Submarine concealment rapid prediction method based on ordered multi-classification Logistic regression model

Abstract

The invention discloses a submarine concealment rapid prediction method based on an ordered multi-classification Logistic regression model. The method comprises the following steps: respectively acquiring attribute data of sea areas passed by sonar of a submarine and an antisubmarine ship on a collision-free route; the concealment level of the submarine is calculated on a section of a vertical sea surface extending from a connecting line of the submarine and the sonar of the anti-submarine naval vessel as a label, the concealment level is calculated by the signal margin between the sonar of the anti-submarine naval vessel and the submarine, and the attribute and the label are used for training an ordered multi-classification Logistic model; and using the ordered multi-classification Logistic model obtained by training to predict the submarine concealment, collecting attribute data of a sea area through which the submarine and the antisubmarine ship sonar pass on another collision-free route, and inputting the collected data into the ordered multi-classification Logistic model and the Bellhop model to respectively obtain a predicted value and a simulation value of the submarine concealment level. And constructing a confusion matrix according to the predicted value and the simulation value, and performing quantitative evaluation on the ordered multi-classification Logistic model by using a multi-classification comprehensive evaluation secondary index under the confusion matrix.

Description

Technical Field

[0001] The invention relates to a method for quickly predicting submarine stealth based on an ordered multi-classification Logistic regression model, and belongs to the technical field of submarine stealth prediction. Background Art

[0002] One of the core challenges in achieving efficient submarine stealth prediction is how to quickly and accurately assess a submarine's stealth level in a dynamic ocean environment. Traditional methods rely on complex acoustic propagation models (such as the Bellhop model) and use active sonar equations to calculate signal margins to categorize stealth levels. While this process can provide relatively accurate predictions, it suffers from low computational efficiency and high resource consumption. The Bellhop model integrates multidimensional environmental data such as seawater temperature, salinity, and seabed topography, relies on high-performance computing platforms, and a single calculation can take up to several minutes, making it difficult to meet the demands of real-time decision-making. Furthermore, traditional methods require manual parameter adjustment and data cleaning, making them ineffective in adapting to the dynamic maneuvers of submarines and sonars, as well as real-time changes in the ocean environment. Meanwhile, while existing machine learning methods have improved computational efficiency to some extent, they still have significant drawbacks, such as a disconnect between prediction results and physical laws, and a difficulty balancing computational speed and accuracy. Existing research has attempted to improve efficiency through simulation optimization or distributed computing, but still faces bottlenecks such as high resource dependence and a single evaluation system, making it difficult to meet the demands of real-time prediction in complex environments. Summary of the Invention

[0003] Purpose of the Invention: In response to the problems and shortcomings of the existing technology, the present invention aims to establish a low-cost, high-efficiency prediction solution for submarine stealth assessment scenarios with limited resources. The present invention proposes a solution based on an ordered multi-classification logistic regression model. The route is determined by seabed terrain collision detection, the model is trained with data on the route, and data such as ocean noise and sound speed profile are integrated to achieve dynamic stealth prediction. This method breaks through the limitations of traditional models such as high complexity and strong resource dependence, providing lightweight and efficient technical support for submarine stealth assessment. It can quickly and accurately predict submarine stealth, and the prediction speed is significantly improved compared to the traditional Bellhop model, which has important application value.

[0004] Technical solution: A rapid prediction method for submarine stealthiness based on an ordered multi-classification logistic regression model. The method determines collision-free routes for anti-submarine ship sonars and submarines, and the submarines and anti-submarine ship sonars respectively sail along the collision-free routes. The temperature, salinity, current, seabed depth and other data of the sea areas surrounding the submarine and anti-submarine ship routes are collected and gridded, and the submarines and anti-submarine ship sonars respectively sail along the routes. The temperature, salinity, current and seabed depth data of the grid bodies passed by the submarine route and the anti-submarine ship sonar route are calculated as attributes. The temperature, salinity, current and seabed depth data of all grid bodies are calculated by the cross-section perpendicular to the sea surface extending from the connection line between the submarine and anti-submarine ship sonars, and the sound propagation loss is input into the Bellhop model. The sound propagation loss is input into the active sonar equation to calculate the signal margin. The signal margin is evenly divided according to size to obtain the submarine stealthiness level as a label. The attributes and labels are used together to train the ordered multi-classification logistic model.

[0005] The trained ordered multi-classification logistic model was used to predict submarine stealth. During the prediction process, an alternate collision-free route was set for the submarine. The submarine and anti-submarine vessel sonars were used to navigate along the route. Temperature, salinity, current, and seabed depth data for the sea areas surrounding the route were collected and gridded. The temperature, salinity, current, and seabed depth data for the grids passed by the submarine and anti-submarine vessel sonars along the route were calculated and input into the ordered multi-classification logistic model to predict the submarine's stealthiness. The ordered multi-classification logistic model was evaluated using a confusion matrix and its secondary multi-classification comprehensive evaluation index.

[0006] The route is obtained by setting the shapes of the anti-submarine ship sonar and submarine routes and the depths of their respective cruising, selecting the endpoints of the line segments on the anti-submarine ship sonar and submarine routes respectively, and constructing the straight line equations of the line segments.

[0007] The method of collecting temperature, salinity, current, and seabed depth data for the sea area surrounding the route refers to collecting temperature, salinity, current, and seabed depth data for a square volume of sea area. The method of framing the square volume range is to find all the longitude, latitude, and depth coordinates of submarines on the route and all the longitude, latitude, and depth coordinates of anti-submarine ship sonar routes, find the minimum and maximum values ​​on each axis of longitude, latitude, and depth, and then frame a square volume using the minimum and maximum values. The collected data is discretized, labeled with longitude, latitude, and depth, and gridded by longitude, latitude, and depth. The data of all grid volumes in the square volume are extracted and used to calculate the sound speed profile as Bellhop input and for training and prediction of the ordered multi-classification logistic model, thereby reducing the amount of calculation.

[0008] The Bellhop model calculates sound propagation loss, providing the sound propagation loss of submarine and anti-submarine ship sonars in any sea area. Specifically, the sound velocity profile and seabed topography of the sea area are calculated as input data for the Bellhop model.

[0009] The sound velocity profile data is calculated as follows: When the submarine and anti-submarine ship sonars reach a certain location along their route (the location is marked with latitude, longitude and depth coordinates), the vertical sea surface profile extending from the line connecting the submarine and anti-submarine ship sonars is called a cross section. Based on a gridded square ocean area, all data (including temperature, salinity, ocean currents, and seabed depth) is calculated for each grid cell that the cross section passes through. This data is then substituted into the sound velocity profile formula to obtain the sound velocity for each grid cell on the cross section.

[0010] The Bellhop model calculates and outputs the sound propagation loss at every point along the entire cross-section. This refers to each grid cell passed through when the square body is cut along the cross-section. The sound propagation loss of the grid cell where the submarine is located is calculated based on the submarine's latitude, longitude, and depth coordinates. The signal margin is calculated based on variables such as the sound propagation loss, the active sonar source level, and ambient noise. As the anti-submarine ship sonar and the submarine continue to navigate along their established route, the position of the spatial connection line between the anti-submarine ship sonar and the submarine changes, generating multiple signal margin values. These signal margin values ​​are sorted from smallest to largest and assigned to multiple levels in ascending order, representing the submarine's stealth level.

[0011] The multivariate interpolation model for constructing the seabed topography of the entire square sea area is as follows: input the longitude and latitude coordinates of a certain location in the square sea area, and output the seabed depth of the location.

[0012] The method for calculating the seabed topography data on the cross section is as follows: when the submarine and anti-submarine ship sonars sail to a certain location along the route, the cross section perpendicular to the sea surface extended from the line connecting the submarine and anti-submarine ship sonars is tangent to the sea surface to form a line segment of the sea surface. Multiple equal points are inserted into the line segment to obtain the latitude and longitude coordinates of all equal points. According to the Haversine formula, the distance from each point to the anti-submarine ship sonar is calculated; the multivariate interpolation model is used to calculate the seabed depth at each point to obtain the seabed topography data on the cross section.

[0013] The method for calculating the seabed topography data along the route is as follows: project the sonar routes of submarines and anti-submarine ships onto the sea surface to form one or more line segments on the sea surface, insert multiple equal points into each line segment, obtain the latitude and longitude coordinates of all equal points, substitute them into the multivariate interpolation model to calculate the seabed depth at each point, and obtain the seabed topography data along the route.

[0014] A submarine's route is set on the premise that it will not collide with seamounts. Calculation methods for the seabed topography along the route can be used to detect whether the submarine's dive depth along the route will collide with seamounts.

[0015] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for rapidly predicting submarine stealth based on an ordered multi-classification logistic regression model are implemented.

[0016] A computer-readable storage medium stores a computer program for executing the above-mentioned submarine stealth fast prediction method based on the ordered multi-classification Logistic regression model.

[0017] Beneficial Effects: By incorporating an ordered multi-classification logistic regression model and a Bellhop model, this present invention enables rapid and accurate prediction of submarine stealth levels along different maneuvering routes. Using this prediction method, submarines can obtain real-time information about their stealth status in complex marine environments, providing a scientific basis for decision-making. This present invention offers significant advantages in computational efficiency, prediction accuracy, model adaptability, operational simplicity, data security, and application prospects. It is widely applicable in a wide range of fields, including modern marine decision-making, marine scientific research, and underwater safety monitoring, possessing significant value and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic diagram of the process of calculating submarine stealth using an ordered multi-classification logistic model according to an embodiment of the present invention; Figure 2 A schematic diagram of a method for finding and collecting all data of a route passing through a grid volume according to an embodiment of the present invention; Figure 3 An aerial view of a "cross"-shaped route according to an embodiment of the present invention; Figure 4 An aerial view of a "Z"-shaped route according to an embodiment of the present invention; Figure 5 A schematic cross-sectional view of a cube enclosing a certain route of an anti-submarine ship and a submarine according to an embodiment of the present invention; Figure 6 Schematic diagram of a method for reducing a seabed terrain database based on gridding according to an embodiment of the present invention; Figure 7 Schematic diagram of the evaluation index module structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The present invention is further illustrated below with reference to specific examples. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, modifications of various equivalent forms of the present invention made by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0020] A method for rapidly predicting submarine stealth based on an ordered multi-classification logistic regression model is proposed. The submarine and anti-submarine ship sonars are respectively sailing along certain collision-free routes. The method includes the following steps: Step 1: Collect temperature, salinity, current, and seabed depth data of the sea area surrounding the submarine and anti-submarine ship sonar routes. Specifically, it refers to collecting temperature, salinity, current, and seabed depth data of a square sea area. The square range framing method is to find all the longitude, latitude, and depth coordinates on the submarine route and all the longitude, latitude, and depth coordinates on the anti-submarine ship sonar route, and find the minimum and maximum values ​​on each axis of the longitude axis, latitude axis, and depth axis respectively, and frame a square using the minimum and maximum values.

[0021] Gridding: Discretize the collected data, annotate the data with longitude, latitude, and depth, and grid the data by longitude, latitude, and depth. Extract the grid data of the area where the anti-submarine ship sonar and submarine maneuvering routes are located. Use this data to calculate the sound speed profile as Bellhop input and for Logistic model training and prediction, reducing the amount of calculation.

[0022] Sound speed profile at any location: Construct a vertical sea surface profile extending from the sonar line connecting the submarine and anti-submarine ship. Calculate all the data (including temperature, salinity, ocean current, and seabed depth) in each grid volume passed through when the profile cuts through the square volume. Substitute all the data in the sound speed profile formula into the formula to obtain the sound speed of each grid on the profile.

[0023] Seabed topography interpolation model: Construct a multivariate interpolation model of seabed depth-latitude and longitude coordinates of the square volume area containing the route. Substitute the longitude and latitude coordinates of any location in the area into the multivariate interpolation model to obtain the seabed depth at that location.

[0024] Collision detection: Through the multivariate interpolation model, it can quickly detect whether the submarine's maneuvering route collides with the seabed terrain to ensure that the submarine route is safe.

[0025] Stealth level marking: The signal margin is calculated based on the Bellhop model, and the signal margin values ​​are sorted from small to large and divided into intervals to obtain the stealth level of the submarine from 1 to 5.

[0026] Step 2: Use the ordered multi-classification logistic regression model to predict submarine stealth. The ordered multi-classification logistic regression model divides the five stealth levels into four binary logistic regression problems, sharing the regression coefficients and only different constant terms. Set the stochastic gradient descent method to optimize the model parameters to ensure fast calculation speed and stable convergence.

[0027] Step 3 trains an ordered multi-class logistic regression model. The independent variables are temperature, salinity, ocean currents, and seafloor depth data collected along the "cross" maneuvering route. The ocean environment data here comes from the grid volumes collected and gridded in Step 1. These grid volumes are located along the routes of submarines and anti-submarine ship sonars. The dependent variable is the stealth level of the submarines along the corresponding routes. The ordered multi-class logistic regression model is trained, and the parameters are optimized using a stochastic gradient descent algorithm to minimize prediction error.

[0028] Step 4: Use the trained ordered multi-classification logistic regression model to predict the submarine stealthiness: Input the ocean environment data on the "Z"-shaped maneuvering route into the trained ordered multi-classification logistic regression model to obtain the submarine stealthiness level prediction result.

[0029] In step 5, we compared the Bellhop model results with the ordered multi-classification logistic regression model predictions using accuracy, macro-average precision, macro-average recall, and a comprehensive evaluation metric (F1) to evaluate the model's performance. This evaluation provides guidance for model optimization and improvement.

[0030] like Figure 1As shown, the rapid submarine stealth prediction process based on an ordered multi-classification logistic regression model includes: a cube A enclosing the "X" route for anti-submarine sonars and submarines, and a cube B enclosing the "Z" route for anti-submarine sonars and submarines. The waters of A and B are adjacent, and submarines along the routes avoid collisions with the seabed or reefs. Ocean environmental data, such as temperature, salinity, currents, and seabed depth, are collected from cubes A and B and then gridded. Ocean environmental data (temperature, salinity, currents, and seabed depth) is calculated for all grids along the X and Z routes for anti-submarine sonars and submarines passing through them. All ocean environmental data (temperature, salinity, currents, and seabed depth) collected along the X route are used as attributes and fed into the Bellhop model to calculate acoustic propagation loss. This acoustic propagation loss is then substituted into the active sonar equation to obtain the signal margin. The signal margin is then evenly divided according to the size of the signal margin to determine the submarine stealth level along the X route as a label. The attributes and labels are used together to train the ordered multi-classification logistic model. The trained ordered multi-classification logistic model can be used for prediction. It accepts all ocean environmental data collected along the "Z" route (temperature, salinity, currents, and seabed depth) and outputs a predicted value for the submarine's stealthiness. This data is also fed into the Bellhop model to calculate acoustic propagation loss. This acoustic propagation loss is then substituted into the active sonar equation to obtain the signal margin. The signal margin is then evenly divided according to its size, resulting in the submarine's stealthiness along the "Z" route as a simulated value. A confusion matrix is ​​constructed based on the predicted and simulated values. The ordered multi-classification comprehensive evaluation secondary indicator derived from the confusion matrix is ​​then used to quantitatively evaluate the ordered multi-classification logistic model.

[0031] The ocean data collection method for the “X” and “Z” shaped routes passing through all grid volumes is to find the endpoints of the line segments on the route. The straight line equation of the line segment is: (1) The longitude coordinate is x , the latitude coordinate is y The longitude and latitude coordinates of the endpoints of a line segment on the route are ( x 1, y 1) and ( x 2, y 2). Under the condition of given depth, according to the longitude interval [ x 1, x 2], and the latitude interval [ y 1, y 2] The mesh body where the above line segment passes through the square body can be found, such as Figure 2As shown, the grid body of the filled part is the grid body in the square body passed by the line segment with endpoints 1 and endpoint 2 as the end points.

[0032] Extract all the data from each line segment in the route that passes through the grid to form the training model dataset and the prediction model dataset. The dataset used for training the model is specifically the construction of the "cross" route path and the collection of ocean data on the path, such as Figure 3 The figure shows an aerial view of the "X"-shaped route. The sonar routes of submarines and anti-submarine ships form two intersecting straight lines. The submarine's navigation path is: a->o->b->o->d->o->c->o->a. The anti-submarine ship's sonar navigation path is: d->o->c->o->a->o->b->o->d. The collection of ocean data on the route includes the seawater temperature when the anti-submarine ship sonar route passes through the grid body, the seawater salinity when the sonar route passes through the grid body, the seawater flow rate when the sonar route passes through the grid body, the seawater temperature when the submarine route passes through the grid body, the seawater salinity when the submarine route passes through the grid body, the seawater flow rate when the submarine route passes through the grid body, the seawater depth when the submarine route passes through the grid body, and the sea level distance between the sonar and the submarine. Among them, the data set used for the prediction model is specifically the "Z"-shaped maneuvering route path construction and the collection of ocean data on the path, such as Figure 4 The image shows an aerial view of a zigzag route. The submarine and anti-submarine ship sonar routes form three zigzag line segments. The submarine's route is: d->c->b->a. The anti-submarine ship's sonar route is: a->b->c->d. Oceanographic data collected along the route includes seawater temperature, salinity, and current velocity for the anti-submarine ship's sonar route through the grid, seawater temperature, salinity, and current velocity for the submarine's route through the grid, seawater depth, and the sea level distance between the sonar and submarine.

[0033] The signal margin (SE) is calculated using the active sonar equation. The signal margin is defined as the amount by which the sonar receiver receives a signal that exceeds the signal required for detection, expressed in dB. The size of the signal margin directly determines whether the sonar can detect underwater objects. SE value The magnitude of the sound signal received by the sonar receiver is the result of the amplification and reduction of multiple sound signals in the ocean, which is the multiplication of multiple signals. After taking the logarithm, the multiplication of multiple sound signals is converted into addition. dB The expression is: (2) The signal margin equation of active sonar after taking the logarithm is: (3) The active sonar equation can be calculated by SL, NL, DI, DT, and TL. In the active sonar mode, SL represents the active sonar source level, which is the sound intensity emitted by the sound source itself. If the anti-submarine ship uses active sonar, the sound source is the sound signal emitted by the sonar transmitter, where P is the sound power emitted by the sonar, ranging from 1 to 40 kW. The sound source level SL is defined as: (4) NL stands for ambient background noise. The ocean ambient noise level NL is mainly composed of wind-induced noise, which is defined as: (5) in, v wind is the wind speed at 10 meters above the sea surface, f is the sonar sound frequency.

[0034] DI stands for Directivity Index, and DT stands for Sonar Signal Detection Threshold. The Directivity Index represents the array gain of a sonar receiver system and is typically a fixed value. When the signal-to-noise ratio of the input acoustic signal exceeds the detection threshold, the sonar detects a target. The detection threshold is also fixed, but varies between sonar models.

[0035] TL stands for sound propagation loss. The sound propagation path is the round trip of sound from the sonar to the submarine, so the propagation loss is multiplied by 2. The sound propagation loss is the loss caused by the round trip of sound from the anti-submarine ship sonar to the submarine when the anti-submarine ship sonar and the submarine are sailing along a predetermined route. It is only necessary to calculate the sound propagation loss on the ocean section determined by the positions of the anti-submarine ship sonar and the submarine (the sound not on the section either cannot be transmitted to the submarine or cannot be transmitted to the receiving sonar of the anti-submarine ship after being reflected by the submarine). The Bellhop model is used for calculation. In order to achieve long-distance propagation, sound waves must be reflected from the seabed. To this end, the input parameters of the Bellhop model must be configured with the sound speed profile (.ssp file) and seabed topography (.bty file) on the line connecting the submarine and the anti-submarine ship to more accurately calculate the sound propagation loss. At this time, the profile (cross section) and the straight line formed by the intersection of the profile and the sea surface are as follows. Figure 5 As shown, the profile equation is parallel to the depth coordinate axis z in space, assuming that the longitude coordinate axis is x , the latitude axis is y , then the profile equation is: (6) The method for calculating the sound velocity profile is to find each mesh volume that the profile passes through when it cuts through the square body. The specific method is: ① List the maximum and minimum values ​​of the longitude and latitude of all grid bodies. If the maximum and minimum values ​​of the longitude of a grid are substituted into the x value of the profile equation, and the resulting y value is within the range of the maximum and minimum values ​​of its latitude, then the profile passes through the grid body; otherwise, it does not pass through.

[0036] ② Collect all the data (including temperature, salinity, ocean current, and seabed depth) that the profile passes through in each grid volume. Substitute the data into the sound velocity profile formula, which uses the Wood empirical model, and the formula can be expressed as: (7) In the above formula, the speed of sound is c Indicated by m / s, temperature is expressed as T Indicated by °C, salinity is expressed in S Indicates, unit ppt, P is the static pressure in atmospheric pressure, P The calculation formula is shown in formula (8).

[0037] (8) In the above formula D Indicates the depth of the current mesh from the sea surface, in meters.

[0038] The data in the Bellhop sound velocity profile file (.ssp) is a two-dimensional table with row labels D and column labels representing the distance (km) projected onto the sea surface from the sonar position along the cross section to the submarine. Calculate the sound velocity using the sound velocity profile calculation formula and write the sound velocity profile data to the required file format.

[0039] In order to save computing time and speed up detection, the huge seabed topography dataset is first reduced, then a multivariate interpolation model is constructed, and finally the seabed depth is calculated.

[0040] like Figure 6 As shown, let the longitude coordinate be x , the latitude coordinate is y ,The specific method of reducing the seabed terrain database is: ① Find the endpoints of multiple line segments on the route and mark them as a, b, c, d, and o respectively.

[0041] ② Find the minimum and maximum values ​​of the longitudes of the above endpoints x min 、 x max ; The maximum and minimum latitudes of the above endpoints y min 、 y max There are 4 values ​​in total.

[0042] ③The above four values ​​are combined into four coordinate points according to longitude and latitude: ( x min , y min )、( x min , y max )、( x max , y min )、( x max , y max ).

[0043] ④ Calculate the grid volume of the sea surface framed by the rectangle formed by the above four coordinate points, and collect the seabed topography data in these grid volumes.

[0044] A multivariate interpolation model of seafloor depth versus longitude and latitude coordinates is constructed based on this dataset. Substituting the coordinates of the points on the projected line segment of the profile equation onto the sea surface, the seafloor depth at these locations can be obtained. The specific steps are as follows: ①The profile passes through the anti-submarine ship sonar and submarine. Assume the coordinates of the anti-submarine ship ( x 1, y 1) and the submarine coordinates ( x 2, y 2) Substitute the cross-section equation to obtain the two-point straight line equation where the cross-section intersects the sea surface: (9) ②In the longitude interval [ x 1, x 2] Interpolate n longitude values ​​in equal parts and substitute these points into the line equation to get n latitude values.

[0045] ③The above n longitude values ​​and n latitude values ​​are sequentially combined into n longitude and latitude coordinate points.

[0046] ④ According to the Haversine formula, calculate the sea surface distance from n longitude and latitude coordinate points to the sound source in sequence.

[0047] ⑤ According to the multivariate interpolation model, calculate the seabed depth corresponding to n coordinate points in turn.

[0048] ⑥ The data in Bellhop's seabed topography file (.bty) has the first column as the distance from the n coordinate points to the sound source, and the second column as the seabed depth at the coordinate points. The data is written into the seabed topography file in the corresponding format.

[0049] The stealth level is based on signal margin. The active sonar source is located at the same position as the anti-submarine sonar. The signal margin from the source to all grids intersected by the active sonar and the anti-submarine sonar is calculated. The calculations reveal that in some grids traversed by the cross-section, submarine mountain ranges block and reflect all acoustic signals from the sonar source. For example, when a submarine is located on one side of a seamount and the anti-submarine sonar is located on the other side, the sonar receives almost no acoustic signal reflected from the submarine. In this case, the signal margin is negative infinity, and the sonar cannot detect the submarine. At this point, the submarine's stealth level is the highest, at level 5. The signal margins of the remaining grids are evenly divided into four equal parts from bottom to top, and the submarine's stealth levels are ranked as level 4, level 3, level 2, and level 1. From level 5 to level 1, stealth decreases, with level 5 making the submarine invisible and level 1 making it visible.

[0050] The submarine route collision detection method primarily involves constructing a multivariate interpolation model of seabed depth and longitude and latitude coordinates based on a reduced seabed topography dataset. Multiple longitude and latitude points are sampled along the submarine's maneuvering route, and the seabed depths at these points are calculated. If any of these longitude and latitude points have a depth less than the submarine's route, the submarine is deemed to have collided with the seabed along this route, requiring either a change in the submarine's diving depth or a reconfiguration of the route's position or shape.

[0051] The model building module builds an ordered multi-classification logistic regression model based on the ocean environment data collected from the "cross" route in Sea Area A and the concealment levels calculated and divided by Bellhop and sonar equations, providing a basic model for the subsequent training model module; The training model module is based on the ocean environment data collected from the "Z"-shaped route in Sea Area B and the trained ordered multi-classification Logistic regression model to predict the stealth level of submarines on the "Z"-shaped route.

[0052] The main responsibility of the index evaluation module is to calculate and divide the concealment level based on the Bellhop and sonar equations on the "Z"-shaped route in the B sea area, and to construct a confusion matrix to calculate the accuracy rate, macro-average precision rate, macro-average recall rate, and comprehensive evaluation indicators to quantitatively evaluate the prediction performance of the ordered multi-classification logistic regression model to ensure its reliability and effectiveness in actual operations.

[0053] Evaluation index module structure and function are as follows Figure 7 As shown. Among them, the prediction results and the output results of the Bellhop model form a confusion matrix. According to the confusion matrix, the accuracy, precision and recall of the prediction model can be calculated. The precision and recall rates are further used to calculate the comprehensive evaluation index of the prediction model ( F 1). Specifically, each grid on the route has a reference value. i , a predicted value j.when i = j When , the Logistic model predicts correctly, T represents the number of samples with correct predictions, and in this embodiment, the sample refers to the concealment level of each grid on the route; when i ≠ j When , the Logistic model makes a prediction error, and F represents the number of samples with incorrect predictions. The calculation formulas for accuracy, macro-average precision, macro-average recall, and comprehensive evaluation index (F1) are as follows: (10) (11) (12) (13) N is the number of samples to be predicted, F ij Indicates the number of samples that the Logistic model classifies into class j but the sample is actually in class i.

[0054] Obviously, those skilled in the art should understand that the various steps and modules of the method for rapid submarine stealth prediction based on an ordered multi-classification logistic regression model of the above-mentioned embodiment of the present invention can be implemented using a general-purpose computing device. They can be concentrated on a single computing device or distributed across a network composed of multiple computing devices. Alternatively, they can be implemented using program code executable by the computing device, so that they can be stored in a storage device and executed by the computing device. In some cases, the steps shown or described can be performed in a different order than herein, or they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. Thus, the embodiments of the present invention are not limited to any specific combination of hardware and software.

Claims

1. A method for rapid submarine stealth prediction based on an ordered multi-classification logistic regression model, characterized in that: The sonar routes of certain submarines and anti-submarine ships are verified to ensure that there are no collisions on the routes. The temperature, salinity, current, and seabed depth data of the sea areas surrounding the routes are collected and gridded. The submarine and anti-submarine ship sonars respectively sail along the routes. The temperature, salinity, current, and seabed depth data of the grid bodies through which the submarine and anti-submarine ship sonar routes pass are calculated as attributes. The temperature, salinity, current, and seabed depth data of all grid bodies are calculated by cutting the vertical sea surface cross-section of the submarine and anti-submarine ship sonar lines. The sound propagation loss is input into the Bellhop model, and the sound propagation loss is input into the active sonar equation to calculate the signal margin. The signal margin is evenly divided according to size to obtain the submarine's stealth level as a label. The attributes and labels are used together to train an ordered multi-classification logistic model. The trained ordered multi-classification logistic model is used to predict submarine stealth. During the prediction process, an alternative collision-free route is sought. Submarines and anti-submarine ship sonars respectively sail along the route, collecting temperature, salinity, current, and seabed depth data for the sea areas surrounding the route and forming a grid. The temperature, salinity, current, and seabed depth data for the grids passed by the submarine and anti-submarine ship sonars on the route are calculated and input into the ordered multi-classification logistic model to predict the submarine's stealth level.

2. The method for rapid submarine stealth prediction based on the ordered multi-classification logistic regression model according to claim 1 is characterized in that: The confusion matrix and its multi-classification comprehensive evaluation secondary indicators were used to evaluate the ordered multi-classification logistic model.

3. The method for rapid submarine stealth prediction based on the ordered multi-classification logistic regression model according to claim 1 is characterized in that: Determine collision-free routes through seabed terrain collision detection.

4. The method for rapid submarine stealth prediction based on an ordered multi-classification logistic regression model according to claim 1 is characterized in that: The route is obtained by setting the shapes of the anti-submarine ship sonar and submarine routes and the depths of their respective cruising, selecting the endpoints of the line segments on the anti-submarine ship sonar and submarine routes respectively, and constructing the straight line equations of the line segments.

5. The method for rapid submarine stealth prediction based on ordered multi-classification logistic regression model according to claim 1 is characterized in that: The collecting of temperature, salinity, current, and seabed depth data of the sea area passed by the surrounding route refers to collecting temperature, salinity, current, and seabed depth data of a square sea area; the square range framing method is to find all longitude, latitude, and depth coordinates of submarines on the route and all longitude, latitude, and depth coordinates of anti-submarine ship sonar routes, respectively find the minimum and maximum values ​​on each axis of longitude, latitude, and depth, and frame a square using the minimum and maximum values; discretize the collected data, mark the collected data with longitude, latitude, and depth, grid the data according to longitude, latitude, and depth, extract the data of all grids in the square, and use them to calculate the sound speed profile as Bellhop input and for training and prediction of the ordered multi-classification logistic model.

6. The method for rapid submarine stealth prediction based on ordered multi-classification logistic regression model according to claim 1 is characterized in that: The Bellhop model calculates sound propagation loss by providing the sound propagation loss of submarine and anti-submarine ship sonars in any sea area; the sound velocity profile and seabed topography of the sea area are calculated as input data for the Bellhop model; the sound velocity profile data is calculated as follows: when the submarine and anti-submarine ship sonars sail along the route to a certain location, the cross section perpendicular to the sea surface is extended from the line connecting the submarine and anti-submarine ship sonars; based on the gridded square sea area, all numbers in each grid body passed through when the cross section cuts through the square body are calculated and substituted into the sound velocity profile formula to obtain the sound velocity of each grid on the cross section; all data in each grid body include temperature, salinity, ocean current, and seabed depth data; The Bellhop model calculates and outputs the sound propagation loss at every point along the entire cross-section. This refers to each grid volume passed through when the square volume is cut along the cross-section. The sound propagation loss of the grid volume where the submarine is located is calculated based on the longitude, latitude and depth coordinates of the submarine. The signal margin is calculated based on the sound propagation loss, the active sonar source level, and the ambient noise. When the anti-submarine ship sonar and the submarine continue to navigate along the established route, the cross-section position of the anti-submarine ship sonar and the submarine changes, generating multiple signal margin values. These signal margin values ​​are sorted from small to large and assigned to multiple levels in ascending order according to the interval, representing the submarine's stealth level.

7. The method for rapid submarine stealth prediction based on ordered multi-classification logistic regression model according to claim 3 is characterized in that: The calculation method of the seabed topography data on the route is as follows: a multivariate interpolation model of the seabed topography of the entire square sea area is constructed as follows: the latitude and longitude coordinates of a certain position in the square sea area are input, and the seabed depth of the position is output; the sonar routes of submarines and anti-submarine ships are projected onto the sea surface to form one or more line segments of the sea surface, and multiple equal points are inserted into each line segment to obtain the latitude and longitude coordinates of all equal points, which are substituted into the multivariate interpolation model to calculate the seabed depth at each point to obtain the seabed topography data on the route.

8. The method for rapid submarine stealth prediction based on ordered multi-classification logistic regression model according to claim 6 is characterized in that: The calculation method of the seabed topography data on the cross section is as follows: a multivariate interpolation model of the seabed topography of the entire square sea area is constructed as follows: the longitude and latitude coordinates of a certain position in the square sea area are input, and the seabed depth of the position is output; when the submarine and anti-submarine ship sonar sail along the route to a certain position, a cross section perpendicular to the sea surface extended from the line connecting the submarine and anti-submarine ship sonar is tangent to the sea surface to form a line segment, multiple equal points are inserted in the line segment, and the longitude and latitude coordinates of all equal points are obtained. According to the Haversine formula, the distance from each point to the anti-submarine ship sonar is calculated; the seabed depth at each point is calculated by substituting it into the multivariate interpolation model to obtain the seabed topography data on the cross section.

9. A computer device, characterized in that: The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for rapid prediction of submarine stealth based on an ordered multi-classification logistic regression model as described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program for executing the submarine stealth fast prediction method based on the ordered multi-classification logistic regression model as described in any one of claims 1 to 8.

Images