A hybrid target posture control method and system

Abstract

The present application provides a method and system for controlling the posture of a hybrid target. The method includes: predicting the scene change trend with a pre-trained neural network model based on terrain data and temperature and humidity data to obtain a scene change trend vector, and adjusting the parameters of the target hybrid model based on the scene change trend vector; calculating the air pressure target value through the target hybrid model after parameter adjustment; and controlling the gas injection or release rate of the bidirectional pressure regulating device of the hybrid target based on the air pressure target value, until the fluctuation amplitude of the actual air pressure value after pressure adjustment is less than the preset fluctuation amplitude threshold, and the stable time is greater than the preset time threshold. The bidirectional pressure regulating device includes an air pump and a pressure relief valve. After the control is completed, the posture of the hybrid target is controlled according to the actual air pressure value. The technical solution provided by the present application controls the posture of the hybrid target through the actual air pressure value of the hybrid target, which can achieve precise stability in a complex environment.

Description

Technical Field

[0001] The present application relates to the technical field of target posture control, and in particular to a method and system for controlling the posture of a hybrid target. Background Art

[0002] In military training, weapons testing, or dynamic simulation scenarios, the coordinated control of mobile physical target vehicles and hybrid targets must meet the requirements of adaptive adjustment in complex environments, especially for inflatable targets. Specifically, the targets need to adapt to the realism requirements of different training scenarios while ensuring stability and durability after air pressure adjustment.

[0003] A typical existing solution for this requirement is an environmental prediction control model based on a single neural network. This solution collects environmental data such as terrain, temperature and humidity, uses pre-trained recurrent neural networks or convolutional neural networks to predict dynamic environmental trends, and directly maps them to posture data.

[0004] The main drawback of existing solutions is their insufficient generalization capabilities in dynamic environments. When processing multimodal coupled data, a single neural network model struggles to simultaneously balance the rule-based reasoning capabilities of fuzzy logic with the global optimization characteristics of nonlinear optimization. This results in excessive sensitivity of prediction results to control parameters. Furthermore, existing solutions are prone to overshoot or response lag in the face of complex environmental disturbances, making them unable to meet the requirements for precise attitude control. Summary of the Invention

[0005] The present application provides a hybrid target posture control method and system to solve the problem in the prior art that overshoot or response lag easily occurs when facing complex environmental disturbances in a dynamic environment, resulting in poor posture control accuracy.

[0006] In a first aspect, the present application provides a method for controlling the posture of a hybrid target, comprising:

[0007] Obtaining the moving speed data of the movable physical target vehicle, the hardness level parameter data and the actual value of the air pressure of the mixed target, and the terrain data and temperature and humidity data of the environment where the movable physical target vehicle is located;

[0008] Based on the terrain data and temperature and humidity data, a pre-trained neural network model is used to predict scene change trends to obtain a scene change trend vector, and parameters of a target hybrid model are adjusted based on the scene change trend vector. The target hybrid model is a hybrid model that integrates a fuzzy logic module and a nonlinear optimization module.

[0009] Calculating an air pressure target value through a target hybrid model with adjusted parameters based on the moving speed data, the hardness level parameter data, the terrain data, and the temperature and humidity data;

[0010] According to the air pressure target value, the gas injection or release rate of the bidirectional pressure regulating device of the mixed target is controlled until the fluctuation amplitude of the actual air pressure value after pressure adjustment is less than the preset fluctuation amplitude threshold, and the stable time is greater than the preset time threshold. After the control is completed, the posture of the mixed target is controlled according to the actual air pressure value. The bidirectional pressure regulating device includes an air pump and a pressure relief valve.

[0011] Optionally, the calculating the air pressure target value based on the moving speed data, the hardness level parameter data, the terrain data, and the temperature and humidity data through a target hybrid model with adjusted parameters includes:

[0012] Aligning sampling timestamps of the movement speed data and the hardness level parameter data to generate a multi-feature matrix;

[0013] performing a second-order polynomial basis expansion on the multi-feature matrix to generate an expanded feature matrix, performing local response normalization on an environmental feature tensor to generate a normalized tensor, and calculating a differential feature vector based on a time series of historical air pressure values, wherein the environmental feature tensor is generated based on the terrain data and the temperature and humidity data;

[0014] Performing weighted concatenation on the extended feature matrix, the normalized tensor, and the differential feature vector to form a mixed feature vector;

[0015] The target mixed model after the parameter adjustment is used to perform nonlinear mapping on the mixed characteristic vector to obtain the air pressure target value.

[0016] Optionally, performing a second-order polynomial basis expansion on the multi-feature matrix to generate an expanded feature matrix, performing local response normalization on the environmental feature tensor to generate a normalized tensor, and calculating a differential feature vector based on a time series of historical air pressure values ​​includes:

[0017] For each feature dimension in the multi-feature matrix, calculating the original eigenvalue on the feature dimension, the product term of each two feature dimensions, and the square term of a single feature dimension, and splicing the original eigenvalue, the product term, and the square term by column to generate an extended feature matrix;

[0018] For each channel of the environmental feature tensor, the square sum of the activation values ​​of the adjacent m channels is selected as the normalization base, the normalization base is weighted by a preset channel attenuation factor, the activation value of the current channel is divided by the weighted normalization base to obtain a normalized value, and the normalized value is nonlinearly compressed using a hyperbolic tangent function to generate a normalized tensor;

[0019] With a fixed window length, the air pressure history value sequences in multiple windows are cut out from the time series of the air pressure history values, and the absolute difference sequence between the air pressure history values ​​at adjacent time points in each window is calculated. The absolute difference sequence is subjected to a sliding average filter to obtain a filtered sequence corresponding to each window; the filtered sequences corresponding to all windows are accumulated and summed to generate a differential feature vector.

[0020] Optionally, the weighted concatenation of the extended feature matrix, the normalized tensor, and the differential feature vector to form a mixed feature vector includes:

[0021] Allocating a weight coefficient related to feature dimension, a weight coefficient related to the number of channels, and a weight coefficient of a fixed value to the extended feature matrix, the normalized tensor, and the differential feature vector, respectively;

[0022] According to the weight coefficient related to the feature dimension, the weight coefficient related to the number of channels, and the weight coefficient of the fixed value, Hadamard product operations are performed on the extended feature matrix, the normalized tensor, and the differential eigenvector to obtain a weighted extended feature matrix, a weighted normalized tensor, and a weighted differential eigenvector;

[0023] Performing format conversion on the weighted extended feature matrix and the weighted normalized tensor respectively to obtain an extended feature vector and a normalized vector;

[0024] Elements at the same position in the extended feature vector, the normalized vector and the weighted differential feature vector are spliced ​​in a preset splicing order to form a mixed feature vector.

[0025] Optionally, performing nonlinear mapping on the mixed feature vector using the target mixed model after parameter adjustment to obtain the air pressure target value includes:

[0026] Extracting a first sub-vector related to the terrain and target hardness, and a second sub-vector related to the environment and movement speed from the mixed feature vector;

[0027] Inputting the first sub-vector and the second sub-vector into a fuzzy logic module and a nonlinear optimization module, respectively; the fuzzy logic module calculates a fuzzy inference result according to a predefined fuzzy logic rule base and in combination with the adjusted weight parameter of the membership function, and converts the fuzzy inference result into a first air pressure compensation value; and the nonlinear optimization module calculates an air pressure reference value based on the adjusted penalty factor and iteration step size of the nonlinear optimization algorithm;

[0028] The first air pressure compensation value and the air pressure reference value are superimposed to obtain an air pressure target value.

[0029] Optionally, predicting a scene change trend using a pre-trained neural network model based on the terrain data and the temperature and humidity data to obtain a scene change trend vector, and adjusting parameters of a target hybrid model based on the scene change trend vector includes:

[0030] Discretizing the terrain data into a terrain elevation matrix according to preset grid units, mapping the temperature and humidity data into a temperature and humidity distribution tensor according to spatial coordinates, and performing cross-modal splicing on the terrain elevation matrix and the temperature and humidity distribution tensor to generate an environmental feature tensor;

[0031] The environmental feature tensor is input into a pre-trained neural network model, combined with the historical environmental feature tensor within the time sliding window, to predict the scene change trend and output a scene change trend vector representing the dynamic evolution of the terrain and the intensity of climate fluctuations;

[0032] According to the terrain-related component in the scene change trend vector, the weight parameter of the membership function in the fuzzy logic rule base corresponding to the fuzzy logic module is adjusted, and according to the climate-related component in the scene change trend vector, the penalty factor and iteration step size of the nonlinear optimization algorithm adopted by the nonlinear optimization module are adjusted.

[0033] Optionally, adjusting a weight parameter of a membership function in a fuzzy logic rule base corresponding to a fuzzy logic module according to a terrain-related component in the scene change trend vector, and adjusting a penalty factor and an iteration step size of a nonlinear optimization algorithm adopted by a nonlinear optimization module according to a climate-related component in the scene change trend vector, includes:

[0034] Analyze the terrain dynamic component in the scene change trend vector to extract the slope change rate and elevation fluctuation intensity. Analyze the climate fluctuation component in the scene change trend vector to extract the temperature and humidity mutation direction and wind speed correlation intensity.

[0035] adjusting, according to the slope change rate and the elevation fluctuation intensity, weight parameters of a membership function of a first rule and a weight parameter of a membership function of a second rule in a fuzzy logic rule base, respectively, wherein the first rule is a rule between slope and hardness compensation, and the second rule is a rule between terrain flatness and air pressure stability;

[0036] According to the temperature and humidity mutation direction and the wind speed correlation strength, the penalty factor and the iteration step size of the nonlinear optimization algorithm are adjusted respectively.

[0037] In a second aspect, the present application provides a hybrid target posture control system, comprising:

[0038] An acquisition module is used to obtain the moving speed data of the movable physical target vehicle, the hardness grade parameter data and the actual value of the air pressure of the mixed target, and the terrain data and temperature and humidity data of the environment where the movable physical target vehicle is located;

[0039] an adjustment module, configured to predict scene change trends using a pre-trained neural network model based on the terrain data and the temperature and humidity data, obtain a scene change trend vector, and adjust parameters of a target hybrid model based on the scene change trend vector, wherein the target hybrid model is a hybrid model that integrates a fuzzy logic module and a nonlinear optimization module;

[0040] a calculation module, configured to calculate an air pressure target value through a target hybrid model with adjusted parameters based on the moving speed data, the hardness level parameter data, the terrain data, and the temperature and humidity data;

[0041] The regulating module is used to control the gas injection or release rate of the bidirectional pressure regulating device of the mixed target according to the air pressure target value until the fluctuation amplitude of the actual air pressure value after pressure regulation is less than the preset fluctuation amplitude threshold and the stable time is greater than the preset time threshold. After the control is completed, the posture of the mixed target is controlled according to the actual air pressure value. The bidirectional pressure regulating device includes an air pump and a pressure relief valve.

[0042] In a third aspect, the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a posture control method for a hybrid target as described in any one of the first aspects.

[0043] In a fourth aspect, the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a hybrid target posture control method as described in any one of the first aspects.

[0044] In the present application, a posture control method for a hybrid target is provided, which includes: obtaining moving speed data of a movable physical target vehicle, hardness level parameter data and actual air pressure value of the hybrid target, and terrain data and temperature and humidity data of the environment in which the movable physical target vehicle is located; predicting the scene change trend through a pre-trained neural network model based on the terrain data and temperature and humidity data to obtain a scene change trend vector, and adjusting the parameters of a target hybrid model based on the scene change trend vector, wherein the target hybrid model is a hybrid model that integrates a fuzzy logic module and a nonlinear optimization module; calculating the air pressure target value through the parameter-adjusted target hybrid model based on the moving speed data, the hardness level parameter data, the terrain data and temperature and humidity data; and controlling the gas injection or release rate of the bidirectional pressure regulating device of the hybrid target based on the air pressure target value until the fluctuation amplitude of the actual air pressure value after pressure adjustment is less than a preset fluctuation amplitude threshold, and the stable time is greater than a preset time threshold, wherein the bidirectional pressure regulating device includes an air pump and a pressure relief valve.

[0045] This application achieves precise control of the hardness of mixed targets in complex scenarios through dynamic environmental perception and hybrid intelligent control mechanisms, thereby achieving precise control and stable maintenance of the posture of mixed targets. Based on the neural network, the scene change trend is predicted and the parameters of the fuzzy logic-nonlinear optimization hybrid model are dynamically adjusted to enhance the robustness of the system to sudden changes in terrain and fluctuations in temperature and humidity, and reduce the sensitivity of control parameters. The target pressure value is calculated through the target hybrid model by integrating multi-source data such as moving speed, target hardness, terrain, temperature and humidity, and taking into account fuzzy rule reasoning and nonlinear global optimization in the calculation process. The dynamic gas injection / release rate control based on the bidirectional pressure regulating device is combined with the preset fluctuation threshold and stable time constraint to ensure that the actual pressure value converges quickly and is stable in the long term, avoiding the overshoot or lag problems of the traditional proportional integral differential controller.

[0046] Furthermore, accurate prediction of target air pressure values ​​is achieved through multi-source time series data fusion and feature engineering optimization. The sampling timestamps of moving speed and hardness grade data are aligned to generate a multi-feature matrix, and the nonlinear characterization capability is enhanced through second-order polynomial basis expansion. The noise interference of terrain / temperature and humidity tensors is suppressed based on local response normalization. The windowed absolute difference sequence is extracted from the historical air pressure values, and the differential eigenvector is generated through sliding average filtering and accumulation. The expanded feature matrix, normalized tensor and differential eigenvector are weightedly spliced ​​into a hybrid eigenvector, and the target air pressure value is generated by a parameter-adaptive fuzzy-nonlinear optimization hybrid model. Through multi-dimensional feature fusion and dynamic feature optimization, the air pressure control accuracy and stability in complex scenarios are improved. The second-order polynomial expansion enhances the model's ability to model the nonlinear coupling relationship between movement speed and hardness parameters; local response normalization and hyperbolic tangent compression reduce the interference of local mutations in the environmental feature tensor and improve generalization; windowed differential feature extraction accurately captures the dynamic change trend of historical air pressure values, combined with sliding average filtering to suppress high-frequency noise; hybrid feature vectors fuse multi-scale information, allowing fuzzy logic and nonlinear optimization modules to work together to achieve rapid convergence of air pressure target values ​​and long-term stable control.

[0047] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0049] Figure 1 A flowchart of a hybrid target posture control method provided in an embodiment of the present application;

[0050] Figure 2 A schematic structural diagram of a hybrid target posture control system provided in an embodiment of the present application;

[0051] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0053] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 11, 12, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0054] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0055] In order to solve the problem that the existing technology is prone to overshoot or response lag when facing complex environmental disturbances, and is difficult to meet the requirements of precise attitude control, the present application provides a posture control method for a hybrid target, which adopts the following ideas: by real-time collection of multi-source data such as target vehicle movement speed, target hardness, environmental terrain, temperature and humidity, a time-space aligned multimodal feature matrix and environmental feature tensor are constructed; a neural network is used to predict the dynamic evolution of terrain and climate mutation trends, and a scene change trend vector is generated to dynamically adjust the parameters of the hybrid model; a hybrid feature vector is generated based on feature expansion, normalization and differential processing, and the air pressure target value is collaboratively mapped through fuzzy logic and nonlinear optimization; finally, the bidirectional pressure regulation device is driven to respond quickly, so that the actual air pressure value is stable within a preset fluctuation threshold, thereby realizing high-precision adaptive control of the target air pressure in a complex dynamic environment, and thus realizing precise attitude control of the hybrid target in this environment.

[0056] Figure 1 A flow chart of a hybrid target posture control method provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0057] S11. Obtain the moving speed data of the movable physical target vehicle, the hardness level parameter data and the actual value of the air pressure of the mixed target, and the terrain data and temperature and humidity data of the environment where the movable physical target vehicle is located.

[0058] Among them, the movement speed data is the real-time movement rate of the movable physical target vehicle under specific terrain. The hardness level parameter data is a quantitative indicator of the air pressure level corresponding to the current hardness of the hybrid target. The actual air pressure value is the real-time measurement value of the target's internal pressure sensor. The terrain data is the three-dimensional terrain feature vector such as the slope and surface roughness of the target vehicle's location. The temperature and humidity data is the synchronous monitoring value of the ambient temperature and relative humidity. The hybrid target is a hybrid target located on the movable physical target vehicle. It can be an inflatable target, so pressure control is required to achieve the hybrid target's posture control.

[0059] In this embodiment, sensors first capture the real-time speed data of the mobile target vehicle. Simultaneously, they read the hardness parameter data and current air pressure value recorded by the hybrid target's built-in pressure sensor. A lidar scanner collects three-dimensional terrain data of the mobile target vehicle's environment, and a temperature and humidity sensor continuously monitors ambient temperature and relative humidity. All this data is transmitted via a bus to the vehicle's central processing unit, forming a multidimensional dataset containing speed, hardness parameters, air pressure, terrain characteristics, and ambient temperature and humidity.

[0060] S12. Based on the terrain data and temperature and humidity data, the scene change trend is predicted by a pre-trained neural network model to obtain a scene change trend vector. The parameters of the target hybrid model are adjusted according to the scene change trend vector. The target hybrid model is a hybrid model that integrates a fuzzy logic module and a nonlinear optimization module.

[0061] The scene change trend prediction can predict the evolution direction of environmental parameters within the next 10 seconds based on the spatiotemporal correlation between terrain and temperature and humidity data. The target hybrid model is a parallel coupled architecture consisting of a fuzzy logic module and a nonlinear optimization module. The fuzzy logic module is a rule-based inference system based on fuzzy set theory, designed to process imprecise, unstructured, or fuzzy input data. The nonlinear optimization module is a mathematical modeling tool used to find the global optimal solution to the objective function under complex constraints.

[0062] In this embodiment, a deep convolutional neural network is used to extract features from terrain elevation data and temperature and humidity time series data, and a pre-trained model is used to predict scene change trends within the next 15 seconds. The predicted 16-dimensional scene change trend vector is input into a hybrid model containing a fuzzy logic controller. The fuzzy logic module processes the terrain slope and temperature and humidity correlation parameters through a membership function, and the nonlinear optimization module uses an improved particle swarm algorithm to dynamically adjust the pressure compensation weight coefficient, ultimately generating a hybrid model configuration parameter set containing multiple key parameters.

[0063] S13. Calculate the target air pressure value through the target hybrid model after parameter adjustment based on the moving speed data, hardness level parameter data, terrain data, and temperature and humidity data.

[0064] Among them, movement speed data, hardness level parameter data, terrain data, and temperature and humidity data can be spliced ​​into a mixed input vector after feature engineering. The air pressure target value is the optimal air pressure setting value that meets the hardness requirements of the current scene.

[0065] In this embodiment, the acquired speed data is normalized with the hardness grade parameter, combined with the terrain elevation gradient and the temperature and humidity compensation coefficient, and then fed into a parameter-adjusted target hybrid model. The model first calculates the base pressure compensation using a fuzzy inference system, then performs a secondary correction using a nonlinear optimizer. Finally, a weighted summation formula is used: speed multiplied by the terrain resistance coefficient plus the humidity factor multiplied by the temperature compensation term. This generates a precise target pressure value, outputting a target pressure command with an accuracy of ±0.5 kPa.

[0066] S14. According to the air pressure target value, the gas injection or release rate of the bidirectional pressure regulating device of the mixed target is controlled until the fluctuation amplitude of the actual air pressure value after pressure adjustment is less than the preset fluctuation amplitude threshold and the stable time is greater than the preset time threshold. After the control is completed, the posture of the mixed target is controlled according to the actual air pressure value. The bidirectional pressure regulating device includes an air pump and a pressure relief valve.

[0067] The gas injection or release rate is used to adjust the torque of the air pump's drive motor, the air pump's power, and other parameters. If the actual air pressure fluctuation is less than a preset fluctuation threshold and the duration of stability is greater than a preset duration threshold, it indicates that the actual air pressure after pressure adjustment is stable at the target pressure value. Furthermore, there is a corresponding relationship between the target pressure value and the attitude parameters. Precise attitude control can be achieved based on the attitude parameters corresponding to the target pressure value.

[0068] In an embodiment of the present application, a control algorithm is used to adjust the bidirectional pressure regulating device based on the difference between the target air pressure value and the actual air pressure value. For example, when the target air pressure value is higher than the actual air pressure value, the injection rate is calculated according to the formula of target difference × 0.8 + historical deviation integral × 0.2, and the air pump is driven by modulation based on the injection rate; when the target air pressure value is lower than the actual air pressure value, the pressure relief valve is controlled to open the release rate proportionally. The air pressure fluctuation amplitude is continuously monitored. When the fluctuation value is less than ±0.3kPa within 30 seconds and remains stable for 120 seconds, the system steady-state locking mechanism is triggered. At the same time, after the control is completed, the posture of the mixed target is controlled according to the actual air pressure value.

[0069] Here's a specific example: In a desert range training scenario, a movable physical target vehicle performs an S-shaped maneuver at a speed of 8 km / h. The temperature and humidity sensors measure an ambient temperature of 42°C and a humidity of 15%. A terrain scan reveals a 5° slope of sand ahead, and the neural network predicts that airflow disturbances will occur in the future. The hybrid model calculates that the target air pressure should be increased from 85kPa to 89kPa. The air pump rapidly increases the pressure in a pulsed mode. When the pressure reaches 88.7kPa and fluctuates by less than 0.25kPa for 125 seconds, the system enters steady-state hold mode. After inflation is complete, the hybrid target's posture is controlled based on the actual air pressure. This entire process enables the hybrid target to maintain its optimal posture in complex environments.

[0070] Alternatively, when the same target corresponds to different actual air pressure values, for the same posture, such as vertical ground control, the required equipment traction force and target mounting frame tightness values ​​will be different. Taking vertical ground as an example, the higher the actual air pressure value, the greater the target's rigidity, and therefore the lower the required equipment traction force. Conversely, the lower the actual air pressure value, the lower the target's rigidity, and therefore the greater the environmental impact. To maintain verticality, a higher equipment traction force value is required. Therefore, adjusting the hybrid target's posture based on the actual air pressure value is accurate and reasonable.

[0071] Here is another specific example: a low-cost, mobile, and easily replaceable physical target vehicle that can meet the diverse needs of different training or testing scenarios. The mobile physical target vehicle mainly consists of a chassis system and an upper inflatable target system, including:

[0072] (1) The chassis system is designed as follows: Steel is used as the main frame material, and a truss structure is adopted, which is welded from steel pipes to form a strong and lightweight frame. Reinforcement beams are designed in the front and rear parts to improve impact resistance. The tires are made of wear-resistant rubber material and are equipped with anti-skid patterns to improve grip on different surfaces. The support legs are designed to be adjustable in height and connected to the chassis by bolts to adapt to uneven surfaces. A simple and efficient drive system is designed, which is driven by an electric motor. The movement path and speed of the target vehicle can be controlled by remote control or preset program.

[0073] (2) The design of the inflatable target system is as follows: a modular target mounting frame is designed to allow for quick disassembly and replacement of different types of targets. The target mounting frame can be connected to the chassis by snaps, bolts or magnets to ensure stability and ease of operation. The inflatable target is made of high-strength, wear-resistant polyvinyl chloride material, formed by heat sealing and sewing processes, and connected to the air pressure regulating device installed on the chassis through a connecting pipe. The hardness and stability of the target can be adjusted as needed. A variety of target shapes and sizes are provided, such as launch vehicles, radar vehicles, power supply vehicles, etc., to meet different training or testing needs. In addition, the control system in the movable physical target vehicle integrates a wireless shortwave radio in the chassis, and realizes the target vehicle's movement control, target replacement and air pressure regulation functions through a remote control or a command center console. Equipped with an intelligent control system, the target vehicle's automatic movement and target change are triggered by a preset program, thereby improving the degree of automation of training or testing. In terms of safety design, the embodiment of the present application adopts a smooth design for the target vehicle and target edges to reduce the risk of injury in the event of an accidental collision. It is also equipped with an emergency stop button and overload protection device to ensure safety during operation.

[0074] By executing S11 to S14, the embodiment of the present application achieves adaptive dynamic regulation of the air pressure of the hybrid target by integrating the environmental prediction model with the intelligent control strategy, thereby achieving precise control of the posture. Based on the ability to predict scene change trends, the hybrid model combined with fuzzy logic and nonlinear optimization improves the control robustness under complex terrain and changing climate conditions. The coordinated control mechanism of the bidirectional pressure regulating device effectively reduces the amplitude of air pressure fluctuations and shortens the system stabilization time, enabling the hybrid target to maintain optimal hardness characteristics during dynamic movement and enhancing its adaptability to different training scenarios.

[0075] In a possible embodiment, S13, calculating the target air pressure value through the target hybrid model after parameter adjustment based on the moving speed data, the hardness level parameter data, the terrain data, and the temperature and humidity data, includes:

[0076] Step 131: Align the sampling timestamps of the movement speed data and the hardness level parameter data to generate a multi-feature matrix.

[0077] Among them, alignment uses a double-threshold comparator as the alignment method, and the multi-feature matrix is ​​a spatiotemporal correlation data matrix generated by aligning the timestamps of heterogeneous sensors such as moving speed and hardness level, which contains cross-dimensional features such as speed mean and hardness change rate.

[0078] In this embodiment, the time synchronization module first aligns the timestamps of the mobile speed data collected by the mobile speed sensor with the hardness parameter data recorded by the hardness parameter sensor. A linear interpolation algorithm is used to resample the data streams with inconsistent sampling frequencies. The mobile speed data points at 0.1-second intervals are matched with the corresponding hardness parameter data points along the time axis to generate a two-dimensional multi-feature matrix containing two columns of features. The row dimension of this matrix represents the time series, while the column dimension represents the data from different sensors, ensuring temporal consistency in subsequent feature fusion.

[0079] Step 132: Perform a second-order polynomial basis expansion on the multi-feature matrix to generate an expanded feature matrix, perform local response normalization on the environmental feature tensor to generate a normalized tensor, and calculate the differential feature vector based on the time series of historical air pressure values. The environmental feature tensor is generated based on terrain data and temperature and humidity data.

[0080] Among them, the extended feature matrix is ​​a high-dimensional matrix generated by nonlinearly combining the original features through second-order polynomial basis functions; local response normalization is an operation that performs nonlinear scaling on environmental features such as terrain slope and curvature; and the differential feature vector is a rate of change vector calculated based on a sliding window of the historical pressure series.

[0081] In this embodiment of the present application, the original two-column feature matrix is ​​expanded into a five-column feature matrix by calculating the squared terms of velocity and hardness, as well as the cross-term of the product of velocity and hardness. Simultaneously, the terrain data and temperature and humidity data are combined into a four-dimensional environmental feature tensor. Using local response normalization, the eigenvalues ​​within three adjacent time windows are normalized around each feature point to generate a normalized tensor. Furthermore, based on the time series of historical pressure values, the first-order differences of adjacent timestamps are calculated to generate a differential feature vector.

[0082] Step 133: Perform weighted concatenation on the extended feature matrix, the normalized tensor, and the differential feature vector to form a mixed feature vector.

[0083] Among them, weighted splicing is an operation that assigns weight coefficients through the entropy weight method and fuses feature vectors of different dimensions in the latent space.

[0084] In this embodiment, a weight coefficient of 0.6 is first assigned to the expanded feature matrix, a weight of 0.3 to the normalized tensor, and a weight of 0.1 to the differential feature vector. A 10-dimensional hybrid feature vector is generated using the feature dimension concatenation formula. Terrain-related features dominate the weight in the normalized tensor, ensuring that environmental factors are not significantly affected by the final feature.

[0085] Step 134 : Perform nonlinear mapping on the mixed feature vector using the target mixed model after parameter adjustment to obtain the target pressure value.

[0086] Among them, nonlinear mapping is an operation of mapping a high-dimensional feature vector to a target air pressure value through a hybrid model.

[0087] In this embodiment, the parameter-adjusted target hybrid model receives a mixed feature vector as input and performs nonlinear mapping via a three-layer fully connected neural network. The first layer uses an activation function to nonlinearly transform the features, the second layer generates intermediate feature weights using an sigmoid function, and the third layer uses a linear regression layer to map the weighted features to a target pressure value. Gradient clipping is performed before the model output to prevent numerical overflow, and the resulting target pressure value strictly matches the current environmental state and the target's dynamic characteristics.

[0088] The following is a specific example: During desert range training, when a mobile target vehicle performs an S-shaped maneuver at 8 km / h, the system aligns the time series data from the mobile speed sensor and the hardness sensor in real time. Using cubic spline interpolation, the speed and target hardness values ​​are unified to a 0.01-second time granularity, generating a 1200×2-dimensional multi-feature matrix. A second-order polynomial basis expansion is then performed, adding squared velocity terms, squared hardness terms, and velocity-hardness cross terms to form a 1200×5 expanded feature matrix. Simultaneously, the environmental feature tensor constructed from a 5° slope, 42°C temperature, and 15% humidity is locally normalized and combined with historical air pressure values ​​to generate a differential feature vector [1,1]. After predicting airflow disturbances, the system dynamically adjusts feature weights: the expanded features are assigned a weight of 0.65 to enhance nonlinear relationships, the normalized environmental features are assigned a weight of 0.3 to highlight the influence of slope, and the differential features are assigned a weight of 0.05. Ultimately, these features are concatenated into a 10-dimensional hybrid feature vector. The target mixture model interprets this vector using a three-layer neural network: the first layer activates the nonlinear coupling between velocity and slope, the second layer uses a weighted S-function to determine if the air pressure needs to be accelerated, and the final output corrects the target pressure from 89kPa to 89.8kPa to compensate for airflow disturbances. The air pump switches to high-frequency pulse mode accordingly, rapidly increasing the air pressure from 85kPa to 88.7kPa in 12 seconds. After the fluctuation stabilized at ±0.25kPa for 125 seconds, the target maintained optimal impact feedback despite dust and sudden airflow fluctuations. This complete closed-loop verification verifies the effectiveness of multi-feature fusion and dynamic compensation.

[0089] By executing steps 131-134, the present embodiment improves the accuracy and robustness of target pressure prediction through multi-dimensional feature fusion and nonlinear transformation. Timestamp alignment ensures temporal consistency of multi-source data, polynomial expansion enhances the model's ability to capture nonlinear relationships, local response normalization effectively suppresses environmental noise interference, and a weighted splicing mechanism strengthens the decision weights of key features. Ultimately, a deep neural network achieves high-precision mapping of complex features to target pressure, enabling target pressure regulation to achieve both rapid response and steady-state maintenance.

[0090] In one possible embodiment, step 132 includes performing a second-order polynomial basis expansion on the multi-feature matrix to generate an expanded feature matrix, performing local response normalization on the environmental feature tensor to generate a normalized tensor, and calculating a differential feature vector based on a time series of historical pressure values. The environmental feature tensor is generated based on terrain data and temperature and humidity data, including:

[0091] Step a1: For each feature dimension in the multi-feature matrix, calculate the original eigenvalue on the feature dimension, the product term of each two feature dimensions, and the square term of a single feature dimension, and concatenate the original eigenvalue, product term, and square term by column to generate an extended feature matrix.

[0092] Among them, the extended feature matrix is ​​a high-dimensional data matrix generated by performing a second-order combination of the original features, which contains the original eigenvalues, pairwise feature product terms and single feature square terms, and is used to capture the nonlinear coupling effects between multidimensional parameters.

[0093] In an embodiment of the present application, each feature dimension of a multi-feature matrix is ​​first traversed, and the original eigenvalues ​​are extracted for each feature dimension. The product terms between each feature dimension and the square terms of each feature dimension are calculated. All original eigenvalues, product terms, and square terms are concatenated by column. For example, the original 2-dimensional feature matrix is ​​expanded to generate a 5-dimensional expanded feature matrix containing the original columns, product columns, and square columns, thereby achieving nonlinear enhancement of the feature space.

[0094] Step a2: For each channel of the environmental feature tensor, select the square sum of the activation values ​​of the adjacent m channels as the normalization base, weight the normalization base with a preset channel attenuation factor, divide the activation value of the current channel by the weighted normalization base to obtain a normalized value, perform nonlinear compression on the normalized value through the hyperbolic tangent function, and generate a normalized tensor.

[0095] Among them, the normalized tensor is a standardized expression of environmental features calculated by weighting the sum of squared activation values ​​across channels, which is used to suppress the numerical offset caused by sudden changes in terrain.

[0096] In this embodiment of the present application, for each channel of the environmental feature tensor, the square sum of the activation values ​​of its m adjacent channels is selected as the normalization cardinality. The normalization cardinality is weighted by a preset channel attenuation factor, and the activation value of the current channel is divided by the weighted normalization cardinality to obtain a preliminary normalized value. The normalized value is then nonlinearly compressed using a hyperbolic tangent function, constraining the numerical range to the interval [-1, 1], ultimately generating a normalized tensor.

[0097] Step a3: Extract multiple windows of historical pressure values ​​from the time series of historical pressure values ​​using a fixed window length. Calculate the absolute difference sequence between adjacent historical pressure values ​​within each window. Perform a sliding average filter on the absolute difference sequence to obtain a filtered sequence corresponding to each window. Accumulate and sum the filtered sequences corresponding to all windows to generate a differential feature vector.

[0098] The differential eigenvector is a vector that extracts the pressure trend using a sliding window and is used to characterize short-term fluctuations caused by sudden environmental changes. Sliding average filtering is a technique used to smooth time series data. Its purpose is to reduce the impact of random noise on the pressure difference sequence between adjacent time points, thereby extracting more stable trend features.

[0099] In this embodiment, multiple sliding window sequences are extracted from a time series of historical air pressure values ​​using a fixed window length. For each window, the absolute differences between the pressure values ​​at adjacent time points are calculated to form an absolute difference sequence. This sequence is then filtered using a sliding average filter to generate a filtered sequence. Finally, the filtered sequences for all windows are cumulatively summed at each time point to generate a differential feature vector representing the intensity of the air pressure change.

[0100] The following is a specific example: During desert range training, when a mobile target vehicle performs an S-shaped maneuver at 8.2 km / h, the system constructs a two-dimensional multi-feature matrix based on the real-time collected movement speed and target hardness. A five-dimensional extended feature matrix is ​​generated by calculating the product of speed × hardness, the square of speed, and the square of hardness. Simultaneously, an environmental feature tensor is constructed using the 5° slope acquired by terrain scanning, the 42°C temperature, and the 15% humidity detected by the temperature and humidity sensors. For the slope channel, the adjacent temperature and humidity channels are selected to calculate a normalized cardinality of 994.5, weighted by a preset channel attenuation factor of 0.5. The original slope value of 5 is divided by this cardinality and compressed to 0.157 using a hyperbolic tangent function to generate an interference-resistant normalized tensor. Based on historical air pressure values, the system captures the sequence with a fixed 10-second window, calculates the absolute difference between adjacent time points, and applies a three-point sliding average filter to obtain a smoothed sequence [0.67, 0.67]. This is then accumulated to generate a differential feature vector value of 1.34, representing the trend of sudden pressure changes. The extended feature matrix, normalized slope and differential features are weighted and fused and then input into the target hybrid model, driving the air pump to quickly increase the air pressure from 85kPa to 89kPa under dust disturbance, and finally stabilized at 88.7kPa through closed-loop control, verifying the synergistic effect of multi-feature expansion, environmental perception normalization and dynamic differential analysis, ensuring that the target maintains the best posture in extreme environments.

[0101] By executing steps a1-a3, this embodiment of the present application enhances the model's ability to capture nonlinear relationships through multi-dimensional feature expansion, utilizes a channel-aware normalization mechanism to improve the interference resistance of environmental features, and combines sliding window differential analysis to accurately extract air pressure trends. The synergistic effect of feature expansion and normalization improves the robustness of target value prediction in complex environments, while dynamic differential feature construction provides a forward-looking adjustment basis for the system, achieving overall coordinated optimization of air pressure control accuracy and response speed.

[0102] In a possible embodiment, step a1, weighted concatenation of the extended feature matrix, the normalized tensor, and the differential feature vector to form a mixed feature vector, includes:

[0103] Step b1: assigning feature dimension-related weight coefficients, channel number-related weight coefficients, and fixed-value weight coefficients to the extended feature matrix, normalized tensor, and differential feature vector, respectively.

[0104] The feature dimension-related weight coefficient is a dynamic weight calculated based on the information entropy of each dimension of the feature matrix, reflecting the importance of data distribution across different feature dimensions. The channel-related weight coefficient is a channel attention vector generated by the attention mechanism based on the number of channels in the input data. The fixed numerical weight coefficient is a predefined constant weight used to ensure the stability of the basic features.

[0105] Step b2: Perform Hadamard product operations on the extended feature matrix, normalized tensor, and differential eigenvector according to the weight coefficients related to the feature dimension, the weight coefficients related to the number of channels, and the weight coefficients of fixed values, to obtain the weighted extended feature matrix, the weighted normalized tensor, and the weighted differential eigenvector.

[0106] Among them, the Hadamard product is a matrix element-level multiplication, which is used to fuse the weight coefficients with the original features element by element. The formula of the Hadamard product is: ,in, is the original expanded feature matrix, is the weight coefficient matrix related to the feature dimension, is the Hadamard product operator, is the weighted extended feature matrix.

[0107] Step b3: Convert the weighted extended feature matrix and the weighted normalized tensor into different formats to obtain an extended feature vector and a normalized vector.

[0108] Among them, format conversion is to convert the high-dimensional tensor into a vector form that matches the dimension of the target feature space.

[0109] Step b4: splicing the elements at the same position in the extended feature vector, the normalized vector and the weighted differential feature vector according to a preset splicing order to form a mixed feature vector.

[0110] A hybrid feature vector is a composite feature vector formed by concatenating feature vectors from different sources in a preset order. Its purpose is to jointly express multi-dimensional, multi-physical features in a unified vector space, enhancing the model's global perception of complex scenarios. For example, if the extended feature vector contains a nonlinear combination of velocity and hardness, the normalized feature vector contains compressed terrain parameters, and the differential feature vector contains the rate of change of air pressure, the hybrid feature vector integrates this cross-dimensional information into a single vector for subsequent model inference.

[0111] The following is a specific example: In the desert shooting range test scenario, first, the extended feature matrix is ​​collected by the target vehicle sensor, and the weight coefficient matrix related to the feature dimension is assigned 0.7, 0.7, 0.6, 0.6; at the same time, the normalized tensor generated by the terrain sensor is assigned channel number-related weight coefficients 0.4, 0.5, 0.3, 0.3, and a fixed numerical weight coefficient of 0.2 is assigned to the differential eigenvector. Secondly, the Hadamard product operation is performed on the extended feature matrix: the speed is 9.5m / s×0.7=6.65, the hardness is 7.8×0.7=5.46, the temperature is 42×0.6=25.2, and the humidity is 12×0.6=7.2 to generate the weighted extended feature matrix 6.65, 5.46, 25.2, 7.2; the Hadamard product operation is performed on the normalized tensor: the slope is 15°×0.4=6.0, the roughness is 0.9×0.5=0.45, the temperature is 42℃×0.3=12.6, and the humidity is 12%×0.3=3.6 to generate the weighted normalized tensor 6.0, 0.45, 12.6, 3.6; the differential eigenvector is 3.2kPa / s×0.2=0.64 to generate the weighted differential eigenvector 0.64. Next, the weighted expanded feature matrix and the normalized tensor were flattened into vectors 6.65, 5.46, 25.2, 7.2 and 6.0, 0.45, 12.6, 3.6, respectively. Finally, the three were concatenated in the order of "expansion, normalization, and difference" to generate a mixed feature vector 6.65, 5.46, 25.2, 7.2, 6.0, 0.45, 12.6, 3.6, 0.64. After inputting into the model, the output target air pressure value was 96.8 kPa, which was only 1.3 kPa lower than the measured value of 98.1 kPa. The air pressure was maintained stable for 28 seconds during a sandstorm, verifying the effectiveness of the weight allocation and feature fusion mechanism.

[0112] By executing steps b1 to b4, the embodiment of the present application achieves efficient fusion of multi-source heterogeneous features by dynamically allocating feature dimensions, number of channels, and fixed weight coefficients, combined with Hadamard product operations and format conversion. The extended feature matrix enhances the expression ability of spatial information, the normalized tensor improves the modeling of inter-channel correlation, and the differential feature vector captures local detail changes. The resulting hybrid feature vector has strong compatibility and is suitable for downstream tasks such as classification and detection, improving the model's ability to represent complex data.

[0113] In a possible embodiment, S13, performing nonlinear mapping on the mixed feature vector using the target mixed model after parameter adjustment to obtain the air pressure target value, includes:

[0114] Step c1: extracting a first sub-vector related to the terrain and target hardness, and a second sub-vector related to the environment and movement speed from the mixed feature vector.

[0115] The first sub-vector extracts a subset of features directly related to terrain and target hardness from the mixed feature vector, used to characterize the physical interaction between the machine and terrain. The second sub-vector extracts a subset of features related to movement speed and environmental parameters, used to characterize the coupling effect between dynamic motion and environmental disturbances.

[0116] Step c2: Input the first sub-vector and the second sub-vector into the fuzzy logic module and the nonlinear optimization module respectively. The fuzzy logic module calculates the fuzzy reasoning result according to the predefined fuzzy logic rule base and the adjusted weight parameter of the membership function, and converts the fuzzy reasoning result into the first air pressure compensation value. The nonlinear optimization module calculates the air pressure reference value based on the adjusted penalty factor and iteration step size of the nonlinear optimization algorithm.

[0117] The fuzzy logic rule base is a set of conditional rules based on expert experience or data-driven, used to describe the nonlinear mapping relationship between input features and output actions (such as air pressure compensation value). The membership function is used to map the precise input value (such as slope = 0.12) to the membership degree (a value between 0 and 1) of the fuzzy linguistic variable (such as "steep" or "flat"). The function formula is: ,in, is the input eigenvalue, is the central value of the fuzzy linguistic variable, is the standard deviation, which controls the width of the function, is a logarithm. The fuzzy inference result is a set of fuzzy outputs calculated by the fuzzy rule base and the membership function, which needs to be defuzzified (such as the center of gravity method) to be converted into an exact value. The first air pressure compensation value is the air pressure adjustment output by the fuzzy logic module, which is used to compensate for the impact of sudden changes in terrain or changes in target hardness on the system. The penalty factor is a coefficient used to penalize violations of constraints in the nonlinear optimization algorithm and controls the strictness of the constraints during the optimization process. The iteration step size is the amplitude of the parameter update at each iteration in the nonlinear optimization algorithm, which affects the convergence speed and stability. The air pressure reference value is the preliminary air pressure target value calculated by the nonlinear optimization module based on the dynamic model and environmental constraints, without considering fuzzy factors such as sudden changes in terrain.

[0118] Step c3: superimpose the first air pressure compensation value and the air pressure reference value to obtain an air pressure target value.

[0119] The air pressure target value refers to the final air pressure setting value obtained by superimposing the first air pressure compensation value output by the fuzzy logic module and the air pressure reference value calculated by the nonlinear optimization module.

[0120] The following is a specific example: In a desert shooting range test, when the system detected that the target vehicle was passing a 5° sand slope at 8.2 km / h, the fuzzy logic module output a +3.2 kPa compensation value based on the "large slope + low humidity" rule combination. At the same time, the nonlinear optimization module calculated a baseline value of 85.8 kPa considering the material expansion effect caused by high temperature. During the superposition processing, the system automatically limited the compensation value to not exceed the safety threshold, and finally generated a target value of 89.0 kPa, driving the air pump to complete the pressurization within 12 seconds, so that the target maintained a working pressure of 88.7±0.2 kPa in the sandstorm environment, verifying the effectiveness of the dual-module collaborative control.

[0121] By executing steps c1 to c3, the embodiment of the present application achieves specialized processing of different modules through intelligent segmentation of characteristic sub-vectors. The fuzzy logic module effectively captures the complex nonlinear relationship between terrain and hardness, and the nonlinear optimization module accurately quantifies the dynamic impact of environment and speed. The synergistic superposition of the dual module outputs maintains baseline stability and has the ability to respond quickly to disturbances.

[0122] In a possible embodiment, S12, predicting a scene change trend using a pre-trained neural network model based on terrain data and temperature and humidity data to obtain a scene change trend vector, and adjusting parameters of a target hybrid model based on the scene change trend vector, includes:

[0123] Step 121: discretize the terrain data into a terrain elevation matrix according to preset grid units, map the temperature and humidity data into a temperature and humidity distribution tensor according to spatial coordinates, perform cross-modal splicing on the terrain elevation matrix and the temperature and humidity distribution tensor, and generate an environmental feature tensor.

[0124] The terrain elevation matrix discretizes continuous terrain elevation data into a two-dimensional matrix according to a preset grid, with each grid cell storing the elevation value at the corresponding location. The temperature and humidity distribution tensor maps the point data from the temperature and humidity sensors into a three-dimensional tensor through spatial interpolation, with channels corresponding to temperature and humidity, respectively. Cross-modal splicing combines the terrain matrix and the temperature and humidity tensor along the channel dimension to form a unified environmental feature tensor that includes terrain, temperature, and humidity.

[0125] Step 122: Input the environmental feature tensor into a pre-trained neural network model, combine it with the historical environmental feature tensor in the time sliding window, predict the scene change trend, and output a scene change trend vector that represents the dynamic evolution of the terrain and the intensity of climate fluctuations.

[0126] The scene change trend vector is a multidimensional vector output by the neural network model, containing components such as the intensity of terrain dynamic evolution and climate fluctuations. The environmental feature tensor integrates multiple environmental parameters into a unified multidimensional data structure for machine learning and model processing.

[0127] Step 123: According to the terrain-related component in the scene change trend vector, the weight parameters of the membership function in the fuzzy logic rule base corresponding to the fuzzy logic module are adjusted, and according to the climate-related component in the scene change trend vector, the penalty factor and iteration step size of the nonlinear optimization algorithm adopted by the nonlinear optimization module are adjusted.

[0128] The terrain-related component is a parameter in the trend vector that describes terrain evolution and is used to adjust the membership function of the fuzzy logic. The climate-related component is a parameter in the trend vector that describes climate fluctuations and is used to adjust the penalty factor and step size of the optimization algorithm.

[0129] The following is a specific example: In a desert range training scenario, a vehicle-mounted lidar scans the terrain, discretizing the continuous elevation data of the sand dunes into 0.5m×0.5m grid cells, generating a two-dimensional terrain elevation matrix encompassing a 5° slope. Simultaneously, real-time data collected by a network of temperature and humidity sensors deployed around the range is mapped into a three-dimensional temperature and humidity distribution tensor aligned with the terrain grid using an inverse distance weighted interpolation algorithm. The terrain matrix is ​​then cross-modally concatenated with the climate tensor along the channel dimension to form an environmental feature tensor that integrates the slope, temperature, and humidity dynamics of the sand dunes. This tensor is then fed into a pre-trained spatiotemporal convolutional neural network, combined with historical data from a 30-minute sliding window, to predict the scenario's trend over the next 10 minutes. The terrain component outputs the increase in dune slope due to wind erosion, while the climate component outputs the fluctuations in local airflow intensity caused by the target vehicle's maneuvering. The weight of the "sand softness" membership function in the fuzzy logic module is adjusted according to the terrain components, and the center value of the Gaussian function is shifted from loose sand to semi-solidified sand to match the dynamic evolution of sand dunes; at the same time, the penalty factor of the nonlinear optimization algorithm is increased according to the climate component, and the iterative step size of the air pressure regulation is shortened, so that the air pump can respond quickly in a pulse mode when the airflow suddenly changes, and maintain the wind pressure stability of the hybrid target after the air pressure reaches the steady-state threshold, ultimately achieving precise attitude control of the target in a complex environment.

[0130] By executing steps 121 to 123, the embodiment of the present application realizes the spatial alignment and joint modeling of terrain and climate data through the fusion of environmental feature tensors, utilizes the spatiotemporal neural network to extract time series features and predict trends of the complex nonlinear relationships of the dynamic scenes of the shooting range, and through the dynamic parameter adjustment mechanism of fuzzy logic and optimization algorithm, enables the system to adapt to the evolution of sand slope and fluctuations of airflow intensity, and optimize the air pressure regulation strategy in real time.

[0131] In one possible embodiment, step 123, adjusting the weight parameter of the membership function in the fuzzy logic rule base corresponding to the fuzzy logic module according to the terrain-related component in the scene change trend vector, and adjusting the penalty factor and iteration step size of the nonlinear optimization algorithm used by the nonlinear optimization module according to the climate-related component in the scene change trend vector, includes:

[0132] Step d1: Analyze the terrain dynamic component in the scene change trend vector to extract the slope change rate and elevation fluctuation intensity, and analyze the climate fluctuation component in the scene change trend vector to extract the temperature and humidity mutation direction and wind speed correlation intensity.

[0133] The slope change rate is the rate of change of terrain slope per unit time, quantifying the dynamic evolution of terrain steepness. The elevation fluctuation intensity is the magnitude or degree of fluctuation of terrain elevation over a specific time period, typically expressed as a standard deviation or rate of change. The temperature and humidity mutation direction is the direction of change in temperature and humidity parameters over a short period of time. The wind speed correlation strength is the statistical correlation strength between wind speed changes and other climate parameters, typically expressed as a correlation coefficient or regression coefficient.

[0134] Step d2: According to the slope change rate and the elevation fluctuation intensity, adjust the weight parameters of the membership function of the first rule and the weight parameters of the membership function of the second rule in the fuzzy logic rule base respectively. The first rule is the rule between slope and hardness compensation, and the second rule is the rule between terrain flatness and air pressure stability.

[0135] Among them, the first rule is to define the association rule between slope and target hardness for air pressure compensation, and the second rule is to define the association rule between terrain flatness and air pressure stability. First, the "steep slope" judgment standard is automatically adjusted based on the slope change rate: when the slope accelerates and becomes steeper, the system will relax the judgment threshold of "steep slope" and increase the weight of related rules, so that the fuzzy logic can trigger air pressure compensation more sensitively. Secondly, based on the intensity of elevation fluctuations, the judgment range of "terrain flatness" is dynamically adjusted: if the ground fluctuates violently, the system will expand the coverage area of ​​"low flatness" and reduce the confidence weight of related rules, thereby reducing misjudgments caused by short-term terrain noise. Through this dynamic adaptation, the model can more accurately balance slope risks and ground stability in scenarios such as rainstorm collapse, improve the response speed of air pressure control, and reduce terrain-related errors.

[0136] Step d3: Adjust the penalty factor and iteration step size of the nonlinear optimization algorithm according to the temperature and humidity mutation direction and the wind speed correlation strength.

[0137] When a sudden change in temperature or humidity is detected, the algorithm strengthens the application of climate constraints by increasing the penalty factor. For example, if the humidity changes in a positive direction, the algorithm proportionally increases the penalty factor, forcing the optimization process to prioritize physical constraints related to humidity and avoid model failure due to excessively high humidity. If the change in direction is negative, the penalty factor is appropriately reduced to reduce excessive constraints on low temperature constraints. Furthermore, the iteration step size is dynamically adjusted based on the strength of the wind speed correlation. If the correlation between wind speed and climate parameters is strong, the step size is reduced to suppress optimization oscillations and ensure stable convergence in windy environments. If the correlation is weak, the step size is increased to accelerate the solution. For example, in a typhoon scenario, when the wind speed correlation strength reaches 0.3, the step size is reduced by 20%, reducing the pressure control response delay from 1.5 seconds to 1.1 seconds and optimizing the stability standard deviation to 0.25 kPa. This adjustment mechanism balances the strength of environmental changes with algorithm robustness in real time, ensuring high accuracy and efficient convergence in complex scenarios.

[0138] Here's a specific example: During dynamic training at a desert range, while a target vehicle performs an S-shaped maneuver at 8 km / h, a lidar scans the dune terrain in real time, extracting the current slope change rate and elevation fluctuation intensity. The temperature and humidity sensors simultaneously detect a sudden drop in ambient temperature from 42°C to 38°C, and the anemometer measures a gust-correlation coefficient of 0.8. Based on the slope change rate, the fuzzy logic module adjusts the μ value of the Gaussian function for the "slope > 5°" range from 5.2 to 6.0, triggering a 15% increase in target hardness compensation. Based on the elevation fluctuation intensity, the weight of "high flatness" in the triangular membership function is increased from 40% to 55%, driving the air pump to increase the base air pressure from 85 kPa to 87 kPa. To address sudden northwestward changes in temperature and humidity, the optimization algorithm's penalty factor, λ, was increased from 1.2 to 2.5 to suppress pressure fluctuations caused by sudden temperature and humidity changes. Combined with a wind speed correlation strength of 0.8, the iteration step size, α, was reduced from 0.1 to 0.0625, enabling the air pump to stabilize the air pressure to 88.7 kPa in pulse mode within 12 seconds, with fluctuations within ±0.25 kPa. This entire mechanism ultimately enabled the hybrid target to maintain optimal posture despite sudden changes in dune slope and gusty wind disturbances.

[0139] By executing steps d1 to d3, the embodiment of the present application realizes the quantitative extraction of dynamic features such as slope mutations and elevation fluctuations through refined analysis of terrain and climate components. Combined with the adaptive adjustment of the fuzzy logic rule library and the dynamic optimization of parameters of the nonlinear optimization algorithm, the system can respond to the evolution of dune slopes and temperature and humidity mutations in real time in complex scenarios such as desert shooting ranges, accurately adjust the target hardness compensation strategy and air pressure stability control, thereby maintaining the target's impact resistance and motion trajectory stability under extreme conditions such as strong wind disturbances and sand deformation, and improving the environmental adaptability and feedback authenticity of mobile target vehicle training simulation.

[0140] Figure 2 A structural diagram of a hybrid target posture control system provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes:

[0141] The acquisition module 21 is used to acquire the moving speed data of the movable physical target vehicle, the hardness level parameter data and the actual value of the air pressure of the mixed target, and the terrain data and temperature and humidity data of the environment where the movable physical target vehicle is located.

[0142] The adjustment module 22 is used to predict the scene change trend based on the terrain data and temperature and humidity data through a pre-trained neural network model, obtain the scene change trend vector, and adjust the parameters of the target hybrid model according to the scene change trend vector. The target hybrid model is a hybrid model that integrates the fuzzy logic module and the nonlinear optimization module.

[0143] The calculation module 23 is used to calculate the target air pressure value based on the moving speed data, hardness level parameter data, terrain data and temperature and humidity data through the target hybrid model with adjusted parameters.

[0144] The regulating module 24 is used to control the gas injection or release rate of the bidirectional pressure regulating device of the mixed target according to the air pressure target value until the fluctuation amplitude of the actual air pressure value after pressure regulation is less than the preset fluctuation amplitude threshold and the stable time is greater than the preset time threshold. The bidirectional pressure regulating device includes an air pump and a pressure relief valve.

[0145] Figure 2 The described hybrid target posture control system can execute Figure 1 The implementation principle and technical effects of the hybrid target posture control method described in the illustrated embodiment will not be elaborated here. The specific manner in which each module and unit performs operations in the hybrid target posture control system in the above embodiment has been described in detail in the embodiment of the method and will not be elaborated here.

[0146] In one possible design, Figure 2 A hybrid target posture control system of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32 .

[0147] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0148] The processing component 32 is used to obtain the moving speed data of the movable physical target vehicle, the hardness level parameter data and the actual air pressure value of the hybrid target, and the terrain data and temperature and humidity data of the environment in which the movable physical target vehicle is located. Based on the terrain data and temperature and humidity data, the scene change trend is predicted by a pre-trained neural network model to obtain a scene change trend vector, and the parameters of the target hybrid model are adjusted according to the scene change trend vector. The target hybrid model is a hybrid model that integrates a fuzzy logic module and a nonlinear optimization module. Based on the moving speed data, hardness level parameter data, terrain data and temperature and humidity data, the air pressure target value is calculated by the target hybrid model after parameter adjustment. According to the air pressure target value, the gas injection or release rate of the bidirectional pressure regulating device of the hybrid target is controlled until the fluctuation amplitude of the actual air pressure value after pressure adjustment is less than the preset fluctuation amplitude threshold, and the stable time is greater than the preset time threshold. The bidirectional pressure regulating device includes an air pump and a pressure relief valve.

[0149] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0150] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as random access memory (RAM), static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0151] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0152] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0153] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0154] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0155] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 A hybrid target posture control method according to the illustrated embodiment.

[0156] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0157] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0158] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for controlling the posture of a hybrid target, characterized in that: include: Obtaining the moving speed data of the movable physical target vehicle, the hardness level parameter data and the actual value of the air pressure of the mixed target, and the terrain data and temperature and humidity data of the environment where the movable physical target vehicle is located; Based on the terrain data and temperature and humidity data, a pre-trained neural network model is used to predict scene change trends to obtain a scene change trend vector, and parameters of a target hybrid model are adjusted based on the scene change trend vector. The target hybrid model is a hybrid model that integrates a fuzzy logic module and a nonlinear optimization module. Calculating an air pressure target value through a target hybrid model with adjusted parameters based on the moving speed data, the hardness level parameter data, the terrain data, and the temperature and humidity data; According to the air pressure target value, the gas injection or release rate of the bidirectional pressure regulating device of the mixed target is controlled until the fluctuation amplitude of the actual air pressure value after pressure regulation is less than the preset fluctuation amplitude threshold and the stable time is longer than the preset time threshold. After the control is completed, the posture of the mixed target is controlled according to the actual air pressure value, and the bidirectional pressure regulating device includes an air pump and a pressure relief valve; The calculating of the air pressure target value by a target hybrid model after parameter adjustment based on the moving speed data, the hardness level parameter data, the terrain data, and the temperature and humidity data includes: Aligning sampling timestamps of the movement speed data and the hardness level parameter data to generate a multi-feature matrix; performing a second-order polynomial basis expansion on the multi-feature matrix to generate an expanded feature matrix, performing local response normalization on an environmental feature tensor to generate a normalized tensor, and calculating a differential feature vector based on a time series of historical air pressure values, wherein the environmental feature tensor is generated based on the terrain data and the temperature and humidity data; Performing weighted concatenation on the extended feature matrix, the normalized tensor, and the differential feature vector to form a mixed feature vector; The target mixed model after the parameter adjustment is used to perform nonlinear mapping on the mixed characteristic vector to obtain the air pressure target value.

2. The method according to claim 1, characterized in that The step of performing a second-order polynomial basis expansion on the multi-feature matrix to generate an expanded feature matrix, performing local response normalization on the environmental feature tensor to generate a normalized tensor, and calculating a differential feature vector based on a time series of historical air pressure values ​​includes: For each feature dimension in the multi-feature matrix, calculating the original eigenvalue on the feature dimension, the product term of each two feature dimensions, and the square term of a single feature dimension, and splicing the original eigenvalue, the product term, and the square term by column to generate an extended feature matrix; For each channel of the environmental feature tensor, the square sum of the activation values ​​of the adjacent m channels is selected as the normalization base, the normalization base is weighted by a preset channel attenuation factor, the activation value of the current channel is divided by the weighted normalization base to obtain a normalized value, and the normalized value is nonlinearly compressed using a hyperbolic tangent function to generate a normalized tensor; With a fixed window length, the air pressure history value sequences in multiple windows are cut out from the time series of the air pressure history values, and the absolute difference sequence between the air pressure history values ​​at adjacent time points in each window is calculated. The absolute difference sequence is subjected to a sliding average filter to obtain a filtered sequence corresponding to each window; the filtered sequences corresponding to all windows are accumulated and summed to generate a differential feature vector.

3. The method according to claim 2, characterized in that The weighted concatenation of the extended feature matrix, the normalized tensor, and the differential feature vector to form a mixed feature vector includes: Allocating a weight coefficient related to feature dimension, a weight coefficient related to the number of channels, and a weight coefficient of a fixed value to the extended feature matrix, the normalized tensor, and the differential feature vector, respectively; According to the weight coefficient related to the feature dimension, the weight coefficient related to the number of channels, and the weight coefficient of the fixed value, Hadamard product operations are performed on the extended feature matrix, the normalized tensor, and the differential eigenvector to obtain a weighted extended feature matrix, a weighted normalized tensor, and a weighted differential eigenvector; Performing format conversion on the weighted extended feature matrix and the weighted normalized tensor respectively to obtain an extended feature vector and a normalized vector; Elements at the same position in the extended feature vector, the normalized vector and the weighted differential feature vector are spliced ​​in a preset splicing order to form a mixed feature vector.

4. The method according to any one of claims 1 to 3, characterized in that The nonlinear mapping of the mixed characteristic vector by the target mixed model after the parameter adjustment to obtain the air pressure target value includes: Extracting a first sub-vector related to the terrain and target hardness, and a second sub-vector related to the environment and movement speed from the mixed feature vector; Inputting the first sub-vector and the second sub-vector into a fuzzy logic module and a nonlinear optimization module, respectively; the fuzzy logic module calculates a fuzzy inference result according to a predefined fuzzy logic rule base and in combination with the adjusted weight parameter of the membership function, and converts the fuzzy inference result into a first air pressure compensation value; and the nonlinear optimization module calculates an air pressure reference value based on the adjusted penalty factor and iteration step size of the nonlinear optimization algorithm; The first air pressure compensation value and the air pressure reference value are superimposed to obtain an air pressure target value.

5. The method according to claim 1, wherein The method includes predicting a scene change trend using a pre-trained neural network model based on the terrain data and the temperature and humidity data to obtain a scene change trend vector, and adjusting parameters of a target hybrid model based on the scene change trend vector, including: Discretizing the terrain data into a terrain elevation matrix according to preset grid units, mapping the temperature and humidity data into a temperature and humidity distribution tensor according to spatial coordinates, and performing cross-modal splicing on the terrain elevation matrix and the temperature and humidity distribution tensor to generate an environmental feature tensor; The environmental feature tensor is input into a pre-trained neural network model, combined with the historical environmental feature tensor within the time sliding window, to predict the scene change trend and output a scene change trend vector representing the dynamic evolution of the terrain and the intensity of climate fluctuations; According to the terrain-related component in the scene change trend vector, the weight parameter of the membership function in the fuzzy logic rule base corresponding to the fuzzy logic module is adjusted, and according to the climate-related component in the scene change trend vector, the penalty factor and iteration step size of the nonlinear optimization algorithm adopted by the nonlinear optimization module are adjusted.

6. The method according to claim 5, characterized in that The adjusting, based on the terrain-related component in the scene change trend vector, the weight parameter of the membership function in the fuzzy logic rule base corresponding to the fuzzy logic module, and based on the climate-related component in the scene change trend vector, the penalty factor and iteration step size of the nonlinear optimization algorithm adopted by the nonlinear optimization module, include: Analyze the terrain dynamic component in the scene change trend vector to extract the slope change rate and elevation fluctuation intensity. Analyze the climate fluctuation component in the scene change trend vector to extract the temperature and humidity mutation direction and wind speed correlation intensity. adjusting, according to the slope change rate and the elevation fluctuation intensity, weight parameters of a membership function of a first rule and a weight parameter of a membership function of a second rule in a fuzzy logic rule base, respectively, wherein the first rule is a rule between slope and hardness compensation, and the second rule is a rule between terrain flatness and air pressure stability; According to the temperature and humidity mutation direction and the wind speed correlation strength, the penalty factor and the iteration step size of the nonlinear optimization algorithm are adjusted respectively.

7. A hybrid target attitude control system, characterized in that: include: An acquisition module is used to obtain the moving speed data of the movable physical target vehicle, the hardness grade parameter data and the actual value of the air pressure of the mixed target, and the terrain data and temperature and humidity data of the environment where the movable physical target vehicle is located; an adjustment module, configured to predict scene change trends using a pre-trained neural network model based on the terrain data and the temperature and humidity data, obtain a scene change trend vector, and adjust parameters of a target hybrid model based on the scene change trend vector, wherein the target hybrid model is a hybrid model that integrates a fuzzy logic module and a nonlinear optimization module; a calculation module, configured to calculate an air pressure target value through a target hybrid model with adjusted parameters based on the moving speed data, the hardness level parameter data, the terrain data, and the temperature and humidity data; a regulating module, configured to control the gas injection or release rate of the bidirectional pressure regulating device of the hybrid target according to the target air pressure value, until the fluctuation amplitude of the actual air pressure value after pressure regulation is less than a preset fluctuation amplitude threshold and the stable duration is greater than a preset duration threshold, and after the control is completed, the posture of the hybrid target is controlled according to the actual air pressure value, the bidirectional pressure regulating device including an air pump and a pressure relief valve; The calculating of the air pressure target value by a target hybrid model after parameter adjustment based on the moving speed data, the hardness level parameter data, the terrain data, and the temperature and humidity data includes: Aligning sampling timestamps of the movement speed data and the hardness level parameter data to generate a multi-feature matrix; performing a second-order polynomial basis expansion on the multi-feature matrix to generate an expanded feature matrix, performing local response normalization on an environmental feature tensor to generate a normalized tensor, and calculating a differential feature vector based on a time series of historical air pressure values, wherein the environmental feature tensor is generated based on the terrain data and the temperature and humidity data; Performing weighted concatenation on the extended feature matrix, the normalized tensor, and the differential feature vector to form a mixed feature vector; The target mixed model after the parameter adjustment is used to perform nonlinear mapping on the mixed characteristic vector to obtain the air pressure target value.

8. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a posture control method for a hybrid target as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, a posture control method of a hybrid target as described in any one of claims 1 to 6 is implemented.

Images