A biomimetic three-dimensional underwater electric field positioning method and device based on a cascading space division algorithm

Abstract

The application provides a kind of bionic three-dimensional underwater electric field positioning method and device based on cascade space division algorithm, solve the technical problems of existing active electric field positioning technology when applied in three-dimensional, model mismatch, insufficient computing power and large spatial positioning error.It includes real-time acquisition of physical perception data through the modified sensor, pre-processing to obtain 3D space electric field signal;3D space electric field signal is input into KNN algorithm for dimension reduction, and the horizontal coordinate is predicted;Determine whether the horizontal coordinate meets the requirements, if so, use the random forest algorithm to predict the depth coordinate with the horizontal coordinate as the priori space constraint, coordinate fusion is carried out with the horizontal coordinate to output three-dimensional positioning result;Otherwise, input into KNN algorithm for dimension reduction, and the horizontal coordinate is predicted.The application can be widely applied in underwater positioning technology field.

Description

Technical Field

[0001] This application belongs to the field of underwater positioning technology, and more specifically, it relates to a biomimetic three-dimensional underwater electric field positioning method and device based on a cascaded spatial partitioning algorithm. Background Technology

[0002] Traditional underwater sensing technologies have significant shortcomings when conducting security monitoring and intrusion detection in critical underwater areas such as ports, waterways, and submarine pipelines. On the one hand, underwater environments are generally turbid with strong background noise, making sonar detection susceptible to multipath interference and echo reverberation, resulting in a significant decrease in detection accuracy. On the other hand, underwater visual systems are limited by light attenuation and scattering, making it difficult to clearly identify targets. Electric eels in nature possess unique biological detection capabilities. They can release weak electrical signals through their electroorgans, causing local electric field distortion when prey, obstacles, or other external objects enter their detection range. By sensing subtle changes in potential on their body surface, electric eels achieve precise active electrolocation in dark, highly turbid water environments.

[0003] Currently, active electric field localization technology, which mimics the perception of electric fish, has become a research hotspot. However, three major challenges remain when moving towards three-dimensional engineering applications: First, the physical model is distorted. The traditional ideal electric dipole model cannot accurately describe the nonlinear distortion caused by the physical size of the electrodes and boundary effects in the near-field environment. Second, there is a contradiction between computing power and accuracy. The computational complexity of inversion in three-dimensional space increases exponentially, and conventional global mapping models (such as single deep neural networks) are prone to getting stuck in local optima during training. Third, there is non-uniform attenuation. The underwater electric field signal attenuates non-linearly and drastically with increasing distance, resulting in a much lower positioning accuracy in the vertical dimension (depth) than in the horizontal dimension, ultimately leading to a large overall spatial positioning error. Summary of the Invention

[0004] The purpose of this application is to provide a biomimetic three-dimensional underwater electric field positioning method and device based on a cascaded spatial partitioning algorithm, so as to solve the technical problems of model mismatch, insufficient computing power and large spatial positioning error in the active electric field positioning technology in three-dimensional applications.

[0005] To achieve the above objectives, a first aspect of this application provides a biomimetic three-dimensional underwater electric field localization method based on a cascaded spatial partitioning algorithm, comprising the following steps: The corrected sensor acquires physical sensing data in real time, and preprocesses it to obtain a 3D spatial electric field signal. The 3D spatial electric field signal is input into the KNN algorithm for dimensionality reduction and prediction of the horizontal coordinates; Determine if the horizontal coordinates meet the requirements. If so, use the horizontal coordinates as a priori spatial constraint, predict the depth coordinates using the random forest algorithm, and fuse them with the horizontal coordinates to output the 3D positioning result. Otherwise, input the result into the KNN algorithm for further dimensionality reduction and predict the horizontal coordinates.

[0006] Preferably, the preprocessing process includes: extracting the target disturbance differential voltage of the sensing node from the physical sensing data, constructing a feature vector, and performing Z-score normalization to obtain the 3D spatial electric field signal.

[0007] Preferably, the process of inputting the KNN algorithm for dimensionality reduction and predicting the horizontal coordinates includes: establishing a global mapping model to reduce the three-dimensional space to a two-dimensional horizontal plane; determining the optimal number of nearest neighbor nodes based on the K-value adaptive optimization mechanism of cross-validation; and combining the inverse distance weighting mechanism to assign weights based on the inverse Euclidean distance between the nearest neighbor samples obtained from the optimal number of nearest neighbor nodes and the query samples, and outputting the predicted horizontal coordinates by weighted average.

[0008] Preferably, the process of determining the optimal number of nearest neighbor nodes includes: measuring sample similarity with Euclidean distance, performing a grid search within the candidate interval through 5-fold cross-validation, and determining the optimal number of nearest neighbor nodes using root mean square error as the evaluation index.

[0009] Preferably, the process of predicting depth coordinates includes: using the horizontal coordinates as a prior spatial constraint, activating the local regression model of the corresponding region using a spatial partitioning mechanism, and using a random forest algorithm to perform high-precision fitting of the nonlinear decay characteristics in the depth direction to obtain the depth coordinates.

[0010] Preferably, the process of performing high-precision fitting includes: randomly sampling with replacement from the training subset of the local regression model based on the Bagging strategy to generate a differentiated training subset and independently grow decision trees; during the growth of each decision tree, finding the optimal splitting feature and splitting threshold based on minimizing the mean square error to adapt to the severe nonlinear decay characteristics in the depth direction; and averaging the prediction results of all decision trees to obtain the depth coordinates.

[0011] Preferably, the formula for the horizontal coordinate is: ; In the formula, The horizontal coordinate is... To find the optimal number of nearest neighbor nodes, The weight of the k-th nearest neighbor sample. In the historical characteristic space, the first The true value of the nearest neighbor samples Axis coordinates In the historical characteristic space, the first The true value of the nearest neighbor samples Axis coordinates.

[0012] Preferably, the formula for depth coordinates is: ; In the formula, These are the depth coordinates ultimately predicted by the random forest. For the first The predicted depth of the target by each decision tree.

[0013] Preferably, a multi-level correction coefficient is introduced to correct the sensor, and a high-fidelity physical forward correction model is established to obtain the corrected sensor.

[0014] The second aspect of this application provides a biomimetic three-dimensional underwater electric field positioning device based on a cascaded spatial partitioning algorithm, including a sensor, which is a regular hexagonal electrode array arranged in multiple layers along the vertical direction for acquiring physical sensing data. The transmitting electrode is located at the center of the regular hexagonal electrode array, and the receiving electrodes are located at the six vertices of the regular hexagonal electrode array.

[0015] The beneficial effects of this application are as follows: This application provides a biomimetic three-dimensional underwater electric field positioning method and device based on a cascaded spatial partitioning algorithm. First, physical sensing data is acquired through a modified sensor and preprocessed to obtain a highly consistent 3D spatial electric field signal, avoiding model mismatch and positioning deviation caused by distortion of the original data, and making the positioning model fit the actual underwater electric field environment. Second, the three-dimensional space is reduced to a two-dimensional horizontal plane through the KNN algorithm, and the horizontal coordinates are predicted. This reduces computational complexity and computing power consumption while improving the real-time positioning performance and providing prior spatial constraints for calculating depth coordinates. Then, the horizontal coordinates are verified and filtered, and high-confidence horizontal coordinates are used as prior constraints to avoid low-precision horizontal coordinates misleading the prediction of depth coordinates, reduce depth positioning errors, and avoid blind searching to further reduce computing power consumption. In addition, the random forest algorithm is used to adapt to the nonlinear characteristics of the depth direction, improving the accuracy and robustness of depth coordinate prediction. Finally, the horizontal coordinates and depth coordinates are fused to output a high-precision three-dimensional positioning result. For horizontal coordinates that do not meet the requirements, they are returned to the KNN algorithm for re-optimization to ensure the robustness and reliability of the positioning results. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A schematic diagram of the overall process of a three-dimensional positioning method based on a cascaded spatial partitioning algorithm provided in an embodiment of this application; Figure 2 A schematic diagram of the logic flow of a cascaded space partitioning algorithm provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a regular hexagonal electrode array provided in an embodiment of this application; Figure 4 A schematic diagram of the structure of a three-dimensional volume detection module constructed from multiple regular hexagonal electrode arrays provided in an embodiment of this application; Figure 5 This is a schematic diagram of the optimal K value adaptive selection curve of the KNN algorithm provided in an embodiment of this application; Figure 6 A comparison diagram of the fitting residuals between a high-fidelity corrected physical model and experimental data provided in an embodiment of this application; Figure 7 A heat map showing the distribution of actual positioning error in three-dimensional space, provided for an embodiment of this application. Detailed Implementation

[0018] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.

[0019] This application provides a biomimetic three-dimensional underwater electric field localization method based on a cascaded spatial partitioning algorithm. By decomposing the three-dimensional problem through dimensionality reduction and decoupling, partitioning and mapping to adapt to regional characteristics, and processing complex three-dimensional nonlinear inverse problems in stages, it achieves high-precision and efficient three-dimensional localization of underwater targets.

[0020] Please see Figure 1 The first embodiment of this application provides a biomimetic three-dimensional underwater electric field localization method based on a cascaded spatial partitioning algorithm, comprising: S1: Physical sensing data is acquired in real time through the corrected sensor and preprocessed to obtain the 3D spatial electric field signal.

[0021] By introducing multi-level correction coefficients such as amplitude coupling, size offset, and environmental attenuation, the sensor is corrected, a high-fidelity physical forward correction model is established, the corrected sensor is obtained, and physical sensing data is acquired in real time.

[0022] Specifically, in actual underwater (or complex fluid) engineering environments, due to the actual physical geometric dimensions of the sensor electrodes (not ideal point sources), the non-uniform absorption of the detection medium, and the influence of system boundary effects, classical ideal physical field (such as electric or magnetic field) attenuation models often exhibit significant systematic biases when predicting spatial potential distributions, failing to accurately characterize the sensor's true field response. To bridge the gap between ideal theoretical models and actual engineering environments, this application innovatively introduces multi-level empirical correction parameters to compensate for the non-ideal characteristics of the sensor electrodes, the attenuation characteristics of the medium, and system boundary effects, thereby correcting the sensor and establishing a high-fidelity physical forward correction model. This yields the corrected sensor, whose spatial potential function... The mathematical expression is: ; In the formula, This is the equivalent amplitude coupling coefficient, used to compensate for the global amplitude difference between the overall system transmit power and the receiver gain. This is the offset of the physical size of the electrode, which serves as a spatial position correction term and is specifically used to characterize and compensate for the spatial field distortion caused by the fact that the electrode is not a perfect point source. , , These are the coordinate components of a Cartesian coordinate system, used to characterize the position of any probe point in three-dimensional space; It is an environmental nonlinear decay exponent that replaces the fixed decay power (such as inverse square or inverse cube) in traditional ideal fields, and is used to dynamically adapt to the true dissipation rate of signal energy in complex underwater environments.

[0023] In a specific embodiment, data was collected through extensive simulation experiments, and the parameters were optimized using the nonlinear least squares method. The final globally optimal parameter values ​​for the sensor in the target environment were obtained as follows: =4.7022、 =0.1178、 =1.034, making the goodness-of-fit R-value of the forward physical correction model 1.034. 2 The accuracy reached 0.987, resulting in a corrected sensor. This corrected sensor was then used to acquire real-time physical sensing data of the underwater environment.

[0024] After acquiring high-precision physical sensing data, preprocessing is performed to construct a 3D spatial electric field signal adapted to the machine learning model.

[0025] Please see Figure 2 Specifically, the differential voltage of the target perturbation at the sensing node is extracted from the physical sensing data to construct a feature vector, and Z-score normalization is performed to eliminate the difference in feature dimensions, thus obtaining the 3D spatial electric field signal. This process includes two main steps: differential feature extraction and dimensionless normalization. Because the raw voltage signal in the physical sensing data acquired by the sensor electrode array contains both the target's disturbance response and environmental background noise (such as water flow disturbance and baseline drift), this step first acquires the background baseline voltage of the system in the absence of a target to effectively remove static interference and highlight the target's position characteristics. And the measured voltage under the target condition Differential calculations are performed on all nodes in the sensor electrode array to extract the absolute differential values ​​(target disturbance differential voltages of the sensing nodes). This operation effectively removes static background interference, highlights the pure target location features, and constructs a multidimensional difference feature vector based on this.

[0026] In multi-channel sensor arrays, the signal intensities captured by each electrode may differ by several orders of magnitude due to varying distances from the excitation source. To ensure absolute comparability of signals from different physical locations and at different energy scales within the feature space, and to prevent large numerical features from dominating subsequent distance metrics (such as Euclidean distance), this step performs Z-score (standard deviation) standardization on the input feature vector. The conversion formula is as follows: ; In the formula, These are the standardized eigenvalues ​​after Z-score standardization. The absolute difference value, The mean of the training set, The standard deviation is given. This step eliminates dimensional differences, significantly accelerating the convergence speed of subsequent machine learning models and yielding 3D spatial electric field signals.

[0027] S2: Input the 3D spatial electric field signal into the KNN algorithm for dimensionality reduction and predict the horizontal coordinates.

[0028] The KNN algorithm is used to retrieve the target in the feature space of a 3D electric field signal to achieve dimensionality reduction and initial localization of the horizontal plane. The initial localization includes adaptive K-value optimization based on cross-validation and a distance-inverse weighting mechanism.

[0029] Specifically, this application establishes a global mapping model, first reducing the dimensionality of the complex 3D spatial positioning task and projecting it onto a 2D horizontal plane. To overcome the poor generalization ability caused by traditional empirical parameter setting, this step introduces a hyperparameter dynamic optimization mechanism. By retrieving the optimal number of signal patterns in the feature space of the 3D spatial electric field signal, the horizontal coordinate region where the target is located is quickly and accurately located. The specific process based on cross-validation K-value adaptive optimization and distance inverse weighting mechanism is as follows: K-value adaptive optimization based on cross-validation uses Euclidean distance to measure sample similarity in the standardized feature space, with the formula as follows: ; In the formula, The distance is Euclidean, representing the difference between the current query version and the version in the historical training set. The absolute similarity of each sample in the feature space. For the current sample to be predicted, at the th The differential voltage characteristic value on each electrode channel For the first volume of the history training set The sample at the th Differential voltage characteristic values ​​on each electrode channel.

[0030] A grid search is performed within the candidate interval K∈[1, 30] using 5-fold cross-validation. The root mean square error (RMSE) is used as the validation set performance evaluation metric to obtain the optimal number of nearest neighbor nodes that minimizes the localization error. The formula is: ; In the formula, To find the optimal number of nearest neighbor nodes, This represents the total number of samples in the validation set during the cross-validation phase. To verify the first set The true horizontal coordinate set of each sample When the set number of nearest neighbors is At that time, the model was tested on the validation set. The predicted estimated value of the horizontal coordinate of each sample.

[0031] Get the global optimal number of nearest neighbor nodes After the nearest neighbor samples, an inverse distance weighting mechanism is introduced. The closer the training sample is to the current query point, the greater its physical influence on location prediction. Weights of the nearest neighbor samples Defined as follows: ; In the formula, For the first The weights of the nearest neighbor samples, For the first The reciprocal of the Euclidean distance between the nearest neighbor sample and the query sample.

[0032] The final output is the predicted high-confidence horizontal coordinates. It is the following formula: ; In the formula, The horizontal coordinate is... To find the optimal number of nearest neighbor nodes, The weight of the k-th nearest neighbor sample. In the historical characteristic space, the first The true value of the nearest neighbor samples Axis coordinates In the historical characteristic space, the first The true value of the nearest neighbor samples Axis coordinates.

[0033] This stage successfully simplifies the three-dimensional nonlinear search into a fast two-dimensional plane retrieval through dynamic optimization and dimensionality reduction mapping, providing accurate prior space constraints for depth direction calculation.

[0034] S3: Determine if the horizontal coordinates meet the requirements. If so, use the horizontal coordinates as a priori spatial constraint, use the random forest algorithm to predict the depth coordinates, and perform coordinate fusion with the horizontal coordinates to output the 3D positioning result; otherwise, input the result into the KNN algorithm for further dimensionality reduction and predict the horizontal coordinates.

[0035] The system determines whether the horizontal coordinates meet the requirements. If so, it uses the horizontal coordinates as a prior spatial constraint and activates the local regression model of the corresponding region using the random forest algorithm to achieve a nonlinear robust estimation of the target depth coordinates. The requirement here is a preset positioning accuracy threshold, set to 5cm. If the horizontal coordinate positioning error is less than the preset accuracy threshold, the system is deemed ready to proceed to the next step. It is worth noting that this application does not limit the preset positioning accuracy threshold; it can be set according to actual conditions.

[0036] Specifically, using the horizontal coordinate as a guide, a spatial partitioning mechanism is employed to activate the corresponding local regression model. Addressing the drastic nonlinear decay characteristics of the target along the depth direction (Z-axis), this step utilizes a random forest regression algorithm comprising 100 independent decision trees for high-precision fitting. The specific steps include: The Random Forest algorithm is based on the Bagging strategy. It performs random sampling with replacement from the training subsets in the local space of the current local regression model to generate 100 differentiated training subsets and independently grows 100 regression decision trees.

[0037] During the growth of each decision tree, to address the nonlinear decay along the Z-axis, the algorithm randomly selects a subset of features at each node and searches for the optimal splitting features based on minimizing the mean squared error (MSE). and splitting threshold The formula is: ; In the formula, The objective function is to minimize the mean square error. To assign voltage feature sample points to the current local sub-region, These are the actual Z-axis coordinates of the sample points in three-dimensional physical space. and This represents the left and right child node regions after the split. and This represents the mean value of the sample depth Z within the corresponding region.

[0038] Let the first The prediction output of each decision tree for the target depth is: The final depth coordinates predicted by the random forest The formula is the average of the prediction results for all 100 trees: ; In the formula, These are the depth coordinates ultimately predicted by the random forest. For the first The predicted depth of the target by each decision tree.

[0039] By integrating and voting from multiple differentiated decision trees, this stage effectively smooths the prediction variance of a single model caused by environmental noise, and achieves robust estimation of Z-axis depth in complex environments.

[0040] The horizontal and depth coordinates of the decoupled simulation are fused to output the final high-precision three-dimensional spatial positioning result of the target.

[0041] This step is the final decision-making and execution layer of the entire spatial partitioning and cascaded localization algorithm. By re-aggregating the results of the previous spatial decoupling and inference, the absolute three-dimensional spatial position of the target in the complex environment is reconstructed, and the high-confidence horizontal coordinates output by the KNN algorithm are extracted. And depth coordinates that are accurately fitted by random forest (RF) based on local level guidance. The predicted values ​​of each dimension obtained through the above decoupling are vectorized and concatenated (spatial reconstruction and cascade fusion) to construct the final spatial location descriptor. The system outputs the final high-precision three-dimensional coordinate vector P of the target, which is mathematically expressed as: ; In the formula, It is a three-dimensional coordinate vector. For the goal The coordinate values ​​along the axis. For the goal The coordinate values ​​along the axis. For the goal The coordinate values ​​along the axis.

[0042] Please see Figure 3The second embodiment of this application provides a biomimetic three-dimensional underwater electric field positioning device based on a cascaded spatial partitioning algorithm, which includes a sensor. The sensor is a regular hexagonal electrode array arranged in multiple layers along the vertical direction for acquiring physical sensing data.

[0043] The transmitting electrode is located at the center of the regular hexagonal electrode array, and the receiving electrodes are located at the six vertices of the regular hexagonal electrode array.

[0044] Specifically, this application constructs a regular hexagonal electrode array based on a regular hexagonal topology to establish a physical hardware foundation for underwater target detection. Based on mechanical principles and spatial tessellation theory, a regular hexagonal topological electrode array is designed.

[0045] Please see Figure 4 In underwater environments, receiving electrodes are deployed at the vertices of a hexagon, while transmitting electrodes are placed at the center. A three-dimensional volumetric detection module is constructed using a multi-layered cascaded array of regular hexagonal electrodes in the vertical direction. The advantages of this structure lie in the extreme utilization of hardware resources and the uniformity of spatial sampling. Based on the mathematical relationship between spatial coverage and node density, the node density of the regular hexagonal grid reaches global optimum while ensuring the same seamless detection area S and the same effective sensing radius R. Theoretical calculations and engineering verification show that, when covering the same detection area, the total number of sensor nodes required by the regular hexagonal array is only about 77% of the number of nodes required by a traditional square array. Furthermore, a regular hexagon is an inscribed regular polygon that can achieve seamless planar tiling and is closest to a circle. In this structure, the maximum physical distance from any detection blind spot point to the nearest sensor is strictly minimized. This strictly minimizes the blind spot distance and reduces the standard deviation of the positioning error to a minimum.

[0046] Thanks to its uniform baseline length (the distance from the center to each vertex is equal to R) and multi-view feature redundancy, this topology can capture higher quality spatial electromagnetic features. System measurements and comparative analysis show that the average three-dimensional positioning accuracy of the hexagonal electrode array constructed in this application exhibits significant advantages: its average accuracy is improved by about 29% compared to the traditional square array and by about 57% compared to the equilateral triangular array, effectively breaking through the positioning accuracy bottleneck in complex underwater environments from a hardware physical perspective.

[0047] Example 1: Physical experimental verification scheme based on near-shore environment.

[0048] To verify the effectiveness and engineering feasibility of the dimensionality reduction decoupling and cascaded spatial partitioning (CSP) positioning strategy proposed in this application, this embodiment provides a physical experimental verification scheme and system platform based on a near-shore environment. Specific implementation steps and configuration details are as follows: (1) Hardware experimental platform construction and environmental parameter configuration.

[0049] The single-positioning module hardware experimental platform established in this embodiment mainly includes: a regular hexagonal electrode array, a mobile power supply, a signal generator, an oscilloscope, a voltage acquisition module, and a portable computer as a data processing terminal. The experiment was conducted in a near-shore marine environment, where the seawater conductivity was measured to be approximately 2.7 S / m. The positioning device adopts a regular hexagonal topology with a horizontal side length of 20 cm. In the vertical depth direction (Z-axis), the receiving electrodes are arranged in a multi-layer cascaded configuration with a fixed interlayer spacing of 4 cm.

[0050] (2) Signal excitation mechanism and detection target setting.

[0051] To effectively suppress electrochemical corrosion (polarization effect) caused by DC signals and considering the skin effect of electromagnetic waves in seawater, an AC signal was used as the excitation source for the central transmitting electrode of the system. The specific parameters of this excitation signal were set as follows: frequency 1 kHz, amplitude ±10 V. A metal sphere with a diameter of 2 cm was introduced into the constructed electric field environment as the target object to be measured.

[0052] (3) Model training and adaptive hyperparameter optimization.

[0053] Algorithm validation employs a strategy of simulation training followed by experimental verification. First, the proposed localization model is preprocessed using a simulation dataset. In the horizontal dimensionality reduction localization stage, the choice of the hyperparameter K plays a decisive role in the model's noise robustness and localization accuracy. If the K value is too small, it can easily lead to overfitting; if it is too large, it can cause excessive smoothing, thus masking local features. Therefore, this embodiment uses a cross-validation strategy to quantitatively analyze the evolution trend of the root mean square error (RMSE) under different K values. Finally, the optimal K value that best balances local accuracy and global stability is selected for subsequent predictions, such as... Figure 5 As shown, K=16.

[0054] (4) Real data calculation and positioning performance analysis.

[0055] To verify the effectiveness of the high-fidelity corrected physical model established in this application, the measured data of a typical electrode (electrode 1) were compared and analyzed with the model prediction results. The fitting residual comparison results are as follows: Figure 6 As shown. By Figure 6 It can be seen that the measured voltage value of electrode 1 changes with parameter a in a highly consistent manner with the fitting curve of the high-fidelity correction model of this application; the corresponding fitting residuals are randomly distributed near zero, with no systematic bias, and the maximum fitting residual is only 0.0464V, which fully demonstrates the model's ability to accurately represent the real electric field distribution underwater.

[0056] Based on this high-fidelity corrected physical model, the real voltage disturbance data collected in real time by the experimental platform is input into the pre-trained Cascaded Spatial Partition (CSP) model for calculation. The three-dimensional positioning error distribution between the actual and predicted target positions is as follows: Figure 7 As shown, the system outputs a root mean square error (RMSE) of approximately 3 cm in three-dimensional space, and a mean absolute error (MAE) strictly controlled within 2.5 cm. Further analysis shows that after compensating for the geometric influence of the target metal sphere's physical dimensions (2 cm in diameter), the system's actual absolute positioning error in three-dimensional space remains consistently within 5 cm. This result fully demonstrates the system's excellent precision positioning capability.

[0057] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0058] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A biomimetic three-dimensional underwater electric field localization method based on a cascaded spatial partitioning algorithm, characterized in that, Includes the following steps: The corrected sensor acquires physical sensing data in real time, and preprocesses it to obtain a 3D spatial electric field signal. The 3D spatial electric field signal is input into the KNN algorithm for dimensionality reduction and to predict the horizontal coordinates; Determine whether the horizontal coordinates meet the requirements. If so, use the horizontal coordinates as a priori spatial constraint, predict the depth coordinates using the random forest algorithm, and perform coordinate fusion with the horizontal coordinates to output the 3D positioning result. Otherwise, input the result into the KNN algorithm for further dimensionality reduction and predict the horizontal coordinates.

2. The biomimetic three-dimensional underwater electric field localization method based on the cascaded spatial partitioning algorithm as described in claim 1, characterized in that, The preprocessing process includes: extracting the target disturbance differential voltage of the sensing node from the physical sensing data, constructing a feature vector, and performing Z-score normalization to obtain the 3D spatial electric field signal.

3. The biomimetic three-dimensional underwater electric field localization method based on the cascaded spatial partitioning algorithm as described in claim 1, characterized in that, The process of inputting the KNN algorithm for dimensionality reduction and predicting horizontal coordinates includes: establishing a global mapping model to reduce the three-dimensional space to a two-dimensional horizontal plane; determining the optimal number of nearest neighbor nodes based on the K-value adaptive optimization mechanism of cross-validation; and combining the inverse distance weighting mechanism to assign weights based on the inverse Euclidean distance between the nearest neighbor samples obtained from the optimal number of nearest neighbor nodes and the query sample, and outputting the predicted horizontal coordinates by weighted averaging.

4. The biomimetic three-dimensional underwater electric field localization method based on the cascaded spatial partitioning algorithm as described in claim 3, characterized in that, The process of determining the optimal number of nearest neighbor nodes includes: measuring sample similarity using Euclidean distance, performing a grid search within the candidate interval using 5-fold cross-validation, and determining the optimal number of nearest neighbor nodes using root mean square error as the evaluation metric.

5. The biomimetic three-dimensional underwater electric field localization method based on the cascaded spatial partitioning algorithm as described in claim 1, characterized in that, The process of predicting the depth coordinates includes: using the horizontal coordinates as a prior spatial constraint, activating the local regression model of the corresponding region using a spatial partitioning mechanism, and using a random forest algorithm to perform high-precision fitting of the nonlinear decay characteristics in the depth direction to obtain the depth coordinates.

6. The biomimetic three-dimensional underwater electric field localization method based on the cascaded spatial partitioning algorithm as described in claim 5, characterized in that, The process of performing the high-precision fitting includes: randomly sampling with replacement from the training subset of the local regression model based on the Bagging strategy to generate a differentiated training subset and independently grow decision trees; during the growth of each decision tree, finding the optimal splitting feature and splitting threshold based on minimizing the mean square error to adapt to the drastic nonlinear decay characteristics in the depth direction; and averaging the prediction results of all decision trees to obtain the depth coordinates.

7. The biomimetic three-dimensional underwater electric field localization method based on the cascaded spatial partitioning algorithm as described in claim 3, characterized in that, The formula for the horizontal coordinate is: ； In the formula, The horizontal coordinate is... To find the optimal number of nearest neighbor nodes, The weight of the k-th nearest neighbor sample. In the historical characteristic space, the first The true value of the nearest neighbor samples Axis coordinates In the historical characteristic space, the first The true value of the nearest neighbor samples Axis coordinates.

8. The biomimetic three-dimensional underwater electric field localization method based on the cascaded spatial partitioning algorithm as described in claim 6, characterized in that, The formula for the depth coordinates is: ； In the formula, These are the depth coordinates ultimately predicted by the random forest. For the first The predicted depth of the target by each decision tree.

9. The biomimetic three-dimensional underwater electric field localization method based on the cascaded spatial partitioning algorithm as described in claim 1, characterized in that, The sensor is corrected by introducing multi-level correction coefficients, and a high-fidelity physical forward correction model is established to obtain the corrected sensor.

10. A biomimetic three-dimensional underwater electric field positioning device based on a cascaded spatial partitioning algorithm, applied to the biomimetic three-dimensional underwater electric field positioning method based on a cascaded spatial partitioning algorithm as described in any one of claims 1-9, comprising a sensor, characterized in that, The sensor is a multi-layered, cascaded array of regular hexagonal electrodes arranged along the vertical direction, used to acquire physical sensing data; The center of the regular hexagonal electrode array is provided with a transmitting electrode, and the six vertices of the regular hexagonal electrode array are provided with receiving electrodes.

Images