Cigarette packet packaging film static electricity monitoring and eliminating system based on big data analysis

Abstract

The invention relates to the technical field of electrostatic monitoring, in particular to a cigarette packet packaging film electrostatic monitoring and eliminating system based on big data analysis. The method specifically comprises the following steps: collecting surface charge data of a packaging film in real time, carrying out high-precision fidelity noise reduction on electrostatic signals in combination with an improved empirical mode decomposition and adaptive threshold denoising algorithm, and reducing the abrupt change signal loss risk by adopting a dynamic window smoothing technology; constructing a Res-GRU-Attn model, performing time sequence modeling on the multi-dimensional feature embedded data, predicting a future static accumulation trend, and reducing an accumulated error through a dynamic error correction mechanism; dynamically adjusting the ion wind intensity and the positive and negative ion output proportion based on the prediction result; pID closed-loop feedback control is combined to optimize elimination parameters in real time; and meanwhile, static tracing and accurate recording are introduced to ensure data reliability and accuracy. Signal processing, intelligent prediction and dynamic control are combined, and the intelligent level of electrostatic monitoring and neutralization in the cigarette packet production process is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrostatic monitoring, and particularly to an electrostatic monitoring and elimination system for cigarette package films based on big data analysis. Background Art

[0002] During the production process of cigarette package films, electrostatic accumulation is likely to cause material adsorption, dust pollution and equipment failures. Traditional electrostatic monitoring technologies mostly use fixed filtering algorithms to process non-stationary signals, resulting in the loss of high-frequency transient impact characteristics. Moreover, prediction models based on static thresholds or single sensors are difficult to cope with the dynamic charge distribution under complex working conditions, and the insufficient prediction accuracy further restricts the elimination efficiency.

[0003] A Chinese invention patent with the publication number CN110856328B discloses a closed-loop electrostatic monitoring and elimination system. Along the running route of the product to be de-electrified, above the guide rail, a first electrostatic monitoring device, a first electrostatic elimination device, a second electrostatic monitoring device, a second electrostatic elimination device and a third electrostatic monitoring device are sequentially arranged; the first electrostatic monitoring device monitors the initial static voltage of the product to be de-electrified; the first electrostatic elimination device performs the first electrostatic elimination on the product to be de-electrified; the second electrostatic monitoring device monitors the remaining static voltage of the product to be de-electrified; the second electrostatic elimination device performs the second electrostatic elimination on the product to be de-electrified; the third electrostatic monitoring device monitors the residual static voltage of the product to be de-electrified; it automatically performs electrostatic monitoring and electrostatic elimination functions on the product to be de-electrified by setting an electrostatic monitoring device at the front and rear positions of each electrostatic elimination device and adopting a closed-loop feedback mode; it can also monitor or evaluate the de-electrification ability of the electrostatic elimination device.

[0004] With the upgrading of industrial intelligence and the complication of the production environment, the demand for high-precision fidelity processing of non-stationary signals, real-time prediction of dynamic charge distribution and intelligent regulation of elimination parameters has increased significantly; at the same time, the rapid development of big data analysis, machine learning and Internet of Things technologies has provided a new technical path for the industrial control field, making it possible to build a comprehensive system integrating signal processing, time series prediction and closed-loop control, thereby solving key technical bottlenecks such as non-stationary signal fidelity, prediction error accumulation, insufficient elimination efficiency and data credibility, and improving the quality control and intelligent level of cigarette production. Summary of the Invention

[0005] The object of the present invention is to propose an electrostatic monitoring and elimination system for cigarette package films based on big data analysis in view of the problems in the background art.

[0006] The technical solution of the present invention: An electrostatic monitoring and elimination system for cigarette package films based on big data analysis, comprising:

[0007] The static electricity monitoring module is used to monitor the static electricity data on the surface of the packaging film in real time and transmit the static electricity data to the static electricity data processing module;

[0008] The static electricity data processing module is used to process non-stationary static electricity signals by combining improved empirical mode decomposition and adaptive threshold denoising, denoise and reconstruct the signals through adaptive threshold, perform dynamic window smoothing based on the static electricity accumulation rate, map the processed data into a finite field matrix through a hash function and generate a static electricity trace record, and transmit {the static electricity trace record, the processed data} to the data analysis and prediction module;

[0009] The data analysis and prediction module is used to ensure the reliability of the received data through the hash check of the static electricity trace record, predict the future charge trend using the Res-GRU-Attn model with attention mechanism and residual connection introduced, dynamically correct the model by combining historical errors to reduce the cumulative error, divide the static electricity risk level based on the dynamic risk threshold and transmit it to the static electricity elimination module;

[0010] The static electricity elimination module is used to generate continuous control factors based on the static electricity risk level, dynamically adjust the output voltage and current of the ion wind, and dynamically regulate the output ratio of positive and negative ions according to the static electricity distribution;

[0011] The adaptive control module is used to monitor the error between the remaining static electricity signal after elimination and the target value in real time through closed-loop PID feedback and dynamically correct the output voltage parameter of the ion wind;

[0012] The data storage and traceability module is used to store and visualize the static electricity data, trend prediction results and ion wind parameters.

[0013] Preferably, the implementation process of denoising and reconstructing the signal through adaptive threshold is as follows:

[0014] S21. Normalize the original static electricity acquisition signal x(t):

[0015] where x(t) represents the original static electricity signal; μ x represents the mean value of the signal; σ x represents the standard value of the signal; x norm (t) represents the standardized signal;

[0016] S22. Process the non-stationary static electricity signal by combining improved empirical mode decomposition and adaptive threshold denoising, and decompose the non-stationary signal x norm (t) into several intrinsic mode function IMF components;

[0017] S23. Identify the high-frequency noise IMF i in the decomposed IMF components and perform denoising using the adaptive threshold function:

[0018]

[0019] where, θ i represents the adaptive threshold of the i-th IMF; represents the denoised IMF; sign() represents the sign function; λ represents the empirical adjustment factor; σ IMFi represents the standard deviation of the i-th IMF; L represents the signal length;

[0020] S2104. Reconstruct the signal by using the denoised IMF components together with the unprocessed medium and low frequency IMFs and the residual term r(t):

[0021] where, m is the number of IMFs determined to be noisy; represents the final denoised signal.

[0022] Preferably, the process of decomposing the non-stationary signal x norm (t) into several intrinsic mode function IMF components is as follows:

[0023] S31. Perform several processes on the standardized signal x norm (t), and add different independent white noises w k each time to obtain several noise-perturbed signals: x k (t) = x norm (t) + α·w k (t), k = 1, 2,..., K;

[0024] where, x k (t) represents adding noise for the k-th time; w k (t) represents the k-th group of white noise with zero mean and unit variance; α represents the noise perturbation amplitude; K represents the total number of iterations;

[0025] S32. Perform traditional empirical mode decomposition on each perturbed signal x k (t) to obtain IMF components:

[0026]

[0027] where, represents the i-th IMF in the k-th group; n k represents the number of IMFs obtained from the k-th decomposition; r k (t) represents the residual component of the k-th time;

[0028] S33. For each component serial number i, average and synthesize all the IMF components obtained from K decompositions:

[0029]

[0030] Among them, IMF i (t) represents the i-th intrinsic mode function; W k represents the weighting coefficient (by default, equal weight W k = 1 / K) in this embodiment;

[0031] S34. The final signal can be restored from multiple averaged IMFs and the residual term:

[0032]

[0033] Among them, IMF i (t) represents the i-th intrinsic mode function; r(t) represents the remaining trend term; n represents the number of IMFs obtained by decomposition.

[0034] Preferably, the generation process of the static trace record is as follows:

[0035] S41. Convert the processed electrostatic monitoring data into a binary data string data;

[0036] S42. Encode the processed electrostatic monitoring data data to obtain the encoded data Cd = H(data) ∈ GF(256) 2×2 ;

[0037] Among them, H() is a preset hash function, and this hash function H() maps the data to a 2×2 matrix; GF(256) 2×2 is a preset matrix field, that is, a 2×2 matrix is selected over the finite field GF(256);

[0038] S43. Calculate the static trace record Ra = B -1 ·A -1 ·Cd ∈ GF(256) 2×2 ;

[0039] Among them, A and B are preset randomly generated invertible matrices, A, B ∈ GF(256) 2×2 , and satisfy det(A) ≠ and det(B) ≠ 0, where det() is the determinant value corresponding to the matrix.

[0040] Preferably, the verification process to ensure the reliability of the received data through the hash verification of the static trace record is as follows:

[0041] S51. Convert the received electrostatic monitoring data x t into a binary string data data';

[0042] S52. Re-encode the data data' to obtain the encoded data Cd' = H(data');

[0043] S53. Calculate the check bit Cb = P·Ra;

[0044] where P is a preset parsing factor, P = A·B ∈ GF(256) 2×2 ;

[0045] S54. If Cb = Cd′, the check passes, indicating that the received static electricity data is reliable and the status record is accurate; otherwise, an alarm is immediately issued.

[0046] Preferably, the Res - GRU - Attn model structure is as follows:

[0047] Output layer: The embedded feature sequence within the time window, i.e., the embedded vector F t ;

[0048] GRU layer:

[0049] z t = σ(W z ·[h t-1 , F t )

[0050] r t = σ(W r ·[h t-1 , F t )

[0051]

[0052] where h t-1 represents the hidden state at the previous moment; z t represents the update gate; r t represents the reset gate; W z represents the weight matrix; σ represents the sigmoid activation function; W r represents the weight matrix; tanh() represents the hyperbolic tangent activation function; W h represents the weight matrix; h t represents the hidden state at the current moment; ⊙ represents element - wise multiplication;

[0053] Residual connection layer: Add the input directly to the GRU output:

[0054] where represents the residual output; W res represents the linear transformation matrix that maps the input to the hidden state dimension;

[0055] Attention layer: Calculate the attention weights for all hidden states within the time window to obtain the final weighted output

[0056]

[0057] e i = v T tanh(W a h i + b a );

[0058]

[0059] Among them, W a , v, b a represent the trainable parameters of the attention network; α i represents the attention weight, that is, the contribution degree of the hidden state h i at the i-th moment to the final output; e i represents the attention score, which is used to represent the importance of the hidden state h i in the entire input sequence; w represents the size of the time window, that is, how long to look back from the current moment t; j = t - w represents the starting moment of the time window, that is, pushing w moments forward from the current moment t to form a time series window;

[0060] Weighted fusion layer: Weightedly fuse the attention output and the residual result to obtain a fused representation

[0061]

[0062] Among them, γ represents the learnable fusion weight, γ ∈ [0, 1];

[0063] Output layer: Perform a non-linear mapping on the fused representation to predict the future static electricity value:

[0064]

[0065] Among them, represents the static electricity value predicted by the model, that is, the static electricity value at the moment t + 1; W o represents the weight matrix of the output layer; b o represents the bias term; ReLU() represents the ReLU activation function.

[0066] Preferably, the regulation process of regulating the positive and negative ion output ratio is as follows:

[0067] S71. Obtain the reference sliding window mean standard deviation trend prediction result and the dynamic risk adjustment coefficient k, and calculate the continuous control factor I ctrl :

[0068]

[0069] Among them, λ′ is an empirical adjustment parameter;

[0070] S72. Set the ion wind output voltage V ion and the output current I ion adjustment formula:

[0071] V ion = V0 + η·I ctrl ;

[0072] I ion = I0 + ζ·I ctrl ;

[0073] Among them, V0 represents the basic working voltage of the ion wind system; I0 represents the basic working current of the ion wind system; η represents the voltage adjustment coefficient; ζ represents the current adjustment coefficient; V ion and I ion respectively represent the ion wind system output voltage and current calculated at the current moment;

[0074] S73. Set the positive ion output ratio P ion and the negative ion output ratio N ion :

[0075]

[0076] N ion = 1 - P ion ;

[0077] Among them, φ represents the ratio adjustment coefficient.

[0078] Preferably, the correction process of dynamically correcting the ion wind output voltage parameter is as follows:

[0079] Define the electrostatic elimination error e(t): e(t) = x post (t) - x target ;

[0080] Among them, x post (t) is the remaining electrostatic signal x post (t) after elimination at the current moment t; x target represents the target electrostatic level;

[0081] PID adjustment output:

[0082] Update the ion wind output voltage V ion,new : V ion,new = V ion + u(t);

[0083] Among them, u(t) represents the PID regulation output; K p , K i , K d respectively represent the proportional, integral, and differential coefficients of the PID controller;

[0084] Preferably, the smoothing process of dynamic window smoothing based on the electrostatic accumulation rate is as follows:

[0085]

[0086]

[0087] Among them, S smooth (t) represents the finally processed smoothing signal of the output; β represents the adjustment coefficient; represents the electrostatic change rate; W t represents the smoothing window length; ΔS represents the change amplitude of the electrostatic signal within the time interval Δt.

[0088] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0089] The present invention designs an electrostatic monitoring and elimination system for cigarette package wrapping films based on big data analysis. It collects surface charge data of the wrapping film in real time through a distributed electrostatic sensor network, effectively processes non-stationary signals by combining improved empirical mode decomposition and adaptive threshold denoising algorithms, reduces noise interference and retains signal mutation characteristics, significantly improving data accuracy; uses an improved GRU network integrating residual connection and attention mechanism to perform time series modeling on multi-dimensional feature data, and combines a dynamic error correction mechanism to achieve high-precision electrostatic trend prediction; synergistically optimizes the output parameters of the ionic wind through the positive and negative ion ratio dynamic regulation equation and PID closed-loop feedback control to improve the electrostatic elimination efficiency; introduces the static tracing and recording technology based on finite field matrix coding, and ensures data accuracy and reliability through hash verification and matrix operations; the system integrates signal processing, intelligent prediction, dynamic control and data verification functions, overcomes the defects of traditional methods in non-stationary signal processing, prediction lag and insufficient elimination efficiency, and significantly improves the real-time performance, accuracy and reliability of electrostatic monitoring and neutralization in the cigarette package production process. Description of the Drawings

[0090] Figure 1 is the system architecture diagram of an electrostatic monitoring and elimination system for cigarette package wrapping films based on big data analysis proposed by the present invention. Detailed Embodiments

[0091] Example 1, as Figure 1 shown, an electrostatic monitoring and elimination system for cigarette package wrapping films based on big data analysis proposed by the present invention includes:

[0092] Electrostatic monitoring module: It adopts multiple-point electrostatic sensors, which are arranged at key points on the conveying path of the packaging film to collect electrostatic data on the surface of the packaging film in real time;

[0093] Electrostatic data processing module: It performs denoising and normalization processing on the collected electrostatic data, eliminates interference signals, and ensures data quality;

[0094] Data analysis and prediction module: Based on big data analysis and machine learning algorithms, it identifies electrostatic accumulation patterns and predicts potential electrostatic risks;

[0095] Electrostatic elimination module: Combining intelligent ion wind technology, it precisely releases positive and negative ions to achieve dynamic neutralization of static electricity;

[0096] Adaptive control module: According to real-time electrostatic monitoring data, it automatically adjusts electrostatic elimination parameters to improve elimination efficiency;

[0097] Data storage and traceability module: It establishes an electrostatic data cloud platform to store historical data and supports trend analysis and production optimization.

[0098] Embodiment 2, A static electricity monitoring and elimination system for cigarette pack packaging film based on big data analysis proposed by the present invention further includes a static electricity monitoring and elimination method for cigarette pack packaging film based on big data analysis, which is the working mode of a static electricity monitoring and elimination system for cigarette pack packaging film based on big data analysis proposed in Embodiment 1, and its specific implementation steps are as follows:

[0099] S1. The electrostatic monitoring module arranges N electrostatic sensors on the conveying path of the packaging film to form a distributed monitoring network, monitors the electrostatic data on the surface of the packaging film in real time, and transmits the electrostatic data x(t) to the electrostatic data processing module using wireless transmission technology (including but not limited to LoRa or Wi-Fi).

[0100] S2. The electrostatic data processing module performs denoising and normalization processing on the collected electrostatic data, and eliminates interference signals to ensure data quality. The specific implementation process is as follows:

[0101] S21. Combining improved empirical mode decomposition (IEEMD) with adaptive threshold denoising (ATD) breaks through the limitations of traditional filtering methods (such as moving average or median filtering) in dealing with non-stationary and non-linear noise, and achieves higher-precision fidelity denoising for the non-stationary and transient shock-containing charge signals in the electrostatic monitoring process of cigarette pack packaging film. Specifically:

[0102] S2101. Normalize the original electrostatic acquisition signal x(t) to eliminate the dimensional difference and the measurement point fluctuation difference:

[0103] Among them, x(t) represents the original electrostatic signal; μx represents the mean value of the signal; σ x represents the standard value of the signal; x norm (t) represents the normalized signal;

[0104] S2102. Using the IEEMD method, decompose the non-stationary signal x norm (t) into multiple Intrinsic Mode Function (IMF) components: The decomposition process is as follows:

[0105] (1) Process the normalized signal x norm (t) multiple times, each time adding different independent white noises w k , to obtain multiple noise-perturbed signals: x k (t) = x norm (t) + α·w k (t), k = 1, 2,..., K;

[0106] Among them, x k (t) represents the addition of noise for the k-th time; w k (t) represents the k-th group of white noise with zero mean and unit variance; α represents the noise perturbation amplitude; K represents the total number of iterations;

[0107] (2) Perform traditional EMD on each perturbed signal x k (t) to obtain a set of IMF components:

[0108]

[0109] Among them, IMF i (k) (t) represents the i-th IMF in the k-th group; n k represents the number of IMFs obtained from the k-th decomposition; r k (t) represents the residual component of the k-th time;

[0110] (3) For each component serial number i, average and synthesize all the IMF components obtained from K decompositions:

[0111]

[0112] Among them, IMF i (t) represents the i-th intrinsic mode function; W k represents the weighting coefficient (in this embodiment, the default equal weight W k = 1 / K);

[0113] (4) The final signal can be restored from multiple averaged IMFs and the residual term:

[0114]

[0115] Among them, IMF i (t) represents the i-th intrinsic mode function; r(t) represents the remaining trend term; n represents the number of IMFs obtained by decomposition;

[0116] S2103. Identify the high-frequency noise IMFs in the decomposed IMF vector i , and perform denoising using an adaptive threshold function:

[0117]

[0118] Among them, θ i represents the adaptive threshold of the i-th IMF; represents the denoised IMF; sign() represents the sign function; λ represents the empirical adjustment factor; represents the standard deviation of the i-th IMF; L represents the signal length;

[0119] S2104. Reconstruct the signal together with the denoised IMF components and the unprocessed medium- and low-frequency IMFs and the residual term r(t):

[0120] Among them, m is the number of IMFs determined to be noisy; represents the final denoised signal;

[0121] S22. Since the electrostatic signal has the characteristic of sudden accumulation, the smoothing process with a fixed window is likely to cause the loss of mutations. Therefore, a dynamic window smoothing based on the electrostatic accumulation rate is adopted, specifically as follows:

[0122]

[0123] Among them, S smooth (t) represents the finally processed smoothed signal output; β represents the adjustment coefficient; represents the electrostatic change rate; W t represents the smoothing window length, which is dynamically adjusted according to the electrostatic change rate; ΔS represents the change amplitude of the electrostatic signal within the time interval Δt, that is, the increment of the electrostatic signal value;

[0124] S23. Generate a static trace record for the processed electrostatic monitoring data, and its generation process is as follows:

[0125] S2301. Convert the processed electrostatic monitoring data into a binary data string data;

[0126] S2302. Encode the processed electrostatic monitoring data data to obtain the encoded data Cd = H(data) ∈ GF(256)2×2 ;

[0127] Among them, H() is a preset hash function, and this hash function H() maps data into a 2×2 matrix; GF(256) 2×2 is a preset matrix field, that is, a 2×2 matrix is selected over the finite field GF(256) (each element occupies 1 byte);

[0128] S2303. Calculate the static tracing standard record Ra = B -1 ·A -1 ·Cd ∈ GF(256) 2×2 ;

[0129] Among them, A and B are preset randomly generated invertible matrices, A, B ∈ GF(256) 2×2 , and satisfy det(A) ≠ and det(B) ≠ 0, where det() is the determinant value corresponding to the matrix;

[0130] S24. Transmit {the static tracing standard record Ra, the processed static electricity monitoring data x t} to the data analysis and prediction module.

[0131] S3. The data analysis and prediction module integrates multi-dimensional feature embedding + time series recursive modeling + dynamic error correction mechanism. When analyzing large-scale static electricity monitoring data of cigarette package wrapping films, it can not only accurately predict the future charge accumulation trend, but also identify abnormal change signals in advance, providing intelligent decision-making support for the adjustment of static electricity elimination strategies. The specific implementation process is as follows:

[0132] S31. Receive {the static tracing standard record Ra, the processed static electricity monitoring data x t}, extract the static tracing standard record Ra and the processed static electricity monitoring data x t from it. To ensure the reliability of static electricity data and the accuracy of status records, a verification is performed based on the static tracing standard record Ra. The verification process is as follows:

[0133] S3101. Convert the received static electricity monitoring data x t into binary string data data′;

[0134] S3102. Perform secondary encoding on the data data′ to obtain the encoded data Cd′ = H(data′);

[0135] S3103. Calculate the check bit Cb = P·Ra;

[0136] Among them, P is a preset parsing factor, P = A·B ∈ GF(256) 2×2 ;

[0137] S3204. If Cb = Cd', the verification passes, indicating that the received static electricity data is reliable and the status record is accurate; otherwise, an alarm is immediately issued.

[0138] S32. Extract the processed static electricity monitoring data and denote it as X = {x1, x2, …, x t , …, x T} (where x t represents the static electricity signal value at any sampling time point);

[0139] S33. Expand the static electricity detection data signal x t into an embedding vector F that includes temporal context features + process features + environmental features t :

[0140] where, Δx t = x t - x t-1 represents the signal difference (change rate); and respectively represent the mean and standard deviation within the sliding window W; T t and H t respectively represent the current temperature and humidity; S t represents the status of the production equipment (including but not limited to on / off, frequency coding information);

[0141] S34. Use an improved GRU network (Res-GRU-Attn) that introduces an attention mechanism and residual connections to perform trend prediction on the embedding sequence, and the model result is:

[0142] S3401. Output layer: The embedding feature sequence within the time window, that is, the embedding vector F t ;

[0143] S3402. GRU layer (basic temporal modeling):

[0144] z t = σ(W z · [h t-1 , F t );

[0145] r t = σ(W r · [h t-1 , F t );

[0146]

[0147] where, h t-1 represents the hidden state at the previous moment; z tRepresents the update gate, which controls whether to retain the hidden information of the previous moment at the current moment; r t Represents the reset gate, which controls whether to retain the hidden information of the previous moment at the current moment; W z Represents the weight matrix, which learns how to combine the current input and the hidden state of the previous moment; [h t-1 ,F t represents the input vector formed by concatenating the hidden state of the previous moment and the current input; σ represents the sigmoid activation function; W r Represents the weight matrix, which learns how to combine the current input and the hidden state of the previous moment; tanh() represents the hyperbolic tangent activation function; W h Represents the weight matrix, which learns how to combine the current input and the hidden state of the previous moment; r t ⊙h t-1 Represents the reset gate r t The adjusted hidden state of the previous moment, which controls the degree of information retention through element-wise multiplication; h t Represents the hidden state of the current moment; ⊙ represents element-wise multiplication, which controls the intensity of information flow;

[0148] S3403. To avoid information decay in time steps, add a residual channel and directly add the input to the GRU output:

[0149] Among them, Represents the residual output; W res Represents the linear transformation matrix that maps the input to the hidden state dimension;

[0150] S3404. Calculate the attention weights for all hidden states within the time window:

[0151]

[0152] e i = v T tanh(W a h i + b a );

[0153] Among them, W a , v, b a Represent the trainable parameters of the attention network; α i Represents the attention weight, that is, the contribution degree of the hidden state h i at the i-th moment to the final output; e i Represents the attention score, which is used to represent h iImportance in the entire input sequence; w represents the size of the time window, that is, how long to look back from the current moment t; j = t - w represents the starting moment of the time window, pushing w moments forward from the current moment t to form a time series window;

[0154] Accordingly: Obtain the final weighted output

[0155] S3405. Concatenate or weighted fuse the attention output and the residual result to obtain a fused representation

[0156]

[0157] Among them, γ represents the learnable fusion weight, γ ∈ [0, 1];

[0158] S3406. Perform a non-linear mapping on the fused representation to predict the future static electricity value:

[0159]

[0160] Among them, represents the static electricity value predicted by the model, that is, the static electricity value (including but not limited to voltage, charge) at the moment t + 1; W o represents the weight matrix of the output layer; b o represents the bias term; ReLU() represents the ReLU activation function;

[0161] S35. To address the problem of possible cumulative errors in static electricity prediction, introduce a historical error correction model (Dynamic Error Feedback):

[0162] Among them, represents the corrected predicted value; κ represents the error learning rate; ε t represents the prediction error at the current moment; represents the predicted value at the current moment;

[0163] S36. Utilize the trend prediction result Based on the dynamic threshold, conduct static electricity risk early warning:

[0164]

[0165] Among them, k represents the dynamic risk adjustment coefficient (taking the value of 3 in this embodiment); and respectively represent the reference moving window mean and standard deviation; RiskLevel represents the output static electricity risk level;

[0166] S37. Take Transmitted to the static elimination module.

[0167] S4. The static elimination module converts the static risk level RiskLevel into a continuous control factor, combines it with the ion wind output parameters, the positive and negative ion ratio, and the closed-loop PID regulation to achieve the full-chain intelligent control from prediction to physical elimination. The specific implementation process is as follows:

[0168] S41. Extract the mean value of the reference sliding window Standard deviation Trend prediction result And the dynamic risk adjustment coefficient k, and calculate a continuous control factor I ctrl , which reflects the relative amplitude of the static electricity exceeding the safety level and provides a quantitative basis for the subsequent control strategy:

[0169] Among them, λ′ is an empirical adjustment coefficient used to amplify the control factor and adjust the elimination response intensity;

[0170] Accordingly: Convert the risk level (high, medium, low) into a continuous numerical control factor I ctrl , which provides a quantitative basis for the parameter setting of the ion wind system to ensure that the subsequent elimination strategy is closely related to the actual risk;

[0171] S42. Set the adjustment formula for the output voltage V ion And the output current I ion :

[0172] V ion = V0 + η·I ctrl ;

[0173] I ion = I0 + ζ·I ctrl ;

[0174] Among them, V0 represents the basic working voltage of the ion wind system to ensure that basic ion output can still be maintained at low risk; I0 represents the basic working current of the ion wind system; η represents the voltage adjustment coefficient, which converts the control factor into a voltage increment; ζ represents the current adjustment coefficient, which converts the control factor into a current increment; V ion And I ion respectively represent the output voltage and current of the ion wind system calculated at the current moment;

[0175] Accordingly: Map the control factor calculated in the previous step directly to the physical output parameters of the ion wind, so that the ion wind output is stronger at high risk, thereby quickly neutralizing the excessive static electricity. The two groups of parameters (voltage and current) jointly regulate the balance of positive and negative ion release to ensure that both over-discharge is avoided and the purpose of static electricity neutralization is achieved during the elimination process;

[0176] S43. During the static electricity elimination process, in addition to adjusting the ion wind intensity, it is also necessary to regulate the output ratio of positive and negative ions. The output ratio is dynamically set according to the currently monitored static electricity distribution (higher positive or negative charge) so that the released ions can more precisely neutralize the target static electricity. Therefore, the positive ion output ratio P ion and the negative ion output ratio N ion are set as follows:

[0177]

[0178] N ion = 1 - P ion ;

[0179] where φ represents the ratio adjustment coefficient;

[0180] Accordingly: After the ion wind output parameters are determined, by finely adjusting the positive and negative ion ratio, precise correction of local static electricity imbalance is achieved, thereby improving the overall effect of static electricity elimination.

[0181] S5. The adaptive control module uses closed-loop feedback (PID control). It continuously monitors the remaining static electricity signal x post (t) after elimination and compares it with the target static electricity level x target . Then, the ion wind parameters are dynamically corrected through the PID controller. Specifically:

[0182] Define the static electricity elimination error e(t): e(t) = x post (t) - x target ;

[0183] PID regulation output:

[0184] Update the ion wind output voltage V ion,new : V ion,new = V ion + u(t);

[0185] where u(t) represents the PID regulation output; K p , K i , K d respectively represent the proportional, integral, and differential coefficients of the PID controller.

[0186] S6. The data storage and traceability module stores {static electricity data x(t), trend prediction results ion wind parameters {ion wind output voltage V ion , output current I ion , positive ion output ratio P ion , negative ion output ratio N ion , updated ion wind output voltage V ion,new}} stored in the built-in database, and based on the built-in visualization device, the static electricity data x(t) and the trend prediction results The ionic wind parameters {ionic wind output voltage V ion , output current I ion , positive ion output ratio P ion , negative ion output ratio N ion , updated ionic wind output voltage V ion,new} are visually displayed.

[0187] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art.

Claims

1. A static electricity monitoring and elimination system for cigarette package wrapping film based on big data analysis, characterized in that Including: An electrostatic monitoring module, which is used to monitor the electrostatic data on the surface of the packaging film in real time and transmit the electrostatic data to the electrostatic data processing module; An electrostatic data processing module, which is used to process non-stationary electrostatic signals by combining improved empirical mode decomposition and adaptive threshold denoising, denoise and reconstruct the signals through adaptive threshold, perform dynamic window smoothing based on the electrostatic accumulation rate, map the processed data into a finite field matrix through a hash function and generate a static trace record, and transmit {the static trace record, the processed data} to the data analysis and prediction module; A data analysis and prediction module, which is used to ensure the reliability of the received data through the hash check of the static trace record, predict the future charge trend by using a Res-GRU-Attn model that introduces an attention mechanism and residual connection, dynamically correct the model by combining historical errors to reduce the cumulative error, divide the electrostatic risk level based on the dynamic risk threshold and transmit it to the electrostatic elimination module; An electrostatic elimination module, which is used to generate a continuous control factor based on the electrostatic risk level, dynamically adjust the output voltage and current of the ion wind, and dynamically regulate the positive and negative ion output ratio according to the electrostatic distribution; An adaptive control module, which is used to monitor the error between the remaining electrostatic signal after elimination and the target value in real time through closed-loop PID feedback and dynamically correct the output voltage parameter of the ion wind; A data storage and traceability module, which is used to store and visualize the electrostatic data, trend prediction results and ion wind parameters.

2. The electrostatic monitoring and elimination system for cigarette package wrapping film based on big data analysis according to claim 1, characterized in that, The implementation process of denoising and reconstructing the signal through adaptive threshold is as follows: S21. Normalize the original electrostatic acquisition signal x(t): Among them, x(t) represents the original electrostatic signal; μ x represents the mean value of the signal; σ x represents the standard value of the signal; x norm (t) represents the signal after normalization; S22. Process the non-stationary electrostatic signal by combining improved empirical mode decomposition and adaptive threshold denoising, and decompose the non-stationary signal x norm (t) into several intrinsic mode function IMF components; S23. Identify the high-frequency noise IMF in the decomposed IMF components and denoise them using an adaptive threshold function: i , and use an adaptive threshold function for denoising: where, θ i represents the adaptive threshold of the i-th IMF; represents the denoised IMF; sign() represents the sign function; λ represents the empirical adjustment factor; represents the standard deviation of the i-th IMF; L represents the signal length; S2104. Reconstruct the signal by using the denoised IMF components together with the untreated medium and low frequency IMFs and the residual term r(t): Among them, m is the number of IMFs determined to be noisy; represents the final signal after denoising.

3. The electrostatic monitoring and elimination system for cigarette package wrapping film based on big data analysis according to claim 2, characterized in that, Decompose the non-stationary signal x norm (t) into a number of Intrinsic Mode Function (IMF) components as follows: S31. Perform several processes on the standardized signal x norm (t), and each time, superimpose different independent white noises w k , to obtain several noise-perturbed signals: x k (t) = x norm (t) + α·w k (t), k = 1, 2,..., K; where x k (t) represents the k-th addition of noise; w k (t) represents the white noise of the k-th group with zero mean and unit variance; α represents the noise perturbation amplitude; K represents the total number of iterations; S32. Perform conventional empirical mode decomposition on each perturbation signal x k (t) to obtain IMF components: Among them, IMF i (k) (t) represents the i-th IMF in the k-th group; n k represents the number of IMFs obtained from the k-th decomposition; r k (t) represents the residual component of the k-th time; S33. For each component serial number i, average and synthesize all the IMF components obtained by K decompositions: Among them, IMF i (t) represents the i-th intrinsic mode function; W k represents the weighting coefficient (in this embodiment, the equal weight W k = 1 / K is default); S34. The final signal can be restored from multiple averaged IMFs and the residual term: Among them, IMF i (t) represents the i-th intrinsic mode function; r(t) represents the remaining trend term; n represents the number of IMFs obtained by decomposition.

4. The electrostatic monitoring and elimination system for cigarette package wrapping film based on big data analysis according to claim 1, characterized in that, The generation process of the static trace record is: S41. Convert the processed electrostatic monitoring data into a binary data string data; S42. Encode the processed electrostatic monitoring data data to obtain the encoded data Cd = H(data) ∈ GF(256) 2×2 ; Among them, H() is a preset hash function, and this hash function H() maps data into a 2×2 matrix; GF(256) 2×2 is a preset matrix field, that is, a 2×2 matrix is selected over the finite field GF(256); S43. Calculate the static trace quasi-record Ra = B -1 ·A -1 ·Cd ∈ GF(256) 2×2 ; Among them, A and B are preset randomly generated invertible matrices, and A, B ∈ GF(256) 2×2 , and satisfy det(A) ≠ and det(B) ≠ 0, where det() is the determinant value corresponding to the matrix.

5. The electrostatic monitoring and elimination system for cigarette package wrapping film based on big data analysis according to claim 4, characterized in that The verification process of ensuring the reliability of the received data through the hash check of the static trace record is as follows: S51. Convert the received static electricity monitoring data x t into binary string data data′; S52. Re-encode the data data′ to obtain the encoded data Cd′ = H(data′); S53. Calculate the check bit Cb = P·Ra; Wherein, P is a preset parsing factor, and P = A·B ∈ GF(256) 2×2 ; S54. If Cb = Cd′, the verification passes, indicating that the received electrostatic data is reliable and the status record is accurate; otherwise, an alarm is immediately issued.

6. The electrostatic monitoring and elimination system for cigarette package wrapping film based on big data analysis according to claim 1, characterized in that, The structure of the Res-GRU-Attn model is: Output layer: The embedded feature sequence within the time window, i.e., the embedded vector F t ; GRU layer: z t = σ(W z · [h t-1 , F t ); r t = σ(W r · [h t-1 , F t ); where h t-1 represents the hidden state at the previous moment; z t represents the update gate; r t represents the reset gate; W z represents the weight matrix; σ represents the sigmoid activation function; W r represents the weight matrix; tanh() represents the hyperbolic tangent activation function; W h represents the weight matrix; h t represents the hidden state at the current moment; ⊙ represents element-wise multiplication; Residual connection layer: directly add the input to the GRU output: Among them, represents the residual output; W res represents the linear transformation matrix that maps the input to the hidden state dimension; Attention layer: Calculate the attention weights for all hidden states within the time window to obtain the final weighted output e i = v T tanh(W a h i + b a ); Among them, W a , v, b a represent the trainable parameters of the attention network; α i represents the attention weight, that is, the contribution degree of the hidden state h i at the i-th moment to the final output; e i represents the attention score, which is used to represent the importance of the hidden state h i in the entire input sequence; w represents the size of the time window, that is, how long to look back from the current moment t; j = t - w represents the starting moment of the time window, that is, pushing w moments forward from the current moment t to form a time series window; Weighted fusion layer: The attention output and the residual result are weighted and fused to obtain a fused representation Among them, γ represents the learnable fusion weight, γ ∈ [0,1]; Output layer: Perform a non-linear mapping on the fused representation to predict the future electrostatic value: Among them, represents the static electricity value predicted by the model, that is, the static electricity value at time t + 1; W o represents the weight matrix of the output layer; b o represents the bias term; ReLU() represents the ReLU activation function.

7. The electrostatic monitoring and elimination system for cigarette pack packaging film based on big data analysis according to claim 1, wherein The regulation process of regulating the positive and negative ion output ratio is as follows: S71. Obtain the mean value of the reference sliding window Standard deviation Trend prediction result And the dynamic risk adjustment coefficient k, calculate the continuous control factor I ctrl : Among them, λ′ is an empirical adjustment parameter; S72. Set the ion wind output voltage V ion and the output current I ion Adjustment formula: V ion = V0 + η·I ctrl ; I ion = I0 + ζ·I ctrl ; Among them, V0 represents the basic operating voltage of the ionic wind system; I0 represents the basic operating current of the ionic wind system; η represents the voltage regulation coefficient; ζ represents the current regulation coefficient; V ion and I ion respectively represent the output voltage and current of the ionic wind system calculated at the current moment; S73. Set the positive ion output ratio P ion and the negative ion output ratio N ion : N ion = 1 - P ion ; Among them, φ represents the ratio adjustment coefficient.

8. The electrostatic monitoring and elimination system for cigarette pack packaging film based on big data analysis according to claim 1, wherein, The correction process of dynamically correcting the output voltage parameter of the ion wind is as follows: Define the electrostatic elimination error e(t): e(t) = x post (t) - x target ; where x post (t) is the remaining static electricity signal x after elimination at the current moment t post (t); x target represents the target static electricity level; PID regulation output: Update the ionic wind output voltage V ion,new : V ion,new = V ion + u(t); Among them, u(t) represents the output of PID regulation; K p , K i , K d represent the proportional, integral, and derivative coefficients of the PID controller, respectively.

9. The electrostatic monitoring and elimination system for cigarette package wrapping film based on big data analysis according to claim 2, characterized in that, The smoothing process of performing dynamic window smoothing based on the electrostatic accumulation rate is as follows: Among them, S smooth (t) represents the smoothed signal of the final processed output; β represents the adjustment coefficient; represents the electrostatic change rate; W t represents the smoothing window length; ΔS represents the change amplitude of the electrostatic signal within the time interval Δt.

Images