3PE anticorrosive coating production process parameter control method based on intelligent algorithm

Abstract

The invention belongs to the technical field of intelligent control, and particularly relates to a 3PE anti-corrosion layer production process parameter control method based on an intelligent algorithm. The method comprises the following steps: firstly, collecting historical data and real-time production data, and obtaining a standardized data set through abnormal value elimination, missing value filling and data normalization processing; thirdly, training by adopting an improved multi-objective optimization algorithm, and outputting a first prediction process parameter containing a parameter optimal interval by taking the yield and the coating uniformity as targets; and then constructing a process parameter dynamic graph model, calculating a node correlation entropy and triggering parameter adjustment according to the deviation degree of real-time parameters and predicted parameters, historical correlation coefficients among the parameters and expert experience, and obtaining optimized production parameters through multiple rounds of iteration. According to the method, parameters can be accurately determined and dynamically adjusted, complex correlation between data is effectively processed, and the production stability and the product quality of the 3PE anticorrosive coating are improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent control technology, and in particular relates to a method for controlling process parameters of a 3PE anti-corrosion layer production based on an intelligent algorithm. Background Art

[0002] 3PE coating is a widely used corrosion protection structure for buried steel pipelines. It consists of three layers: epoxy powder, adhesive, and polyethylene. The epoxy powder layer provides excellent adhesion and corrosion resistance. The adhesive layer connects the epoxy powder and polyethylene, strengthening the interlayer bond. The polyethylene layer provides excellent mechanical properties, weather resistance, and corrosion resistance, effectively protecting the pipeline from erosion by external factors such as soil, moisture, and chemicals, thereby extending its service life.

[0003] During the production of 3PE anti-corrosion coatings, process parameters have a crucial impact on product quality. However, existing methods for controlling these parameters have limitations. Traditional methods struggle to fully account for the numerous and complex process parameters and their interrelationships, making it difficult to accurately determine optimal production parameters. This results in instability in product yield and coating uniformity. Furthermore, the lack of an effective dynamic adjustment mechanism prevents timely optimization of parameters in the face of real-time parameter fluctuations during production, impacting production efficiency and product quality. Summary of the Invention

[0004] In view of the technical problems existing in the above-mentioned background technology, the present invention proposes a method for controlling the production process parameters of 3PE anti-corrosion layer based on an intelligent algorithm.

[0005] In order to achieve the above object, the technical solution adopted by the present invention includes the following steps:

[0006] S1. Collect the current real-time production data of historical data in the production of 3PE anti-corrosion layer;

[0007] S2. Preprocessing the collected historical data, wherein the preprocessing includes removing outliers, filling missing values, and normalizing the historical data to obtain a standardized data set;

[0008] S3. Using an improved multi-objective optimization algorithm to train the standardized data set, taking yield rate and coating uniformity as optimization objectives, and outputting first predicted process parameters containing an optimal parameter range;

[0009] S4. Construct a dynamic graph model of process parameters, encode the deviation between the real-time production parameters and the first predicted process parameters as node features, annotate edge weights based on the historical correlation coefficient between parameters combined with expert experience, calculate node correlation entropy, and trigger parameter adjustments based on the deviation of core nodes in the graph. Propagate adjustment instructions to neighboring nodes according to edge weights. Finally, after multiple rounds of adjustment iterations, output the optimized production parameters.

[0010] The specific steps of finally obtaining the optimized production parameters in step S4 are as follows:

[0011] S41. First, collect production parameters in real time Calculate the deviation from the first predicted process parameter. The deviation calculation method is: in is to take the midpoint value of the first predicted process parameter interval;

[0012] S42, the top six nodes ranked by correlation entropy are recorded as core nodes. When the deviation degree d of the core node i ≥0.15, adjust the accuracy according to the equipment Converge to the target interval and get the first-level adjustment parameter, which is calculated as follows: in It is expressed as a first-level adjustment parameter, and the remaining adjusted parameters are obtained by diffusing the adjustment instructions according to the edge weights;

[0013] S43. Constraint check is performed on the adjusted parameters. If the parameter value is within the physical range, it is directly adopted. If it is out of bounds, the nearest boundary value is taken to form the secondary adjustment parameter.

[0014] S44, after one round of adjustment, recalculate the core node deviation, and perform iterative calculations until the d of all core nodes i ≤0.1 Output all final optimized production parameters.

[0015] Preferably, the historical data and production data in step S1 include preheating temperature, medium frequency heating temperature, post-heating temperature, pulling speed, coating thickness, adhesive layer ratio, epoxy powder spraying powder particle size, epoxy powder spraying powder curing temperature, epoxy powder spraying speed, adhesive extrusion temperature, adhesive extrusion speed, adhesive extrusion temperature, polyethylene extrusion temperature, steel pipe speed, cooling time, and cooling temperature.

[0016] Preferably, the specific implementation method of data preprocessing in step S1 includes outlier detection and missing value filling;

[0017] The outlier detection method is to construct an anomaly detection graph. For each parameter node, directly related nodes with edge weights ≥ 0.6 in the graph are extracted to form local related clusters. The Mahalanobis distance between real-time data and historical mean is calculated in each related cluster, and a threshold is set. If the calculated Mahalanobis distance threshold is greater than the threshold, it is determined that the parameter is abnormal. All node data in the related cluster determined to be abnormal are removed.

[0018] The missing value filling is to fill the data position after the outliers are removed, and the bidirectional strong correlation nodes are screened for the missing positions according to the positive order of the graph edge weights. The bidirectional strong correlation nodes are based on the weights of the structural edges in the graph. Starting from the missing value parameters, the parameters with direct causal influence on them are first screened, and the top three are selected in descending order of the edge weights. The parameters affected by their direct feedback are screened again, and the top three are selected in descending order of the edge weights to form a set of 6 strongly correlated nodes including causality and feedback. Finally, the filling value obtained by interpolation is checked to see whether it meets the process constraints. If the filling value is within a reasonable range, it is directly adopted. If it exceeds the upper limit, the maximum value boundary is taken. If it is lower than the lower limit, the minimum value boundary is taken.

[0019] Preferably, the step S3 uses an improved multi-objective optimization algorithm to train the standardized data set, takes yield rate and coating uniformity as optimization targets, and outputs the first predicted process parameters containing the optimal parameter range as follows:

[0020] S31. First, build a deep solution generation network, input the historical standardized parameters, and output the initial candidate solution so that the initial distribution of the population preferentially covers the historical efficient area. The calculation method of the initial candidate solution is: init =clip(σ(W2σ(W1X std +b1)+b2)⊙(x max -x min )+x min ,x min ,x max ), where W2, W1 represent weight matrices, b1, b2 represent bias vectors, σ represents activation function, and x max ,x min are the maximum and minimum values of the physical interval of the process parameters, ⊙ represents element-by-element multiplication, and clip is the interval clipping function;

[0021] S32. Set an adaptive weight network, input the target statistical characteristics of the current population, and output a dynamic weight ω(t). The dynamic weight output is calculated as follows: Where Ω represents the adaptive weight matrix, σ Y ,σ U are the standard deviation of good products and coating uniformity, σ YU Represents the correlation between two goals;

[0022] S33. Construct an objective function based on the dynamic weight. The objective function is specifically: f(x)=ω Y (1-Y(x))+ω Y ·U(x)+λ·D HD (x), where Y(x) is the yield rate prediction value of parameter combination x, U(x) represents the coating uniformity prediction value, and D HD (x) represents the diversity metric based on Hausdorff distance, which characterizes the spatial distribution difference between solution x and other solutions in the population, and λ represents the diversity adjustment coefficient;

[0023] S34, then generate non-traditional parameter combination x based on the target value (Y, U) gen , specifically implemented by using the adversarial network, the generator G inputs noise z and the conditional target output x gen =G(z,Y * ,U * ), where Y * ,U * represents the desired target yield and target coating uniformity; the discriminator D distinguishes the true solution from the generated solution, and the optimization objective is: Among them, P real represents the real data distribution, P (z) represents the probability distribution of noise z,

[0024] S35. Finally, based on the Pareto frontier solution, the kernel density estimation is used to determine the top 80% continuous intervals of density, and the sub-intervals that violate the process hard constraints are eliminated. After 1000 Monte Carlo simulations, the intervals with target fluctuations ≤ 5% are retained, and the first predicted process parameters containing the optimal interval are output.

[0025] Preferably, for the generated solutions in step S34, the generated solutions are replaced by the solutions with the smallest crowding distance in the population at a ratio of 30% every 10 generations, so that the algorithm can explore the low probability parameter region.

[0026] Preferably, the training loss function of the deep solution generation network in step S31 is: Where MSE is the mean square error, which measures the difference in solutions. This item represents the repulsive force between the generated solution and the existing solution, avoiding duplicate solutions.

[0027] Preferably, in step S4, edge weights are annotated using historical correlation coefficients between parameters combined with expert experience, and the specific implementation method for calculating correlation entropy is as follows: first, the Pearson correlation coefficient ρ between the parameters is calculated, and the prior knowledge of the process experts on the edges is introduced to obtain the expert correction coefficient δ. The two are averaged to obtain the final edge weight, and the weight is calculated to obtain the correlation entropy. The correlation entropy is calculated as follows: Where N(i) is the set of neighbor nodes, β ij Represents the edge weight from node i to node j.

[0028] Compared with the prior art, the advantages and positive effects of the present invention are:

[0029] 1. Build an anomaly detection graph to identify outliers. By calculating the Mahalanobis distance, we can eliminate abnormally correlated clusters to ensure data accuracy. We can filter bidirectionally strongly correlated nodes based on graph edge weights to fill in missing values. We verify the rationality of the data through process constraints and effectively handle strong data coupling, providing a reliable data foundation for subsequent analysis and optimization.

[0030] 2. Construct a deep solution generation network to ensure that the initial candidate solutions cover the historically efficient areas and improve the quality of the optimization starting point. Set the adaptive weight network to output dynamic weights, adjust the weights according to the target statistical characteristics, and balance the yield rate and coating uniformity. Combine the adversarial network to generate non-traditional parameter combinations, explore low-probability parameter areas, and avoid local optimality. Based on Pareto frontier solution screening, the precise optimal parameter interval is determined through kernel density estimation, elimination of illegal intervals, and Monte Carlo simulation verification.

[0031] 3. The deviation between real-time parameters and predicted parameters is set as the node feature, and the edge weights are labeled based on the historical correlation coefficient and expert experience, and the correlation entropy is calculated. Adjustments are triggered based on the deviation of the core node, and instructions are diffused to neighboring nodes according to the edge weights. Production parameters are optimized through multiple rounds of iterations to respond promptly to parameter fluctuations in real-time production and adapt to complex parameter coupling relationships. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0033] Figure 1 The figure is a structural flow chart of a method for controlling process parameters of 3PE anti-corrosion layer production based on intelligent algorithm. DETAILED DESCRIPTION

[0034] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0035] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0036] In order to solve the problem that the optimal production parameters cannot be accurately determined in the 3PE anti-corrosion layer production line, the present invention provides a 3PE anti-corrosion layer production process parameter control method based on an intelligent algorithm. Through data preprocessing, multi-objective optimization algorithm and dynamic graph model, accurate control and dynamic adjustment of parameters are achieved, production stability and product quality are improved, and the overall process is as follows: Figure 1 shown.

[0037] First, we collected current, real-time production data from historical data on 3PE anti-corrosion coating production. This data included preheating temperature, intermediate frequency heating temperature, post-heating temperature, pulling speed, coating thickness, adhesive layer ratio, epoxy powder coating particle size, epoxy powder coating curing temperature, epoxy powder coating speed, adhesive extrusion temperature, adhesive extrusion speed, adhesive extrusion temperature, polyethylene extrusion temperature, steel pipe speed, cooling time, and cooling temperature. These various data points are correlated and mutually influential, resulting in strong coupling between them.

[0038] Considering the strong coupling between data, for example, preheating temperature affects the melting state of epoxy powder and, together with the medium-frequency heating temperature, determines coating adhesion; the coupling between pulling speed and polyethylene extrusion temperature affects coating thickness uniformity; and the correlation between adhesive extrusion speed and the adhesive layer ratio determines interlayer bonding strength. Parameter fluctuations are transmitted through a chain of physical and chemical reactions, and a single parameter anomaly can cause degradation of multiple indicators. Traditional methods struggle to quantify these complex relationships. Therefore, it is necessary to construct a parameter network through graphical models, explore causal and feedback relationships, and provide a correlation basis for intelligent control. The specific implementation method of the present invention for data preprocessing includes outlier detection and missing value filling; the implementation method of the outlier detection is to construct an anomaly detection graph, for each parameter node, extract directly associated nodes with edge weights ≥0.6 in the graph to form a local association cluster; calculate the Mahalanobis distance between real-time data and historical mean in each association cluster, set a threshold, and if the calculated Mahalanobis distance threshold is greater than the threshold, it is determined that the parameter is abnormal; all node data in the association cluster determined to be abnormal are eliminated; missing value filling is to fill the data position after the outliers are eliminated, and the missing position is screened for bidirectional strong associated nodes according to the positive order of the graph edge weights. The bidirectional strong associated nodes are based on the weights of the structural edges in the graph, starting from the missing value parameters, first screening the parameters with direct causal influence on them, and selecting the top three in descending order of edge weights, then screening the parameters affected by their direct feedback, and selecting the top three in descending order of edge weights, to form a set of 6 strongly associated nodes including causal and feedback, and finally testing whether the interpolated filling value meets the process constraints. If the filling value is within a reasonable range, it is directly adopted; if it exceeds the upper limit, the maximum value boundary is taken; if it is below the lower limit, the minimum value boundary is taken.

[0039] After processing the data, considering that traditional single-objective optimization cannot balance yield rate and coating uniformity, and the initial solution distribution is random, resulting in low optimization efficiency and easy to fall into local optimality. The present invention uses an improved multi-objective optimization algorithm to train the standardized data set, with yield rate and coating uniformity as optimization objectives, and outputs the first predicted process parameters containing the optimal interval of parameters. Specifically, a deep solution generation network is first constructed, and the training loss function of the deep solution generation network is: Where MSE is the mean square error, which measures the difference in solutions. This term represents the repulsive force between the generated solution and the existing solution, avoiding duplicate solutions. Then input the standardized parameters of the history and output the initial candidate solution so that the initial distribution of the population preferentially covers the historical efficient area. The calculation method of the initial candidate solution is: init =clip(σ(W2σ(W1X std +b1)+b2)⊙(x max -x min )+x min ,x min ,x max), where W2, W1 represent weight matrices, b1, b2 represent bias vectors, σ represents activation function, and x max ,x min are the maximum and minimum values of the physical interval of the process parameters, ⊙ represents element-by-element multiplication, and clip is the interval clipping function. Set the adaptive weight network, input the target statistical characteristics of the current population, and output the dynamic weight ω(t). The calculation method of the dynamic weight output is: Where Ω represents the adaptive weight matrix, σ Y ,σ U are the standard deviation of good products and coating uniformity, σ YU Represents the correlation between the two objectives; construct an objective function based on the dynamic weight, and the objective function is specifically: f(x)=ω Y (1-Y(x))+ω Y ·U(x)+λ·D HD (x), where Y(x) is the yield rate prediction value of parameter combination x, U(x) represents the coating uniformity prediction value, and D HD (x) represents the diversity metric based on Hausdorff distance, which characterizes the spatial distribution difference between solution x and other solutions in the population, and λ represents the diversity adjustment coefficient. Then, with the target value (Y, U) as the condition, a non-traditional parameter combination x is generated. gen , specifically implemented by using the adversarial network, the generator G inputs noise z and the conditional target output x gen =G(z,Y * ,U * ), where Y * ,U * represents the desired target yield and target coating uniformity; the discriminator D distinguishes between the true solution and the generated solution, where the generated solution is replaced by the solution with the smallest crowding distance in the population at a ratio of 30% every 10 generations, allowing the algorithm to explore the low-probability parameter area. The optimization goal is: Among them, P real represents the real data distribution, P (z) Represents the probability distribution of noise z. Finally, based on the Pareto frontier solution, the kernel density estimation is used to determine the continuous intervals of the top 80% of the density, and the sub-intervals that violate the hard constraints of the process are eliminated. After 1000 Monte Carlo simulations, the intervals with target fluctuations ≤ 5% are retained, and the first predicted process parameters containing the optimal interval are output.

[0040] After obtaining the predicted parameters, we finally consider that parameter deviations in real-time production are difficult to respond to in a timely manner, and the coupling relationship between parameters is complex. Traditional parameter-by-parameter adjustments are prone to trigger chain reactions. Therefore, the present invention constructs a process parameter dynamic graph model, encodes the deviation between the real-time production parameters and the first predicted process parameters as node features, and annotates the edge weights with the historical correlation coefficients between the parameters combined with expert experience, calculates the node correlation entropy, and triggers parameter adjustments based on the deviation of the core nodes in the graph. The adjustment instructions are diffused to the neighboring nodes according to the edge weights, and finally, after multiple rounds of adjustment iterations, the optimized production parameters are output. Specifically, the production parameters are first collected in real time. Calculate the deviation from the first predicted process parameter. The deviation calculation method is: in The midpoint value of the first predicted process parameter interval is taken; the top six nodes ranked by correlation entropy are recorded as core nodes. When the deviation degree d of the core node is i ≥0.15, adjust the accuracy according to the equipment Converge to the target interval and get the first-level adjustment parameter, which is calculated as follows: in It is expressed as a first-level adjustment parameter, and the adjustment instruction is diffused according to the edge weight to obtain the remaining adjusted parameters; the adjusted parameters are constrained and checked. If the parameter value is within the physical range, it is directly adopted. If it is out of bounds, the nearest boundary value is taken to form the second-level adjustment parameter; after one round of adjustment, the core node deviation is recalculated and iterative calculation is performed until the d of all core nodes is i ≤0.1 Output all final optimized production parameters.

[0041] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any person skilled in the art may utilize the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes for application in other fields. However, any simple modification, equivalent change, and modification of the above embodiments made in accordance with the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for controlling process parameters of 3PE anti-corrosion layer production based on intelligent algorithm, characterized in that: The following steps are involved: S1. Collect the current real-time production data of historical data in the production of 3PE anti-corrosion layer; S2. Preprocessing the collected historical data, wherein the preprocessing includes removing outliers, filling missing values, and normalizing the historical data to obtain a standardized data set; S3. Using an improved multi-objective optimization algorithm to train the standardized data set, taking yield rate and coating uniformity as optimization objectives, and outputting first predicted process parameters containing an optimal parameter range; S4. Construct a dynamic graph model of process parameters, encode the deviation between the real-time production parameters and the first predicted process parameters as node features, annotate edge weights based on the historical correlation coefficient between parameters combined with expert experience, calculate node correlation entropy, and trigger parameter adjustments based on the deviation of core nodes in the graph. Propagate adjustment instructions to neighboring nodes according to edge weights. Finally, after multiple rounds of adjustment iterations, output the optimized production parameters. The specific steps of finally obtaining the optimized production parameters in step S4 are as follows: S41, first collect production parameters in real time Calculate the deviation from the first predicted process parameter. The deviation calculation method is: in is to take the midpoint value of the first predicted process parameter interval; S42, the top six nodes ranked by correlation entropy are recorded as core nodes. When the deviation degree d of the core node i ≥0.15, adjust the accuracy according to the equipment Converge to the target interval and get the first-level adjustment parameter, which is calculated as follows: in It is expressed as a first-level adjustment parameter, and the remaining adjusted parameters are obtained by diffusing the adjustment instructions according to the edge weights; S43. Constraint check is performed on the adjusted parameters. If the parameter value is within the physical range, it is directly adopted. If it is out of bounds, the nearest boundary value is taken to form the secondary adjustment parameter. S44, after one round of adjustment, recalculate the core node deviation, and perform iterative calculations until the d of all core nodes i ≤0.1 Output all final optimized production parameters.

2. The method for controlling process parameters of 3PE anti-corrosion layer production based on intelligent algorithm according to claim 1, characterized in that: The historical data and production data in step S1 include preheating temperature, medium frequency heating temperature, post-heating temperature, pulling speed, coating thickness, adhesive layer ratio, epoxy powder spraying powder particle size, epoxy powder spraying powder curing temperature, epoxy powder spraying speed, adhesive extrusion temperature, adhesive extrusion speed, adhesive extrusion temperature, polyethylene extrusion temperature, steel pipe speed, cooling time, and cooling temperature.

3. The method for controlling process parameters of 3PE anti-corrosion layer production based on intelligent algorithm according to claim 1, characterized in that: The specific implementation method of the data preprocessing in step S1 includes outlier detection and missing value filling; The outlier detection method is to construct an anomaly detection graph. For each parameter node, directly related nodes with edge weights ≥ 0.6 in the graph are extracted to form local related clusters. The Mahalanobis distance between real-time data and historical mean is calculated in each related cluster, and a threshold is set. If the calculated Mahalanobis distance threshold is greater than the threshold, it is determined that the parameter is abnormal. All node data in the related cluster determined to be abnormal are removed. The missing value filling is to fill the data position after the outliers are removed, and the bidirectional strong correlation nodes are screened for the missing positions according to the positive order of the graph edge weights. The bidirectional strong correlation nodes are based on the weights of the structural edges in the graph. Starting from the missing value parameters, the parameters with direct causal influence on them are first screened, and the top three are selected in descending order of the edge weights. The parameters affected by their direct feedback are screened again, and the top three are selected in descending order of the edge weights to form a set of 6 strongly correlated nodes including causality and feedback. Finally, the filling value obtained by interpolation is checked to see whether it meets the process constraints. If the filling value is within a reasonable range, it is directly adopted. If it exceeds the upper limit, the maximum value boundary is taken. If it is lower than the lower limit, the minimum value boundary is taken.

4. The method for controlling process parameters of 3PE anti-corrosion layer production based on intelligent algorithm according to claim 1, characterized in that: The step S3 uses an improved multi-objective optimization algorithm to train the standardized data set, takes yield rate and coating uniformity as optimization targets, and outputs the first predicted process parameters containing the optimal parameter range as follows: S31. First, build a deep solution generation network, input the historical standardized parameters, and output the initial candidate solution so that the initial distribution of the population preferentially covers the historical efficient area. The calculation method of the initial candidate solution is: init =clip(σ(W2σ(W1X std +b1)+b2)⊙(x max -x min )+x min ,x min ,x max ), where W2, W1 represent weight matrices, b1, b2 represent bias vectors, σ represents activation function, and x max ,x min are the maximum and minimum values of the physical interval of the process parameters, ⊙ represents element-by-element multiplication, and clip is the interval clipping function; S32. Set an adaptive weight network, input the target statistical characteristics of the current population, and output a dynamic weight ω(t). The dynamic weight output is calculated as follows: Where Ω represents the adaptive weight matrix, σ Y ,σ U are the standard deviation of good products and coating uniformity, σ YU Represents the correlation between two goals; S33. Construct an objective function based on the dynamic weight. The objective function is specifically: f(x)=ω Y (1-Y(x))+ω Y ·U(x)+λ·D HD (x), where Y(x) is the yield rate prediction value of parameter combination x, U(x) represents the coating uniformity prediction value, and D HD (x) represents the diversity metric based on Hausdorff distance, which characterizes the spatial distribution difference between solution x and other solutions in the population, and λ represents the diversity adjustment coefficient; S34, then generate non-traditional parameter combination x based on the target value (Y, U) gen , specifically implemented by using the adversarial network, the generator G inputs noise z and the conditional target output x gen =G(z,Y * ,U * ), where Y * ,U * represents the desired target yield and target coating uniformity; the discriminator D distinguishes the true solution from the generated solution, and the optimization objective is: Among them, P real represents the real data distribution, P (z) represents the probability distribution of noise z, S35. Finally, based on the Pareto frontier solution, the kernel density estimation is used to determine the top 80% continuous intervals of density, and the sub-intervals that violate the process hard constraints are eliminated. After 1000 Monte Carlo simulations, the intervals with target fluctuations ≤ 5% are retained, and the first predicted process parameters containing the optimal interval are output.

5. The method for controlling process parameters of 3PE anti-corrosion layer production based on intelligent algorithm according to claim 4, characterized in that: For the generated solutions in step S34, the generated solutions are replaced by 30% of the solutions with the smallest crowding distance in the population every 10 generations, so that the algorithm can explore the low-probability parameter area.

6. The method for controlling process parameters of 3PE anti-corrosion layer production based on intelligent algorithm according to claim 4, characterized in that: The training loss function of the deep solution generation network in step S31 is: Where MSE is the mean square error, which measures the difference in solutions. This item represents the repulsive force between the generated solution and the existing solution, avoiding duplicate solutions.

7. The method for controlling process parameters of 3PE anti-corrosion layer production based on intelligent algorithm according to claim 1, characterized in that: In step S4, edge weights are annotated based on historical correlation coefficients between parameters combined with expert experience. The specific implementation method for calculating correlation entropy is as follows: first, the Pearson correlation coefficient ρ between parameters is calculated, and the prior knowledge of process experts on edges is introduced to obtain the expert correction coefficient δ. The two are averaged to obtain the final edge weight, and the weight is calculated to obtain correlation entropy. The correlation entropy is calculated as follows: Where N(i) is the set of neighbor nodes, β ij Represents the edge weight from node i to node j.

Images