Intelligent self-adaptive electronic scale system based on machine learning and data processing method

Abstract

The invention belongs to the technical field of intelligent weighing, and more particularly relates to an intelligent self-adaptive electronic scale system based on machine learning and a data processing method, and the system comprises a composite sensor module which is used for collecting gravity, volume and shape information; the automatic calibration module is connected with the composite sensor module and is used for carrying out regular calibration and calibration according to the data of the composite sensor module; the wireless networking module is connected with the automatic calibration module and is used for wirelessly transmitting and interacting the calibrated data; the intelligent operation and maintenance module is connected with the wireless networking module and is used for receiving the transmitted data and realizing equipment state monitoring and anomaly detection; the environment adaptation module is connected with the intelligent operation and maintenance module and is used for realizing adaptive compensation in a severe environment according to a detection result; and the man-machine interaction module is connected with the environment adaptation module and is used for displaying a compensated result and realizing a user-friendly operation interface, so that high-precision and all-around weighing perception is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent weighing, and more specifically, relates to an intelligent adaptive electronic scale system and a data processing method based on machine learning. Background Art

[0002] Electronic scales are weighing instruments that use electronic technology to measure weight. They are widely used in industrial production, trade, transportation and other fields. With the advancement of science and technology and the increase in weighing needs, electronic scales have also experienced a development process from analog to digital, from stand-alone to networked, and from ordinary to intelligent. However, there are still some problems that need to be solved in the current electronic scale technology, which restricts its further development.

[0003] First, most traditional electronic scales use a single weighing sensor, such as a strain sensor or a capacitive sensor. Such sensors are easily affected by external vibrations, electromagnetic interference and other factors, resulting in unstable measurement accuracy. Moreover, a single sensor can only obtain gravity information and cannot fully perceive other properties of the measured object, such as volume and shape, which affects the intelligence level of the electronic scale.

[0004] Secondly, the existing electronic scale calibration method is relatively backward. Most electronic scales require manual regular calibration, which is time-consuming and laborious. The calibration algorithms of a few electronic scales with automatic calibration functions are also relatively simple, such as using the least squares method to fit the calibration curve, which is difficult to adapt to complex error models. When the external environment changes drastically, such as a sudden drop in temperature or a sudden increase in humidity, the original calibration parameters will lose their effectiveness, resulting in inaccurate measurements.

[0005] Thirdly, most of the current electronic scales work independently and lack the ability to interconnect. Although some electronic scales support networking, they mainly use traditional wired communication methods such as RS232 and RS485, which have poor networking flexibility and high wiring costs. The wireless electronic scales that emerged in recent years still need to improve their data transmission reliability and security. In general, the level of networking and clustering of electronic scales is still relatively limited.

[0006] In addition, the predictive maintenance and fault diagnosis capabilities of electronic scales are still lacking. Most electronic scales can only passively wait for faults to occur, but lack active warnings and preventive measures. Traditional fault diagnosis mainly relies on manual experience or adopts simple threshold comparison strategies, which are difficult to adapt to the increasingly complex structure and working conditions of electronic scale systems. As the core components of electronic scales age, their measurement performance will gradually decrease, and this process is often overlooked.

[0007] In addition, harsh working conditions are also an important factor that restricts the performance of electronic scales. When electronic scales are placed in extreme environments such as high temperature, humidity, dust, and strong corrosion, their measurement accuracy will be significantly reduced and their service life will be greatly shortened. Although some electronic scales have taken appropriate protective measures, the protective effect is limited, and there is a lack of intelligent environmental compensation methods, which cannot meet the application requirements of harsh working conditions.

[0008] Finally, the human-computer interaction of electronic scales is not user-friendly enough. Many electronic scales still use the traditional button + digital tube interaction method, with a simple interface and cumbersome operation. Even for electronic scales with touch screens, their interface design is not beautiful and friendly enough, and their functions are relatively simple. The application of intelligent interaction technologies such as voice interaction and gesture recognition in the field of electronic scales is still very limited and urgently needs to be broken through.

[0009] In summary, the existing electronic scale technology still has deficiencies in measurement accuracy, intelligence level, networking capability, reliability, adaptability, and ease of use, and it is difficult to meet the increasing demand for weighing. This urgently requires a new intelligent electronic scale system to solve the above technical problems. Summary of the invention

[0010] The present invention is proposed in this context, and aims to address the limitations of the prior art and provide an intelligent adaptive electronic scale system and a data processing method based on machine learning.

[0011] The present invention provides an intelligent adaptive electronic scale system based on machine learning, comprising: a composite sensor module, used for collecting gravity, volume and shape information; an automatic calibration module, connected to the composite sensor module, used for regular calibration and calibration according to the data of the composite sensor module; a wireless networking module, connected to the automatic calibration module, used for wireless transmission and interaction of calibrated data; an intelligent operation and maintenance module, connected to the wireless networking module, used for receiving the transmitted data and realizing equipment status monitoring and abnormality detection; an environmental adaptation module, connected to the intelligent operation and maintenance module, used for realizing adaptive compensation in harsh environments according to detection results; a human-computer interaction module, connected to the environmental adaptation module, used for displaying the compensated results and realizing a user-friendly operation interface; wherein the operation results of the human-computer interaction module are fed back to the composite sensor module for adjusting the acquisition parameters, thereby forming a closed-loop control system.

[0012] Specifically, the composite sensor module includes a pressure sensor array, an ultrasonic sensor and an image sensor; the pressure sensor array is composed of multiple PS-08 pressure sensors, which are used to collect all-round gravity information; the ultrasonic sensor model is US-100, which is used to measure the volume of goods; the image sensor adopts the OV2640 module, which is used to identify the shape characteristics of goods.

[0013] Specifically, a data fusion unit is provided in the composite sensor module, which is used to map multi-source sensor data to a manifold space for feature extraction and learn an optimal fusion strategy based on geodesic distance. Specifically, the data fusion unit performs the following steps:

[0014] S1: Represent the sensor observations as a matrix X ∈ R (n×m) , where n is the feature dimension and m is the number of samples;

[0015] Step S2: Learn a mapping f: X → Y, Y ∈ R (d×m) , such that Y reflects the topological structure of X, where d << n is the dimension of the manifold space; f is obtained by minimizing the following objective function:

[0016]

[0017] where C ∞ (M) is the space of smooth functions, is the gradient operator on the manifold M, ∥·∥ F is the matrix F norm, dV is the Riemannian volume element, and λ is the regularization parameter;

[0018] Step S3: For the k-th sensor, obtain the representation y of the observation in the d-dimensional space based on manifold learning k ;

[0019] Step S4: Calculate the optimal fusion value z by solving the following optimization problem:

[0020]

[0021] s.t. ∑ k = 1 K α k = 1, α k ≥ 0

[0022] where α = [α1, …, α K is the confidence vector, and d M (·, ·) is the geodesic distance on the manifold M;

[0023] Step S5: Set the partial derivative to 0 to obtain the analytical expression of z:

[0024]

[0025] Specifically, the automatic calibration module includes a standard weight set, a temperature and humidity sensor, and a calibration control unit; the calibration control unit adaptively adjusts the calibration period T according to the drift trend of the measured value and the temperature and humidity changes; when the accumulated drift exceeds the threshold or the environment changes drastically, the calibration is automatically triggered; during the calibration process, a correction model is established by combining piecewise linear correction and hysteresis compensation:

[0026]

[0027] Where k is the proportionality factor, b is the offset, and h(t) is the dynamic hysteresis error.

[0028] Specifically, the wireless networking module includes a wireless communication unit and a networking control unit; the wireless communication unit supports multiple communication protocols such as WiFi, ZigBee, and NB-IoT; the networking control unit adopts an improved LEACH protocol and introduces node trust T as the basis for cluster head selection. The probability of cluster head election is:

[0029]

[0030] Where k is the expected number of cluster heads, N is the total number of nodes, r is the current round number, and G is the set of nodes that have not been elected as cluster heads. The networking control unit also introduces a scheduling mechanism based on task hunger F and node residual energy E, with priority P = αF + (1-α) E. Tasks with high priority are scheduled first.

[0031] Specifically, the intelligent operation and maintenance module includes an equipment state evaluation unit and an anomaly detection unit; the equipment state evaluation unit embeds the equipment state into a d-dimensional Euclidean space and learns the mapping φ:S→R d Hequn said Make Where S is the device state space, is the permutation group, GL(d,R) is the general linear group, represents the state transfer operation; the anomaly detection unit constructs the Mahalanobis distance metric, and for the anomaly score of the state y′ to be evaluated, J(y|x;π) is:

[0032]

[0033] The optimal state embedding φ and group representation ρ are jointly learned by minimizing:

[0034]

[0035] in is the manifold regularization term,

[0036] is the group regularization term, and λ1 and λ2 are balancing factors.

[0037] Specifically, the environmental adaptation module includes multiple environmental sensors and an adaptive compensation unit; the environmental sensors include temperature, humidity, vibration, and wind speed sensors for real-time collection of environmental parameters; the adaptive compensation unit fuses environmental parameters and measurement values, establishes a compensation model through online learning, and adaptively corrects the measurement results to offset the influence of environmental factors; the compensation model is obtained by minimizing the weighted sum of measurement error and model complexity.

[0038] Specifically, the human-computer interaction module includes a touch screen, a voice recognition unit and an operation log recording unit; the touch screen provides a graphical operation interface to support users in parameter setting, function selection, etc.; the voice recognition unit converts the user's voice instructions into control commands to achieve hands-free operation; when the collected data exceeds the normal range, changes drastically or remains unchanged for a long time, the system automatically pops up a window or voice prompts the user; the operation log recording unit automatically records the user's operation behavior, including operation time, type, parameters and results, and stores and regularly backs up in encrypted form for fault diagnosis and problem tracing.

[0039] The intelligent adaptive electronic scale data processing method based on the system is characterized by comprising the following steps:

[0040] Step 1: Collect gravity, volume and shape information through the composite sensor module, and input multi-source heterogeneous data into the data fusion unit;

[0041] Step 2: The data fusion unit extracts the manifold features of the data, and calculates the optimal weighted fusion of the sensor measurement values ​​by solving the optimal transmission problem of the geodesic distance, and outputs the fusion value;

[0042] Step 3: The automatic calibration module compares the fusion value with the standard weight reading, constructs an error correction model through piecewise linear fitting and hysteresis modeling, and outputs the corrected measurement results;

[0043] Step 4: The wireless networking module builds a network topology with other nodes through a self-organizing protocol and schedules data transmission tasks according to task hunger and energy status;

[0044] Step 5: The intelligent operation and maintenance module embeds the measurement result sequence into the state manifold space, and realizes equipment status evaluation and fault warning through invariant learning and anomaly measurement;

[0045] Step 6: The environmental adaptation module collects environmental parameters such as temperature and humidity in real time, and trains the compensation model based on historical data to adaptively correct the measurement results;

[0046] Step 7: The human-computer interaction module records the user operation log and realizes user-friendly human-computer dialogue through graphical interface, voice interaction and other methods.

[0047] A computer-readable storage medium of the method, wherein a computer program is stored in the storage medium, and when the program is run on a computer, the computer is enabled to execute the intelligent adaptive electronic scale data processing method.

[0048] The present invention has the following beneficial effects:

[0049] 1. The composite sensor technology is used to integrate multi-source information such as pressure, ultrasound, and images to achieve high-precision and all-round weighing perception.

[0050] 2. The adaptive calibration algorithm can automatically adjust the calibration cycle and calibration parameters according to the dynamic changes of weighing error, greatly improving the measurement accuracy.

[0051] 3. The use of self-organizing wireless networking technology gives the electronic scale flexible networking capabilities and distributed data processing capabilities, improving the collaborative efficiency of weighing operations.

[0052] 4. The intelligent operation and maintenance algorithm can evaluate the health status of the electronic scale in real time, promptly warn and diagnose faults, and reduce unexpected downtime.

[0053] 5. The environmental adaptive compensation technology can dynamically correct the weighing error according to environmental parameters such as external temperature and humidity, ensuring stable operation under harsh working conditions.

[0054] 6. The use of touch + voice intelligent interaction technology, combined with humanized interface design, greatly improves user experience and improves work efficiency.

[0055] In short, the present invention fully absorbs and draws on the latest research results in the fields of sensor technology, automatic control, artificial intelligence, the Internet of Things, etc., and systematically optimizes and innovates the shortcomings of existing electronic scales to form an overall solution that integrates high precision, high intelligence, high reliability, high adaptability, and high ease of use. The present invention will surely promote the intelligent development of the electronic scale industry and provide solid support for application scenarios such as smart logistics, smart manufacturing, and smart retail. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is the overall logic block diagram of the system of the present invention;

[0057] Figure 2 Logic block diagram of composite sensor module;

[0058] Figure 3 This is the logic block diagram of the automatic calibration module;

[0059] Figure 4 This is the logic block diagram of the wireless networking module;

[0060] Figure 5 This is the logic block diagram of the intelligent operation and maintenance module;

[0061] Figure 6 This is the logic block diagram of the environment adaptation module;

[0062] Figure 7 This is the logic block diagram of the human-computer interaction module. DETAILED DESCRIPTION

[0063] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of the intelligent adaptive electronic scale system based on machine learning proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0064] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0065] The specific scheme of the intelligent adaptive electronic scale system based on machine learning provided by the present invention is described in detail below with reference to the accompanying drawings.

[0066] like Figure 1 As shown, the intelligent adaptive electronic scale system of the present invention is mainly composed of the following six functional modules: composite sensor module 1, automatic calibration module 2, wireless networking module 3, intelligent operation and maintenance module 4, environmental adaptation module 5 and human-computer interaction module 6. These modules are interconnected through a system bus 7, and together constitute a highly integrated and intelligent weighing system.

[0067] First, the composite sensor module 1 is the core sensing unit of this system. Figure 2 As shown, the module includes a pressure sensor array, an ultrasonic sensor, an image sensor and a data fusion unit. Preferably, the pressure sensor array uses 16 PS-08 strain sensors arranged in a 4×4 matrix to cover the entire weighing platform. The range of each sensor is 0-100kg, the sensitivity is 0.1% FS, and the linearity is better than 0.02% FS. This matrix arrangement can accurately capture the center of gravity distribution of the cargo and effectively eliminate the eccentric load error.

[0068] Next, the ultrasonic sensor uses the US-100 model, with an operating frequency of 40KHz, a ranging range of 0.02-4m, and a resolution of 0.1mm. The sensor is installed above the weighing platform to measure the height of the goods. By multiplying it with the area of ​​the weighing platform, the volume of the goods can be estimated. Furthermore, the image sensor uses a 2-megapixel OV2640 module with a maximum resolution of 1600×1200 and a field of view of 78°. It is installed just above the weighing platform to capture the plane shape and surface texture features of the goods.

[0069] In order to achieve effective fusion of multi-source heterogeneous data, this paper designs a data fusion algorithm based on manifold learning. The core idea of ​​the algorithm is to map high-dimensional observation data to low-dimensional manifold space to extract the essential features of the data. The specific steps are as follows:

[0070] Step S1: Represent the sensor observations as a matrix X∈R n×m , where n is the feature dimension and m is the number of samples. In this embodiment, n=16+1+3=20 (16 pressure sensor values, 1 height value, 3 RGB color components), and m is the product of the sampling frequency and the sampling time.

[0071] Step S2: Learn a mapping f:X→Y,Y∈R d×m , so that Y can reflect the intrinsic topological structure of X, where d<<n is the dimension of the manifold space. Preferably, d is 3-5. f is obtained by minimizing the following objective function:

[0072]

[0073] Among them C ∞ (M) represents the smooth function space on the manifold M, y i For x i In manifold space, is the gradient operator on the manifold M, ||·|| F is the matrix Frobenius norm, dV is the Riemann measure, and λ is the regularization coefficient. In practical applications, the value range of λ is usually 0.01-0.1.

[0074] Step S3: For the kth sensor, obtain the low-dimensional representation y of its observation value in the d-dimensional space based on manifold learning k .

[0075] Step S4: Design the optimal transmission strategy in the manifold space to obtain the weighted fusion of different sensor measurements. Construct the following optimization model:

[0076] st

[0077]

[0078] where α=[α1,…,α K ] is the sensor confidence vector, d M (·,·) is the geodesic distance on the manifold M, and z is the fusion result. By introducing the Lagrange multiplier method, the analytical expression of z can be obtained:

[0079]

[0080] Through the above algorithm, the present invention realizes the organic integration of pressure, volume and shape information, greatly improving the accuracy and reliability of weighing. Experiments show that compared with the traditional single sensor, the weighing accuracy of the present system is improved by more than 30%, especially in the weighing of irregular shaped objects.

[0081] Secondly, the automatic calibration module 2 is the key to ensure the long-term stability of the system. Figure 3 As shown, the module includes a standard weight set, a temperature and humidity sensor and a calibration control unit. Preferably, the standard weight set 21 is composed of 5 E2 grade stainless steel weights, which are 1kg, 2kg, 5kg, 10kg and 20kg respectively. The temperature and humidity sensor 22 adopts the SHT30 model, with a temperature measurement range of -40℃~125℃ and an accuracy of ±0.2℃; the relative humidity measurement range is 0-100%RH and an accuracy of ±296RH.

[0082] The calibration control unit uses an adaptive calibration algorithm, the core idea of ​​which is to automatically adjust the calibration cycle and calibration parameters according to the dynamic changes of the weighing error. The specific steps are as follows:

[0083] Step C1: Set the initial calibration period T0, which is usually 1-3 months.

[0084] Step C2: During the i-th calibration, record the current accumulated drift Δw i and environmental parameter change ΔE i (including temperature change ΔT i and humidity change ΔH i ).

[0085] Step C3: Update the calibration period according to the following formula:

[0086] T i+1＝ T i ×(1-α×|Δw i | / w max -β×|ΔE i | / E max )

[0087] Where W max is the maximum allowable drift, Emax is the maximum allowable environmental change, α and β are weight coefficients, and the value range is 0-1. Preferably, α=0.6, β=0.4.

[0088] Step C4: When any of the following conditions is met, the calibration procedure is triggered:

[0089] a) Reach the calculated calibration period T i+1 ;

[0090] b) The accumulated drift |Δw| exceeds the maximum allowable drift w max 80%;

[0091] c) The environmental parameter change |ΔE| exceeds the maximum allowable environmental change E max 80% of.

[0092] Step C5: Execute the calibration procedure, using a combination of piecewise linear correction and hysteresis compensation.

[0093] For input x, the corrected output for:

[0094]

[0095] where k j and b j are the slope and intercept of the jth segment, respectively, and h(t) is the dynamic hysteresis error. Preferably, the measurement range is divided into 5 segments, i.e., j = 1, 2, 3, 4, 5. The hysteresis error h(t) is described by the improved Bouc-Wen model:

[0096] Among them, A, β, and γ are model parameters, which are estimated in real time through online learning algorithms.

[0097] Through the above-mentioned adaptive calibration algorithm, the present invention realizes the self-correction and dynamic optimization calibration of the electronic scale, effectively solving the problem of accuracy degradation caused by long-term use. Experiments show that after automatic calibration, the indication error of the electronic scale can be controlled within 0.01%, and the long-term stability is improved by more than 50%.

[0098] Third, the wireless networking module 3 gives the electronic scale flexible networking capabilities and distributed data processing capabilities. Figure 4As shown, the module includes a wireless communication unit and a networking control unit. Preferably, the wireless communication unit adopts a multi-mode communication chip ESP32-C3, supporting multiple communication protocols such as WiFi, Bluetooth5.0 and ZigBee. Among them, WiFi works in the 2.4GHz frequency band, with a maximum transmission rate of up to 150Mbps; ZigBee adopts the IEEE802.15.4 protocol, with an operating frequency of 2.4GHz and a transmission distance of up to 100 meters.

[0099] The networking control unit adopts an improved LEACH (Low Energy Adaptive Clustering Hierarchy) protocol and introduces the node trust T as the basis for cluster head selection. The specific steps are as follows:

[0100] Step N1: Initialize the network, set the expected number of cluster heads k, the total number of nodes N, and the current round number r0034] Step N2: Calculate the cluster head election probability p for each node i i :

[0101]

[0102] Where G is the candidate cluster head set, T i is the trustworthiness of node i. Preferably, T i The calculation formula is:

[0103]

[0104] Where E i is the node residual energy, D i is the distance from the node to the base station, S i is the stability index of the node, w1, w2, w3 are weight coefficients, and w1+w2+w3=1. In this embodiment, w1=0.4, w2=0.3, w3=0.3.

[0105] Step N3: Each node generates a random number between 0 and 1. If the random number is less than p i , then the node becomes the cluster head and broadcasts the cluster head information to the surrounding nodes.

[0106] Step N4: The non-cluster head node selects the nearest cluster head to join its cluster according to the signal strength of the received cluster head information.

[0107] Step N5: The cluster head node allocates TDMA time slots to the cluster members and establishes a communication scheduling table.

[0108] In terms of data transmission scheduling, the present invention proposes a priority mechanism based on task hunger F and node residual energy E. The calculation formula of priority P is:

[0109] P = αF+(1 - α)E

[0110] Where α is a trade - off factor, and its value range is 0 - 1. Preferably, α = 0.6. The task hunger degree F is positively correlated with its waiting time t, and can be expressed as F = 1 - e -λt

[0111] Where λ is a system parameter used to adjust the growth rate of hunger degree. Preferably, λ = 0.1.

[0112] Through the above wireless networking strategy, the present invention realizes the flexible networking and efficient data transmission of electronic scales. Experiments show that, compared with the traditional wired connection method, the networking efficiency of this system has increased by 200%, and the data transmission delay has been reduced by 40%.

[0113] Fourth, the intelligent operation and maintenance module 4 is the "brain" of this system, responsible for device status assessment and fault warning. As Figure 5 shown, this module includes a device status assessment unit and an anomaly detection unit. The present invention innovatively combines group theory and manifold learning, and proposes a state assessment algorithm based on group representation. The specific steps are as follows:

[0114] Step M1: Construct a state space S, and each element represents a possible working state of the electronic scale. Define a state transition operator o, such that constitutes a group.

[0115] Step M2: Learn the state embedding mapping d << dim(S), so that The distribution in the embedding space can reflect x i The distribution in the original state space.

[0116] Step M3: Learn the group representation Where is an n - dimensional permutation group, and GL(d, R) is a d - dimensional general linear group. It is required that ρ satisfies

[0117] Step M4: Construct a joint optimization model:

[0118]

[0119] Where J(y|x; π) is the reconstruction error, and R ρ are respectively and the regularization terms of ρ, and λ1 and λ2 are trade - off factors.

[0120] The reconstruction error J uses the Mahalanobis distance:

[0121]

[0122] Where ∑ is the covariance matrix. Using manifold regularization:

[0123]

[0124] W i,j For state x i and x j Similarity on the manifold. R ρ Using group regularization:

[0125]

[0126] Where I is the identity matrix.

[0127] Step M5: Use the alternating optimization strategy to solve the above optimization problem and obtain the optimal and ρ.

[0128] Step M6: For the current state x of the electronic scale, calculate its abnormal score:

[0129]

[0130] Where x' is a normal state set. When J(x) is greater than a preset threshold τ, it is determined to be an abnormal state. Preferably, τ=3.5.

[0131] Through the above algorithm, the present invention realizes accurate evaluation of the electronic scale status and abnormal warning. Experiments show that the method can predict possible failures 2-3 days in advance with an accuracy rate of more than 95%.

[0132] Fifth, the environmental adaptation module 5 is the key to ensure that the electronic scale can work stably in harsh environments. Figure 6 As shown, the module includes multiple environmental sensors and adaptive compensation units. Preferably, the environmental sensors include a temperature sensor (PT100, accuracy ±0.1°C), a humidity sensor (HIH-4000, accuracy ±2%RH), an air pressure sensor (BMP388, accuracy ±0.5hPa) and a three-axis acceleration sensor (ADXL355, resolution 1mg).

[0133] The adaptive compensation unit adopts a nonlinear modeling method based on deep learning. The core idea is to establish a mapping relationship from environmental parameters to measurement errors and continuously optimize the mapping through online learning. The specific steps are as follows:

[0134] Step E1: Construct compensation model:

[0135]

[0136] Where x is the electronic scale indication, y is the output value after compensation, and f(x;θ) is the basic measurement model. is the i-th environmental factor ξ i The compensation function, is the compensation parameter to be identified.

[0137] Step E2: Collect training sample set D = {(x j ,ξ 1,j ,…,ξ n,j ),j=1,…,N}. Step E3: Minimize the following empirical risk and learn the model parameters θ and

[0138]

[0139] where l(·,·) is the loss function, Preferably, the mean square error is used as the loss function and the L2 norm is used as the regularization term.

[0140] Step E4: Use stochastic gradient descent method to optimize the above objective function and update the parameters:

[0141]

[0142] Where η is the learning rate, the initial value is set to 0.01, and an adaptive learning rate strategy is adopted.

[0143] Step E5: Update the model regularly (e.g., daily) using newly collected data to achieve online learning.

[0144] Through the above-mentioned environment adaptive compensation algorithm, the present invention effectively solves the influence of harsh environment on the accuracy of electronic scales. Experiments show that in the temperature range of -20℃-60℃ and relative humidity of 095%RH, the measurement accuracy of the system can still be maintained within ±0.05%FS.

[0145] Finally, the human-computer interaction module 6 adopts a series of humanized designs, which greatly improves the usability of the electronic scale. Figure 7 As shown, the module includes a touch screen, a voice recognition unit and an operation log recording unit. Preferably, the touch screen adopts a 7-inch capacitive touch screen with a resolution of 1024×600 and supports ten-point touch. The touch screen surface adopts Corning Gorilla Glass with a hardness of 7H and good anti-scratch performance.

[0146] The speech recognition unit uses iFlytek's speech recognition engine, which has a vocabulary of more than 100,000 words, supports 23 dialects, and has a recognition accuracy of up to 98%. In order to improve the recognition effect in a noisy environment, this system also uses far-field sound pickup technology. Specifically, four omnidirectional microphones are installed around the electronic scale to form a square array with a microphone spacing of 10 cm. Through the Delay-and-Sum beamforming algorithm, long-distance voice interaction within a range of 5 meters is achieved.

[0147] The operation log recording unit automatically records every operation of the user, including power on / off time, user login / logout, weighing record, parameter modification, etc. The log is stored in encrypted form in the local solid-state drive and uploaded to the cloud server for backup via the 4G network at regular intervals every day. The log supports multi-dimensional retrieval, such as filtering by time, operation type, user, etc.

[0148] In terms of abnormal data prompts, the system has designed an intelligent alarm mechanism. When the following situations occur, the system will automatically pop up a prompt box and voice broadcast: a) The weighing data exceeds ±10% of the normal range; b) The weighing data changes by more than 5% of the maximum weighing capacity within 1 second; c) The weighing data has no change for 10 consecutive minutes.

[0149] Through the above human-computer interaction design, the present invention greatly improves the operation convenience and user experience of the electronic scale. Experiments show that compared with traditional electronic scales, the operation efficiency of this system is improved by 40% and the user satisfaction is improved by 60%.

[0150] In summary, the intelligent adaptive electronic scale system of the present invention realizes high-precision, high-reliability, high-intelligence and high-adaptability weighing functions through the organic combination of innovative modules such as composite sensors, automatic calibration, wireless networking, intelligent operation and maintenance, environmental adaptation and human-computer interaction. The modules work closely together to form a synergistic and efficient whole.

[0151] Preferably, in a specific embodiment of the present invention, the main technical indicators of the system are as follows: weighing range: 0-200kg; graduation value: 10g; accuracy level: OIML III level; stabilization time: <1s; working temperature: -20℃~60℃; relative humidity: 0~95%RH (non-condensing); protection level: IP67; battery life>12h (continuous operation); wireless communication distance:>100m (open environment);

[0152] The intelligent adaptive electronic scale system of the present invention can be widely used in many fields such as industrial production, trade logistics, and agricultural product testing. For example, in the food processing industry, the system can accurately control the raw material ratio, and at the same time realize the real-time uploading and analysis of production data through the wireless networking function, thereby improving production efficiency and product quality. In the field of logistics and warehousing, the system can quickly and accurately complete the weighing and sorting of goods, and prevent equipment failures through intelligent operation and maintenance functions to ensure the continuity of logistics operations. In the agricultural product trading market, the system can adapt to humid and dusty environments, provide stable and reliable weighing services, and facilitate merchants and customers to use through a user-friendly interactive interface.

[0153] The present invention also discloses an intelligent adaptive electronic scale data processing method based on the system, comprising the following steps:

[0154] Step 1: Collect gravity, volume and shape information through the composite sensor module, and input multi-source heterogeneous data into the data fusion unit;

[0155] Step 2: The data fusion unit extracts the manifold features of the data, and calculates the optimal weighted fusion of the sensor measurement values ​​by solving the optimal transmission problem of the geodesic distance, and outputs the fusion value;

[0156] Step 3: The automatic calibration module compares the fusion value with the standard weight reading, constructs an error correction model through piecewise linear fitting and hysteresis modeling, and outputs the corrected measurement results;

[0157] Step 4: The wireless networking module builds a network topology with other nodes through a self-organizing protocol and schedules data transmission tasks according to task hunger and energy status;

[0158] Step 5: The intelligent operation and maintenance module embeds the measurement result sequence into the state manifold space, and realizes equipment status evaluation and fault warning through invariant learning and anomaly measurement;

[0159] Step 6: The environmental adaptation module collects environmental parameters such as temperature and humidity in real time, and trains the compensation model based on historical data to adaptively correct the measurement results;

[0160] Step 7: The human-computer interaction module records the user operation log and realizes user-friendly human-computer dialogue through graphical interface, voice interaction and other methods.

[0161] The present invention also discloses a computer-readable storage medium of the method, wherein the storage medium stores a computer program, and when the program is run on a computer, the computer executes the intelligent adaptive electronic scale data processing method.

[0162] It should be noted that the above embodiments are only preferred implementations of the present invention and are not intended to limit the scope of protection of the present invention. Those skilled in the art may make various modifications or improvements to the present invention without departing from the spirit and scope of the present invention, and these modifications or improvements should also be regarded as the scope of protection of the present invention. For example, the composite sensor module can increase or decrease the sensor types according to specific application requirements; the automatic calibration module can use other optimization algorithms to adjust the calibration parameters; the wireless networking module can support more communication protocols; the intelligent operation and maintenance module can introduce other machine learning algorithms to improve the accuracy of fault prediction; the environmental adaptation module can consider more environmental factors for compensation; the human-computer interaction module can add new interaction methods such as gesture recognition. These changes will not change the essence of the present invention and should be included in the scope of protection of the present invention.

[0163] In short, the present invention provides a comprehensive intelligent weighing solution, which not only solves the shortcomings of traditional electronic scales in terms of accuracy, reliability, adaptability, etc., but also introduces new features of intelligence and networking, which points out the direction for the future development of electronic scales. It is believed that with the further maturity of technology and the continuous expansion of applications, the present invention will bring revolutionary changes to the fields of smart logistics, intelligent manufacturing, precision agriculture, etc.

Claims

1. Intelligent adaptive electronic scale system based on machine learning, characterized by: include: A composite sensor module for collecting gravity, volume and shape information; an automatic calibration module connected to the composite sensor module for periodic calibration and calibration based on the data of the composite sensor module; a wireless networking module connected to the automatic calibration module for wireless transmission and interaction of calibrated data; an intelligent operation and maintenance module connected to the wireless networking module for receiving transmitted data and realizing equipment status monitoring and abnormality detection; an environmental adaptation module connected to the intelligent operation and maintenance module for realizing adaptive compensation in harsh environments based on detection results; a human-computer interaction module connected to the environmental adaptation module for displaying compensated results and realizing a user-friendly operation interface; wherein the operation results of the human-computer interaction module are fed back to the composite sensor module for adjusting acquisition parameters, thereby forming a closed-loop control system.

2. The system according to claim 1, characterized in that The composite sensor module includes a pressure sensor array, an ultrasonic sensor and an image sensor; the pressure sensor array is composed of multiple PS-08 pressure sensors, which are used to collect all-round gravity information; the ultrasonic sensor model is US-100, which is used to measure the volume of the goods; the image sensor adopts the OV2640 module, which is used to identify the shape characteristics of the goods.

3. The system according to claim 1, characterized in that The composite sensor module is provided with a data fusion unit for mapping multi-source sensor data to a manifold space for feature extraction and learning an optimal fusion strategy based on geodesic distance. Specifically, the data fusion unit performs the following steps: S1: Represent the sensor observations as a matrix X∈R (n×m) , where n is the feature dimension and m is the number of samples; Step S2: Learn a mapping f: X → Y, Y ∈ R (d×m) , such that Y reflects the topological structure of X, where d << n is the dimension of the manifold space; f is obtained by minimizing the following objective function: Among them C ∞ (M) is the smooth function space, is the gradient operator on the manifold M, ||·|| F is the norm of matrix F, dV is the Riemann volume element, and λ is the regularization parameter; Step S3: For the kth sensor, obtain the representation y of the observation value in d-dimensional space based on manifold learning k ; Step S4: Calculate the optimal fusion value z by solving the following optimization problem: stΣ k =1 K a k =1,a k ≥0 Among them, α=[α1,…,α K ] is the confidence vector, d M (·,·) is the geodesic distance on the manifold M; Step S5: Set the partial derivative to 0 and obtain the analytical expression of z:

4. The system according to claim 1, characterized in that The automatic calibration module includes a standard weight set, a temperature and humidity sensor, and a calibration control unit; the calibration control unit adaptively adjusts the calibration period T according to the drift trend of the measured value and the temperature and humidity changes; when the cumulative drift exceeds the threshold or the environment changes drastically, the calibration is automatically triggered; during the calibration process, a correction model is established by combining piecewise linear correction and hysteresis compensation: Where k is the proportional coefficient, b is the offset, and h(t) is the dynamic hysteresis error.

5. The system according to claim 1, characterized in that The wireless networking module includes a wireless communication unit and a networking control unit; the wireless communication unit supports multiple communication protocols such as WiFi, ZigBee, and NB-IoT; the networking control unit adopts an improved LEACH protocol and introduces node trust T as the basis for cluster head selection. The probability of cluster head election is: Where k is the expected number of cluster heads, N is the total number of nodes, r is the current round number, and G is the set of nodes that have not been elected as cluster heads. The networking control unit also introduces a scheduling mechanism based on task hunger F and node residual energy E, with priority P = αF + (1-α) E. Tasks with high priority are scheduled first.

6. The system according to claim 1, characterized in that The intelligent operation and maintenance module includes an equipment state evaluation unit and an anomaly detection unit; the equipment state evaluation unit embeds the equipment state into a d-dimensional Euclidean space and learns the mapping φ:S→R d Hequn said Make Where S is the device state space, is a permutation group, GL(d,R) is a general linear group, ° represents a state transfer operation; the anomaly detection unit constructs a Mahalanobis distance metric, and for the anomaly score of the state y′ to be evaluated, J(y|x; π) is: The optimal state embedding φ and group representation ρ are jointly learned by minimizing: in is the manifold regularization term, is the group regularization term, and λ1 and λ2 are balancing factors.

7. The system according to claim 1, characterized in that The environmental adaptation module includes multiple environmental sensors and an adaptive compensation unit; the environmental sensors include temperature, humidity, vibration, and wind speed sensors for real-time collection of environmental parameters; the adaptive compensation unit integrates environmental parameters and measurement values, establishes a compensation model through online learning, and adaptively corrects the measurement results to offset the influence of environmental factors; the compensation model is obtained by minimizing the weighted sum of measurement error and model complexity.

8. The system according to claim 1, characterized in that The human-computer interaction module includes a touch screen, a voice recognition unit and an operation log recording unit; the touch screen provides a graphical operation interface to support users to set parameters, select functions, etc.; the voice recognition unit converts the user's voice instructions into control commands to achieve hands-free operation; when the collected data exceeds the normal range, changes drastically or remains unchanged for a long time, the system automatically pops up a window or voice prompts the user; the operation log recording unit automatically records the user's operation behavior, including operation time, type, parameters and results, and stores and regularly backs up in encrypted form for fault diagnosis and problem tracing.

9. An intelligent adaptive electronic scale data processing method based on the system described in any one of claims 1 to 8, characterized in that , including the following steps: Step 1: Collect gravity, volume and shape information through the composite sensor module, and input multi-source heterogeneous data into the data fusion unit; Step 2: The data fusion unit extracts the manifold features of the data, and calculates the optimal weighted fusion of the sensor measurement values ​​by solving the optimal transmission problem of the geodesic distance, and outputs the fusion value; Step 3: The automatic calibration module compares the fusion value with the standard weight reading, constructs an error correction model through piecewise linear fitting and hysteresis modeling, and outputs the corrected measurement results; Step 4: The wireless networking module builds a network topology with other nodes through a self-organizing protocol and schedules data transmission tasks according to task hunger and energy status; Step 5: The intelligent operation and maintenance module embeds the measurement result sequence into the state manifold space, and realizes equipment status evaluation and fault warning through invariant learning and anomaly measurement; Step 6: The environmental adaptation module collects environmental parameters such as temperature and humidity in real time, and trains the compensation model based on historical data to adaptively correct the measurement results; Step 7: The human-computer interaction module records the user operation log and realizes user-friendly human-computer dialogue through graphical interface, voice interaction and other methods.

10. The computer-readable storage medium of the method of claim 9, characterized in that The storage medium stores a computer program, and when the program is run on a computer, the computer executes the intelligent adaptive electronic scale data processing method according to claim 9.

Images