An Intelligent Load Prediction and Regulation Method in a Dynamic Weighing System

Abstract

The present invention relates to the field of dynamic weighing technology, and discloses an intelligent load prediction and adjustment method in a dynamic weighing system, including: separating the weighing signal into a spatial domain static load feature (pressure point distribution sparsity, center of gravity geometric vector) and a time domain dynamic inertial interference component (acceleration coupling noise) through spatio-temporal double-domain decoupling, eliminating the spectral aliasing problem of traditional filtering algorithms from the signal source; predicting the future interference trend based on the real-time motion state of the vehicle, generating a feed-forward compensation signal through a pre-trained dynamic basis function library, and actively canceling the dynamic error; combining the spatial consistency detection of the sensor array and lightweight online learning to dynamically correct the decoupling model parameters. The present invention combines spatial geometric analysis with kinematic feed-forward prediction, breaks out of the traditional signal processing framework, realizes the separation of error sources and real-time compensation, and significantly improves the dynamic weighing accuracy under transient conditions such as sudden stops and speed changes.

Description

Technical Field

[0001] The present invention relates to an intelligent load prediction and adjustment method in a dynamic weighing system, belonging to the technical field of dynamic weighing. Background Art

[0002] Existing dynamic weighing systems usually rely on filtering algorithms to remove noise interference. However, when traditional filtering algorithms process high-frequency noise and low-frequency signals, spectral aliasing is likely to occur, resulting in distortion of static load signals. This makes it impossible to meet the high-precision application scenarios in transient working conditions such as sudden stops and speed changes. Especially in fields such as e-commerce warehousing, automated logistics, and intelligent manufacturing, weighing errors pose significant challenges to work efficiency and cost control.

[0003] In addition, most existing dynamic weighing methods are based on the processing of single-sensor signals and fail to effectively consider the spatial distribution characteristics of the sensor array and the motion state of the vehicle, resulting in insufficient stability and reliability of the system in complex dynamic environments. Even with the support of high-precision sensors, traditional technologies are still difficult to achieve rapid adaptation to different vehicle types and dynamic environments, and the update process is complex and costly. Therefore, existing dynamic weighing technologies still have technical bottlenecks in dealing with rapidly changing dynamic loads of vehicles, improving weighing accuracy, and adapting to diverse application requirements. Summary of the Invention

[0004] The present invention provides an intelligent load prediction and adjustment method in a dynamic weighing system, and its main purpose is to solve the problem of insufficient weighing accuracy caused by inertial interference and dynamic error in the dynamic weighing system.

[0005] To achieve the above object, an intelligent load prediction and adjustment method in a dynamic weighing system provided by the present invention includes the following steps:

[0006] S1, Signal acquisition: Obtain the real-time output signals of the weighing sensor array and synchronously receive the motion state data of the vehicle, where the motion state data includes acceleration, speed, and attitude information;

[0007] S2, Spatiotemporal double-domain decoupling:

[0008] Spatial domain analysis: Based on the spatial distribution characteristics of the weighing sensor array, extract the spatial domain features of the static load, where the spatial domain features include the sparsity of pressure point distribution and the geometric vector of the center of gravity position;

[0009] Temporal domain analysis: According to the real-time motion state data of the vehicle, construct a dynamic inertial interference model to generate temporal domain noise components; the dynamic inertial interference model is expressed as:

[0010] ,

[0011] Among them, is the noise component in the time domain, is the real-time acceleration data, is the real-time velocity data, is the acceleration change rate, , , are the inertial interference coefficients fitted by calibrating data of the standard test vehicle;

[0012] S3, Feedforward compensation: Matching the type of the vehicle through a pre-trained dynamic basis function library, predicting the interference evolution trend at future moments based on the dynamic inertial interference model, generating a feedforward compensation signal, and reversely superimposing the feedforward compensation signal on the real-time output signal to obtain a corrected static load value;

[0013] S4, Closed-loop self-healing calibration: When it is detected that the deviation between the corrected static load value and the preset threshold exceeds the limit, tracing the residual error distribution pattern based on the spatial consistency characteristics of the sensor array, and adjusting the weight parameters of the spatio-temporal dual-domain decoupling through the gradient descent method to complete dynamic error suppression.

[0014] As a preferred embodiment, in step S2: The spatial domain analysis uses a sparse coding algorithm to separate the static load component in the signal of the weighing sensor array, where the basis function of the sparse coding is generated from experimental data of typical load distributions; The time domain analysis constructs a second-order inertial interference equation through the acceleration and velocity data of the vehicle to calculate the dynamic noise component in real time.

[0015] As a preferred embodiment, the pre-training method of the dynamic basis function library includes: Extracting the common motion-load coupling characteristics of forklifts, AGVs, and conveyor belts through transfer learning to generate a basic feature vector; When a new vehicle type is added, fine-tuning the basic feature vector through online learning based on the mapping relationship between 10-20 groups of real-time motion states and weighing data.

[0016] As a preferred embodiment, in the closed-loop self-healing calibration: Detecting the residual error distribution pattern through the spatial consistency of the sensor array to locate the inertial interference components that are not fully decoupled; Using a lightweight online learning algorithm to adjust the weight parameters of the spatio-temporal dual-domain decoupling, where the weight parameters include the basis function weights of the spatial domain sparse coding and the coefficients of the time domain inertial interference model.

[0017] As a preferred embodiment, the matching method of the dynamic basis function library includes: Invoking the pre-trained forklift motion basis function, AGV motion basis function, or conveyor belt motion basis function according to the vehicle type; When the vehicle type is not pre-stored, matching the closest basis function group based on the spectral characteristics of the real-time motion state data.

[0018] As a preferred embodiment, in the spatio-temporal dual-domain decoupling: the priority of spatial-domain feature extraction is higher than that of time-domain interference modeling, and the decoupling operations of the two are synchronously completed within a 10-ms cycle; the geometric vector of the spatial-domain features is calculated through the topological relationship of the sensor array, and the topological relationship includes the adjacent sensor spacing and the pressure distribution gradient.

[0019] As a preferred embodiment, the method for generating the feedforward compensation signal includes: predicting the amplitude of the inertial interference component within the next 5 - 10 ms according to the acceleration change rate of the vehicle; scaling the predicted value through the migration adaptation coefficient of the dynamic basis function library to generate a compensation signal matching the current vehicle.

[0020] As a preferred embodiment, the triggering condition for the closed-loop self-healing calibration is that the fluctuation amplitude of the corrected static load value exceeds 1.5 times the preset threshold within three consecutive sampling periods; the learning rate of the gradient descent method is dynamically adjusted according to the distribution pattern of the residual error.

[0021] As a preferred embodiment, it further includes the update of the dynamic basis function library, and the update mechanism of the dynamic basis function library includes: triggering a global optimization of the basis function library every time 100 groups of mapping relationships between the vehicle motion state and the weighing data are newly added; during the optimization process, the common features of the historical basis function groups are retained, and only the feature vectors conflicting with the newly added data are updated.

[0022] As a preferred embodiment, the method for initializing the weight parameters of the spatio-temporal dual-domain decoupling is: based on the calibration data of the standard test vehicle, fitting the initial weights through the least squares method; the standard test vehicle includes the reference motion load curves of three types of equipment, namely forklifts, AGVs, and conveyor belts.

[0023] Compared with the problems described in the background art, the beneficial effects of the present invention are as follows: By combining spatial geometry analysis and kinematic feedforward prediction, the dynamic weighing signal is decoupled into static load characteristics in the spatial domain (such as the sparsity of pressure point distribution and the geometric vector of the center of gravity) and dynamic inertial interference components in the time domain (such as acceleration coupling noise). Error separation is achieved at the signal source, avoiding the distortion of static load information caused by time-frequency aliasing in traditional filtering algorithms. Through the collaborative decoupling of spatial geometric features and kinematic prediction, the signal coupling effect is fundamentally eliminated, and the weighing stability under dynamic conditions is improved. The motion state of the vehicle is used as a feedforward variable, and the future interference trend is predicted based on the real-time motion state of the vehicle, which is more applicable to transient scenarios such as sudden stops and speed changes. Weighing accuracy is ensured in the form of pre-judgment. Through the migration of common features across devices and lightweight online learning, zero-cost and rapid adaptation of new vehicles is achieved (such as fine-tuning with 10 sets of samples). At the same time, combined with the spatial consistency detection of the sensor array, the residual error distribution pattern is automatically traced and the decoupling model is dynamically corrected, achieving a low-cost usage effect, which is more conducive to meeting the requirements of e-commerce warehousing scenarios for high-density and high-dynamic weighing. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flowchart of the intelligent load prediction and adjustment method of the present invention;

[0025] Figure 2 is a flowchart of the intelligent load prediction and adjustment method in the dynamic weighing system of the present invention;

[0026] Figure 3 is a process diagram of error detection and correction in the dynamic weighing system of the present invention.

[0027] The implementation, functional characteristics, and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0029] The embodiment of the present application provides an intelligent load prediction and adjustment method in a dynamic weighing system. The method includes the following steps: S1, signal acquisition: Obtain the real-time output signal of the weighing sensor array and synchronously receive the motion state data of the vehicle. The motion state data includes acceleration, speed, and attitude information;

[0030] S2, spatio-temporal dual-domain decoupling: Spatial domain analysis: Based on the spatial distribution characteristics of the weighing sensor array, extract the spatial domain characteristics of the static load. The spatial domain characteristics include the sparsity of pressure point distribution and the geometric vector of the center of gravity position; Time domain analysis: According to the real-time motion state data of the vehicle, construct a dynamic inertial interference model to generate time domain noise components. The dynamic inertial interference model is expressed as:

[0031] ,

[0032] Among them, is the noise component in the time domain, is the real-time acceleration data. The real-time acceleration data comes from the acceleration sensors in the weighing sensor array. During the data acquisition process, the real-time acceleration signal is filtered to remove static interference, ensuring that it only reflects the dynamic acceleration changes. is the real-time speed data. The real-time speed data is obtained in real-time by the speed sensor and is appropriately filtered to reduce static errors. The accuracy of the speed data directly affects the accuracy of the interference prediction model. Therefore, during the operation of the system, the calibration and maintenance of the speed sensor are crucial; is the acceleration change rate. This parameter is obtained from the differential equation of the vehicle's motion state and is specifically calculated as the time derivative of the acceleration signal, reflecting the vehicle's speed change trend. The acceleration change rate provides immediate feedback on the vehicle's dynamic characteristics and is the key to constructing the dynamic inertial interference model; , , are the inertial interference coefficients fitted from the calibration data of the standard test vehicle; these coefficients are obtained by fitting the experimental data of the standard test vehicle. Through regression analysis, the least squares method is used to fit the acceleration, speed, and acceleration change rate to ensure that these coefficients can accurately reflect the inertial characteristics of the vehicle under different dynamic loads. This process avoids overfitting through multiple cross-validations, ensuring that the model can adapt to different vehicle types.

[0033] S3, Feedforward compensation: Match the type of the vehicle through a pre-trained dynamic basis function library, predict the interference evolution trend at future moments based on the dynamic inertial interference model, generate a feedforward compensation signal, and reverse-stack the feedforward compensation signal to the real-time output signal to obtain a corrected static load value; S4, Closed-loop self-healing calibration: When it is detected that the deviation between the corrected static load value and the preset threshold exceeds the limit, trace the residual error distribution pattern based on the spatial consistency characteristics of the sensor array, and adjust the weight parameters of the spatio-temporal dual-domain decoupling through the gradient descent method to complete dynamic error suppression.

[0034] As a preferred implementation, in step S2: In the spatial domain analysis, a sparse coding algorithm is used to separate the static load component in the signal of the weighing sensor array, where the basis function of the sparse coding is generated from the experimental data of the typical load distribution; in the time domain analysis, a second-order inertial interference equation is constructed from the acceleration and speed data of the vehicle to calculate the dynamic noise component in real time.

[0035] As a preferred embodiment, the pre-training method of the dynamic basis function library includes: extracting the common motion-load coupling features of forklifts, AGVs, and conveyor belts through transfer learning to generate basic feature vectors; when a new vehicle type is added, fine-tuning the basic feature vectors through online learning based on the mapping relationship between 10-20 groups of real-time motion states and weighing data of the new vehicle type.

[0036] As a preferred embodiment, in the closed-loop self-healing calibration: detecting the residual error distribution pattern through the spatial consistency of the sensor array to locate the inertial interference components that are not fully decoupled; using a lightweight online learning algorithm to adjust the weight parameters of the spatio-temporal dual-domain decoupling, where the weight parameters include the basis function weights of spatial-domain sparse coding and the coefficients of the time-domain inertial interference model.

[0037] As a preferred embodiment, the matching method of the dynamic basis function library includes: calling the pre-trained forklift motion basis function, AGV motion basis function, or conveyor belt motion basis function according to the vehicle type; when the vehicle type is not pre-stored, matching the closest basis function group based on the spectral characteristics of the real-time motion state data.

[0038] As a preferred embodiment, in the spatio-temporal dual-domain decoupling: the priority of spatial-domain feature extraction is higher than that of time-domain interference modeling, and the decoupling operations of the two are synchronously completed within a 10 ms cycle; the geometric vectors of the spatial-domain features are calculated through the topological relationship of the sensor array, and the topological relationship includes the adjacent sensor spacing and the pressure distribution gradient.

[0039] As a preferred embodiment, the generation method of the feedforward compensation signal includes: predicting the amplitude of the inertial interference component within the next 5-10 ms according to the acceleration change rate of the vehicle; scaling the predicted value through the transfer adaptation coefficient of the dynamic basis function library to generate a compensation signal matching the current vehicle.

[0040] As a preferred embodiment, the triggering condition of the closed-loop self-healing calibration is that the fluctuation amplitude of the corrected static load value exceeds 1.5 times the preset threshold within three consecutive sampling periods; the learning rate of the gradient descent method is dynamically adjusted according to the distribution pattern of the residual error.

[0041] As a preferred embodiment, it further includes the update of the dynamic basis function library, where the update mechanism of the dynamic basis function library includes: triggering a global optimization of the basis function library every time 100 groups of mapping relationships between vehicle motion states and weighing data are added; during the optimization process, the common features of the historical basis function groups are retained, and only the feature vectors conflicting with the new data are updated.

[0042] As a preferred embodiment, the method for initializing the weight parameters of spatio-temporal decoupling is as follows: based on the calibration data of a standard test vehicle, the initial weights are fitted by the least squares method; the standard test vehicle includes the reference motion load curves of three types of equipment, namely forklifts, AGVs, and conveyor belts.

[0043] See Figure 2 and Figure 3 , Figure 2 is a flowchart of the intelligent load prediction and adjustment method in the dynamic weighing system of the present invention. First, sensor signals and motion state data are input. Next, spatial domain decoupling is performed. The spatial domain analysis includes sparse coding to separate static load components, basis functions from experimental data, and calculation of topological relationships (such as spacing / gradient). Then, temporal domain decoupling is performed. The temporal domain analysis includes constructing a third-order inertia equation and real-time calculation of dynamic noise components. The entire decoupling process is synchronously completed within a 10-ms cycle, and finally, static load characteristics and temporal domain noise components are output. Figure 3 is a process diagram of error detection and correction in the dynamic weighing system of the present invention. First, the system detects the deviation between the static load value and the adjusted value. According to the deviation value, the system enters the error detection module for detection. Then, the system performs spatial consistency detection to determine whether the responses of the sensor array are consistent. Based on this, the system updates relevant system parameters and restores stability. As the error detection progresses, the system adjusts the model parameters by a progressive order reduction method to reduce the impact of errors and finally ensure the stability and accuracy of the system. In addition, the system adjusts the spatio-temporal decoupling weights to gradually correct and optimize the weighing accuracy.

[0044] Example 1: During implementation, multiple weighing sensor arrays are first installed on the vehicle, and real-time motion state data of the vehicle are synchronously collected. The motion state data include acceleration, speed, and attitude information. Through these data, the system can comprehensively reflect all factors affecting weighing during the dynamic operation of the vehicle. While collecting signals, the system ensures the data synchronization of all sensors to avoid measurement errors caused by inconsistent data time series. In the weighing sensor array, the output signal of each sensor and the motion state data are transmitted into the signal processing module of the system together. Then, based on the spatial distribution characteristics of the sensor array, the system uses a sparse coding algorithm to separate the static load component and the dynamic noise component in the weighing signal. The extraction of spatial domain features is mainly characterized by the geometric vector of the center of gravity position and the sparsity of the pressure point distribution to represent the spatial distribution of the load. Specifically, the center of gravity geometric vector is calculated through the distance between adjacent sensors and the pressure distribution gradient. The spatial domain features are analyzed prior to the time domain features to ensure their priority is higher than the modeling of dynamic inertial interference. Subsequently, using the acceleration, speed, and attitude information of the vehicle, the system constructs a dynamic inertial interference model in real time. This model calculates and extracts the noise component in the time domain through the motion state data of the vehicle. The output signal of each sensor, after being decoupled in the spatio-temporal domain, can effectively separate the static load from the dynamic inertial interference, thus avoiding errors caused by spectral aliasing in traditional filtering algorithms.

[0045] The system generates corresponding compensation signals through a pre-trained dynamic basis function library, such as according to the vehicle type and motion state. The generation of the compensation signal is based on the motion state of the vehicle and, in combination with the previous dynamic inertial interference model, predicts the interference evolution trend at future moments. The feedforward compensation signal actively corrects the dynamic load error through reverse superposition, thereby effectively improving the real-time accuracy of the weighing system. When the vehicle type changes or a new vehicle is added, the system will fine-tune the dynamic adaptation coefficients in the basis function library through online learning. Specifically, the system updates the basis function library through transfer learning based on the mapping relationship between 10 - 20 groups of real-time motion data and weighing data of the newly added vehicle to ensure that the new vehicle can be quickly adapted at zero cost and maintain the weighing accuracy.

[0046] When the system detects that the deviation between the corrected static load value and the preset threshold exceeds the limit, it enters the closed-loop self-healing calibration mechanism. At this time, based on the spatial consistency characteristics of the sensor array, the system locates the inertial interference components that are not fully decoupled by tracing the residual error distribution pattern. Through the gradient descent method, the system dynamically adjusts the weight parameters of the spatio-temporal decoupling, so as to achieve the suppression of dynamic errors. Through the spatial consistency detection algorithm of the sensor array, the system can timely detect the errors caused by hardware failures or external interferences and effectively correct them. The learning rate of the gradient descent method is dynamically adjusted according to the distribution pattern of the residual errors to ensure the accuracy and stability of error calibration. The system is divided into four major modules: the signal acquisition module, the spatio-temporal decoupling module, the compensation signal generation module, and the closed-loop calibration module. Each module has independent functions and responsibilities, and the data flow and control logic between the modules are closely connected. The specific process is as follows: Signal acquisition module: Real-time acquisition of load cell signals and synchronous collection of the motion state data of the vehicle; Spatio-temporal decoupling module: Perform signal analysis in the spatial domain and the time domain to separate the static load from the dynamic interference; Compensation signal generation module: Generate a feedforward compensation signal based on the dynamic basis function library and the kinematic model; Closed-loop calibration module: Automatically trigger closed-loop calibration when the error exceeds the limit, and adjust the decoupling model parameters through the gradient descent method to correct the dynamic error. This embodiment combines spatial geometric analysis and kinematic feedforward prediction, breaking through the bottleneck of signal decoupling in traditional dynamic weighing systems. In practical applications, the system can effectively improve the dynamic weighing accuracy, especially having obvious advantages in transient working conditions such as sudden stops and speed changes. The modular design of the system makes the technical solution easy to implement and maintain, and can be automatically adapted according to different vehicle types without additional hardware support, having good market application prospects. Through precise signal decoupling and real-time dynamic compensation, the present invention can provide a high-precision and high-stability dynamic weighing solution in complex weighing environments, especially in high-density and high-dynamic e-commerce warehousing environments.

[0047] Embodiment 2: In the signal acquisition stage of this embodiment, first, ensure that the real-time motion state data of the weighing sensor array and the vehicle (including acceleration, speed, and attitude information) can be synchronously acquired. These data are simultaneously acquired by the weighing sensor and the inertial measurement unit (IMU) and transmitted to the system. In a specific implementation, a timing synchronization mechanism is adopted to ensure the timing consistency of the sensor data and avoid weighing errors caused by signal delay or data timing disorder. The optimized spatio-temporal decoupling step is divided into two sub-steps: spatial domain feature extraction and time domain interference modeling. Spatial domain feature extraction: On the original basis, further clarify the specific calculation method of the spatial domain features. The spatial domain features of the static load include the sparsity of the pressure point distribution and the geometric vector of the center of gravity. During implementation, by analyzing the positional relationship of the weighing sensor array (sensor spacing and pressure distribution gradient of adjacent sensors), the geometric vector of the static load is accurately calculated. This part no longer relies on traditional filtering algorithms but adopts a geometric analysis method based on the topology of the sensor array. By algorithmically analyzing the local load distribution of each sensor, the center of gravity position of the entire array is then calculated. Time domain interference modeling: Further improve the time domain model and clarify the generation process of the dynamic inertial interference signal. During this process, the acceleration, speed, and attitude data of the vehicle act together to construct a second-order inertial interference equation. By calculating the noise component of the acceleration signal in real time, the weight of the time domain noise component is dynamically adjusted. The source and role of each variable have been clarified to ensure that the model can adapt to the inertial interference of different vehicles under different working conditions.

[0048] The specific method for generating the feedforward compensation signal is as follows: Through a pre-trained dynamic basis function library, based on the real-time motion state data of the vehicle (acceleration, rate of change of speed, etc.), accurately predict the change trend of the inertial interference signal within the next 5 - 10 milliseconds, and then based on the predicted interference trend, correct the weighing signal by means of reverse superposition. For each vehicle type, the basis functions in the dynamic basis function library have been pre-trained through transfer learning and are fine-tuned online according to the newly added vehicle data to ensure that the compensation signal can be dynamically adapted according to the specific characteristics of the vehicle, and predict the inertial interference amplitude within the next few milliseconds based on the real-time motion state data of the vehicle. Through the pre-trained dynamic basis function library, the motion-load coupling characteristics of each vehicle type are extracted, and the coefficients in the basis function library are fine-tuned through transfer learning. The system reversely superposes the compensation signal based on the prediction result of the rate of change of acceleration to correct the weighing signal. This process accurately predicts the future dynamic error and avoids weighing deviation caused by inertial interference.

[0049] The optimized closed-loop self-healing calibration mechanism specifically includes the following steps, such as error detection and traceability: When the deviation between the corrected static load value and the preset threshold exceeds the predetermined range, the system locates the inertial interference components that are not fully decoupled by analyzing the spatial consistency characteristics of the sensor array. By tracing the residual error distribution pattern, the rapid identification of dynamic errors is ensured; Gradient descent method for dynamic adjustment: Based on the detected residual error distribution pattern, the gradient descent method is used to adjust the weight parameters of the spatio-temporal decoupling. The key parameters include the basis function weights of the spatial domain sparse coding and the coefficients of the inertial interference model in the time domain. The learning rate and adjustment step size of the gradient descent method will be dynamically optimized according to the error distribution pattern to ensure the accuracy and stability of the adjustment process. And for the spatial domain feature extraction, a sparse coding algorithm is adopted. Through the topological relationship of the sensor array, the sparsity of the pressure point distribution and the centroid geometric vector are analyzed. The time domain interference modeling is carried out through a dynamic inertial interference model, which models the acceleration, velocity and their change rates, calculates the dynamic noise components in real time, and uses these noise components to predict the future dynamic interference trend. Through this method, the static load and the dynamic interference signal can be effectively decoupled, and both belong to the extended implementation methods known to those of ordinary skill in the art.

[0050] The dynamic basis function library is updated in real time during the implementation process. Whenever 100 new mapping relationships between the vehicle motion state and the weighing data are added, the global optimization process of the basis function library is triggered. During the optimization process, the common characteristics of the historical basis function groups are retained, and only the eigenvectors conflicting with the new data are adjusted. This optimization process can ensure the zero-cost and rapid adaptation of new vehicle types without increasing the system hardware cost. During the implementation process of the entire system, the following specific optimization paths are adopted to ensure the accuracy and efficiency of the algorithm, that is, a sparse coding algorithm is used to extract the static load component in the spatial domain analysis to ensure that when the signal is decoupled, it is not affected by the spectral aliasing problem in the traditional filtering algorithm. The time domain interference model can accurately predict and suppress the dynamic interference signal in the transient working condition by modeling the second-order equations of the acceleration and velocity data.

[0051] During system operation, all real-time data (acceleration, speed, acceleration change rate) is transmitted into the system from various sensors through a synchronization mechanism. Through pre-set models and algorithms, this data is analyzed in real time and necessary conversions are performed to ensure the temporal consistency of the data. The data of each sensor will undergo a decoupling and correction process in the system to ensure that each variable can directly contribute to the improvement of weighing accuracy. For example, extraction of spatial domain features: Through the relative position relationship of the sensor array (such as sensor spacing and pressure distribution gradient), the system extracts the spatial domain features of the static load. The extraction of these features takes precedence over the construction of the dynamic interference model to ensure that in spatio-temporal decoupling, the static load can be accurately processed preferentially; Establishment of the dynamic inertial interference model: By combining the acceleration, speed, and attitude information of the vehicle, a dynamic inertial interference model is constructed. This model identifies and predicts the trend of dynamic errors through temporal data analysis and dynamically adjusts the weight coefficients to ensure that the interference signals of the data of each sensor can be effectively eliminated. These all belong to the extended implementation methods known to those of ordinary skill in the art.

[0052] Embodiment 3: This embodiment provides an intelligent load prediction and adjustment method in a further optimized dynamic weighing system. In this embodiment, first, it is ensured that the weighing sensor array synchronously collects the motion state data (acceleration, speed, attitude) of the vehicle. The signal acquisition adopts a timing synchronization mechanism to ensure the temporal consistency of the sensor data and avoid errors caused by disordered data timing. The output signal of each sensor will be transmitted to the signal processing module in real time, and this module further processes the signal to ensure the effectiveness of each signal source.

[0053] Space-time Dual-domain Decoupling Optimization: The optimization of the space-time dual-domain decoupling process includes two parts: spatial domain analysis and temporal domain interference modeling. Spatial Domain Analysis: In this embodiment, a sparse coding algorithm is adopted to analyze the spatial characteristics of static loads through the topological relationship of the sensor array. Specifically, the sensor spacing and the pressure distribution gradient are used to calculate the geometric vector of the centroid position, and the sparsity of the pressure point distribution is combined to extract the spatial domain features. In this way, the spatial domain feature extraction takes precedence over the temporal domain interference modeling, ensuring that the static load signal can be accurately extracted and avoiding distortion caused by spectral aliasing. Temporal Domain Analysis: During the temporal domain decoupling process, a dynamic inertial interference model is constructed by combining the real-time motion state data of the vehicle (acceleration, speed, and attitude information). This model is based on the time series of acceleration and speed, and calculates the dynamic noise component through a second-order inertial interference equation. This step ensures accurate modeling of the noise component and interference suppression by dynamically adjusting the weight coefficient in real time. Optimization of Feedforward Compensation and Dynamic Basis Function Library: Active correction of dynamic errors is achieved through feedforward compensation. In this embodiment, the dynamic basis function library uses a pre-training method to generate basic feature vectors to adapt to the coupled characteristics of the loads and motion states of different vehicle types. The real-time motion state data of each vehicle (acceleration, speed, attitude change) is input into the dynamic basis function library, and the system fine-tunes the basis function through a transfer learning mechanism to adapt to the dynamic behavior of the new vehicle. Generation of Feedforward Compensation Signal: The compensation signal generated by the dynamic basis function library predicts the amplitude of inertial interference within the next 5-10 milliseconds based on the acceleration change rate of the vehicle, and corrects the weighing signal through reverse superposition. This process plays a feedforward role in the system, effectively avoiding weighing errors caused by inertial interference and ensuring the stability of the system under transient conditions such as sudden stops and speed changes. Closed-loop Self-healing Calibration and Error Correction: When the deviation between the corrected static load value and the preset threshold exceeds the limit, the system automatically enters the closed-loop self-healing calibration mode.

[0054] Spatial Consistency Detection: Through the spatial consistency characteristics of the sensor array, the system analyzes the distribution pattern of the residual error and locates the inertial interference components that are not fully decoupled. According to the results of the spatial consistency detection, the system will dynamically adjust the weight parameters of the spatio-temporal dual-domain decoupling to optimize the error suppression process. At this time, a lightweight online learning algorithm is used to adjust the weight parameters to ensure that the dynamic error of the system can be effectively suppressed. Dynamic Adjustment of the Gradient Descent Method: To improve the accuracy of closed-loop calibration, the gradient descent method is used to dynamically adjust the weight parameters. By analyzing the distribution pattern of the residual error, the learning rate of the gradient descent method will be adjusted in real time according to the characteristics of the error to ensure the accuracy and stability of the error correction process. Update Mechanism of the Dynamic Basis Function Library: To enhance the adaptability and stability of the system, this embodiment introduces an update mechanism for the dynamic basis function library. When 100 new sets of mapping relationships between vehicle motion states and weighing data are added, the system triggers a global optimization of the basis function library. During the optimization process, the common characteristics of the historical basis function groups are retained, and only the feature vectors that conflict with the new data are updated to ensure the rapid adaptation of new vehicle types without increasing the hardware burden of the system.

[0055] Embodiment 4: This embodiment further describes the formula used in the dynamic inertial interference model:

[0056] ,

[0057] Where: : The time-domain noise component, which is the final output of the interference signal; : The real-time acceleration data of the vehicle, which comes from the acceleration sensors of the weighing sensor array; : The real-time speed data of the vehicle, which is obtained through speed sensors and has been filtered to remove static interference; : The acceleration change rate, which is obtained through the differential equation of the vehicle motion state and reflects the dynamic trend of the vehicle speed change; coefficient , , are inertial interference coefficients obtained by fitting the calibration data of the standard test vehicle to ensure that the model can accurately reflect the dynamic behavior of different vehicles. The numerical value and calculation method of each coefficient are determined through regression analysis of experimental data to ensure its compliance with practical applications. In this embodiment, the inertial interference coefficient It is obtained by fitting the calibration data of standard test vehicles, which include three typical vehicles: forklifts, AGVs (Automated Guided Vehicles), and conveyor belts. The criteria for selecting these vehicles are based on their wide applications in the fields of e-commerce warehousing and automated logistics. The calibration process includes measuring the acceleration, speed, and attitude changes of each vehicle under static and dynamic conditions. Data acquisition uses high-precision inertial measurement units (IMUs) and weighing sensor arrays to ensure high accuracy and low error in each data acquisition. During the fitting process, for example, the least squares method can be used for regression analysis to match the collected motion state data (such as acceleration, speed, and attitude changes) with the weighing data, and calculate the optimal inertial interference coefficient. Multiple cross-validations are used during the fitting process to avoid overfitting and ensure that the model can adapt to different dynamic load conditions. Finally, the obtained inertial interference coefficient reflects the inertial characteristics of the vehicle in a dynamic environment through the calibration data of the standard vehicle, and can effectively eliminate dynamic interferences caused by acceleration changes, speed fluctuations, etc. in practical applications.

[0058] In practical applications, the inertial interference coefficient will be used to correct the interference error in the dynamic weighing system. Especially under transient conditions such as sudden stops and speed changes of the vehicle, by applying these coefficients to the dynamic inertial interference model, the system can perform feed-forward compensation based on real-time data acquisition, effectively improving the weighing accuracy and stabilizing the output result of the system. The specific implementation method is as follows: Step 1: Signal acquisition. The signal acquisition module synchronously acquires the dynamic load data of the vehicle through a high-precision weighing sensor array, and cooperates with an inertial measurement unit (IMU) to obtain acceleration, speed, and attitude information. All data ensures temporal consistency through a timing synchronization mechanism, thereby eliminating errors caused by data timing confusion.

[0059] Step 2: Spatiotemporal decoupling. In this step, the spatial domain feature extraction calculates the topological relationship of the sensor array to determine the local pressure distribution of each sensor, and further extracts the spatial domain features of the static load (such as the sparsity of the pressure point distribution and the geometric vector of the center of gravity). These spatial domain features are analyzed prior to the time domain interference modeling to ensure the accurate extraction of static load data and avoid distortion caused by spectral aliasing. The time domain interference modeling analyzes the acceleration and speed data by constructing a second-order inertial interference equation to calculate the dynamic noise component. The key to this part is to calculate the trend of the interference signal through real-time acceleration data and the rate of change of speed, and then provide input for the feed-forward compensation module.

[0060] Step 3: Feedforward compensation. During the feedforward compensation process, the future interference trend is predicted based on the previous dynamic inertia interference model. Through the pre-trained dynamic basis function library, the system can generate targeted compensation signals for different vehicle types (such as forklifts, AGVs, conveyor belts, etc.). This compensation signal is reversely superimposed on the real-time output signal to correct the static load value, thereby eliminating the influence of inertia interference. The generation of the compensation signal depends on the prediction models, which are continuously fine-tuned and optimized through transfer learning algorithms based on the kinematic characteristics of the vehicle and historical load data to adapt to the dynamic behavior of new vehicles.

[0061] Step 4: Closed-loop self-healing calibration. When it is detected that the deviation of the corrected static load value exceeds the limit, the system will trigger the closed-loop self-healing calibration mechanism. This mechanism locates the uncompletely decoupled inertia interference components by using the residual error distribution pattern based on the spatial consistency characteristics of the sensor array. The gradient descent method is used to adjust the weight parameters of the spatio-temporal double-domain decoupling to suppress the dynamic error. This calibration step can not only make real-time adjustments according to the spatial consistency detection of the sensor array, but also optimize the parameters during the online learning process to ensure that the error remains within an acceptable range during long-term operation, and all belong to the extended implementation methods known to those of ordinary skill in the art.

[0062] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent load prediction and regulation method in a dynamic weighing system, characterized in that, It includes the following steps: S1, Signal acquisition: Obtain the real-time output signals of the weighing sensor array and synchronously receive the motion state data of the vehicle, where the motion state data includes acceleration, speed, and attitude information; S2, Spatiotemporal dual-domain decoupling: Spatial domain analysis: Based on the spatial distribution characteristics of the weighing sensor array, extract the spatial domain features of the static load, where the spatial domain features include the sparsity of the pressure point distribution and the geometric vector of the center of gravity position; Temporal domain analysis: According to the real-time motion state data of the vehicle, construct a dynamic inertial interference model to generate a temporal domain noise component; the dynamic inertial interference model is expressed as: ， Among them, is the noise component in the time domain, is the real-time acceleration data, is the real-time velocity data, is the acceleration change rate, , , are the inertial interference coefficients fitted by calibrating the data of the standard test vehicle; S3, Feedforward compensation: Match the type of the vehicle through a pre-trained dynamic basis function library, predict the interference evolution trend at future moments based on the dynamic inertial interference model, generate a feedforward compensation signal, and reversely superimpose the feedforward compensation signal on the real-time output signal to obtain a corrected static load value; S4, Closed-loop self-healing calibration: When it is detected that the deviation between the corrected static load value and the preset threshold exceeds the limit, trace the residual error distribution pattern based on the spatial consistency characteristics of the sensor array, and adjust the weight parameters of the spatiotemporal dual-domain decoupling through the gradient descent method to complete dynamic error suppression; Among them, in step S2: The spatial domain analysis uses a sparse coding algorithm to separate the static load component in the signals of the weighing sensor array, where the basis function of the sparse coding is generated from the experimental data of typical load distributions; the temporal domain analysis constructs a second-order inertial interference equation through the acceleration and speed data of the vehicle to calculate the dynamic noise component in real time; in the closed-loop self-healing calibration: Detect the residual error distribution pattern through the spatial consistency of the sensor array to locate the inertial interference components that are not fully decoupled; Use a lightweight online learning algorithm to adjust the weight parameters of the spatiotemporal dual-domain decoupling, and the weight parameters include the basis function weights of the spatial domain sparse coding and the coefficients of the temporal domain inertial interference model.

2. The intelligent load prediction and adjustment method in the dynamic weighing system according to claim 1, wherein The pre-training method of the dynamic basis function library includes: Extract the common motion load coupling characteristics of forklifts, AGVs, and conveyor belts through transfer learning to generate a basic feature vector; When a new vehicle type is added, fine-tune the basic feature vector through online learning based on the mapping relationship between its 10-20 groups of real-time motion states and weighing data.

3. The intelligent load prediction and regulation method in the dynamic weighing system according to claim 1, characterized in that, The matching method of the dynamic basis function library includes: Call the pre-trained forklift motion basis function, AGV motion basis function, or conveyor belt motion basis function according to the vehicle type; When the vehicle type is not pre-stored, match the closest basis function group based on the spectral characteristics of the real-time motion state data.

4. The intelligent load prediction and regulation method in the dynamic weighing system according to claim 1, characterized in that, In the spatiotemporal dual-domain decoupling: The priority of the spatial domain feature extraction is higher than that of the temporal domain interference modeling, and the decoupling operations of the two are synchronously completed within a 10 ms cycle; The geometric vector of the spatial domain feature is calculated through the topological relationship of the sensor array, and the topological relationship includes the adjacent sensor spacing and the pressure distribution gradient.

5. The intelligent load prediction and regulation method in the dynamic weighing system according to claim 4, characterized in that, The generation method of the feedforward compensation signal includes: Predict the amplitude of the inertial interference component within the next 5-10 ms according to the acceleration change rate of the vehicle; Scale the predicted value through the transfer adaptation coefficient of the dynamic basis function library to generate a compensation signal matching the current vehicle.

6. The intelligent load prediction and regulation method in the dynamic weighing system according to claim 5, characterized in that, The triggering condition for the closed-loop self-healing calibration is that the fluctuation amplitude of the corrected static load value exceeds 1.5 times the preset threshold within three consecutive sampling periods; the learning rate of the gradient descent method is dynamically adjusted according to the distribution pattern of the residual error.

7. The intelligent load prediction and regulation method in the dynamic weighing system according to claim 1, characterized in that It also includes the update of the dynamic basis function library, and the update mechanism of the dynamic basis function library includes: every time 100 groups of mapping relationships between the vehicle motion state and the weighing data are newly added, a global optimization of the basis function library is triggered; during the optimization process, the common features of the historical basis function groups are retained, and only the feature vectors conflicting with the newly added data are updated.

8. The intelligent load prediction and regulation method in the dynamic weighing system according to claim 1, characterized in that, The initialization method of the weight parameters for the spatio-temporal dual-domain decoupling is: based on the calibration data of the standard test vehicle, the initial weights are fitted by the least squares method; the standard test vehicle includes the reference motion load curves of three types of equipment, namely forklifts, AGVs, and conveyor belts.

Images