Liquid level sensor device of pesticide preparation tank

Abstract

The invention discloses a liquid level sensor device of a pesticide preparation tank, and relates to the technical field of liquid level measurement. The device comprises a first acquisition module for acquiring internal component parameters of a pesticide preparation tank and working condition parameters of a preparation process, and an initial mounting position determination module for determining an initial mounting position of a radar liquid level sensor according to related parameters; after the second acquisition module acquires liquid level data and an original radar signal, the change amplitude calculation module calculates the change amplitude of the stirring speed, and if the change amplitude exceeds a threshold value, the dynamic adjustment module dynamically adjusts the position of the sensor; and the liquid level compensation value generation module generates a compensation value by using a trained liquid level dynamic compensation model of correlation mapping of the stirring speed and the measurement error in combination with the liquid level data, the original radar signal and the working condition parameters, and finally, the dynamic fusion module fuses the initial liquid level data and the compensation value to obtain an accurate liquid level value. According to the method, the accuracy of liquid level measurement in the pesticide preparation process can be effectively improved by dynamically adjusting the position of the sensor and based on multi-parameter dynamic compensation.

Description

Technical Field

[0001] This invention belongs to the technical field, specifically relating to a liquid level sensor device for pesticide mixing tanks. Background Technology

[0002] In modern agricultural production, the precise formulation of pesticides is a key link in ensuring the effectiveness of pest and disease control, reducing environmental pollution, and lowering pesticide residues in agricultural products. As the core equipment for achieving precise dilution and mixing of pesticides, the real-time and accurate monitoring of the liquid level inside the pesticide mixing tank directly determines the accuracy of the pesticide concentration ratio. If the liquid level measurement deviation is too large, it may result in insufficient effective pesticide components, affecting the control effect, or even cause crop damage or ecological pollution due to excessive concentration.

[0003] Patent application CN119394405A discloses an online real-time liquid level sensor liquid level height measurement method and system, including: real-time fitting of triangular and linear waves output by the liquid level sensor converter; determining the current mode by the trend of the fitted linear wave; if the trend of the fitted linear wave is upward, it is determined that the current mode is filling; if the trend of the fitted linear wave is downward, it is determined that the current mode is discharging; finding the maximum or minimum value of the fitted triangular wave, updating the maximum or minimum value in the liquid level height calculation formula and the number of sensor sections; selecting the appropriate liquid level height calculation formula according to the parity of the number of sensor sections, and calculating the real-time liquid level height; this method updates the maximum and minimum values ​​in the liquid level height calculation formula in real time during the acquisition process, which can effectively reduce the liquid level height calculation error caused by the change of dielectric constant, and improve the accuracy and real-time performance of liquid level height measurement. However, this invention relies solely on fitting and pattern determination of triangular and linear waves to update calculation parameters. Under conditions of stirring and fluctuation, irregular fluctuations in the liquid surface will cause deviations in the fitting of triangular waves, and the dynamic changes in dielectric constant cannot be accurately compensated, resulting in low accuracy of liquid level measurement. Summary of the Invention

[0004] The purpose of this invention is to solve the problem of low accuracy in liquid level measurement during pesticide preparation, and to propose a liquid level sensor device for pesticide preparation tanks.

[0005] This invention provides a pesticide preparation tank level sensor device, the device comprising:

[0006] The first acquisition module is used to acquire the internal component parameters of the pesticide preparation tank and to acquire the operating condition parameter data of the pesticide preparation process in real time; the operating condition parameter data includes stirring speed, liquid concentration and liquid temperature;

[0007] An initial installation position determination module is used to determine the initial installation position of the radar level sensor based on the internal component parameters, the stirring speed, and the liquid concentration.

[0008] The second acquisition module is used to acquire the liquid level data and raw radar signal from the radar liquid level sensor.

[0009] The variation amplitude calculation module is used to calculate the variation amplitude of the stirring speed based on the stirring speed to obtain the variation amplitude value;

[0010] The dynamic adjustment module is used to dynamically adjust the initial installation position of the radar liquid level sensor to the target installation position if the change amplitude value exceeds the preset change amplitude threshold, and then return to the second acquisition module.

[0011] The liquid level compensation value generation module is used to input the liquid level data, the original radar signal and the operating condition parameter data into the trained liquid level dynamic compensation model to obtain the liquid level compensation value; the liquid level dynamic compensation model establishes a correlation mapping between the stirring speed and the radar liquid level gauge measurement error;

[0012] The dynamic fusion module is used to dynamically fuse the initial liquid level data with the liquid level compensation value to obtain an accurate liquid level value.

[0013] Optionally, the initial installation location determination module includes:

[0014] The candidate installation area acquisition module is used to determine the interference area of ​​internal components that needs to be excluded when installing the radar liquid level sensor based on the parameters of the internal components, and thus obtain the candidate installation area.

[0015] An interference comprehensive coefficient generation module is used to perform a correlation analysis on the stirring speed and the liquid concentration to obtain an interference comprehensive coefficient;

[0016] The offset calculation module is used to calculate the installation position offset value based on the interference synthesis coefficient of the radar liquid level sensor in the candidate installation area.

[0017] An available installation area selection module is used to determine the initial installation position of the radar level sensor based on the available installation area and the installation position offset value.

[0018] Optionally, the interference synthesis coefficient generation module includes:

[0019] pass The interference synthesis coefficient is obtained; where ε(c) is a function of the dielectric constant corresponding to liquid concentration c and , The dielectric constant of the pure solvent is and Here, w(v) is the concentration effect coefficient calibrated in the experiment, and w(v) is the turbulent kinematic viscosity function corresponding to the stirring speed v. , This represents the viscosity coefficient corresponding to the stirring speed v. Let L be the liquid density, and L be the characteristic length of radar wave propagation, determined experimentally.

[0020] Optionally, the dynamic adjustment module:

[0021] The feature parameter extraction module is used to extract liquid surface fluctuation feature parameters from the change amplitude value to obtain a set of fluctuation feature parameters; the liquid surface fluctuation feature parameters include fluctuation frequency, fluctuation amplitude, and fluctuation energy.

[0022] The installation location evaluation function generation module is used to model the correlation between the initial installation location and the measurement error of the radar liquid level sensor based on the fluctuation characteristic parameter set to obtain the installation location evaluation function;

[0023] The optimization solution module is used to optimize the installation location evaluation function using the gradient descent method to obtain the target installation location;

[0024] A sensor moving module is used to move the radar level sensor to the target installation position.

[0025] Optionally, the installation location evaluation function generation module includes:

[0026] The installation location evaluation function is: ;in, This is a comprehensive evaluation value for the installation location of the radar level sensor. and Here, f is the weighting coefficient, f is the frequency of liquid surface ripples, A is the amplitude of liquid surface ripples, and E is the energy of liquid surface ripples. This refers to the distance from the installation location of the radar level sensor to the liquid surface. This is the distance threshold.

[0027] Optionally, the training process of the preset liquid level dynamic compensation model includes:

[0028] Simultaneously acquire historical raw radar signals, operating parameters, and liquid level data to construct a spatiotemporally aligned training sample set;

[0029] Spatial features and parameter correlation features are obtained by parallel processing of the original radar signal and the operating parameters;

[0030] The spatial features and the parameter-related features are interactively fused to obtain environmental adaptive fusion features;

[0031] By improving the LSTM model, a time-series dependency analysis of the environmental adaptive fusion features is performed to obtain a dynamic error feature sequence;

[0032] A local compensation sequence is obtained by constructing a multi-level strategy from the dynamic error feature sequence;

[0033] The robust compensation parameters are obtained by progressively optimizing the local compensation sequence.

[0034] The performance of the robust compensation parameters was verified using a test set to obtain the trained dynamic liquid level compensation model.

[0035] Optionally, parallel processing of the original radar signal and the operating parameters to obtain spatial features and parameter correlation features includes:

[0036] The original radar signal is subjected to intrinsic mode separation to obtain a set of signal components;

[0037] An enhanced feature set is obtained by reconstructing the operating parameters using an operating condition parameter reconstruction algorithm.

[0038] A correlation matrix is ​​obtained by performing cross-modal correlation analysis on the signal component set and the enhanced feature set;

[0039] Feature selection parameters are obtained by adjusting the weight coefficients of the relationship matrix based on the changes in the operating parameters.

[0040] Based on the feature selection parameters, dual-channel feature extraction is performed on the signal component set and the enhanced feature set to obtain spatial features and parameter correlation features.

[0041] Optionally, improvements to the LSTM model structure include:

[0042] The gating unit includes a forget gate and an output gate. The forget gate is used to filter historical cell states, and the output gate is used to control the output of the current cell state.

[0043] The cell state unit is used to receive the historical cell state after being filtered by the forget gate at the previous moment. At the same time, it receives input information and directly participates in the update of the cell state. The integration path of new information does not go through the on / off control of the independent input gate.

[0044] The hidden state unit is used to receive the output of the output gate and the cell state activated by the tanh function. The current hidden state is obtained by multiplying the two and then passed to the next time step. The input information and the hidden state of the previous time step are first used to calculate the forget gate and the output gate. At the same time, the input information directly participates in the update of the cell state. The update logic of the cell state is the sum of the forgotten historical state and the new candidate state that has not been gated by the input. The new candidate state is generated by the input information and the hidden state of the previous time step through the tanh function.

[0045] Optionally, the dynamic fusion includes:

[0046] The spectral feature extraction module is used to extract the spectral features of the stirring speed to obtain the liquid surface fluctuation feature spectrum;

[0047] The difference calculation module is used to calculate the difference between the liquid concentration, liquid temperature and stirring speed to obtain the change in concentration, change in temperature and change in stirring speed;

[0048] The attenuation quantification module is used to quantify the attenuation of the radar wave propagation path based on the concentration change to obtain the signal propagation characteristic curve.

[0049] The feature layer fusion module is used to perform feature layer fusion of the liquid surface wave feature spectrum and the signal propagation characteristic curve to obtain a fused feature vector;

[0050] A dynamic weighting module is used to dynamically weight the fused feature vector according to the liquid temperature change to obtain an environment-adaptive feature matrix.

[0051] The feature-guided fusion module is used to perform feature-guided fusion on the initial liquid level data and liquid level compensation value according to the environmental adaptive feature matrix to obtain the intermediate fused liquid level value.

[0052] The phase calibration level value generation module is used to perform phase compensation on the intermediate fusion level value according to the change in stirring speed to obtain the phase calibration level value;

[0053] The precise liquid level value generation module is used to obtain the precise liquid level value by dynamically smoothing the phase calibration liquid level value through a multi-scale sliding window.

[0054] Optionally, the dynamic weighting module includes:

[0055] The segmentation module is used to segment the thermal disturbance condition range according to the direction and magnitude of the liquid temperature change; the thermal disturbance condition range includes a rapid heating range and a rapid cooling range;

[0056] The weight allocation strategy generation module is used to apply a corresponding weight allocation strategy to the thermal disturbance condition range; the rapid heating range adopts a strategy of suppressing liquid surface fluctuation characteristic weights, and the rapid cooling range adopts a strategy of enhancing medium characteristic characteristic weights.

[0057] A linear scaling module is used to linearly scale the strength of the weight allocation strategy based on the absolute value of the liquid temperature change to obtain the weight allocation coefficient.

[0058] The environment adaptive feature matrix generation module is used to reconstruct the fused feature vector by weighting according to the weight allocation coefficients to obtain the environment adaptive feature matrix.

[0059] The beneficial effects of this invention are as follows: This invention proposes a liquid level sensor device for pesticide preparation tanks. A first acquisition module acquires parameters of the internal components of the pesticide preparation tank and the operating conditions during the preparation process. An initial installation position determination module determines the initial installation position of the radar liquid level sensor based on these parameters. A second acquisition module acquires liquid level data and the original radar signal. A change amplitude calculation module calculates the change amplitude of the stirring speed. If the change exceeds a threshold, a dynamic adjustment module dynamically adjusts the sensor position. A liquid level compensation value generation module uses a trained dynamic compensation model that correlates stirring speed with measurement error, combined with liquid level data, the original radar signal, and operating conditions, to generate a compensation value. Finally, a dynamic fusion module fuses the initial liquid level data and the compensation value to obtain an accurate liquid level value. This method, through dynamic sensor position adjustment and multi-parameter-based dynamic compensation, effectively improves the accuracy of liquid level measurement during pesticide preparation. Attached Figure Description

[0060] The present invention will now be further described with reference to the accompanying drawings.

[0061] Figure 1 This is a schematic diagram of the structure of a pesticide preparation tank level sensor device provided in an embodiment of the present invention;

[0062] Figure 2 This is a structural diagram of a pesticide preparation tank level sensor device provided in an embodiment of the present invention. Detailed Implementation

[0063] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0064] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] This invention provides a liquid level sensor device for pesticide mixing tanks. See also... Figure 1 , Figure 1 A schematic diagram of a pesticide preparation tank level sensor device provided in an embodiment of the present invention includes:

[0066] The first acquisition module is used to acquire the internal component parameters of the pesticide preparation tank and to acquire the operating condition parameter data of the pesticide preparation process in real time.

[0067] The initial installation location determination module is used to determine the initial installation location of the radar level sensor based on internal component parameters, stirring speed, and liquid concentration.

[0068] The second acquisition module is used to acquire the liquid level data and raw radar signal from the radar liquid level sensor.

[0069] The variation amplitude calculation module is used to calculate the variation amplitude of the stirring speed based on the stirring speed and obtain the variation amplitude value.

[0070] The dynamic adjustment module is used to dynamically adjust the initial installation position of the radar liquid level sensor to the target installation position if the change value exceeds the preset change value threshold, and then return to the second acquisition module.

[0071] The liquid level compensation value generation module is used to input liquid level data, raw radar signals and operating condition parameter data into the trained liquid level dynamic compensation model to obtain liquid level compensation values.

[0072] The dynamic fusion module is used to dynamically fuse the initial liquid level data with the liquid level compensation value to obtain an accurate liquid level value.

[0073] The operating parameters include stirring speed, liquid concentration, and liquid temperature.

[0074] A dynamic compensation model for liquid level is established to map the correlation between stirring speed and radar level gauge measurement error.

[0075] Based on the pesticide preparation tank liquid level sensor device provided in this embodiment of the invention, the following steps are taken: A first acquisition module acquires the internal component parameters and operational parameters of the pesticide preparation tank; an initial installation position determination module determines the initial installation position of the radar liquid level sensor based on these parameters; a second acquisition module acquires the liquid level data and the original radar signal; a change amplitude calculation module calculates the change amplitude of the stirring speed; if it exceeds a threshold, a dynamic adjustment module dynamically adjusts the sensor position; a liquid level compensation value generation module uses a trained dynamic compensation model that correlates stirring speed with measurement error, combined with the liquid level data, the original radar signal, and operational parameters, to generate a compensation value; finally, a dynamic fusion module fuses the initial liquid level data and the compensation value to obtain an accurate liquid level value. This method, through dynamic adjustment of the sensor position and multi-parameter-based dynamic compensation, can effectively improve the accuracy of liquid level measurement during pesticide preparation.

[0076] In one implementation, the stirring speed is collected by a speed sensor installed on the impeller shaft; the liquid concentration is obtained by a capacitive concentration sensor installed at 1 / 5 of the liquid level in the tank, at the edge of the turbulent zone; the liquid temperature is collected redundantly by a thermocouple sensor installed in the stable zone in the middle of the tank at 1 / 3 of the liquid level; the raw radar signal is acquired by the radar level sensor with its built-in acquisition unit in the candidate installation area on the top of the tank; the collected stirring speed, liquid concentration, liquid temperature and raw radar signal are transmitted to a computer in real time through a compatible data transmission line to provide basic data for subsequent accurate liquid level calculation.

[0077] In one embodiment, the initial installation location determination module includes:

[0078] The candidate installation area acquisition module is used to determine the interference area of ​​internal components that needs to be excluded when installing the radar liquid level sensor based on the parameters of the internal components, and thus obtain the candidate installation area.

[0079] The interference comprehensive coefficient generation module is used to perform correlation analysis on stirring speed and liquid concentration to obtain the interference comprehensive coefficient;

[0080] The offset calculation module is used to calculate the installation position offset value of the radar liquid level sensor in the candidate installation area based on the interference synthesis coefficient.

[0081] An available installation area selection module is used to determine the initial installation position of the radar level sensor based on the available installation area and the installation position offset value.

[0082] In one implementation, the interference comprehensive coefficient generation module reconstructs the phase space of the time-series data of stirring speed and liquid concentration, analyzes the correlation between their trajectories in phase space, and constructs a coupled model that integrates the chaotic fluctuation characteristics of the liquid surface caused by stirring speed and the changes in the electromagnetic properties of the medium caused by liquid concentration, based on the influence model of liquid concentration on the dielectric constant of the medium. Through this model, a nonlinear correlation analysis is performed on stirring speed and liquid concentration to generate a chaotic interference comprehensive coefficient that can comprehensively characterize the interference of both on radar measurement. This coefficient is no longer a simple linear superposition, but can accurately reflect the degree of nonlinear interference under the coupling of multiple factors, providing a precise quantitative basis with chaotic characteristics for subsequent interference compensation.

[0083] In one implementation, the offset calculation module employs the ant colony optimization algorithm from biomimetic optimization algorithms. Using the interference synthesis coefficient as the pheromone source, the position offset of the radar level sensor within the candidate installation area is used as the optimization path for the ant colony. Simultaneously, a fitness function is constructed by introducing constraints on radar measurement accuracy, allowing the ant colony to intelligently optimize within the candidate installation area and calculate the position offset that minimizes radar measurement error, thus obtaining the installation position offset value. This offset calculation method based on biomimetic intelligent optimization realizes a shift from passive coefficient-based calculation to proactive intelligent optimization with accuracy as the goal, significantly improving the optimization effect of the offset value on radar measurement accuracy.

[0084] In one embodiment, the interference synthesis coefficient generation module includes:

[0085] pass The interference synthesis coefficient is obtained; where ε(c) is a function of the dielectric constant corresponding to liquid concentration c and , The dielectric constant of the pure solvent is and Here, w(v) is the concentration effect coefficient calibrated in the experiment, and w(v) is the turbulent kinematic viscosity function corresponding to the stirring speed v. , This represents the viscosity coefficient corresponding to the stirring speed v. Let L be the liquid density, and L be the characteristic length of radar wave propagation, determined experimentally.

[0086] In one implementation, the interference comprehensive coefficient is calculated by coupling the dielectric constant function corresponding to the liquid concentration with the turbulent kinematic viscosity function corresponding to the stirring speed. A model is constructed based on the radar wave propagation characteristic length, which can accurately quantify the influence of liquid concentration on radar wave propagation characteristics and the effect of stirring speed on the turbulent state of the liquid surface. This achieves comprehensive quantification of multi-physics interference factors, breaking through the limitations of traditional single-factor or linear superposition interference assessment. The interference comprehensive coefficient can more realistically reflect the degree of nonlinear interference in radar measurement under complex working conditions in pesticide preparation tanks. This provides a scientific and accurate quantitative basis for the precise optimization of the radar liquid level sensor installation position and the dynamic compensation of measurement errors, thereby significantly improving the accuracy and stability of radar liquid level measurement under dynamic and complex working conditions.

[0087] In one embodiment, the dynamic adjustment module includes:

[0088] The feature parameter extraction module is used to extract liquid surface fluctuation feature parameters from the change amplitude value to obtain a set of fluctuation feature parameters; the liquid surface fluctuation feature parameters include fluctuation frequency, fluctuation amplitude and fluctuation energy.

[0089] The installation location evaluation function generation module is used to model the correlation between the initial installation location and the measurement error of the radar liquid level sensor based on the set of fluctuation characteristic parameters to obtain the installation location evaluation function.

[0090] The optimization solution module is used to optimize the installation location evaluation function using the gradient descent method to obtain the target installation location;

[0091] The sensor moving module is used to move the radar level sensor to the target installation location.

[0092] In one implementation, the formula for calculating the fluctuation frequency is: Where g is the acceleration due to gravity. The equivalent elastic modulus is obtained by coupling the stirring speed v and the liquid viscosity μ. The height of the liquid layer. This is the correction factor for the stirring speed relative to the frequency. and The formula for calculating the fluctuation amplitude is obtained from experimental fitting. ,in, Let x be the incident wave height, and x be the horizontal coordinate. Let λ be the incident wavelength in deep water, t be time, T be the period of the reflected wave, and η(c) be a correction function of the liquid concentration c on the amplitude, obtained through experimental fitting; the formula for calculating the wave energy is... ,in, Let ρ be the liquid density, g be the gravitational acceleration, A be the amplitude of the wave, and λ be the wavelength.

[0093] In one implementation, a global chaotic search is performed in the solution space of the installation position evaluation function using the ergodicity of the chaotic mapping to avoid getting trapped in local optima. Simultaneously, a fractal iterative algorithm is introduced to refine the fractal dimension of the potential optimal region obtained from the chaotic search, accurately identifying the nonlinear correlation details between position and error. Furthermore, a self-learning neural synapse of the operating condition algorithm is implanted, transforming the dynamic changes in stirring speed and liquid concentration into synaptic weight update signals within the algorithm. This allows the optimization process to adapt to the dynamic characteristics of liquid surface fluctuations in real time. The final target installation position is not a static minimum error point, but a dynamically robust optimal solution that continuously achieves optimal measurement accuracy under dynamic operating conditions. This represents a creative leap from conventional numerical optimization to chaotic fractals and neural self-learning.

[0094] In one embodiment, the installation location evaluation function generation module includes:

[0095] The installation location evaluation function is: ;in, This is a comprehensive evaluation value for the installation location of the radar level sensor. and Here, f is the weighting coefficient, f is the frequency of liquid surface ripples, A is the amplitude of liquid surface ripples, and E is the energy of liquid surface ripples. This refers to the distance from the installation location of the radar level sensor to the liquid surface. This is the distance threshold.

[0096] In one implementation, the installation location evaluation function deeply integrates several core factors, including the frequency, amplitude, and energy of liquid surface fluctuations, as well as the distance from the radar installation location to the liquid surface, using weighting coefficients. , It rationally allocates the influence weight of each factor. Unlike traditional methods that consider only a single factor or simple linear relationships, it precisely quantifies the combined effect of the dynamic fluctuations of the liquid surface and the spatial parameters of the installation location on radar measurement accuracy, thereby evaluating the obtained... It can truly reflect the measurement adaptability of the radar liquid level sensor in the face of dynamic liquid surface conditions at the current installation position, providing a reliable quantitative standard for finding the optimal installation position in the future, thereby effectively improving the measurement accuracy and stability of the radar under complex liquid surface fluctuation conditions, and ensuring that the sensor installation position is always the most conducive to measurement.

[0097] In one embodiment, the training process of the preset liquid level dynamic compensation model includes:

[0098] Simultaneously acquire historical raw radar signals, operating parameters, and liquid level data to construct a spatiotemporally aligned training sample set;

[0099] Spatial features and parameter correlation features are obtained by parallel processing of the original radar signal and operating parameters;

[0100] Interactive fusion of spatial features and parameter correlation features yields environmentally adaptive fusion features;

[0101] By improving the LSTM model, a time-series dependency analysis of the adaptive fusion features of the environment is performed to obtain a dynamic error feature sequence;

[0102] A local compensation sequence is obtained by constructing a multi-level strategy from the dynamic error feature sequence;

[0103] Robust compensation parameters are obtained by incrementally optimizing the local compensation sequence;

[0104] The performance of the robust compensation parameters was verified using a test set to obtain the trained dynamic liquid level compensation model.

[0105] In one implementation, a heterogeneous parallel computing module is built. On the one hand, the original radar signal is decomposed into sub-components such as multipath reflection and clutter scattering in the spatial domain using a wavefront decomposition algorithm, and the spatial distribution characteristics of radar wave propagation, such as reflection point density and spatial gradient of scattering intensity, are extracted. On the other hand, the operating parameters are analyzed using a multi-parameter coupling matrix to construct a coupling correlation model between the stirring turbulence intensity and the change of liquid dielectric constant, and the nonlinear correlation characteristics between parameters are extracted. Through this parallel feature extraction, the spatial physical characteristics of radar measurement are accurately captured, and the multi-field coupling correlation of operating parameters is quantified. This achieves parallel deconstruction of multi-physics field features from signal to parameters, providing dual-dimensional feature support with both spatial physical meaning and operating condition coupling mechanism for subsequent fusion.

[0106] In one implementation, the local compensation sequence is first input into a digital twin simulation platform to simulate the compensation effect under different working conditions. The basic compensation parameters are then preliminarily optimized using a genetic algorithm. Subsequently, the basic compensation parameters are deployed to an actual measurement system, and residual data after compensation is collected under real working conditions. The residual data is then fed back to the optimization module, and the weights and thresholds of the compensation parameters are iteratively updated based on a dynamic constraint adjustment strategy using Bayesian optimization. This closed-loop process of simulation optimization, actual measurement verification, residual feedback, and parameter iteration is repeatedly executed, ultimately yielding robust compensation parameters that maintain stable compensation accuracy under various extreme dynamic working conditions. This achieves a progressive robustness enhancement throughout the entire process, from theoretical simulation optimization to actual scenario verification iteration, ensuring that the compensation parameters have continuous and reliable error compensation capabilities in complex dynamic environments.

[0107] In one embodiment, parallel processing of the raw radar signal and operating parameters to obtain spatial features and parameter correlation features includes:

[0108] The original radar signal is subjected to intrinsic mode separation to obtain the signal component set;

[0109] An enhanced feature set is obtained by reconstructing the features of the operating parameters using an operating parameter reconstruction algorithm.

[0110] A correlation matrix is ​​obtained by performing correlation analysis on the signal component set and the enhanced feature set through cross-modal correlation analysis;

[0111] Feature selection parameters are obtained by adjusting the weight coefficients of the relationship matrix based on changes in operating parameters.

[0112] Based on the feature selection parameters, dual-channel feature extraction is performed on the signal component set and the enhanced feature set to obtain spatial features and parameter correlation features.

[0113] In one implementation, an adaptive intrinsic mode decomposition algorithm is used to adaptively decompose multiple physically meaningful signal components based on the time-frequency characteristics of the original radar signal. Each component corresponds to a different spatial propagation or reflection mechanism in the radar measurement scenario, realizing the creative separation from a single signal to multiple physical mechanism signal components, and providing a basic unit with clear physical meaning for subsequent spatial feature extraction.

[0114] In one implementation, a dynamic mapping neural unit for operating conditions is constructed. This unit has a pre-trained correlation model built into the operating condition parameters such as stirring speed and liquid concentration and the weight coefficients of the relation matrix. This model is trained with a large amount of experimental data under operating conditions and can quantify the influence of different operating condition changes on the correlation strength between radar signals and operating condition parameters. When changes in operating condition parameters are detected in real time, such as a sudden increase in stirring speed or a step change in liquid concentration, the neural unit will convert the change in operating conditions into the weight adjustment value of the corresponding correlation item in the relation matrix according to the pre-trained model. This generates feature selection parameters that can dynamically change with the operating conditions and accurately match the current correlation strength, realizing the shift from static fixed weights to dynamic adaptive weight adjustment driven by operating conditions, and ensuring that the feature selection always remains consistent with the multimodal correlation characteristics under real-time operating conditions.

[0115] In one embodiment, the structure of the improved LSTM model includes:

[0116] The gating unit includes a forget gate and an output gate. The forget gate is used to filter historical cell states, and the output gate is used to control the output of the current cell state.

[0117] The cell state unit is used to receive the historical cell state after being filtered by the forget gate at the previous moment. At the same time, it receives input information and directly participates in the update of the cell state. The integration path of new information does not go through the on / off control of the independent input gate.

[0118] The hidden state unit is used to receive the output of the output gate and the cell state activated by the tanh function. The current hidden state is obtained by multiplying the two and then passed to the next time step. The input information and the hidden state of the previous time step are first used to calculate the forget gate and the output gate. At the same time, the input information directly participates in the update of the cell state. The update logic of the cell state is the sum of the forgotten historical state and the new candidate state that has not been gated by the input. The new candidate state is generated by the input information and the hidden state of the previous time step through the tanh function.

[0119] In one implementation, structurally, only the forget gate and output gate are retained while the independent input gate is eliminated. This simplifies the number of model parameters, reducing the computational complexity of training and inference, and avoids the redundant computational overhead that traditional input gates may bring, making it particularly suitable for resource-constrained edge computing scenarios. In terms of information processing, new information directly participates in cell state updates, and the switch-like control without input gates makes the cell state's response to real-time inputs more direct and efficient. At the same time, the design of forgotten historical states and ungated new candidate states in the cell state update logic enhances the ability to capture current input features while ensuring basic filtering of historical dependency information, thus improving the model's sensitivity to short-term temporal fluctuations. In terms of performance balance, the hidden state unit generates the result through the product operation of the output gate and the tanh-activated cell state. This inherits the basic modeling ability of LSTM for long-term dependencies and achieves a balance between computational efficiency and prediction accuracy through structural simplification.

[0120] In one embodiment, dynamic fusion includes:

[0121] The spectral feature extraction module is used to extract the spectral features of the stirring speed to obtain the liquid surface fluctuation feature spectrum;

[0122] The difference calculation module is used to calculate the difference between liquid concentration, liquid temperature and stirring speed to obtain the change in concentration, change in temperature and change in stirring speed;

[0123] The attenuation quantization module is used to quantify the attenuation of the radar wave propagation path based on the concentration change to obtain the signal propagation characteristic curve.

[0124] The feature layer fusion module is used to perform feature layer fusion of the liquid surface wave feature spectrum and the signal propagation characteristic curve to obtain a fused feature vector.

[0125] The dynamic weighting module is used to dynamically weight the fused feature vector according to the change in liquid temperature to obtain an environment-adaptive feature matrix.

[0126] The feature-guided fusion module is used to perform feature-guided fusion of the initial liquid level data and the liquid level compensation value based on the environmental adaptive feature matrix to obtain the intermediate fused liquid level value.

[0127] The phase calibration level value generation module is used to obtain the phase calibration level value by performing phase compensation on the intermediate fused liquid level value according to the change in stirring speed.

[0128] The precise liquid level value generation module is used to obtain a precise liquid level value by dynamically smoothing the phase calibration liquid level value through a multi-scale sliding window.

[0129] In one implementation, a nonlinear function of the liquid dielectric constant as a function of concentration is first obtained through experimental fitting. ,in The dielectric constant of the pure solvent is b are experimental calibration coefficients; then, based on the correlation formula between dielectric constant and radar wave attenuation coefficient... Where k is a frequency-dependent constant set by technicians, and ω is the radar operating frequency, the concentration change is substituted into the calculation to obtain the real-time attenuation coefficient; finally, a signal propagation characteristic curve that can continuously reflect the attenuation of radar waves on the propagation path is generated, realizing the full physical link quantization from concentration change to radar wave attenuation, providing a signal transmission attenuation characterization with both theoretical accuracy and practical physical significance for subsequent feature fusion.

[0130] In one implementation, guided by an environmental adaptive feature matrix, the initial liquid level data and liquid level compensation values ​​are fused at the feature level to obtain an intermediate fused liquid level value. This achieves an innovative leap from data-level fusion to feature-to-data dual-driven fusion. Feature-level fusion can adjust the fusion weight of the compensation value according to the severity of liquid surface fluctuations in the feature matrix, and optimize the credibility weight of the initial data according to the degree of signal transmission disturbance. Based on the change in stirring speed and combined with the disturbance mechanism of stirring turbulence on radar wave phase, phase compensation is performed on the intermediate fused liquid level value. The resulting phase-calibrated liquid level value can accurately eliminate the phase error caused by the change in stirring speed, achieving a creative breakthrough from simple phase compensation to dynamic coupling compensation from stirring speed to phase.

[0131] In one embodiment, the dynamic weighting module includes:

[0132] The partitioning module is used to divide the thermal disturbance condition range according to the direction and magnitude of the liquid temperature change; the thermal disturbance condition range includes a rapid heating range and a rapid cooling range;

[0133] The weight allocation strategy generation module is used to apply corresponding weight allocation strategies to the thermal disturbance condition range; the rapid heating range adopts a strategy of suppressing liquid surface fluctuation characteristic weights, and the rapid cooling range adopts a strategy of enhancing medium characteristic characteristic weights.

[0134] The linear scaling module is used to linearly scale the strength of the weight allocation strategy based on the absolute value of the liquid temperature change to obtain the weight allocation coefficients.

[0135] The environment adaptive feature matrix generation module is used to reconstruct the fused feature vector by weighting it according to the weight allocation coefficients to obtain the environment adaptive feature matrix.

[0136] In one implementation, the dynamic weighting module divides the thermal disturbance condition range, formulates a targeted weight allocation strategy, linearly scales the weight intensity based on the temperature change, and reconstructs the feature matrix. This enables precise adaptation to the thermal disturbance conditions caused by liquid temperature changes, allowing the feature matrix to dynamically adjust the weight ratio of liquid surface fluctuations and medium characteristics according to different scenarios of rapid heating or cooling. This enhances the adaptability of the features to the temperature disturbance environment and lays a feature foundation that can dynamically change with temperature for subsequent accurate liquid level calculation.

[0137] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A liquid level sensor device for a pesticide preparation tank, characterized in that, The device includes: The first acquisition module is used to acquire the internal component parameters of the pesticide preparation tank and to acquire the operating condition parameter data of the pesticide preparation process in real time; the operating condition parameter data includes stirring speed, liquid concentration and liquid temperature; An initial installation position determination module is used to determine the initial installation position of the radar level sensor based on the internal component parameters, the stirring speed, and the liquid concentration. The second acquisition module is used to acquire the liquid level data and raw radar signal from the radar liquid level sensor. The variation amplitude calculation module is used to calculate the variation amplitude of the stirring speed based on the stirring speed to obtain the variation amplitude value; The dynamic adjustment module is used to dynamically adjust the initial installation position of the radar liquid level sensor to the target installation position if the change amplitude value exceeds the preset change amplitude threshold, and then return to the second acquisition module. The liquid level compensation value generation module is used to input the liquid level data, the original radar signal and the operating condition parameter data into the trained liquid level dynamic compensation model to obtain the liquid level compensation value; the liquid level dynamic compensation model establishes a correlation mapping between the stirring speed and the radar liquid level gauge measurement error; The dynamic fusion module is used to dynamically fuse the initial liquid level data with the liquid level compensation value to obtain an accurate liquid level value.

2. The pesticide preparation tank level sensor device according to claim 1, characterized in that, The initial installation location determination module includes: The candidate installation area acquisition module is used to determine the interference area of ​​internal components that needs to be excluded when installing the radar liquid level sensor based on the parameters of the internal components, and thus obtain the candidate installation area. An interference comprehensive coefficient generation module is used to perform a correlation analysis on the stirring speed and the liquid concentration to obtain an interference comprehensive coefficient; The offset calculation module is used to calculate the installation position offset value based on the interference synthesis coefficient of the radar liquid level sensor in the candidate installation area. An available installation area selection module is used to determine the initial installation position of the radar level sensor based on the available installation area and the installation position offset value.

3. The pesticide preparation tank level sensor device according to claim 2, characterized in that, The interference synthesis coefficient generation module includes: pass The interference synthesis coefficient is obtained; where ε(c) is a function of the dielectric constant corresponding to liquid concentration c and , The dielectric constant of the pure solvent is and Here, w(v) is the concentration effect coefficient calibrated in the experiment, and w(v) is the turbulent kinematic viscosity function corresponding to the stirring speed v. , This represents the viscosity coefficient corresponding to the stirring speed v. Let L be the liquid density, and L be the characteristic length of radar wave propagation, determined experimentally.

4. The pesticide preparation tank level sensor device according to claim 1, characterized in that, The dynamic adjustment module: The feature parameter extraction module is used to extract liquid surface fluctuation feature parameters from the change amplitude value to obtain a set of fluctuation feature parameters; the liquid surface fluctuation feature parameters include fluctuation frequency, fluctuation amplitude, and fluctuation energy. The installation location evaluation function generation module is used to model the correlation between the initial installation location and the measurement error of the radar liquid level sensor based on the fluctuation characteristic parameter set to obtain the installation location evaluation function; The optimization solution module is used to optimize the installation location evaluation function using the gradient descent method to obtain the target installation location; A sensor moving module is used to move the radar level sensor to the target installation position.

5. The pesticide preparation tank level sensor device according to claim 4, characterized in that, The installation location evaluation function generation module includes: The installation location evaluation function is: ;in, This is a comprehensive evaluation value for the installation location of the radar level sensor. and Here, f is the weighting coefficient, f is the frequency of liquid surface ripples, A is the amplitude of liquid surface ripples, and E is the energy of liquid surface ripples. This refers to the distance from the installation location of the radar level sensor to the liquid surface. This is the distance threshold.

6. The pesticide preparation tank level sensor device according to claim 1, characterized in that, The training process of the preset liquid level dynamic compensation model includes: Simultaneously acquire historical raw radar signals, operating parameters, and liquid level data to construct a spatiotemporally aligned training sample set; Spatial features and parameter correlation features are obtained by parallel processing of the original radar signal and the operating parameters; The spatial features and the parameter-related features are interactively fused to obtain environmental adaptive fusion features; By improving the LSTM model, a time-series dependency analysis of the environmental adaptive fusion features is performed to obtain a dynamic error feature sequence; A local compensation sequence is obtained by constructing a multi-level strategy from the dynamic error feature sequence; The robust compensation parameters are obtained by progressively optimizing the local compensation sequence. The performance of the robust compensation parameters was verified using a test set to obtain the trained dynamic liquid level compensation model.

7. The pesticide preparation tank level sensor device according to claim 6, characterized in that, The spatial features and parameter correlation features obtained by parallel processing of the original radar signal and the operating parameters include: The original radar signal is subjected to intrinsic mode separation to obtain a set of signal components; An enhanced feature set is obtained by reconstructing the operating parameters using an operating condition parameter reconstruction algorithm. A correlation matrix is ​​obtained by performing cross-modal correlation analysis on the signal component set and the enhanced feature set; Feature selection parameters are obtained by adjusting the weight coefficients of the relationship matrix based on the changes in the operating parameters. Based on the feature selection parameters, dual-channel feature extraction is performed on the signal component set and the enhanced feature set to obtain spatial features and parameter correlation features.

8. The pesticide preparation tank level sensor device according to claim 7, characterized in that, The improved structure of the LSTM model includes: The gating unit includes a forget gate and an output gate. The forget gate is used to filter historical cell states, and the output gate is used to control the output of the current cell state. The cell state unit is used to receive the historical cell state after being filtered by the forget gate at the previous moment. At the same time, it receives input information and directly participates in the update of the cell state. The integration path of new information does not go through the on / off control of the independent input gate. The hidden state unit is used to receive the output of the output gate and the cell state activated by the tanh function. The current hidden state is obtained by multiplying the two and then passed to the next time step. The input information and the hidden state of the previous time step are first used to calculate the forget gate and the output gate. At the same time, the input information directly participates in the update of the cell state. The update logic of the cell state is the sum of the forgotten historical state and the new candidate state that has not been gated by the input. The new candidate state is generated by the input information and the hidden state of the previous time step through the tanh function.

9. The pesticide preparation tank level sensor device according to claim 1, characterized in that, The dynamic fusion includes: The spectral feature extraction module is used to extract the spectral features of the stirring speed to obtain the liquid surface fluctuation feature spectrum; The difference calculation module is used to calculate the difference between the liquid concentration, liquid temperature and stirring speed to obtain the change in concentration, change in temperature and change in stirring speed; The attenuation quantification module is used to quantify the attenuation of the radar wave propagation path based on the concentration change to obtain the signal propagation characteristic curve. The feature layer fusion module is used to perform feature layer fusion of the liquid surface wave feature spectrum and the signal propagation characteristic curve to obtain a fused feature vector; A dynamic weighting module is used to dynamically weight the fused feature vector according to the liquid temperature change to obtain an environment-adaptive feature matrix. The feature-guided fusion module is used to perform feature-guided fusion on the initial liquid level data and liquid level compensation value according to the environmental adaptive feature matrix to obtain the intermediate fused liquid level value. The phase calibration level value generation module is used to perform phase compensation on the intermediate fusion level value according to the change in stirring speed to obtain the phase calibration level value; The precise liquid level value generation module is used to obtain the precise liquid level value by dynamically smoothing the phase calibration liquid level value through a multi-scale sliding window.

10. The pesticide preparation tank level sensor device according to claim 9, characterized in that, The dynamic weighting module includes: The segmentation module is used to segment the thermal disturbance condition range according to the direction and magnitude of the liquid temperature change; the thermal disturbance condition range includes a rapid heating range and a rapid cooling range; The weight allocation strategy generation module is used to apply a corresponding weight allocation strategy to the thermal disturbance condition range; the rapid heating range adopts a strategy of suppressing liquid surface fluctuation characteristic weights, and the rapid cooling range adopts a strategy of enhancing medium characteristic characteristic weights. A linear scaling module is used to linearly scale the strength of the weight allocation strategy based on the absolute value of the liquid temperature change to obtain the weight allocation coefficient. The environment adaptive feature matrix generation module is used to reconstruct the fused feature vector by weighting according to the weight allocation coefficients to obtain the environment adaptive feature matrix.

Images