An integrated sensor module for dynamic weighing system

Abstract

The present invention relates to the field of measurement technology, and in particular to an integrated sensing module for a dynamic weighing system, including a physical domain, an environmental domain and a virtual domain, which realizes dynamic adjustment of parameters through cross-domain interaction, so as to cope with high-frequency vibrations and rapid temperature difference changes in complex environments. The physical domain is responsible for collecting multi-source data, the environmental domain perceives external interference in real time, and the virtual domain realizes time series feature extraction and dynamic model updating by combining the optimization capabilities of recursive neural networks and reinforcement learning. This technical solution significantly improves the real-time and adaptability of the system, can accurately predict environmental change trends, adjust the weighing strategy in advance, and control the error rate at an extremely low level. At the same time, by combining the negative feedback mechanism with the optimization of the objective function, the sampling frequency is dynamically adjusted to reduce energy consumption while ensuring data processing efficiency. The deep integration of wavelet decomposition, dynamic normalization algorithm and neural network model greatly enhances the accuracy and stability of data processing.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic weighing measurement, and particularly to an integrated sensing module for a dynamic weighing system. Background Art

[0002] Dynamic weighing technology is a key technology for real-time measurement of the weight of vehicles during driving, and is widely used in scenarios such as logistics transportation, highway toll stations, and industrial production lines. Its basic principle is to collect the weight information of vehicles or objects through sensors and transmit it to a data processing unit for calculation and analysis. However, in complex environments, such as rapid temperature difference changes and vibration interference on highways, and diverse cargo types in logistics transportation scenarios, traditional dynamic weighing technology faces many challenges in terms of real-time performance and accuracy.

[0003] Currently, to solve these problems, the following several methods are usually adopted in the industry:

[0004] 1. Static optimization model: By statically optimizing the parameters of weighing sensors under fixed conditions to adapt to specific application scenarios. However, this method is difficult to respond to dynamic changes in the external environment in real time, such as drastic fluctuations in temperature difference or changes in vehicle vibration frequency, resulting in an increase in weighing errors.

[0005] 2. Redundant design of sensor arrays: By increasing the number and types of sensors to improve the comprehensiveness and redundancy of data acquisition. However, this method will significantly increase the system cost and maintenance complexity, and the fusion of redundant data may also introduce more noise, further reducing the stability of the system.

[0006] 3. Centralized data processing architecture: By centrally transmitting all the collected data to a central processing unit for unified analysis and processing. This method has a significant delay in processing high-dimensional and multi-source data and is difficult to meet the real-time requirements of dynamic weighing systems.

[0007] In addition, to address the above problems, some industries have begun to introduce dynamic compensation strategies, such as correcting data through a simple regularization feedback mechanism. However, due to the lack of in-depth perception and modeling of environmental interference, this strategy cannot achieve precise compensation in high-complexity application scenarios, ultimately resulting in insufficient reliability and adaptability of system operation. Summary of the Invention

[0008] In view of the deficiencies of the prior art, the present invention provides the following technical solutions:

[0009] An integrated sensing module for a dynamic weighing system, the module comprising:

[0010] Physical domain sensor integration unit, used to collect multi-source data generated during dynamic weighing, where:

[0011] The multi-source data includes the weighing data of the object, environmental vibration data, and temperature data;

[0012] The physical domain sensor integration unit includes a strain gauge sensor, an acceleration sensor, and a temperature sensor;

[0013] The physical domain sensor integration unit preprocesses the collected data through an edge computing unit, and the preprocessing includes signal filtering, data normalization, and frequency domain characteristic extraction;

[0014] Environmental domain monitoring unit, used to collect and process environmental parameters in real time, where:

[0015] The environmental parameters include environmental vibration frequency, temperature gradient, and humidity fluctuation;

[0016] The unit extracts the environmental vibration frequency characteristics through the short-time Fourier transform STFT, calculates the temperature gradient and humidity change rate through the differential algorithm, and outputs the environmental interference characteristics;

[0017] Virtual domain data fusion and optimization unit, used to build a multi-domain collaborative dynamic twin model based on the data of the physical domain sensor integration unit and the environmental domain monitoring unit, where:

[0018] The dynamic twin model includes:

[0019] A dynamic prediction module based on the recurrent neural network RNN, used to predict the impact of vibration on the weighing data according to the environmental interference characteristics;

[0020] An optimization module based on reinforcement learning RL, used to generate a dynamic compensation strategy, where:

[0021] The dynamic compensation strategy is generated by fusing historical data and real-time data, and is used to dynamically adjust the sensitivity and sampling frequency of the sensor;

[0022] The data fusion formula is: Where, Refers to the data fusion result after weighted summation, representing the comprehensive calculated value of multi-source sensor data; Represents the preprocessed data of the th sensor, Represents the weight coefficient of the corresponding sensor, and the weight coefficient The calculation formula of is:

[0023] ;

[0024] Where, Represents the sensor sensitivity factor; Represents different sensor numbers; Represents the real-time correlation between the sensor and environmental data, calculated by the correlation coefficient formula: ;

[0025] Where, is and the covariance of environmental data , and are respectively and standard deviations;

[0026] The dynamic compensation unit is used to adjust the sampling parameters of the sensor in real time according to the dynamic compensation strategy generated by the virtual domain data fusion and optimization unit, where:

[0027] The dynamic compensation includes adjusting the sensitivity, sampling frequency and data processing priority of the sensor;

[0028] The dynamic compensation unit and the physical domain sensor integration unit form a closed-loop feedback;

[0029] The distributed computing unit is used to support the collaborative work of each unit, where: the edge computing unit is used to process the data of the physical domain sensor integration unit and the environmental domain monitoring unit; the central computing unit is used to run the multi-domain collaborative twin model and generate the dynamic compensation strategy; the cloud optimization unit is used to integrate historical data and large-scale environmental characteristic data to optimize the twin model.

[0030] As an improvement of the above technical solution, the wavelet decomposition algorithm in the edge computing unit includes at least three levels of decomposition, and the decomposition basis function selects the Daubechies wavelet basis, which is specifically implemented through the following steps: performing discrete wavelet transform on the collected original signal; extracting high-frequency components for vibration feature analysis; combining low-frequency components for tilt signal correction, and finally generating dynamic feature signals.

[0031] As an improvement of the above technical solution, the normalization processing in the environmental domain monitoring unit adopts the maximum-minimum normalization algorithm, through the formula:

[0032] Calculated, where and The values of are determined according to the real-time data statistical results of the environmental domain monitoring unit.

[0033] As an improvement to the above technical solution, the model training method of the recurrent neural network combined with the attention mechanism in the virtual domain data fusion and optimization unit includes the following steps: constructing a multi-dimensional data set containing historical weighing data, vibration characteristics, and environmental parameters; designing a recurrent neural network structure with an attention mechanism for dynamic weight allocation; optimizing the network weights through the error backpropagation algorithm, and the objective function is to minimize the root mean square value of the prediction error.

[0034] As an improvement to the above technical solution, the calculation steps of the weight coefficient include: calculating the mean and standard deviation of the real-time collected data to determine the initial weight; constructing a covariance matrix to analyze the correlation between each input feature; normalizing the weight based on the gradient descent method, and the range is limited between 0 and 1.

[0035] As an improvement to the above technical solution, the sensor sampling frequency adjustment algorithm in the dynamic compensation unit adopts negative feedback control, where the feedback signal is the deviation between the vibration amplitude and the standard threshold, and it is specifically implemented through the following method: calculating the amplitude of the real-time vibration signal; comparing the set threshold to determine the sampling frequency adjustment amplitude; dynamically updating the sampling frequency within the range of 100Hz to 1kHz.

[0036] As an improvement to the above technical solution, the message queue mechanism in the distributed computing unit is implemented through an event-driven mode, where the event triggers include the following conditions: the sensor signal exceeds the set threshold; a notification signal is generated after the data fusion task is completed.

[0037] As an improvement to the above technical solution, the cut-off frequency range of the high-pass filter in the physical domain sensor integration unit is 10Hz to 50Hz, which is used to remove low-frequency interference signals.

[0038] As an improvement to the above technical solution, the message queue mechanism is implemented based on the MQTT protocol, and specifically includes the following steps: realizing data classification transmission through multi-topic publishing; the subscriber parses and validates the received data; ensuring data synchronization and consistency of multiple sensor nodes.

[0039] The beneficial effects of the present invention are as follows: Through the multi-domain collaborative twin model of the physical domain, environmental domain and virtual domain, the problem that traditional weighing technology is difficult to respond to complex environmental changes in real time is solved. When dealing with high-frequency vibrations and rapid temperature differences, the model dynamically adjusts parameters through cross-domain interactions, so that the dynamic weighing system can still operate accurately under multi-dimensional interference. This multi-domain collaborative technical design significantly improves the real-time and adaptability of the system; by combining the time series feature extraction capability of the recursive neural network (RNN) and the adaptive optimization capability of reinforcement learning (RL), the real-time update of the dynamic model is achieved, which not only improves the system's processing efficiency for multi-dimensional data, but also can actively predict environmental change trends and adjust the weighing strategy in advance, so that the error rate of the system in complex industrial environments is significantly reduced; and the sampling frequency is dynamically adjusted by combining the negative feedback mechanism with the objective function optimization to ensure data While improving processing efficiency, energy consumption is reduced. In particular, the feedback signal is used to correct system deviations in real time, avoiding the waste of resources in the traditional method of frequent reconfiguration, achieving a high balance between efficiency and energy consumption, and organically combining wavelet decomposition with dynamic normalization algorithm and neural network optimization model to achieve deep integration of multiple algorithms in dynamic weighing, greatly improving the accuracy and stability of data processing; and through the close coordination of physical domain sensor integration units, edge computing units and distributed computing units, a technical closed loop is constructed, especially the application of message queue mechanism in multi-node data synchronization, ensuring the real-time performance and data consistency of the system, and enhancing the reliability of the dynamic weighing system in a complex distributed environment. Brief Description of the Figures

[0040] Figure 1 This is the architecture diagram of the multi-domain collaborative twin model of the present invention;

[0041] Figure 2 This is the flow chart of the dynamic weighing system of the present invention. Specific implementation method

[0042] The following describes the implementation of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific implementations. The details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0043] The existing battery cabinets are closed storage to prevent the batteries from contacting the outside world. Long-term closed storage can easily lead to moisture inside the battery cabinet that cannot be discharged. In addition, closed storage can easily lead to internal temperature rise and erosion of the batteries due to lack of ventilation.

[0044] To resolve this issue, see Figure 1 - Figure 2, an integrated sensing module for a dynamic weighing system, the module comprising:

[0045] A physical domain sensor integration unit for collecting multi-source data generated during the dynamic weighing process, wherein:

[0046] The multi-source data includes the weighing data of the object, the environmental vibration data, and the temperature data;

[0047] The physical domain sensor integration unit includes a strain gauge sensor, an acceleration sensor, and a temperature sensor;

[0048] The physical domain sensor integration unit preprocesses the collected data through an edge computing unit, and the preprocessing includes signal filtering, data normalization, and frequency domain characteristic extraction;

[0049] An environmental domain monitoring unit for collecting and processing environmental parameters in real time, wherein:

[0050] The environmental parameters include environmental vibration frequency, temperature gradient, and humidity fluctuation;

[0051] The unit extracts the environmental vibration frequency characteristics through the short-time Fourier transform STFT, calculates the temperature gradient and the humidity change rate through the differential algorithm, and outputs the environmental interference characteristics;

[0052] A virtual domain data fusion and optimization unit for constructing a multi-domain collaborative digital twin model based on the data of the physical domain sensor integration unit and the environmental domain monitoring unit, wherein:

[0053] The digital twin model includes:

[0054] A dynamic prediction module based on the recurrent neural network RNN for predicting the influence of vibration on the weighing data according to the environmental interference characteristics;

[0055] An optimization module based on reinforcement learning RL for generating a dynamic compensation strategy, wherein:

[0056] The dynamic compensation strategy is generated by fusing historical data and real-time data and is used to dynamically adjust the sensitivity and sampling frequency of the sensor;

[0057] The data fusion formula is: Wherein, Refers to the data fusion result after weighted summation, representing the comprehensive calculated value of multi-source sensor data; Represents the preprocessed data of the th sensor, Represents the weight coefficient of the corresponding sensor, and the weight coefficient The calculation formula of is:

[0058] ;

[0059] Among them, is the weight of the th sensor, which is calculated by weighting the sensitivity of each sensor and the correlation with environmental data . The calculation formula of the weight sums up all sensors (from to ) to obtain a normalized result; represents the sensor sensitivity factor; represents different sensor numbers or data sources, which can be understood as the index of the sensor, ranging from 1 to n; represents the real-time correlation between the sensor and environmental data, which is calculated by the correlation coefficient formula: ;

[0060] Among them, is the and environmental data covariance, and are respectively the and standard deviations;

[0061] Sensor data ( ): represents the preprocessed data of each sensor, including weighing data, environmental vibration data, and temperature data; Environmental data ( ): refers to various interference factor data related to the environment, including vibration frequency, temperature, and humidity; Weight coefficient ( ): is dynamically calculated according to the sensor sensitivity and the correlation with environmental data.

[0062] The dynamic compensation unit is used to adjust the sampling parameters of the sensor in real time according to the dynamic compensation strategy generated by the virtual domain data fusion and optimization unit, where:

[0063] The dynamic compensation includes adjusting the sensitivity, sampling frequency, and data processing priority of the sensor;

[0064] The dynamic compensation unit forms a closed-loop feedback with the physical domain sensor integration unit;

[0065] The distributed computing unit is used to support the collaborative work of each unit, where: the edge computing unit is used to process the data of the physical domain sensor integration unit and the environmental domain monitoring unit; the central computing unit is used to run the multi-domain collaborative twin model and generate a dynamic compensation strategy; the cloud optimization unit is used to integrate historical data and large-scale environmental characteristic data to optimize the twin model.

[0066] As an improvement to the above technical solution, the wavelet decomposition algorithm in the edge computing unit includes at least three levels of decomposition. The decomposition basis function is selected as the Daubechies wavelet basis, and it is specifically implemented through the following steps: performing discrete wavelet transform on the collected original signal; extracting high-frequency components for vibration feature analysis; and combining low-frequency components for tilt signal correction to finally generate dynamic feature signals.

[0067] As an improvement to the above technical solution, the normalization process in the environmental domain monitoring unit uses the maximum-minimum normalization algorithm, through the formula:

[0068] Calculated, where and The values of are determined according to the real-time data statistics results of the environmental domain monitoring unit.

[0069] As an improvement to the above technical solution, the model training method of the recurrent neural network combined with the attention mechanism in the virtual domain data fusion and optimization unit includes the following steps: constructing a multi-dimensional data set containing historical weighing data, vibration features, and environmental parameters; designing a recurrent neural network structure with an attention mechanism for dynamic weight allocation; and optimizing the network weights through the error backpropagation algorithm, with the objective function being the minimization of the root mean square value of the prediction error.

[0070] As an improvement to the above technical solution, the calculation steps of the weight coefficient include: calculating the mean and standard deviation of the real-time collected data to determine the initial weight; constructing a covariance matrix to analyze the correlation between input features; and normalizing the weight based on the gradient descent method, with the range limited between 0 and 1.

[0071] As an improvement to the above technical solution, the sensor sampling frequency adjustment algorithm in the dynamic compensation unit uses negative feedback control, where the feedback signal is the deviation between the vibration amplitude and the standard threshold, and it is specifically implemented through the following method: calculating the amplitude of the real-time vibration signal; comparing with the set threshold to determine the sampling frequency adjustment amplitude; and dynamically updating the sampling frequency within the range of 100 Hz to 1 kHz.

[0072] As an improvement to the above technical solution, the objective function of the closed-loop control module is:

[0073] Where represents the system error, represents the system energy consumption, and The value of is obtained by fitting the training data. The specific optimization goal is to achieve a balance between system error and energy consumption. The way to achieve this balance is to introduce a dynamic optimization process, which performs double optimization of error and energy consumption by combining reinforcement learning (RL) and recurrent neural network (RNN). The specific optimization path is as follows: the design of the objective function: in this technical solution, the core objective function of system optimization is: ;

[0074] where represents the system error, represents the system energy consumption, and are the weight coefficients of error and energy consumption respectively. The purpose of designing this objective function is to reduce the energy consumption of the system while ensuring that the system error is as low as possible. The specific values of the weight coefficients and are dynamically adjusted according to the real-time state of the system to achieve the optimal balance between error and energy consumption.

[0075] Dynamic adjustment of weight coefficients: To achieve a balance between error and energy consumption, and weight coefficients need to be dynamically adjusted according to real-time data. The specific adjustment strategy is as follows:

[0076] Error impact adjustment: When the system error ( ) exceeds the set threshold, the weight coefficient will increase, and the system will give priority to reducing the error, which may be achieved by increasing the sampling frequency or enhancing the sensor sensitivity. At this time, the energy consumption may increase slightly, but the system accuracy is guaranteed first.

[0077] Energy consumption impact adjustment: When the energy consumption of the system ( ) exceeds the set threshold, will increase, and the system will give priority to reducing the energy consumption by reducing the sampling frequency or adjusting the sensor sensitivity, while moderately allowing the error to increase. In this case, the system optimizes the sampling frequency and sensitivity to keep the energy consumption within an acceptable range.

[0078] Adaptive adjustment: When the system is in the normal operation state and both the error and energy consumption are within the predetermined range, and values will be balanced through learning historical data to ensure that the system always maintains the optimal balance between error and energy consumption under different working conditions.

[0079] Reinforcement Learning and Objective Function Optimization: The system optimizes the objective function through the Reinforcement Learning (RL) algorithm. The states in reinforcement learning include sensor sensitivity, sampling frequency, current error, and energy consumption. The actions include adjusting the sampling frequency and sensitivity of the sensors. The reward function is based on the objective function to evaluate the quality of the optimization strategy: ;

[0080] Among them, and are balance coefficients used to control the priorities of error and energy consumption. When the error or energy consumption is high, the system will dynamically adjust and values through the RL algorithm to achieve the best balance between error and energy consumption.

[0081] Combination of Recurrent Neural Network (RNN) and Reinforcement Learning, RNN for Time-Series Feature Extraction: The Recurrent Neural Network (RNN) is used to analyze the time-series features of historical data and real-time data to help the system predict the impact of environmental changes on weighing data. The output of the RNN will be used as the input of the RL algorithm to provide information on how to adjust the sensor sensitivity and sampling frequency.

[0082] RL for Dynamic Policy Adjustment: Through the reinforcement learning algorithm, according to real-time data (such as temperature, humidity, vibration, etc.) and historical data, dynamically adjust the sensitivity and sampling frequency of the sensors. The output of RL will directly affect the balance between error and energy consumption in the system.

[0083] Feedback Mechanism and Closed-Loop Control: The system adopts a negative feedback control mechanism to monitor the changes in error and energy consumption of sensor data in real time, ensuring that the system automatically adjusts parameters during operation to maintain the balance between error and energy consumption. The feedback mechanism adjusts the sampling frequency and sensitivity of the system by calculating the real-time difference between error and energy consumption to achieve dynamic compensation. Through this optimization strategy, the system can significantly reduce energy consumption while ensuring high-precision weighing, ensuring the best performance and efficiency in a complex industrial environment.

[0084] As an improvement of the above technical solution, the message queue mechanism in the distributed computing unit is implemented through an event-driven mode, where the event triggers include the following conditions: the sensor signal exceeds the set threshold; a notification signal is generated after the data fusion task is completed.

[0085] As an improvement of the above technical solution, the cut-off frequency range of the high-pass filter in the physical domain sensor integration unit is 10Hz to 50Hz, which is used to remove low-frequency interference signals.

[0086] As an improvement of the above technical solution, the message queue mechanism is implemented based on the MQTT protocol, specifically including the following steps: realizing data classification transmission through multi-topic publishing; the subscription end parses and validates the received data; ensuring data synchronization and consistency of multiple sensor nodes.

[0087] Embodiment 1:

[0088] In this embodiment, the highway dynamic weighing system is used as a specific application scenario to elaborate in detail how the technical solution of the present invention realizes high-precision dynamic weighing in a complex environment and significantly improves real-time performance and system stability. For example, at a certain highway toll station, the vehicle passing speeds vary, and the environment changes drastically, including significant day-night temperature differences, complex road surface vibrations, etc. The traditional static weighing method cannot meet the real-time detection requirements, and the existing dynamic weighing technologies have low precision and poor stability under high-frequency vibrations and temperature changes. This embodiment is based on the dynamic twin model of multi-domain collaboration and combines multi-algorithm optimization to solve the above problems.

[0089] First, the multi-source sensor network in the physical domain is used to collect wheel pressure, road surface vibration, and temperature data in real time. Specifically, highly sensitive strain sensors and vibration sensors are embedded in the toll station ground, and infrared temperature sensors are set in the measurement area to ensure the multi-dimensionality and accuracy of the collected data.

[0090] The collected data is preliminarily processed by the edge computing unit. The wavelet decomposition algorithm is used to denoise the vibration signal, and the dynamic normalization algorithm is used to smooth the multi-dimensional data differences to avoid data distortion. The high-priority data after processing is transmitted to the central computing unit in real time, and the low-priority data enters the regular compensation path.

[0091] In the central computing unit, the virtual domain of the twin model calculates the vehicle load in real time. The model extracts time series features based on the recurrent neural network (RNN) and combines reinforcement learning (RL) to dynamically adjust the model parameters to adapt to the changing states of high-speed vehicles. For example, when a vehicle passes over a speed bump, the system quickly identifies the interference through the reinforcement learning algorithm and adjusts the weight parameters to eliminate the impact of instantaneous vibrations.

[0092] In addition, this embodiment introduces the environmental domain as a compensation layer to capture the real-time information of day-night temperature differences and humidity changes through environmental sensors to compensate for the errors between the physical domain and the virtual domain. Through the negative feedback mechanism, the real-time parameters of the environmental domain directly correct the dynamic calculation process of the twin model, further improving the overall accuracy of the system.

[0093] The multi-domain collaborative architecture of this technical solution, especially with the support of the dynamic resource scheduling mechanism, significantly reduces data processing latency. Experimental data shows that in the high-speed dynamic weighing scenario at 120 km / h, the weighing error of the system is controlled within 0.03%, the data processing latency is reduced by 40%, and the energy consumption is decreased by about 25%.

[0094] In summary, through the deep combination of the physical domain, environmental domain and virtual domain, as well as the real-time dynamic calculation method with multi-algorithm optimization, this embodiment realizes the high precision, low latency and high reliability of the dynamic weighing system in complex scenarios.

[0095] Example 2:

[0096] In a large intelligent logistics center, for example, materials are dynamically sorted through an automated conveyor belt. The environmental temperature changes greatly, and the vibration frequency of the conveyor belt fluctuates significantly due to different loads. The traditional weighing system cannot effectively eliminate vibration interference, resulting in insufficient data accuracy and poor real-time performance. The present invention solves this problem through a multi-domain collaborative twin model of the physical domain, environmental domain and virtual domain, combined with a multi-level anti-interference and dynamic optimization strategy.

[0097] Experimental setup: High-precision strain sensors and vibration sensors are arranged on a logistics sorting line, for example, to construct a physical domain data acquisition system in combination with temperature and humidity sensors. The edge computing unit uses wavelet decomposition to denoise the vibration signal and adopts a dynamic normalization algorithm to smooth data differences. The virtual domain extracts the dynamic parameters of the materials through a recurrent neural network (RNN) and realizes the adaptive optimization of the model parameters in combination with a reinforcement learning (RL) algorithm. The environmental domain compensates for the real-time changes in temperature and humidity and dynamically corrects the weighing model in the virtual domain.

[0098] Comparative experiment and results: The performance of the traditional static weighing system, conventional dynamic weighing system and the dynamic weighing system of the present invention was compared in the experiment. Under the conditions of an environmental temperature fluctuation range of -5°C to 40°C and a vibration frequency of 10 Hz to 100 Hz, the weighing error, data processing latency and energy consumption performance were tested. The results show that the weighing error of the present invention is controlled within 0.02%, significantly lower than the 0.15% error of the traditional dynamic weighing system; the data processing latency is reduced by about 45%, from 150 ms of the traditional system to 80 ms of the system of the present invention; the system energy consumption is reduced by about 30%, due to the collaborative work of the edge computing unit and the dynamic resource scheduling mechanism. Through multi-domain collaboration and dynamic optimization, this embodiment of the present invention realizes a significant improvement in dynamic weighing accuracy, and at the same time has extremely high real-time performance and environmental adaptability. Especially when dealing with high-frequency vibration and drastic temperature difference changes, the system dynamically adjusts the weighing strategy through the feedback mechanism and model layer compensation, effectively solving the instability problem in the traditional system.

[0099] Example 3:

[0100] In this embodiment, the physical domain sensor integration unit continues to be responsible for collecting multi-source data generated during the dynamic weighing process. To improve the data collection accuracy, this unit uses strain gauge sensors, acceleration sensors, and temperature sensors. Their distribution and layout take into account the interference factors of different environments on the weighing system. The specific layout methods are as follows: strain gauge sensors: installed under the weighing platform to detect the weight of the object; acceleration sensors: placed at both ends of the weighing platform to monitor the vibration information of the platform in real time; temperature sensors: used to collect the ambient temperature for dynamic compensation during subsequent data processing; the data is preprocessed in real time through the edge computing unit, and the preprocessing steps include signal filtering, data normalization, frequency domain feature extraction, and the preliminary application of the wavelet decomposition algorithm. Here, a three-level decomposition scheme of wavelet decomposition is particularly emphasized. The Daubechies wavelet basis used can effectively extract the high-frequency part in the vibration signal for feature analysis.

[0101] To further improve the perception and adaptability to environmental changes, the environmental domain monitoring unit is responsible for real-time monitoring and processing of external environmental parameters. It includes environmental vibration frequency monitoring: extracting the environmental vibration frequency characteristics through the short-time Fourier transform (STFT); temperature gradient and humidity fluctuation monitoring: using the differential algorithm to calculate the temperature gradient and humidity change rate to ensure the accurate tracking of environmental changes; the collection and analysis of all environmental parameters are preliminarily preprocessed through the edge computing unit to reduce data latency.

[0102] And the core of the virtual domain data fusion and optimization unit lies in constructing a multi-domain collaborative dynamic twin model. Through the combination of the recurrent neural network (RNN) and reinforcement learning (RL), dynamic data compensation and adjustment are realized. The specific process is as follows: recurrent neural network (RNN): used to extract the long-term dependencies in time series data and predict the impact of environmental changes on weighing data; reinforcement learning (RL): based on the fusion of historical data and real-time data, dynamically adjust the sensitivity and sampling frequency of the sensors; data fusion formula: ;

[0103] where represents the preprocessed data of the th sensor, is the sensor weight coefficient, and the calculation formula of the weight coefficient is:

[0104] ;

[0105] where is the sensor sensitivity factor, represents different sensor numbers; is the real-time correlation between the sensor and the environmental data. The correlation is calculated by the following covariance formula: ;

[0106] wherein, is the covariance between the sensor data and the environmental data, and are the standard deviations of the two respectively.

[0107] The dynamic compensation unit adopts a negative feedback control algorithm to adjust the sensor sampling frequency and sensitivity in real time. This compensation unit forms a closed-loop feedback with the physical domain sensor integration unit to ensure real-time adjustment of the sensor performance to cope with environmental interference; and sampling frequency adjustment: according to the deviation between the vibration amplitude and the preset threshold, the sampling frequency is dynamically adjusted within the range of 100 Hz to 1 kHz.

[0108] The distributed computing unit supports the collaborative work between modules. The edge computing unit performs preliminary processing on the data, the central computing unit is responsible for running the multi-domain collaborative twin model and generating compensation strategies, and the cloud optimization unit integrates historical data for large-scale optimization. The message queue mechanism is implemented based on the MQTT protocol, and data is published through multiple topics to ensure data synchronization and consistency. Also, in terms of hardware requirements, for example, the physical domain sensor integration unit requires strain gauge sensors, acceleration sensors, and temperature sensors with high precision, and is equipped with edge computing devices to support real-time preprocessing of data; while the software requirements are that the system needs to support the real-time computing capabilities of RNN and RL algorithms, and at the same time has a data fusion and feedback control mechanism. Its execution steps are as follows: First step, the sensor collects data and performs preprocessing through the edge computing unit; Second step, the environmental domain monitoring unit monitors and processes external interference information in real time; Third step, the virtual domain data fusion and optimization unit adjusts the compensation strategy according to the real-time data; Finally, the dynamic compensation unit adjusts the sampling frequency in real time through the negative feedback mechanism to ensure the stability and high precision of the system.

[0109] Embodiment 4:

[0110] This embodiment is based on the technical solution of Embodiment 3. Through experimental verification and supplementary description, it demonstrates the implementation path and effect of the technical solution in practical applications. For example, in a dynamic weighing system in a certain logistics park, to verify the effectiveness of the technical solution, the following experimental design was carried out, and its experimental conditions:

[0111] Ambient temperature: -10°C to 50°C; Vibration frequency: 5Hz to 200Hz; Vehicle speed: 0 to 120 km / h; In terms of system settings, sensor configuration: including high-sensitivity strain gauge sensors, triaxial acceleration sensors, and temperature and humidity sensors; Edge computing unit: running wavelet decomposition algorithm and dynamic normalization algorithm; Central computing unit: executing multi-domain collaborative twin model; Test data: comparing the weighing accuracy, latency, and energy consumption of traditional static weighing systems, existing dynamic weighing systems, and the system of the present invention.

[0112] The experimental results show that for weighing accuracy: the weighing error of the system of the present invention is 0.02%, which is better than that of the traditional system (0.15%) and the existing dynamic system (0.05%); For data latency: the latency is reduced from 120 ms of the traditional system to 80 ms; For energy consumption optimization: the dynamic compensation strategy reduces the overall energy consumption by 30%.

[0113] And to further illustrate the implementation path of the Recurrent Neural Network (RNN) and Reinforcement Learning (RL) in the dynamic compensation strategy, the following specific steps are adopted in this embodiment. Data preprocessing: The collected weighing data and environmental parameters (vibration frequency, temperature and humidity) are normalized by the edge computing unit; Feature extraction: Wavelet decomposition is used to denoise the high-frequency signal; RNN model construction: A time series feature dataset is constructed, including historical weighing data and real-time environmental parameters; Model structure: The hidden layer of the RNN uses GRU (Gated Recurrent Unit) to improve the training efficiency; Training objective: Minimize the Mean Squared Error (MSE) to optimize the time series prediction of the weighing data; And RL strategy optimization, State: The current sampling frequency and weighing error of the system; Action: Adjust the sensor sampling frequency and sensitivity; Reward function: According to the weighing error and energy consumption optimization target, dynamically balance the system performance. Its combination logic is that the RNN predicts the impact of environmental parameter changes on the weighing data and provides a reference input; The RL generates an optimal compensation strategy in real time according to the prediction result of the RNN.

[0114] At the same time, to adapt to large-scale data calculation, this embodiment simplifies the description of the data fusion formula, clarifies the applicable scope and defines the calculation logic. Fusion formula: ;

[0115] Where: is the sensor weight, which has been clearly defined in Embodiment 3. When there are a large number of sensors, a parallel computing mode is adopted to process the weight coefficient and data weighted summation in partitions.

[0116] : The weight coefficient of the th sensor, indicating the importance of this sensor in data fusion. The weight is determined by the sensitivity of the sensor and its Joint decision.

[0117] : The correlation between the sensor and environmental data, which represents the degree of correlation between the output signal of the sensor and the environmental interference signal. Its value usually ranges from 0 to 1, and the larger the value, the higher the correlation.

[0118] : The sensitivity factor of the sensor, which represents the response degree of the sensor to the input signal. The range of the sensor sensitivity factor is limited (0.1 to 1.0) to ensure calculation stability. The present invention strengthens the following technical effects by introducing experimental data and optimizing logic: for example, by combining RNN and RL, the adaptability of the system to environmental changes is significantly improved; it can predict in advance the impact of vibration and temperature changes on weighing accuracy and adjust sampling parameters; the dynamic compensation strategy effectively reduces system power consumption, especially prominent in high-frequency vibration scenarios. Experimental results prove that the system of the present invention can maintain high precision and low latency in various complex scenarios.

[0119] Example 5:

[0120] This example introduces a method for generating a multi-objective dynamic compensation strategy, which ensures the accuracy and real-time performance of the system in high-dynamic scenarios through a reinforcement learning (RL) model and a real-time feedback mechanism. For example, in traditional compensation strategies, minimizing error is the main goal. This example further incorporates system energy consumption optimization into the objective function and uses the following multi-objective optimization formula: ;

[0121] Where: : System error, calculated from the real-time and predicted errors of weighing data; : System energy consumption, estimated by the sensor sampling frequency and the power consumption of the computing node; and : Respectively represent the weight coefficients of error and energy consumption, obtained by dynamic fitting of historical data, with a range of 0.1 to 1.0.

[0122] For its specific implementation steps, state variable design: The sensitivity factor, sampling frequency, and system error of the current sensor are used as the input states of the reinforcement learning model; action space design: Define the action as the dynamic adjustment of the sensor sampling frequency and sensitivity; reward function design: Among them, and are balance coefficients, used to dynamically adjust the priority between error and energy consumption.

[0123] And, in another optimization path, such as the dynamic adaptation method of variable weights, to improve the fusion accuracy of multi-source data, in this embodiment, the calculation formula of the weight coefficient is improved, such as the distributed calculation of the weight coefficient, the weight coefficient The calculation formula is:

[0124] ;

[0125] Where: : The real-time correlation between sensor data and environmental data; : The sensitivity factor, and its dynamic value is calculated by the following formula: ; : The adjustment coefficient, which controls the sensitivity adjustment rate; : The standard deviation of sensor data; : The mean value of multi-sensor data; in its multi-threaded calculation optimization, the weight calculation adopts distributed parallel processing. After partitioning the sensor data, the weights are calculated independently, and finally merged by the central node to reduce the calculation delay.

[0126] And this embodiment takes the logistics sorting center as the experimental scenario to verify the optimization path:

[0127] Experimental environment, temperature range: -20°C to 50°C; vibration frequency: 10Hz to 120Hz; vehicle speed: 0 to 120km / h; sensor configuration: high-sensitivity strain gauge sensors, triaxial acceleration sensors, and temperature and humidity sensors.

[0128] Its experimental design and data are that the experimental samples include 10 vehicles in total, including small cars, medium-sized trucks, and heavy trucks, with a load range from 1 ton to 40 tons. They are tested under static, low-speed (10km / h), medium-speed (60km / h), and high-speed (120km / h) scenarios. Each group of conditions is tested 10 times, and the average value is taken.

[0129] The experimental results show the weighing accuracy: the error is reduced from 0.03% of the traditional system to 0.015%, and the standard deviation is 0.005%; the processing delay: is reduced from the traditional 90ms to 60ms, mainly due to the optimization of the distributed computing mechanism; the energy consumption: the average energy consumption under multiple conditions is reduced by 20%, and can reach up to 30% at most, especially prominent in the high-frequency vibration scenario.

[0130] Example 6:

[0131] To further optimize the dynamic weighing system mentioned in the present invention, this embodiment proposes a dynamic compensation strategy, with the focus on enhancing the response accuracy to environmental changes and improving its real-time performance and energy efficiency through system-level optimization algorithms. This embodiment particularly emphasizes how to accurately capture environmental disturbances and perform dynamic compensation on the weighing results under complex environmental conditions to maximize the reliability and stability of the system.

[0132] In this embodiment, the dynamic weighing system still collaborates based on modules such as the physical domain sensor integration unit, environmental domain monitoring unit, virtual domain data fusion and optimization unit, dynamic compensation unit, and distributed computing unit. However, in this embodiment, the dynamic compensation algorithm for the data processing part is particularly strengthened, and an environmental adaptive feedback mechanism is introduced, enabling the system to automatically adjust parameters and maintain a high-precision weighing function when environmental conditions change.

[0133] In this system, the physical domain sensor integration unit continues to undertake the task of collecting multi-source data, including the weighing data of objects, environmental vibration data, and temperature data. All sensors are installed in protective devices that can withstand complex environmental changes to ensure their stability.

[0134] The environmental domain monitoring unit analyzes environmental parameters through the short-time Fourier transform (STFT) and differential algorithms, monitors temperature differences, humidity changes, and vibration frequencies in real time, and generates corresponding environmental interference characteristics. These characteristics are then transmitted to the virtual domain data fusion and optimization unit for subsequent compensation strategy design.

[0135] The edge computing unit is responsible for the preliminary preprocessing of the collected data. This process includes operations such as signal filtering, data normalization, and frequency domain feature extraction, especially the denoising process of high-frequency vibration signals. The wavelet decomposition algorithm is fully applied in this process to separate the high-frequency part of the vibration signal and reduce the error caused by vibration.

[0136] The dynamic compensation unit dynamically adjusts the sampling frequency and sensitivity of the sensor according to the environmental interference data provided by the edge computing unit through a negative feedback control mechanism. Its purpose is to ensure that when the environmental interference is large, the system can automatically adjust the sampling frequency and reduce the measurement error. This feedback mechanism ensures that the system has high adaptability and accuracy.

[0137] The virtual domain data fusion and optimization unit combines the recurrent neural network (RNN) and reinforcement learning (RL) algorithms to optimize the data fusion process of the sensor. The RNN network extracts the influence of environmental changes on the weighing data through time series analysis, while RL optimizes the compensation strategy through training. Specifically, the data fusion formula is: ;

[0138] Among them, is the preprocessed data of the i-th sensor, is the weight coefficient of the corresponding sensor, and the calculation formula is as follows:

[0139] ;

[0140] is the real-time correlation between the sensor and the environmental data, and the covariance formula is:

[0141] ;

[0142] In the above formula, represents the environmental data, is the sensor data, and the covariance is used to measure the relationship between environmental interference and sensor data, while the standard deviation and are used to standardize the covariance. And in this embodiment, reinforcement learning (RL) is used to generate a dynamic compensation strategy. The state variables of RL include the current sensor sensitivity, sampling frequency, and system error; the action variables are to adjust the sampling frequency and sensitivity; the reward function combines the system error and the energy consumption optimization goal to generate an optimal strategy. The formula is as follows: ;

[0143] Among them, and are balance coefficients, which are used to dynamically adjust the priority between error and energy consumption according to the current environmental state and system requirements. At the same time, in order to ensure the real-time performance and data consistency of the system, the distributed computing unit works in coordination through the event-driven mode and the message queue mechanism. The data in the system is classified and transmitted through the MQTT protocol to ensure data synchronization and consistency among multiple sensor nodes.

[0144] Embodiment 7:

[0145] This embodiment further optimizes the key parts of the dynamic weighing system, and focuses on solving the challenges in aspects such as data fusion, dynamic compensation, feedback regulation, and system optimization, ensuring that the system can operate stably, efficiently, and accurately under complex environmental conditions. Specifically, the system uses high-precision strain gauge sensors, acceleration sensors, and temperature and humidity sensors to collect multi-source data related to weighing in real time. The signals collected by all sensors will be preprocessed by the edge computing unit. The preprocessing steps include:

[0146] Signal filtering: A high-pass filter is used to remove low-frequency noise signals, ensuring that only the valid information related to the weighing data is transmitted; Data normalization: The output data of all different sensors will be standardized through the maximum-minimum normalization algorithm to ensure that the data of different sensors can be compared on the same scale; Feature extraction: The wavelet decomposition algorithm is used to decompose the vibration signal into three levels, and the high-frequency components are extracted for further vibration analysis, while the low-frequency components are used to correct the tilt signals in the data; Through the above steps, the system can eliminate the noise and instability in the data, ensuring the consistency and reliability of all the collected data.

[0147] To address the impact of the external environment (such as temperature changes, humidity fluctuations, and vibration interference) on the weighing data, the environmental domain monitoring unit collects environmental data in real time, including vibration frequency, temperature gradient, and humidity fluctuations. The environmental data is processed through the following methods:

[0148] Short-time Fourier transform (STFT): It is used to extract the environmental vibration frequency characteristics, identify and quantify the impact of vibrations at different frequencies on the weighing accuracy; Differential algorithm: Calculate the temperature gradient and humidity change rate, monitor the changes in environmental conditions in real time, and generate environmental interference characteristics; These environmental data will be fused with the physical domain (sensor data) through the following steps:

[0149] Weighted fusion: The data of each sensor will be assigned different weights according to its sensitivity and correlation with environmental interference. Sensors with a greater environmental impact will receive higher weights; Data fusion formula: The weighted data will be fused into the final weighing data through the weighted average method to ensure that the weighing result accurately reflects the impact of environmental changes; Through this data fusion technology, the system can more accurately handle the interference of complex environments on the weighing data and improve the accuracy of the overall weighing system.

[0150] At the same time, the system introduces a dynamic compensation strategy to adjust the sampling frequency and sensitivity of the sensors in real time through a feedback mechanism to ensure that the weighing accuracy is not affected by environmental interference:

[0151] Dynamic compensation strategy: When the system detects changes in the environmental vibration frequency or temperature fluctuations, the compensation strategy will automatically adjust the sampling frequency of the sensors. For example, when the vibration increases, the system will increase the sampling frequency to capture more details and ensure accuracy; Negative feedback control: The system continuously monitors the deviation between the sensor data and the standard value through a negative feedback mechanism and automatically adjusts the parameters to reduce errors. The feedback signal determines the adjustment range of the sampling frequency by calculating the deviation between the amplitude of the real-time vibration signal and the set threshold, and the adjustment range is from 100Hz to 1kHz; This mechanism enables the system to respond to environmental changes in real time, automatically adjust the working parameters, and ensure the weighing accuracy.

[0152] In addition, on the basis of dynamic compensation, this embodiment introduces an energy efficiency optimization strategy to ensure that the system can reduce energy consumption while maintaining high precision, such as the balance between error and energy consumption: while the system monitors the weighing error and energy consumption in real time, it balances the two through an adaptive adjustment strategy. Specifically, when the system error increases, the sampling frequency will be preferentially increased to improve the weighing accuracy; while when the energy consumption increases, the sampling frequency is reduced to save energy; Energy efficiency optimization formula: The system ensures low energy consumption without sacrificing accuracy during operation through an adaptive adjustment mechanism. The adjustment process matches the changes in sampling frequency and sensitivity with real-time data to ensure optimal energy efficiency. This optimization strategy reduces unnecessary energy consumption while the system operates in a high-precision mode.

[0153] To further improve the adaptability and optimization ability of the system, this embodiment combines a Recurrent Neural Network (RNN) and Reinforcement Learning (RL). Recurrent Neural Network (RNN): used to analyze time series data, especially to extract long-term dependencies from historical weighing data and environmental parameters, helping the system predict the impact of future environmental changes on weighing results; Reinforcement Learning (RL): based on the time series features extracted by the RNN, the reinforcement learning algorithm is used to optimize the compensation strategy. The system adaptively adjusts the sensor sensitivity and sampling frequency through reinforcement learning to dynamically generate an optimal compensation strategy; through the combination of RNN and RL, the system can optimize data collection and compensation strategies under different environmental conditions, improving the accuracy and efficiency of the system.

[0154] Embodiment 8:

[0155] This embodiment details how to use a multi-domain collaborative model to optimize data fusion and dynamic compensation strategies to ensure accurate weighing of the system in a complex environment. For example, the data collection and preprocessing module is mainly responsible for the collection and preprocessing of multi-source data such as the weight of the object, environmental vibration, and temperature during the dynamic weighing process. To ensure data quality, high-precision strain gauge sensors, acceleration sensors, and temperature sensors are used and installed at appropriate positions on the weighing platform, near the vibration source, and in the environment.

[0156] The collection process is as follows: Weighing data collection: The strain gauge sensor is directly installed under the weighing platform to detect the wheel pressure in real time and convert it into weight data; Environmental vibration data collection: The acceleration sensors are arranged at both ends of the weighing platform to collect vibration signals generated by the passing of the vehicle; Temperature data collection: The temperature sensor is installed in the surrounding environment to monitor temperature fluctuations in real time to ensure compensation under temperature difference changes.

[0157] The data preprocessing process is as follows: The wavelet decomposition algorithm is used to denoise the vibration signal, and the short-time Fourier transform (STFT) is used to extract the environmental vibration frequency characteristics; The data normalization uses the maximum-minimum normalization algorithm, through the formula:

[0158] Ensure that the various sensor data collected are compared and fused under the same dimension.

[0159] Then, adopt a multi-domain collaborative data fusion method to integrate the collected data from the physical domain, environmental domain, and virtual domain into a unified model. To process high-dimensional data and eliminate redundant noise, the weighted average method is used to weight the data of each sensor, and the formula is as follows: ;

[0160] Where, is the preprocessed data of the th sensor, is the weight coefficient of this sensor, defined as:

[0161] ;

[0162] Where, is the real-time correlation between the th sensor and the environmental data , and the calculation formula is:

[0163] ;

[0164] After the data fusion is completed, the system automatically adjusts the sampling frequency and sensitivity of the sensors based on the combined action of reinforcement learning (RL) and recurrent neural network (RNN). The compensation process uses negative feedback control to ensure that the system can make real-time adjustments when the environment changes:

[0165] Feedback mechanism: Monitor the amplitude of the vibration signal in real time, compare it with the set threshold, and dynamically adjust the sampling frequency of the sensor according to the error to ensure the measurement accuracy; Sampling frequency adjustment formula: Assume that the amplitude of the vibration signal is , the standard threshold is , and the adjustment range of the sampling frequency is from 100 Hz to 1 kHz:

[0166] ;

[0167] Where the Adjust function is used to dynamically calculate the required sampling frequency based on the difference between the current data of the sensor and the set threshold . The implementation method of this function can be based on simple proportional control or adjust the frequency according to the signal amplitude difference, aiming to improve the response ability of the system in a rapidly changing environment. For example, when the system detects a large signal difference, it will automatically increase the sampling frequency to improve the accuracy of capturing data details.

[0168] The role of reinforcement learning (RL) in this embodiment is to dynamically adjust the weight coefficient, sampling frequency, and sensitivity, enabling the system to achieve the best balance between error and energy consumption. The state variables include the current sensor sensitivity, sampling frequency, error, and energy consumption. The reward function is designed as: ;

[0169] where represents the weighing error of the system, represents the energy consumption of the system, and are balance coefficients that adjust the priorities of error and energy consumption according to the real-time environment.

[0170] The role of the RNN in this embodiment is to analyze time-series data, extract the relationship between historical data and real-time data, especially to predict the impact of environmental changes on weighing data. The output of the RNN is passed to the RL module to guide the dynamic adjustment of the sampling frequency and sensitivity.

[0171] Its hardware requirements are such as high-precision strain gauge sensors, acceleration sensors, and temperature and humidity sensors, which can collect multi-source data generated during the dynamic weighing process in real time; the edge computing unit is required to have strong data processing capabilities to be able to process sensor data in real time and perform preprocessing; the central computing unit needs to run a multi-domain collaborative twin model and dynamically adjust system parameters in combination with the RL optimization strategy. Also, its software requirements are that the system needs to support real-time calculations based on algorithms of recurrent neural network (RNN) and reinforcement learning (RL), and have data fusion, dynamic compensation, and feedback control mechanisms to be able to adjust the sensor sampling frequency and sensitivity in real time to ensure the accuracy and stability of the system in complex environments, all of which belong to the extended implementation methods known to those of ordinary skill in the art.

[0172] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes completed by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. An integrated sensor module for a dynamic weighing system, characterized in that: This module includes: The physical domain sensor integration unit is used to collect multi-source data generated during dynamic weighing, including: The multi-source data includes object weighing data, environmental vibration data and temperature data; The physical domain sensor integration unit includes a strain gauge sensor, an acceleration sensor and a temperature sensor; The physical domain sensor integration unit preprocesses the collected data through the edge computing unit, and the preprocessing includes signal filtering, data normalization and frequency domain characteristic extraction; The environmental domain monitoring unit is used to collect and process environmental parameters in real time, including: The environmental parameters include environmental vibration frequency, temperature gradient and humidity fluctuation; The unit extracts the environmental vibration frequency characteristics through short-time Fourier transform (STFT), calculates the temperature gradient and humidity change rate through a differential algorithm, and outputs the environmental interference characteristics; The virtual domain data fusion and optimization unit is used to build a multi-domain collaborative dynamic twin model based on the data of the physical domain sensor integration unit and the environmental domain monitoring unit, wherein: The dynamic twin model includes: A dynamic prediction module based on a recurrent neural network (RNN) is used to predict the impact of vibration on weighing data based on the characteristics of environmental disturbances; An optimization module based on reinforcement learning (RL) is used to generate dynamic compensation strategies, where: The dynamic compensation strategy is generated based on the fusion of historical data and real-time data, and is used to dynamically adjust the sensitivity and sampling frequency of the sensor; The data fusion formula is: in, It refers to the data fusion result after weighted summation, which represents the comprehensive calculation value of multi-source sensor data; Indicates Preprocessing data of sensors, Indicates the weight coefficient of the corresponding sensor, weight coefficient The calculation formula is: in, represents the sensor sensitivity factor; Indicates different sensor numbers; Indicates the real-time correlation between sensor and environmental data, calculated by the correlation coefficient formula: in, for and environmental data The covariance of and They are and The standard deviation of The dynamic compensation unit is used to adjust the sampling parameters of the sensor in real time according to the dynamic compensation strategy generated by the virtual domain data fusion and optimization unit, wherein: The dynamic compensation includes adjusting the sensitivity, sampling frequency and data processing priority of the sensor; The dynamic compensation unit and the physical domain sensor integration unit form a closed-loop feedback; Distributed computing units are used to support the collaborative work of various units, including: the edge computing unit is used to process data from the physical domain sensor integration unit and the environmental domain monitoring unit; the central computing unit is used to run the multi-domain collaborative twin model and generate dynamic compensation strategies; the cloud optimization unit is used to integrate historical data and large-scale environmental characteristic data to optimize the twin model.

2. The integrated sensor module for a dynamic weighing system according to claim 1, characterized in that: The wavelet decomposition algorithm in the edge computing unit includes at least three levels of decomposition. The decomposition basis function uses the Daubechies wavelet basis, which is implemented through the following steps: performing discrete wavelet transform on the collected original signal; extracting high-frequency components for vibration feature analysis; and combining low-frequency components to perform tilt signal correction.

3. The integrated sensor module for a dynamic weighing system according to claim 1, characterized in that: The normalization process in the environmental domain monitoring unit adopts the maximum and minimum normalization algorithm, through the formula Calculate, where and The values ​​of are the statistical results of the real-time data collected by the environmental domain monitoring unit, which are the minimum and maximum values ​​in the statistical result data respectively.

4. The integrated sensor module for a dynamic weighing system according to claim 1, characterized in that: The model training method of the recursive neural network combined with the attention mechanism in the virtual domain data fusion and optimization unit includes the following steps: constructing a multidimensional data set containing historical weighing data, vibration characteristics and environmental parameters; designing a recursive neural network structure with an attention mechanism for dynamic weight allocation; optimizing the network weights through the error back propagation algorithm, and the objective function is to minimize the root mean square value of the prediction error.

5. The integrated sensor module for a dynamic weighing system according to claim 4, characterized in that: The calculation steps of the weight coefficient include: calculating the mean and standard deviation of the real-time collected data to determine the initial weight; constructing a covariance matrix to analyze the correlation between the input features; and normalizing the weights based on the gradient descent method, with the range limited to between 0 and 1.

6. The integrated sensor module for a dynamic weighing system according to claim 1, characterized in that: The sensor sampling frequency adjustment algorithm in the dynamic compensation unit adopts negative feedback control, in which the feedback signal is the deviation between the vibration amplitude and the standard threshold. It is implemented in the following ways: calculating the amplitude of the real-time vibration signal; comparing with the set threshold to determine the sampling frequency adjustment amplitude; dynamically updating the sampling frequency in the range of 100Hz to 1kHz.

7. The integrated sensor module for a dynamic weighing system according to claim 1, characterized in that: The message queue mechanism in the distributed computing unit is implemented through an event-driven model, where event triggers include the following conditions: sensor signals exceed the set threshold; A notification signal is generated after the data fusion task is completed.

8. The integrated sensor module for a dynamic weighing system according to claim 1, characterized in that: The cut-off frequency range of the high-pass filter in the physical domain sensor integration unit is 10 Hz to 50 Hz, which is used to remove low-frequency interference signals.

9. The integrated sensor module for a dynamic weighing system according to claim 1, characterized in that: The message queue mechanism is implemented based on the MQTT protocol, which specifically includes the following steps: realizing data classification transmission through multi-topic publishing; the subscriber parses and verifies the received data; and ensures data synchronization and consistency of multiple sensor nodes.

Images