A weighing target value regulation system based on raw material information feedback

Abstract

The application discloses a kind of based on raw material information feedback's weighing target value regulation and control system, specifically related to industrial automatic control technical field, raw material characteristic identification module exports characteristic feature group, its internal pre-set correlation screening model is according to feedback screening high correlation feature, constructs change matrix and generates dynamic compensation coefficient, and optimizes characteristic feature group output.Target value determination module determines weighing target value accordingly.After real-time weighing, deviation calculation module calculates the deviation of actual weighing and target value.Regulation and control execution module adjusts feeding equipment according to deviation, and uses dynamic compensation coefficient to correct adjustment amount, so that actual weighing approaches target value;The application constructs change matrix and generates dynamic compensation coefficient by screening high correlation feature, improves compensation pertinence;Regulation and control result is fed back to optimize characteristic identification, realizes self-learning closed loop;Dynamic compensation coefficient is used to correct adjustment amount, so that feeding equipment is self-adapting adjustment, significantly improves weighing precision, dynamic stability and repeatability.

Description

Technical Field

[0001] This invention relates to the field of industrial automatic control technology, and more specifically, to a weighing target value control system based on raw material information feedback. Background Technology

[0002] During dynamic batching and weighing processes, the physical properties of raw materials can fluctuate due to changes in batch size, origin, storage conditions, or ambient temperature and humidity. These changes directly affect the conveying stability of the feeding equipment and the accuracy of the actual weighing. Existing methods for controlling the weighing target value mostly rely on pre-set fixed target values ​​or adjust based solely on the deviation between real-time weighing data and the target value. While these methods are effective when raw material properties are relatively stable, they become less effective when raw material properties change significantly, such as when the material becomes viscous or loose. Relying solely on deviation feedback often results in a lag, making it difficult to quickly bring the actual weighing close to the target value.

[0003] Some systems attempt to detect the characteristics of the raw materials before weighing and make a one-time correction to the target value or feeding parameters accordingly. However, this correction is an open-loop compensation and does not feed back the deviation information generated during subsequent control processes to the characteristic identification stage. Therefore, the accuracy of characteristic identification cannot be continuously improved with the increase of control cycles, and the system is also unable to automatically adapt to long-term drift of raw material characteristics. In addition, when multiple physical properties change simultaneously and their effects on weighing deviations are coupled, conventional proportional-integral-derivative control or simple linear compensation methods are difficult to effectively distinguish the degree of influence of different characteristics, resulting in limited control effects. Therefore, this invention proposes a weighing target value control system based on raw material information feedback to solve the above problems. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: A weighing target value control system based on raw material information feedback includes: The raw material characteristic identification module is used to identify the characteristics of raw materials and output the characteristic feature group of raw materials. The raw material characteristic identification module has a preset correlation screening model. The preset correlation screening model is used to receive feedback from the control results, compare each feature category in the characteristic feature group with a preset correlation threshold, screen out at least one feature category whose correlation is greater than the preset correlation threshold, construct a preset change matrix based on the screened at least one feature category, and generate dynamic compensation coefficients according to the preset change matrix. At the same time, the output of the characteristic feature group is optimized using the dynamic compensation coefficients. The target value determination module is used to determine the weighing target value by calling the preset target value determination model based on the optimized characteristic feature group; The real-time weighing module is used to weigh the raw materials in real time and obtain the actual weighing data; The deviation calculation module is used to calculate the deviation value based on the actual weighing data and the target value determined by the target value determination module. The control execution module is used to adjust the operating parameters of the feeding equipment based on the deviation value, so that the actual weighing data approaches the target weighing value. The control execution module is also used to receive the dynamic compensation coefficient generated by the raw material characteristic identification module, and use the dynamic compensation coefficient to correct the adjustment amount of the operating parameters of the feeding equipment. It also feeds back the control results to the raw material characteristic identification module to form a closed-loop control.

[0005] In a preferred embodiment, the characteristic group of the raw material is obtained by combining at least three characteristic categories: density, moisture, and flowability.

[0006] In a preferred embodiment, the generation of dynamic compensation coefficients by the preset correlation screening model is achieved through the following steps: Obtain the deviation value sequence from historical control results and the measured value sequence of each feature category in the characteristic feature group; Calculate the Pearson correlation coefficient between the measured value sequence and the deviation value sequence for each feature category, and mark the feature categories with a correlation coefficient greater than a preset correlation threshold as highly correlated features; When there are multiple highly correlated features selected, the measured values ​​of all highly correlated features at the current time are used to form a vector, and the preset change matrix is ​​constructed as a symmetric positive semi-definite matrix, where the element in the i-th row and j-th column of the matrix is ​​the product of the covariance of highly correlated feature i and highly correlated feature j divided by their respective standard deviations; when there is only one highly correlated feature selected, the preset change matrix is ​​constructed as a one-by-one matrix, and the unique element of the matrix is ​​one. Solve the linear equation system, where the coefficient matrix of the linear equation system is a preset change matrix, the unknown vector is the change increment vector, and the constant term is the difference between the current measured value vector of the preset convergence step size multiplied by the highly correlated feature and the preset standard feature vector. The dynamic compensation coefficient is obtained by mapping the length of the change increment vector to a closed interval between zero and one.

[0007] In a preferred embodiment, the preset relevance screening model is also used to perform the following operations: When the correlation coefficient between all feature categories in the feature group and the deviation value sequence formed by multiple deviation values ​​in the historical control results is less than or equal to the preset correlation threshold, the preset correlation screening model abandons the construction of the preset change matrix and generates a flag bit with no high correlation feature and passes it to the control execution module. When the control execution module receives a flag bit with no high correlation characteristics, the dynamic compensation coefficient does not participate in the correction of the adjustment amount, but only adjusts the operating parameters of the feeding equipment based on the deviation value. When at least one highly relevant feature is selected, the preset correlation screening model optimizes the output of the feature group using the generated dynamic compensation coefficient. Specifically, the value of each highly relevant feature in the current feature group is multiplied by the dynamic compensation coefficient to obtain the optimized value of the highly relevant feature. The values ​​of non-highly relevant features remain unchanged. The optimized feature group is then reconstructed from all the optimized feature values ​​and output to the target value determination module.

[0008] In a preferred embodiment, the preset target value determination model is a nonlinear mapping model based on a radial basis function neural network. This model uses the optimized characteristic feature group as the input layer node and the weighing target value as the output layer node. The center vector and width parameters of the hidden layer nodes of the network are predetermined by historical batching data through a preset clustering algorithm.

[0009] In a preferred embodiment, when the deviation calculation module calculates the deviation value, it subtracts the target weight value from the actual weighing data to obtain a signed difference value, then takes the absolute value of the signed difference value as the absolute deviation, and compares the absolute deviation with a preset dead zone threshold. If the absolute deviation is less than the preset dead zone threshold, the deviation value is set to zero; if the absolute deviation is greater than or equal to the preset dead zone threshold, the signed difference value is retained as the deviation value.

[0010] In a preferred embodiment, the control execution module corrects the adjustment amount using a dynamic compensation coefficient through the following steps: The first step is to calculate the initial adjustment amount by calling the preset proportional-integral-derivative (PID) controller inside the control execution module based on the deviation value of the current control cycle. The calculation method of the PID controller is as follows: multiply the deviation value of the current control cycle by a preset proportional coefficient to obtain the proportional term value; multiply the sum of the deviation values ​​of all historical control cycles before the current control cycle by a preset integral coefficient to obtain the integral term value; multiply the difference between the deviation value of the current control cycle and the deviation value of the previous control cycle by a preset derivative coefficient to obtain the derivative term value; and use the sum of the proportional term value, integral term value, and derivative term value as the initial adjustment amount. This initial adjustment amount represents the percentage of the feeder motor speed to be adjusted. The second step is to obtain the historical final adjustment value of each historical control cycle within the last five control cycles, and for each historical control cycle, calculate the theoretical adjustment value of the historical control cycle according to the calculation method in the first step using the same preset proportional integral derivative controller based on the historical deviation value corresponding to the historical control cycle. Then, calculate the ratio of the historical final adjustment value to the theoretical adjustment value, and use all ratios to form a historical correction sequence. The third step is to calculate the exponentially weighted moving average of the historical correction sequence, where the exponential decay factor of the exponentially weighted moving average is recalibrated by the sum of the dynamic compensation coefficient and the preset benchmark decay factor. The fourth step is to construct a correction function. The function value of the correction function is the dynamic compensation coefficient multiplied by the negative preset smoothing factor of the natural constant e multiplied by the absolute value of the deviation ratio, where the deviation ratio is the ratio of the deviation value of the current control cycle to the preset maximum allowable deviation. The fifth step is to multiply the initial adjustment amount by the function value of the correction function to obtain the corrected final adjustment amount for the current control cycle, and then drive the motor speed of the feeding device according to the corrected final adjustment amount.

[0011] In a preferred embodiment, the control results fed back from the control execution module to the raw material characteristic identification module include: The steady-state deviation value after this adjustment, the actual response time of the feeding equipment during the adjustment process, and the actual correction effect after the application of the dynamic compensation coefficient are identified. The raw material characteristic identification module updates the sample database inside the preset correlation screening model based on the adjustment result, and recalculates the Pearson correlation coefficient between the measured value sequence and the deviation value sequence of each feature category in the sample database. The recalculated correlation coefficient is used as the updated correlation confidence of each feature category for comparison in the next correlation screening process.

[0012] In a preferred embodiment, the control execution module also monitors the current rotation speed and current feed flow rate of the feeder in real time during the adjustment of the operating parameters of the feeder, and compares the current rotation speed with the preset upper speed threshold and the preset lower speed threshold, and compares the current feed flow rate with the preset upper flow rate threshold and the preset lower flow rate threshold; if any one of them exceeds the corresponding threshold range, a preset safety protection mode is triggered, and the feeder is forcibly switched to a preset constant speed operation state.

[0013] The technical effects and advantages of this invention are as follows: This invention utilizes a preset correlation screening model within the raw material characteristic identification module to compare each characteristic category in a characteristic feature group with a preset correlation threshold. It then filters out at least one characteristic category whose correlation exceeds the threshold and constructs a preset change matrix based on the selected characteristic categories to generate dynamic compensation coefficients. This invention can distinguish one or more characteristic categories most strongly correlated with weighing deviation from multiple raw material characteristics and construct a change matrix based on the covariance relationship between these characteristics, thereby generating physically meaningful dynamic compensation coefficients. These dynamic compensation coefficients are used to optimize the output of the characteristic feature group and to correct the adjustment of the feeding equipment's operating parameters. This ensures that the compensation action is not blindly or equally applied to all characteristics, but rather selectively and emphatically targets the key characteristics that truly affect weighing deviation for precise compensation. This mechanism significantly improves the targeting of control and avoids additional disturbances introduced by ineffective compensation of irrelevant characteristics, thus maintaining high weighing accuracy even under changing raw material characteristics.

[0014] This invention feeds back the control results of the control execution module to the raw material characteristic identification module, enabling the preset correlation screening model to continuously optimize the output of the characteristic feature group based on the actual control results, forming a complete closed-loop control. Through the feedback loop, the actual deviation value of each control, the response time of the feeding equipment, and the correction effect of the dynamic compensation coefficient are all returned to the preset correlation screening model to update the sample database and recalculate the Pearson correlation coefficient between each feature category and the deviation value sequence. As the number of controls increases, the preset correlation screening model becomes increasingly accurate in judging the correlation of each feature category, and the output quality of the characteristic feature group continuously improves, thus making the input information relied upon by the target value determination module more reliable. This self-learning and self-optimization capability allows the system to automatically adapt to the slow drift or batch changes of raw material characteristics during long-term operation, eliminating the need for frequent manual calibration.

[0015] In this invention, the control and execution module adjusts the operating parameters of the feeding equipment using the deviation value, while simultaneously receiving a dynamic compensation coefficient generated by the raw material characteristic identification module. This dynamic compensation coefficient is then used to correct the adjustment amount of the feeding equipment's operating parameters. The dynamic compensation coefficient originates from the correlation analysis between raw material characteristics and deviation, reflecting the degree of deviation between the current physical characteristics of the raw material and the standard state. When the dynamic compensation coefficient is small, it indicates a large deviation between the raw material characteristics and the standard state, requiring a strong attenuation of the initial adjustment amount to avoid over-adjustment. When the dynamic compensation coefficient is close to one, it indicates that the raw material characteristics are close to the standard state, and the initial adjustment amount can be used directly. This method of dynamically correcting the adjustment amount based on raw material characteristic information enables the feeding equipment to adaptively adjust its movement amplitude according to the characteristics of different batches of raw materials. This ensures that the weighing quickly approaches the target value while effectively suppressing overshoot and oscillation, significantly improving the dynamic response stability of the system and the repeatability of the final weighing results. Attached Figure Description

[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of a weighing target value control system based on raw material information feedback in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] Reference Figure 1 The following examples were obtained: Example 1: A weighing target value control system based on raw material information feedback, comprising: The raw material characteristic identification module is used to identify the characteristics of raw materials and output the characteristic feature group of raw materials. The raw material characteristic identification module has a preset correlation screening model. The preset correlation screening model is used to receive feedback from the control results, compare each feature category in the characteristic feature group with a preset correlation threshold, screen out at least one feature category whose correlation is greater than the preset correlation threshold, construct a preset change matrix based on the screened at least one feature category, and generate dynamic compensation coefficients according to the preset change matrix. At the same time, the output of the characteristic feature group is optimized using the dynamic compensation coefficients. The target value determination module is used to determine the weighing target value by calling the preset target value determination model based on the optimized characteristic feature group; The real-time weighing module is used to weigh the raw materials in real time and obtain the actual weighing data; The deviation calculation module is used to calculate the deviation value based on the actual weighing data and the target value determined by the target value determination module. The control and execution module is used to adjust the operating parameters of the feeding equipment based on the deviation value, so that the actual weighing data approaches the target weighing value. The control and execution module is also used to receive the dynamic compensation coefficient generated by the raw material characteristic identification module, and use the dynamic compensation coefficient to correct the adjustment amount of the operating parameters of the feeding equipment. It also feeds back the control results to the raw material characteristic identification module to form a closed-loop control.

[0019] The raw material's characteristic set is derived from at least three categories: density, moisture content, and flowability. The raw material characteristic identification module has a pre-set spectral sensor array that emits light with wavelengths from 900 nm to 1700 nm to illuminate the raw material. It directly outputs the density characteristic value based on the intensity change of the reflectance spectrum, without requiring an additional model. For example, corn flour has a reflectance of 42% at a wavelength of 1100 nm. The correspondence between this reflectance and density is pre-calibrated using experimental data. The calibration result shows that every 2% change in reflectance corresponds to a density change of 0.01 g / cm³, therefore the density characteristic is 0.65 g / cm³. The wavelength range of the spectral sensor array is selected based on empirical values ​​of the light absorption characteristics of common raw materials, and the density output accuracy is 0.5%.

[0020] The raw material characteristic identification module has a preset dielectric constant sensor. This sensor measures the capacitance value of the raw material at a 10 MHz excitation frequency. The capacitance value is directly converted into a humidity characteristic value. The conversion factor is given in the equipment manual as 0.3% humidity per picofarad of capacitance. For example, if the measured capacitance value is 47 picofarads, the humidity characteristic is 14.1%. The measurement range of this sensor is set to 5% to 35% based on statistical results, with an accuracy of 0.5%.

[0021] The raw material characteristic identification module has a preset angle of repose measurement unit. A laser profilometer scans the inclined surface of the raw material pile and automatically calculates the angle of repose. This angle value is directly used as a reverse indicator of the flowability characteristic. For ease of subsequent processing, the angle of repose is converted into a flowability characteristic value between zero and one using a preset linear mapping formula. The mapping formula is: flowability equals one minus the ratio of the angle of repose divided by 90 degrees. For example, if the angle of repose is 38 degrees, the flowability characteristic is 0.578; if the angle of repose is 50 degrees, the flowability characteristic is 0.444. The angle resolution of the angle of repose measurement unit is 0.1 degrees, and the repeatability error is 2% based on experimental data. The density characteristic, moisture characteristic, and flowability characteristic are combined into a characteristic feature group. For example, the characteristic feature group for the corn flour mentioned above is [0.65, 14.1, 0.578]. This characteristic feature group is output to a preset correlation screening model.

[0022] The characteristic feature group of raw materials can be expanded to include other physical or chemical properties according to specific application scenarios. The following are some examples of other features that can be included in the characteristic feature group. These features can also be acquired by preset sensors or detection units and participate in correlation screening and dynamic compensation coefficient generation: Temperature characteristics: The temperature of raw materials affects their internal molecular activity and surface adhesion, thereby altering the flow characteristics during feeding and the zero drift of the weighing sensor. For example, in the food or chemical industries, refrigerated powdered raw materials are at low temperatures and easily absorb moisture and clump, leading to decreased flowability and affecting feeding speed and weighing accuracy. Temperature characteristics can be obtained through a preset infrared temperature sensor or a contact-type resistance temperature sensor. The sampling frequency is typically 1 Hz, and the measurement range is set from -10 degrees Celsius to 50 degrees Celsius depending on the type of raw material, with a typical resolution of 0.1 degrees Celsius.

[0023] Particle size characteristics: The particle size distribution of raw materials directly affects their bulk density, porosity, and angle of repose, thus impacting the filling efficiency and weighing stability of the feeding equipment. For example, raw materials with larger particle diameters are prone to periodic pulsation in screw feeders, while fine powdery raw materials are prone to wall adhesion or scattering. Particle size characteristics can be obtained using a preset laser particle size analyzer or vibrating screen device, with a typical detection range of 10 micrometers to 5 millimeters. The output is the median diameter value corresponding to the cumulative distribution percentage, for example, a median diameter of 200 micrometers.

[0024] Electrostatic charge characteristics: Some powdery raw materials generate static electricity due to friction during pneumatic or screw conveying. Static electricity accumulation causes the material to adhere to the inner wall of the pipe or the surface of the feeder, resulting in the actual feed rate deviating from the set value. For example, when plastic powder or flour is conveyed in a dry environment, the electrostatic potential can reach several kilovolts, significantly affecting weighing repeatability. Electrostatic charge characteristics can be obtained through a preset electrostatic potential sensor or charge meter, with a typical measurement range of -10 kilovolts to +10 kilovolts and a resolution of 10 volts.

[0025] Ambient humidity characteristics: In addition to the moisture content of the raw materials themselves, the relative humidity of the surrounding air also affects the moisture absorption rate of the raw material surface, especially during open feeding or short-term storage. Ambient humidity can be obtained through preset temperature and humidity sensors, with a typical measurement range of 10% to 90% and a resolution of 1%.

[0026] The introduction of the above feature categories requires the use of corresponding sensors or detection units. After normalizing the detected values, they are combined with density, humidity, and flowability to form an extended characteristic feature group. The preset correlation screening model will recalculate the correlation coefficient between all feature categories and the deviation value sequence. Only feature categories with a correlation coefficient greater than the preset correlation threshold (such as 0.3) will be marked as highly correlated features and participate in the subsequent construction of the change matrix and the generation of dynamic compensation coefficients.

[0027] The generation of dynamic compensation coefficients by the preset correlation screening model is achieved through the following steps: A pre-set correlation screening model is used to obtain the deviation value sequence from historical control results and the measured value sequence of each feature category in the characteristic feature group. The deviation value sequence is the deviation value at the end of the last twenty control operations. The window length is determined based on experimental data and covers typical cycles of raw material characteristic changes. The measured value sequence and the deviation value sequence correspond one-to-one in time. For example, in the most recent control operation, the deviation value was +5 kg, while the measured density was 0.65 g / cm³, the measured humidity was 14.3, and the measured flowability was 0.8.

[0028] Calculate the Pearson correlation coefficient between the measured value series and the deviation value series for each feature category. The Pearson correlation coefficient is the product of the covariance of the two series divided by their respective standard deviations, and its value ranges from -1 to +1. A preset correlation threshold of 0.3 is set based on statistical experience; this threshold is used to distinguish the strength of the correlation. If the correlation coefficient between the density measured value series and the deviation value series is 0.6, it is greater than 0.3, and the density feature is marked as a highly correlated feature. If the correlation coefficient of the humidity feature is 0.1, it is less than 0.3 and is not marked as a highly correlated feature. If the correlation coefficient of the mobility feature is -0.4, its absolute value is greater than 0.3 and it is negatively correlated, it is also marked as a highly correlated feature, because a negative correlation also indicates a significant impact.

[0029] When multiple highly correlated features are selected, the measured values ​​of all highly correlated features at the current time are constructed into a vector. For example, the measured density value of 0.65 and the measured flow rate value of 0.8 form a two-dimensional vector [0.65, 0.8]. The model constructs the preset change matrix as a symmetric positive semi-definite matrix, where the number of rows and columns is equal to the number of highly correlated features. The element in the i-th row and j-th column of the matrix is ​​the product of the covariance of highly correlated feature i and highly correlated feature j divided by their respective standard deviations; this product is the Pearson correlation coefficient. For the case of two highly correlated features, the preset change matrix is ​​a second-order symmetric matrix, where all diagonal elements are the value 1, and the off-diagonal elements are the correlation coefficients of density and flow rate, such as -0.4. When only one highly correlated feature is selected, the model constructs the preset change matrix as a one-by-one matrix, where the unique element is the value 1, to maintain consistency in mathematical operations.

[0030] Solve the linear equation system where the coefficient matrix is ​​a preset change matrix, the unknown vector is the change increment vector, and the constant term is the difference between the current measured value vector and the preset standard eigenvector, multiplied by a preset convergence step size. The preset convergence step size is set to 0.1 based on experimental data of the control system's response speed, and this step size controls the magnitude of each adjustment. The preset standard eigenvector is the characteristic value obtained by calibrating the equipment under standard raw material conditions, for example, a density standard value of 0.70 and a flowability standard value of 0.60. This vector originates from the calibration experiment before the equipment leaves the factory. The difference vector is the result of subtracting the standard eigenvector from the current measured value vector; for example, [0.65 minus 0.70, 0.8 minus 0.60] equals [-0.05, 0.20]. The constant term equals 0.1 multiplied by this difference vector, resulting in [-0.005, 0.020]. Solving the linear equation system yields the value of the change increment vector; for example, the solution result is [-0.004, 0.018].

[0031] The dynamic compensation coefficient is obtained by mapping the length of the increment vector to a closed interval between zero and one. The length of the increment vector is the square root of the sum of the squares of its components. For example, a length of √0.00016 plus 0.000324 equals √0.000484, which is approximately 0.022. The model uses a linear mapping formula: the mapped value equals the length of the increment vector divided by a preset maximum length, which is set to 0.1 based on historical data statistics. If the length exceeds 0.1, it is set to 1. 0.022 divided by 0.1 equals 0.22, so the dynamic compensation coefficient is 0.22. If the length of the increment vector is zero or negative, the mapping result is zero; if the length is greater than or equal to 0.1, the mapping result is one. This dynamic compensation coefficient is used for subsequent optimization of the characteristic feature group's output and correction adjustment.

[0032] The preset relevance screening model is also used to perform the following operations: The preset correlation screening model checks the correlation coefficients between all feature categories in the characteristic feature group and the deviation value sequence. This preset correlation threshold is set to 0.3 based on statistical experience. When the correlation coefficients of all feature categories are less than or equal to 0.3, it indicates that no feature significantly affects the deviation. For example, the Pearson correlation coefficients between density, humidity, and fluidity and the deviation value sequence are 0.1, 0.2, and -0.15, respectively, all below 0.3. In this case, the model abandons the construction of the preset change matrix because the change matrix cannot provide effective compensation when high-correlation features are lacking. Instead, it generates a flag indicating no high-correlation features. This flag is a binary signal; a value of 1 represents no high-correlation features, and this flag is passed to the control execution module.

[0033] When the control execution module receives a flag indicating no highly correlated features, it confirms that there are no highly correlated features in the current control cycle. This module ignores the dynamic compensation coefficient and does not participate in the correction of the adjustment amount. The module adjusts the operating parameters of the feeding device only based on the deviation value. At this point, the adjustment method degenerates into pure feedback control. This pure feedback control uses a preset pure feedback proportional coefficient, which is derived from the no-load test data at the time of equipment delivery. A typical value is, for example, 0.01, indicating that each kilogram deviation corresponds to a 1% speed adjustment. For example, if the deviation is positive 5 kilograms, the adjustment amount is 5 multiplied by 0.01, which equals 5%, i.e., reducing the feeding motor speed by 5%. No additional correction using the dynamic compensation coefficient is used. This strategy is determined based on experimental data. When the correlation of all features is weak, introducing a compensation coefficient may introduce noise; therefore, a complete degeneration into proportional control is chosen.

[0034] When at least one highly relevant feature is selected, the preset correlation screening model optimizes the output of the characteristic feature group using the generated dynamic compensation coefficient. The optimization method involves multiplying the dynamic compensation coefficient by the value of each highly relevant feature in the current characteristic feature group to obtain the optimized highly relevant feature value. The values ​​of non-highly relevant features remain unchanged. For example, if the dynamic compensation coefficient is 0.22, and the current highly relevant features are density 0.65 and flowability 0.8, then the optimized density becomes 0.65 multiplied by 0.22 equals 0.143, and the optimized flowability becomes 0.8 multiplied by 0.22 equals 0.176. The non-highly relevant feature humidity of 14.3 remains unchanged. The model reconstructs the characteristic feature group from all the optimized feature values; for example, the new characteristic feature group is [0.143, 14.3, 0.176]. This optimized characteristic feature group is output to the target value determination module to recalculate the weighing target value. The principle behind this optimization is that the dynamic compensation coefficient reflects the direction of deviation between the current raw material characteristics and the standard characteristics. Multiplying by this coefficient can weaken the weight of highly relevant features in the target value determination, avoiding over-compensation.

[0035] The target value determination model is a nonlinear mapping model based on a radial basis function (RBF) neural network. A RBF neural network is a three-layer feedforward network where hidden layer neurons use radial basis functions as activation functions. The value of the radial basis function depends on the distance between the input vector and the center vector. The model uses optimized feature sets as input layer nodes, with the number of input layer nodes equal to the number of feature categories in the optimized feature set. For example, if the optimized feature set includes density, humidity, and fluidity, then the input layer has three nodes. The model uses the target weight value as the output layer node, with only one node in the output layer. The output value is the desired target weight value, such as 100 kilograms.

[0036] The center vector and width parameters of the hidden layer nodes in the network are pre-determined using a preset clustering algorithm based on historical batching data. The historical batching data consists of data pairs between optimized characteristic feature groups and actual weighing target values ​​recorded during the past one hundred successful batching processes; the amount of data is set based on empirical values. The preset clustering algorithm uses K-means clustering, with the number of clusters set to eight based on experimental data, indicating that the historical data is divided into eight typical raw material categories. The center vector is the centroid of each cluster, i.e., the average value of all optimized characteristic feature groups within that cluster. For example, the center vector of the first cluster is [0.60, 12.5, 0.70]. The width parameter is the standard deviation of each feature within each cluster multiplied by a preset expansion coefficient. The expansion coefficient is recommended to be 1.5 according to the equipment manual. The width parameter determines the influence range of the radial basis function; for example, the width parameter of the first cluster is [0.08, 1.2, 0.05]. When a new optimized characteristic set is input, the model calculates the Euclidean distance between the input and each center vector, then converts it into the hidden layer output using radial basis functions. Finally, the hidden layer output is multiplied by the output layer weights and summed to obtain the target weighing value. The output layer weights are obtained using the least squares method based on historical batching data. For example, when the optimized characteristic set [0.143, 14.3, 0.176] is input, the model outputs a target weighing value of 98.5 kg.

[0037] When calculating the deviation value, the deviation calculation module subtracts the target weight value from the actual weighing data to obtain the difference. The absolute value of this difference is then taken as the absolute deviation. For example, if the actual weighing data is 105 kg and the target weight value is 100 kg, the difference is positive 5 kg, and the absolute deviation is 5 kg. If the actual weighing data is 95 kg and the target weight value is 100 kg, the difference is negative 5 kg, and the absolute deviation is also 5 kg. This absolute deviation is a non-negative number used for comparison with the preset dead zone threshold.

[0038] This module compares the absolute deviation with a preset dead zone threshold. The preset dead zone threshold is set based on empirical values ​​of the weighing sensor's noise level and the equipment's mechanical vibration, for example, to two kilograms. This threshold represents the allowable weighing error range. If the absolute deviation is less than the preset dead zone threshold, for example, an absolute deviation of 1.5 kilograms less than two kilograms, the deviation value is set to zero. Setting it to zero means that the current error is considered within an acceptable range, and no subsequent adjustment action is triggered. If the absolute deviation is greater than or equal to the preset dead zone threshold, for example, an absolute deviation of five kilograms greater than or equal to two kilograms, the original difference is retained as the deviation value. This original difference may be positive or negative, for example, retaining +5 kilograms or -5 kilograms. Retaining the signed difference reflects whether the actual weighing is too high or too low, so that the control execution module can decide to increase or decrease the feeding amount. For example, +5 kilograms indicates that the actual weight is overweight, requiring a reduction in feeding; -5 kilograms indicates that the actual weight is underweight, requiring an increase in feeding.

[0039] The control execution module uses dynamic compensation coefficients to correct the adjustment amount through the following steps: The control execution module first calls a preset proportional-integral-derivative (PID) controller to calculate the initial adjustment based on the deviation value of the current control cycle. The preset proportional coefficient, preset integral coefficient, and preset derivative coefficient are all derived from step response experiments during equipment commissioning. For example, the proportional coefficient is set to 0.5, the integral coefficient to 0.1, and the derivative coefficient to 0.05. During calculation, the deviation value of the current control cycle is multiplied by the preset proportional coefficient to obtain the proportional term value. For example, if the deviation value is +5 kg, the proportional term is 5 multiplied by 0.5, which equals 2.5. The sum of the deviation values ​​of all previous control cycles is multiplied by the preset integral coefficient to obtain the integral term value. Assuming the sum is 10 kg, the integral term is 10 multiplied by 0.1, which equals 1.0. The difference between the deviation value of the current control cycle and the deviation value of the previous control cycle is multiplied by the preset derivative coefficient to obtain the derivative term value. If the previous deviation value was 3 kg, the difference is 2 kg, and the derivative term is 2 multiplied by 0.05, which equals 0.1. The sum of the proportional, integral, and derivative terms is used as the initial adjustment amount, i.e., 2.5 plus 1.0 plus 0.1 equals 3.6. This initial adjustment amount indicates that the motor speed of the feeding equipment needs to be increased by 3.6%.

[0040] The control execution module acquires the historical final adjustment value for each historical control cycle within the last five control cycles. The historical final adjustment value refers to the corrected final adjustment value calculated in step five of that historical cycle. For example, the historical final adjustment values ​​for the past five cycles are 3.2%, 2.8%, 3.5%, 3.0%, and 2.9%, respectively. For each historical control cycle, based on the corresponding historical deviation value, the theoretical adjustment value for that historical cycle is calculated using the same preset proportional-integral-derivative controller according to the calculation method in step one. For example, if the historical deviation value for a certain historical cycle is positive 4 kg, the cumulative sum is 8 kg, and the difference is 1 kg, then the theoretical adjustment value is 4 multiplied by 0.5 plus 8 multiplied by 0.1 plus 1 multiplied by 0.05 equals 2.0 plus 0.8 plus 0.05 equals 2.85. Then, the ratio of the historical final adjustment value to the theoretical adjustment value is calculated. For example, 3.2 divided by 2.85 is approximately equal to 1.12. All five ratios constitute the historical correction sequence, such as [1.12, 0.98, 1.05, 1.02, 0.96].

[0041] The control execution module calculates the exponentially weighted moving average of the historical correction sequence. The formula for the exponentially weighted moving average is: current average = smoothing factor multiplied by current value + (smoothing factor multiplied by the previous average value) / (smoothing factor multiplied by the previous average value). The smoothing factor is typically set to 0.3. The exponential decay factor of this exponentially weighted moving average is recalibrated by the sum of the dynamic compensation coefficient and the preset baseline decay factor. The preset baseline decay factor is set to 0.2 based on experience. If the dynamic compensation coefficient is, for example, 0.22, then the decay factor is 0.2 + 0.22 = 0.42. The larger this decay factor is, the faster the weight of the historical data decays. The smoothing factor in the exponentially weighted moving average is recalibrated using this decay factor. The smoothing factor is equal to one minus the negative power of the natural constant e, for example, one minus 0.657 = 0.343. Then, the exponentially weighted moving average of the historical correction sequence is calculated based on this smoothing factor, for example, yielding an average value of 1.03.

[0042] The control execution module constructs a correction function. The function value of this correction function is the dynamic compensation coefficient multiplied by the negative preset smoothing factor of the natural constant e, multiplied by the absolute value of the deviation ratio raised to a power. The preset smoothing factor is set to 2.0 based on experimental data. The deviation ratio is the ratio of the deviation value of the current control cycle to the preset maximum permissible deviation. The preset maximum permissible deviation is set at 5% of the equipment's rated feeding capacity; for example, if it is 10 kg, and the current deviation value is 5 kg, then the deviation ratio is 0.5. The natural constant e is calculated as -2.0 multiplied by 0.5 raised to the power of 0.5, i.e., -1.0 raised to the power of 0.368. The function value of the correction function is 0.22 multiplied by 0.368, which equals 0.0811.

[0043] The control execution module multiplies the initial adjustment amount by the function value of the correction function to obtain the corrected final adjustment amount for the current control cycle. The initial adjustment amount is 3.6%, multiplied by 0.0811, resulting in 0.292%, which is the corrected final adjustment amount. The control execution module drives the motor speed of the feeding device according to this corrected final adjustment amount, for example, decreasing or increasing the motor speed by 0.292%. If the corrected final adjustment amount is positive, the speed is increased; if it is negative, the speed is decreased. This drive command is output to the feeding motor driver via a pulse width modulation signal.

[0044] The control results fed back from the control execution module to the raw material characteristic identification module include: The control execution module feeds back three control results to the raw material characteristic identification module. The first is the steady-state deviation value after the adjustment is completed. This value is defined as the absolute value of the difference between the actual weighing data and the target weighing value at the end of the adjustment. The sampling time is the average value of five consecutive seconds after the feeder speed stabilizes. For example, the steady-state deviation value is 0.3 kg. The second item is the actual response time of the feeding equipment during the adjustment process. This time starts from when the control execution module sends the drive signal and ends when the actual weighing data first enters the range of 5% of the steady-state deviation value. For example, the actual response time is 1.2 seconds. The third item is the actual correction effect indicator after the dynamic compensation coefficient is applied. This indicator is a binary flag. When the final adjustment amount after correction makes the steady-state deviation value decrease by more than 20% compared with the previous adjustment, it is marked as one; otherwise, it is zero. For example, if the deviation drops from 5 kg to 3.8 kg after a certain adjustment, it is marked as one.

[0045] The raw material characteristic identification module updates the sample database within the preset correlation screening model based on the control results. For example, this sample database stores complete records of the last two hundred controls, including the characteristic feature group before each control, the deviation value sequence during the control process, and the above three feedback results. The database capacity is set according to the experimental data to ensure coverage of the seasonal variation cycle of the raw materials. For example, after a certain control, the steady-state deviation value of 0.3 kg, the actual response time of 1.2 seconds, and the correction effect indicator are stored in the corresponding record.

[0046] The raw material characteristic identification module recalculates the Pearson correlation coefficient between the measured value sequences and deviation value sequences of each characteristic category in the sample database. This module extracts the measured density value sequences and corresponding deviation value sequences from the sample database for the most recent one hundred adjustments, and calculates the correlation coefficient according to the Pearson formula. For example, the correlation coefficient between density and deviation is 0.65. Similarly, the correlation coefficient between humidity and deviation is calculated to be 0.12, and the correlation coefficient between flowability and deviation is -0.38. This calculation is performed once every ten adjustments to reduce the computational burden.

[0047] The raw material characteristic identification module uses the recalculated correlation coefficients as the updated correlation confidence scores for each feature category. These confidence scores directly replace the original correlation coefficient values ​​and are used for comparison in the next correlation screening process. For example, if the original density correlation confidence score was 0.55 and it is updated to 0.65, then 0.65 will be compared with the preset correlation threshold of 0.3 in the next screening. If the confidence score of a feature category is updated from 0.28 to 0.31, it changes from a non-highly correlated feature to a highly correlated feature. This update mechanism enables the preset correlation screening model to adaptively track long-term changes in raw material characteristics.

[0048] During the adjustment of the operating parameters of the feeding equipment, the control and execution module monitors the current rotational speed and current feed flow rate of the feeding equipment in real time. The current rotational speed is obtained through a Hall effect speed sensor installed on the motor shaft, with a sampling frequency of 20 times per second and a measurement unit of revolutions per minute. The current feed flow rate is obtained through a solid flow meter installed at the discharge port, with a sampling frequency of 10 times per second and a measurement unit of kilograms per hour.

[0049] The module compares the current rotational speed with preset upper and lower speed thresholds. The upper speed threshold is set according to the upper limit of the motor's safe operating range in the equipment manual. For example, if the motor's rated speed is 1500 rpm, the upper speed threshold is between 90% and 95% of the rated speed, with a typical value of 1425 rpm. The lower speed threshold is between 5% and 10% of the rated speed, with a typical value of 75 rpm. The module also compares the current feed flow rate with preset upper and lower flow rate thresholds. The upper flow rate threshold is set according to 80% to 90% of the feeder's maximum conveying capacity. For example, if the maximum conveying capacity is 10 tons per hour, the upper flow rate threshold is between 8 and 9 tons per hour, with a typical value of 8.5 tons per hour. The lower flow rate threshold is set according to the minimum flow rate at which the feeder operates stably. For example, if the minimum stable flow rate is 0.5 tons per hour, the lower flow rate threshold is between 0.4 and 0.6 tons per hour, with a typical value of 0.5 tons per hour.

[0050] If the current rotational speed exceeds the preset upper speed limit or falls below the preset lower speed limit, or if the current feed flow rate exceeds the preset upper flow rate limit or falls below the preset lower flow rate limit, a preset safety protection mode is triggered. For example, if the current rotational speed is 1450 rpm and exceeds the upper limit of 1425 rpm, safety protection is immediately triggered. This module forces the feeding equipment to switch to a preset constant speed operating state. The preset constant speed operating state means that the motor runs at a fixed speed of 50% of its rated speed. This speed value comes from the equipment's no-load safety test data, with a typical value of 750 rpm. At this time, the system pauses closed-loop control and issues an audible and visual alarm signal until the operator manually resets it. This safety protection mode ensures that the equipment does not damage mechanical parts under extreme operating conditions.

[0051] The above-mentioned models or function formulas are all dimensionless and numerical calculations. The models or function formulas are obtained by software simulation based on a large amount of collected data to obtain the most recent real situation. The preset parameters in the models or function formulas are set by those skilled in the art according to the actual situation.

[0052] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0053] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0054] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0055] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A weighing target value control system based on raw material information feedback, characterized in that, include: The raw material characteristic identification module is used to identify the characteristics of raw materials and output the characteristic feature group of raw materials; The raw material characteristic identification module has a preset correlation screening model. The preset correlation screening model is used to receive feedback from the control results, compare each characteristic category in the characteristic feature group with the preset correlation threshold, screen out at least one characteristic category whose correlation is greater than the preset correlation threshold, construct a preset change matrix based on the screened at least one characteristic category, and generate dynamic compensation coefficients according to the preset change matrix. At the same time, the output of the characteristic feature group is optimized using the dynamic compensation coefficients. The target value determination module is used to determine the weighing target value by calling the preset target value determination model based on the optimized characteristic feature group; The real-time weighing module is used to weigh the raw materials in real time and obtain the actual weighing data; The deviation calculation module is used to calculate the deviation value based on the actual weighing data and the target value determined by the target value determination module. The control and execution module is used to adjust the operating parameters of the feeding equipment based on the deviation value, so that the actual weighing data approaches the target weighing value. The control and execution module is also used to receive the dynamic compensation coefficient generated by the raw material characteristic identification module, and use the dynamic compensation coefficient to correct the adjustment amount of the operating parameters of the feeding equipment; The control results are fed back to the raw material characteristic identification module to form a closed-loop control.

2. The weighing target value control system based on raw material information feedback according to claim 1, characterized in that, The characteristic group of raw materials is obtained by combining at least three characteristic categories: density, moisture, and flowability.

3. The weighing target value control system based on raw material information feedback according to claim 1, characterized in that, The generation of dynamic compensation coefficients by the preset correlation screening model is achieved through the following steps: Obtain the deviation value sequence and the measured value sequence of each feature category in the characteristic feature group from the historical control results; Calculate the Pearson correlation coefficient between the measured value sequence and the deviation value sequence for each feature category, and mark the feature categories with a correlation coefficient greater than a preset correlation threshold as highly correlated features; When there are multiple highly correlated features selected, the measured values ​​of all highly correlated features at the current time are used to form a vector, and the preset change matrix is ​​constructed as a symmetric positive semi-definite matrix, where the element in the i-th row and j-th column of the matrix is ​​the product of the covariance of highly correlated feature i and highly correlated feature j divided by their respective standard deviations; when there is only one highly correlated feature selected, the preset change matrix is ​​constructed as a one-by-one matrix, and the unique element of the matrix is ​​one. Solve the linear equation system, where the coefficient matrix of the linear equation system is a preset change matrix, the unknown vector is the change increment vector, and the constant term is the difference between the current measured value vector of the preset convergence step size multiplied by the highly correlated feature and the preset standard feature vector. The dynamic compensation coefficient is obtained by mapping the length of the change increment vector to a closed interval between zero and one.

4. The weighing target value control system based on raw material information feedback according to claim 3, characterized in that, The preset relevance screening model is also used to perform the following operations: When the correlation coefficient between all feature categories in the feature group and the deviation value sequence formed by multiple deviation values ​​in the historical control results is less than or equal to the preset correlation threshold, the preset correlation screening model abandons the construction of the preset change matrix and generates a flag bit with no high correlation feature and passes it to the control execution module. When the control execution module receives a flag bit with no high correlation characteristics, the dynamic compensation coefficient does not participate in the correction of the adjustment amount, but only adjusts the operating parameters of the feeding equipment based on the deviation value. When at least one highly relevant feature is selected, the preset correlation screening model optimizes the output of the feature group using the generated dynamic compensation coefficient. Specifically, the value of each highly relevant feature in the current feature group is multiplied by the dynamic compensation coefficient to obtain the optimized value of the highly relevant feature. The values ​​of non-highly relevant features remain unchanged. The optimized feature group is then reconstructed from all the optimized feature values ​​and output to the target value determination module.

5. The weighing target value control system based on raw material information feedback according to claim 1, characterized in that, The preset target value determination model is a nonlinear mapping model based on radial basis function neural network. This model uses the optimized characteristic feature group as the input layer node and the weighing target value as the output layer node. The center vector and width parameters of the hidden layer nodes of the network are predetermined by historical batching data through a preset clustering algorithm.

6. The weighing target value control system based on raw material information feedback according to claim 1, characterized in that, When calculating the deviation value, the deviation calculation module subtracts the target weight value from the actual weighing data to obtain a signed difference value. Then, it takes the absolute value of the signed difference value as the absolute deviation value and compares it with the preset dead zone threshold. If the absolute deviation is less than the preset dead zone threshold value, the deviation value is set to zero. If the absolute deviation is greater than or equal to the preset dead zone threshold value, the signed difference value is retained as the deviation value.

7. The weighing target value control system based on raw material information feedback according to claim 1, characterized in that, The control execution module uses dynamic compensation coefficients to correct the adjustment amount through the following steps: The first step is to calculate the initial adjustment amount by calling the preset proportional-integral-derivative (PID) controller inside the control execution module based on the deviation value of the current control cycle. The calculation method of the PID controller is as follows: multiply the deviation value of the current control cycle by the preset proportional coefficient to obtain the proportional term value; multiply the sum of the deviation values ​​of all historical control cycles before the current control cycle by the preset integral coefficient to obtain the integral term value; multiply the difference between the deviation value of the current control cycle and the deviation value of the previous control cycle by the preset derivative coefficient to obtain the derivative term value; and use the sum of the proportional term value, integral term value, and derivative term value as the initial adjustment amount. This initial adjustment amount represents the percentage of the motor speed of the feeding device to be adjusted; The second step is to obtain the historical final adjustment value of each historical control cycle within the last five control cycles, and for each historical control cycle, calculate the theoretical adjustment value of the historical control cycle according to the calculation method in the first step using the same preset proportional integral derivative controller based on the historical deviation value corresponding to the historical control cycle. Then, calculate the ratio of the historical final adjustment value to the theoretical adjustment value, and use all ratios to form a historical correction sequence. The third step is to calculate the exponentially weighted moving average of the historical correction sequence, where the exponential decay factor of the exponentially weighted moving average is recalibrated by the sum of the dynamic compensation coefficient and the preset benchmark decay factor. The fourth step is to construct a correction function. The function value of the correction function is the dynamic compensation coefficient multiplied by the negative preset smoothing factor of the natural constant e multiplied by the absolute value of the deviation ratio, where the deviation ratio is the ratio of the deviation value of the current control cycle to the preset maximum allowable deviation. The fifth step is to multiply the initial adjustment amount by the function value of the correction function to obtain the corrected final adjustment amount for the current control cycle, and then drive the motor speed of the feeding device according to the corrected final adjustment amount.

8. The weighing target value control system based on raw material information feedback according to claim 1, characterized in that, The control results fed back from the control execution module to the raw material characteristic identification module include: The steady-state deviation value after this adjustment, the actual response time of the feeding equipment during the adjustment process, and the actual correction effect after the application of the dynamic compensation coefficient are identified. The raw material characteristic identification module updates the sample database inside the preset correlation screening model based on the adjustment result, and recalculates the Pearson correlation coefficient between the measured value sequence and the deviation value sequence of each feature category in the sample database. The recalculated correlation coefficient is used as the updated correlation confidence of each feature category for comparison in the next correlation screening process.

9. The weighing target value control system based on raw material information feedback according to claim 1, characterized in that, During the process of adjusting the operating parameters of the feeding equipment, the control and execution module also monitors the current speed and current feed flow rate of the feeding equipment in real time, and compares the current speed with the preset upper speed threshold and the preset lower speed threshold, and compares the current feed flow rate with the preset upper flow rate threshold and the preset lower flow rate threshold. If any item exceeds the corresponding threshold range, the preset safety protection mode will be triggered, forcibly switching the feeding equipment to the preset constant speed operation state.

Images