Environmental protection evaluation method and system based on integrated printer operation parameters

Abstract

The invention provides an environmental protection evaluation method and system based on operation parameters of an integrated printer, and relates to the technical field of green printing, and the method comprises the steps: inputting the operation parameters of the integrated printer into a neural network algorithm, calculating environmental protection index scores of VOCs, wastewater, energy consumption, solid waste and the like, and generating a printing technology optimization suggestion based on the environmental protection index scores. And the self-adaptive fuzzy PID control algorithm is adopted to automatically adjust the operation parameters, and it is ensured that the environmental protection index score reaches the preset threshold value, so that environmental protection optimization of the printing process is achieved, the resource utilization efficiency is improved, and environmental pollution is reduced.

Description

Technical Field

[0001] The present invention relates to green printing technology, and in particular to an environmental protection evaluation method and system based on integrated printing press operating parameters. Background Art

[0002] As an important part of modern printing technology, the optimization of the operating parameters of integrated printing presses directly affects the achievement of environmental protection indicators. The existing environmental assessment methods have the following shortcomings: Traditional environmental assessment methods often rely on manual data collection and analysis, which is inefficient and difficult to adapt to the rapidly changing production environment, resulting in insufficient real-time and accuracy of environmental indicators.

[0003] Existing technologies lack intelligent means for optimizing printing processes and are unable to automatically generate optimization suggestions based on real-time data, resulting in the failure to effectively control environmental indicators during the production process.

[0004] Most existing control algorithms are static adjustments and cannot be dynamically adjusted according to changes in environmental protection indicators. As a result, the operating parameters of the integrated printing press cannot adapt to changes in environmental protection requirements in a timely manner, affecting the overall environmental protection effect. Summary of the Invention

[0005] The embodiments of the present invention provide an environmental assessment method and system based on integrated printing press operating parameters, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention, Provides an environmental assessment method based on integrated printing press operating parameters, including: Using the collected operating parameters of the integrated printing press as input layer nodes of a neural network algorithm, and calculating an environmental index score based on the neural network algorithm, the environmental index score including a VOCs emission index score, a wastewater discharge index score, an energy consumption index score, and a solid waste generation index score; generating printing process optimization suggestions based on the environmental index scores; Based on the printing process optimization suggestion, an adaptive fuzzy PID control algorithm is used to automatically adjust the operating parameters of the integrated printing press, including: establishing an adaptive fuzzy PID controller with an environmental index score deviation value as an input and an operating parameter adjustment value as an output; fuzzifying the environmental index score deviation value and the deviation change rate, respectively, and obtaining a fuzzy reasoning result based on a preset fuzzy rule library; defuzzifying the fuzzy reasoning result using a center of gravity method, and dynamically adjusting the PID controller; and converting the defuzzified output into an adjustment instruction for the operating parameter; After completing the parameter adjustment, the operating parameters are collected again, and the environmental protection index scores are calculated and evaluated until the environmental protection index scores all reach corresponding preset thresholds.

[0007] In an optional embodiment, The steps of calculating the environmental protection index score based on the neural network algorithm include: Mapping the operating parameter to a range from negative one to positive one using a mapping function; Construct a neural network structure with three hidden layers. The first hidden layer includes fourteen neurons and uses the ReLU activation function. The second hidden layer includes ten neurons and uses the ReLU activation function. The third hidden layer includes six neurons and uses the ReLU activation function. The output layer includes four output nodes and uses the Sigmoid activation function. According to the operating parameters, the environmental protection index score is calculated by the neural network structure, wherein: The VOCs emission index scores are: ; w T 、w H are the weight coefficients of temperature parameter and ink layer thickness parameter respectively, 、 These are the influence functions of temperature and ink layer thickness on VOCs, T is the temperature parameter, and H is the ink layer thickness parameter; Wastewater discharge indicators are calculated based on paper moisture content and printing pressure, energy consumption indicators are calculated based on ink roller speed and ink supply, and solid waste generation indicators are calculated based on printing speed; The neural network structure is trained using a comprehensive loss function, wherein the comprehensive loss function includes an environmental index prediction loss term, a parameter coupling loss term, a time series continuity loss term, and a process constraint loss term; The calculation results of the environmental index scores are evaluated in real time, including: data validity verification and outlier processing of the operating parameters; calculation of confidence intervals of the environmental index scores; and retraining the neural network structure when the prediction error of any environmental index score exceeds a preset error threshold.

[0008] In an optional embodiment, The steps of the comprehensive loss function including the environmental index prediction loss term, the parameter coupling loss term, the time series continuity loss term and the process constraint loss term include: The environmental index prediction loss term adopts a piecewise function structure. When the prediction error is less than a first preset threshold, a quadratic loss is used for calculation; when the prediction error is greater than or equal to the first preset threshold, a linear loss is used for calculation; the parameter coupling loss term is calculated based on a first coupling constraint between printing speed and ink layer thickness, a second coupling constraint between printing pressure and paper moisture content, and a third coupling constraint between ink roller speed and ink supply; the temporal continuity loss term is calculated based on the difference between the environmental index prediction values at adjacent moments; the process constraint loss term is calculated based on the physical limitations of the operating parameters. When the predicted value of the operating parameter exceeds the corresponding feasible interval, a constraint penalty value is calculated based on the distance between the predicted value of the operating parameter and the boundary of the feasible interval; During the training process of the neural network structure, an adaptive weight adjustment mechanism is used to dynamically adjust the weight coefficients of each loss term in the comprehensive loss function.

[0009] In an optional embodiment, The steps of dynamically adjusting the weight coefficients of the loss items in the comprehensive loss function using an adaptive weight adjustment mechanism include: Construct a time series feature matrix to record the historical sequence of loss items, and calculate weighted features through the attention mechanism; establish a memory matrix to store historical weight configurations and update based on similarity; calculate the loss item gradient correlation matrix to generate an adjustment vector; input the attention features, memory information, and correlation vector into the GRU network, selectively process them through the gating mechanism, and dynamically generate the loss item weight coefficient, specifically including: Construct a time series feature matrix for each loss item, wherein the time series feature matrix includes a historical value sequence of each loss item within a preset time window; calculate an attention score based on the time series feature matrix, and generate an attention weighted feature based on the attention score, wherein the attention weighted feature is obtained by calculating the similarity between the query matrix and the key matrix; construct a memory matrix to store historical weight configuration information, wherein each memory unit in the memory matrix includes a weight configuration vector, a corresponding loss value, and a timestamp information; update the memory matrix based on the similarity between the current weight configuration and the historical memory unit in the memory matrix; calculate a gradient correlation matrix between loss items, wherein the gradient correlation matrix is calculated based on the inner product and norm of the gradient vector of each loss item, and generate a correlation adjustment vector based on the gradient correlation matrix; The attention weighted features, the updated information of the memory matrix, and the correlation adjustment vector are input into a weight generation network constructed based on a gated recurrent unit. The weight generation network selectively processes the input features through an update gate and a reset gate to generate weight coefficients for each loss term. An adaptive learning rate is calculated based on historical weight changes, and the weight coefficients for each loss term are updated in combination with the adaptive learning rate. The parameters of the neural network structure are optimized and updated based on the calculation result of the comprehensive loss function until the calculation result of the comprehensive loss function is less than a preset score threshold.

[0010] In an optional embodiment, An adaptive fuzzy PID controller is established with the environmental index score deviation value as input and the operating parameter adjustment value as output; the environmental index score deviation value and the deviation change rate are fuzzified respectively, and a fuzzy reasoning result is obtained according to a preset fuzzy rule base; the fuzzy reasoning result is defuzzified using the center of gravity method. The steps of dynamically adjusting the PID controller include: Collect historical data sequences of environmental index scores; process the historical data sequences using an LSTM network with a differential attention layer, enhance mutation characteristics by calculating the weight distribution of the environmental index score difference sequence, generate prediction trajectories and prediction confidence, and set three-level warning thresholds, including mild, moderate, and severe; calculate the deviation value and deviation change rate between the predicted trajectory and the actual environmental index score; divide the value ranges of the deviation value and deviation change rate into multi-level fuzzy subsets and establish corresponding membership functions; fuzzify the deviation value and deviation change rate based on the membership function to obtain the corresponding fuzzy quantity; Constructing a fuzzy rule base, the fuzzy rule base including a correspondence between a combination of the deviation value and the deviation change rate and an increment of a PID controller parameter, the fuzzy rule base setting a control parameter adjustment strategy based on a relationship between the deviation value and the three-level warning threshold and an absolute value of the deviation change rate; Based on the fuzzy quantity, the activation degree of each fuzzy rule is calculated, and a fuzzy reasoning operation is performed to obtain a comprehensive fuzzy reasoning result; the fuzzy reasoning result is defuzzified by using the center of gravity method to obtain a PID controller parameter increment for dynamically optimizing control performance; the control parameter is updated in real time according to the PID controller parameter increment and the control quantity is calculated; a feedforward compensation term is introduced, the feedforward compensation term is calculated based on the second-order derivative of the deviation value, and the feedforward compensation term is superimposed on the control quantity to obtain a total control quantity; the operating parameter is adjusted based on the total control quantity, and the adjustment amount of the operating parameter is proportional to the total control quantity.

[0011] In an optional embodiment, The step of setting the adjustment strategy of the control parameter based on the relationship between the deviation value and the three-level warning threshold and the absolute value of the deviation change rate by the fuzzy rule base includes: An adjustment strategy is determined according to a preset fuzzy rule base, wherein when the current deviation value is greater than the severe warning threshold and the absolute value of the deviation change rate is less than the second preset threshold, the integral coefficient is adjusted to a preset maximum value; when the current deviation value is between the medium warning threshold and the severe warning threshold and the absolute value of the deviation change rate is less than the second preset threshold, the integral coefficient is increased according to a first preset ratio; when the current deviation value is less than the minor warning threshold and the absolute value of the deviation change rate is greater than the second preset threshold, the differential coefficient is adjusted to a preset maximum value; when the current deviation value is between the minor warning threshold and the medium warning threshold and the absolute value of the deviation change rate is greater than the second preset threshold, the differential coefficient is increased according to a second preset ratio; Dynamically adjusting the third-level warning threshold based on the accuracy and prediction confidence of the predicted trajectory; The proportional coefficient, integral coefficient and differential coefficient of the PID controller are updated according to the preset PID controller parameter increment, the prediction error of the predicted trajectory is calculated, the network parameters of the LSTM network are optimized, and the predicted trajectory and prediction confidence are regenerated using the optimized network parameters.

[0012] In an optional embodiment, Introducing a feedforward compensation term, wherein the feedforward compensation term is calculated based on the second-order derivative of the deviation value, comprises the following steps: Construct multi-timescale window data, calculate high-frequency features through derivatives in the short term, extract frequency and periodic features in the medium term, and use Transformer to predict trends in the long term. Calculate performance indicators based on the integrated squared error of each time scale, and use exponential normalization to obtain compensation weights. Update compensation parameters through dynamic step size factor optimization, specifically including: Construct short-term time window data, medium-term time window data, and long-term time window data containing environmental protection index scores and operating parameters; calculate the first-order derivative and second-order derivative of the environmental protection index score based on the short-term time window data, obtain the high-frequency component feature vector, and calculate the short-term compensation amount; perform spectrum analysis and pattern recognition on the medium-term time window data, extract the dominant frequency characteristics and periodic characteristics, and calculate the medium-term compensation amount; use the Transformer network based on the self-attention mechanism to perform time series prediction on the long-term time window data, obtain trend prediction results and prediction confidence, and calculate the long-term compensation amount; The integrated square error of the environmental protection index score is calculated in the short-term evaluation time interval, the medium-term evaluation time interval, and the long-term evaluation time interval to obtain the corresponding performance index; the compensation weight of each time scale is calculated based on the performance index using the exponential normalization method, the sum of the compensation weights is 1 and is inversely proportional to the corresponding performance index; the compensation amount of each time scale is multiplied by the corresponding compensation weight and the sum is obtained to obtain the feedforward compensation amount; A corresponding dynamic step factor is calculated based on the performance index of each time scale, and the dynamic step factor decays exponentially with the increase of the performance index; the compensation parameters of each time scale are dynamically optimized and updated using the dynamic step factor, and the feedforward compensation amount is recalculated based on the updated compensation parameters.

[0013] According to a second aspect of the embodiments of the present invention, Provide an environmental assessment system based on integrated printing press operating parameters, including: The first unit is configured to use the collected operating parameters of the integrated printing press as input layer nodes of a neural network algorithm, and calculate an environmental index score based on the neural network algorithm, wherein the environmental index score includes a VOCs emission index score, a wastewater discharge index score, an energy consumption index score, and a solid waste generation index score; The second unit is used to generate printing process optimization suggestions based on the environmental index score; The third unit is configured to automatically adjust the operating parameters of the integrated printing press using an adaptive fuzzy PID control algorithm based on the printing process optimization suggestion, including: establishing an adaptive fuzzy PID controller with an environmental index score deviation value as an input and an operating parameter adjustment value as an output; fuzzifying the environmental index score deviation value and the deviation change rate, respectively, and obtaining a fuzzy inference result based on a preset fuzzy rule base; defuzzifying the fuzzy inference result using a center of gravity method and dynamically adjusting the PID controller; and converting the defuzzified output into an adjustment instruction for the operating parameter; The fourth unit is used to re-collect the operating parameters after completing the parameter adjustment, and calculate and evaluate the environmental protection index scores until the environmental protection index scores all reach the corresponding preset thresholds.

[0014] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0015] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0016] By combining the operating parameters of the integrated printing press with a neural network algorithm, the present invention can achieve accurate assessment of environmental indicators, thereby effectively reducing emissions of VOCs, wastewater, energy consumption, and solid waste, and improving environmental protection levels; the printing process optimization suggestions generated based on the environmental indicator scores can provide enterprises with scientific improvement directions, promote the green transformation of the production process, reduce production costs, and improve resource utilization efficiency.

[0017] The application of the adaptive fuzzy PID control algorithm of the present invention enables the operating parameters of the integrated printing press to be adjusted dynamically in real time, ensuring that the environmental index score continues to meet the preset threshold, thereby enhancing the stability and environmental compliance of the production process.

[0018] The present invention achieves rapid response and precise control of environmental indicator fluctuations through multi-time-scale collaborative analysis and second-order derivative feedforward compensation, combined with dynamic weight allocation and Transformer network trend prediction, effectively reducing the risk of exceeding the standard and improving the system steady-state precision and prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Schematic diagram of a flow chart of an environmental assessment method based on integrated printing press operating parameters according to an embodiment of the present invention; Figure 2 This is a schematic structural diagram of an environmental assessment system based on integrated printing press operating parameters according to an embodiment of the present invention; Figure 3 This is the adaptive fuzzy PID control flow chart of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0021] The following combination Figure 1 、 Figure 2 and Figure 3 The technical solution of the present invention is described in detail with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0022] Figure 1 FIG. 1 is a flow chart of an environmental assessment method based on integrated printing press operating parameters according to an embodiment of the present invention. Figure 1 As shown, the method includes: S1. Using the collected operating parameters of the integrated printing press as input layer nodes of a neural network algorithm, and calculating an environmental index score based on the neural network algorithm, the environmental index score includes a VOCs emission index score, a wastewater discharge index score, an energy consumption index score, and a solid waste generation index score; S2. generating printing process optimization suggestions based on the environmental index score; S3. Based on the printing process optimization suggestion, an adaptive fuzzy PID control algorithm is used to automatically adjust the operating parameters of the integrated printing press, including: establishing an adaptive fuzzy PID controller with the environmental index score deviation as input and the operating parameter adjustment as output; fuzzifying the environmental index score deviation and the deviation change rate, respectively, and obtaining a fuzzy inference result based on a preset fuzzy rule base; defuzzifying the fuzzy inference result using a center of gravity method and dynamically adjusting the PID controller; and converting the defuzzified output into an adjustment instruction for the operating parameter; S4. After completing the parameter adjustment, the operating parameters are collected again, and the environmental protection index scores are calculated and evaluated until the environmental protection index scores all reach the corresponding preset thresholds.

[0023] For example, parameters are collected during the operation of an integrated printing press, with a sampling period of 2 seconds. The operating parameters include printing speed, printing pressure, ink layer thickness, paper moisture content, printing temperature, ink roller speed, and ink supply. Specifically, the collected data ranges from 6,000 to 8,000 sheets / hour, printing pressure from 0.2 to 0.4 MPa, ink layer thickness from 0.8 to 1.2 microns, paper moisture content from 6% to 8%, printing temperature from 20 to 28°C, ink roller speed from 150 to 200 rpm, and ink supply from 25 to 35 mL / min. These parameters are collected in real time using high-precision sensors to establish an operating parameter database.

[0024] Constructing deep neural networks for environmental indicators evaluation.

[0025] Thresholds are determined and optimization recommendations are generated for the environmental indicator assessment results. The VOCs emission index score threshold is set at 85 points. When the score falls below the threshold, the printing temperature is lowered from 26°C to 23°C, the ink layer thickness is reduced from 1.1 microns to 0.8 microns, and the measured VOCs concentration is reduced by 32%. The wastewater discharge index score threshold is set at 80 points. When the score falls below the threshold, the paper moisture content testing interval is shortened from 30 minutes to 10 minutes, and the printing pressure is adjusted from 0.35 MPa to 0.28 MPa, reducing wastewater generation by 28%. The energy consumption index score threshold is set at 75 points. When the score falls below the threshold, the ink roller speed is reduced from 180 rpm to 160 rpm, and the ink supply is adjusted from 32 mL / min to 28 mL / min, reducing overall energy consumption by 21%. The solid waste generation index score threshold is set at 78 points. When the score falls below the threshold, the printing speed is maintained at a stable level of 7,200 sheets / hour, reducing the scrap rate by 25%.

[0026] Implement an adaptive fuzzy PID controller for automatic parameter adjustment. The environmental index score deviation and rate of change are categorized into seven levels: large negative, medium negative, small negative, zero, small positive, medium positive, and large positive. A fuzzy rule library containing 49 rules is constructed based on expert experience. For example, when the deviation is large negative and the rate of change is large negative, the proportional coefficient is increased by 50%, the integral time constant is increased by 40%, and the differential time constant is decreased by 30%. Defuzzification is performed using the center of gravity method to determine the adjustment values for the PID controller parameters.

[0027] After adjustment by the controller, the system monitors and evaluates the operating parameters in real time.

[0028] The invention has strong adaptability and stability. In printing tasks of different batches and specifications, the fluctuation range of environmental index scores is controllable, and various parameters are always maintained in the optimal operating range, achieving green and efficient operation of the printing process.

[0029] In an optional embodiment, The steps of calculating the environmental protection index score based on the neural network algorithm include: Mapping the operating parameter to a range from negative one to positive one using a mapping function; Construct a neural network structure with three hidden layers. The first hidden layer includes fourteen neurons and uses the ReLU activation function. The second hidden layer includes ten neurons and uses the ReLU activation function. The third hidden layer includes six neurons and uses the ReLU activation function. The output layer includes four output nodes and uses the Sigmoid activation function. According to the operating parameters, the environmental protection index score is calculated by the neural network structure, wherein: The VOCs emission index scores are: ; w T 、w H are the weight coefficients of temperature parameter and ink layer thickness parameter respectively, 、 These are the influence functions of temperature and ink layer thickness on VOCs, T is the temperature parameter, and H is the ink layer thickness parameter; Wastewater discharge indicators are calculated based on paper moisture content and printing pressure, energy consumption indicators are calculated based on ink roller speed and ink supply, and solid waste generation indicators are calculated based on printing speed; The neural network structure is trained using a comprehensive loss function, wherein the comprehensive loss function includes an environmental index prediction loss term, a parameter coupling loss term, a time series continuity loss term, and a process constraint loss term; The calculation results of the environmental index scores are evaluated in real time, including: data validity verification and outlier processing of the operating parameters; calculation of confidence intervals of the environmental index scores; and retraining the neural network structure when the prediction error of any environmental index score exceeds a preset error threshold.

[0030] For example, a mapping function is used to convert the operating parameters to [-1, 1] to ensure the normalization of the input data. This process is achieved through linear transformation, ensuring that parameters of different dimensions can be effectively compared and processed within the same range. In this way, the neural network can better learn and adapt to the characteristics of the input data.

[0031] A neural network architecture with three hidden layers was constructed. The first hidden layer contained fourteen neurons and used the ReLU activation function to introduce nonlinear features. The second hidden layer contained ten neurons and also used the ReLU activation function. The third hidden layer contained six neurons and also used the ReLU activation function. The output layer had four output nodes and used the Sigmoid activation function to facilitate multi-classification problems. This structural design aims to improve the model's ability to fit complex data through multi-level feature extraction.

[0032] Based on the operating parameters, a neural network structure calculates environmental performance scores. Specifically, the VOC emission score calculation relies on weighted coefficients for temperature and ink thickness. The impact of temperature and ink thickness on VOCs is quantified using specific influence functions. This process ensures that the model accurately reflects the impact of different parameters on environmental performance.

[0033] The wastewater discharge index score is calculated based on the paper moisture content and printing pressure. The paper moisture content and printing pressure are converted through the moisture content influence function and pressure influence function respectively, and then multiplied by the corresponding weight coefficients and summed. The energy consumption index score is calculated based on the ink roller speed and ink supply. The ink roller speed and ink supply are converted through the speed influence function and ink supply influence function respectively, and then multiplied by the corresponding weight coefficients and summed. The solid waste generation index score is calculated based on the printing speed. The printing speed is converted through the speed influence function and multiplied by the corresponding weight coefficient. The temperature influence function is implemented using a segmented mapping approach: when the temperature is below the ideal process temperature, it increases with increasing temperature; when the temperature is within the ideal process temperature range, it maintains the optimal influence value; and when the temperature exceeds the ideal process temperature, it decreases with increasing temperature. Similarly, the ink layer thickness influence function f_H(H) employs a similar segmented mapping approach. The moisture content influence function increases when the paper moisture content is too low, remains stable when the moisture content is moderate, and decreases when the moisture content is too high. The pressure influence function increases when the printing pressure is low, remains stable when the pressure is moderate, and decreases when the pressure is too high. The speed influence function reflects the relationship between ink roller speed and energy consumption: it changes slowly in the low speed range, increases approximately linearly in the medium speed range, and increases rapidly in the high speed range. The ink supply quantity influence function represents the relationship between ink supply quantity and energy consumption, showing a nearly exponential growth trend. The speed impact function describes the impact of printing speed on solid waste generation. It changes slowly in the low-speed range, remains relatively stable in the medium-speed range, and rises rapidly in the high-speed range, reflecting the inherent correlation between speed and scrap rate. These impact functions are fitted by historical production data and can effectively reflect the actual correspondence between various process parameters and environmental indicators.

[0034] A comprehensive loss function is used to train the neural network. This loss function includes environmental indicator prediction loss, parameter coupling loss, time series continuity loss, and process constraint loss. Through multi-dimensional loss evaluation, it is ensured that the model can take into account the optimization of various indicators during the training process and improve the overall performance. The Adam optimizer is used for training the neural network structure. The initial learning rate of the Adam optimizer is 0.001 and the batch size is 32. Dropout is set during training to prevent overfitting. The Dropout rate is 0.3. At the same time, L2 regularization is used, and the L2 regularization coefficient is 0.01. The calculated environmental indicator scores are evaluated in real time. First, the operating parameters are validated and outliers are processed to ensure the reliability of the input data. Next, confidence intervals are calculated for the environmental indicator scores to assess the credibility of the predictions. If the prediction error for any environmental indicator score exceeds a preset error threshold, the system automatically triggers a retraining of the neural network structure to improve the model's accuracy and stability.

[0035] The printing process optimization recommendations in this invention refer to specific technical solutions generated based on the results of environmental indicator score analysis, used to guide the adjustment of integrated printing press operating parameters. These solutions include parameter optimization target values, adjustment directions, and adjustment ranges for different environmental indicators. The printing process optimization recommendations are derived by comparing environmental indicator scores with preset thresholds to identify environmental indicators requiring optimization. For example, when the VOCs emission index score falls below a preset threshold (e.g., 85), VOCs emissions are marked as requiring optimization; when the wastewater discharge index score falls below a preset threshold (e.g., 80), wastewater discharge is also marked as requiring optimization, and so on. For each environmental indicator requiring optimization, the key influencing parameters and their optimization targets are determined based on historical production data. For VOCs emissions, the primary influencing parameters are printing temperature and ink layer thickness, and the optimization targets are lowering printing temperature and reducing ink layer thickness. For wastewater discharge, the primary influencing parameters are paper moisture content and printing pressure, and the optimization targets are optimizing the paper moisture content testing frequency and reducing printing pressure. For energy consumption, the main influencing parameters are the roller speed and ink supply, and the optimization direction is to reduce the roller speed and optimize the ink supply. For solid waste generation, the main influencing parameter is printing speed, and the optimization direction is to stabilize the printing speed.

[0036] Based on the difference between the environmental index score and the preset threshold, combined with the results of parameter sensitivity analysis, the adjustment range for each parameter is calculated. The adjustment range calculation formula is: Adjustment range = Basic adjustment amount × (Preset threshold - Actual score) / Preset threshold × Parameter sensitivity coefficient. The basic adjustment amount is a pre-set parameter adjustment baseline value based on production experience (for example, the basic adjustment amount for temperature parameters is ±3°C, and the basic adjustment amount for ink layer thickness is ±0.2 microns). The parameter sensitivity coefficient reflects the impact of parameter changes on environmental indicators and is calculated as the ratio of the parameter change rate to the environmental index score change rate.

[0037] The resulting printing process optimization recommendations include the environmental indicators to be optimized and their current scores; the current values, target values, and adjustment directions of each influencing parameter; the priority and magnitude of parameter adjustments; and the expected degree of improvement in environmental performance. For example, when the VOCs emission index score is 75 (below the preset threshold of 85), the generated printing process optimization recommendations include reducing the printing temperature from 26°C to 23°C (an 11.5% reduction) and reducing the ink layer thickness from 1.1 microns to 0.8 microns (a 27.3% reduction), with an expected VOCs concentration reduction of approximately 32%.

[0038] The generated printing process optimization suggestions will serve as the input of the adaptive fuzzy PID control algorithm to automatically adjust the operating parameters of the integrated printing press to achieve the optimization goal of environmental protection indicators.

[0039] The present invention improves the adaptability of the neural network to input data by standardizing operating parameters, thereby enhancing the model's learning ability and prediction accuracy. The multi-level neural network structure effectively extracts features from complex data, improves the calculation accuracy of environmental indicators, and ensures the reliability of evaluation results. The comprehensive loss function takes into account the optimization of various indicators and improves overall performance. The real-time evaluation mechanism and automatic retraining function ensure the continuous optimization of the model in practical applications, enabling timely response to environmental changes and improving the environmental protection level of the production process.

[0040] In an optional embodiment, The steps of the comprehensive loss function including the environmental index prediction loss term, the parameter coupling loss term, the time series continuity loss term and the process constraint loss term include: The environmental index prediction loss term adopts a piecewise function structure. When the prediction error is less than a first preset threshold, a quadratic loss is used for calculation; when the prediction error is greater than or equal to the first preset threshold, a linear loss is used for calculation; the parameter coupling loss term is calculated based on a first coupling constraint between printing speed and ink layer thickness, a second coupling constraint between printing pressure and paper moisture content, and a third coupling constraint between ink roller speed and ink supply; the temporal continuity loss term is calculated based on the difference between the environmental index prediction values at adjacent moments; the process constraint loss term is calculated based on the physical limitations of the operating parameters. When the predicted value of the operating parameter exceeds the corresponding feasible interval, a constraint penalty value is calculated based on the distance between the predicted value of the operating parameter and the boundary of the feasible interval; During the training process of the neural network structure, an adaptive weight adjustment mechanism is used to dynamically adjust the weight coefficients of each loss term in the comprehensive loss function.

[0041] For example, the environmental indicator prediction loss term is designed using a piecewise function structure. Specifically, when the prediction error is less than a preset first threshold, a quadratic loss calculation method is used to ensure that the model is penalized less severely within a small error range, thereby encouraging the model to make more refined adjustments within this range. When the prediction error is greater than or equal to the first threshold, a linear loss calculation method is used instead, imposing a greater penalty on the model as the error increases, prompting the model to make necessary adjustments.

[0042] Secondly, the calculation of the parameter coupling loss term is based on multiple coupling constraints. Specifically, the coupling relationship between printing speed and ink layer thickness is first considered to ensure their coordination during the production process. Secondly, the coupling constraint between printing pressure and paper moisture content is also taken into account to avoid print quality issues caused by improper pressure. Finally, the coupling relationship between ink roller speed and ink supply is equally important to ensure stable and uniform ink supply. By integrating these coupling constraints, losses caused by parameter incompatibility can be effectively reduced.

[0043] Next, the temporal continuity loss term is calculated based on the difference between the environmental indicator predictions at adjacent moments. By comparing the predictions at adjacent moments, we can assess the model's stability and continuity over the time series, ensuring consistency across the time dimension and avoiding sudden changes or unreasonable fluctuations in predictions.

[0044] Finally, the process constraint penalty term is calculated based on the physical limitations of the operating parameters. When the predicted value of an operating parameter exceeds the feasible interval, the distance between the predicted value and the feasible interval boundary is calculated and used as the constraint penalty. This mechanism ensures that the model output is feasible in actual production, avoiding production accidents or waste of resources caused by inaccurate predictions.

[0045] During the training of the neural network structure, an adaptive weight adjustment mechanism is used to dynamically adjust the weight coefficients of each loss term in the comprehensive loss function. By monitoring the performance of each loss term in real time, its weight can be flexibly adjusted to better balance the impact of each loss term during training, ultimately improving the overall performance of the model.

[0046] The present invention designs environmental index prediction loss terms with a piecewise function structure, which can flexibly adjust the penalty intensity of the model within different error ranges, thereby improving the prediction accuracy and stability of the model; the introduction of parameter coupling loss terms ensures the coordination between various parameters in the production process, reduces the losses caused by parameter incoordination, and improves production efficiency and product quality; the adaptive mechanism of dynamically adjusting the weight of the loss term enables the model to be optimized in real time during the training process, improves the adaptability and generalization ability of the model, and ultimately achieves better prediction results.

[0047] In an optional embodiment, The steps of dynamically adjusting the weight coefficients of the loss items in the comprehensive loss function using an adaptive weight adjustment mechanism include: Construct a time series feature matrix to record the historical sequence of loss items, and calculate weighted features through the attention mechanism; establish a memory matrix to store historical weight configurations and update based on similarity; calculate the loss item gradient correlation matrix to generate an adjustment vector; input the attention features, memory information, and correlation vector into the GRU network, selectively process them through the gating mechanism, and dynamically generate the loss item weight coefficient, specifically including: Construct a time series feature matrix for each loss item, wherein the time series feature matrix includes a historical value sequence of each loss item within a preset time window; calculate an attention score based on the time series feature matrix, and generate an attention weighted feature based on the attention score, wherein the attention weighted feature is obtained by calculating the similarity between the query matrix and the key matrix; construct a memory matrix to store historical weight configuration information, wherein each memory unit in the memory matrix includes a weight configuration vector, a corresponding loss value, and a timestamp information; update the memory matrix based on the similarity between the current weight configuration and the historical memory unit in the memory matrix; calculate a gradient correlation matrix between loss items, wherein the gradient correlation matrix is calculated based on the inner product and norm of the gradient vector of each loss item, and generate a correlation adjustment vector based on the gradient correlation matrix; The attention weighted features, the updated information of the memory matrix, and the correlation adjustment vector are input into a weight generation network constructed based on a gated recurrent unit. The weight generation network selectively processes the input features through an update gate and a reset gate to generate weight coefficients for each loss term. An adaptive learning rate is calculated based on historical weight changes, and the weight coefficients for each loss term are updated in combination with the adaptive learning rate. The parameters of the neural network structure are optimized and updated based on the calculation result of the comprehensive loss function until the calculation result of the comprehensive loss function is less than a preset score threshold.

[0048] Illustratively, this embodiment describes in detail the steps of dynamically adjusting the weight coefficients of each loss term in the comprehensive loss function around the adaptive weight adjustment mechanism.

[0049] First, a time series feature matrix is constructed to record the historical sequence of loss items. Each row of this matrix represents the loss item value at a point in time, while the columns correspond to different loss items. By collecting historical data, a feature matrix containing multiple time windows is formed. Next, weighted features are calculated using the attention mechanism. Specifically, an attention score is first calculated for each loss item at the current moment. These scores reflect the importance of each loss item at that moment. Then, attention-weighted features are generated by calculating the similarity between the query matrix and the key matrix.

[0050] Next, a memory matrix is established to store historical weight configurations. Each memory cell contains a weight configuration vector, a corresponding loss value, and a timestamp. The information in the memory matrix is updated by calculating the similarity between the current weight configuration and the historical memory cells in the memory matrix. This process ensures the effective use of historical information and makes weight adjustments more accurate.

[0051] Next, we calculate the gradient correlation matrix between the loss terms. This matrix is calculated by taking the inner product of the gradient vectors of each loss term and their norms, reflecting the correlation between the loss terms. Based on the gradient correlation matrix, we generate a correlation adjustment vector to facilitate dynamic weight adjustments in subsequent steps.

[0052] The attention-weighted features, memory matrix updates, and relevance adjustment vectors are then fed into a weight generation network based on gated recurrent units. This network selectively processes the input features through update and reset gates to generate weight coefficients for each loss term. These weight coefficients are generated based on the features and historical information extracted in the previous steps, ensuring dynamic and adaptable weights.

[0053] Finally, an adaptive learning rate is calculated based on historical weight changes, and the weight coefficients of each loss term are updated using this learning rate. By optimizing and updating the results of the comprehensive loss function, the parameters of the neural network structure are adjusted until the result of the comprehensive loss function is less than the preset score threshold. This process ensures continuous optimization and performance improvement of the model.

[0054] The dynamic adjustment of weight coefficients in the present invention can effectively improve the learning efficiency of the model, making the weight distribution between different loss items of the model more reasonable, thereby improving the overall performance; by introducing the attention mechanism and memory matrix, the model can better capture the relationship between historical information and the current state, enhancing the adaptability and flexibility of the model; the calculation of the adaptive learning rate combined with the weight update can accelerate model convergence, reduce training time, and at the same time improve the model's generalization ability on new data.

[0055] In an optional embodiment, An adaptive fuzzy PID controller is established with the environmental index score deviation value as input and the operating parameter adjustment value as output; the environmental index score deviation value and the deviation change rate are fuzzified respectively, and a fuzzy reasoning result is obtained according to a preset fuzzy rule base; the fuzzy reasoning result is defuzzified using the center of gravity method. The steps of dynamically adjusting the PID controller include: Collect historical data sequences of environmental index scores; process the historical data sequences using an LSTM network with a differential attention layer, enhance mutation characteristics by calculating the weight distribution of the environmental index score difference sequence, generate prediction trajectories and prediction confidence, and set three-level warning thresholds, including mild, moderate, and severe; calculate the deviation value and deviation change rate between the predicted trajectory and the actual environmental index score; divide the value ranges of the deviation value and deviation change rate into multi-level fuzzy subsets and establish corresponding membership functions; fuzzify the deviation value and deviation change rate based on the membership function to obtain the corresponding fuzzy quantity; Constructing a fuzzy rule base, the fuzzy rule base including a correspondence between a combination of the deviation value and the deviation change rate and an increment of a PID controller parameter, the fuzzy rule base setting a control parameter adjustment strategy based on a relationship between the deviation value and the three-level warning threshold and an absolute value of the deviation change rate; Based on the fuzzy quantity, the activation degree of each fuzzy rule is calculated, and a fuzzy reasoning operation is performed to obtain a comprehensive fuzzy reasoning result; the fuzzy reasoning result is defuzzified by using the center of gravity method to obtain a PID controller parameter increment for dynamically optimizing control performance; the control parameter is updated in real time according to the PID controller parameter increment and the control quantity is calculated; a feedforward compensation term is introduced, the feedforward compensation term is calculated based on the second-order derivative of the deviation value, and the feedforward compensation term is superimposed on the control quantity to obtain a total control quantity; the operating parameter is adjusted based on the total control quantity, and the adjustment amount of the operating parameter is proportional to the total control quantity.

[0056] Exemplarily, this embodiment executes the control process in the following order: data collection → prediction analysis → deviation calculation → fuzzification processing → rule reasoning → defuzzification → control calculation → execution adjustment.

[0057] First, collect a historical data series of environmental performance indicators. This data can be obtained through sensors or monitoring systems, recording the environmental performance indicators at different points in time. The integrity and accuracy of this historical data are the foundation for subsequent analysis.

[0058] Next, a long short-term memory (LSTM) network with a differential attention layer is used to process the historical data series. The LSTM network used consists of two hidden layers, each with 64 neurons, with the differential attention layer placed after the LSTM layer. LSTM networks effectively capture long-term dependencies in time series data. By calculating the weight distribution of the difference series of environmental indicator scores, the mutation characteristics are enhanced, and a prediction trajectory and prediction confidence are generated. At this point, three warning thresholds are set: minor, moderate, and severe, to facilitate subsequent deviation analysis. These three warning thresholds are determined based on the statistical characteristics of the historical data: the minor warning threshold is set at 1.2 times the historical fluctuation range, the moderate warning threshold at 1.5 times, and the severe warning threshold at 2 times. These thresholds directly influence the selection weights of fuzzy rules: when the deviation value exceeds the severe warning threshold, the rule group emphasizing rapid response is activated; when the deviation value is in the moderate warning range, a balance is maintained between response speed and stability; and when the deviation value only exceeds the minor warning threshold, rules emphasizing stability are prioritized.

[0059] After obtaining the predicted trajectory, the deviation and rate of change between the trajectory and the actual environmental indicator score are calculated. The deviation reflects the gap between the prediction and the actual score, while the rate of change indicates the trend of this gap. This information is used in subsequent fuzzification processing.

[0060] The range of deviation values and deviation change rates is divided into multiple fuzzy subsets, and corresponding membership functions are established. Deviation values are typically divided into seven fuzzy subsets: "negative large," "negative medium," "negative small," "zero," "positive small," "positive medium," and "positive large." A similar division is used for the deviation change rate. For example, for the "zero" fuzzy subset, a triangular membership function can be used, with the center at 0 and the left and right bases at -0.1 and 0.1, respectively. For the "positive small" and "negative small" fuzzy subsets, a trapezoidal membership function can be used, with the range of ±(0.05-0.3). The membership function describes the membership of each fuzzy subset and maps specific deviation values and rates of change to fuzzy quantities. For example, a deviation value may belong to both the "positive small" and "positive medium" fuzzy subsets, but with different memberships. For example, its membership to "positive small" is 0.7, while its membership to "positive medium" is 0.3.

[0061] A "pre-set fuzzy rule base" refers to a control knowledge base defined during the design phase. It contains a series of empirical rules that guide how to adjust PID parameters based on the deviation state. These rules follow the following empirical principles: when the deviation is large, the proportional coefficient Kp should be increased to speed up the response; when the deviation is close to zero but the rate of change is large, the differential coefficient Kd should be increased to prevent overshoot; and when the deviation remains stable at a small value for a long time, the integral coefficient Ki should be appropriately increased to eliminate static error. For example, "If the deviation is large and positive and the rate of change is small, then Kp should be increased, Ki should remain unchanged, and Kd should be slightly decreased." Based on the fuzzy rule base template defined during the design phase and combined with the characteristic parameters of the current operating environment, a specific fuzzy rule base suitable for the current operating conditions is instantiated. This two-stage "template + instantiation" design model ensures consistent control strategies while providing adaptability to diverse environmental conditions. The specific fuzzy rule base consists of multiple if-then rules, such as "IF (deviation is positive or neutral) AND (deviation rate of change is small and negative) THEN (Kp increment is small and positive) (Ki increment is zero) (Kd increment is small and negative)." The fuzzy rule base is designed based on the relationship between the deviation value and the three-level warning thresholds, as well as the absolute value of the deviation change rate, to define the control parameter adjustment strategy. Specifically, when the deviation value approaches or exceeds the severe warning threshold, the corresponding rule will make a larger parameter adjustment; when the absolute value of the deviation change rate is large, the rule will increase the adjustment weight of the Kd parameter. Through these rules, dynamic adjustment of PID controller parameters is achieved.

[0062] Based on the fuzzy variables, the activation level of each fuzzy rule is calculated, and fuzzy reasoning is performed. The activation level of each rule is determined by the minimum membership level of each premise in its conditional part (IF part). Multiple rules may be activated simultaneously, and their conclusions (THEN parts) are weighted and synthesized according to their respective activation levels to obtain a comprehensive fuzzy reasoning result. The results of fuzzy reasoning provide the basis for the increment of PID controller parameters.

[0063] The center of gravity method is used to defuzzify the fuzzy inference results and obtain the PID controller parameter increments used to dynamically optimize control performance. The center of gravity method is a commonly used defuzzification method. Its basic idea is to treat each output fuzzy set as objects with different masses distributed at different locations. The center of mass of these objects is then calculated as the final precise output value. In actual calculations, the activation level of each rule is first multiplied by the eigenvalue of its corresponding output fuzzy set. Then, all these products are added and divided by the sum of all activation levels. The result is the precise defuzzified value. This method comprehensively considers the influence of all activated rules and produces a smooth and continuous control output. This increment is used to update the control parameters in real time and calculate the control variable.

[0064] A feedforward compensation term is introduced, calculated based on the second-order derivative of the deviation. The feedforward compensation term is added to the control variable to obtain the total control variable. The total control variable is used to adjust the operating parameters, ensuring that the adjustment of the operating parameters is proportional to the total control variable.

[0065] The present invention can improve the real-time monitoring and adjustment capabilities of environmental protection indicators, ensure the effectiveness of environmental protection measures, enhance the system's response capability to emergencies through the adaptive fuzzy PID controller, reduce environmental risks, and achieve dynamic optimization of control parameters, thereby improving the overall operating efficiency and stability of the system.

[0066] In an optional embodiment, The step of setting the adjustment strategy of the control parameter based on the relationship between the deviation value and the three-level warning threshold and the absolute value of the deviation change rate by the fuzzy rule base includes: An adjustment strategy is determined according to a preset fuzzy rule base, wherein when the current deviation value is greater than the severe warning threshold and the absolute value of the deviation change rate is less than the second preset threshold, the integral coefficient is adjusted to a preset maximum value; when the current deviation value is between the medium warning threshold and the severe warning threshold and the absolute value of the deviation change rate is less than the second preset threshold, the integral coefficient is increased according to a first preset ratio; when the current deviation value is less than the minor warning threshold and the absolute value of the deviation change rate is greater than the second preset threshold, the differential coefficient is adjusted to a preset maximum value; when the current deviation value is between the minor warning threshold and the medium warning threshold and the absolute value of the deviation change rate is greater than the second preset threshold, the differential coefficient is increased according to a second preset ratio; Dynamically adjusting the third-level warning threshold based on the accuracy and prediction confidence of the predicted trajectory; The proportional coefficient, integral coefficient and differential coefficient of the PID controller are updated according to the preset PID controller parameter increment, the prediction error of the predicted trajectory is calculated, the network parameters of the LSTM network are optimized, and the predicted trajectory and prediction confidence are regenerated using the optimized network parameters.

[0067] Exemplarily, this embodiment revolves around the adjustment strategy of the fuzzy rule base, and describes in detail how to set the adjustment strategy of the control parameters based on the relationship between the deviation value and the three-level warning threshold and the absolute value of the deviation change rate.

[0068] First, the system analyzes the relationship between the current deviation and the warning threshold based on a pre-set fuzzy rule base. Specifically, when the current deviation exceeds the critical warning threshold and the absolute value of the deviation change rate is less than a second pre-set threshold, the system adjusts the integral coefficient to a preset maximum value. This strategy aims to rapidly enhance the system's responsiveness even in the presence of large, but stable, deviations.

[0069] For example, assuming that the severe warning threshold is ten, the current deviation value is twelve, and the absolute value of the deviation change rate is 0.5, the system sets the integral coefficient to the maximum value to improve the control effect.

[0070] If the current deviation falls between the moderate and severe warning thresholds and the absolute value of the deviation change rate is less than a second preset threshold, the system will increase the integral coefficient by a first preset ratio. This strategy aims to moderately adjust for moderate deviations to avoid overreaction.

[0071] For example, if the medium warning threshold is eight, the current deviation value is nine, and the absolute value of the deviation change rate is 0.3, the system will increase the integral coefficient according to a first preset ratio (eg, 10 percent).

[0072] When the current deviation is less than the minor warning threshold and the absolute value of the deviation change rate is greater than a second preset threshold, the system adjusts the differential coefficient to a preset maximum value. This strategy aims to enhance the system's sensitivity in situations where the deviation is small but changes dramatically.

[0073] For example, assuming the minor warning threshold is five, the current deviation value is four, and the absolute value of the deviation change rate is one, the system sets the differential coefficient to the maximum value to improve its responsiveness to rapid changes.

[0074] If the current deviation falls between the minor and moderate warning thresholds and the absolute value of the deviation change rate exceeds a second preset threshold, the system will increase the differential coefficient by a second preset ratio. This strategy aims to moderately adjust for minor deviations to improve system stability.

[0075] For example, if the current deviation value is six and the absolute value of the deviation change rate is 0.7, the system will increase the differential coefficient according to a second preset ratio (eg, 5 percent).

[0076] Based on the accuracy and confidence of the predicted trajectory, the system will dynamically adjust the three-level warning thresholds. This process requires real-time monitoring of the prediction results and adjusting the thresholds according to actual conditions to ensure the flexibility and adaptability of the system.

[0077] For example, if the accuracy of the predicted trajectory improves, the system may increase the warning threshold accordingly to reduce false alarms.

[0078] Based on the preset PID controller parameter increments, the system will update the proportional, integral, and differential coefficients of the PID controller. At this time, the system will calculate the prediction error of the predicted trajectory and use this error to optimize the network parameters of the LSTM network.

[0079] For example, if the prediction error is 0.2, the system will adjust the parameters of the PID controller based on the error to improve the control accuracy.

[0080] Finally, using the optimized network parameters, the system will regenerate the predicted trajectory and prediction confidence. This process ensures the continuous optimization and improvement of the system in a dynamic environment.

[0081] For example, after parameter optimization, the new predicted trajectory generated by the system shows higher accuracy and confidence, improving the overall control effect.

[0082] The present invention improves the system's response speed to deviations and enhances the sensitivity and stability of control; dynamically adjusts the warning threshold to ensure the system's adaptability and flexibility in different environments; and by optimizing LSTM network parameters, improves the accuracy and confidence of predicted trajectories and enhances the overall performance of the system.

[0083] In an optional embodiment, Introducing a feedforward compensation term, wherein the feedforward compensation term is calculated based on the second-order derivative of the deviation value, comprises the following steps: Construct multi-timescale window data, calculate high-frequency features through derivatives in the short term, extract frequency and periodic features in the medium term, and use Transformer to predict trends in the long term. Calculate performance indicators based on the integrated squared error of each time scale, and use exponential normalization to obtain compensation weights. Update compensation parameters through dynamic step size factor optimization, specifically including: Construct short-term time window data, medium-term time window data, and long-term time window data containing environmental protection index scores and operating parameters; calculate the first-order derivative and second-order derivative of the environmental protection index score based on the short-term time window data, obtain the high-frequency component feature vector, and calculate the short-term compensation amount; perform spectrum analysis and pattern recognition on the medium-term time window data, extract the dominant frequency characteristics and periodic characteristics, and calculate the medium-term compensation amount; use the Transformer network based on the self-attention mechanism to perform time series prediction on the long-term time window data, obtain trend prediction results and prediction confidence, and calculate the long-term compensation amount; The integrated square error of the environmental protection index score is calculated in the short-term evaluation time interval, the medium-term evaluation time interval, and the long-term evaluation time interval to obtain the corresponding performance index; the compensation weight of each time scale is calculated based on the performance index using the exponential normalization method, the sum of the compensation weights is 1 and is inversely proportional to the corresponding performance index; the compensation amount of each time scale is multiplied by the corresponding compensation weight and the sum is obtained to obtain the feedforward compensation amount; A corresponding dynamic step factor is calculated based on the performance index of each time scale, and the dynamic step factor decays exponentially with the increase of the performance index; the compensation parameters of each time scale are dynamically optimized and updated using the dynamic step factor, and the feedforward compensation amount is recalculated based on the updated compensation parameters.

[0084] For example, first, multi-timescale window data is constructed. Short-term window data extracts high-frequency features by calculating the first- and second-order derivatives of the environmental index score, and a short-term compensation value is obtained by weighted combination of these derivative values. These features can reflect rapid changes in the environmental index score. Medium-term window data uses spectrum analysis and pattern recognition to extract frequency and periodic features. A medium-term compensation value is calculated by weighted combination of the dominant frequency values and periodic feature values, helping to identify cyclical changes in the environmental index score. Long-term window data uses a Transformer network based on the self-attention mechanism for trend prediction. The long-term compensation value is calculated by taking the weighted difference between the predicted value and the current value, which can capture the long-term trend of changes in the environmental index score.

[0085] Next, performance indicators are calculated based on the integrated squared error at each time scale. The short-term, medium-term, and long-term performance indicators reflect the accuracy of the environmental indicator scores at different time scales. These performance indicators are converted into compensation weights using exponential normalization, ensuring that the sum of the compensation weights is 1 and that they are inversely proportional to the corresponding performance indicator.

[0086] Next, the compensation parameters are updated using a dynamic step factor. The dynamic step factor adjusts based on performance metrics, decaying exponentially as the performance metric increases. This mechanism ensures that the compensation parameter updates can flexibly adapt to changes at different time scales.

[0087] Next, the compensation amount is calculated for each time scale. Short-term compensation is calculated based on the high-frequency component eigenvector, medium-term compensation is calculated by extracting dominant frequency and periodic features, and long-term compensation relies on the trend prediction results of the Transformer network. The compensation amount for each time scale is multiplied by the corresponding compensation weight and summed to obtain the final feedforward compensation amount.

[0088] Finally, the corresponding dynamic step size factor is calculated based on the performance indicators of each time scale, and the compensation parameters of each time scale are dynamically optimized and updated. The updated compensation parameters are used to recalculate the feedforward compensation, forming a closed-loop feedback mechanism to ensure the accuracy and real-time performance of the compensation.

[0089] This method uses a collaborative analysis across three timescales: short-term, medium-term, and long-term. This enables the control system to rapidly respond to fluctuations across different timescales. Compared to single-timescale analysis, this significantly shortens the system's response time to sudden fluctuations, improves fluctuation suppression, and reduces steady-state errors.

[0090] By introducing a feedforward compensation term based on the second-order derivative, the present invention enables the system to predict the changing trend of environmental protection indicators. The advance compensation mechanism enables the adjustment to be started in time and effectively controls the extent of exceeding the standard. Compared with traditional PID control, the risk of exceeding the standard is significantly reduced.

[0091] The present invention uses an exponential normalization method to dynamically assign compensation weights, enabling the control strategy to adaptively adjust based on performance at different time scales. This system exhibits lower overshoot, shorter settling time, and higher steady-state accuracy when responding to periodic fluctuations.

[0092] This paper introduces a Transformer network to handle long-term trend forecasting, fully leveraging the temporal characteristics of historical data. Compared with traditional temporal forecasting methods, the accuracy of long-term environmental indicators forecasts is significantly improved, providing a more reliable basis for control decisions.

[0093] Figure 2FIG. 1 is a structural diagram of an environmental assessment system based on integrated printing press operating parameters according to an embodiment of the present invention. Figure 2 As shown, the system includes: The first unit is configured to use the collected operating parameters of the integrated printing press as input layer nodes of a neural network algorithm, and calculate an environmental index score based on the neural network algorithm, wherein the environmental index score includes a VOCs emission index score, a wastewater discharge index score, an energy consumption index score, and a solid waste generation index score; The second unit is used to generate printing process optimization suggestions based on the environmental index score; The third unit is configured to automatically adjust the operating parameters of the integrated printing press using an adaptive fuzzy PID control algorithm based on the printing process optimization suggestion, including: establishing an adaptive fuzzy PID controller with an environmental index score deviation value as an input and an operating parameter adjustment value as an output; fuzzifying the environmental index score deviation value and the deviation change rate, respectively, and obtaining a fuzzy inference result based on a preset fuzzy rule base; defuzzifying the fuzzy inference result using a center of gravity method and dynamically adjusting the PID controller; and converting the defuzzified output into an adjustment instruction for the operating parameter; The fourth unit is used to re-collect the operating parameters after completing the parameter adjustment, and calculate and evaluate the environmental protection index scores until the environmental protection index scores all reach the corresponding preset thresholds.

[0094] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0095] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0096] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An environmental assessment method based on integrated printing press operating parameters, characterized in that: include: Using the collected operating parameters of the integrated printing press as input layer nodes of a neural network algorithm, and calculating an environmental index score based on the neural network algorithm, the environmental index score including a VOCs emission index score, a wastewater discharge index score, an energy consumption index score, and a solid waste generation index score; generating printing process optimization suggestions based on the environmental index scores; Based on the printing process optimization suggestion, an adaptive fuzzy PID control algorithm is used to automatically adjust the operating parameters of the integrated printing press, including: establishing an adaptive fuzzy PID controller with an environmental index score deviation value as an input and an operating parameter adjustment value as an output; fuzzifying the environmental index score deviation value and the deviation change rate, respectively, and obtaining a fuzzy reasoning result based on a preset fuzzy rule library; defuzzifying the fuzzy reasoning result using a center of gravity method, and dynamically adjusting the PID controller; and converting the defuzzified output into an adjustment instruction for the operating parameter; After completing the parameter adjustment, the operating parameters are collected again, and the environmental protection index scores are calculated and evaluated until the environmental protection index scores all reach corresponding preset thresholds.

2. The method according to claim 1, characterized in that The steps of calculating the environmental protection index score based on the neural network algorithm include: Mapping the operating parameter to a range from negative one to positive one using a mapping function; Construct a neural network structure with three hidden layers. The first hidden layer includes fourteen neurons and uses the ReLU activation function. The second hidden layer includes ten neurons and uses the ReLU activation function. The third hidden layer includes six neurons and uses the ReLU activation function. The output layer includes four output nodes and uses the Sigmoid activation function. According to the operating parameters, the environmental protection index score is calculated by the neural network structure, wherein: The VOCs emission index scores are: ； w T 、w H are the weight coefficients of temperature parameter and ink layer thickness parameter respectively, 、 These are the influence functions of temperature and ink layer thickness on VOCs, T is the temperature parameter, and H is the ink layer thickness parameter; Wastewater discharge indicators are calculated based on paper moisture content and printing pressure, energy consumption indicators are calculated based on ink roller speed and ink supply, and solid waste generation indicators are calculated based on printing speed; The neural network structure is trained using a comprehensive loss function, wherein the comprehensive loss function includes an environmental index prediction loss term, a parameter coupling loss term, a time series continuity loss term, and a process constraint loss term; The calculation results of the environmental index scores are evaluated in real time, including: data validity verification and outlier processing of the operating parameters; calculation of confidence intervals of the environmental index scores; and retraining the neural network structure when the prediction error of any environmental index score exceeds a preset error threshold.

3. The method according to claim 2, characterized in that The steps of the comprehensive loss function including the environmental index prediction loss term, the parameter coupling loss term, the time series continuity loss term and the process constraint loss term include: The environmental index prediction loss term adopts a piecewise function structure. When the prediction error is less than a first preset threshold, a quadratic loss is used for calculation; when the prediction error is greater than or equal to the first preset threshold, a linear loss is used for calculation; the parameter coupling loss term is calculated based on a first coupling constraint between printing speed and ink layer thickness, a second coupling constraint between printing pressure and paper moisture content, and a third coupling constraint between ink roller speed and ink supply; the temporal continuity loss term is calculated based on the difference between the environmental index prediction values at adjacent moments; the process constraint loss term is calculated based on the physical limitations of the operating parameters. When the predicted value of the operating parameter exceeds the corresponding feasible interval, a constraint penalty value is calculated based on the distance between the predicted value of the operating parameter and the boundary of the feasible interval; During the training process of the neural network structure, an adaptive weight adjustment mechanism is used to dynamically adjust the weight coefficients of each loss term in the comprehensive loss function.

4. The method according to claim 3, characterized in that The steps of dynamically adjusting the weight coefficients of the loss items in the comprehensive loss function using an adaptive weight adjustment mechanism include: Constructing a time series feature matrix for each loss item, the time series feature matrix containing a historical value sequence of each loss item within a preset time window; calculating an attention score based on the time series feature matrix, and generating an attention weighted feature based on the attention score; constructing a memory matrix to store historical weight configuration information, and updating the memory matrix based on the similarity between the current weight configuration and the historical memory units in the memory matrix; calculating a gradient correlation matrix between the loss items, and generating a correlation adjustment vector based on the gradient correlation matrix; The attention weighted features, the updated information of the memory matrix and the correlation adjustment vector are input into the weight generation network to generate the weight coefficients of each loss item; the adaptive learning rate is calculated based on the historical weight changes, and the weight coefficients of each loss item are updated in combination with the adaptive learning rate.

5. The method according to claim 1, wherein An adaptive fuzzy PID controller is established with the environmental index score deviation value as input and the operating parameter adjustment value as output; the environmental index score deviation value and the deviation change rate are fuzzified respectively, and the fuzzy reasoning result is obtained according to the preset fuzzy rule base; The steps of defuzzifying the fuzzy inference result by using the center of gravity method and dynamically adjusting the PID controller include: Collect historical data series of environmental index scores, use LSTM network processing to generate prediction trajectories and prediction confidence, and set three-level warning thresholds; calculate the deviation value and deviation change rate between the predicted trajectory and the actual environmental index score; and obtain the corresponding fuzzy value based on the deviation value and deviation change rate; Constructing a fuzzy rule base, the fuzzy rule base including a correspondence between a combination of the deviation value and the deviation change rate and an increment of a PID controller parameter, the fuzzy rule base setting a control parameter adjustment strategy based on a relationship between the deviation value and the three-level warning threshold and an absolute value of the deviation change rate; The activation degree of the rule is calculated based on the fuzzy quantity and reasoning is performed. The center of gravity method is used to convert the reasoning result into a PID parameter increment. The controller parameters are updated according to the parameter increment and the control quantity is calculated. A feedforward compensation term is introduced and the feedforward compensation term is superimposed on the control quantity to obtain the total control quantity. The operating parameters are adjusted based on the total control quantity.

6. The method according to claim 5, characterized in that The step of setting the adjustment strategy of the control parameter based on the relationship between the deviation value and the three-level warning threshold and the absolute value of the deviation change rate by the fuzzy rule base includes: An adjustment strategy is determined according to a preset fuzzy rule base, wherein when the current deviation value is greater than the severe warning threshold and the absolute value of the deviation change rate is less than the second preset threshold, the integral coefficient is adjusted to a preset maximum value; when the current deviation value is between the medium warning threshold and the severe warning threshold and the absolute value of the deviation change rate is less than the second preset threshold, the integral coefficient is increased according to a first preset ratio; when the current deviation value is less than the minor warning threshold and the absolute value of the deviation change rate is greater than the second preset threshold, the differential coefficient is adjusted to a preset maximum value; when the current deviation value is between the minor warning threshold and the medium warning threshold and the absolute value of the deviation change rate is greater than the second preset threshold, the differential coefficient is increased according to a second preset ratio; Dynamically adjusting the third-level warning threshold based on the accuracy and prediction confidence of the predicted trajectory; The proportional coefficient, integral coefficient and differential coefficient of the PID controller are updated according to the preset PID controller parameter increment, the prediction error of the predicted trajectory is calculated, the network parameters of the LSTM network are optimized, and the predicted trajectory and prediction confidence are regenerated using the optimized network parameters.

7. The method according to claim 5, characterized in that The steps to introduce the feedforward compensation term include: Construct short-term time window data, medium-term time window data, and long-term time window data containing environmental protection index scores and operating parameters; calculate the first-order derivative and second-order derivative of the environmental protection index score based on the short-term time window data, obtain the high-frequency component feature vector, and calculate the short-term compensation amount; perform spectrum analysis and pattern recognition on the medium-term time window data, extract the dominant frequency characteristics and periodic characteristics, and calculate the medium-term compensation amount; use the Transformer network to perform time series prediction on the long-term time window data, obtain trend prediction results and prediction confidence, and calculate the long-term compensation amount; Calculate the integrated square error of the environmental protection index score in different evaluation time intervals to obtain the corresponding performance index; calculate the compensation weight of each time scale based on the performance index; multiply the compensation amount of each time scale by the corresponding compensation weight and sum them to obtain the feedforward compensation amount; The corresponding dynamic step size factor is calculated according to the performance indicators of each time scale, and the compensation parameters are optimized and updated.

8. An environmental assessment system based on integrated printing press operating parameters, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is configured to use the collected operating parameters of the integrated printing press as input layer nodes of a neural network algorithm, and calculate an environmental index score based on the neural network algorithm, wherein the environmental index score includes a VOCs emission index score, a wastewater discharge index score, an energy consumption index score, and a solid waste generation index score; The second unit is used to generate printing process optimization suggestions based on the environmental index score; The third unit is configured to automatically adjust the operating parameters of the integrated printing press using an adaptive fuzzy PID control algorithm based on the printing process optimization suggestion, including: establishing an adaptive fuzzy PID controller with an environmental index score deviation value as an input and an operating parameter adjustment value as an output; fuzzifying the environmental index score deviation value and the deviation change rate, respectively, and obtaining a fuzzy inference result based on a preset fuzzy rule base; defuzzifying the fuzzy inference result using a center of gravity method and dynamically adjusting the PID controller; and converting the defuzzified output into an adjustment instruction for the operating parameter; The fourth unit is used to re-collect the operating parameters after completing the parameter adjustment, and calculate and evaluate the environmental protection index scores until the environmental protection index scores all reach the corresponding preset thresholds.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images