Anti-deviation control method for wireless charging system based on equipment posture recognition

Abstract

The application provides a wireless charging system anti-deviation control method based on equipment posture recognition. First, the electrical information of the primary side system under different roll angles and docking distances of the equipment is collected, specifically the voltage, current and phase difference between the two of the parallel capacitor. Based on a data-driven algorithm, the voltage, current and phase difference between the two of the parallel capacitor are taken as input information, and the roll angle and docking distance of the equipment are taken as output information. An equipment docking posture parameter identification model is established. The docking posture of the equipment is identified before the stable charging starts, so that the equipment can enter the charging process in the best posture. During the stable charging process, the posture of the equipment is continuously monitored, and problems such as unstable charging and efficiency reduction caused by equipment deviation due to factors such as sea current impact are dealt with in real time.

Description

Technical Field

[0001] This invention belongs to the field of underwater wireless charging technology, specifically relating to an anti-offset control method for a wireless charging system based on equipment attitude recognition. Background Technology

[0002] With the increasing demand for marine resource development and marine area maintenance, autonomous underwater vehicles (AUVs), as key equipment, are directly constrained by their endurance, which limits their operational range and efficiency. Currently, AUVs mainly rely on their onboard batteries for power, resulting in short range and limited operational time. To improve their continuous operational capabilities, underwater wireless charging technology has become a research hotspot. This technology avoids the defects of traditional plug-in charging, such as interface exposure and contact sparks, through contactless energy transfer. Combined with underwater charging base stations, it enables rapid in-situ energy replenishment for AUVs.

[0003] However, the energy transfer efficiency of underwater wireless charging systems is highly dependent on the docking status between the AUV and the charging base station. Specifically, the coupling coefficient and transmission efficiency of the magnetic coupling mechanism are affected by the docking distance and docking angle, manifested as follows: 1. When the AUV and the charging base station are perfectly aligned, the coupling coefficient and energy transfer efficiency decrease linearly with increasing docking distance, resulting in lower energy transfer efficiency. 2. Due to ocean current disturbances or AUV motion imbalance in the marine environment, the relative roll angle between the primary and secondary couplers may shift. The larger the shift angle, the linearly lower the coupling coefficient and energy transfer efficiency, resulting in lower energy transfer efficiency. Experiments show that at a standard docking distance of 15cm, the theoretical values ​​of the system coupling coefficient and energy transfer efficiency decrease linearly with increasing relative roll angle between the AUV and the charging base station; even in a perfectly aligned state, the theoretical values ​​of the system coupling coefficient and energy transfer efficiency also decrease linearly with increasing docking distance.

[0004] In existing technologies, mechanical limiting structures are typically used to ensure optimal docking. Specifically, this involves using a recovery cage, mechanical structures within the cage, and distance sensors to force the AUV to dock at a fixed angle and distance. For example, adding a physical structure to the AUV that matches the recovery cage ensures that the AUV enters the recovery cage at a predetermined angle. However, this requires custom design for AUVs with different diameters and fluid shapes, resulting in poor versatility and the potential for damaging the AUV's structure. Furthermore, during docking, the AUV cannot directly enter the recovery cage at the designed angle, making it prone to collisions that could damage the system's physical structure.

[0005] Furthermore, due to the limitations of underwater communication, charging base stations cannot directly obtain the real-time attitude parameters of the AUV (such as roll angle and docking distance). Traditional control schemes based on sensor feedback require the addition of an extra control module on the AUV side, which not only increases the size and weight of the AUV but may also affect its hydrodynamic performance. Summary of the Invention

[0006] Invention Concept: Based on the shortcomings of traditional underwater wireless charging system anti-drift technology, the applicant considers that even without a communication link between the charging base station and the AUV, the electrical information on the charging base station side of the wireless charging system will change when the AUV's attitude changes. Through the mapping relationship between the two, the electrical information on the charging base station side can be used to indirectly obtain the AUV's attitude information. Based on the foregoing considerations, this invention provides an anti-drift control method for wireless charging systems based on equipment attitude identification. It utilizes a data-driven algorithm to identify the relative roll angle and docking distance between the AUV and the charging base station through the electrical information of the primary system, ensuring that the AUV always enters the charging process in the optimal docking attitude and maintains this optimal docking attitude during charging. This achieves stable and efficient charging of the AUV, solving the problems of insufficient anti-drift capability and limitations in AUV attitude perception in existing underwater wireless charging system anti-drift technology.

[0007] To achieve the above objectives, the technical solution provided by this invention is:

[0008] An anti-offset control method for a wireless charging system based on equipment attitude recognition includes the following steps:

[0009] Step 1: Collect primary-side electrical parameters, including the voltage in the primary-side compensation circuit. Current and the phase difference between the two ;

[0010] Step 2: Input the primary side electrical parameters into the pre-built equipment docking attitude parameter identification model to obtain the current relative attitude parameters between the equipment and the charging base station. The relative attitude parameters include the relative roll angle between the equipment and the charging base station. and docking distance ;

[0011] Among them, the equipment docking attitude parameter identification model is stored in the primary side MCU and is used to identify the relative attitude parameters based on the collected primary side electrical parameters.

[0012] The acquired relative roll angle and docking distance are compared with their respective target values. If the difference reaches the preset accuracy threshold, return to step 1 and re-acquire the original side electrical parameters for judgment; otherwise, proceed to step 3.

[0013] Step 3: Based on the relative attitude parameters obtained in Step 2, the PID control module calculates and generates control commands for adjusting the equipment attitude, and sends the control commands to the charging base station servo mechanism.

[0014] Step 4: The charging base station servo mechanism adjusts the equipment attitude according to the control command until the target docking attitude is reached. At this time, the equipment and the charging base station maintain stable docking.

[0015] Furthermore, the process of pre-constructing the equipment docking attitude parameter identification model in step 2 includes the following steps:

[0016] Step S1: Establish the neural network model architecture, which includes an input layer, an output layer, and a decision tree layer; the input layer is used to extract features from the primary electrical parameters of the input.

[0017] The decision tree layer is used to map the extracted feature vectors to the relative pose parameter space according to a loss function dependent on seawater depth; the formula for the loss function dependent on seawater depth is:

[0018]

[0019]

[0020]

[0021]

[0022] Among them, the environmental weighted loss coefficient dependent on seawater depth Determined by seawater depth;

[0023]

[0024] In the formula, The total number of data samples. Represented as the first One data sample; To predict loss for roll angle, To predict the loss for docking distance, Environmentally weighted losses depending on seawater depth This is a gradient penalty term that depends on the input and output. and These are the predicted and actual values ​​of the roll angle, respectively. and These are the predicted and actual values ​​of the docking distance, respectively. , , , The weights of the roll angle prediction loss, docking distance prediction loss, seawater depth-based weighted loss, and input-output dependent gradient penalty term are set to initial values. , , , Furthermore, the weighting coefficients are dynamically adjusted according to changes in seawater depth.

[0025]

[0026] express right The weight of the prediction results express right The weight of the prediction results express right The weight of the prediction results express right The weight of the prediction results express right Pre

[0027] The weight of the impact of the measurement results express right The weight of the impact of the prediction results;

[0028] The output layer consists of two neurons, corresponding to the recognition values ​​of relative roll angle and docking distance, respectively;

[0029] Step S2: Based on the measurement data obtained in the air environment, which includes primary side electrical parameters and relative attitude parameters under different target attitudes, establish a reference mapping relationship;

[0030] Using the benchmark mapping relationship, adaptive filtering calibration is performed on similar data measured in marine environments at different depths. A marine environment sample dataset is then established based on the primary-side electrical parameters in the calibrated marine environment and their corresponding actual relative attitude parameters.

[0031] The marine environmental sample dataset was preprocessed and randomly divided into training and test sets;

[0032] Step S3: Using the preprocessed primary side electrical parameters as input and the corresponding preprocessed true relative attitude as the expected output, the mean square error is used as the loss function. The neural network model established in step S1 is trained using the genetic programming algorithm with the goal of minimizing the loss function until the loss function converges, thus obtaining the initial training model.

[0033] The error between the predicted results of the training set and the corresponding true relative pose is calculated and used as the first accuracy indicator.

[0034] Step S4: Use the initial trained model to predict the test set, calculate the error between the prediction result and the corresponding true relative pose, and use it as the second accuracy indicator;

[0035] When the difference between the second accuracy metric and the first accuracy metric exceeds a preset difference threshold, the following optimization process is executed:

[0036] Using the minimum mean square error of the prediction results on the test set as the optimization objective, the Bayesian optimization algorithm is used to tune the key hyperparameters of the model and determine the optimal combination of hyperparameters.

[0037] Step S5: Retrain the model using the determined optimal hyperparameter combination to obtain the optimized equipment docking attitude parameter identification model.

[0038] Furthermore, in step S2, the specific process of constructing the marine environmental sample dataset is as follows:

[0039] The equipment roll angle data during wireless charging is collected at a first interval within a preset first angle range to form a first dataset. The docking distance data between the equipment and the charging base station is collected at a second interval within a preset second distance range to form a second dataset. The two datasets are then randomly combined to generate multiple experimental combination schemes.

[0040] All test combinations were executed sequentially in air and marine environments at different depths;

[0041] Primary-side electrical parameters were collected synchronously in each experiment. The primary-side electrical parameters and corresponding relative attitude parameters obtained in the air environment were used as the control dataset, while the primary-side electrical parameters and corresponding relative attitude parameters obtained in the marine environment at different depths were used as the marine environment experimental dataset.

[0042] The primary-side electrical parameters from the experimental dataset are input into a pre-trained adaptive deep learning filter. The adaptive deep learning filter performs calculations and outputs the actual relative attitude parameters corresponding to the primary-side electrical parameters in the marine environment experimental dataset. The actual relative attitude parameters include the actual roll angle. and actual docking distance The actual roll angle is determined by the active roll angle adjustment component. and roll angle underwater environmental noise interference component With the corresponding first weight Second weight The actual docking distance is obtained through fusion calculation; it is determined by the active adjustment component of the docking distance. and the underwater environmental noise interference component at docking distance With the corresponding third weight and the fourth weight The fusion calculation yielded:

[0043]

[0044]

[0045] Among them, the active adjustment component refers to the attitude achieved by the charging base station servo mechanism after control according to control commands; the second weight and the fourth weight The adaptive deep learning filter is adaptively determined based on one or more factors of the current marine environment, such as depth, pressure, salinity, and temperature; and the adaptive deep learning filter is trained using a control dataset and an experimental dataset.

[0046] The primary-edge electrical parameters and corresponding actual relative attitude parameters obtained in marine environments at different depths are used as marine environment sample datasets.

[0047] Furthermore, the preset first angle range is [-90°, 90°], and the first interval is 3°; the preset second distance range is [0cm, 40cm], and the second interval is 2cm.

[0048] Furthermore, in step S4, the key hyperparameters include the maximum depth of a single decision tree, the number of iterations, the number of sample data used in each iteration, the proportion of features used in each tree, the learning rate, and the minimum loss required for the split point.

[0049] Furthermore, during the tuning of key hyperparameters of the model using the Bayesian optimization algorithm, the tuning direction of the key hyperparameters is guided by the model's prediction performance on the training and test sets:

[0050] If the prediction errors of both the training set and the test set are higher than the preset error threshold, the hyperparameter tuning direction is to increase the decision depth of a single tree, increase the number of iterations, or increase the number of sample data used in each iteration.

[0051] If the prediction error of the training set is low while the prediction error of the test set is high, the hyperparameter tuning direction should be: reduce the learning rate, reduce the proportion of features used per tree, or increase the minimum loss required for the split point.

[0052] Furthermore, the termination condition for the optimization process is: the prediction error of the optimized model on both the training set and the test set is less than 3%, and the error difference is less than 0.01%.

[0053] Furthermore, after the equipment and charging base station have successfully connected and entered a stable charging state, the following steps are performed:

[0054] The primary-side MCU collects primary-side electrical parameters at a preset sampling period and inputs the primary-side electrical parameters into the equipment docking attitude parameter identification model to monitor the docking attitude of the equipment and the charging base station in real time.

[0055] The real-time attitude data output by the equipment docking attitude parameter identification model is compared with the target attitude value to determine whether the current attitude deviation exceeds the first preset threshold. If the attitude deviation exceeds the first preset threshold, steps 3 and 4 are executed to adjust the equipment attitude in order to maintain stable docking.

[0056] Simultaneously, anomaly monitoring is performed on the collected primary-side electrical parameters: if the collected parameters at the current time... and If all values ​​are not zero, and the differences between electrical parameters at adjacent sampling times are within the corresponding safety thresholds, the wireless charging system is determined to be in normal condition, and the charging state is maintained.

[0057] If any difference in primary-side electrical parameters exceeds the corresponding safety threshold, it is determined to be an abnormal state, and the primary-side MCU immediately controls the wireless charging system to stop charging.

[0058] Furthermore, the safety threshold is set at 20%.

[0059] Furthermore, the preset sampling period is 1 second.

[0060] The advantages of this invention are:

[0061] The proposed method for anti-offset control of a wireless charging system based on equipment attitude identification obtains the voltage, current, and phase difference information of the primary-side wireless charging system by collecting historical data of different roll angles and docking distances of the equipment. A data-driven algorithm is then used to establish an equipment docking attitude parameter identification model. The docking attitude of the equipment is monitored in real time, and the attitude information is predicted using the parameter identification model. The attitude of the equipment is then controlled and adjusted by the servo mechanism of the charging base station, thereby ensuring that the equipment enters the charging process in the optimal docking attitude. This ensures the system's resistance to interference from factors such as ocean current impacts during the charging process, enabling stable and efficient charging.

[0062] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0063] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0064] Figure 1 This is a schematic diagram of the anti-offset control method of the wireless charging system of the present invention;

[0065] Figure 2 This is a structural diagram of a wireless charging system;

[0066] Figure 3 This is an equivalent circuit model diagram of a wireless charging system with an LCC-S topology under eddy current effect;

[0067] Figure 4 This is a schematic diagram showing the docking positions of the AUV and the charging base station at different roll angles;

[0068] Figure 5 This is a schematic diagram showing the docking positions of the AUV and the charging base station at the minimum and maximum docking distances.

[0069] Figure 6 This is a schematic diagram of the training process of the AUV docking attitude parameter identification model constructed in this invention;

[0070] Figure 7 This is a schematic diagram of the AUV and charging base station docking control process based on the anti-offset control method of the wireless charging system of the present invention.

[0071] Figure 8 This is a comparison chart of the AUV roll angle identification results and actual offset based on the AUV docking attitude parameter identification model constructed using this invention. Detailed Implementation

[0072] The embodiments of the present invention are described in detail below. These embodiments are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0073] This embodiment uses the docking control of an autonomous underwater vehicle (AUV) and an underwater charging base station as an example to describe the control scheme of the present invention.

[0074] Reference Figure 1 and Figure 2 In an underwater wireless charging system, the DC input voltage signal is converted into an AC signal by an inverter. After passing through the primary compensation circuit and the transmitting coil, the AC signal is transmitted to the secondary coil and the compensation circuit. Then, the AC signal is converted into DC power by a rectifier to charge the load battery.

[0075] Reference Figure 3 In the wireless power transmission system employing the LCC-S compensated topology of this invention, equivalent eddy current impedance is generated on the primary and secondary sides due to the influence of seawater eddy current effect. , This reduces the coupling coefficient of the wireless power transmission system, thus affecting the mutual inductance coefficient. (0< <1), the phase difference between the primary and secondary currents is no longer 90°, meaning a phase shift occurs in the mutual inductance branch. When the relative roll angle between the AUV and the charging base station... and docking distance When changes occur, and Changes occur, causing changes in the mutual inductance value of the wireless charging system, and affecting the voltage of the primary-side compensation circuit. Current and the phase difference between the two The relationship between them has changed.

[0076] Based on the above analysis, in order to accurately obtain the relative roll angle between the AUV and the charging base station and docking distance In this embodiment, an AUV docking parameter identification model was constructed, which identifies the voltage in the primary-side compensation circuit. (i.e., the voltage across the parallel capacitor), current and the phase difference between the two As an input variable, the relative roll angle between the AUV and the charging base station and docking distance The relative roll angle during charging is accurately identified using a parameter identification model. and docking distance The identification results are input into the primary-side MCU. The primary-side MCU outputs control commands based on the PID control module and sends the control commands to the base station servo mechanism. The base station servo mechanism adjusts the docking status of the AUV according to the control commands until the target docking posture is achieved.

[0077] The specific construction process of the AUV docking attitude parameter identification model is as follows:

[0078] Step S1: Establish the neural network model architecture, including an input layer, an output layer, and a decision tree layer. The input layer is used to extract features from the primary electrical parameters of the input. The decision tree layer is used to map the extracted feature vectors to the relative attitude parameter space according to the seawater depth-dependent loss function. The output layer is used to output the prediction result, and the output layer includes two neurons, corresponding to the relative roll angle respectively. and docking distance The identification value.

[0079] Specifically, the process of building the decision tree layer is as follows:

[0080] To extract the feature weights of the influence of each input variable on the output variable, this invention constructs a posture-environment coupled feature fusion influence model based on the seawater depth-dependent loss function shown in formula (1), where The total loss function consists of four parts, namely, roll angle prediction loss. Docking distance prediction loss Environmental weighted loss depending on seawater depth and the gradient penalty term for input-output dependencies , , , The mean squared error loss function is adopted, and the specific calculation formulas are shown in formulas (2), (3), and (4), respectively, where the coefficient of the environmental weighted loss dependent on seawater depth is... Determined by the depth of the seawater.

[0081] (1)

[0082] (2)

[0083] (3)

[0084] (4)

[0085] (5)

[0086] In the formula, where The total number of data samples. Represented as the first One data sample. and These are the predicted and actual values ​​of the roll angle, respectively. and These are the predicted and actual values ​​of the docking distance, respectively. , , , The predicted loss for roll angle is respectively Docking distance prediction loss Weighted loss based on seawater depth Gradient penalty term for input-output dependency The weights, set initial values , , , As shown in formula (6), , , and The weighting coefficient is dynamically adjusted according to the seawater depth, and the adjustment formula is shown in equation (6):

[0087] (6)

[0088] Gradient penalty term for input-output dependency The definition is shown in formula (7), which is mainly used to calculate the influence weights of input layer variables on output layer variables. In the process of determining model weights, the optimization objective is to minimize the loss function. The weight allocation that increases with the number of splits of the input layer variables and minimizes the loss function is selected, that is, the weights are determined. , , , , , The value, Indicates the primary voltage For roll angle The weight of the prediction results Represents the primary current For roll angle The weight of the prediction results Indicates phase difference For roll angle The weight of the prediction results express docking distance The weight of the prediction results express right The weight of the prediction results Indicates phase difference docking distance The weight of the impact of the prediction results.

[0089] (7)

[0090] Step S2: Establish sample data for subsequent training of the neural network model constructed in Step S1. The specific process is as follows:

[0091] First, refer to Figure 4 Within the roll angle range of [-90°, 90°], a roll angle data point is taken at 3° intervals, resulting in a first dataset containing 60 roll angle data points. (Refer to...) Figure 5 Within the docking distance range of [0cm, 40cm], a docking distance data point is taken at 2cm intervals to obtain a second dataset containing 20 docking distance data points. The discrete data points in the first and second datasets are then shorthanded and paired to obtain 200 experimental combination schemes. Using the roll angle and corresponding docking distance in the experimental scheme as the target attitude, all experimental combination schemes are executed in an air environment, and the primary side electrical parameters are measured under different target attitudes. This embodiment obtains a total of 200 sets of control experimental data. Based on the measured data of the primary side electrical parameters and the corresponding relative attitude data, a benchmark mapping relationship is established to form a control dataset.

[0092] Then, in marine environments at depths below 1m and depths of 1000m, 2000m, 3000m, 4000m, and 5000m, the above-mentioned experimental schemes were executed to measure the primary side electrical parameters under different target attitudes. Based on the measurement data of the primary side electrical parameters and the corresponding target relative attitude data, a mapping relationship was established to form a marine environment experimental dataset.

[0093] Due to factors such as ocean current impact and seawater pressure, the actual docking state achieved by the charging base station servo mechanism after issuing control commands to the AUV differs from the target attitude, specifically the actual roll angle. and docking distance There are active adjustment components and underwater environmental noise interference components. The active adjustment component is the roll angle achieved by the charging base station servo mechanism after control according to the control target. and docking distance The impact of underwater environmental noise includes the effects of factors such as water flow impact and biofouling. and To ensure the control accuracy of the base station servo mechanism, this embodiment utilizes the benchmark mapping relationship of the comparison dataset to perform adaptive filtering calibration on marine environmental test data measured at different depths. A marine environmental sample dataset is then established using the calibrated primary-edge electrical parameters and their corresponding actual relative attitude parameters in the marine environment. The specific calibration process is detailed below:

[0094] First, an adaptive deep learning filter was trained using a control dataset and a partial marine environmental experimental dataset to learn the mapping relationship between the primary side electrical participation docking attitude parameters in the control dataset.

[0095] The primary-side electrical parameters from the experimental dataset are then input into a pre-trained deep learning filter. The adaptive deep learning filter calculates and outputs the actual relative attitude parameters corresponding to the primary-side electrical parameters in the marine environment experimental dataset. The actual relative attitude parameters include the actual roll angle. and actual docking distance The actual roll angle is determined by the active roll angle adjustment component. and roll angle underwater environmental noise interference component With the corresponding first weight Second weight The actual docking distance is obtained through fusion calculation; it is determined by the active adjustment component of the docking distance. and the underwater environmental noise interference component at docking distance With the corresponding third weight and the fourth weight The result is obtained through fusion calculation, and the specific calculation formula is as follows:

[0096]

[0097]

[0098] In the formula, at different seawater depths, due to differences in seawater pressure, salinity, and temperature, the second weight... and the fourth weight It varies with seawater depth and is adaptively determined by an adaptive deep learning filter based on one or more factors of the current marine environment, including depth, pressure, salinity, and temperature.

[0099] A marine environment sample dataset was established using the calibrated primary-side electrical parameters of the marine environment and their corresponding actual relative attitude parameters.

[0100] Data preprocessing: The marine environmental sample dataset was normalized to eliminate the influence of different units on model training. Then, the data was randomly shuffled and divided into training and test sets in a 4:1 ratio.

[0101] Step S3: Train the neural network model using the preprocessed training set of 1200 sets. Specifically, the primary side electrical parameters are used as input layer variables, and the corresponding preprocessed true relative attitudes (actual roll angle and docking distance) are used as the expected output. Mean squared error is used as the loss function. The neural network model established in Step S1 is trained based on the Genetic Programming (GP) algorithm with the goal of minimizing the loss function, until the loss function converges, resulting in the initial trained model. After training, the error between the predicted results of the training set and the corresponding true relative attitudes is calculated as the first accuracy metric. This first accuracy metric is used to determine the subsequent model optimization process.

[0102] Step S4: Model Optimization: Use the initial trained model to predict the divided test set, and calculate the error between the predicted result and the corresponding true relative pose, as the second accuracy index. To ensure the neural network model achieves optimal prediction performance, the first accuracy index and the second accuracy index are compared. When the difference between the second accuracy index and the first accuracy index is greater than a preset difference threshold (set to 0.01% in this embodiment), the following optimization process is performed: Taking the minimum mean square error of the neural network model's prediction results on the test set as the optimization objective, the key hyperparameters of the model are tuned using the Bayesian optimization algorithm to determine the optimal hyperparameter combination, thereby improving the anti-interference and robustness of the neural network model. The key hyperparameters include the maximum depth of a single decision tree, the number of iterations, the number of sample data used in each iteration, the proportion of features used in each tree, the learning rate, and the minimum loss required for the split point.

[0103] In the process of tuning key hyperparameters, the direction of tuning is guided by the prediction performance of the neural network model on the training and test sets, specifically as follows:

[0104] If the prediction errors of both the training set and the test set are higher than the preset error threshold (e.g., higher than 3%, which can be determined according to the actual prediction accuracy requirements), then the hyperparameter tuning direction is to increase the decision depth of a single tree, increase the number of iterations, or increase the number of sample data used in each iteration.

[0105] If the prediction error on the training set is low while the prediction error on the test set is high, the hyperparameter tuning direction should be: reduce the learning rate, reduce the proportion of features used per tree, or increase the minimum loss required for split points. For example, the prediction error on the test set is greater than 1% compared to the prediction error on the training set.

[0106] The termination condition for the model optimization process is: the prediction error of the optimized model on both the training set and the test set is lower than the preset error threshold (set to 3% in this embodiment), and the difference between the prediction error of the test set and the prediction error of the training set meets the preset difference threshold (set to 0.01% in this embodiment).

[0107] Step S5: Retrain the model using the optimal hyperparameter combination determined in step S4 to obtain the optimized AUV docking attitude parameter identification model.

[0108] Reference Figure 7 The process of controlling the docking attitude of an AUV using the established AUV docking attitude parameter identification model includes the following steps:

[0109] Step 1: When the AUV needs to recharge during underwater missions, it first docks with the charging base station. During docking, the wireless charging system is activated at low power. The AC signal acquisition module located on the primary side collects the electrical parameters of the primary side, including the voltage in the primary side compensation circuit. Current and the phase difference between the two .

[0110] Step 2: The AC signal acquisition module inputs the primary-side electrical parameters into the AUV docking attitude parameter identification model stored in the primary-side MCU, and outputs the relative attitude parameters between the current AUV and the charging base station, including the relative roll angle between the AUV and the charging base station. and docking distance .

[0111] The acquired relative roll angle and docking distance are compared with their respective target values. If the difference reaches the preset accuracy threshold (in this embodiment, the accuracy threshold is set to 3%), then return to step 1 to proceed to the next sampling time and re-acquire the primary side electrical parameters for judgment; otherwise, proceed to step 3. In this embodiment, the sampling interval is set to 1 second.

[0112] Step 3: Based on the relative attitude parameters obtained in Step 2, the PID control module calculates and generates control commands to adjust the AUV's attitude, and sends these commands to the charging base station servo mechanism. The PID control module calculation process is as follows:

[0113] The actual values ​​of the relative roll angle and docking distance are compared with their respective target values ​​to obtain the roll angle error and distance error. The roll angle error and distance error are input into the corresponding PID control algorithm to calculate the control quantity for the base station servo mechanism. The control command includes the amount of movement in the roll and distance directions of the AUV.

[0114] Step 4: The charging base station servo mechanism adjusts the AUV's attitude according to the control commands until the target docking attitude is achieved. At this point, the AUV maintains a stable docking with the charging base station and begins high-power charging.

[0115] After the AUV and the charging base station complete the docking, the charging state is stable. The primary-side MCU collects data every 1 second. , , The information input parameter identification model monitors the docking attitude of the AUV and the charging base station in real time and compares it with the corresponding target value. If the deviation from the target value exceeds 3%, the control methods in steps 3 and 4 are used to control the AUV's attitude to maintain stable docking between the AUV and the charging base station. Simultaneously, the collected primary-side electrical parameters are evaluated in real time: if the differences between electrical parameters at adjacent sampling times are all within the corresponding safety threshold (set to 20% in this embodiment), and and If the value is not 0, it indicates that the wireless charging system is in a normal state and will continue to charge. If the difference of any primary-side electrical parameter exceeds the corresponding safety threshold, it indicates that the wireless charging system has malfunctioned. The primary-side MCU will immediately stop the charging process to prevent damage to other electronic devices in the system or to the AUV.

[0116] During the docking process between the AUV and the charging base station, the method of this invention is used to identify the relative roll angle between the AUV and the charging base station. The identification results are as follows: Figure 8As shown in the figure, the analysis yields an accuracy of 95.6%. This demonstrates that the model can accurately identify the relative roll angle between the AUV and the charging base station, thereby sending precise control commands to the servo mechanism at the charging base station. This calibrates the docking offset between the AUV and the charging base station, preventing system detuning due to self- and mutual inductance parameter shifts caused by docking offset, which could lead to decreased charging efficiency or even safety hazards.

[0117] In summary, the method of this invention can effectively solve the problem of precise docking control between AUV and charging base station during underwater wireless charging, as well as the problem of inaccurate docking caused by ocean current impact and AUV rolling motion during charging. It avoids the need to install physical structures on the charging base station and AUV side to achieve alignment, and can monitor the electrical status of the system in real time to prevent system abnormalities or malfunctions, thus effectively improving the safety of the system.

[0118] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A method for anti-offset control of a wireless charging system based on equipment attitude recognition, characterized in that, Includes the following steps: Step 1: Collect primary-side electrical parameters, including the voltage in the primary-side compensation circuit. Current and the phase difference between the two ; Step 2: Input the primary side electrical parameters into the pre-constructed equipment docking attitude parameter identification model to obtain the current relative attitude parameters between the equipment and the charging base station. The relative attitude parameters include the relative roll angle between the equipment and the charging base station. and docking distance The equipment docking attitude parameter identification model is stored in the primary side MCU and is used to identify the relative attitude parameters based on the collected primary side electrical parameters. The acquired relative roll angle and docking distance are compared with their respective target values. If the difference reaches the preset accuracy threshold, return to step 1 and re-acquire the original side electrical parameters for judgment; otherwise, proceed to step 3. The equipment docking attitude parameter identification model employs a neural network model, comprising an input layer, an output layer, and a decision tree layer. The input layer extracts features from the input primary-side electrical parameters. The decision tree layer maps the extracted feature vectors to the relative attitude parameter space based on a seawater depth-dependent loss function. The seawater depth-dependent loss function is calculated using the following formula: in, The environmental weighted loss coefficient, which depends on seawater depth, is determined by seawater depth. Indicates different depths of seawater; The total number of data samples. Represented as the first One data sample; To predict loss for roll angle, To predict the loss for docking distance, Environmentally weighted losses depending on seawater depth This is a gradient penalty term that depends on the input and output. and These are the predicted and actual values ​​of the roll angle, respectively. and These are the predicted and actual values ​​of the docking distance, respectively. , , , The weights of the roll angle prediction loss, docking distance prediction loss, seawater depth-based weighted loss, and input-output dependent gradient penalty term are set to initial values. , , , Furthermore, the weighting coefficients are dynamically adjusted according to changes in seawater depth. In the formula, express right The weight of the prediction results express right The weight of the prediction results express right The weight of the prediction results express right The weight of the prediction results express right The weight of the prediction results express right The weight of the impact of the prediction results; The output layer includes two neurons, which correspond to the identification values ​​of the relative roll angle and the docking distance, respectively. Step 3: Based on the relative attitude parameters obtained in Step 2, the PID control module calculates and generates control commands for adjusting the equipment attitude, and sends the control commands to the charging base station servo mechanism; Step 4: The charging base station servo mechanism controls and adjusts the equipment attitude according to the control command until the target docking attitude is reached. At this time, the equipment and the charging base station maintain stable docking.

2. The anti-offset control method for a wireless charging system according to claim 1, characterized in that, Step 2, which involves pre-constructing the equipment docking attitude parameter identification model, includes the following steps: Step S1: Establish the neural network model, which includes an input layer, an output layer, and a decision tree layer; Step S2: Based on the measurement data obtained in the air environment, which includes primary side electrical parameters and relative attitude parameters under different target attitudes, establish a reference mapping relationship; Using the aforementioned benchmark mapping relationship, adaptive filtering and calibration are performed on similar data measured in marine environments at different depths, and a marine environment sample dataset is established based on the primary-side electrical parameters in the calibrated marine environment and their corresponding actual relative attitude parameters. The marine environment sample dataset was preprocessed and randomly divided into training and test sets; Step S3: Using the preprocessed primary side electrical parameters as input and the corresponding preprocessed true relative attitude as the expected output, the mean square error is used as the loss function. The neural network established in step S1 is trained using a genetic programming algorithm with the goal of minimizing the loss function until the loss function converges, thus obtaining the initial training model. The error between the predicted results of the training set and the corresponding true relative pose is calculated and used as the first accuracy indicator. Step S4: Use the initial training model to predict the test set, and calculate the error between the prediction result and the corresponding true relative pose as a second accuracy index; When the difference between the second accuracy index and the first accuracy index exceeds a preset difference threshold, the following optimization process is executed: Using the minimum mean square error of the prediction results on the test set as the optimization objective, the key hyperparameters of the model are tuned using the Bayesian optimization algorithm to determine the optimal combination of hyperparameters; Step S5: Retrain the model using the determined optimal hyperparameter combination to obtain the optimized equipment docking attitude parameter identification model.

3. The anti-offset control method for a wireless charging system according to claim 2, characterized in that, In step S2, the specific process of constructing the marine environmental sample dataset is as follows: The equipment roll angle data during wireless charging is collected at a first interval within a preset first angle range to form a first dataset. The docking distance data between the equipment and the charging base station is collected at a second interval within a preset second distance range to form a second dataset. The two datasets are then randomly combined to generate multiple experimental combination schemes. All test combinations were executed sequentially in air and marine environments at different depths; Primary-side electrical parameters were collected synchronously in each experiment. The primary-side electrical parameters and corresponding relative attitude parameters obtained in the air environment were used as the control dataset, while the primary-side electrical parameters and corresponding relative attitude parameters obtained in the marine environment at different depths were used as the marine environment experimental dataset. The primary-side electrical parameters in the experimental dataset are input into a pre-trained adaptive deep learning filter, which performs calculations and outputs the actual relative attitude parameters corresponding to the primary-side electrical parameters in the marine environment experimental dataset. The actual relative attitude parameters include the actual roll angle. and actual docking distance The actual roll angle is determined by the active roll angle adjustment component. and roll angle underwater environmental noise interference component With the corresponding first weight Second weight The actual docking distance is obtained through fusion calculation; the docking distance is actively adjusted by the docking distance component. and the underwater environmental noise interference component at docking distance With the corresponding third weight and the fourth weight The fusion calculation yielded: Among them, the active adjustment component refers to the attitude achieved by the charging base station servo mechanism after control according to control commands; the second weight and the fourth weight The adaptive deep learning filter is adaptively determined based on one or more factors of the current marine environment, including depth, pressure, salinity, and temperature; and the adaptive deep learning filter is trained using the control dataset and the experimental dataset. The primary-edge electrical parameters and corresponding actual relative attitude parameters obtained in marine environments at different depths are used as marine environment sample datasets.

4. The anti-offset control method for a wireless charging system according to claim 3, characterized in that, The preset first angle range is The first interval is 3°; The preset second distance range is [0cm, 40cm], and the second interval is 2cm.

5. The anti-offset control method for a wireless charging system according to claim 2, characterized in that, In step S4, the key hyperparameters include the maximum depth of a single decision tree, the number of iterations, the number of sample data used in each iteration, the proportion of features used in each tree, the learning rate, and the minimum loss required for the split point.

6. The anti-offset control method for a wireless charging system according to claim 5, characterized in that, When tuning the key hyperparameters of a model using the Bayesian optimization algorithm, the tuning direction of these key hyperparameters is guided by the model's prediction performance on the training and test sets. If the prediction errors of both the training set and the test set are higher than the preset error threshold, the hyperparameter tuning direction is to increase the decision depth of a single tree, increase the number of iterations, or increase the number of sample data used in each iteration. If the prediction error of the training set is low while the prediction error of the test set is high, the hyperparameter tuning direction should be: reduce the learning rate, reduce the proportion of features used per tree, or increase the minimum loss required for the split point.

7. The anti-offset control method for a wireless charging system according to claim 2, characterized in that, The termination condition of the optimization process is: the prediction error of the optimized model on both the training set and the test set is less than 3%, and the error difference is less than 0.01%.

8. The anti-offset control method for a wireless charging system according to claim 1, characterized in that, After the equipment and charging base station have successfully connected and entered a stable charging state, perform the following steps: The primary-side MCU collects the primary-side electrical parameters at a preset sampling period and inputs the primary-side electrical parameters into the equipment docking attitude parameter identification model to monitor the docking attitude of the equipment and the charging base station in real time. The real-time attitude data output by the equipment docking attitude parameter identification model is compared with the target attitude value to determine whether the current attitude deviation exceeds the first preset threshold. If the attitude deviation exceeds the first preset threshold, steps 3 and 4 are executed to adjust the equipment attitude in order to maintain stable docking. Simultaneously, anomaly monitoring is performed on the collected primary-side electrical parameters: if the collected parameters at the current time... and If all values ​​are not zero, and the differences between electrical parameters at adjacent sampling times are within the corresponding safety thresholds, the wireless charging system is determined to be in normal condition, and the charging state is maintained. If any difference in primary-side electrical parameters exceeds the corresponding safety threshold, it is determined to be an abnormal state, and the primary-side MCU immediately controls the wireless charging system to stop charging.

9. The anti-offset control method for a wireless charging system according to claim 8, characterized in that, The safety threshold is set to 20%.

10. The anti-offset control method for a wireless charging system according to claim 8, characterized in that, The preset sampling period is 1 second.

Images