Prediction and adjustment method of raw material moisture content for particleboard production

Abstract

The present invention relates to the field of intelligent control technology, specifically to a method for predicting and adjusting the moisture content of raw materials for particleboard production. In the present invention, by obtaining the disturbance parameter set and moisture content detection sequence record of the raw material drying section in the particleboard production process, combining the prediction sequence with the disturbance feature sequence to perform symbol matching, amplitude sorting, and perform disturbance direction and impact intensity marking, the raw material moisture content prediction not only has the dynamic trend judgment capability, but also can trace the source of the disturbance, adjust the amplitude of the prediction sequence through the disturbance direction and amplitude mark, realize the prediction correction of the disturbance response and the detection of trend reversal anomalies, and at the same time introduce local residual analysis to construct a trend residual sorting table, and perform weight sorting of the correction results with the abnormal inflection point as the core to guide the scaling and direction adjustment of the moisture content correction trend, and finally generate adjustment instructions for hot air temperature, heating source power and drying channel humidity through the degree of deviation between the prediction result and the target set value.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a method for predicting and regulating the moisture content of raw materials used in particleboard production. Background Art

[0002] The field of intelligent control technology involves precise parameter control and real-time adjustment of various industrial processes to improve the automation level and production efficiency of system operations. It includes real-time monitoring of process parameters, prediction and control of key factors, dynamic adjustment of production processes, and intelligent analysis and optimization of equipment operating status.

[0003] Among them, the traditional method for predicting and adjusting the moisture content of raw materials used in particleboard production refers to the use of drying experiments in the particleboard production process to determine the moisture content of raw materials. By physical sampling and combined with weighing analysis after heating with drying equipment, the change data of the moisture content of raw materials is obtained, and then the moisture status of the raw materials is adjusted according to manual experience or preset moisture content standards.

[0004] The existing technology relies on physical sampling combined with weighing analysis after heating by drying equipment to obtain raw material moisture content data during the particleboard production process. This method can only perform manual inference based on static detection results at a single time point, and cannot capture the direction and amplitude of the impact of disturbance factors on the raw material moisture content during the production process in real time. It is easy to cause the prediction results to lag and respond to disturbance changes in a timely manner. In actual operation, manual adjustment only relies on experience or preset standards, which is prone to lag or excessive moisture content adjustment. For example, the nonlinear coupling effect of changes in hot air temperature, heating source power or drying channel humidity on the raw material moisture content during production cannot be timely fed back and corrected, resulting in large fluctuations in the raw material moisture content and poor drying uniformity, which in turn causes the risk of board quality fluctuations in subsequent pressing and forming processes. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method for predicting and adjusting the moisture content of raw materials for particleboard production.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a method for predicting and adjusting the moisture content of raw materials for particleboard production, comprising the following steps:

[0007] S1: Obtain the disturbance parameter set of the real-time detection of the particleboard raw material drying section and the moisture content detection sequence records of the specified period, and build a moisture content prediction model using the XGBoost algorithm. The model predicts based on the disturbance parameter set and the moisture content detection sequence records, and outputs the initial moisture content prediction sequence;

[0008] S2: performing sign matching and amplitude sorting on the initial water content prediction sequence and the disturbance parameter set, and outputting a disturbance direction and impact intensity flag set;

[0009] S3: adjusting the amplitude of the initial water content prediction sequence using the disturbance direction and the impact intensity flag set, and correcting the water content prediction using the LSTM algorithm, and outputting a disturbance correction prediction sequence;

[0010] S4: performing reversal detection on the signs of increases and decreases in the water content prediction values ​​in the disturbance-corrected prediction sequence, extracting abnormal response inflection points and calculating local trend residuals, and constructing an abnormal trend residual ranking table;

[0011] S5: performing amplitude scaling and direction adjustment on the disturbance correction prediction sequence according to the abnormal trend residual sorting table, and generating a particleboard raw material drying instruction according to the adjustment result.

[0012] As a further solution of the present invention, the initial moisture content prediction sequence includes a time step prediction value sequence, a predicted change trend, and a predicted time axis; the disturbance direction and impact intensity mark set includes a disturbance direction mark, an impact intensity mark, and a disturbance parameter weight; the disturbance correction prediction sequence is specifically a predicted moisture content value sequence, a disturbance correction amplitude, and a prediction correction time node; the abnormal trend residual sorting table specifically refers to the abnormal response inflection point position, the local trend residual, and the residual sorting weight; the particleboard raw material drying instruction adjustment includes the hot air temperature adjustment amplitude, the heating source power adjustment amplitude, and the drying channel humidity adjustment setting.

[0013] As a further solution of the present invention, the steps for obtaining the initial moisture content prediction sequence are specifically as follows:

[0014] S111: Obtain a disturbance parameter set detected in real time in the drying section of the particleboard raw material, and a moisture content detection sequence recorded as a moisture content detection sequence of the particleboard raw material at the same time, perform maximum and minimum normalization processing on the disturbance parameter set, and generate a normalized disturbance parameter sequence, wherein the disturbance parameter set includes a hot air temperature change rate, a heating source power fluctuation amplitude, and an instantaneous offset of a humidity gradient in the drying channel;

[0015] S112: performing time step concatenation on the normalized disturbance parameter sequence and the moisture content detection sequence and using them as input features of XGBoost, using the moisture content detection value of the next time step as a supervision label, calculating feature gain through error back propagation and completing training iterations to generate a disturbance and moisture content feature gain sequence;

[0016] S113: performing weighted calculation on the normalized disturbance parameter sequence and the moisture content detection sequence records according to the disturbance and moisture content characteristic gain sequence to generate an initial moisture content prediction sequence.

[0017] As a further solution of the present invention, the steps of obtaining the disturbance direction and impact intensity flag set are specifically as follows:

[0018] S211: Obtain the increase or decrease difference between the predicted value of each time step and the predicted value of the previous time step from the initial water content prediction sequence, and calculate the sign of each difference to generate a water content prediction change sign sequence;

[0019] S212: performing element-by-element sign comparison based on the moisture content predicted change sign sequence and the normalized disturbance parameter sequence, determining that identical signs are forward and different signs are reverse, and marking a direction flag for each disturbance parameter to obtain a disturbance direction flag sequence;

[0020] S213: Sort the absolute values ​​of the normalized disturbance parameter sequence by amplitude according to the disturbance direction flag sequence, and output a disturbance direction and influence intensity flag set in combination with the direction flag.

[0021] As a further solution of the present invention, the step of obtaining the disturbance correction prediction sequence is specifically as follows:

[0022] S311: obtaining the disturbance direction flag and amplitude sorting weight of each time step based on the disturbance direction and impact intensity flag set, weighting the direction flag and sorting weight with the time step prediction value of the initial water content prediction sequence to generate a disturbance weighted prediction sequence;

[0023] S312: Continuously concatenate the predicted values ​​of each time step in the disturbance weighted prediction sequence and match them with the corresponding time step numbers to form a disturbance correction input sequence;

[0024] S313: Input the disturbance-corrected input sequence into the LSTM model, combine the time step prediction value of the specified period with the current input vector, calculate the corrected moisture content prediction value of the current time step, and output the disturbance-corrected prediction sequence.

[0025] As a further solution of the present invention, the steps for obtaining the abnormal trend residual ranking table are specifically as follows:

[0026] S411: Based on the continuous increase and decrease trend of the corrected moisture content prediction value at each time step in the disturbance correction prediction sequence, the positive and negative signs of the differences between the prediction values ​​at adjacent time steps are calculated to generate a moisture content increase and decrease sign sequence;

[0027] S412: Based on the water content increase / decrease sign sequence, compare the difference between the sign of the current time step and the sign of the previous time step, determine the position where the positive changes to negative or the negative changes to positive as the sign reversal point, and select the sign reversal time step position to generate an abnormal response inflection point position sequence;

[0028] S413: Based on the abnormal response inflection point position sequence, the disturbance correction prediction sequence within a fixed time window before and after the inflection point is selected, linear fitting is performed, and the fitting residual value of each time step is calculated. The residual value is sorted to obtain an abnormal trend residual sorting table.

[0029] As a further solution of the present invention, the steps for obtaining the particleboard raw material drying instruction are specifically as follows:

[0030] S511: Based on the time step position with a specified ranking in the abnormal trend residual sorting table, the corrected water content prediction value of the corresponding time step is selected, and the ranking position is used as a weight factor to adjust the scaling coefficient and direction sign of the prediction value to obtain a prediction sequence after scaling and direction adjustment;

[0031] S512: Based on the prediction sequence after the scaling direction adjustment, calculating the deviation between the corrected moisture content prediction value and the target moisture content setting value at each time step, and performing amplitude classification according to the deviation to generate a moisture content prediction deviation classification result;

[0032] S513: Based on the moisture content prediction deviation classification result, the deviation amplitude is calculated to adjust the corresponding hot air temperature amplitude, the heating source power increase / decrease amplitude and the drying channel humidity control setting parameters, and the particleboard raw material drying adjustment instruction is generated.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are:

[0034] In the present invention, by obtaining the disturbance parameter set and moisture content detection sequence record of the raw material drying section in the particleboard production process, combining the prediction sequence with the disturbance feature sequence for symbol matching, amplitude sorting, and disturbance direction and influence intensity marking, the raw material moisture content prediction not only has the dynamic trend judgment ability, but also can trace the source of the disturbance. The amplitude of the prediction sequence is adjusted by the disturbance direction and amplitude mark, and the prediction correction of the disturbance response and the detection of trend reversal anomalies are realized. At the same time, local residual analysis is introduced to construct a trend residual sorting table. The correction results are weighted with the abnormal inflection point as the core, and the scaling and direction adjustment of the moisture content correction trend are guided. Finally, the adjustment instructions of the hot air temperature, heating source power and drying channel humidity are generated according to the degree of deviation between the prediction result and the target set value. The moisture content prediction and adjustment can adaptively respond to changes in multi-source disturbances and have the capabilities of trend correction, abnormal inflection point identification, disturbance tracing, error feedback and linkage adjustment of drying process parameters. The dynamic linkage, intelligent correction and precise feedback of the raw material moisture content prediction and adjustment process are realized, and the accuracy of the prediction correction and the adaptability of the drying process are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the main steps of the present invention;

[0036] Figure 2 This is a flow chart of step S1 of the present invention;

[0037] Figure 3 This is a flow chart of step S2 of the present invention;

[0038] Figure 4 This is a flow chart of step S3 of the present invention;

[0039] Figure 5 This is a flow chart of step S4 of the present invention;

[0040] Figure 6 This is a flow chart of step S5 of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0042] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0043] See also Figure 1 The present invention provides a technical solution: a method for predicting and adjusting the moisture content of raw materials for particleboard production, comprising the following steps:

[0044] S1: Obtain the disturbance parameter set of the real-time detection of the particleboard raw material drying section and the moisture content detection sequence records of the specified period, and build a moisture content prediction model using the XGBoost algorithm. The model predicts based on the disturbance parameter set and the moisture content detection sequence records, and outputs the initial moisture content prediction sequence;

[0045] S2: Perform sign matching and amplitude sorting on the initial water content prediction sequence and the disturbance parameter set, and output the disturbance direction and impact intensity sign set;

[0046] S3: The amplitude of the initial water content prediction sequence is adjusted by the disturbance direction and impact intensity flag set, and the water content prediction is corrected by the LSTM algorithm, and the disturbance correction prediction sequence is output;

[0047] S4: Perform reversal detection on the sign of the increase or decrease of the water content prediction value in the disturbance correction prediction sequence, extract the abnormal response inflection point and calculate the local trend residual, and construct the abnormal trend residual ranking table;

[0048] S5: performing amplitude scaling and direction adjustment on the disturbance correction prediction sequence according to the abnormal trend residual sorting table, and generating particleboard raw material drying instructions according to the adjustment results;

[0049] The initial moisture content prediction sequence includes the time step prediction value sequence, the predicted change trend, and the predicted time axis. The disturbance direction and impact intensity mark set includes the disturbance direction mark, the impact intensity mark, and the disturbance parameter weight. The disturbance correction prediction sequence specifically includes the predicted moisture content value sequence, the disturbance correction amplitude, and the prediction correction time node. The abnormal trend residual sorting table specifically refers to the abnormal response inflection point position, the local trend residual, and the residual sorting weight. The particleboard raw material drying instruction adjustment includes the hot air temperature adjustment amplitude, the heating source power adjustment amplitude, and the drying channel humidity adjustment setting.

[0050] See also Figure 2 , the specific steps for obtaining the initial moisture content prediction sequence are:

[0051] S111: Obtain a disturbance parameter set detected in real time in the drying section of the particleboard raw material, and a moisture content detection sequence recorded as a moisture content detection sequence of the particleboard raw material at the same time, perform maximum and minimum normalization processing on the disturbance parameter set, and generate a normalized disturbance parameter sequence, wherein the disturbance parameter set includes a hot air temperature change rate, a heating source power fluctuation amplitude, and an instantaneous offset of a humidity gradient in the drying channel;

[0052] When obtaining the disturbance parameter set of the real-time detection of the particleboard raw material drying section, it is necessary to monitor and collect the environmental and equipment parameters of the drying section in real time during the particleboard raw material drying process, including the hot air temperature change rate, the heating source power fluctuation amplitude and the instantaneous offset of the drying channel humidity gradient. The hot air temperature change rate can be measured by thermocouples to obtain the change in hot air temperature per minute. For example, by sampling temperature data T1=150℃, T2=152℃, T3=151.5℃ in real time, the average temperature change rate within one minute is calculated to be (152-150) / 1=2℃ / min, and the heating source power fluctuation amplitude is obtained by The current and voltage are sampled and the power is calculated. Assuming that the maximum power within one minute is 15kW and the minimum power is 12kW, the fluctuation amplitude is 15-12=3kW. The instantaneous offset of the humidity gradient in the drying channel is measured by humidity sensors installed at different locations in the drying channel. For example, if the inlet humidity is 45% and the outlet humidity is 20%, and at a certain moment the inlet humidity changes to 46% and the outlet humidity to 19%, the instantaneous offset is (46-19)-(45-20)=2%. The above disturbance parameters are filled with missing values ​​through linear interpolation and denoised to form unified disturbance parameter data.

[0053] S112: The normalized perturbation parameter sequence and the moisture content detection sequence are spliced ​​in time steps and used as input features of XGBoost. The moisture content detection value of the next time step is used as the supervision label. The feature gain is calculated through error back propagation and training iterations are completed to generate a perturbation and moisture content feature gain sequence.

[0054] In the process of splicing the normalized disturbance parameter sequence and the moisture content detection sequence, the disturbance parameter needs to be normalized first. For example, the original value range of the hot air temperature change rate is 0 to 10. , the current value is 2 , normalized to 0.2, the heating source power fluctuation range is 0 to 5kW, the current value is 3kW, normalized to 0.6, the humidity gradient instantaneous offset range is 0 to 5%, the current value is 2%, normalized to 0.4. The moisture content detection value does not need to be normalized, for example The moisture content at this moment is 6%. The time is 5.8%, The input feature sequence formed by splicing is 、 、 , set the moisture content detection value of the next time step as the supervision label, for example The moisture content is 5.5% at this moment, so the label .

[0055] The error back propagation mechanism is used in the training process, and the loss function is the mean square error (MSE), which is expressed as:

[0056] ;

[0057] in: : The total number of samples, for example, if the number of time steps is 5, then , determined by counting; : Sample serial number index, indicating the samples, ranging from 1 to Integer between ; : No. The actual moisture content of each sample is measured by a moisture content detector, such as , ; : No. The predicted moisture content of the samples is obtained by the XGBoost model, for example , ; :Indicates the The square of the sample prediction error, for example .

[0058] In the XGBoost model, the core of the training process is feature gain calculation, and the gain formula is:

[0059] ;

[0060] in: : Indicates the gain value of the current feature when it is used to split the node. The larger the value, the more suitable the feature is as a splitting condition. : is the first-order gradient sum, calculated as ,in: : The set of samples contained in the current node, such as Indicates that the node contains the first and second samples; : No. The gradient of the sample, the first-order partial derivative, is calculated as follows: ,in The loss function is in the form of MSE ,but ,For example , ,but ; : is the second-order gradient sum, calculated as ,in: : No. Hessian value of samples, MSE loss , is a constant value; : Regularization term, used to control model complexity, the value is usually between 0.1 and 1.0, for example, here we set .

[0061] Assume that the current node contains two samples, , ,but , , substituting into the formula we get , which represents the information gain generated by splitting the node using this feature.

[0062] After all feature gains are calculated, the gain results of each feature are recorded to form a disturbance and moisture content feature gain sequence. For example, if the hot air temperature change rate feature gain is 0.5, the heating source power fluctuation amplitude feature gain is 0.6, the humidity gradient instantaneous offset feature gain is 0.4, and the moisture content feature gain is 0.3, then the sequence is .

[0063] S113: performing weighted calculation on the normalized disturbance parameter sequence and the moisture content detection sequence records according to the disturbance and moisture content characteristic gain sequence to generate an initial moisture content prediction sequence;

[0064] When performing weighted calculations on the normalized perturbation parameter sequence and moisture content detection sequence records based on the perturbation and moisture content feature gain sequence, it is first necessary to clarify the meaning of the perturbation and moisture content feature gain sequence. Each value in this sequence represents the weighted contribution of the corresponding feature in the prediction. For example, the hot air temperature change rate feature gain is 0.5, the heating source power fluctuation amplitude feature gain is 0.6, the humidity gradient instantaneous offset feature gain is 0.4, and the moisture content feature gain is 0.3, which respectively represent the relative weights of these four features. These gain values ​​are used as weighting coefficients to perform weighted calculations on the normalized perturbation parameter sequence and moisture content detection sequence records at each time step.

[0065] For example, for the first time step, the normalized perturbation parameter is , the moisture content test value is 6.0%, then the weighted calculation process is: the weighted result of the hot air temperature change rate is , the weighted result of the heating source power fluctuation amplitude is , the weighted result of the instantaneous offset of the humidity gradient is , the weighted result of moisture content detection value is , the weighted results are added together to get The 2.42 is the first value of the initial moisture content prediction sequence after weighted calculation.

[0066] For the second time step, the normalized perturbation parameter is The moisture content test value is 5.8%, and the weighted result of the hot air temperature change rate is , the weighted result of the heating source power fluctuation amplitude is , the weighted result of the instantaneous offset of the humidity gradient is , the weighted result of moisture content detection value is , the weighted sum is .

[0067] Similarly, for the third time step, the normalized perturbation parameter is , the moisture content test value is 5.6%, and the weighted result of the hot air temperature change rate is , the weighted result of the heating source power fluctuation amplitude is , the weighted result of the instantaneous offset of the humidity gradient is , the weighted result of moisture content detection value is , the weighted sum is .

[0068] See also Figure 3 , the specific steps for obtaining the disturbance direction and impact intensity flag set are:

[0069] S211: Obtain the increase or decrease difference between the predicted value of each time step and the predicted value of the previous time step from the initial water content prediction sequence, and calculate the sign mark of each difference to generate a water content prediction change sign sequence;

[0070] To calculate the difference between each time step's predicted value and the previous time step's predicted value from the initial moisture content prediction sequence, the equation is listed for each time step. For example, if the initial moisture content prediction sequence is [2.42, 2.48, 2.42, 2.50], the difference at the second time step is 2.48 - 2.42 = 0.06, the difference at the third time step is 2.42 - 2.48 = -0.06, and the difference at the fourth time step is 2.50 - 2.42 = 0.08. The sign of each difference is determined by determining its positive or negative sign. For example, if 0.06 > 0, it is marked as +1; if -0.06 < 0, it is marked as -1; and if 0.08 > 0, it is marked as +1. The resulting sequence of moisture content prediction change signs is [+1, -1, +1], which indicates the direction of the predicted change at each time step.

[0071] S212: performing element-by-element sign comparison based on the predicted water content change sign sequence and the normalized disturbance parameter sequence, determining that identical signs are positive and different signs are negative, and marking a direction flag for each disturbance parameter to obtain a disturbance direction flag sequence;

[0072] When comparing the sign of the predicted moisture content change sequence with the normalized disturbance parameter sequence element by element, first obtain the sign of each disturbance parameter. For example, if the normalized disturbance parameter sequence is [0.2, -0.6, 0.4], the sign is marked as [+1, -1, +1]. The predicted moisture content change sign sequence is [+1, -1, +1]. When the two sequences are compared element by element, the first item +1 and +1 have the same sign and are marked as positive. The second item -1 and +1 have different signs and are marked as negative. The third item +1 and +1 have the same sign and are marked as positive. Therefore, the direction mark sequence for this time step is [forward, negative, forward]. Each disturbance parameter obtains a direction mark based on whether the sign is the same or different, and the direction of the disturbance parameter at each time step is marked in turn.

[0073] S213: sorting the absolute values ​​of the normalized disturbance parameter sequence by amplitude according to the disturbance direction flag sequence, and outputting a disturbance direction and impact intensity flag set in combination with the direction flag;

[0074] When sorting the absolute values ​​of the normalized disturbance parameter sequence by amplitude according to the disturbance direction mark sequence, the absolute value is calculated first. For example, if the normalized disturbance parameter sequence is [0.2, -0.6, 0.4], the absolute value is [0.2, 0.6, 0.4]. Sorting by size yields the order 0.6>0.4>0.2, and the corresponding disturbance parameters are heating source power fluctuation amplitude>humidity gradient instantaneous offset>hot air temperature change rate. Combined with the direction mark sequence [forward, reverse, forward], the disturbance direction and impact intensity mark set is obtained as heating source power fluctuation amplitude (reverse), humidity gradient instantaneous offset (forward), and hot air temperature change rate (forward). The above calculation is repeated for each time step, gradually generating a disturbance direction and impact intensity mark set, forming a complete disturbance direction and intensity mark sequence.

[0075] See also Figure 4 , the specific steps for obtaining the disturbance-corrected prediction sequence are:

[0076] S311: Based on the disturbance direction and impact intensity flag set, obtain the disturbance direction flag and amplitude sorting weight of each time step, weight the direction flag and sorting weight with the time step prediction value of the initial water content prediction sequence, and generate a disturbance weighted prediction sequence;

[0077] When obtaining the disturbance direction flag and amplitude ranking weight for each time step based on the disturbance direction and impact intensity flag set, it is necessary to clarify the meaning and acquisition method of each element. The disturbance direction flag refers to the direction flag of each disturbance parameter at the current time step, for example, "forward" is marked as +1, and "reverse" is marked as -1. The amplitude ranking weight is derived from the weight values ​​assigned after sorting the absolute values ​​of the disturbance parameters. The weights are assigned in descending order according to the sorting order. For example, in the sorting result, the maximum value weight is set to 0.5, the second largest value weight is set to 0.3, the minimum value weight is set to 0.2, and so on. For a certain time step, for example, the disturbance direction flag is [forward, reverse, forward], the corresponding value is marked as [+1, -1, +1], and the amplitude sorting result is the heating source power fluctuation amplitude (0.6), the humidity gradient instantaneous offset (0.4), and the hot air temperature change rate (0.2). The assigned weights are [0.5, 0.3, 0.2]. The direction flag is element-wise multiplied by the sorting weight to obtain the directional weight of each disturbance parameter. For example, the direction flag of the hot air temperature change rate at the second time step is +1, and the sorting weight is 0.2, resulting in a product of +1 × 0.2 = +0.2; the direction flag of the heating source power fluctuation amplitude is -1, and the sorting weight is 0.5, resulting in a product of -1 × 0.5 = -0.5; the direction flag of the humidity gradient instantaneous offset is +1, and the sorting weight is 0.3, resulting in a product of +1 × 0.3 = +0.3. These disturbance directional weights are weighted with the time step prediction value of the initial moisture content prediction sequence. For example, if the initial moisture content prediction value is 2.48, the disturbance weighting result is 2.48 + 0.2 - 0.5 + 0.3 = 2.48. The same operation is performed for each time step to generate a disturbance weighted prediction sequence.

[0078] S312: Continuously concatenate the predicted values ​​of each time step in the disturbance weighted prediction sequence and match them with the corresponding time step numbers to form a disturbance correction input sequence;

[0079] When continuously concatenating the predicted values ​​for each time step in the perturbation-weighted prediction sequence, it is necessary to directly concatenate the perturbation-weighted prediction values ​​for each time step into a long sequence in the order of the time steps. Taking the perturbation-weighted prediction sequence as an example, if the time step prediction values ​​are [2.48, 2.45, 2.50, 2.46], the continuous concatenation result is [2.48, 2.45, 2.50, 2.46], without the need for additional calculations. The operation of matching the time step numbers is to add the corresponding number to each time step, for example, the first time step is numbered 1, the second time step is numbered 2, the third time step is numbered 3, and the fourth time step is numbered 4. The predicted values ​​are matched one-to-one with the time step numbers to form the final perturbation-corrected input sequence, for example, [(1, 2.48), (2, 2.45), (3, 2.50), (4, 2.46)]. The whole process is carried out in sequence according to the time steps, and the predicted value and number information are retained at each time step to form a complete disturbance correction input sequence.

[0080] S313: Input the disturbance-corrected input sequence into the LSTM model, combine the time step prediction value of the specified period with the current input vector, calculate the corrected moisture content prediction value of the current time step, and output the disturbance-corrected prediction sequence;

[0081] When the perturbation-corrected input sequence is input into the LSTM model, the composition of the input vector is first clarified. The input vector for each time step is ,in is the time step number, is the perturbation weighted prediction value, for example, the input sequence is The LSTM model processes the input sequence time-step by time step and calculates the corrected moisture content prediction value based on the recursive structure. The core calculation formula is:

[0082] ;

[0083] in, : represents the time step The corrected moisture content prediction value is calculated by the LSTM model and output step by step according to the recursive formula at each time step. For example, at the current time step Time , from the previous time step With the current input Decide; : represents the previous time step The LSTM output value is the corrected moisture content prediction value of the previous time step, which comes from the recursive calculation result of the model. For example, ; : represents the disturbance weighted prediction value, which comes from the disturbance weighted prediction sequence of S311, for example , obtained by weighting the disturbance direction flag, sorting weight and initial water content prediction value; : is the weight matrix of the state of the previous time step, which represents the weighted influence of the memory ability learned by the model on the previous state. It is generally obtained through training. When not trained, the initial value can be set through experimental experience. For example, according to the historical fluctuation range, , indicating a strong memory dependency, and the value range is usually , in actual use, it can be obtained by adjusting the parameters of the verification set; : is the weight matrix of the current input vector, which indicates the sensitivity of the model to the current disturbance input. It is usually obtained through training. The initial value of the experiment can be set to , the recommended range is set at , adjusted according to the actual disturbance amplitude ratio; : is a bias term used to adjust the model output baseline, usually initialized to a small constant, such as , the setting basis is to ensure that there is no excessive deviation in the initial state; : is the hyperbolic tangent activation function, which maps the weighted sum to interval, ensuring smooth convergence of model output, and the activation function is selected by default for the LSTM internal structure.

[0084] Assume that the output of the previous time step is , the weighted prediction value of the disturbance at the current time step , the weight matrix is ​​set to 、 , bias term , then substitute into the formula to calculate:

[0085] ;

[0086] calculate . To map to the actual moisture content range , it needs to be denormalized, the formula is:

[0087] Corrected moisture content prediction value ;

[0088] Substitution have to: .

[0089] Therefore, the current time step The corrected moisture content prediction value is 9.98%. Each time step is calculated according to this process, for example hour Depend on and Calculate and generate the complete disturbance correction prediction sequence in sequence.

[0090] See also Figure 5 , the specific steps for obtaining the abnormal trend residual sorting table are:

[0091] S411: Based on the continuous increase and decrease trend of the corrected moisture content prediction value at each time step in the disturbance correction prediction sequence, the positive and negative signs of the differences between the prediction values ​​at adjacent time steps are calculated to generate a moisture content increase and decrease sign sequence;

[0092] When the corrected moisture content prediction value at each time step in the disturbance-corrected prediction sequence shows a continuous increase or decrease, it is necessary to calculate the difference between the prediction values ​​at adjacent time steps and extract their signs. First, list the disturbance-corrected prediction sequence time step by time step, for example, [9.98, 9.85, 10.00, 9.90]. The difference between the prediction values ​​at adjacent time steps is calculated as follows: 9.85 - 9.98 = -0.13 for the second time step, 10.00 - 9.85 = 0.15 for the third time step, and 9.90 - 10.00 = -0.10 for the fourth time step. The sign of the difference is determined based on the sign of the difference: if the difference is greater than zero, it is marked as +1; if the difference is less than zero, it is marked as -1; and if the difference is equal to zero, it is marked as 0. In the above example, the difference at the second time step is -0.13, with a sign of -1; the difference at the third time step is 0.15, with a sign of +1; and the difference at the fourth time step is -0.10, with a sign of -1. The final generated sequence of moisture content increase and decrease symbols is [-1, +1, -1], which is used to mark the direction of change of the predicted value at each time step.

[0093] S412: Based on the sign sequence of water content increase and decrease, the sign of the current time step is compared with the sign of the previous time step, and the position from positive to negative or from negative to positive is determined as the sign reversal point. The sign reversal time step position is selected to generate an abnormal response inflection point position sequence;

[0094] Based on the sign sequence of water content changes, when comparing the sign of the current time step with the sign of the previous time step, it is necessary to determine whether a sign reversal has occurred. This is checked time-step by time-step. For example, if the sign sequence of water content changes is [-1, +1, -1], starting from the second time step, the sign of the second time step (+1) is compared with the sign of the first time step (-1). The signs are different, so the second time step is a sign reversal point. The sign of the third time step (-1) is compared with the sign of the second time step (+1). The signs are different, so the third time step is also a sign reversal point. The sign reversal time steps are screened out to generate a sequence of abnormal response inflection point locations. For example, [2, 3] indicates that sign reversals were detected in the second and third time steps, which are potential abnormal inflection point locations.

[0095] S413: Based on the abnormal response inflection point position sequence, a disturbance correction prediction sequence within a fixed time window before and after the inflection point is selected, a linear fit is performed, and the fitting residual value of each time step is calculated. The residual values ​​are sorted to obtain an abnormal trend residual sorting table;

[0096] When selecting a disturbance-corrected prediction sequence within a fixed time window before and after the inflection point based on the abnormal response inflection point location sequence, the time window length must be set first. For example, a window length of 1 indicates selecting one time step before and after the inflection point. Taking the abnormal response inflection point location sequence [2, 3] as an example, for the inflection point at the second time step, the disturbance-corrected prediction values ​​for the first, second, and third time steps are selected, for example, [9.98, 9.85, 10.00]. For the inflection point at the third time step, the disturbance-corrected prediction values ​​for the second, third, and fourth time steps are selected, for example, [9.85, 10.00, 9.90]. A linear fit is performed within the selected window, using the formula: predicted value = slope × time step number + intercept. The fitting line parameters are calculated using the least squares method. Taking the inflection point of the second time step as an example, the time steps are numbered [1, 2, 3], the predicted values ​​are [9.98, 9.85, 10.00], and the fitting line parameters are calculated. Then, the fitting residual value for each time step is calculated (i.e., the difference between the predicted value and the fitted line value). For example, the residual for time step 1 is 9.98 - the fitted value, the residual for time step 2 is 9.85 - the fitted value, and the residual for time step 3 is 10.00 - the fitted value). All residual values ​​are sorted to obtain a ranking table of abnormal trend residuals. For example, the residual for the first time step is 0.05, the residual for the second time step is -0.10, and the residual for the third time step is 0.07. The absolute value is sorted from largest to smallest: second time step > third time step > first time step, ultimately forming a complete ranking table of abnormal trend residuals.

[0097] See also Figure 6 , the specific steps for obtaining the particleboard raw material drying instruction are:

[0098] S511: Based on the time step position with a specified ranking in the abnormal trend residual sorting table, select the corrected water content prediction value of the corresponding time step, and use the sorting position as a weight factor to adjust the scaling coefficient and direction sign of the prediction value to obtain a prediction sequence after scaling and direction adjustment;

[0099] When selecting the corrected moisture content prediction value for a time step with a specified ranking in the abnormal trend residual ranking table, the ranking in the ranking table must be determined. For example, the top two time steps are selected as references. Taking the abnormal trend residual ranking table as an example, if the first-ranked time step is the second time step and the second-ranked time step is the third time step, the corrected moisture content prediction values ​​for these two time steps are selected. For example, the second time step is 9.85, and the third time step is 10.00. The ranking position is used as a weighting factor. For example, if the weighting factor is set to the proportion of the last-ranked value, the first-ranked time step has a weight of 1, and the second-ranked time step has a weight of 0.5. When adjusting the scaling factor, the corrected prediction value is multiplied by the weighting factor. For example, the second time step is adjusted to 9.85 × 1 = 9.85, and the third time step is adjusted to 10.00 × 0.5 = 5.00. Direction sign adjustment is performed by reversing the sign of the prediction deviation. For example, if the moisture content prediction value is greater than the target value, the direction sign is +1; if it is less than the target value, the direction sign is -1. The adjusted sign is applied to the prediction value. For example, if the direction sign of the second time step is +1, the adjusted value is 9.85×+1=9.85; if the direction sign of the third time step is -1, the adjusted value is 5.00×-1=-5.00. This generates a scaled direction-adjusted prediction sequence.

[0100] S512: Based on the prediction sequence after the scaling direction adjustment, the deviation between the corrected moisture content prediction value and the target moisture content setting value at each time step is calculated, and the magnitude is graded according to the deviation to generate a moisture content prediction deviation classification result;

[0101] When calculating the deviation between the corrected moisture content prediction value and the target moisture content set value at each time step based on the scaled prediction sequence, it is first necessary to clarify the basis for setting the target moisture content set value. The target moisture content set value is usually determined based on the particleboard production process standard. For example, the national standard requires a particleboard factory moisture content range of 6% to 10%. To ensure production safety and product performance, the target set value is often set at 8.00% as the process setting benchmark.

[0102] When calculating the deviation of each time step, directly list the formula. For example, if the corrected predicted value of the second time step is 9.85 and the target value is 8.00, then the deviation is 9.85-8.00=+1.85; if the adjusted predicted value of the third time step is -5.00 and the target value is 8.00, then the deviation is -5.00-8.00=-13.00.

[0103] The deviation range is set based on the process sensitivity range and the safety control line. Considering the effect of particleboard moisture content on product performance, the following grading ranges are set:

[0104] An absolute deviation value between 0 and 1.0 indicates that the fluctuation is within the allowable range and is considered a "slight deviation." This is based on the fact that a moisture content deviation within 1% has little effect on the mechanical properties of the product.

[0105] An absolute deviation value greater than 1.0 and less than 3.0 indicates a certain degree of deviation, which is classified as "moderate deviation". The basis is that when the deviation exceeds 1%, the product performance begins to fluctuate significantly, and the process needs attention;

[0106] An absolute deviation value greater than 3.0 and less than 5.0 indicates a high degree of deviation and is considered "significant deviation." This deviation may cause surface cracking, warping, and other issues.

[0107] An absolute deviation value greater than 5.0 indicates a serious deviation, which is classified as a "serious deviation". The reason is that the moisture content is far beyond the normal process control line, and immediate measures must be taken to adjust the drying process.

[0108] In this example, the deviation at the second time step is +1.85, with an absolute value of 1.85, falling between 1.0 and 3.0, and is classified as "moderate deviation." The deviation at the third time step is -13.00, with an absolute value of 13.00, exceeding 5.0, and is classified as "severe deviation." The resulting moisture content prediction deviation classification is [(2, moderate deviation), (3, severe deviation)].

[0109] S513: Based on the moisture content prediction deviation classification result, the deviation range is calculated and the corresponding hot air temperature adjustment range, the heating source power increase / decrease range and the drying channel humidity control setting parameters are adjusted to generate the particleboard raw material drying adjustment instruction;

[0110] Based on the moisture content prediction deviation classification results, when calculating the hot air temperature adjustment range, heating source power increase / decrease range, and drying channel humidity control setting parameters corresponding to the deviation range, first clarify the meaning and setting basis of each adjustment parameter:

[0111] Hot air temperature adjustment range: It is expressed as the value of changing the hot air temperature in degrees Celsius. It is mainly determined by the degree of deviation of moisture content. The greater the deviation, the greater the adjustment range.

[0112] Heating source power increase or decrease range: expressed as the heating source output power adjustment value, in kilowatts (kW), set according to the relationship between drying load and moisture content;

[0113] Drying channel humidity control setting parameter: represents the correction value of the drying channel humidity setting, the unit is percentage (%), which is achieved by adjusting the channel moisture removal amount.

[0114] The core formula for calculating the adjustment range is the linear proportional adjustment model, which is set as follows:

[0115] , , .

[0116] in: : Hot air temperature adjustment range (unit: °C), used to adjust the drying medium temperature; : The increase or decrease range of the heating source power (unit: kW), used to adjust the heat source output; : Drying channel humidity control setting parameter adjustment value (unit: %), used to modify the channel humidity setting; : Hot air temperature adjustment coefficient, set based on experience, for example, 3.0℃ / %, means that for every 1% increase in moisture content deviation, the hot air temperature needs to be adjusted by 3.0℃; : Heating source power adjustment coefficient, set according to equipment capacity and process design, for example 1.5kW / ; : Humidity setting adjustment coefficient, set according to dehumidification capacity and process sensitivity, for example 0.8% / ; : The absolute deviation between the predicted moisture content and the target moisture content set value. For example, the deviation in the second time step is 1.85%.

[0117] Taking the second time step as an example, , substitute into the formula to calculate:

[0118] , , .

[0119] Taking the third time step as an example, , substitute into the formula to calculate:

[0120] , , .

[0121] The resulting drying control instructions for the particleboard raw material are: in the second time step, the hot air temperature is increased by 5.55°C, the heating source power is increased by 2.775kW, and the drying channel humidity setting is increased by 1.48%; in the third time step, the hot air temperature is increased by 39.00°C, the heating source power is increased by 19.50kW, and the drying channel humidity setting is increased by 10.40%. This calculation method accurately calculates the drying control parameters for each time step's deviation, combined with the preset adjustment coefficient, achieving precise control.

[0122] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for predicting and adjusting the moisture content of raw materials for particleboard production, characterized in that: The following steps are involved: S1: Obtain the disturbance parameter set of the real-time detection of the particleboard raw material drying section and the moisture content detection sequence records of the specified period, and build a moisture content prediction model using the XGBoost algorithm. The model predicts based on the disturbance parameter set and the moisture content detection sequence records and outputs the initial moisture content prediction sequence. The disturbance parameter set includes the hot air temperature change rate, the heating source power fluctuation amplitude, and the instantaneous offset of the drying channel humidity gradient; S2: performing sign matching and amplitude sorting on the initial water content prediction sequence and the disturbance parameter set, and outputting a disturbance direction and impact intensity flag set; S3: adjusting the amplitude of the initial water content prediction sequence using the disturbance direction and the impact intensity flag set, and correcting the water content prediction using the LSTM algorithm, and outputting a disturbance correction prediction sequence; The steps for obtaining the disturbance correction prediction sequence are specifically as follows: S311: obtaining the disturbance direction flag and amplitude sorting weight of each time step based on the disturbance direction and impact intensity flag set, weighting the direction flag and sorting weight with the time step prediction value of the initial water content prediction sequence to generate a disturbance weighted prediction sequence; S312: Continuously concatenate the predicted values ​​of each time step in the disturbance weighted prediction sequence and match them with the corresponding time step numbers to form a disturbance correction input sequence; S313: Input the disturbance-corrected input sequence into the LSTM model, combine the time step prediction value of the specified period with the current input vector, calculate the corrected moisture content prediction value of the current time step, and output the disturbance-corrected prediction sequence; S4: performing reversal detection on the signs of increases and decreases in the water content prediction values ​​in the disturbance-corrected prediction sequence, extracting abnormal response inflection points and calculating local trend residuals, and constructing an abnormal trend residual ranking table; S5: performing amplitude scaling and direction adjustment on the disturbance correction prediction sequence according to the abnormal trend residual sorting table, and generating a particleboard raw material drying instruction according to the adjustment result.

2. The method for predicting and adjusting the moisture content of raw materials for particleboard production according to claim 1, characterized in that: The initial moisture content prediction sequence includes a time step prediction value sequence, a predicted change trend, and a predicted time axis. The disturbance direction and impact intensity flag set includes a disturbance direction flag, an impact intensity flag, and a disturbance parameter weight. The disturbance correction prediction sequence specifically includes a predicted moisture content value sequence, a disturbance correction amplitude, and a prediction correction time node. The abnormal trend residual sorting table specifically refers to the abnormal response inflection point position, the local trend residual, and the residual sorting weight. The particleboard raw material drying instruction adjustment includes the hot air temperature adjustment amplitude, the heating source power adjustment amplitude, and the drying channel humidity adjustment setting.

3. The method for predicting and adjusting the moisture content of raw materials for particleboard production according to claim 1, characterized in that: The steps for obtaining the initial moisture content prediction sequence are specifically as follows: S111: obtaining a disturbance parameter set detected in real time during the drying section of the particleboard raw material, and a moisture content detection sequence recorded as a moisture content detection sequence of the particleboard raw material at the same time, performing maximum and minimum normalization processing on the disturbance parameter set, and generating a normalized disturbance parameter sequence; S112: performing time step concatenation on the normalized disturbance parameter sequence and the moisture content detection sequence and using them as input features of XGBoost, using the moisture content detection value of the next time step as a supervision label, calculating feature gain through error back propagation and completing training iterations to generate a disturbance and moisture content feature gain sequence; S113: performing weighted calculation on the normalized disturbance parameter sequence and the moisture content detection sequence records according to the disturbance and moisture content characteristic gain sequence to generate an initial moisture content prediction sequence.

4. The method for predicting and adjusting the moisture content of raw materials for particleboard production according to claim 3, characterized in that: The steps for obtaining the disturbance direction and impact intensity flag set are specifically as follows: S211: Obtain the increase or decrease difference between the predicted value of each time step and the predicted value of the previous time step from the initial water content prediction sequence, and calculate the sign of each difference to generate a water content prediction change sign sequence; S212: performing element-by-element sign comparison based on the moisture content predicted change sign sequence and the normalized disturbance parameter sequence, determining that identical signs are forward and different signs are reverse, and marking a direction flag for each disturbance parameter to obtain a disturbance direction flag sequence; S213: Sort the absolute values ​​of the normalized disturbance parameter sequence by amplitude according to the disturbance direction flag sequence, and output a disturbance direction and influence intensity flag set in combination with the direction flag.

5. The method for predicting and adjusting the moisture content of raw materials for particleboard production according to claim 1, wherein: The steps for obtaining the abnormal trend residual sorting table are specifically as follows: S411: Based on the continuous increase and decrease trend of the corrected moisture content prediction value at each time step in the disturbance correction prediction sequence, the positive and negative signs of the differences between the prediction values ​​at adjacent time steps are calculated to generate a moisture content increase and decrease sign sequence; S412: Based on the water content increase / decrease sign sequence, compare the difference between the sign of the current time step and the sign of the previous time step, determine the position where the positive changes to negative or the negative changes to positive as the sign reversal point, and select the sign reversal time step position to generate an abnormal response inflection point position sequence; S413: Based on the abnormal response inflection point position sequence, the disturbance correction prediction sequence within a fixed time window before and after the inflection point is selected, linear fitting is performed, and the fitting residual value of each time step is calculated. The residual value is sorted to obtain an abnormal trend residual sorting table.

6. The method for predicting and adjusting the moisture content of raw materials for particleboard production according to claim 5, characterized in that: The specific steps for obtaining the particleboard raw material drying instruction are: S511: Based on the time step position with a specified ranking in the abnormal trend residual sorting table, the corrected water content prediction value of the corresponding time step is selected, and the ranking position is used as a weight factor to adjust the scaling coefficient and direction sign of the prediction value to obtain a prediction sequence after scaling and direction adjustment; S512: Based on the prediction sequence after the scaling direction adjustment, calculating the deviation between the corrected moisture content prediction value and the target moisture content setting value at each time step, and performing amplitude classification according to the deviation to generate a moisture content prediction deviation classification result; S513: Based on the moisture content prediction deviation classification result, the deviation amplitude is calculated to adjust the corresponding hot air temperature amplitude, the heating source power increase / decrease amplitude and the drying channel humidity control setting parameters, and the particleboard raw material drying adjustment instruction is generated.

Images