A corrugated paper moisture content real-time compensation method based on multi-source data fusion

Abstract

The application discloses a corrugated base paper moisture content real-time compensation method based on multi-source data fusion, and belongs to the technical field of industrial process control, which comprises the following steps: acquiring equipment running length, glue penetration parameters, temperature and humidity, corrugated base paper tension change data and production line speed data to generate original monitoring data; performing cross-validation and cleaning on the original monitoring data to generate dynamic cleaning data; performing double-mode moisture content prediction on the dynamic cleaning data; dynamically assigning weights to the prediction results to generate dynamic compensation parameters; generating a real-time compensation signal based on the parameters; and correcting the real-time compensation signal according to the obtained compensated moisture content feedback to obtain a moisture content compensation value. The application adopts a double-mode fusion prediction technology of a physical model and a time series model, combines a closed-loop hierarchical compensation and feedback adaptive correction technology, can realize accurate and stable control of the corrugated base paper moisture content, and effectively overcomes the coupling influence of the glue penetration timeliness and temperature and humidity disturbance.

Description

Technical Field

[0001] This invention relates to the field of industrial process control technology, and in particular to a method for real-time compensation of moisture content in corrugated base paper based on multi-source data fusion. Background Technology

[0002] In the production of corrugated base paper, precise control of moisture content directly affects paper strength, flatness, and subsequent processing performance, making it a core component of the paper industry's automatic control system. Current technologies typically rely on a single sensor to monitor moisture content, combined with a fixed-parameter model for steam injection compensation.

[0003] Current methods primarily utilize data from temperature, humidity, and velocity sensors to predict moisture content changes using a linear regression model, and trigger steam nozzle operation based on a preset threshold. Some improvements incorporate equipment uptime as a correction factor to mitigate performance drift during long-term production.

[0004] Existing technologies are highly sensitive to instantaneous disturbances such as tension fluctuations, which can easily lead to data distortion; they do not fully consider the nonlinear decay characteristics of glue penetration parameters as the equipment ages; the steam injection volume control of the actuator has response lag and overshoot, and lacks closed-loop verification of the actual injection effect. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a real-time moisture content compensation method for corrugated base paper based on multi-source data fusion. It employs a dual-mode fusion prediction technique combining physical and time-series models, along with closed-loop hierarchical compensation and feedback adaptive correction techniques. This method enables precise steady-state control of the moisture content of corrugated base paper, effectively overcoming the coupling effects of glue penetration timeliness and temperature and humidity disturbances.

[0006] The above objectives can be achieved through the following approach:

[0007] A real-time moisture content compensation method for corrugated base paper based on multi-source data fusion includes: acquiring raw monitoring data such as equipment runtime, glue penetration parameters, temperature and humidity, corrugated base paper tension changes, and production line speed; performing cross-validation and cleaning on the raw monitoring data to generate dynamic cleaning data; performing dual-mode moisture content prediction on the dynamic cleaning data; dynamically assigning weights to the prediction results to generate dynamic compensation parameters; generating a real-time compensation signal based on these parameters; and correcting the real-time compensation signal based on the acquired compensated moisture content feedback to obtain the moisture content compensation value. This invention employs a dual-mode fusion prediction technology combining a physical model and a time series model, combined with closed-loop hierarchical compensation and feedback adaptive correction technology, enabling precise steady-state control of the moisture content of corrugated base paper and effectively overcoming the coupling effects of glue penetration timeliness and temperature and humidity disturbances.

[0008] Optionally, the step of cross-validating and cleaning the original monitoring data to generate dynamic cleaning data includes: if the stress change data of the corrugated base paper does not exceed the preset tension fluctuation threshold range, then the original monitoring data is re-verified to obtain the moisture content deviation coefficient; the original monitoring data is corrected according to the moisture content deviation coefficient to generate dynamic cleaning data.

[0009] Optionally, the step of predicting the moisture content of the dynamic cleaning data and generating a dual-mode prediction result includes: establishing a physical prediction sub-model based on the temperature and humidity data and the operating speed data of the corrugated base paper production line; constructing a time series prediction sub-model based on the equipment running time data and the glue penetration parameters; and coupling the output results of the physical prediction sub-model and the time series prediction sub-model to generate a dual-mode prediction result.

[0010] Optionally, the step of dynamically weighting the dual-mode prediction results to generate dynamic compensation parameters includes: calculating the first prediction error value of the physical prediction sub-model and the second prediction error value of the time series prediction sub-model to obtain the weight coefficients of the physical prediction sub-model and the time series prediction sub-model; when the corrugated paper force change data is detected to exceed the tension fluctuation threshold range, the weight coefficient of the physical prediction sub-model is increased to obtain the physical dominant weight allocation parameter; when the glue penetration attenuation rate corresponding to the equipment running time data exceeds the timeliness threshold of the glue penetration parameter, the weight coefficient of the time series prediction sub-model is increased to obtain the time series dominant weight allocation parameter; and the physical dominant weight allocation parameter and the time series dominant weight allocation parameter are weighted linearly superimposed to generate dynamic compensation parameters.

[0011] Optionally, generating a real-time compensation signal based on the dynamic compensation parameters includes: obtaining a voltage control signal based on the dynamic compensation parameters; adjusting the opening frequency and injection volume of the steam nozzle according to the voltage control signal to obtain actual injection volume data of the steam nozzle; and performing physical feature feedback correction on the actual injection volume data of the steam nozzle to generate a real-time compensation signal.

[0012] Optionally, the step of obtaining the compensated moisture content data and performing feedback correction on the real-time compensation signal to obtain the moisture content compensation value includes: calculating the cumulative deviation of the moisture content target value based on the compensated moisture content data; obtaining the deviation perception correction coefficient based on the cumulative deviation of the moisture content target value; adjusting the hidden layer node connection weights of the time series prediction sub-model based on the deviation perception correction coefficient to obtain an adaptive time series weight matrix; and performing feedback correction on the real-time compensation signal according to the adaptive time series weight matrix to obtain the moisture content compensation value.

[0013] Optionally, the step of feedback correction of the real-time compensation signal based on the adaptive time-series weight matrix to obtain the water content compensation value includes: reconstructing the time-series prediction model and iterating the dynamic weights of the adaptive time-series weight matrix to obtain optimized compensation parameters; and adjusting the feedback gain and dynamically compensating the real-time compensation signal based on the optimized compensation parameters to obtain the water content compensation value.

[0014] Optionally, the step of correcting the original monitoring data based on the moisture content deviation coefficient to generate dynamic cleaning data includes: obtaining multi-channel correction gain parameters based on the moisture content deviation coefficient; and performing multi-factor coupling correction on the temperature and humidity data, corrugated base paper production line operating speed data, and glue penetration parameters based on the multi-channel correction gain parameters to generate dynamic cleaning data.

[0015] Optionally, obtaining the voltage control signal based on the dynamic compensation parameters includes: performing a timing correction on the dynamic compensation parameters to obtain compensation correction parameters; and performing a nonlinear adaptive mapping transformation on the compensation correction parameters to obtain the voltage control signal.

[0016] Based on the same inventive concept, this invention also provides a real-time moisture content compensation system for corrugated base paper based on multi-source data fusion. The system includes: a multi-parameter acquisition module for acquiring equipment runtime data, glue penetration parameters, temperature and humidity data, corrugated base paper force change data, and corrugated base paper production line operating speed data, generating raw monitoring data; a dynamic cleaning module for performing cross-validation cleaning processing on the raw monitoring data, generating dynamic cleaning data; a dual-mode prediction module for predicting the moisture content of the dynamic cleaning data, generating dual-mode prediction results; a weight allocation module for dynamically allocating weights to the dual-mode prediction results, generating dynamic compensation parameters; a signal generation module for generating a real-time compensation signal based on the dynamic compensation parameters; and a feedback correction module for acquiring the compensated moisture content data, performing feedback correction on the real-time compensation signal, and obtaining a moisture content compensation value.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] By employing a multi-source data cross-validation and cleaning mechanism, sensor measurement noise and abnormal fluctuation interference are effectively eliminated, significantly improving the accuracy and consistency of the original monitoring data and providing a reliable data foundation for moisture content control.

[0019] By combining a physical model with real-time correlation of environmental parameters and a time series model with equipment aging trends, a dual-mode prediction mechanism is established to comprehensively capture short-term dynamic disturbances and long-term degradation factors, thereby enhancing the robustness and adaptability of moisture content prediction.

[0020] An adaptive weight allocation strategy based on prediction error and working condition is adopted to dynamically optimize the contribution ratio of the physical model and the time series model, ensuring the rationality of compensation parameter decisions under complex working conditions such as sudden tension changes or glue aging.

[0021] By using closed-loop control of steam injection quantity and graded threshold compensation, the actuator deviation is accurately corrected, achieving rapid response and stable execution of moisture content compensation. At the same time, historical deviation feedback is used to correct the weight of the prediction model, continuously improving the long-term control accuracy of the system.

[0022] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a real-time moisture content compensation method for corrugated base paper based on multi-source data fusion, according to an embodiment of the present invention.

[0025] Figure 2 This is a diagram showing the tension variation and threshold range of an embodiment of the present invention.

[0026] Figure 3 This is a correction diagram of the moisture content deviation coefficient according to an embodiment of the present invention.

[0027] Figure 4 This is a diagram of the physical prediction sub-model of an embodiment of the present invention.

[0028] Figure 5 This is a diagram showing the generation of dynamic compensation parameters according to an embodiment of the present invention.

[0029] Figure 6 This is an adaptive temporal weight matrix diagram according to an embodiment of the present invention.

[0030] Figure 7 This is a schematic diagram of a real-time moisture content compensation system for corrugated base paper based on multi-source data fusion, according to an embodiment of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Reference Figure 1 One embodiment of the present invention proposes a real-time compensation method for the moisture content of corrugated base paper based on multi-source data fusion. It adopts a dual-mode fusion prediction technology of physical model and time series model, combined with closed-loop hierarchical compensation and feedback adaptive correction technology, which can achieve accurate steady-state control of the moisture content of corrugated base paper and effectively overcome the coupling effect of glue penetration time and temperature and humidity disturbance.

[0033] The method described in this embodiment specifically includes:

[0034] Acquire equipment runtime data, glue penetration parameters, temperature and humidity data, corrugated base paper force change data, and corrugated base paper production line operating speed data to generate raw monitoring data;

[0035] The original monitoring data is cross-validated and cleaned to generate dynamic cleaned data;

[0036] Moisture content prediction is performed on the dynamic cleaning data to generate dual-mode prediction results;

[0037] Dynamic weight allocation is performed on the dual-mode prediction results to generate dynamic compensation parameters;

[0038] Based on the aforementioned dynamic compensation parameters, a real-time compensation signal is generated;

[0039] Based on the compensated moisture content data, the real-time compensation signal is corrected by feedback to obtain the moisture content compensation value.

[0040] Specifically, a programmable logic controller (PLC) is deployed on-site to collect real-time data on the equipment's runtime via its built-in timer. This data records the cumulative continuous operating time of the corrugated cardboard production line since its startup. Simultaneously, an online adhesive viscometer monitors adhesive penetration parameters. This viscometer measures the flow resistance of the adhesive through a capillary tube and converts it into a standardized adhesive penetration coefficient based on Newton's law of fluid dynamics. Ambient temperature and humidity data are collected by high-precision temperature and humidity sensors installed in the paper feeding area of ​​the production line. The temperature sensor measures air temperature based on thermocouple principles, while the humidity sensor uses a capacitive sensing element to obtain relative humidity values. Corrugated base paper tension changes are acquired in real-time by a cantilever tension sensor installed at the floating roller bearing seat along the paper web transport path. This sensor detects the mechanical changes in paper web tension using a strain gauge bridge. The corrugated base paper production line's operating speed data is collected by a rotary encoder installed at the end of the guide roller bearing. This encoder calculates the actual linear speed by multiplying the number of roller rotations per unit time by a pre-calibrated roller circumference value. The above five types of data are transmitted to the central controller via industrial Ethernet. After data alignment according to millisecond-level timestamps, they are stored in a designated data table in the real-time database in a structured format. The table fields include five dimensions: equipment runtime, glue penetration coefficient, ambient temperature, ambient humidity, and production line speed, thus forming a raw monitoring data set with time series characteristics.

[0041] Optionally, the step of performing cross-validation cleaning on the original monitoring data to generate dynamically cleaned data includes:

[0042] If the stress change data of the corrugated base paper does not exceed the preset tension fluctuation threshold range, the original monitoring data is re-checked to obtain the moisture content deviation coefficient.

[0043] The original monitoring data is corrected based on the moisture content deviation coefficient to generate dynamic cleaning data.

[0044] Specifically, the tension change and threshold range are as follows: Figure 2 As shown, firstly, tension sensor data is used to collect real-time data on the stress changes of the corrugated base paper. This data characterizes the instantaneous mechanical fluctuations during paper web operation. The acquired tension change data is compared with a tension fluctuation threshold range, which needs to be determined through pre-experimental calibration based on the basis weight parameters of the corrugated base paper and the mechanical characteristics of the production line. If the tension does not exceed the threshold range, a verification and testing process is initiated: the compensated moisture content data in the original monitoring database is retrieved simultaneously, and the actual moisture content of the corrugated base paper is obtained in real-time using an infrared online moisture meter. The moisture content deviation coefficient is then calculated. :

[0045] ,

[0046] in This represents the actual moisture content measured online by an infrared moisture meter. This represents the compensated moisture content, read from the original monitoring database. Temperature and humidity correction gain parameters, production line operating speed correction gain parameters, and adhesive penetration correction gain parameters are generated. Each parameter is generated based on the incremental product of the moisture content deviation coefficient and the historical coupling coefficient. The moisture content deviation coefficient correction is as follows: Figure 3 As shown in the figure. Subsequently, the original multi-source monitoring data are collaboratively adjusted: the environmental temperature and humidity data are multiplied by the temperature and humidity correction gain parameter, the production line operating speed data is multiplied by the production line operating speed correction gain parameter, and the glue penetration parameter is multiplied by the glue penetration correction gain parameter to form a dynamic cleaning dataset for subsequent processing modules to call.

[0047] For example, during the operation of a production line, the tension sensor reading of 0.98 Newtons is lower than the lower threshold of 1.5 Newtons, triggering a verification check. An infrared moisture meter measures the actual moisture content of the paper feeding area to be 6.1%, while the compensated moisture content record in the original database is 6.6%, resulting in a deviation coefficient of 0.5%. Based on the regression calibration relationship of historical data, temperature and humidity correction gains of 1.025, speed correction gains of 1.055, and glue penetration correction gains of 1.040 are generated. The original monitoring data is updated: temperature 28°C to 28.7°C, humidity 70%RH to 71.75%RH, production line speed 125 m / min to 131.875 m / min, and glue penetration parameter 22 to 22.880. This calibrated dynamic cleaning data package is then transmitted to the subsequent prediction module.

[0048] Optionally, the step of predicting the moisture content of the dynamic cleaning data and generating a dual-mode prediction result includes:

[0049] Based on the temperature and humidity data and the operating speed data of the corrugated base paper production line, a physical prediction sub-model is established.

[0050] Based on the equipment runtime data and the glue penetration parameters, a time series prediction sub-model is constructed.

[0051] The outputs of the physical prediction sub-model and the time series prediction sub-model are coupled and calculated to generate a dual-model prediction result.

[0052] Specifically, a physical prediction sub-model is first established based on temperature and humidity data and corrugated base paper production line operating speed data. The physical prediction sub-model is as follows: Figure 4 As shown, this model calculates predicted values ​​based on the principle of thermodynamic diffusion and the heat balance equation relating paper moisture absorption to ambient temperature and humidity. The physical prediction sub-model expression is as follows:

[0053] ,

[0054] The temperature difference between the ambient temperature and the set temperature is obtained in real time via a temperature sensor. The ambient humidity is acquired in real time via a humidity sensor. The operating speed of the corrugated base paper production line is collected via an encoder. and This is an empirical coefficient, obtained through regression calibration using historical temperature, humidity, and speed data.

[0055] The physical prediction sub-model is established by following these steps: First, data acquisition and preprocessing: Under steady-state operating conditions on the production line, continuously collect 72 hours of temperature and humidity data, production line speed data, and synchronous infrared moisture content data; perform moving average filtering (window width 5 minutes) on the raw data to eliminate instantaneous noise; construct the dataset: input features are... ( =Ambient temperature - 25℃ (Base temperature) =Relative humidity, =Production line speed), the output label is the measured moisture content. ;

[0056] Then, empirical coefficient regression calibration is performed: the model is solved using regularized multiple linear regression (RidgeRegression) as follows:

[0057] ,

[0058] in The temperature and humidity are coupled influencing factors. For production line speed sensitivity coefficient, loss function :

[0059] ,

[0060] This is the regularization coefficient, set to 0.01 to prevent overfitting. The optimization algorithm is: least squares closed-form solution:

[0061] ,

[0062] in For an N×3 characteristic matrix: column 1 = ,Column 2= Column 3 = 1; for ×1 moisture content matrix; It is an identity matrix; the constructed test set has a value greater than 0.85 and a maximum absolute error less than 0.5%. Finally, the coefficients are solidified and deployed: the calibrated... , The data is written to the PLC's read-only memory and recalibrated every 30 days based on new data (trigger condition: cumulative prediction error > 10%). Simultaneously, a time-series prediction sub-model is constructed based on equipment runtime data and glue penetration parameters. The network topology of the time-series prediction sub-model includes: an input layer (2D vectors: equipment runtime, glue penetration parameters); an LSTM layer (single layer with 16 neurons, activation function, forget gate bias initial value); a Dropout layer (dropout rate 0.2, to prevent overfitting); a fully connected layer (8 neurons, ReLU activation); and an output layer (1 neuron, predicted moisture content value, linear activation). The data source is time-series data from 200 hours of continuous equipment operation (sampling interval 1 minute). Data augmentation is performed by generating sequence samples through window sliding (window length 60, step size 1). Maximum and minimum values ​​are normalized for equipment runtime, and Z-score standardization is applied to the glue penetration parameters. This model uses a Long Short-Term Memory (LSTM) network to capture the time-series trends of equipment aging and glue penetration decay. The inputs are equipment runtime data and glue penetration parameters. The time series prediction sub-model is calculated through forward propagation of the neural network. The LSTM weights are optimized offline using the initial training data. Finally, the outputs of the two sub-models are... and Perform coupled calculations to generate dual-mode prediction results The coupling uses an initial equal-weight linear superposition, expressed as:

[0063] ,

[0064] This is the weighting coefficient, ensuring that the sum of the weights is 1.

[0065] For example, on a corrugated base paper production line, dynamic cleaning data includes temperature and humidity data of 24.5℃ and 63%RH, production line operating speed of 115 meters per minute, equipment operating time of 385 minutes, and glue penetration parameter 17. When establishing the physical prediction sub-model, pre-calibrated empirical coefficients are used. =0.02 and =0.5, calculate the temperature difference. Substituting into the formula, we get When constructing the time series prediction sub-model, the device runtime of 385 minutes and the glue parameter of 17 are input into the pre-trained LSTM network, and its output is: =6.7%. Equal-weighted stacking is used to generate dual-mode prediction results during coupled calculations. This result serves as the input for subsequent dynamic weight allocation.

[0066] Optionally, the step of dynamically weighting the dual-mode prediction results to generate dynamic compensation parameters includes:

[0067] Calculate the first prediction error value of the physical prediction sub-model and the second prediction error value of the time series prediction sub-model to obtain the weight coefficients of the physical prediction sub-model and the time series prediction sub-model;

[0068] When the detected force change data of the corrugated base paper exceeds the tension fluctuation threshold range, the weight coefficient of the physical prediction sub-model is increased to obtain the physical dominant weight allocation parameter;

[0069] When the glue penetration decay rate corresponding to the device runtime data exceeds the timeliness threshold of the glue penetration parameter, the weight coefficient of the time series prediction sub-model is increased to obtain the time series dominant weight allocation parameter.

[0070] The physical dominant weight allocation parameter and the temporal dominant weight allocation parameter are weighted and linearly superimposed to generate dynamic compensation parameters.

[0071] Specifically, firstly, the first prediction error value of the physical prediction sub-model and the second prediction error value of the time series prediction sub-model are calculated. The prediction error value is defined as the absolute difference between the model output value and the actual moisture content detection value. Initial weighting coefficients are determined based on the ratio of the recent mean errors of the two models. Let the weighting coefficient of the physical prediction sub-model be... The weighting coefficients of the time series prediction sub-model are Initially satisfied When the detected force variation data of the corrugated base paper exceeds the tension fluctuation threshold range, it indicates increased mechanical vibration interference. In this case, the weight coefficient of the physical prediction sub-model is increased through a weighting factor. Adjust the weighting coefficients of the physics prediction sub-model; the weighting factor formula is as follows:

[0072] ,

[0073] in The difference between the tension exceeding the threshold value. As the tension reference constant, calibrated through production line mechanical testing, the adjusted weight coefficients of the physical prediction sub-model are:

[0074] ,

[0075] Simultaneously reduce the weight coefficients of the time series prediction sub-model:

[0076] ,

[0077] When the glue penetration degradation rate calculated from the equipment runtime variance data exceeds the timeliness threshold, it indicates that the glue performance has degraded, as indicated by the percentage of degradation exceeding the limit. Calculate the exponential gain The formula is:

[0078] ,

[0079] Improve the time series prediction sub-model to obtain the time-series dominant weight allocation parameters:

[0080] ,

[0081] The physical dominant weight allocation parameters are:

[0082] .

[0083] Finally, the two dominant state parameters are weighted and linearly superimposed. When the tension change data exceeds the tension fluctuation threshold range, a dynamic compensation parameter is output. The dynamic compensation parameter is generated as follows: Figure 5 As shown:

[0084] ,

[0085] When the glue penetration attenuation rate exceeds the timeliness threshold of the glue penetration parameter, a dynamic compensation parameter is output:

[0086] .

[0087] For example, in a corrugated base paper production line, the physical prediction sub-model outputs... The output of the time series prediction sub-model Initial weight setting When the tension sensor detects a value of 2.3 Newtons, exceeding the threshold range of 1.5-2.0 Newtons, the calculation... Take the calibration constant Weighting factors Then the physical dominant weight If the value exceeds 1, the system will automatically truncate it to 1. The time-series weights are updated synchronously. Dynamic compensation parameter output After three hours, the variance of the equipment's operating time indicates the glue penetration decay rate. Exceeding the 80% threshold, the timing weight boosting mechanism is triggered to calculate the gain. Order-dominant weight allocation parameters The truncation is set to 0.92, and the physics-dominant weight allocation parameter is adjusted to 0.08. The output is then... .

[0088] Optionally, generating a real-time compensation signal based on the dynamic compensation parameters includes:

[0089] Based on the aforementioned dynamic compensation parameters, a voltage control signal is obtained;

[0090] Based on the voltage control signal, the opening frequency and injection volume of the steam nozzle are adjusted to obtain the actual injection volume data of the steam nozzle.

[0091] The actual injection volume data of the steam nozzle is corrected by physical feature feedback to generate a real-time compensation signal.

[0092] Specifically, the dynamic compensation parameter, which is the percentage of moisture content, is first processed by the equipment operation attenuation timing correction unit. The attenuation factor calculated from the equipment operation duration variance data is defined as follows: It is determined by the ratio of the equipment's continuous operating time to its lifespan coefficient, and the correction formula is as follows:

[0093] ,

[0094] in To dynamically compensate and correct parameters, For dynamic compensation parameters, The data was obtained by fitting the historical operating database of the equipment. Next, a nonlinear adaptive mapping transformation was performed to... Converted to voltage control signal :

[0095] ,

[0096] in the formula This is the voltage value. As the baseline moisture content, This is the voltage conversion factor, obtained through equipment calibration tests. According to... Control the steam nozzle drive circuit and adjust the opening frequency. and injection volume The two satisfy a linear relationship:

[0097]

[0098] ,

[0099] in , This represents the nozzle characteristic constant. Finally, the actual opening degree data of the steam nozzle is collected. With steam pressure data The actual injection volume is calculated using fluid dynamics equations. :

[0100] ,

[0101] This is the flow coefficient, obtained from the nozzle's factory specifications. With command injection volume Closed-loop comparison, when the deviation exceeds the threshold The system generates steam pressure compensation increments in real time, and finally outputs a real-time compensation signal that includes frequency, opening degree and pressure compensation.

[0102] For example, in a corrugated base paper production line, dynamic compensation parameters The attenuation factor is calculated based on the variance of the current device's operating time. After correction Take the reference moisture content. Calibration coefficient Calculate voltage signal Based on nozzle characteristic constants Adjust the on / off frequency Command injection volume In actual operation, the pressure sensor detects... The opening sensor provides feedback on the actual opening degree. Calculate the actual injection volume The deviation from the command value reached Exceeding the threshold This generates a pressure compensation incremental signal, and finally outputs a composite compensation command to adjust the nozzle opening to... .

[0103] Optionally, the step of obtaining the compensated moisture content data and performing feedback correction on the real-time compensation signal to obtain the moisture content compensation value includes:

[0104] Based on the compensated moisture content data, the cumulative deviation of the moisture content target value is calculated;

[0105] Based on the cumulative deviation of the target moisture content, a deviation perception correction coefficient is obtained;

[0106] Based on the aforementioned bias-aware correction coefficient, the hidden layer node connection weights of the time series prediction sub-model are adjusted to obtain an adaptive time series weight matrix.

[0107] Based on the adaptive time-series weight matrix, the real-time compensation signal is corrected by feedback to obtain the moisture content compensation value.

[0108] Specifically, adaptive time-series weight matrix adjustment is as follows: Figure 6 As shown, the cumulative deviation of the target moisture content value is first obtained. This deviation is obtained by subtracting the compensated moisture content setpoint sequence from the actual measured value sequence point by point and then summing the absolute values:

[0109] ,

[0110] In the formula This represents the cumulative percentage of deviation. For a moment The target value for moisture content, For a moment The actual moisture content measured value, This represents the number of consecutive sampling points. Based on... Calculate the bias perception correction coefficient :

[0111] ,

[0112] The proportionality coefficient was obtained through regression analysis using a historical deviation database. Input the hidden layer weight adjustment module of the time series prediction sub-model and define three fixed time periods. , , ( =1h、 =2h、 =4h, calculated based on production line operating characteristics), calculate the time period within each time period. Accumulated value:

[0113] ,

[0114] For time period The cumulative intensity, take , , The maximum value determines the gradient factor for weight adjustment. The model is then initialized with the weight gradient sign matrix saved during training. ,in This represents the gradient of the loss function with respect to the weights (obtained through offline backpropagation). Elements are taken as ±1 or 0. Weights are updated using gradient-sensitive direction compensation:

[0115] ,

[0116] The output is the adaptive time-series weight matrix. This is the original weight matrix. To set the learning rate, Adjust the gradient factor for the determined maximum weight. After loading the time series prediction sub-model, the dynamic weight allocation is re-executed to generate new compensation parameters, and finally the moisture content compensation value is output through the steam nozzle control link.

[0117] For example, in the corrugated base paper production line, the target moisture content sequence after compensation is [6.5%, 6.5%, 6.5%], and the actual measured values ​​at three consecutive sampling points are [6.7%, 6.6%, 6.8%]. The cumulative deviation is calculated. =|6.7%-6.5%|+|6.6%-6.5%|+|6.8%-6.5%|=0.6%. Take the proportionality coefficient. =0.02 0.012. Piecewise integration setting =0~1h( =0.008), =1~3h ( =0.010), =3~7h ( =0.012), determine the weight adjustment gradient factor. The value is 0.012. Load the original weight matrix. and symbol matrix ,set up =0.01 Update the weight matrix. After loading the new weights, the time series prediction sub-model runs on the input device for 400 minutes with glue parameter 18, and the output moisture content is corrected from 7.0% to 6.65%. Combined with the physical sub-model output of 6.60%, a compensation parameter of 6.63% is generated through dynamic weight allocation, and the actual moisture content stably converges to the target range.

[0118] Optionally, the step of feedback correction of the real-time compensation signal based on the adaptive time-series weight matrix to obtain the moisture content compensation value includes:

[0119] The adaptive time-series weight matrix is ​​reconstructed using a time-series prediction model and subjected to dynamic weight iteration to obtain optimized compensation parameters.

[0120] Based on the optimized compensation parameters, the real-time compensation signal feedback gain is adjusted and dynamically compensated to obtain the moisture content compensation value.

[0121] Specifically, firstly, an adaptive time-series weight matrix is ​​obtained and loaded into the time-series prediction sub-model to replace the initial weight matrix. The model prediction is then re-executed based on the current equipment runtime data and glue penetration parameters to obtain optimized prediction values. These optimized prediction values ​​are input into the dynamic weight allocation module to recalculate the real-time prediction error values ​​of the physical prediction sub-model and the time-series prediction sub-model. The prediction error value is the absolute difference between the model output value and the moisture content measured by the infrared online moisture meter. Weight coefficients are adjusted based on equipment operating status fluctuation data: when the corrugated paper tension change data exceeds the preset tension fluctuation threshold range, the weight coefficient of the physical prediction sub-model is increased by a weight factor; when the glue penetration attenuation rate exceeds the timeliness threshold, the weight coefficient of the time-series prediction sub-model is increased by an exponential gain to generate optimized compensation parameters. These optimized compensation parameters are then input into the voltage conversion stage to perform attenuation correction and nonlinear mapping. Based on the nozzle characteristic equation, the theoretical injection volume command is calculated using the following formula:

[0122] ,

[0123] in This is the commanded injection volume; This refers to the opening frequency of the steam nozzle; These are nozzle-specific constants, obtained through nozzle factory calibration. Simultaneous acquisition of actual nozzle opening data and steam pressure data is used to calculate the actual injection volume based on fluid dynamics formulas. :

[0124] ,

[0125] For flow coefficient, Indicates the steam pressure difference. This represents the density of steam, when... When the threshold is exceeded, an opening compensation increment is generated. Finally, the steam pressure control component in the original real-time compensation signal is superimposed with... A moisture content compensation value is formed.

[0126] For example, an adaptive time-series weight matrix is ​​loaded into the time-series prediction sub-model in a corrugated base paper production line, with inputs of 385 minutes of equipment operation and glue parameter 21. The model's original output moisture content of 7.30% is optimized to 7.03%. The physical prediction sub-model outputs 7.28%. The current glue penetration decay rate of 45% is detected (below the timeliness threshold of 80%) and the tension is 1.8 Newtons (within the threshold). The dynamic weight allocation maintains an equal weight setting, generating an optimized compensation parameter of 7.15%. The nozzle frequency command is converted to 18Hz, corresponding to the theoretical spray volume. =152L / min. Actual measured steam pressure 0.85MPa, opening degree 88%, calculated... =149L / min. A deviation of 3L / min exceeds the threshold, generating an opening compensation increment. =+2%, after adjusting the pressure control component, the output moisture content compensation value controls the nozzle opening to increase to 90%, and finally achieves the moisture content in the paper feeding area to be stable within the target value range of ±0.3%.

[0127] Optionally, the step of correcting the original monitoring data based on the moisture content deviation coefficient to generate dynamic cleaning data includes:

[0128] Based on the moisture content deviation coefficient, the multi-channel correction gain parameters are obtained;

[0129] Based on the multi-channel correction gain parameters, the temperature and humidity data, the corrugated base paper production line operating speed data, and the glue penetration parameters are subjected to multi-factor coupling correction to generate dynamic cleaning data.

[0130] Specifically, the moisture content deviation coefficient is first obtained. It is calculated from the difference between the actual moisture content measured and the compensated moisture content data. Based on Generate multi-channel corrected gain parameters:

[0131] ,

[0132] In the formula To correct the gain parameters for temperature and humidity, The temperature and humidity coupling coefficient is used; similarly, the speed correction gain parameter for the corrugated base paper production line is calculated. :

[0133] ,

[0134] Glue penetration correction gain parameter :

[0135]

[0136] The coupling coefficient for production line operating speed. The adhesive penetration coupling coefficient is obtained through speed encoder and adhesive parameter history drift regression, thereby achieving multi-factor coupling correction: environmental temperature and humidity data multiplied by... Production line operating speed data multiplied by , glue penetration parameter multiplied by This forms a dynamic cleaning data set.

[0137] For example, the tension detection value of the corrugated base paper production line is 0.98 Newtons (not exceeding the threshold range of 1.5-2.0 Newtons), the infrared moisture meter measured the moisture content at 6.1%, and the compensated moisture content in the original monitoring data was 6.6%. Calculate the moisture content deviation coefficient. =|6.1%-6.6%|=0.5%. Take the historical drift regression coefficient. Calculate the gain parameters: , , Corrected original monitoring data: Ambient temperature 28°C and humidity 70%RH updated to... The production line operating speed has been updated from 125 meters per minute. Meters per minute, glue penetration parameter 22 updated to The final output is dynamically cleaned data for use by subsequent predictive models.

[0138] Optionally, obtaining the voltage control signal based on the dynamic compensation parameters includes:

[0139] The dynamic compensation parameters are corrected by time-series correction of the equipment operation attenuation coefficient to obtain the compensation correction parameters;

[0140] A nonlinear adaptive mapping transformation is performed on the compensation correction parameters to obtain the voltage control signal.

[0141] Specifically, the first step is to obtain the dynamic compensation parameters. This parameter represents the percentage value for moisture content compensation. Time-series correction is performed on the equipment operation attenuation coefficient, and the equipment operation attenuation factor is defined as follows: (Obtained by fitting the equipment's historical operating database, representing the degree of equipment performance degradation), the correction formula is:

[0142] ,

[0143] In the formula To compensate for the correction parameters. Secondly, for Perform a nonlinear adaptive mapping transformation, defining the reference moisture content as... The voltage conversion factor is (Obtained through equipment calibration tests), the conversion formula is:

[0144] ,

[0145] In the formula This is a voltage control signal.

[0146] For example, in a corrugated base paper production line, the output dynamic compensation parameters The current equipment runtime data analysis yields the equipment runtime degradation factor. Calculate the compensation correction parameters. Reference moisture content Calibrate voltage conversion factor Generate voltage control signal Volt. This signal is input to the steam nozzle drive circuit, adjusting the opening frequency to 15Hz to optimize the nozzle injection volume and ensure that subsequent moisture content compensation is controlled within the process target range.

[0147] Based on the same inventive concept, such as Figure 7 As shown, the present invention also provides a real-time moisture content compensation system for corrugated base paper based on multi-source data fusion, the system comprising:

[0148] The multi-parameter acquisition module is used to acquire equipment runtime data, glue penetration parameters, temperature and humidity data, corrugated base paper force change data, and corrugated base paper production line operating speed data, and generate raw monitoring data.

[0149] The dynamic cleaning module is used to perform cross-validation cleaning processing on the original monitoring data to generate dynamic cleaned data.

[0150] The dual-mode prediction module is used to predict the moisture content of the dynamic cleaning data and generate dual-mode prediction results.

[0151] The weight allocation module is used to dynamically allocate weights to the dual-mode prediction results and generate dynamic compensation parameters.

[0152] The signal generation module is used to generate a real-time compensation signal based on the dynamic compensation parameters;

[0153] The feedback correction module is used to acquire the compensated moisture content data, perform feedback correction on the real-time compensation signal, and obtain the moisture content compensation value.

[0154] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0155] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A corrugated base paper moisture content real-time compensation method based on multi-source data fusion, characterized in that, The method comprises: obtaining equipment runtime data, glue penetration parameters, temperature and humidity data, corrugated base paper tension change data, and corrugated base paper production line running speed data to generate original monitoring data; cross-validation cleaning processing is performed on the original monitoring data to generate dynamic cleaning data; moisture content prediction is performed on the dynamic cleaning data to generate a dual-mode prediction result, which includes: based on the temperature and humidity data and the corrugated base paper production line running speed data, a physical prediction sub-model is established; based on the equipment runtime data and the glue penetration parameters, a time series prediction sub-model is constructed; the output results of the physical prediction sub-model and the time series prediction sub-model are coupled to generate a dual-mode prediction result; dynamic weight allocation is performed on the dual-mode prediction result to generate a dynamic compensation parameter, which includes: calculating a first prediction error value of the physical prediction sub-model and a second prediction error value of the time series prediction sub-model to obtain the weight coefficients of the physical prediction sub-model and the time series prediction sub-model; when it is detected that the corrugated base paper tension change data exceeds the tension fluctuation threshold range, the weight coefficient of the physical prediction sub-model is increased to obtain a physical dominant weight allocation parameter; when the equipment runtime data corresponds to a glue penetration decay rate that exceeds the timeliness threshold of the glue penetration parameters, the weight coefficient of the time series prediction sub-model is increased to obtain a time series dominant weight allocation parameter; the physical dominant weight allocation parameter and the time series dominant weight allocation parameter are subjected to weighted linear superposition calculation to generate a dynamic compensation parameter; based on the dynamic compensation parameter, a real-time compensation signal is generated, which includes: based on the dynamic compensation parameter, a voltage control signal is obtained; according to the voltage control signal, the opening frequency and the injection amount of the steam nozzle are adjusted to obtain steam nozzle actual injection amount data; the steam nozzle actual injection amount data is subjected to physical characteristic feedback correction to generate a real-time compensation signal; obtaining compensation moisture content data, feedback correcting the real-time compensation signal to obtain a moisture content compensation value.

2. The corrugated base paper moisture content real-time compensation method based on multi-source data fusion according to claim 1, characterized in that, The cross-validation cleaning processing of the original monitoring data to generate dynamic cleaning data comprises: if the corrugated base paper tension change data does not exceed the preset tension fluctuation threshold range, the original monitoring data is rechecked and detected to obtain a moisture content deviation coefficient; the original monitoring data is corrected based on the moisture content deviation coefficient to generate dynamic cleaning data.

3. The corrugated base paper moisture content real-time compensation method based on multi-source data fusion according to claim 1, characterized in that, The compensation moisture content data is obtained, the real-time compensation signal is feedback corrected, and a moisture content compensation value is obtained, which comprises: based on the compensation moisture content data, a moisture content target value cumulative deviation amount is calculated; based on the moisture content target value cumulative deviation amount, a deviation awareness correction coefficient is obtained; based on the deviation awareness correction coefficient, the hidden layer node connection weight of the time series prediction sub-model is adjusted to obtain an adaptive time series weight matrix; based on the adaptive time series weight matrix, the real-time compensation signal is feedback corrected to obtain a moisture content compensation value.

4. The corrugated base paper moisture content real-time compensation method based on multi-source data fusion according to claim 3, characterized in that, The feedback correction of the real-time compensation signal according to the adaptive time sequence weight matrix includes: The adaptive time sequence weight matrix is reconstructed and iterated by a time sequence prediction model to obtain optimized compensation parameters; Based on the optimized compensation parameters, the real-time compensation signal feedback gain is adjusted and dynamically compensated to obtain the water cut compensation value.

5. The corrugated base paper moisture content real-time compensation method based on multi-source data fusion according to claim 2, characterized in that, The original monitoring data is corrected according to the water cut deviation coefficient to generate dynamic cleaning data, which includes: Based on the water cut deviation coefficient, a multi-channel correction gain parameter is obtained; Based on the multi-channel correction gain parameter, the temperature and humidity data, the corrugated paper production line running speed data and the glue penetration parameter are coupled and corrected to generate dynamic cleaning data.

6. The corrugated base paper moisture content real-time compensation method based on multi-source data fusion according to claim 1, characterized in that, The voltage control signal is obtained based on the dynamic compensation parameter, which includes: The dynamic compensation parameter is time-corrected by a device running attenuation coefficient to obtain a compensation correction parameter; The compensation correction parameter is executed by a non-linear adaptive mapping conversion to obtain a voltage control signal.

7. A corrugated paper moisture content real-time compensation system based on multi-source data fusion, applied to a corrugated paper moisture content real-time compensation method based on multi-source data fusion according to any one of claims 1-6, characterized in that, The system includes: A multi-parameter acquisition module for acquiring device running time data, glue penetration parameters, temperature and humidity data, corrugated paper tension change data and corrugated paper production line running speed data to generate original monitoring data; A dynamic cleaning module for cross-validation cleaning of the original monitoring data to generate dynamic cleaning data; A dual-mode prediction module for water cut prediction of the dynamic cleaning data to generate a dual-mode prediction result, which includes: establishing a physical prediction sub-model based on the temperature and humidity data and the corrugated paper production line running speed data; constructing a time series prediction sub-model based on the device running time data and the glue penetration parameter; coupling and calculating the output results of the physical prediction sub-model and the time series prediction sub-model to generate a dual-mode prediction result; A weight distribution module for dynamic weight distribution of the dual-mode prediction result to generate a dynamic compensation parameter, which includes: calculating the first prediction error value of the physical prediction sub-model and the second prediction error value of the time series prediction sub-model to obtain the weight coefficients of the physical prediction sub-model and the time series prediction sub-model; when the corrugated paper tension change data exceeds the tension fluctuation threshold range, the weight coefficient of the physical prediction sub-model is increased to obtain a physical dominant weight distribution parameter; when the glue penetration attenuation rate corresponding to the device running time data exceeds the timeliness threshold of the glue penetration parameter, the weight coefficient of the time series prediction sub-model is increased to obtain a time sequence dominant weight distribution parameter; the physical dominant weight distribution parameter and the time sequence dominant weight distribution parameter are weighted and linearly superimposed to generate a dynamic compensation parameter; The signal generation module is configured to generate a real-time compensation signal based on the dynamic compensation parameter, wherein the signal generation module comprises: a voltage control signal generation unit configured to obtain a voltage control signal based on the dynamic compensation parameter; a steam nozzle adjustment unit configured to adjust an opening frequency and a spraying amount of the steam nozzle according to the voltage control signal to obtain actual spraying amount data of the steam nozzle; and a physical characteristic feedback correction unit configured to perform physical characteristic feedback correction on the actual spraying amount data of the steam nozzle to generate the real-time compensation signal. The feedback correction module is configured to obtain the compensated water content data, perform feedback correction on the real-time compensation signal to obtain a water content compensation value.

Images