as x and the output as... When the time variable t is introduced, the Gaussian process model at time (t+1) can be expressed as follows according to the Bayesian online algorithm: (the values ​​at test point x are obtained respectively) * The estimated value and covariance at point K t (x, x′) represents the covariance function of the model at time t.

[0067]

[0068]

[0069] Where, x * and x t+1 Let represent the test point and the new input at time (t+1), respectively; and Let K represent the estimated value of the test point at time (t+1), the estimated value of the test point at time t, and the covariance at time (t+1), respectively; t (x * x t+1 ), K t (x * x * ), K t (x * x t+1 ), K t (xt+1 x * ) is calculated from the covariance function at time t; q (t+1) and r (t+1) They are respectively the expected likelihood The first and second derivatives of the logarithm, assuming the expected likelihood follows a normal distribution, are:

[0070]

[0071]

[0072] in, for The noise figure; This represents the approximate logarithm of the likelihood function at time t;

[0073] Then, at time t, the parameterization of the approximate posterior Gaussian process can be expressed as:

[0074]

[0075]

[0076] In the formula,

[0077] α t =[α t (1),…,α t (t)] T

[0078] C t ={C t (ij)} i,j＝1,t

[0079]

[0080] Where K0(x,x′) represents the covariance function at time t; α t and C t They represent the approximate parameters defined by α. t+1 =T t+1 (α t )+q (t+1) s t+1 , It is obtained through iterative calculation, where the operator T t+1 and U t+1 This indicates incrementing the length or size of a vector or matrix by one and filling empty spaces with 0, e t+1 =[0,0,…,1] T This represents a unit vector of length (t+1).

[0081] Specifically, the step of initializing the initial position and initial target waypoint of the autonomous underwater vehicle based on the target observation area further includes:

[0082] The mean and covariance functions of the sparse online Gaussian process model are initialized to construct the ocean dynamic characteristic model, and initial observation data corresponding to the initial position are collected. The initial observation data consists of target characteristic data collected by multiple onboard sensors at the initial AUV spatial position. These data are characterized by one or more physical quantities representing the ocean features / phenomena of interest, such as temperature characteristics representing the dynamic temperature field of seawater, water depth characteristics representing topography, and salinity characteristics corresponding to the topography. Temperature and salinity characteristics can be collected by a Conductivity Temperature Depth (CTD) sensor, and water depth characteristics can be obtained by an altimeter sensor. It should be noted that the method of obtaining the AUV's spatial position depends on the actual operating state of the AUV. When the AUV is on the water surface, it can be obtained from GPS data, while underwater, it can be obtained from a Doppler velocity measurement (DVL) inertial navigation system or an ultra-short baseline (USBL) acoustic navigation system. Details are omitted here.

[0083] The construction process of the above-mentioned ocean dynamic characteristic model can be understood as obtaining it through the initialization of a sparse online Gaussian process model. In this embodiment, the mean function of the sparse online Gaussian process model is preferably defined as a constant and initialized to 0, and the covariance function is defined as having automatic correlation determination.

[0084] The Automatic Relevance Determination (ARD) distance metric uses a squared exponential (SE) covariance function, initialized as a squared exponential covariance function K0(x,x′), where the corresponding initialization hyperparameter θ = [σ]. f ,σ ω [l1,l2] T Where, σ f σ ω l1 and l2 represent the signal variance, noise variance, and one-dimensional and two-dimensional length scales, respectively.

[0085] In practical applications after initialization, during the observation process of an AUV moving from one waypoint to the next target waypoint, real-time data from multiple different sensors are acquired at a certain sampling frequency (a relatively low sampling frequency, with one set of data collected for a preset step size), i.e., the spatial position x at a certain time t. t Observational data obtained at the location By observing the spatial location data of the AUV x tAs input to the ocean dynamics model at time t, and the target feature data collected by each sensor As the output at time t, update the sparse online Gaussian process model according to the steps described below.

[0086] S12. When the autonomous underwater vehicle reaches the initial target waypoint, it collects current real-time observation data. The current real-time observation data includes the vehicle's spatial position and the corresponding target feature observation data. It should be noted that the methods for collecting the vehicle's spatial position and target feature observation data at the initial target waypoint and other subsequent target waypoints can refer to the relevant description of the initial observation data above, and will not be repeated here.

[0087] S13. Based on the current real-time observation data, update the pre-constructed ocean dynamic feature model, and obtain the target feature prediction result based on the updated ocean dynamic feature model; wherein, updating the ocean dynamic feature model can be understood as the process of using newly acquired current real-time observation data and historically acquired observation data together to train the model and update the relevant parameters of the model; specifically, the step of updating the pre-constructed ocean dynamic feature model based on the current real-time observation data includes:

[0088] The space position of the vehicle and the target feature observation data in the current real-time observation data are used as the input and output of the ocean dynamic feature model, respectively, to update the parameters of the ocean dynamic feature model; that is, assuming the previous data acquisition time is t, then with the addition of new data at time t+1, the parameters shown in equations (5)-(6) in the Gaussian process are recursively updated according to equations (7)-(9):

[0089] α t+1 =T t+1 (α t )+q (t+1) s t+1 k x (7)

[0090]

[0091] s t+1 =T t+1 (C t k t+1 )+e t+1 (9)

[0092] In the formula, s t+1 Process variables introduced to simplify the expression of operational expressions; operator T t+1 This means that by adding a 0 to the end of the vector, the length of the vector is increased from t to t+1; the operator U t+1This means that by adding 0s to fill empty spaces, the t-dimensional matrix is ​​increased to t+1 dimensions; e t+1 q represents a unit vector whose (t+1)th data element is 1; (tt+1) and r (t+1) Let represent the first and second derivatives of the logarithm of the expected likelihood, respectively.

[0093] In practical applications, considering the limitations of AUV's actual storage capacity and computing power, a data container of a limited size is defined, which is the basic vector set used to update the ocean dynamic characteristic model. (Basic Vector) is used to contain the dataset input to the model. Preferably, this implementation evaluates the importance of each collected target feature, and only adds the collected data to the basic vector set when it is deemed important, for model updates and target feature prediction. Specifically, before updating the parameters of the ocean dynamic feature model, the following steps are also included:

[0094] Based on the spacecraft's spatial position in the current real-time observation data, the position covariance and Gram matrix are calculated. Then, based on the position covariance, the Gram matrix, and the target feature observation data, the deviation coefficient corresponding to the current real-time observation data is calculated. The deviation coefficient is expressed as:

[0095]

[0096] In the formula,

[0097]

[0098] K t ={K0(x i ,x j )} i,j＝1,t

[0099] in, K represents the position covariance at time t+1; t γ represents the Gram matrix at time t; t+1 This represents the deviation coefficient at time t+1;

[0100] Based on the deviation coefficient and the preset tolerance parameter, it is determined whether the current real-time observation data is important to the ocean dynamic characteristic model; wherein, the preset tolerance parameter can be understood as determining whether the new data is relevant to the basic vector set. Is the difference in the existing data large enough to reach a threshold (whether it needs to be added)? (The specific value can be set according to the actual application requirements);

[0101] If the current real-time observation data is important, then the current real-time observation data is added to the basic vector set, and the parameters of the ocean dynamic feature model are updated according to the updated basic vector set; specifically, when γ t+1 When the value is greater than the tolerance parameter ∈, it indicates that the new data is related to the basic vector set. The existing data in the model have significant differences, so the latest real-time observation data needs to be added to the basic vector set and the model needs to be updated according to the parameter update formulas shown in equations (7)-(8).

[0102] If the current real-time observation data is not important, there is no need to add it to the basic vector set. Instead, the existing basic vector set can be reorganized based on feature space decomposition projection using the current real-time observation data to update the parameters of the ocean dynamic feature model. The reorganization calculation based on feature space decomposition projection can be understood as using the concept of projection from inputs to feature space to decompose the new data (new feature vectors) into a component contained in the first t feature subspaces and a residual component orthogonal to it. The length of the residual component is used to determine the error of the new input projection in the feature space. When it is less than a set threshold, the projection in the spanned subspace of the data within the first t basic vector sets is calculated, and the parameters obtained from this projection replace the actual input. Based on this replacement, the new data will be processed by the model without increasing the size of the basic vector set. Specifically, when the new data calculates γ... t+1 When the new data is less than the set tolerance parameter ∈, it indicates that the new data is... The data in the dataset has relatively small differences, so there is no need to add new data. In this process, a renormalization calculation can be performed using a method such as equation (11). Make new data without adding In the case of model parameters as shown in equations (7)-(8), and Update the solution to achieve higher efficiency:

[0103]

[0104] In the formula,

[0105]

[0106] Q t =K0(X t ,X t ) -1

[0107] in, and Q t Let represent the projected coordinates at (t+1) and the inverse of the Gram matrix at time t, respectively;

[0108] Furthermore, considering the capacity limitation of the basic vector set, if new data is added... After the set is completed, if the threshold d set by the basic vector set is exceeded, then the data with less influence in the basic vector set needs to be deleted to ensure the capacity stability of the basic vector set, thereby ensuring that the sparse online Gaussian process has high solution efficiency; based on this, this embodiment preferably adopts a method to measure the importance of each data element in the basic vector set in order to filter out the element with the smallest score for deletion; specifically, the data element importance measurement formula is shown in equation (12):

[0109]

[0110] Where, α t+1 (i) and Q t+1 (i, i) represent the model parameters α at time (t+1). t+1 The i-th element and the i-th row and i-th column element of the inverse matrix of the Gram matrix at time (t+1);

[0111] After finding and deleting the element with the smallest score using equation (12), it is necessary to adjust the parameter α. t+1 C t+1 and Q t+1 Update according to the formulas shown in equations (13)-(15):

[0112]

[0113]

[0114]

[0115] c j =C t+1 (j,j) (16)

[0116] Among them, C (t) It is C t+1 The t-dimensional matrix whose j-th row and j-th column are deleted; C j C t+1 Delete the j-th column of the j-th element; other symbols in the formula can be found in the explanatory description given above, and will not be repeated here;

[0117] It should be noted that updating the ocean dynamic characteristic model in the above manner also includes updating the constant value of the mean function (updated to the average value of all existing observation data) and the hyperparameter θ. For specific update methods, please refer to the existing technology implementation, which will not be elaborated here.

[0118] S14. Based on the target feature prediction result, calculate the current step size information, and based on the current step size information, determine whether the target feature observation is sufficient; wherein, determining whether the target feature observation is sufficient can be understood as using the model's prediction result to determine whether the observation of dynamic features is sufficient under the AUV's mobility capability (using AUV dynamic simulation to achieve control from the current state to the expected target waypoint, mobility capability limitations include airspeed, minimum turning radius, etc.) and energy consumption limitations (AUV's battery capacity is limited, limiting the total mission duration); wherein, the target feature prediction result can be understood as the spatial distribution of the target feature physical quantity and the covariance matrix of the quantified uncertainty obtained at a certain moment, including the mean and covariance matrix of the target feature physical quantity; specifically, the step of calculating the current step size information based on the target feature prediction result and determining whether the target feature observation is sufficient based on the current step size information includes:

[0119] The target feature prediction result and the path corresponding to the current step size are input into a preset path information content objective function to obtain the corresponding current step size information content. The path information content objective function includes three terms: exploration, exploitation, and their combined effect, along with their respective weight coefficients. Exploration represents the AUV's exploration of the location area during path planning, while exploitation represents the AUV's application of observed data to make it more reliable. Considering the balance between exploration and exploitation in information path planning, this embodiment preferably sets the path information content objective function as follows:

[0120]

[0121] Wherein, P represents all possible paths within the target observation area; Indicates the currently planned path; x represents the spatial location; It represents the absolute value of the gradient of the observed data of the target feature, and represents the intensity of the change of dynamic features; The normalized Gaussian process covariance represents the uncertainty of the observed region; α1 represents the product of the absolute value of the gradient of the target feature observation data and the normalized Gaussian process covariance, representing the combined effect of exploration and development; α2 and α3 represent the weighting coefficients of the absolute value of the gradient of the target feature observation data, the normalized Gaussian process covariance, and the product of the absolute value of the gradient of the target feature observation data and the normalized Gaussian process covariance, respectively. This indicates the amount of information within the currently planned path; it should be noted that... and All of these can be calculated from the mean and covariance matrix of the target characteristic physical quantities, which will not be detailed here;

[0122] If the current step size information is less than a preset information threshold, the target feature observation is considered sufficient; otherwise, the target feature observation is considered insufficient. The information threshold can be determined according to actual application needs and is not specifically limited here. For example, assuming the information threshold is ξ, when the calculated current step size information is less than ξ, the dynamic feature observation is considered sufficient and the observation task can be terminated; when the calculated current step size information is greater than or equal to ξ, the dynamic feature observation is considered insufficient, and local information path planning within the next running step size R is required to continue the observation task.

[0123] S15. If the target features are sufficiently observed, the observation task ends; otherwise, the next target waypoint is obtained through local information path planning, and observation continues. The next target waypoint can be understood as the next observation position that an AUV needs to reach, with a distance of R from the current location. To ensure the scientific rationality of the selection of the next observation position, this embodiment preferably uses an objective function that measures the amount of information within a step size to calculate the amount of information, and determines the new optimal target waypoint for acquiring ocean dynamic features by maximizing the amount of information. Specifically, the step of obtaining the next target waypoint through local information path planning includes:

[0124] Based on the current spatial position of the vehicle and the preset operating step size, a set of candidate target waypoints is obtained; the distance between each candidate target waypoint in the set and the current spatial position of the vehicle is the preset operating step size; wherein, the method for determining each candidate target waypoint in the set can be determined according to actual application requirements, for example, the distance between the waypoint and the AUV's location X... now Let P be the set of candidate target waypoints at a distance of step length R from the current space position of the vehicle (i.e., the position of the previous target waypoint). next ={X next1 ,X next2 ,…,X nextn The specific number of points n in the candidate target waypoint set is not specifically limited here;

[0125] Based on the path information objective function, the candidate Dubins path information corresponding to the current space position of the aircraft and each candidate target waypoint is obtained respectively; wherein, the Dubins path can be understood as a path that is relatively close to the Dubins path, and is a path planned with a circle with the turning radius as the radius and a straight line added when considering the heading angle and position of the current point, which satisfies the kinematic model of the AUV; correspondingly, the path information objective function is shown in Equation (17), and the specific calculation process of the candidate Dubins path information is as follows:

[0126] Select X nowTo X nexti n within the path (i = 1…n) s The set of equally divided points is used to calculate the prediction results and covariance Ω at each equally divided point using the ocean dynamic feature model. These are used as the feature values ​​and uncertainty estimates of the target features, respectively. Substituting them into the objective function of equation (17) will yield the information content estimate of each candidate path in the next step.

[0127] The candidate target waypoint corresponding to the largest candidate class Durbins path information is taken as the next target waypoint;

[0128] After determining the next target waypoint using the above methods and steps, the AUV needs to be moved to the next target waypoint for observation. The ocean dynamic characteristic model is then updated using real-time observation data. Within the limited mission time or energy consumption of the AUV, the aforementioned steps are repeated to conduct a new round of target characteristic observation. It should be noted that in practical applications, the AUV will repeatedly execute the "planning-guided sampling - sampling-model construction - model update planning" process within the limited mission time or energy consumption. Subsequent cyclic operations are carried out in the same manner until it is confirmed that the target characteristics of the target observation area have been sufficiently observed, and the observation mission ends.

[0129] This application embodiment initializes the initial position and initial target waypoint of the autonomous underwater vehicle (AUV) based on the target observation area, initializes the mean function and covariance function of the sparse online Gaussian process model, and constructs the ocean dynamic feature model. Then, it begins real-time observation of target features. When the AUV reaches the initial target waypoint, it collects current real-time observation data including the vehicle's spatial position and corresponding target feature observation data. The pre-constructed ocean dynamic feature model is updated based on the current real-time observation data, and the target feature prediction result is obtained based on the updated sparse online Gaussian process model. The current step size information is calculated based on the target feature prediction result, and the sufficiency of target feature observation is determined based on the current step size information. If the target feature observation is sufficient, the observation task ends; otherwise, the next target waypoint is obtained through local information path planning, and observation continues. This provides a solution. Figure 4 The ocean dynamic feature observation method based on information path planning shown here effectively solves the application defects of existing ocean dynamic feature observation methods. It not only gives full play to the autonomy of AUV through information path planning, realizing the most reasonable path planning for obtaining ocean dynamic feature observation under the constraints of AUV mobility and energy consumption, but also ensures high solution efficiency through sparse online Gaussian process and can effectively improve the accuracy of ocean dynamic feature observation, providing a reliable guarantee for obtaining high-quality ocean dynamic feature observation data.

[0130] To illustrate the application scenarios and effects of the marine dynamic characteristic observation method based on information path planning provided by this invention, this application also provides application examples of underwater observation of sea ice thickness and seawater temperature field.

[0131] In one embodiment, an ocean dynamics observation method based on information path planning is applied in the underwater ice thickness observation scenario. In this task, the maximum observation step size for the AUV is set to 25, and the AUV running step size R = 15m is used for local information path planning within the maximum observation step size. The AUV can enter the sea ice, and the distance from the AUV to the sea ice floor is obtained using an upward altimeter. Combined with the depth data of the AUV, the underwater ice thickness is observed. The specific steps are as follows:

[0132] In the simulation, the observation area was defined as a square region of 100m x 100m, and the actual underwater ice thickness in the observation area was as follows: Figure 5 As shown, the observation area is divided into 100*100 data grids at equal intervals along the longitude and latitude directions, and the initial position of the AUV is initialized to X. t＝1 = (0,0), initial target waypoint is X next-ini = (20, 10);

[0133] Initialize the sparse online Gaussian process model and parameters. The mean function is defined as a constant and initialized to 0. The covariance function is defined as the squared exponential covariance function K0(x,x′). Initialize the hyperparameter θ = [σ]. f ,σ ω [l1,l2] T Let θ = [0.5, 0.4, 0.15, 0.15] T ;

[0134] During the observation process from the AUV to the next target waypoint, real-time depth data from the CTD sensor and real-time distance data between the AUV and the ice bottom from the altimeter sensor are acquired. The difference between the two is the underwater ice thickness data, i.e., the distance at horizontal position x at time t. t Observational data obtained at the location

[0135] For new data acquired by the sensor at time t, before updating the model, it is necessary to determine whether this new input pair should be added to the basic vector set. In (Basic Vector), calculate the corresponding deviation coefficient γ. t+1 And according to γ t+1 The model update method is selected based on the relationship between the value of the set tolerance parameter ∈ and the value of the parameter.

[0136] After the AUV reaches the next target waypoint, the constant value of the mean function and the hyperparameter θ are updated based on the observed values, where the constant value of the mean function is updated to the average value of all existing observation data. Simultaneously, based on the path information objective function shown in equation (17) (where α1 = 2, α2 = 2, α3 = 1), the model's prediction results are used to calculate the information content within the local path based on the objective function, in order to select the next target waypoint. After determining the new next target waypoint, the above steps are repeated within the maximum number of observation steps, that is, the model is updated using real-time observation data from the next target waypoint, and a locally optimal information path planning is made after reaching the target waypoint. Finally, the following is obtained: Figure 5 The observation paths of the AUV at t = 100, 200, 350, and 500 s are shown. The black dots represent the observation paths of the AUV. The initial observation paths of the AUV are mainly used to explore unknown areas. After discovering areas with thicker underwater ice or larger gradients, the observation paths mainly revolve around these areas, which can effectively ensure the comprehensiveness and accuracy of ice thickness characteristic observations.

[0137] In one embodiment, an ocean dynamics feature observation method based on information path planning is applied in a seawater temperature field observation scenario. The specific steps are as follows:

[0138] In the simulation, the observation area was defined as a 5km*5km square area, a dynamically changing seawater temperature field was constructed, and the observation area was divided into a 100*100 data grid at equal intervals along the longitude and latitude directions.

[0139] Based on the same implementation steps and parameter settings as the aforementioned sea ice underwater ice thickness observation scenario, the final result is... Figure 6 The observation paths of the AUV and the prediction results of the seawater temperature field model at t = 250, 350, 450, and 600 s are shown. It can be seen that in the process of observing the dynamic characteristics of the temperature field, the AUV can not only focus on the region of interest with a large temperature gradient, but also effectively focus on the unexplored region with greater uncertainty, thus effectively improving the comprehensiveness and accuracy of temperature field observation.

[0140] It should be noted that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders.

[0141] In one embodiment, such as Figure 7 As shown, an ocean dynamic characteristic observation system based on information path planning is provided, the system comprising:

[0142] Initialization module 1 is used to initialize the initial position and initial target waypoint of the autonomous underwater vehicle according to the target observation area, and to start real-time observation of target features;

[0143] Data acquisition module 2 is used to acquire current real-time observation data when the autonomous underwater vehicle reaches the target waypoint; the current real-time observation data includes the vehicle's spatial position and corresponding target feature observation data;

[0144] Feature prediction module 3 is used to update the pre-constructed ocean dynamic feature model based on the current real-time observation data, and to obtain the target feature prediction result based on the updated ocean dynamic feature model.

[0145] Feature diagnosis module 4 is used to calculate the current step size information based on the target feature prediction result, and to determine whether the target feature observation is sufficient based on the current step size information.

[0146] The path planning module 5 is used to end the observation task when the target features are sufficiently observed, and to obtain the next target waypoint through local information path planning when the observation is insufficient, and to continue the observation.

[0147] Specific limitations regarding the information path planning-based marine dynamic characteristic observation system can be found in the limitations of the information path planning-based marine dynamic characteristic observation method described above; the corresponding technical effects are equivalent and will not be repeated here. Each module in the aforementioned information path planning-based marine dynamic characteristic observation system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0148] Figure 8 An internal structural diagram of a computer device is shown in one embodiment. This computer device may specifically be a terminal or a server. Figure 8As shown, the computer device includes a processor, memory, network interface, display, camera, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for observing ocean dynamic characteristics based on information path planning. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0149] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or arrange components with the same effect.

[0150] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0151] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0152] In summary, the present invention provides a method and system for observing ocean dynamic features based on information path planning. The method, after initializing the initial position and initial target waypoint of an autonomous underwater vehicle (AUV) according to the target observation area, begins real-time observation of target features. When the AUV reaches the initial target waypoint, it collects current real-time observation data including the vehicle's spatial position and corresponding target feature observation data. Based on this data, it updates a pre-constructed ocean dynamic feature model, obtains target feature prediction results based on the updated sparse online Gaussian process model, calculates the current step size information based on the prediction results, and then... The method involves determining whether the target feature observations are sufficient. If the target feature observations are sufficient, the observation task is terminated; otherwise, the next target waypoint is obtained through local information path planning, and the observation continues. This method combines information path planning with a sparse online Gaussian process model and follows a cyclical observation concept of "planning-guided sampling - sampling-model construction - model-updating planning." This approach not only fully leverages the autonomy of AUVs through information path planning, achieving the most suitable path planning for obtaining ocean dynamic feature observations under the constraints of AUV mobility and energy consumption, but also ensures high solution efficiency through the sparse online Gaussian process and effectively improves the accuracy of ocean dynamic feature observations, providing a reliable guarantee for obtaining high-quality ocean dynamic feature observation data.

[0153] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0154] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A method for observing dynamic ocean features based on information path planning, characterized in that, The method includes the following steps: Based on the target observation area, initialize the initial position and initial target waypoint of the autonomous underwater vehicle, and begin real-time observation of target features; When the autonomous underwater vehicle reaches the initial target waypoint, it collects current real-time observation data; the current real-time observation data includes the vehicle's spatial position and corresponding target feature observation data. Based on the current real-time observation data, the pre-constructed ocean dynamic feature model is updated, and the target feature prediction result is obtained based on the updated ocean dynamic feature model. Based on the target feature prediction results, calculate the current step size information content, and based on the current step size information content, determine whether the target feature observation is sufficient; If sufficient target features are observed, the observation mission ends; otherwise, the next target waypoint is obtained through local information path planning, and observation continues.

2. The ocean dynamic characteristic observation method based on information path planning as described in claim 1, characterized in that, The step of initializing the initial position and initial target waypoint of the autonomous underwater vehicle based on the target observation area includes: The target observation area is divided into equally spaced grids according to a preset size to obtain the corresponding region grid. The origin of the region grid is initialized to the initial position of the autonomous underwater vehicle; The initial target waypoint is obtained based on the initial position, preset running step size, and heading angle; the initial target waypoint is represented as: X next-ini =(Rcosψ,Rsinψ) Among them, X next-ini R represents the initial target waypoint; R represents the preset running step size; ψ represents the heading angle corresponding to the initial position.

3. The ocean dynamic characteristic observation method based on information path planning as described in claim 2, characterized in that, The ocean dynamics model is a sparse online Gaussian process model. The step of initializing the initial position and initial target waypoint of the autonomous underwater vehicle based on the target observation area further includes: The mean function and covariance function of the sparse online Gaussian process model are initialized to construct the ocean dynamic characteristic model, and the initial observation data corresponding to the initial position are collected.

4. The ocean dynamic characteristic observation method based on information path planning as described in claim 3, characterized in that, The step of updating the pre-constructed ocean dynamic characteristic model based on the current real-time observation data includes: The space position of the vehicle in the current real-time observation data and the target feature observation data are used as the input and output of the ocean dynamic feature model, respectively, to update the parameters of the ocean dynamic feature model.

5. The ocean dynamic characteristic observation method based on information path planning as described in claim 4, characterized in that, Before the step of updating the parameters of the ocean dynamic characteristic model, the method further includes: Based on the space position of the vehicle in the current real-time observation data, calculate the position covariance and Gram matrix, and based on the position covariance, the Gram matrix and the target feature observation data, calculate the deviation coefficient corresponding to the current real-time observation data; Based on the deviation coefficient and the preset tolerance parameter, determine whether the current real-time observation data is important to the ocean dynamic characteristic model; If the current real-time observation data is important, then the current real-time observation data is added to the basic vector set, and the parameters of the ocean dynamic feature model are updated according to the updated basic vector set; If the current real-time observation data is not important, there is no need to add the current real-time observation data to the basic vector set. Instead, the existing basic vector set can be recalculated based on feature space decomposition projection using the current real-time observation data to update the parameters of the ocean dynamic feature model.

6. The ocean dynamic characteristic observation method based on information path planning as described in claim 5, characterized in that, The target feature prediction results include the mean and covariance matrix of the target features; The steps of calculating the current step size information based on the target feature prediction result, and determining whether the target feature observation is sufficient based on the current step size information include: The target feature prediction result and the path corresponding to the current step size are input into a preset path information content objective function to obtain the corresponding current step size information content; the path information content objective function is expressed as: Wherein, P represents all possible paths within the target observation area; Indicates the currently planned path; x represents the spatial location; This represents the absolute value of the gradient of the target feature observation data; This represents the covariance of a normalized Gaussian process. α1 represents the product of the absolute value of the gradient of the target feature observation data and the normalized Gaussian process covariance; α2 and α3 represent the weight coefficients of the absolute value of the gradient of the target feature observation data, the normalized Gaussian process covariance, and the product of the absolute value of the gradient of the target feature observation data and the normalized Gaussian process covariance, respectively. This indicates the amount of information within the currently planned path; If the current step size information is less than the preset information threshold, the target feature observation is determined to be sufficient; otherwise, the target feature observation is determined to be insufficient.

7. The ocean dynamic characteristic observation method based on information path planning as described in claim 6, characterized in that, The step of obtaining the next target waypoint through local information path planning includes: Based on the current spatial position of the vehicle and the preset operating step size, a set of candidate target waypoints is obtained; the distance between each candidate target waypoint in the set and the current spatial position of the vehicle is the preset operating step size. Based on the objective function of the path information, the candidate class Dobbins path information corresponding to the current space position of the vehicle and each candidate target waypoint is obtained respectively. The candidate target waypoint corresponding to the largest candidate type Dobins path information is taken as the next target waypoint.

8. A marine dynamic characteristic observation system based on information path planning, characterized in that, The system includes: The initialization module is used to initialize the initial position and initial target waypoint of the autonomous underwater vehicle according to the target observation area, and to start real-time observation of target features; The data acquisition module is used to collect real-time observation data when the autonomous underwater vehicle reaches the target waypoint; the real-time observation data includes the vehicle's spatial position and the corresponding target feature observation data; The feature prediction module is used to update the pre-constructed ocean dynamic feature model based on the current real-time observation data, and to obtain the target feature prediction result based on the updated ocean dynamic feature model. The feature diagnosis module is used to calculate the current step size information based on the target feature prediction result, and to determine whether the target feature observation is sufficient based on the current step size information. The path planning module is used to end the observation task when sufficient target features are observed, and to obtain the next target waypoint through local information path planning when insufficient features are observed, and to continue observation.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images