Intelligent weighing measurement method and system based on adaptive dynamic evolution

Abstract

The application provides a kind of intelligent weighing measurement method and system based on adaptive dynamic evolution, comprising the following steps: obtaining the original weight data sequence of medicinal materials and multi-source disturbance variables, and preprocessing;Determine the weighted reference sequence, and construct a multivariate dynamic evolution model according to the weighted reference sequence, the preprocessed original weight data sequence and multi-source disturbance variables;Determine the original weight prediction value corresponding to the abnormal data point according to the multivariate dynamic evolution model;Determine the residual between the model weight prediction value and the actual weight measurement value, and evaluate the current accuracy of the multivariate dynamic evolution model according to the residual;The beneficial effects of the present application are: the present application realizes missing data completion and real-time error correction through dynamic cumulative transformation, adaptive reference sequence generation, multi-source disturbance compensation and residual rolling self-correction mechanism, significantly improves the continuity and reliability of moisture content control in traditional Chinese medicine drying process.

Description

Technical Field

[0001] This invention relates to the field of intelligent weighing and measurement technology, and in particular to an intelligent weighing and measurement method and system based on adaptive dynamic evolution. Background Technology

[0002] In the production process of intelligent Chinese herbal medicine drying equipment, the dynamic weighing system is crucial for monitoring real-time changes in the moisture content of medicinal materials. Affected by hot air disturbances, drum turning, and mechanical vibrations, weighing sensors cannot operate continuously and stably, requiring shutdowns for static measurements, leading to production discontinuity. High temperature and humidity environments accelerate sensor aging, causing corrosion and insulation problems, reducing sensor lifespan and measurement accuracy. Drum turning causes uneven material distribution, creating local pressure differences and affecting the consistency of sensor readings. External collisions may damage sensors, and electromagnetic interference can lead to data loss or errors. Existing measurement methods rely on static weighing during shutdowns, resulting in discontinuous production, short sensor lifespan, and low prediction accuracy with small sample sizes. Therefore, there is an urgent need for a highly reliable weighing measurement model that possesses strong noise resistance, adaptability, and real-time drift correction capabilities even with very few effective samples.

[0003] Existing measurement error control methods include: linear regression models, time series analysis, Kalman filters, and neural network models. 1. Linear regression models predict and correct measured values ​​by establishing a linear relationship between input variables and output results, improving data accuracy and consistency. They are simple in technique and computationally efficient. 2. Time series analysis models predict future measured values ​​and detect anomalies by analyzing time-dependent patterns in historical data, thereby improving system accuracy and stability. They are suitable for processing time-dependent data. 3. Kalman filters are used for real-time estimation and correction of sensor data. By fusing multi-source information and reducing noise interference, they provide more accurate and smoother measurement results. Their advantage lies in handling noisy data. 4. Neural network models automatically adapt to and correct errors in measurement data by learning complex nonlinear relationships, thereby improving prediction accuracy and system robustness. Their advantages include strong nonlinear mapping capabilities and strong adaptive learning ability.

[0004] The drawbacks of existing technologies are as follows: 1. Linear regression models are suitable for linear relationships. For nonlinear data, prediction accuracy drops significantly. They are sensitive to outliers and noise, rely on large amounts of input data, and are difficult to adapt to dynamically changing systems. 2. Time series analysis models typically require large amounts of historical data, have limited ability to process non-stationary and nonlinear data, and model selection and parameter tuning are complex. 3. Kalman filters require the assumption that the dynamic behavior of the system and observation errors are linear and follow a Gaussian distribution, which limits their application in non-Gaussian noise environments. 4. Neural network models require large amounts of high-quality data for training, are prone to overfitting, and lack transparency in their internal operation, making them difficult to interpret. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide an intelligent weighing measurement method and system based on adaptive dynamic evolution, which solves the problems of existing measurement methods relying on static weighing during shutdown, resulting in discontinuous production, short sensor lifespan, and low prediction accuracy under small sample conditions.

[0006] To achieve the above and other related objectives, the present invention provides the following technical solution:

[0007] An intelligent weighing measurement method based on adaptive dynamic evolution includes the following steps: acquiring the original weight data sequence of medicinal materials and multi-source disturbance variables rapidly collected in a very short time by N sets of diagonally arranged weighing sensors during the production process of traditional Chinese medicine; preprocessing the original weight data sequence and obtaining abnormal data points of the original weight data sequence based on the preprocessing results; using an adaptive nearest neighbor weighting method and dynamically generating a weighted benchmark sequence based on the volatility of preceding data; constructing a multivariate dynamic evolution model based on the weighted benchmark sequence, the preprocessed original weight data sequence, and the multi-source disturbance variables; and estimating the unknown parameters in the multivariate dynamic evolution model.

[0008] The analytical expression of the equation is solved based on the established multivariate dynamic evolution model, and the predicted value of the cumulative sequence at any time is obtained based on the analytical expression. Then, the original weight prediction value corresponding to the abnormal data point is determined based on the predicted value of the cumulative sequence at any time. The residual between the model weight prediction value and the actual weight measurement value is determined. The current accuracy of the multivariate dynamic evolution model is evaluated based on the residual. Based on the evaluation result, it is determined whether to output a complete and verified weighing data sequence and the weight value within the prediction interval, or to trigger an adaptive correction strategy according to a preset priority order.

[0009] In one embodiment of the present invention, preprocessing the original weight data sequence and obtaining abnormal data points of the original weight data sequence based on the preprocessing result includes: determining the weight change rate between the weight value at the current sampling time and the weight value at the previous sampling time, and obtaining a preset threshold corresponding to the weight change rate; confirming the relationship between the weight change rate and the preset threshold, determining whether to mark the weight value at the current sampling time as erroneous data based on the confirmation result, and marking the erroneous data as erroneous abnormal data points, while also marking missing data in the original weight data sequence as missing abnormal data points.

[0010] In one embodiment of the present invention, the step of employing an adaptive nearest neighbor weighting method and dynamically generating a weighted benchmark sequence based on the volatility of preceding data includes: generating a weighted benchmark sequence according to the following formula: ;in, For the first The weighted benchmark value of each point is the basis for the subsequent construction of the dynamic evolution equation; These are weighting coefficients. , For data volatility, It is a regulatory factor, controlling Sensitivity to volatility; This represents the cumulative value at the current sampling time. This represents the cumulative value at the previous sampling time.

[0011] In one embodiment of the present invention, the step of constructing a multivariate dynamic evolution model based on the weighted benchmark sequence, the preprocessed original weight data sequence, and multi-source perturbation variables, and estimating the unknown parameters in the multivariate dynamic evolution model, includes: converting the preprocessed original weight data sequence into a monotonically increasing sequence, and constructing a first-order differential equation to describe the dynamic evolution law of the sequence based on the monotonically increasing sequence and the weighted benchmark sequence; introducing ambient temperature, relative humidity, and vibration data as multi-source perturbation variables, thereby extending the first-order differential equation into a multivariate dynamic evolution model, and estimating the unknown parameters in the multivariate dynamic evolution model, wherein the least squares parameters are used to estimate the unknown parameters in the multivariate dynamic evolution model.

[0012] In one embodiment of the present invention, the introduction of ambient temperature, relative humidity, and vibration data as multi-source disturbance variables to extend the first-order differential equation into a multivariable dynamic evolution model includes: determining the multivariable dynamic evolution model according to the following formula:

[0013] in, The cumulative sequence represents the rate of change over time, reflecting the rate of weight gain. For the front The first-order cumulative sum of the original weight values; This is the system evolution coefficient, which controls the influence of the cumulative sequence itself on the rate of change; It is a gray action quantity, representing the inherent driving term within the system; This is a cumulative sequence of ambient temperatures; This is a cumulative sequence of relative humidity. This is a cumulative sequence of vibrational accelerations; These are the disturbance coefficients for temperature, humidity, and vibration, respectively.

[0014] In one embodiment of the present invention, the process of solving the analytical expression of the equation based on an established multivariate dynamic evolution model, obtaining the predicted value of the accumulated sequence at any time based on the analytical expression, and then determining the original weight prediction value corresponding to the abnormal data point based on the predicted value of the accumulated sequence at any time includes: solving the analytical expression of the equation and determining the original weight prediction value corresponding to the abnormal data point according to the following formula: , ;

[0015] in, It is the first The cumulative sequence prediction value at each sampling time; It is the accumulated value of the first sampling point; It is the steady-state value of the accumulated sequence; It is an exponential decay / growth factor. Indicates from the first sampling time to the... The number of time steps elapsed at each sampling moment; Indicates the first The original weight prediction value at each sampling time; It is the cumulative predicted value from the previous sampling time.

[0016] In one embodiment of the present invention, determining the residual between the predicted model weight and the actual measured weight, evaluating the current accuracy of the multivariate dynamic evolution model based on the residual, and determining whether to output a complete and verified weighing data sequence and the weight value within the prediction interval, or to trigger an adaptive correction strategy according to a preset priority order, includes: using a sliding window to calculate the residual between the predicted model weight and the actual measured weight, and calculating the mean and standard deviation of the residual based on the residual;

[0017] The model accuracy is determined based on the standard deviation of the residuals and the standard deviation of the original weight data sequence. It is then determined whether the model accuracy and the mean of the residuals meet the pass rate judgment condition. If yes, a complete and verified weighing data sequence and the weight values ​​within the prediction interval are output. If no, an adaptive correction strategy is triggered according to a preset priority order. After correction, the process returns to the step of using the adaptive nearest neighbor weighting method and dynamically generating a weighted benchmark sequence based on the volatility of the preceding data. This process is iterated until the pass rate judgment condition is met.

[0018] In one embodiment of the present invention, the pass rate determination criteria include: and ;in, For model accuracy; The mean of the residuals; The allowable threshold for the residual mean is determined by process requirements; the adaptive correction strategy includes: first priority: adjusting the weighting coefficients of the baseline sequence; second priority: updating external input variables and recalculating the correlation; third priority: increasing the amount of modeling data; fourth priority: clearing historical residuals, re-collecting data and running the model.

[0019] An intelligent weighing and measurement system based on adaptive dynamic evolution includes: a data acquisition module for acquiring raw weight data sequences of medicinal materials and multi-source disturbance variables rapidly collected in a very short time by N sets of diagonally arranged weighing sensors during the production of traditional Chinese medicine; a data preprocessing module for preprocessing the raw weight data sequences and obtaining abnormal data points based on the preprocessing results; and a dynamic evolution analysis module for dynamically generating a weighted benchmark sequence using an adaptive nearest neighbor weighting method and based on the volatility of preceding data, constructing a multivariate dynamic evolution model based on the weighted benchmark sequence, the preprocessed raw weight data sequences, and the multi-source disturbance variables, and estimating the unknown parameters in the multivariate dynamic evolution model.

[0020] The prediction and verification module is used to solve the analytical expression of the equation based on the established multivariate dynamic evolution model, obtain the predicted value of the cumulative sequence at any time based on the analytical expression, and then determine the original weight prediction value corresponding to the abnormal data point based on the predicted value of the cumulative sequence at any time. The adaptive correction module is used to determine the residual between the model weight prediction value and the actual weight measurement value, evaluate the current accuracy of the multivariate dynamic evolution model based on the residual, and determine whether to output a complete and verified weighing data sequence and the weight value within the prediction interval, or to trigger the adaptive correction strategy according to the preset priority order based on the evaluation result.

[0021] As described above, the intelligent weighing measurement method and system based on adaptive dynamic evolution of the present invention has the following beneficial effects: The present invention rapidly collects raw weight, temperature, humidity, and vibration data within a very short time, and uses cumulative transformation to convert the noisy sequence into a monotonically increasing sequence, mutually weakening high-frequency random interference during the accumulation process; then, it adaptively generates a weighted benchmark sequence based on the data volatility, enabling the model to dynamically follow transient processes such as material turning and collapse; simultaneously, temperature, humidity, and vibration are introduced as multi-source disturbance variables into the dynamic evolution equation, and the parameters are estimated using the least squares method to achieve real-time compensation for environmental drift; for missing or erroneous labeled data, accurate prediction and repair are performed through the analytical solution of the model; finally, a sliding window residual rolling verification is adopted, and if the accuracy decreases, the weighting coefficients are adjusted online according to priority, the disturbance variables are updated, or the data volume is expanded, without the need for shutdown and retraining; thus, a closed loop of "acquisition-transformation-compensation-prediction-self-correction" is formed, achieving continuous, noise-resistant, and adaptive weighing data output with very few effective samples, significantly improving the reliability of moisture content control in the drying process of traditional Chinese medicine. Attached Figure Description

[0022] Figure 1 The flowchart shown is a process for an intelligent weighing measurement method based on adaptive dynamic evolution according to the first embodiment of the present invention.

[0023] Figure 2 The flowchart shown is an overall flowchart of the intelligent weighing measurement method based on adaptive dynamic evolution in the first embodiment of the present invention.

[0024] Figure 3 The diagram shown is a schematic representation of the overall structure of the intelligent weighing measurement system based on adaptive dynamic evolution according to the second embodiment of the present invention. Detailed Implementation

[0025] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. It should be noted that, unless otherwise specified, the following embodiments and features described herein can be combined with each other.

[0026] The first embodiment of this invention relates to an intelligent weighing measurement method based on adaptive dynamic evolution, specifically an anti-interference and highly reliable intelligent weighing measurement method for traditional Chinese medicine drying equipment. It is particularly suitable for continuous weighing measurement under complex working conditions such as hot air drying, drum turning, high temperature and humidity, and vibration interference. The process is as follows: Figure 1 As shown, the details are as follows:

[0027] Step 101: Obtain the original weight data sequence of medicinal materials and multi-source perturbation variables rapidly collected by N sets of diagonally arranged weighing sensors in a very short time during the production process of traditional Chinese medicine.

[0028] Specifically, the raw weighing data is acquired as follows: In this invention, N is four. Four sets of diagonally arranged weighing sensors continuously collect data for 10 seconds at a sampling rate of 0.05 seconds per sample, resulting in a total of 200 raw weight data sequences. Simultaneously, the ambient temperature is also collected. relative humidity Vibration acceleration As a multi-source input, it provides the original time series and multi-source perturbation input for subsequent modeling; among them, the mathematical expression of the original weight data sequence is: ,in, The original weight data sequence is a collection or vector containing 200 elements, and the superscript (0) represents the original sequence without cumulative transformation. Indicates the first Weight values ​​(in kg) at each sampling time; Indicates time The hot air or ambient temperature (in °C) is used to compensate for thermal expansion or sensor temperature drift. Indicates time The humidity (in %RH) is used to correct for the effect of humidity on sensor insulation and signal. Indicates time The equipment vibration amplitude (unit: m / s²) is used to measure the disturbances caused by material turning, collisions, etc.

[0029] The above parameters together constitute the basic input of the model: W is the main variable weight sequence. The data is a multi-source disturbance variable; the collected raw data is not used directly, but enters the data preprocessing stage in step 102.

[0030] Step 102: Preprocess the original weight data sequence and obtain the outlier data points of the original weight data sequence based on the preprocessing results.

[0031] Specifically, firstly, the weight change rate between the weight value at the current sampling time and the weight value at the previous sampling time is determined, and a preset threshold corresponding to the weight change rate is obtained; then, the relationship between the weight change rate and the preset threshold is confirmed, and based on the confirmation result, it is determined whether the weight value at the current sampling time is marked as erroneous data, and the erroneous data is marked as erroneous abnormal data points, while the missing data in the original weight data sequence is also marked as missing abnormal data points.

[0032] More specifically, data preprocessing includes: performing a reasonableness check on the data and removing outliers that clearly exceed the sensor's measurement range; firstly, calculating the weight change rate between two adjacent sampling points. ,in, This represents the rate of weight change between two adjacent sampling times (also known as sampling points). This represents the weight value at the current moment (in kg). This represents the weight value (in kg) at the previous moment. This represents the time difference between two adjacent sampling points (unit: kg / s); the formula reflects the rate of change of the weight of the medicinal material per unit time; during the drying process of traditional Chinese medicine, due to moisture evaporation, the weight should decrease slowly and monotonically, therefore... It is usually small and positive, but changes occur abruptly due to violent collisions or vibrations. It will be abnormally large; if If a preset threshold is set, the data is marked as erroneous and will be replaced by the model's predicted value.

[0033] Based on threshold verification of the rate of change, and combined with the characteristic of monotonically decreasing weight in the drying process of traditional Chinese medicine, the model can identify jump noise caused by collision and vibration, and its performance is better than that of ordinary statistical filtering.

[0034] Step 103: Adaptive proximity weighting method is used to dynamically generate a weighted benchmark sequence based on the volatility of preceding data.

[0035] Specifically, a baseline sequence is established: traditional baseline sequences use the neighbor mean with fixed weights, which is poorly adaptable to rapidly changing data; this invention proposes an adaptive neighbor weighting method to generate a weighted baseline sequence for subsequent modeling.

[0036] The adaptive nearest neighbor weighting formula is: ,in, Indicates the first The weighted baseline value of each point is the basic sequence for constructing the dynamic evolution equation, and the original accumulated sequence is adaptively smoothed. This represents the cumulative value at the current point. This represents the cumulative value of the previous sampling point, and the superscript (1) indicates the first-order cumulative sequence; This represents the weighting coefficient, with a value range of [0.5, 0.9]. It determines the weight ratio of the current point to the previous point in the baseline value.

[0037] The weighting coefficients are dynamically and adaptively adjusted based on data volatility, and the formula is as follows: ,in, This represents the volatility of the data, such as the standard deviation of the most recent data points. It is a regulatory factor, controlling Sensitivity to volatility; when data fluctuates drastically due to factors such as material turnover, Approaching 0.9, the current value is more trusted; when the data is stable, it approaches 0.5, reverting to the traditional mean; value range analysis: when the data is completely stable ( This means that the current point and historical points have equal weights, degenerating into the traditional mean method. This is the most conservative and smoothest estimate, suitable for ideal working conditions without interference; however, it does not completely discard historical information when the data fluctuates drastically. ≠1), because the weight changes continuously, completely trusting the current value will amplify the noise at a single point. The upper limit of 0.9 means that the current point accounts for 90% of the weight, and the historical point retains 10% of the inertial constraint to avoid unreasonable jumps in the baseline value.

[0038] Through extensive on-site testing of traditional Chinese medicine drying, when At times, the model is overly sensitive to vibration and electromagnetic interference, easily misinterpreting instantaneous spikes as true weight changes, leading to divergence in subsequent predictions; 0.9 is an empirical threshold that balances response speed and noise resistance, therefore The range is [0.5, 0.9] within the common volatility range; the purpose of this range design is: 1. to provide adaptive smoothing, allowing the model to approach mean filtering when stationary. ) and rapid follow during dramatic changes ( The first option is to adaptively adjust between 0 and 1, without the need for manual threshold setting or switching rules; the second option is to suppress extreme values, eliminating the two extremes of 0 and 1, preventing the model from completely ignoring the past or the present, and maintaining the stability of the model calculation.

[0039] Step 104: Construct a multivariate dynamic evolution model based on the weighted baseline sequence, the preprocessed original weight data sequence, and the multi-source perturbation variables, and estimate the unknown parameters in the multivariate dynamic evolution model.

[0040] Specifically, the preprocessed raw weight data sequence is first transformed into a monotonically increasing sequence, and a first-order differential equation is constructed based on the monotonically increasing sequence and the weighted benchmark sequence to describe the dynamic evolution of the sequence. Then, ambient temperature, relative humidity, and vibration data are introduced as multi-source disturbance variables, thereby extending the first-order differential equation into a multivariate dynamic evolution model, and estimating the unknown parameters in the multivariate dynamic evolution model. The least squares parameter estimation method is used to estimate the unknown parameters in the multivariate dynamic evolution model.

[0041] More specifically, establish a multivariate dynamic evolution equation: the preprocessed original weight sequence Transform into a monotonically increasing sequence To reduce high-frequency random noise, a first-order differential equation is constructed to describe the dynamic evolution of the sequence. Based on this, ambient temperature is introduced. ,humidity ,vibration As a multi-source perturbation variable, it is extended into a multivariate dynamic evolution model; and the unknown parameters in the model are estimated; the equation reflects the influence of external environmental disturbances on weight change, providing an accurate mathematical expression for subsequent missing data prediction and drift correction;

[0042] (1) Preprocessed data accumulation and transformation: ,in, Indicates the first The original weight values ​​(in kg) of each sampling point, from step 101, have been preprocessed in step 102 (outliers are marked but not yet replaced). Indicates the preceding The first-order cumulative sum of the original data points (in kg), with the superscript (1) indicating a single accumulation; the equation transforms the original noisy fluctuation sequence into a monotonically increasing smooth sequence, and the high-frequency random disturbances weaken each other during the accumulation process, which is beneficial for subsequent differential equation modeling.

[0043] (2) Construct the first-order differential equation: ,in, This represents the rate of change of the cumulative sequence over time (unit: kg / s), reflecting the rate of weight gain. This represents the system development coefficient (dimensionless), controlling for the influence of the cumulative sequence itself on the rate of change. This indicates that the system has inertia; the larger the value, the stronger the inhibitory effect of the current accumulated value on the rate of change. It is the gray action quantity (unit kg / s), which represents the inherent driving term within the system and is related to the intrinsic characteristics of the medicinal material, such as the initial weight and drying rate.

[0044] (3) Introduce multi-source disturbance variable temperature ,humidity ,vibration This can be expanded into a multivariate dynamic evolution model: ,in, Indicates ambient temperature The cumulative sequence, i.e. The unit is ℃·s; Relative humidity The cumulative sequence, in %RH·s; Represents vibration acceleration The cumulative sequence, in m / s²·s; These are perturbation coefficients, representing the intensity and direction of the influence of temperature, humidity, and vibration on the weight evolution process, respectively. The multivariate evolution model incorporates external environmental factors as additional driving terms into the differential equation, enabling the model to compensate for temperature and humidity drift and vibration interference, thereby improving prediction accuracy.

[0045] (4) Least squares parameter estimation: ,in, It is the system development coefficient. It is the gray action quantity. B is the perturbation coefficient, representing temperature, humidity, vibration intensity, and direction, respectively; B is the data matrix, consisting of the accumulated sequence and the perturbation sequence, with each row corresponding to a sampling point. The typical construction method is as follows: ,in, It is the weighted benchmark sequence generated in step 103, which replaces the direct accumulated values ​​to improve robustness; Y is the observation vector, which consists of the differences between adjacent accumulated values: That is, the original weight value , It is the standard form of the least squares solution, which minimizes the sum of squared prediction errors of the model.

[0046] This step transforms the engineering problem into a mathematical optimization problem, and automatically identifies the system dynamics and environmental coupling parameters from a finite sample using the least squares method, providing an analytically solvable model for subsequent prediction and adaptive correction.

[0047] Step 105: Solve the analytical expression of the equation based on the established multivariate dynamic evolution model, obtain the predicted value of the cumulative sequence at any time based on the analytical expression, and then determine the original weight prediction value corresponding to the abnormal data point based on the predicted value of the cumulative sequence at any time.

[0048] Specifically, missing data prediction involves using an established multivariate dynamic evolution model to solve the analytical expression of the equation, obtaining the predicted value of the cumulative sequence at any given time. Then, for data points marked as missing or erroneous, their time indices are substituted into the analytical solution formula to calculate the predicted cumulative value, restoring the original weight prediction value. The predicted value is then used to fill in the missing positions or replace erroneous data. The prediction process requires only model parameters and a small amount of historical data, without the need for manual intervention.

[0049] (1) Find the analytical solution for the multivariate dynamic evolution model: ,in, It is the first The cumulative sequence prediction value (in kg) of each sampling point. It is the accumulated value of the first sampling point. It is the system development coefficient, which controls the growth inertia of the cumulative sequence; It is a gray action quantity, representing the inherent driving term within the system; It is an exponential decay / growth factor. Indicates from the first sampling point to the... The number of time steps elapsed for each sampling point, if If the influence of the initial conditions decays exponentially over time, the system tends to a steady state. It is the steady-state value of the cumulative sequence, when hour, This is reflected in the original domain, where the weight will tend to a balance value, such as the final dry weight of the medicinal material;

[0050] Considering the impact of multi-source disturbances is reflected in the parameters In the estimated values, the disturbance term is incorporated into the data matrix, thus changing the... For the sake of simplicity, this paper provides a solution in the basic form for the estimated value.

[0051] (2) Substitute the time points marked as missing into the above formula to calculate the predicted value: ,in, Indicates the first The original predicted weight (in kg) of each sampling point, which is the value that needs to be filled in the missing positions or replaced with incorrect data; the superscript (0) represents the original domain. It is the first The cumulative predicted value of each step; It is the previous step ( The cumulative predicted value;

[0052] because Therefore, the difference between adjacent accumulated values ​​is equal to the current original value: Substituting the predicted cumulative value into the equation yields the predicted original value. This step enables accurate online completion of missing or erroneous data with very few valid samples, ensuring continuous output of complete data.

[0053] Step 106: Determine the residual between the model weight prediction value and the actual weight measurement value, evaluate the current accuracy of the multivariate dynamic evolution model based on the residual, and determine whether to output a complete and verified weighing data sequence and the weight value within the prediction interval, or to trigger an adaptive correction strategy according to the preset priority order, based on the evaluation result.

[0054] Specifically, firstly, a sliding window is used to calculate the residual between the predicted weight value and the actual measured weight value, and the mean and standard deviation of the residual are calculated based on the residual. Then, the model accuracy is determined based on the standard deviation of the residual and the standard deviation of the original weight data sequence, and it is judged whether the model accuracy and the mean of the residual meet the pass rate judgment condition. If yes, a complete and verified weighing data sequence and the weight value within the prediction interval are output. If no, an adaptive correction strategy is triggered according to a preset priority order. After correction, the process returns to the step of using the adaptive nearest neighbor weighting method and dynamically generating a weighted benchmark sequence based on the volatility of the preceding data. This process is iterated until the pass rate judgment condition is met.

[0055] More specifically, adaptive correction: This step is the core of achieving long-term operation without drift. Its main task is to calculate the residual between the model's predicted value and the actual measured value, and to evaluate the model's current accuracy based on the mean and standard deviation of the residual. If the accuracy meets the preset threshold, the result is directly output to the host computer; if not, the adaptive correction strategy is triggered according to the preset priority order, and the process returns to step 103 to retrain the model. The entire process is completed online without downtime, thus ensuring that the weighing system maintains high reliability under conditions such as sensor aging and drastic environmental changes.

[0056] (1) Use a sliding window (window) Calculate the residual between the actual measured value and the predicted value: ,in, It is the first The prediction residual for each sampling point is the difference between the actual measured value and the model prediction value; a positive value indicates that the actual weight is greater than the prediction (the model underestimates), and a negative value indicates that the actual weight is less than the prediction (the model overestimates). Indicates the first The original actual weight value of each sampling point; Indicates the first The model predicts weight values ​​for each sampling point; only the most recent values ​​are used. points ( Calculate residuals to ensure a rapid response to changes in current operating conditions, while avoiding the influence of outdated historical data on judgment.

[0057] (2) Calculate the mean of the residuals and standard deviation : ,in, It is the residual mean, which reflects whether there is a systematic bias in the model's predictions. Ideally, it is close to 0. It is the residual standard deviation, which reflects the degree of fluctuation in prediction error. The smaller the standard deviation, the more stable the model. It is the length of the sliding window. In this invention, it is set to 5. If the window is too small, it is easily affected by random noise. If the window is too large, it will reduce the sensitivity to sudden changes. 5 is a practical experience value.

[0058] (3) Evaluate the model accuracy C: ,in, The standard deviation of the original data series is calculated using the following formula: Where N is the sequence length of 200, This represents the original average weight.

[0059] (4) Criteria for determining pass rate: and ,in, It is the allowable threshold for the mean residual, which is determined by process requirements; Judgment logic: If both conditions are met at the same time → the prediction accuracy is qualified → the result is directly output to the host computer; If either condition is not met → adaptive correction is triggered.

[0060] Adaptive Correction Strategy Priority Table

[0061]

[0062] After correction, return to step 103, regenerate the baseline sequence, and retrain the model. The entire correction process is completed online without interrupting the drying production task.

[0063] It should be noted that the present invention also includes step 107, output and loop: the complete and verified weighing data sequence and the weight values ​​within the prediction interval are output to the host computer. The present invention continues to collect the next set of data samples, forming a sliding time window, and repeats steps 101 to 106, thereby achieving continuous, adaptive, and highly reliable weighing measurement; please refer to the above overall process. Figure 2 .

[0064] In practical applications, the present invention will be further described in detail below with reference to embodiments, but the implementation of the present invention is not limited thereto; 1. Embodiment 1: The system of the present invention is installed on the intelligent drying equipment of a certain Chinese herbal medicine processing enterprise; Equipment parameters: drum diameter 1.3m, rotation speed 8rpm, hot air temperature 80±5℃, relative humidity 65±10%. Four sets of weighing sensors with a range of 0-50kg and an accuracy of 0.01kg;

[0065] (1) After starting the equipment, data was collected for the first 10 seconds at 0.05s / time, for a total of 200 samples, and temperature, humidity and vibration data were collected simultaneously; (2) Data preprocessing revealed that the weight jumps (change rate exceeding 5kg / s) at points 78 and 79 were caused by vibration spikes, and were marked as erroneous data; (3) The fluctuation rate of the first 3 points was calculated. =0.23, determine the weighting coefficient. =0.62, generating a dynamic baseline sequence; (4) After cumulative transformation, construct a multivariate dynamic evolution equation and solve it using the least squares method. =0.031, =42.6; (5) Predict the cumulative values ​​at points 78 and 79, and after subtraction and restoration, obtain the predicted weight values ​​of 32.15kg and 31.98kg respectively, replacing the original erroneous values; (6) Sliding window residual verification: C=0.21, =0.03kg, all less than the threshold, judged as qualified prediction, and output the complete sequence to the host computer; (7) ran continuously for 24 hours, during which adaptive correction was triggered 3 times (all of which were priority 1 adjustments) No output interruption occurred, the weighing data integrity rate was 99.8%, and the moisture content control deviation was ≤0.5%.

[0066] 2. Example 2: When the ambient temperature suddenly rises from 80℃ to 95℃, model validation shows C=0.42, triggering adaptive correction priority 2: recalculating the correlation between temperature and weight (from 0.73 to 0.89), and updating the multivariate model parameters. After correction, C recovers to 0.28, and the system continues to operate normally.

[0067] The above description is merely a field embodiment of the present invention and is not intended to limit the present invention in any way. All equivalent variations and modifications made within the scope of the claims of this invention are covered by the patent.

[0068] Furthermore, the technical innovations and key points of this invention are as follows:

[0069] 1. Small sample prediction mechanism: A small sample prediction model is established, which can generate an exponential law of weight decay with only a small number of original samples, and a weighted neighbor mean background value is introduced to avoid the model error introduced by the traditional mean method; 2. Strong noise resistance transformation: Instead of relying on large sample smoothing filtering, the noisy original sequence is transformed into a monotonically growing sequence, so that the noise weakens each other during the accumulation process, and differential equations are used to suppress high-frequency random interference.

[0070] 3. Adaptive capability: The benchmark sequence dynamically selects weighting coefficients based on data volatility, enabling the model to automatically adapt to non-stationary processes such as drum turning and material collapse; 4. Multi-source information fusion correction: Temperature, humidity, and vibration amplitude are introduced as associated input variables to construct a multivariate dynamic evolution model. External compensation is performed on the model benchmark during the parameter estimation stage to eliminate environmental drift; 5. Real-time drift correction and feedback: The accuracy of the predicted value is evaluated using a residual rolling window. When the residual exceeds the threshold, a dynamic drift compensation factor is triggered to adjust the model parameters online without stopping the machine for retraining.

[0071] Furthermore, the scope of protection of this invention is as follows: 1. This invention aims to propose a highly reliable weighing measurement model technology. Differences in parameters such as data acquisition, parameter verification, parameter estimation, and model analytical solutions, as well as parameter changes involved in the model algorithm process, should all be within the scope of protection; 2. This invention is applicable to applications in the field of traditional Chinese medicine drying, intelligent drying equipment, etc., including applications in the sub-sectors of traditional Chinese medicine drying, but is not limited to these applications.

[0072] The second embodiment of the present invention relates to an intelligent weighing measurement system based on adaptive dynamic evolution. Please refer to [link to relevant documentation]. Figure 3 ,include:

[0073] The data acquisition module is used to acquire the original weight data sequence of medicinal materials and multi-source disturbance variables in the process of traditional Chinese medicine production, which are rapidly collected by N sets of diagonally arranged weighing sensors in a very short time.

[0074] The data preprocessing module is used to preprocess the original weight data sequence and obtain outlier data points in the original weight data sequence based on the preprocessing results.

[0075] The dynamic evolution analysis module is used to dynamically generate a weighted benchmark sequence based on the volatility of preceding data using an adaptive proximity weighting method. Based on the weighted benchmark sequence, the preprocessed original weighted data sequence, and multi-source perturbation variables, a multivariate dynamic evolution model is constructed, and the unknown parameters in the multivariate dynamic evolution model are estimated.

[0076] The prediction and verification module is used to solve the analytical expression of the equation based on the established multivariate dynamic evolution model, obtain the predicted value of the cumulative sequence at any time based on the analytical expression, and then determine the original weight prediction value corresponding to the abnormal data point based on the predicted value of the cumulative sequence at any time.

[0077] The adaptive correction module is used to determine the residual between the model's predicted weight value and the actual measured weight value. Based on the residual, it evaluates the current accuracy of the multivariate dynamic evolution model and determines whether to output a complete and verified weighing data sequence and the weight value within the prediction interval, or to trigger the adaptive correction strategy according to the preset priority order.

[0078] It should be noted that this system also includes four sets of weighing sensors and an environmental sensor group. The four sets of weighing sensors are symmetrically installed at the bottom of the drying equipment to collect the weight data of the medicinal materials. The environmental sensor group includes temperature and humidity sensors and vibration sensors to collect multi-source disturbance information. Therefore, this system consists of four sets of weighing sensors, an environmental sensor group, a data acquisition module, a dynamic evolution equation, a missing data prediction module, an adaptive correction module, and an output section. During the production process, the weighing sensors collect the weight data of the medicinal materials, establish a dynamic evolution equation to draw a weighing curve, the model fills in the missing data, and adaptively corrects the erroneous data to generate a complete weight prediction curve. The effective weight value of the prediction interval is then output to the host computer system.

[0079] It is not difficult to see that this embodiment is a system implementation corresponding to the first embodiment, and this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0080] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by this invention; however, this does not mean that other units are absent from this embodiment.

[0081] In summary, the beneficial effects of this invention are as follows: 1. Resistance to small sample size and strong noise: High-precision prediction can be started with only 200 samples in 10 seconds, which is superior to traditional methods that require a large amount of historical data; 2. Adaptability: The background value weighting coefficient is adjusted in real time according to data fluctuations, without the need for manual parameter tuning, and can automatically adapt to transient processes such as material turning and dropping; 3. Multi-source interference suppression: Temperature, humidity, and vibration are introduced as model inputs, so that the predicted values ​​can reflect the drift caused by the environment and realize real-time drift correction; 4. Closed-loop reliability: The residual sliding window plus adaptive correction strategy ensures that the model does not drift over long-term operation, and solves the problem of accuracy reduction caused by sensor aging and corrosion; 5. Lightweight computation: No complex iterative calculation model is required, which is suitable for deployment in embedded industrial controllers and meets the real-time requirements of traditional Chinese medicine drying equipment.

[0082] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. All equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this invention should still be covered by the claims of this invention.

Claims

1. An intelligent weighing measurement method based on adaptive dynamic evolution, characterized in that, Includes the following steps: The raw weight data sequence of medicinal materials and multi-source disturbance variables in the process of traditional Chinese medicine production are rapidly collected in a very short time by N sets of diagonally arranged weighing sensors. The raw weight data sequence is preprocessed, and abnormal data points of the raw weight data sequence are obtained based on the preprocessing results. An adaptive proximity weighting method is adopted and a weighted benchmark sequence is dynamically generated based on the volatility of preceding data. A multivariate dynamic evolution model is constructed based on the weighted benchmark sequence, the preprocessed original weighted data sequence, and multi-source perturbation variables, and the unknown parameters in the multivariate dynamic evolution model are estimated. The analytical expression of the equation is solved based on the established multivariate dynamic evolution model, and the predicted value of the cumulative sequence at any time is obtained based on the analytical expression. Then, the original weight prediction value corresponding to the abnormal data point is determined based on the predicted value of the cumulative sequence at any time. The residual between the model's predicted weight and the actual measured weight is determined. The current accuracy of the multivariate dynamic evolution model is evaluated based on the residual. Based on the evaluation result, it is determined whether to output a complete and verified weighing data sequence and the weight values ​​within the prediction interval, or to trigger an adaptive correction strategy according to a preset priority order.

2. The intelligent weighing measurement method based on adaptive dynamic evolution according to claim 1, characterized in that: The preprocessing of the original weight data sequence and the determination of outlier data points based on the preprocessing results include: Determine the rate of change of weight between the weight value at the current sampling time and the weight value at the previous sampling time, and obtain a preset threshold corresponding to the rate of change of weight; Confirm the relationship between the weight change rate and the preset threshold. Based on the confirmation result, determine whether to mark the weight value at the current sampling time as erroneous data and mark the erroneous data as erroneous abnormal data points. At the same time, mark the missing data in the original weight data sequence as missing abnormal data points.

3. The intelligent weighing measurement method based on adaptive dynamic evolution according to claim 1, characterized in that: The method of using adaptive proximity weighting and dynamically generating a weighted benchmark sequence based on the volatility of preceding data includes: The weighted benchmark sequence is generated according to the following formula: ； in, For the first The weighted benchmark value of each point is the basis for the subsequent construction of the dynamic evolution equation; These are weighting coefficients. , For data volatility, It is a regulatory factor, controlling Sensitivity to volatility; This represents the cumulative value at the current sampling time. This represents the cumulative value at the previous sampling time.

4. The intelligent weighing measurement method based on adaptive dynamic evolution according to claim 3, characterized in that: The step of constructing a multivariate dynamic evolution model based on the weighted baseline sequence, the preprocessed original weight data sequence, and multi-source perturbation variables, and estimating the unknown parameters in the multivariate dynamic evolution model, includes: The preprocessed original weight data sequence is transformed into a monotonically increasing sequence, and a first-order differential equation is constructed based on the monotonically increasing sequence and the weighted benchmark sequence to describe the dynamic evolution of the sequence. Ambient temperature, relative humidity, and vibration data are introduced as multi-source disturbance variables, thereby extending the first-order differential equation into a multivariate dynamic evolution model, and estimating the unknown parameters in the multivariate dynamic evolution model. The unknown parameters in the multivariate dynamic evolution model are estimated using least squares parameters.

5. The intelligent weighing measurement method based on adaptive dynamic evolution according to claim 4, characterized in that: The introduction of ambient temperature, relative humidity, and vibration data as multi-source disturbance variables expands the first-order differential equation into a multivariate dynamic evolution model, including: The multivariate dynamic evolution model is determined according to the following formula: ； in, The cumulative sequence represents the rate of change over time, reflecting the rate of weight gain. For the front The first-order cumulative sum of the original weight values; This is the system development coefficient, which controls the influence of the cumulative sequence itself on the rate of change; It is a gray action quantity, representing the inherent driving term within the system; This is a cumulative sequence of ambient temperatures; This is a cumulative sequence of relative humidity. This is a cumulative sequence of vibrational accelerations; These are the disturbance coefficients for temperature, humidity, and vibration, respectively.

6. The intelligent weighing measurement method based on adaptive dynamic evolution according to claim 5, characterized in that: The process of solving the analytical expression of the equation based on the established multivariate dynamic evolution model, obtaining the predicted value of the accumulated sequence at any time based on the analytical expression, and then determining the original weight prediction value corresponding to the abnormal data point based on the predicted value of the accumulated sequence at any time includes: The analytical expression of the equation and the original weight prediction value corresponding to the abnormal data point are determined according to the following formula: ， ； in, It is the first The cumulative sequence prediction value at each sampling time; It is the accumulated value of the first sampling point; It is the steady-state value of the accumulated sequence; It is an exponential decay / growth factor. Indicates from the first sampling time to the... The number of time steps elapsed at each sampling moment; Indicates the first The original weight prediction value at each sampling time; It is the cumulative predicted value from the previous sampling time.

7. The intelligent weighing measurement method based on adaptive dynamic evolution according to claim 1, characterized in that: The residual between the predicted weight value and the actual weight measurement value is determined. Based on the residual, the current accuracy of the multivariate dynamic evolution model is evaluated. Based on the evaluation result, it is determined whether to output a complete and validated weighing data sequence and the weight values ​​within the prediction interval, or to trigger an adaptive correction strategy according to a preset priority order, including: The residual between the predicted weight value and the actual weight measurement value is calculated using a sliding window, and the mean and standard deviation of the residual are calculated based on the residual. The model accuracy is determined based on the standard deviation of the residuals and the standard deviation of the original weight data sequence, and it is then determined whether the model accuracy and the mean of the residuals meet the pass rate judgment conditions. If yes, output a complete and verified weighing data sequence and the weight values ​​within the prediction interval; if no, trigger the adaptive correction strategy according to the preset priority order, and after correction, return to the step of using the adaptive nearest neighbor weighting method and dynamically generating a weighted benchmark sequence based on the volatility of the preceding data, and continue to iterate until the pass rate judgment condition is met.

8. The intelligent weighing measurement method based on adaptive dynamic evolution according to claim 7, characterized in that: The pass rate determination criteria include: and ;in, For model accuracy; The mean of the residuals; The allowable threshold for the residual mean is determined by process requirements; the adaptive correction strategy includes: first priority: adjusting the weighting coefficients of the baseline sequence; second priority: updating external input variables and recalculating the correlation; third priority: increasing the amount of modeling data; fourth priority: clearing historical residuals, re-collecting data and running the model.

9. An intelligent weighing and measurement system based on adaptive dynamic evolution, characterized in that: include: The data acquisition module is used to acquire the original weight data sequence of medicinal materials and multi-source disturbance variables in the process of traditional Chinese medicine production, which are rapidly collected by N sets of diagonally arranged weighing sensors in a very short time. A data preprocessing module is used to preprocess the original weight data sequence and obtain outlier data points in the original weight data sequence based on the preprocessing results. The dynamic evolution analysis module is used to dynamically generate a weighted benchmark sequence based on the volatility of preceding data using an adaptive proximity weighting method, construct a multivariate dynamic evolution model based on the weighted benchmark sequence, the preprocessed original weighted data sequence, and multi-source perturbation variables, and estimate the unknown parameters in the multivariate dynamic evolution model. The prediction and verification module is used to solve the analytical expression of the equation based on the established multivariate dynamic evolution model, obtain the predicted value of the cumulative sequence at any time based on the analytical expression, and then determine the original weight prediction value corresponding to the abnormal data point based on the predicted value of the cumulative sequence at any time. The adaptive correction module is used to determine the residual between the model's predicted weight value and the actual measured weight value, evaluate the current accuracy of the multivariate dynamic evolution model based on the residual, and determine whether to output a complete and verified weighing data sequence and the weight value within the prediction interval, or to trigger the adaptive correction strategy according to a preset priority order, based on the evaluation result.

Images