Water environment unmanned ship autonomous navigation method, device and equipment based on concentration gradient and storage medium

Abstract

The invention discloses a water environment unmanned ship autonomous navigation method, device and equipment based on concentration gradient and a storage medium. The method comprises the following steps: collecting concentration data in a water environment through a sensor carried by an unmanned ship; a time-average concentration gradient rapid inversion model is constructed, the time-average concentration gradient rapid inversion model comprises a simple model and a PINN model which can be directly calculated based on the currently collected concentration data, and the weight ratio of the two models is dynamically adjusted so as to process the collected concentration data to obtain a time-average concentration field of the water body; determining the expected navigation path of the next step according to the concentration gradient direction of the time-average concentration field obtained by inversion; and adjusting the moving direction and speed of the unmanned ship in real time based on the expected navigation path in combination with the motion characteristics of the unmanned ship and the water environment characteristics. According to the invention, efficient pollutant traceability and accurate water quality monitoring in the water environment can be realized.

Description

Technical Field

[0001] The present disclosure relates to the technical field of water environment monitoring and autonomous navigation, and in particular to a method, device, equipment and storage medium for autonomous navigation of an unmanned ship in a water environment based on a concentration gradient. Background Art

[0002] With the progress of industrialization and population growth, pollution, overexploitation and climate change jointly threaten global water resources, and water environment issues have become a global focus. Effective monitoring and management of sewage outlets is one of the basic tasks of water environment protection. Due to the wide distribution and hidden location of sewage outlets, coupled with the increase in underwater sewage discharge, monitoring work faces many challenges. Taking a certain area as an example, in 2015, about 12% of COD emissions (20,330 tons) and 14% of NH3-N emissions (1,749 tons) exceeded the maximum allowable pollution load (MAPL) of the water functional area. Therefore, strengthening water environment monitoring, especially the inspection and monitoring of sewage outlets, is of great significance to achieving water quality improvement and ecological restoration.

[0003] With the advancement of automation and intelligence, robots are increasingly used in the field of environmental water conservancy. Robotic systems such as underwater autonomous vehicles (AUVs) and unmanned surface vessels (USVs) are widely studied to achieve automated and continuous monitoring of water quality parameters such as temperature, salinity, dissolved oxygen, pH, and turbidity. These systems can access areas that are difficult for humans to reach, such as the deep sea or remote waters, providing environmental scientists with unprecedented monitoring capabilities.

[0004] The role of concentration gradients in water quality monitoring and pollutant tracing is crucial, especially when combined with autonomous patrol water quality monitoring robots. Concentration gradients, that is, the spatial variation in the concentration of one or more chemical components in water, provide key information for identifying pollution sources and understanding pollution spread. With the development of water quality monitoring robot technology, these autonomous mobile devices are able to implement efficient and accurate concentration gradient monitoring and analysis in water bodies. Through the collected data, combined with geographic information systems (GIS) and environmental models, concentration field inversion and pollutant tracing can be achieved. This technology not only improves the efficiency and accuracy of water quality monitoring, but also plays an important role in the rapid response and handling of pollution incidents.

[0005] Existing water quality monitoring technologies mainly use unmanned boats and other equipment to realize automatic monitoring of water quality parameters. These systems usually include modules such as sensor data acquisition, data transmission and processing, and navigation control. In practical applications, it is necessary to consider the complex dynamic conditions in the water environment (such as flow rate changes, obstacle interference, etc.), as well as technical requirements such as data processing and real-time transmission. In addition, due to factors such as sparse sampling points and the influence of local water mixing on sampling equipment, the monitoring efficiency and accuracy of the system need to be further improved.

[0006] Therefore, there is an urgent need for an autonomous navigation technology for unmanned boats in the water environment, which can achieve efficient pollutant tracing and water quality monitoring, and improve the scientific and refined level of water environment protection. Summary of the invention

[0007] The present invention provides an autonomous navigation method for an unmanned boat in a water environment based on concentration gradient. By combining a simple model with a physical information neural network PINN (Physics-Informed Neural Networks) model to dynamically invert the water concentration field, efficient pollutant tracing and accurate water quality monitoring in the water environment can be achieved.

[0008] According to an embodiment of the present disclosure, a method for autonomous navigation of an unmanned ship in a water environment based on a concentration gradient is proposed, comprising:

[0009] Collect concentration data in the water environment through sensors carried by unmanned vessels;

[0010] Constructing a time-averaged concentration gradient rapid inversion model, the time-averaged concentration gradient rapid inversion model includes a simple model that can be directly calculated based on currently collected concentration data and a physical information neural network PINN model, and dynamically adjusting the weight ratio of the two models to process the collected concentration data to obtain the time-averaged concentration field of the water body;

[0011] Determine the next desired navigation path based on the concentration gradient direction of the time-averaged concentration field obtained by inversion;

[0012] Based on the expected navigation path, combined with the motion characteristics of the unmanned ship and the environmental characteristics, the moving direction and speed of the unmanned ship are adjusted in real time.

[0013] In some possible embodiments, the PINN model uses the control equation of the hydrodynamic-water quality model as a physical constraint in the training process, and the loss function includes a data fitting term and a physical constraint term. The data fitting term is used to measure the deviation between the predicted concentration value and the observed concentration value of the observed data sampling point, and the physical constraint term is used to measure the fit between the predicted concentration value of the constraint sampling point and the physical law.

[0014] In some possible implementations, the data fitting term is determined according to the following formula:

[0015] Among them, Θ is the PINN network parameter, x (i) is the spatial coordinate point, is the number of observation data sampling points, For PINN network in x (i) The predicted concentration value at g(x (i)) is in x (i) The concentration value collected at It means to find the square of L2 norm.

[0016] In some possible implementations, the method further includes:

[0017] A temporal and spatial concentration variation data set is obtained by LES simulation, wherein the temporal and spatial concentration variation data set includes flow velocity and concentration values ​​of each grid point in the flow field at different times;

[0018] Constraint sampling points are selected based on the spatiotemporal concentration variation data set, wherein the constraint sampling points in areas with large flow velocity or concentration gradient variations are denser than the constraint sampling points in areas with small flow velocity or concentration gradient variations.

[0019] In some possible implementations, the physical constraint term L is determined according to the following formula: Ω (Θ):

[0020]

[0021] Among them, N Ω is the number of sampling points under constraint conditions, x (i) is a spatial coordinate point, v represents the physical parameter of the control equation used to establish the physical constraint, D(x (i) ; v) represents x (i) Substituting the predicted concentration value at into the residual calculated by the control equation, It means to find the square of L2 norm.

[0022] In some possible implementations, under complex flow field conditions, the physical constraint term is obtained by substituting the predicted concentration value of the constraint condition sampling point into the large eddy simulation (LES) calculation residual.

[0023] In some possible implementations, the time-averaged concentration gradient fast inversion model integrates the outputs of the simple model and the PINN model according to the following formula:

[0024]

[0025] Among them, w simple +w PINN =1, is the final time-averaged concentration field output by the time-averaged concentration gradient fast inversion model, This is the calculation result of the simple model. is the calculation result of PINN model, w simple and w PINN is the weight coefficient.

[0026] In some possible implementations, dynamically adjusting the weight ratio of the two models includes:

[0027] As the number of sampling points increases exist The weight ratio in .

[0028] In some possible implementations, dynamically adjusting the weight ratio of the two models includes:

[0029] Calculate the concentration gradient direction error and concentration inversion error of the calculation results of the simple model;

[0030] If any one of the concentration gradient direction error and the concentration inversion error exceeds the corresponding preset threshold, then w is set simple =0,w PINN =1.

[0031] In some possible implementations, adjusting the moving direction and speed of the unmanned ship in real time includes:

[0032] Monitor the surrounding environment in real time through sensors to identify and avoid obstacles in the water;

[0033] Adjust the steering angle and acceleration according to the unmanned ship's own inertia characteristics;

[0034] Adjust course and speed according to current water conditions;

[0035] Reduce travel speed in areas of high concentration gradient to increase sampling density.

[0036] According to one embodiment of the present disclosure, a concentration gradient-based autonomous navigation device for an unmanned ship in a water environment is also proposed, comprising:

[0037] A real-time acquisition unit, used to collect concentration data in the water environment through sensors carried by unmanned vessels;

[0038] A model inversion unit is used to construct a time-averaged concentration gradient fast inversion model, wherein the time-averaged concentration gradient fast inversion model includes a simple model that can be directly calculated based on the currently collected concentration data and a physical information neural network PINN model, and dynamically adjusts the weight ratio of the two models to process the collected concentration data to obtain the time-averaged concentration field of the water body;

[0039] A path determination unit, used to determine the next desired navigation path according to the concentration gradient direction of the time-averaged concentration field obtained by inversion;

[0040] The real-time navigation unit is used to adjust the moving direction and speed of the unmanned ship in real time based on the expected navigation path and in combination with the motion characteristics of the unmanned ship and the environmental characteristics.

[0041] According to one embodiment of the present disclosure, an electronic device is provided, the device comprising a memory and a processor, the memory being used to store computer instructions executable on the processor, the processor being used to implement any of the above methods when executing the computer instructions.

[0042] According to an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, any of the above methods is implemented.

[0043] The proposed solution for autonomous navigation of unmanned boats in water environments based on concentration gradients significantly improves the accuracy and computational efficiency of inversion results through innovative concentration field inversion strategies. The proposed solution combines a simple model with a PINN model, which can quickly give a preliminary estimate when there is less sampled data, and gradually transition to a more accurate PINN model as more data is collected, thus avoiding the limitations that may be caused by a single model.

[0044] The design of the PINN model proposed in this application introduces physical constraints. By calculating the residuals at the constraint sampling points, the prediction results are ensured to conform to the hydrodynamic-water quality laws. At the same time, the large eddy simulation (LES) is used to optimize the distribution of constraint points, further improving the prediction accuracy in complex flow fields. This constraint mechanism based on physical laws significantly enhances the generalization ability and prediction reliability of the model.

[0045] In terms of navigation control, the present invention dynamically plans the navigation path based on the concentration gradient direction, comprehensively considers the motion characteristics of the unmanned ship and the water environment characteristics, improves the sampling efficiency through a reasonable speed control strategy, and realizes efficient autonomous navigation.

[0046] The proposed water environment unmanned boat autonomous navigation solution based on concentration gradient has good practicality, simple system structure, easy implementation, strong environmental adaptability, and can meet the needs of actual water environment monitoring. This solution provides a new technical means for water quality monitoring, improves the efficiency of pollution source tracing, and is conducive to the development of water environment pollution control work.

[0047] Other features and advantages of the present disclosure are described in detail below. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the specification and, together with the description, serve to explain the principles of the specification.

[0049] Figure 1 A flow chart of an autonomous navigation method for an unmanned ship in a water environment based on a concentration gradient according to an embodiment of the present disclosure is shown.

[0050] Figure 2 A schematic diagram of a technical framework according to an exemplary embodiment of the present disclosure is shown.

[0051] Figure 3 A schematic diagram of a traceability line according to an exemplary embodiment of the present disclosure is shown.

[0052] Figure 4 It is a schematic diagram of the structure of an electronic device shown in at least one embodiment of the present disclosure. DETAILED DESCRIPTION

[0053] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0054] The disclosed embodiments may be applied to a computer system / server that may operate with numerous other general or special computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations suitable for use with a computer system / server include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, networked personal computers, minicomputer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the above systems, and the like.

[0055] Computer systems / servers may be described in the general context of computer system executable instructions (such as program modules) executed by a computer system. In general, program modules may include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers may be implemented in a distributed cloud computing environment where tasks are performed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules may be located on local or remote computing system storage media including storage devices.

[0056] Figure 1 A flow chart of a method for autonomous navigation of an unmanned ship in a water environment based on a concentration gradient according to an embodiment of the present disclosure is shown. As shown in the figure, the method includes steps 1 to 4.

[0057] Step 1: Collect concentration data in the water environment through sensors carried by unmanned boats.

[0058] In this embodiment, the concentration data in the water environment can be collected by the sensor carried by the unmanned boat. The unmanned boat can be equipped with a high-frequency sampling concentration sensor to record the instantaneous concentration data in real time. Due to the turbulence in the water environment, the concentration value fluctuates with high frequency and irregularity over time and space. This concentration pulsation characteristic will interfere with the inversion of the concentration gradient. Therefore, the sampling frequency of the sensor can be reasonably set, and the water flow rate, turbulence intensity, ambient temperature and other conditions in the experiment can be recorded.

[0059] During the acquisition process, the quality of concentration field data is affected by many factors, such as the choice of sampling frequency affects the temporal resolution of the data, the motion trajectory of the unmanned ship affects the spatial distribution of the sampling points, and local water mixing interferes with the sampling concentration. In order to improve the quality of sampling data, the appropriate sampling frequency can be selected according to the concentration variation characteristics, the sampling density can be increased in the high concentration gradient area, and the key areas of the concentration field can be covered by sampling points with reasonable spatial distribution.

[0060] The collected concentration data can be transmitted to the processing module through a real-time communication system. The sampling frequency and data transmission scheme are matched with the calculation rate of the inversion model to ensure the real-time performance of the system.

[0061] Specifically, in some embodiments, a monitoring-inversion real-time communication model can be established. The model can comprehensively consider the sampling frequency and accuracy of sensors such as concentration, speed, and position, and combine the calculation rate of the time-averaged concentration gradient rapid inversion model to specify a suitable data transmission scheme. For key data such as concentration, speed, and position, a preliminary data judgment can be performed first to ensure the validity of the data. The preliminary judgment link can promptly detect and eliminate abnormal data, avoid invalid or erroneous data from affecting the learning effect of the PINN network, and thus improve the overall performance of the system. This real-time communication mechanism runs through the entire process of unmanned ship navigation, ensuring the continuity and reliability of data collection, transmission, and processing.

[0062] In one example, the initial judgment stage can screen the data through physical rationality checks, time series anomaly detection, spatial correlation verification, etc.

[0063] For example, it is possible to check whether the concentration value is within the sensor range, whether the concentration change rate exceeds the physically possible maximum value, whether the concentration difference between adjacent sampling points conforms to the diffusion law, etc.

[0064] For example, the moving average and standard deviation can be used to detect mutation points, data points that exceed 3 times the standard deviation can be marked, and multiple consecutive abnormal points may indicate sensor failure, etc.

[0065] For example, the concentration correlation of nearby sampling points can be calculated, abnormal jumps in space can be detected, and the diffusion trend predicted by the hydrodynamic model can be verified.

[0066] When the data meets the above criteria at the same time, it will be used as qualified sampling data in the initial judgment stage and input into the subsequent time-averaged concentration gradient rapid inversion model for PINN network learning and reasoning. Through this multi-dimensional data quality control, the overall navigation performance can be effectively improved.

[0067] Step 2, construct a time-averaged concentration gradient rapid inversion model, which includes a simple model that can be directly calculated based on the currently collected concentration data and a physical information neural network PINN model, and dynamically adjusts the weight ratio of the two models to process the collected concentration data to obtain the time-averaged concentration field of the water body.

[0068] The characteristics of unmanned ship concentration measurement, such as non-fixed sampling points, short sampling time, and local mixing of water bodies, can be considered to analyze the incomplete statistical characteristics of the instantaneous value changes of pollutant emissions in turbulence. Combined with deep learning technologies such as PINN, the hydrodynamic-water quality simulation method combining artificial intelligence and numerical simulation is explored to construct a rapid inversion model of time-averaged concentration gradient for unmanned ship concentration measurement.

[0069] The time-averaged concentration gradient rapid inversion model proposed in this embodiment adopts a strategy that combines the simple model and the PINN model. The simple model can use the interpolation method to estimate the concentration field of the existing sampling point data. Optional interpolation methods include nearest neighbor interpolation, linear interpolation or Kriging interpolation. The simple model has a fast calculation speed and can directly obtain preliminary results based on the currently collected concentration data. It is suitable for the early stage of tracing with less sampling data. The PINN model is a physical information neural network based on deep learning technology. Its core principle is to use automatic differentiation technology to embed partial differential equations into the loss function of the neural network, and use the control equations of the hydrodynamic-water quality model as physical constraints in the training process. After accumulating enough sampling data, the PINN model can fully learn the distribution characteristics of the concentration field and meet the physical constraints. At this time, its prediction performance is usually better than the simple model.

[0070] By combining the inversion results of the two models and dynamically adjusting the weight ratio, it is possible to use a simple model for rapid calculation when the number of sampling points is small, avoiding delays in tracing due to waiting for the PINN model to converge. As data accumulates, the contribution of the PINN model can be gradually improved through the weight adjustment mechanism, achieving a smooth transition from rapid response to high-precision calculation. In addition, the PINN model introduces physical constraints to ensure that the calculation results conform to the hydrodynamic-water quality laws, thereby improving the reliability of the inversion results. This time-averaged concentration gradient rapid inversion model not only guarantees the needs of each stage in tracing, but also improves the calculation efficiency and inversion accuracy, and can well meet the needs of real-time navigation of unmanned ships.

[0071] In some embodiments, the PINN model may adopt a feedforward fully connected neural network structure, including an input layer, a hidden layer, and an output layer. Each neuron is linearly transformed through corresponding weights and biases, and then subjected to the action of a nonlinear activation function. Considering the need to solve the second-order partial derivatives, Tanh, which can be infinitely differentiable, can be selected as the activation function. During the model training process, the automatic differentiation AD technology can be used to calculate the derivative of the network output function to the input variable. AD technology can avoid the calculation error caused by the differential operation of the grid scale by decomposing the complex analytical function into a combination of a series of elementary operations, thereby improving the calculation accuracy.

[0072] In some embodiments, the PINN model uses the control equation of the hydrodynamic-water quality model as a physical constraint in the training process, and the loss function includes a data fitting term and a physical constraint term. The data fitting term is used to measure the deviation between the predicted concentration value and the observed concentration value of the observed data sampling point, and the physical constraint term is used to measure the fit between the predicted concentration value of the constraint sampling point and the physical law.

[0073] According to this embodiment, the PINN model introduces the control equations of the hydrodynamic-water quality model as physical constraints during the training process. In some examples, the control equations may include the one-dimensional Saint-Venant equations (continuity equation and momentum equation) describing the motion of water flow and the one-dimensional convection-diffusion equation describing the solute transport.

[0074] The loss function of the PINN model consists of a data fitting term and a physical constraint term. The data fitting term is calculated at the observation sampling point, and the prediction accuracy of the model is evaluated by comparing the deviation between the predicted concentration value and the actual observed value (i.e., the collected concentration data). The physical constraint term is calculated at the preset constraint sampling points. These points do not require actual collection of concentration data. Instead, the predicted concentration values ​​of the model at these points are substituted into the control equation, and the residual value is calculated to evaluate the degree of conformity of the prediction results to the physical laws.

[0075] This design enables the PINN model to simultaneously learn the true concentration distribution characteristics and satisfy the basic physical laws, improving the prediction accuracy and generalization ability of the model. The division of labor and cooperation between the two types of sampling points ensures that the model output is consistent with both actual observations and theoretical requirements.

[0076] In some examples, the data fitting term may be determined according to the following formula:

[0077]

[0078] Among them, Θ is the PINN network parameter, x (i) is the spatial coordinate point of the observation data sampling point, is the number of observation data sampling points, For PINN network in x (i) The predicted concentration value at g(x (i) ) is in x (i) The concentration value collected at It means to find the square of L2 norm.

[0079] is the time-averaged concentration value predicted by the neural network, g(x (i) ) is the instantaneous concentration value (including pulsation) actually observed. Through the above data fitting terms, the difference between the instantaneous concentration value and the time-averaged concentration value (i.e., the concentration pulsation) is treated as the observation error, which is beneficial to improving the inversion stability. The L2 norm can effectively measure the prediction deviation, which is convenient for optimization and derivation. Incorporating the data fitting terms designed as above into the loss function can effectively learn the real concentration distribution characteristics.

[0080] In some examples, the physical constraint term L may be determined according to the following formula: Ω (Θ):

[0081]

[0082] Among them, N Ω is the number of sampling points under constraint conditions, x (i) is the spatial coordinate point of the constraint condition sampling point, v represents the physical parameter of the control equation used to establish the physical constraint, D(x (i) ; v) represents x (i) Substituting the predicted concentration value at into the residual calculated by the control equation, It means to find the square of L2 norm.

[0083] According to the physical constraints designed as above, the predicted concentration values ​​of the constraint sampling points are substituted into the control equation to calculate the residuals. The residuals represent the degree of conformity of the prediction results to the physical laws, and the L2 norm square is used to measure the size of the residuals, thereby ensuring that the prediction results follow the basic laws of the hydrodynamic-water quality model. In addition, the introduction of physical constraints through residual calculation can enhance the generalization ability of the model.

[0084] In some implementations, the distribution of constraint condition sampling points may be determined according to the following method:

[0085] A spatiotemporal concentration variation data set is obtained by large eddy simulation (LES), wherein the data set includes flow velocity and concentration values ​​of each grid point in the flow field at different times;

[0086] Constraint sampling points are selected based on the spatiotemporal concentration variation data set, wherein the constraint sampling points in areas with large flow velocity or concentration gradient variations are denser than the constraint sampling points in areas with small flow velocity or concentration gradient variations.

[0087] According to this embodiment, a complete spatiotemporal concentration variation data set can be first obtained through large eddy simulation (LES). This data set records the concentration value and flow velocity information of each grid point in the calculation domain at different times, and can reflect the concentration distribution and flow characteristics in the flow field. Based on the data set generated by LES, areas where physical quantities in the flow field change dramatically can be identified. Therefore, in areas where the flow velocity changes greatly or the concentration gradient is significant, denser constraint sampling points can be arranged to ensure that the prediction results of these key areas meet the physical constraints; and in areas where the physical quantities change slowly, the sampling point density can be appropriately reduced. This non-uniformly distributed sampling strategy helps to improve the prediction accuracy of the PINN model in complex flow fields.

[0088] In other examples, under complex flow conditions, the physical constraint term L Ω (Θ) can be obtained by substituting the predicted concentration value of the constraint sampling point into the residual calculated by large eddy simulation (LES). When the flow field is complex, it may be difficult to reliably calculate the physical constraint terms directly through the control equation. In this case, large eddy simulation can be used to calculate the residual as a measure of whether the predicted value meets the physical law. It can solve the problem of insufficient control equations in complex flow fields, and LES can more accurately simulate the physical characteristics of complex flow fields, ensuring the applicability of the PINN model under complex conditions.

[0089] In some embodiments, the outputs of the simple model and the PINN model may be combined according to the following formula:

[0090]

[0091] Among them, w simple +w PINN =1, is the final time-averaged concentration field output by the time-averaged concentration gradient fast inversion model, This is the calculation result of the simple model. is the calculation result of PINN model, w simple for and w PINN is the weight coefficient.

[0092] In some embodiments, the number of sampling points can be increased. exist The weight ratio in, that is, increasing w PINN At the same time, w is reduced accordingly simple , thereby achieving smooth switching and avoiding instability caused by sudden switching, while ensuring the inversion accuracy and continuity of the inversion results.

[0093] In some embodiments, the concentration gradient direction error and the concentration inversion error of the calculation result of the simple model can be calculated; if any one of the concentration gradient direction error and the concentration inversion error exceeds the corresponding preset threshold, then w is set simple =0,w PINN = 1. The concentration gradient direction error can be obtained by calculating the angle between the gradient direction obtained by the simple model and the reference direction, and the reference direction can be calculated from the measured data at adjacent moments. The concentration inversion error can be measured by calculating the root mean square error between the simple model calculation value and the collected concentration data.

[0094] According to this embodiment, when any error exceeds a preset threshold, for example, the concentration gradient direction error exceeds the concentration gradient direction error threshold, and / or the concentration inversion error exceeds the concentration inversion error threshold, it can be considered that the performance of the simple model has significantly deteriorated. At this time, in order to prevent it from continuing to affect the results, the weight coefficient of the simple model can be directly set to zero, and the output result of the PINN model is fully adopted, which is conducive to further improving the inversion performance.

[0095] Step 3: Determine the next desired navigation path according to the concentration gradient direction of the time-averaged concentration field obtained by inversion.

[0096] The concentration gradient direction can be calculated based on the time-averaged concentration field obtained by real-time inversion, and the next desired navigation path can be determined based on the concentration gradient direction, thereby guiding the unmanned ship to move in the direction of increasing concentration and gradually approach the pollution source. Since the concentration field is processed by time-average, this navigation method can avoid interference from instantaneous concentration fluctuations and improve the stability and reliability of tracing.

[0097] In practical applications, other goals can also be used to optimize the navigation path. For example, a more uniform sampling coverage of the target area can be considered, so that more comprehensive concentration field data can be obtained, the accuracy of the inversion results can be improved, and misjudgment due to local high concentrations can be avoided.

[0098] Step 4: Based on the expected navigation path and in combination with the motion characteristics of the unmanned ship and the water environment characteristics, the moving direction and speed of the unmanned ship are adjusted in real time.

[0099] The desired navigation path can be tracked by adjusting the moving direction and speed of the unmanned ship.

[0100] In practical applications, the steering angle and acceleration can be reasonably adjusted based on the inertial characteristics of the unmanned ship. Since the unmanned ship cannot suddenly change its motion state, its dynamic characteristics can be comprehensively considered when turning or changing speed.

[0101] The course and speed can be adjusted according to the current water flow. The unmanned boat will be affected by the change of flow speed during movement, and the deviation caused by the water flow needs to be compensated in real time.

[0102] At the same time, sensors can be used to monitor the surrounding environment in real time, identify and avoid obstacles in the water, such as tree stumps, protruding stones, etc., to ensure navigation safety.

[0103] The travel speed can also be reduced in areas of high concentration gradient. Areas with large concentration gradients usually correspond to areas where pollutant diffusion is more significant, and the data in these areas are more critical for accurately locating pollution sources. Reducing the speed can obtain more sampling data points within a unit distance, that is, collect more concentration data, and the increase in sampling density helps the PINN model to more accurately invert the concentration field of the area, and further provide more reliable gradient direction information to guide subsequent navigation. This speed control strategy improves the accuracy of tracing by obtaining more data in key areas. Those skilled in the art can judge whether the concentration gradient in a certain area has reached a high concentration gradient based on experience, real-time conditions and tasks.

[0104] Based on the planned navigation path, a comprehensive control strategy that takes into account the motion characteristics of the unmanned ship and environmental factors is adopted, which not only ensures the safety and reliability of navigation, but also improves the efficiency of sampling.

[0105] The concentration gradient-based autonomous navigation method for unmanned boats in water environments proposed in this embodiment significantly improves the efficiency and reliability of tracing. According to this embodiment, a strategy of dynamically combining a simple model and a PINN model is adopted to quickly obtain a preliminary estimate when there is less sampled data, avoiding delays in tracing due to waiting for the model to converge. As the data increases, it gradually transitions to a more accurate PINN model, achieving an organic combination of rapid response and high accuracy.

[0106] The design of the PINN model introduces physical constraints. By calculating the residuals at the constraint sampling points, the prediction results are ensured to conform to the hydrodynamic-water quality laws. At the same time, the large eddy simulation is used to optimize the distribution of constraint points, further improving the prediction accuracy in complex flow fields. This constraint mechanism based on physical laws significantly enhances the generalization ability and prediction reliability of the model.

[0107] In terms of navigation control, this embodiment determines the expected navigation path based on the concentration gradient, and makes real-time adjustments based on the motion characteristics of the unmanned ship and the characteristics of the water environment. By reducing the speed in high concentration gradient areas and increasing the sampling density, it not only ensures the safety of navigation, but also improves the efficiency of data collection. This navigation solution that comprehensively considers multiple factors significantly improves the accuracy and efficiency of pollution source positioning.

[0108] Figure 2 A schematic diagram of a technical framework according to an exemplary embodiment of the present disclosure is shown.

[0109] The top blue module in the figure shows the combination of the basic technical support modules - large eddy simulation (LES) and physical information neural network (PINN).

[0110] The middle part contains four main technical parts (marked by dotted boxes):

[0111] Highly sensitive water quality information collection system, including water quality sampling and detection functions;

[0112] The time-averaged concentration field inversion method based on instantaneous measurement data, the concentration field inversion under the condition of measurement points (for example, using a simple model), and the fast inversion correction method when the monitoring point data increases (for example, gradually switching to the PINN model), that is, through the monitoring-inversion real-time communication model, the expected navigation path is planned based on high-precision water quality concentration detection to obtain the direction and speed control;

[0113] The robot autonomous navigation method based on concentration gradient integrates navigation planning and control, and autonomously plans a reasonable cruise route based on concentration gradient and inspection requirements. It also explores the accuracy of inspection traceability based on the OpenFOAM virtual simulation platform;

[0114] The scheme can be tested through indoor water pool experiments, autonomous navigation robot design and testing, and field water body testing.

[0115] like Figure 2 As shown at the bottom, the tested technical solutions can be applied to the intelligent monitoring and traceability of water environment.

[0116] The following combination Figure 3 The schematic diagram of the tracing route shown is an exemplary illustration of the tracing process of a certain unmanned ship.

[0117] In the first step, the unmanned ship is positioned at the starting point B and the monitoring system is activated. The concentration data of the current position is collected in real time through sensors and input into the time-averaged concentration gradient rapid inversion model after preliminary screening.

[0118] In the second step, the inversion model dynamically adjusts the weight ratio of the simple model and the PINN model according to the current sampling data. In the early stage of sampling, the interpolation results of the simple model are mainly relied on. As the data accumulates, the weight of the PINN model is gradually increased to obtain the time-averaged concentration field of the water body.

[0119] The third step is to calculate the concentration gradient direction based on the time-averaged concentration field obtained by inversion and determine the expected navigation path. The movement direction and speed are adjusted in real time based on the motion characteristics of the unmanned ship and the current water environment characteristics. During the process, the speed is reduced in the high concentration gradient area to increase the sampling density.

[0120] The fourth step is to continuously evaluate the concentration value at the current location during the sampling process. If a local maximum value is detected, it indicates that the pollution source may be close. At this time, intensive sampling is performed in this area to confirm the location of the pollution source. If the concentration value at point B is the maximum value monitored, it can be inferred that point A is the possible location of the sewage outlet, and the patrol process is terminated after confirmation to complete the tracing task.

[0121] The dynamic inversion and concentration gradient-based adaptive navigation strategy provided by the present invention can effectively improve the accuracy and efficiency of tracing.

[0122] The present disclosure also provides an autonomous navigation device for an unmanned ship in a water environment based on a concentration gradient, comprising:

[0123] A real-time acquisition unit, used to collect concentration data in the water environment through sensors carried by unmanned vessels;

[0124] A model inversion unit is used to construct a time-averaged concentration gradient fast inversion model, wherein the time-averaged concentration gradient fast inversion model includes a simple model that can be directly calculated based on the currently collected concentration data and a physical information neural network PINN model, and dynamically adjusts the weight ratio of the two models to process the collected concentration data to obtain the time-averaged concentration field of the water body;

[0125] A path determination unit, used to determine the next desired navigation path according to the concentration gradient direction of the time-averaged concentration field obtained by inversion;

[0126] The real-time navigation unit is used to adjust the moving direction and speed of the unmanned ship in real time based on the expected navigation path and in combination with the motion characteristics of the unmanned ship and the environmental characteristics.

[0127] For other details and advantages of this embodiment, please refer to the above description.

[0128] Figure 4 An electronic device provided for at least one embodiment of the present disclosure comprises a memory and a processor, wherein the memory is used to store computer instructions executable on the processor, and the processor is used to implement the concentration gradient-based water environment unmanned ship autonomous navigation method described in any embodiment or implementation method of the present disclosure when executing the computer instructions.

[0129] At least one embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for autonomous navigation of an unmanned ship in a water environment based on a concentration gradient as described in any embodiment or implementation of the present disclosure.

[0130] It should be understood by those skilled in the art that one or more embodiments of the present specification may be provided as a method, system or computer program product. Therefore, one or more embodiments of the present specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, one or more embodiments of the present specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0131] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the data processing device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0132] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0133] The embodiments of the subject matter and functional operations described in this specification may be implemented in the following: digital electronic circuits, tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or a combination of one or more of them. The embodiments of the subject matter described in this specification may be implemented as one or more computer programs, i.e., one or more modules in computer program instructions encoded on a tangible non-temporary program carrier to be executed by a data processing device or to control the operation of the data processing device. Alternatively or additionally, the program instructions may be encoded on an artificially generated propagation signal, such as a machine-generated electrical, optical or electromagnetic signal, which is generated to encode information and transmit it to a suitable receiver device for execution by a data processing device. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.

[0134] The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform corresponding functions by operating on input data and generating output. The processes and logic flows can also be performed by special purpose logic circuits, such as FPGAs (field programmable gate arrays) or ASICs (application-specific integrated circuits), and the apparatus can also be implemented as special purpose logic circuits.

[0135] Computers suitable for executing computer programs include, for example, general and / or special microprocessors, or any other type of central processing unit. Typically, the central processing unit will receive instructions and data from a read-only memory and / or a random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, the computer will also include one or more large-capacity storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, or the computer will be operably coupled to this large-capacity storage device to receive data from it or to transmit data to it, or both. However, the computer does not necessarily have such a device. In addition, the computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name a few.

[0136] Computer readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including, for example, semiconductor memory devices (e.g., EPROM, EEPROM and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD ROM and DVD-ROM disks. The processor and memory may be supplemented by, or incorporated in, special purpose logic circuitry.

[0137] Although this specification includes many specific implementation details, these should not be interpreted as limiting the scope of any invention or the scope of protection claimed, but are mainly used to describe the features of the specific embodiments of specific inventions. Certain features described in multiple embodiments in this specification may also be implemented in combination in a single embodiment. On the other hand, the various features described in a single embodiment may also be implemented separately in multiple embodiments or in any suitable sub-combination. In addition, although features may work in certain combinations as described above and even initially claim protection, one or more features from the claimed combination may be removed from the combination in some cases, and the claimed combination may point to a sub-combination or a variation of a sub-combination.

[0138] Similarly, although operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring that these operations be performed in the particular order shown or performed sequentially, or requiring that all illustrated operations be performed to achieve the desired results. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

[0139] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the particular order or sequential order shown to achieve the desired results. In some implementations, multitasking and parallel processing may be advantageous.

[0140] The above description is merely a preferred embodiment of one or more embodiments of the present specification and is not intended to limit one or more embodiments of the present specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present specification shall be included in the scope of protection of one or more embodiments of the present specification.

Claims

1. A method for autonomous navigation of an unmanned ship in a water environment based on concentration gradient, characterized in that: include: Collect concentration data in the water environment through sensors carried by unmanned vessels; Constructing a time-averaged concentration gradient rapid inversion model, the time-averaged concentration gradient rapid inversion model includes a simple model that can be directly calculated based on currently collected concentration data and a physical information neural network PINN model, and dynamically adjusting the weight ratio of the two models to process the collected concentration data to obtain the time-averaged concentration field of the water body; Determine the next desired navigation path based on the concentration gradient direction of the time-averaged concentration field obtained by inversion; Based on the expected navigation path, combined with the motion characteristics of the unmanned ship and the environmental characteristics, the moving direction and speed of the unmanned ship are adjusted in real time.

2. The method according to claim 1, characterized in that The PINN model uses the control equation of the hydrodynamic-water quality model as the physical constraint condition in the training process. The loss function includes a data fitting term and a physical constraint term. The data fitting term is used to measure the deviation between the predicted concentration value and the observed concentration value of the observation data sampling point, and the physical constraint term is used to measure the fit between the predicted concentration value of the constraint condition sampling point and the physical law.

3. The method according to claim 2, characterized in that The data fitting term is determined according to the following formula Among them, Θ is the PINN network parameter, x (i) is the spatial coordinate point, is the number of observation data sampling points, For PINN network in x (i) The predicted concentration value at g(x (i) ) is in x (i) The concentration values ​​collected at It means to find the square of L2 norm.

4. The method according to claim 2, characterized in that: The method further comprises: A spatiotemporal concentration variation data set is obtained by large eddy simulation, wherein the spatiotemporal concentration variation data set includes flow velocity and concentration values ​​of each grid point in the flow field at different times; Constraint sampling points are selected based on the spatiotemporal concentration variation data set, wherein the constraint sampling points in areas with large flow velocity or concentration gradient variations are denser than the constraint sampling points in areas with small flow velocity or concentration gradient variations.

5. The method according to claim 2, characterized in that: The physical constraint term L is determined according to the following formula Ω (Θ): Among them, N Ω is the number of sampling points under constraint conditions, x (i) is a spatial coordinate point, v represents the physical parameter of the control equation used to establish the physical constraint, D(x (i) ; v) represents x (i) Substituting the predicted concentration value at into the residual calculated by the control equation, It means to find the square of L2 norm.

6. The method according to claim 2, characterized in that Under complex flow field conditions, the physical constraint term is obtained by substituting the predicted concentration value of the constraint condition sampling point into the large eddy simulation calculation residual.

7. The method according to claim 1, characterized in that The time-averaged concentration gradient rapid inversion model combines the outputs of the simple model and the PINN model according to the following formula: Among them, w simple +w PINN =1, is the final time-averaged concentration field output by the time-averaged concentration gradient fast inversion model, The calculation results of the simple model are: is the calculation result of PINN model, w simple and w PINN is the weight coefficient.

8. The method according to claim 7, characterized in that Dynamically adjusting the weight ratio of the two models includes: As the number of sampling points increases exist The weight ratio in .

9. The method according to claim 7, characterized in that: Dynamically adjusting the weight ratio of the two models includes: Calculate the concentration gradient direction error and concentration inversion error of the calculation results of the simple model; If any one of the concentration gradient direction error and the concentration inversion error exceeds the corresponding preset threshold, then w is set simple =0,w PINN =1.

10. The method according to claim 1, characterized in that Real-time adjustment of the moving direction and speed of the unmanned ship includes: Monitor the surrounding environment in real time through sensors to identify and avoid obstacles in the water; Adjust the steering angle and acceleration according to the unmanned ship's own inertia characteristics; Adjust course and speed according to current water conditions; Reduce travel speed in areas of high concentration gradient to increase sampling density.

11. An autonomous navigation device for an unmanned ship in a water environment based on a concentration gradient, characterized in that: include: A real-time acquisition unit, used to collect concentration data in the water environment through sensors carried by unmanned vessels; A model inversion unit is used to construct a time-averaged concentration gradient fast inversion model, wherein the time-averaged concentration gradient fast inversion model includes a simple model that can be directly calculated based on the currently collected concentration data and a physical information neural network PINN model, and dynamically adjusts the weight ratio of the two models to process the collected concentration data to obtain the time-averaged concentration field of the water body; A path determination unit, used to determine the next desired navigation path according to the concentration gradient direction of the time-averaged concentration field obtained by inversion; The real-time navigation unit is used to adjust the moving direction and speed of the unmanned ship in real time based on the expected navigation path and in combination with the motion characteristics of the unmanned ship and the environmental characteristics.

12. An electronic device, characterized in that: The device comprises a memory and a processor, wherein the memory is used to store computer instructions executable on the processor, and the processor is used to implement the method according to any one of claims 1 to 10 when executing the computer instructions.

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

Images