Sensor simulation design method and motion perception method based on lateral line density optimization

Abstract

The present invention discloses a sensor simulation design method and a motion perception method based on lateral line density optimization. In the sensor simulation design, the array resolution of an artificial lateral line sensor array is used as a key factor in associating flow field perception capability with fish lateral line characteristics. The array resolution of the artificial lateral line sensor array is optimized by the maximum adjacent pressure difference and the array sensitivity or the variance of the pressure difference between adjacent pressure sensors. This solves the problem that the experimental results of the flow field environment perceived mainly by the flow field perception model in the traditional technical solution are quite different from the expected results, and improves the accuracy of the artificial lateral line array in perceiving the environment. At the same time, the present invention optimizes the resolution of the sensor array, establishes a quantitative relationship between the array layout and the data capture performance, reduces the demand for pressure sensors on the sensor array under the condition of meeting the established requirements, and reduces the testing cost in the process of target motion perception.

Description

Technical Field

[0001] The present invention belongs to the technical field of lateral line perception of underwater vehicles, and in particular relates to a simulation method based on lateral line density optimization and a target motion perception method. Background Art

[0002] Traditional unmanned underwater vehicles (UUVs) use sonar and optical means as their main detection methods, and are mainly concerned with targets within a large range, with little consideration for the perception of their own environment. However, with the increase in UUV mission modes, the service environment has become complex, especially in complex flow fields such as shallow waters and near-wall operations. Due to the "multi-path effect" or turbid waters, acoustic and optical means are difficult to cope with the near-field target perception, positioning and navigation needs of UUVs under complex flow conditions. Therefore, it is necessary to explore new underwater near-field target perception, positioning and navigation technologies. As a near-field perception method, the lateral line perception of fish is not affected by the complex acoustic and optical environment, and can perceive the disturbance of the external environment in a timely manner, and realize behaviors such as obstacle avoidance, predation, avoidance of natural enemies, and cluster activities. The near-field target perception, positioning and navigation of unmanned underwater vehicles based on the bionic lateral line perception principle is also expected to develop into a new detection method with strong anti-interference ability in complex flow fields, providing situation input for the near-field target perception, obstacle avoidance, navigation control, etc. of UUVs in complex watersheds.

[0003] Inspired by the unique near-field perception ability of the lateral line of fish, researchers have tried to replicate this perception method in engineering applications and develop bionic artificial lateral line arrays for UUVs, in order to achieve UUVs with perception and maneuverability comparable to that of fish, which has become a hot topic in bionic engineering research. At present, there have been some explorations and experiments in the research of bionic fish lateral line perception at home and abroad, from sensor arrays to perception methods. The structural size, sensitivity, basic distribution form of lateral line arrays, and perception principles of fish lateral line perception units have been basically understood. Through experimental or simulation analysis, the test mechanism conclusions of some bionic fish lateral lines have been given from an empirical perspective, and a preliminary artificial test perception system has been designed based on MEMS sensors.

[0004] However, there are still many imperfections in the existing research results. There are currently two problems: 1. The correlation between flow field perception ability and fish lateral line characteristics has not been fully clarified, and there are many idealized constraints on the conditions for establishing flow field perception models, so that the experimental results of using flow field perception models as the main means to perceive the flow field environment and conduct related data analysis are quite different from the expected results; 2. The flow field environment perception method based on experimental verification does not have model bias problems, and the neural network method is the most effective among many methods, but most of the current perception networks are based on time and space separation for perception, and cannot simultaneously reflect the implicit relationship between sensors in time and space. The fish lateral line sensor actually perceives pressure at different body positions based on time sequence perception reasoning. Summary of the invention

[0005] The object of the present invention is to provide a sensor simulation design method and a motion perception method based on lateral line density optimization.

[0006] In a first aspect, the present invention provides a sensor simulation design method based on lateral line density optimization, which comprises the following steps:

[0007] Step 1: In the simulation model, multiple pressure sensors are arranged in sequence along the motion direction of the motion source as a sensor array; the working environment and working parameters of the simulation experiment are set, and the sensor array captures the environmental pressure changes caused by the movement of the motion source.

[0008] Step 2: Set multiple sensor arrays with different array resolutions. The simulation model simulates the motion posture of the motion source to imitate the working environment of the artificial lateral line in the real environment. Perform underwater motion simulation on different artificial lateral line sensor arrays to obtain the simulated pressure measured by each pressure sensor in the artificial lateral line sensor array.

[0009] Step 3: Calculate the maximum adjacent pressure difference MAXΔP corresponding to the simulated pressure measured by different artificial lateral line sensor arrays in step 2, as well as the array sensitivity AS or the variance Var of the pressure difference between adjacent pressure sensors. Among the artificial lateral line sensor arrays whose maximum adjacent pressure difference MAXΔP meets the accuracy requirements, select the array resolution corresponding to the artificial lateral line sensor array with the largest array sensitivity AS or variance Var as the optimal array resolution ρ best .

[0010] Step 4: Optimal array resolution ρ obtained from step 3 best The pressure sensors are arranged to obtain an artificial lateral line sensor.

[0011] Preferably, in the step 2, the array resolutions of the plurality of artificial side line sensor arrays are set to increase in sequence; the minimum array resolution ρ minThe corresponding artificial lateral line sensor array only includes two pressure sensors located at both ends of the lateral line simulation area. max In the corresponding artificial lateral line sensor array, the center distance between adjacent pressure sensors is equal to the width of the pressure sensors.

[0012] Preferably, in step three, the method for determining whether the maximum adjacent pressure difference MAXΔP meets the accuracy requirement is: after all sensors receive pressure data, if the maximum adjacent pressure difference MAXΔP is greater than or equal to a preset threshold, the maximum adjacent pressure difference MAXΔP meets the accuracy requirement.

[0013] Preferably, in the step 3, at any time after all pressure sensors in the sensor array receive pressure data, the array sensitivity AS or the variance Var of the pressure difference between adjacent pressure sensors are compared, and the array resolution corresponding to the artificial lateral line sensor array with the largest array sensitivity AS or variance Var is obtained as the optimal array resolution ρ best .

[0014] Preferably, in the step 2, the speed of the motion source is 0.3m / s to 1m / s; the distance d between the motion source and the artificial side line sensor array satisfies: 0.3L≤d≤λ / 2π; wherein λ is the wavelength of underwater sound waves; and L is the length of the artificial side line sensor array.

[0015] Preferably, in step 2, the pressure sensors in the same sensor array are arranged in sequence at equal intervals.

[0016] Preferably, in step 3, the expression of the maximum adjacent pressure difference MAXΔP is:

[0017] MAXΔP=max{|(P n+1 -P n )|}

[0018] Among them, P n is the pressure of the nth pressure sensor; n=1,...,N-1; N is the number of pressure sensors.

[0019] The expression of array sensitivity AS is:

[0020]

[0021] Wherein, ΔS is the distance between adjacent pressure sensors.

[0022] The variance Var of the pressure difference between adjacent pressure sensors can measure the discreteness of the pressure difference between adjacent pressure sensors. The expression of the variance Var is:

[0023]

[0024] in, is the average value of the pressure sensor pressure.

[0025] Preferably, in step 2, 3-5 groups of different difference increasing array resolutions are set.

[0026] In a second aspect, the present invention provides a method for sensing target motion, which comprises the following steps:

[0027] Step 1: Construct a data set including the pressure data measured by the artificial lateral line sensor and the position of the motion source. The artificial lateral line sensor is obtained by the above-mentioned design method.

[0028] Step 2: Build a neural network as a space-time perception model, use the data set to train the space-time perception model, use the pressure data as the input of the space-time perception model, and use the motion source position at the last time step as the output of the space-time perception model.

[0029] Step 3: Use the trained spatiotemporal perception model to perceive the position of the motion source based on the pressure data measured by the pressure sensor.

[0030] Preferably, in step one, if the pressure data collected by the pressure sensor is NaN, the median value of the left and right pressure sensor data is used to replace NaN; if the pressure sensor whose pressure data is NaN is at the end point of the sensor array, the data of the previous time step is used to supplement it; and data with a pressure less than 0.01pa is replaced by 0.

[0031] Preferably, in the step 2, the data set is preprocessed before training; the preprocessing method is: the data collected by the pressure sensor is subjected to FFT calculation, and the spatial features of the pressure data of the sensor array at the same time step are extracted from the frequency domain for normalization.

[0032] Preferably, in step 2, the spatiotemporal perception model adopts a ConvLSTM network.

[0033] The present invention has the following beneficial effects:

[0034] 1. The present invention uses the array resolution of the artificial lateral line sensor array as the key factor associating the flow field perception ability with the lateral line characteristics of the fish, and optimizes the arrangement density of the pressure sensors in the artificial lateral line sensor array by the maximum adjacent pressure difference and the array sensitivity or the variance of the pressure difference between adjacent pressure sensors. This solves the problem that the experimental results of the traditional technical solution that uses the flow field perception model as the main means to perceive the flow field environment are quite different from the expected results, and improves the accuracy of the artificial lateral line array in sensing the environment.

[0035] 2. The present invention optimizes the resolution of the sensor array and establishes a quantitative relationship between the array layout and the data capture performance, so that the sensor array reduces the demand for pressure sensors while meeting the established requirements, thereby reducing the testing cost during the target motion perception process.

[0036] 3. The spatiotemporal perception model constructed by the present invention can simultaneously reflect the implicit relationship between pressure sensors in time and space. Compared with the traditional perception network that perceives based on the separation of time and space, the present invention is more practical and has higher practicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is the overall flow chart of the present invention.

[0038] Figure 2 This is a working schematic diagram of the artificial lateral line sensor array in the present invention.

[0039] Figure 3 The process of building a model for the present invention.

[0040] Figure 4 Flow chart for designing the sensor array resolution in the present invention.

[0041] Figure 5 1 is a graph showing the change in array sensitivity of the artificial lateral line sensor array under different working conditions in Example 1 of the present invention; wherein, (a) is a graph showing the change in array sensitivity at a distance of 10 cm and a speed of 0.1 m / s; (b) is a graph showing the change in array sensitivity at a distance of 10 cm and a speed of 0.5 m / s; (c) is a graph showing the change in array sensitivity at a distance of 10 cm and a speed of 1 m / s; (d) is a graph showing the change in array sensitivity at a distance of 100 cm and a speed of 0.1 m / s; (e) is a graph showing the change in array sensitivity at a distance of 100 cm and a speed of 0.5 m / s; and (f) is a graph showing the change in array sensitivity at a distance of 100 cm and a speed of 1 m / s.

[0042] Figure 61 is a graph showing the variance change of adjacent pressure differences of the artificial lateral line sensor array under different working conditions in Example 1 of the present invention; wherein, (a) is a graph showing the variance change of adjacent pressure differences at a distance of 10 cm and a speed of 0.1 m / s; (b) is a graph showing the variance change of adjacent pressure differences at a distance of 10 cm and a speed of 0.5 m / s; (c) is a graph showing the variance change of adjacent pressure differences at a distance of 10 cm and a speed of 1 m / s; (d) is a graph showing the variance change of adjacent pressure differences at a distance of 100 cm and a speed of 0.1 m / s; (e) is a graph showing the variance change of adjacent pressure differences at a distance of 100 cm and a speed of 0.5 m / s; and (f) is a graph showing the variance change of adjacent pressure differences at a distance of 100 cm and a speed of 1 m / s.

[0043] Figure 7 1 is a graph showing the maximum adjacent pressure difference change curve of the artificial lateral line sensor array under different working conditions in Example 1 of the present invention; wherein, (a) is a graph showing the maximum adjacent pressure difference change curve when the distance is 10 cm and the speed is 0.1 m / s; (b) is a graph showing the maximum adjacent pressure difference change curve when the distance is 10 cm and the speed is 0.5 m / s; (c) is a graph showing the maximum adjacent pressure difference change curve when the distance is 10 cm and the speed is 1 m / s; (d) is a graph showing the maximum adjacent pressure difference change curve when the distance is 100 cm and the speed is 0.1 m / s; (e) is a graph showing the maximum adjacent pressure difference change curve when the distance is 100 cm and the speed is 0.5 m / s; (f) is a graph showing the maximum adjacent pressure difference change curve when the distance is 100 cm and the speed is 1 m / s.

[0044] Figure 8 This is a flow chart of data preprocessing in the present invention.

[0045] Fig. 9 This is the ConvLSTM network structure diagram in the present invention.

[0046] Fig.10 This is a loss function convergence diagram of the ConvLSTM network structure training process in Example 1 of the present invention.

[0047] Fig.11 : This is the relationship between the distance from the target source to the array center and the prediction error in Example 1 of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described below in conjunction with the accompanying drawings.

[0049] like Figure 1 As shown, a sensor simulation design method and motion perception method based on lateral line density optimization include the following steps:

[0050] Step 1: Figure 2 and Figure 3As shown in the figure, the experimental environment is built in the COMSOL platform. The experiment is planned to be carried out in a still water area. The water area is centered and is a rectangular area with a width of 200 cm and a height of 150 cm. The center coordinates are set to (0,0). The vertex at the lower left corner of the water area is the pressure constraint point, the four walls are open boundaries, and the fluid domain is laminar flow. A lateral line simulation area is set at the center of the lower wall at 25 cm. A row of domain point pressure probes are set in the lateral line simulation area to simulate the pressure sensor as an artificial lateral line sensor array. The artificial lateral line sensor array captures the water pressure changes caused by the movement of the motion source to simulate the working environment of the artificial lateral line sensor in the real environment. The sensor array length is 100 cm. A 5 cm diameter aluminum ball is used as the moving source. The aluminum ball is within the range of λ / 2π (λ is the wavelength of underwater sound waves) above the sensor. The geometric position and speed of the moving source are adjustable. The curved motion of the target source is simplified into a small segment of linear motion.

[0051] The target source in the near field is analyzed and studied for the flow field changes caused by the target source in simple linear motion. When the moving source starts to move in the flow field, the flow field in its influence range will change. First, the streamlines will be deformed. As the dipole moves, the streamlines will no longer remain perfect circles, but will be deformed, showing changes relative to the direction of the dipole's movement. At the same time, the velocity distribution will also be affected, and the velocity distribution around the dipole will also change. The velocity increases on the side close to the dipole's moving direction, while the velocity decreases on the side away from the moving direction. Most importantly, due to the change in velocity, according to the Bernoulli equation, the pressure distribution will also change. The pressure in the area in front of the moving source will decrease, and the pressure will increase in the rear. Experiments were conducted on the "fine" and "ultra-fine" mesh units of the model to verify the mesh independence. When the initial position of the moving source is the coordinate (-50,0), and the moving speeds are 0.1m / s, 0.5m / s, and 1m / s, respectively, the pressure values ​​recorded by the sensor array are shown in Table 1. It can be seen from Table 1 that the difference between the mean pressure values ​​measured in the two groups of experiments is very small, and the experiment has nothing to do with the degree of mesh refinement.

[0052] Table 1. Verification and comparison results of grid independence

[0053]

[0054]

[0055] Step 2: Figure 4As shown in the figure, the sensor array with a length of 100 cm is set with 10, 20, 50, and 100 sensor arrays for simulation experiments, that is, the array resolution ρ of the sensor array is 0.1, 0.2, 0.5, and 1. The pressure sensors are arranged from (-50, 0) to (50, 0). Ignoring the initial physical acceleration process of the target source movement, the distance d between the target source and the sensor array is 10 cm and 100 cm, respectively, and the target source moves uniformly in the x-axis direction at a speed of 0.1 m / s, 0.5 m / s, and 1 m / s, respectively, for simulation.

[0056] The array sensitivity AS (Array Sensitivity), the variance Var of the pressure difference between adjacent pressure sensors, and the maximum adjacent pressure difference MAXΔP in the simulation process are defined to design the array resolution ρ of the sensor array. The expression of array sensitivity AS is:

[0057]

[0058] Where N is the number of pressure sensors; ΔS is the spacing between adjacent pressure sensors; P n is the pressure of the nth pressure sensor; n=1,...,N.

[0059] The variance Var of the pressure difference between adjacent pressure sensors can measure the discreteness of the pressure difference between adjacent pressure sensors. The expression of the variance Var is:

[0060]

[0061] in, is the average value of the pressure sensor pressure.

[0062] The expression of the maximum adjacent pressure difference MAXΔP is:

[0063] MAXΔP=max{|(P n+1 -P n )|}

[0064] The larger the array sensitivity AS and the variance Var of the adjacent pressure difference, the more diverse the data is and the better the timing characteristics are; the larger the maximum adjacent pressure difference MAXΔP is, the more the changes in adjacent pressure sensors can meet the accuracy requirements of the pressure sensor.

[0065] Step 3: The simulation experiment collects the time series curves of the maximum adjacent pressure difference MAXΔP, array sensitivity AS and the variance Var of the adjacent pressure difference within 8 seconds. Figure 5 , Figure 6 , Figure 7 As shown. Figure 5 and Figure 6As can be seen from (a) in the figure, when the speed is low and the distance is short, the denser the array (the larger the array resolution), the smaller the fluctuation of the array sensitivity AS and the variance Var of the adjacent pressure difference in time, and the collected data can easily obtain the characteristics of the time series. Figure 5 As can be seen from (d) in the figure, when the speed remains unchanged and the distance increases, but when all sensors capture the pressure changes caused by the motion source (the sensors are in a stable state), the array sensitivity AS of the high-resolution array is significantly higher than that of the low-resolution array; when the near-field sensing range of the artificial lateral line is exceeded, the longer the distance and the faster the speed, the smaller the impact of different resolutions. It can be seen that the higher the array resolution, the higher the array sensitivity AS, the more diverse the data, the smoother the time series curve, and the better the time series characteristics are preserved.

[0066] In addition, it is also necessary to consider whether the changes in adjacent pressure sensors caused by the array resolution can meet the requirements of sensor accuracy to avoid the situation where adjacent pressure sensors have the same values. Figure 7 It can be seen from the data that when the distance is too far and the speed is too small, the maximum adjacent pressure difference MAXΔP is small and does not meet the requirements. When the distance and speed are moderate and the number of pressure sensors is 10, 20, and 50, when the sensor is in a stable state, the maximum adjacent pressure difference MAXΔP ≥ 0.1pa exists, indicating that the appropriate resolution can meet the accuracy requirements of the pressure sensor; the array sensitivity AS and the variance Var of the adjacent pressure difference corresponding to the sensor array with 100 pressure sensors are relatively perfect, but the maximum adjacent pressure difference MAXΔP is too small, that is, there is actually a large amount of invalid data, the cost and power consumption of data acquisition are high, and the computing resource consumption of the subsequent prediction algorithm is also large. Relatively speaking, the array sensitivity AS and the variance Var of the adjacent pressure difference of the array with 50 pressure sensors in various working conditions are very close to the curve of the array with a resolution of 100, and its maximum adjacent pressure difference MAXΔP also mostly meets the accuracy requirements of the sensor array, and it has high resolution accuracy without excessively wasting resources.

[0067] Therefore, in this embodiment, the performance of the sensor array is the best when the resolution is 50. In order to balance accuracy and cost, the number of artificial side line array sensors is set to 50.

[0068] Step 4: Figure 8As shown in the figure, the target source is set within a rectangular range of 150cm×120cm in the center of the water area. An initial starting point is selected every 10cm, and the target source moves uniformly in different directions at different speeds. The speed is 0.1m / s~1m / s, the step length is 0.1m / s, and the direction is an integer multiple of 45 degrees, for a total of eight directions. The domain point probe of the circular target is set to collect its position information. The coordinates are the coordinates of the geometric center of the circle. A total of 1s of target position data is collected, the time step is 0.1s, and there are 10 time steps. The pressure data set obtained by the sensor has a total of 14400 sets of data. Some data will cause the pressure data to be collected as NaN (Not a Number) because the target source passes through the grid where the sensor probe is located. The median value of the left and right sensor data is used to replace NaN; if the sensor is an endpoint, the data of the previous time step is used to supplement it. The data with a pressure less than 0.01pa is replaced by 0. The pressure value here is only a relative value. Because the data is collected under an absolutely ideal environment, there is no need to denoise the data. The data collected by the sensor is calculated by FFT (Fast Fourier Transform), and the spatial characteristics of the pressure data of the sensor array at the same time step are extracted from the frequency domain, and normalized to ensure reasonable data distribution. After normalization, the acquired data set is divided into a training set and a validation set in a ratio of 7:3.

[0069] Step 5: Fig. 9 As shown in the figure, the ConvLSTM network structure is used as the spatiotemporal perception model, and the data is analyzed based on the time dimension and space dimension. The spatiotemporal perception model is trained using the training set obtained in step 3. During the training process, the batch size is set to 16, the 50 sets of pressure data collected are the input data, and the position of the last time step is the output label. The expression of the loss function MSEloss is:

[0070]

[0071] Where N is the training batch; x pi and pi are the x and y coordinates of the predicted target source respectively; ti and ti are the real coordinates of the target source respectively.

[0072] Step 6: Input the test set into the trained ConvLSTM network structure and use the average error Evaluate the ConvLSTM network structure and average error The expression is:

[0073]

[0074] from Fig.10It can be seen that during the training and testing process, the loss function of the ConvLSTM network converged to 39 in a very short epoch, that is, the average error Because the actual position of the geometric circle of the target source is larger than the coordinates of its center, the predicted result with an average error of only 7 cm is very close to the actual position. Fig.11 It can be seen that the prediction accuracy decreases approximately linearly with the increase of the distance d between the target source and the sensor array. This result is more consistent with the true theory of the lateral line of fish and verifies the effectiveness of the near-field target perception algorithm of this experiment.

Claims

1. A sensor simulation design method based on lateral line density optimization, characterized in that: The following steps are involved: Step 1: In the simulation model, multiple sensors are arranged in sequence along the motion direction of the motion source as a sensor array; the working environment and working parameters are set, and the sensor array captures the environmental pressure changes caused by the movement of the motion source; Step 2: setting multiple sensor arrays with different array resolutions; performing underwater motion simulation on different artificial lateral line sensor arrays respectively to obtain simulated pressures measured by each pressure sensor in the artificial lateral line sensor array; Step 3: Calculate the maximum adjacent pressure difference MAXΔP corresponding to the simulated pressure measured by different artificial lateral line sensor arrays in step 2, as well as the array sensitivity AS or the variance Var of the pressure difference between adjacent sensors; among the artificial lateral line sensor arrays whose maximum adjacent pressure difference MAXΔP meets the accuracy requirements, select the array resolution corresponding to the artificial lateral line sensor array with the largest array sensitivity AS or variance Var as the optimal array resolution ρ best ; The expression of the maximum adjacent pressure difference MAXΔP is: MAXΔP=max{|(P n+1 -P n )|} Among them, P n is the pressure of the nth sensor; n=1,...,N-1; N is the number of sensors; The expression of array sensitivity AS is: Where, ΔS is the distance between adjacent sensors; The variance Var of the pressure difference between adjacent sensors can measure the discreteness of the pressure difference between adjacent sensors. The expression of the variance Var is: in, is the average value of the sensor pressure; Step 4: Optimal array resolution ρ obtained from step 3 best The pressure sensors are arranged to obtain an artificial lateral line sensor.

2. The sensor simulation design method based on lateral line density optimization according to claim 1 is characterized in that: In the step 2, the array resolutions of the plurality of artificial side line sensor arrays are set to increase in sequence; the minimum array resolution ρ min The corresponding artificial lateral line sensor array only includes two pressure sensors located at both ends of the lateral line simulation area; the maximum array resolution ρ max In the corresponding artificial lateral line sensor array, the center distance between adjacent pressure sensors is equal to the width of the pressure sensors.

3. The sensor simulation design method based on lateral line density optimization according to claim 1 is characterized in that: In the step three, the method for judging whether the maximum adjacent pressure difference MAXΔP meets the accuracy requirement is: after all sensors receive pressure data, if the maximum adjacent pressure difference MAXΔP is greater than or equal to a preset threshold, the maximum adjacent pressure difference MAXΔP meets the accuracy requirement.

4. The sensor simulation design method based on lateral line density optimization according to claim 1 is characterized in that: In the step 3, at any time after all pressure sensors in the sensor array receive pressure data, the array sensitivity AS or the variance Var of the pressure difference between adjacent sensors is compared, and the array resolution corresponding to the artificial lateral line sensor array with the largest array sensitivity AS or variance Var is obtained as the optimal array resolution ρ best .

5. The sensor simulation design method based on lateral line density optimization according to claim 1 is characterized in that: In the step 2, the speed of the motion source is 0.3m / s to 1m / s; The distance d between the motion source and the artificial side line sensor array satisfies: 0.3L≤d≤λ / 2π; where λ is the wavelength of underwater sound waves; and L is the length of the artificial side line sensor array.

6. The sensor simulation design method based on lateral line density optimization according to claim 1 is characterized in that: In the step 2, the pressure sensors in the same sensor array are arranged in sequence at equal intervals.

7. The sensor simulation design method based on lateral line density optimization according to claim 1 is characterized in that: The number of sensor arrays with different array resolutions set in step 2 is 3 to 5.

8. A method for sensing target motion, characterized in that: The following steps are involved: Step 1, constructing a data set including pressure data measured by an artificial lateral line sensor and the position of the motion source; The artificial lateral line sensor is obtained by the sensor simulation design method based on lateral line density optimization described in claim 1; Step 2: Build a neural network as a space-time perception model, use the data set to train the space-time perception model, use the pressure data as the input of the space-time perception model, and use the motion source position as the output of the space-time perception model; Step 3: Use the trained spatiotemporal perception model to perceive the location of the underwater motion source based on the pressure data measured by the sensor.

9. A method for sensing target motion according to claim 8, characterized in that: In the step 1, if the pressure data collected by the sensor is NaN, the median value of the left and right sensor data is used to replace NaN; if the sensor whose pressure data is NaN is at the end point of the sensor array, the data of the previous time step is used to supplement it; the data with a pressure less than 0.01pa is replaced by 0.

Images