Additive feeding control method and device for plastic production line, equipment and medium

Abstract

The invention belongs to the technical field of intelligent control, and discloses an additive feeding control method and device for a plastic production line, equipment and a medium, material sensing data are collected in real time in production for preprocessing, key feature data are extracted from the preprocessed data for standardization and normalization processing, and real-time feature vectors are obtained; inputting the real-time feature vector into a prediction model to obtain a first predicted feeding amount of each type of additive, inputting the real-time feature vector and the first predicted feeding amount of each type of additive into a reinforcement learning model to obtain a second predicted feeding amount of each type of additive and an accumulated reward, and when the accumulated reward reaches a preset threshold value, obtaining a second predicted feeding amount of each type of additive; and according to the second predicted feeding amount, various additives are controlled to be fed, so that the predicted feeding amount of various additives can be predicted based on the material sensing data, monitored in real time, of the plastic production line, the predicted feeding amount is further optimized, the adaptive capacity can be improved, and the additive feeding control accuracy is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent control, and in particular relates to an additive feeding control method and device, equipment and storage medium for a plastic production line. Background Art

[0002] In plastic production lines, such as the production of polyvinyl chloride (PVC), some additives such as heat stabilizers, plasticizers, lubricants, fillers, antioxidants, and ultraviolet absorbers are usually added for mixed production. The control of these additives is an important link that affects product quality, production efficiency, and cost.

[0003] Traditional additive dosing control methods usually use a single feedback control mechanism based on PID (proportional-integral-differential) control, which relies on preset parameters and linear system assumptions and is difficult to cope with complex process changes and multivariable dosing control requirements. Especially in key processing stages such as mixing and melting, traditional dosing systems lack adaptive capabilities and are unable to handle changes in raw material characteristics, environmental conditions, and internal disturbances in the system, resulting in inaccurate dosing and excessive dosage fluctuations, which in turn affect product consistency and performance. At the same time, PID control systems rely on manual setting and adjustment of parameters, which increases the difficulty of operation and the risk of human intervention, and reduces overall production efficiency. Summary of the invention

[0004] The purpose of the present invention is to provide a method and device, equipment, and storage medium for controlling additive dosing in a plastic production line, which do not rely on human intervention, can improve the adaptive ability, and thus improve the accuracy of additive dosing control.

[0005] The first aspect of the present invention discloses a method for controlling the feeding of additives into a plastic production line, comprising:

[0006] In the plastic production process, material sensor data of the plastic production line is collected in real time, and the material sensor data is preprocessed to obtain preprocessed data;

[0007] Extracting a plurality of key feature data from the preprocessed data, and performing standardization and normalization processing on the plurality of key feature data to obtain a real-time feature vector;

[0008] The real-time feature vector is input into a pre-trained prediction model to obtain a first predicted feed amount of each type of additive; wherein the real-time feature vector includes the temperature of the current mixed material, the viscosity of the current mixed material, the pressure of the production equipment, the type of the added additive, the feed amount of the added additive, the feed speed of the added additive and the processing time of the added additive; the current mixed material includes raw materials and the added additive;

[0009] Input the real-time feature vector and the first predicted feed amount of each type of additive into a pre-trained reinforcement learning model to obtain the second predicted feed amount of each type of additive and the accumulated reward; wherein the accumulated reward refers to the accumulated reward value of taking the second predicted feed amount under the current state represented by the real-time feature vector;

[0010] When the accumulated reward reaches a preset threshold, the feeding of various additives is controlled according to the second predicted feeding amount.

[0011] In some embodiments, after controlling the feeding of various additives, the method further comprises:

[0012] Obtain multiple actual quality indicators of the finished plastic product after feeding;

[0013] Calculate an average value of deviation values ​​between the plurality of actual quality indicators and their corresponding preset standard indicators, marking the average deviation;

[0014] When the average deviation is greater than a specified difference, covariance analysis is performed on the real-time feature vector to obtain a target feature vector, and the target feature vector is input into the prediction model to obtain a third predicted feed amount of each type of additive;

[0015] Inputting the target feature vector and the third predicted feeding amount of each type of additive into the reinforcement learning model to obtain a fourth predicted feeding amount of each type of additive;

[0016] When feeding next time, various additives are fed according to the fourth predicted feeding amount.

[0017] In some embodiments, after obtaining the actual quality index of the finished plastic product after feeding, the method further includes:

[0018] Calculating a reward function indicator of a reinforcement learning model according to the actual quality indicator;

[0019] When the reward function indicator is less than a preset indicator value, the reinforcement learning model is optimized and trained.

[0020] In some embodiments, calculating a reward function indicator of a reinforcement learning model according to the actual quality indicator includes:

[0021] The reward function indicator of the reinforcement learning model is calculated by the following formula:

[0022]

[0023] Among them, R represents the reward function indicator, y i is the ith actual quality indicator, is the preset standard indicator corresponding to the ith actual quality indicator, and n is the number of actual quality indicators.

[0024] In some embodiments, in the process of controlling the feeding of various additives according to the fourth predicted feeding amount, the method further includes:

[0025] Determine whether the actual quality index of the finished plastic product is abnormal. If abnormal, mark it as an abnormal quality index;

[0026] The corresponding target additive is determined according to the abnormal quality index, and the fourth predicted feeding amount of the target additive is adjusted to obtain a fifth predicted feeding amount, and the target additive is fed according to the fifth predicted feeding amount.

[0027] The second aspect of the present invention discloses an additive feeding control device for a plastic production line, comprising:

[0028] The collection unit is used to collect material sensor data of the plastic production line in real time during the plastic production process;

[0029] A preprocessing unit, used for preprocessing the material sensing data to obtain preprocessing data;

[0030] An extraction unit, used for extracting a plurality of key feature data from the preprocessed data, and performing standardization and normalization processing on the plurality of key feature data to obtain a real-time feature vector;

[0031] A first prediction unit is used to input the real-time feature vector into a pre-trained prediction model to obtain a first predicted feed amount of each type of additive; wherein the real-time feature vector includes the temperature of the current mixed material, the viscosity of the current mixed material, the pressure of the production equipment, the type of the added additive, the feed amount of the added additive, the feed speed of the added additive and the processing time of the added additive; the current mixed material includes raw materials and added additives;

[0032] A first optimization unit is used to input the real-time feature vector and the first predicted feed amount of each type of additive into a pre-trained reinforcement learning model to obtain a second predicted feed amount of each type of additive and a cumulative reward; wherein the cumulative reward refers to the cumulative reward value of taking the second predicted feed amount under the current state represented by the real-time feature vector;

[0033] The first control unit is used to control the feeding of various additives according to the second predicted feeding amount when the accumulated reward reaches a preset threshold.

[0034] In some embodiments, it also includes:

[0035] An acquisition unit, used for acquiring a plurality of actual quality indicators of the finished plastic product after the first control unit controls various additives to be added;

[0036] A calculation unit, used to calculate an average value of deviation values ​​between the plurality of actual quality indicators and their corresponding preset standard indicators, marked as an average deviation;

[0037] An analysis unit, configured to perform covariance analysis on the real-time feature vector to obtain a target feature vector when the average deviation is greater than a specified difference;

[0038] The second prediction unit is further used to input the target feature vector into the prediction model to obtain a third predicted feed amount of each type of additive;

[0039] The second optimization unit is further used to input the target feature vector and the third predicted feeding amount of each type of additive into the reinforcement learning model to obtain a fourth predicted feeding amount of each type of additive;

[0040] The second control unit is used to control the feeding of various additives according to the fourth predicted feeding amount during the next feeding.

[0041] In some embodiments, it also includes:

[0042] A measuring unit, configured to calculate a reward function indicator of a reinforcement learning model according to a plurality of actual quality indicators of the finished plastic product after the obtaining unit obtains the actual quality indicators;

[0043] The optimization training unit is used to perform optimization training on the reinforcement learning model when the reward function indicator is less than a preset indicator value.

[0044] The third aspect of the present invention discloses an electronic device, comprising a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the additive feeding control method for the plastic production line disclosed in the first aspect.

[0045] A fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the additive dosing control method for a plastic production line disclosed in the first aspect.

[0046] The beneficial effect of the present invention lies in that, during the plastic production process, material sensor data of the plastic production line is collected in real time for preprocessing to obtain preprocessing data, multiple key feature data are extracted from the preprocessing data for standardization and normalization to obtain real-time feature vectors; the real-time feature vectors are input into the prediction model to obtain the first predicted feeding amount of each type of additive; wherein the real-time feature vector includes the temperature of the current mixed material, the viscosity of the current mixed material, the pressure of the production equipment, the type of the added additive, the feeding amount of the added additive, the feeding speed of the added additive and the processing time of the added additive; the current mixed material includes raw materials and added additives; then the real-time feature vector and the first predicted feeding amount of each type of additive are input into the reinforcement learning model to obtain the second predicted feeding amount of each type of additive and the accumulated reward, when the accumulated reward reaches a preset threshold, the feeding of each type of additive is controlled according to the second predicted feeding amount, so that the predicted feeding amount of each type of additive can be predicted based on the material sensor data of the plastic production line monitored in real time, and the predicted feeding amount can be further optimized, so that there is no need to rely on human intervention, the adaptive ability can be improved, and the accuracy of additive feeding control can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings herein show specific examples of the technical solutions described in the present invention, and together with the specific implementation methods, constitute a part of the specification, and are used to explain the technical solutions, principles and effects of the present invention.

[0048] Unless otherwise specified or defined, the same reference numerals in different drawings represent the same or similar technical features, and the same or similar technical features may also be represented by different reference numerals.

[0049] Figure 1 It is a flow chart of an additive feeding control method for a plastic production line disclosed in an embodiment of the present invention;

[0050] Figure 2 It is a structural schematic diagram of an additive feeding control device for a plastic production line disclosed in an embodiment of the present invention;

[0051] Figure 3 It is a structural schematic diagram of an electronic device disclosed in an embodiment of the present invention.

[0052] Description of reference numerals:

[0053] 201, acquisition unit; 202, preprocessing unit; 203, extraction unit; 204, first prediction unit; 205, first optimization unit; 206, first control unit; 301, memory; 302, processor. DETAILED DESCRIPTION

[0054] Unless otherwise specified or defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art. In the case of combining the technical solution of the present invention with realistic scenarios, all technical and scientific terms used herein may also have meanings corresponding to the purpose of implementing the technical solution of the present invention. The "first, second..." used herein is only used to distinguish the names and does not represent a specific quantity or order. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0055] It should be noted that when a component is considered to be "fixed to" another component, it can be directly fixed to the other component or there can be a central component; when an component is considered to be "connected to" another component, it can be directly connected to the other component or there can be a central component at the same time; when an component is considered to be "installed on" another component, it can be directly installed on the other component or there can be a central component at the same time. When an component is considered to be "set on" another component, it can be directly set on the other component or there can be a central component at the same time.

[0056] Unless otherwise specified or defined, the "said" and "the" used in this document refer to the technical features or technical contents mentioned or described before the corresponding position, and the technical features or technical contents may be the same as or similar to the technical features or technical contents mentioned therein. In addition, the terms "including" and "having" and any variations thereof used in this document are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0057] The embodiment of the present invention discloses a method for controlling the feeding of additives in a plastic production line, which can be implemented by computer programming. The execution subject of the method can be an electronic device such as a computer, a laptop, a tablet computer, or an additive feeding control device for a plastic production line embedded in an electronic device, and the present invention is not limited to this. In order to facilitate the understanding of the present invention, the specific embodiments of the present invention will be described in more detail below with reference to the drawings of the specification.

[0058] like Figure 1 As shown, the method includes the following steps 110 to 150:

[0059] 110. In the plastic production process, the material sensor data of the plastic production line is collected in real time, and the material sensor data is preprocessed to obtain preprocessed data.

[0060] Among them, material sensor data includes temperature information, viscosity information, pressure information, weighing information, flow information and additive information, etc. Additive information includes the type of additives added, feed amount, feed speed, processing time, etc.

[0061] A variety of sensors can be installed on the plastic production line, such as temperature sensors, viscometer sensors, pressure sensors, weighing sensors, flow sensors, machine vision sensors, etc. The temperature sensor can be specifically an RTD sensor, a thermocouple, etc. The pressure sensor can be specifically a piezoelectric sensor, the flow sensor can be specifically an electromagnetic flowmeter, and the machine vision sensor can be specifically an industrial camera or an industrial barcode scanner, etc. The industrial camera is used to capture images for image recognition to obtain additive information, and the industrial barcode scanner is used to scan QR codes to obtain additive information.

[0062] Specifically, the method of preprocessing the material sensor data to obtain the preprocessed data may include:

[0063] After the collected material sensor data is cleaned, filtered, processed for outliers and missing values, the pre-processed data is obtained. Among them, the data cleaning process mainly converts the units of the material sensor data collected by different sensors to ensure that the temperature, pressure, additive information and other data use consistent measurement units and appropriate data types; the data filtering process mainly uses the Kalman filter to filter the material sensor data after data cleaning, thereby filtering out some sensor data with large noise interference, which has great advantages especially in multi-sensor fusion scenarios and large amounts of time series data; the outlier value processing mainly sets the normal range of material sensor data of different sensors, and identifies and excludes outliers that exceed the normal range. The data that exceeds the normal range can be recorded for manual approval; considering that the timed reporting of data may cause data loss due to the communication terminal, the missing value processing mainly checks whether there are missing values ​​in the collected material sensor data. If there are missing values, the interpolation method is used to supplement them to make the data transition smooth.

[0064] 120. Extract multiple key feature data from the preprocessed data, and standardize and normalize the multiple key feature data to obtain a real-time feature vector.

[0065] In the present invention, the multiple key characteristic data include the temperature and viscosity of the current mixed material (including raw materials and additives), the pressure of the production equipment, and the type, amount, speed and processing time of the additives; wherein,

[0066] Current temperature of the mixed material: Entered as multiple continuous variables, representing the real-time temperature values ​​at different processing locations during the mixing or melting stage, in degrees Celsius (°C).

[0067] Viscosity of the current mixed material: Input as multiple continuous variables, representing the viscosity of the material at different processing positions during the melting or mixing stage, reflecting the rheological properties of the material in real time, in Pascal seconds (Pa s).

[0068] Production equipment pressure: input as multiple continuous variables, representing the real-time pressure inside the production equipment at different processing locations, in Pascal (Pa).

[0069] Type of additives: Input as a coding variable. For different additives, their types can be represented by one-hot encoding. The type of additives can be some of the six types of additives (including heat stabilizers, plasticizers, lubricants, fillers, antioxidants, and ultraviolet absorbers).

[0070] The amount of additives added: input as a discrete variable or continuous variable, representing the actual weight of each additive added each time, accurately reflecting the difference in feeding in each cycle, in kilograms (kg).

[0071] Feeding rate of the added additives: As a continuous variable input, it represents the rate at which various added additives are added. This is a dynamic variable related to time, reflecting the real-time feeding status in the production process.

[0072] Processing time of added additives: As a relative time input, it indicates the cumulative time that various added additives participate in production and processing after being put into use, reflecting the dynamic characteristics of the production process.

[0073] In step 120, multiple key feature data are standardized and normalized to obtain a real-time feature vector, including the following steps S11 to S13 (not shown):

[0074] S11. Standardize each key feature data to obtain standardized data corresponding to each key feature data.

[0075] The standardized formula can be the following formula (1):

[0076]

[0077] Among them, x is the key feature data of the original input, μ is the mean of the key feature data, σ is the standard deviation of the key feature data, and X′ is the standardized data after standardization.

[0078] S12. Scale each standardized data to a numerical range corresponding to each key feature data to obtain normalized data corresponding to each key feature data.

[0079] First, according to the characteristics of different key feature data, the corresponding numerical intervals are delineated, which is suitable for the case where the feature values ​​vary greatly but still vary within a limited interval. Then, each standardized data is scaled to the corresponding numerical interval, as shown in the following formula (2):

[0080]

[0081] Among them, X is the normalized data of the original input, X min and X max are the minimum and maximum values ​​corresponding to the standardized data, respectively, and X′ is the normalized data after normalization.

[0082] S13. The normalized data corresponding to each key feature data are concatenated to form a real-time feature vector.

[0083] The real-time feature vector is recorded as X = [x 1 , x 2 , ..., x 7 ], respectively representing: the temperature of the current mixed material (including raw materials and additives), the viscosity of the current mixed material, the pressure of the production equipment, the type of additives, the amount of additives, the feeding speed of additives and the processing time of additives.

[0084] 130. Input the real-time feature vector into a pre-trained prediction model to obtain the first predicted feed amount of each type of additive.

[0085] The prediction model can be a model based on an artificial neural network (ANN). The task of the ANN model is to output the first predicted feed amount of each type of additive in real time according to the input real-time feature vector. The ANN model can specifically adopt a multilayer perceptron (MLP) structure, and the model structure design is as follows:

[0086] (1) Input layer: a total of 7 nodes, corresponding to the 7 key feature data in the real-time feature vector.

[0087] (2) Hidden layer: including the first hidden layer and the second hidden layer, where:

[0088] The first hidden layer: contains 64 nodes and uses the ReLU activation function h 1 =ReLU(W 1 ·X+b 1 ), where W 1 is the weight matrix from the input layer to the first hidden layer, b 1 is the bias, X represents the real-time feature vector of the input (i.e., the 7 node inputs in the input layer), h 1is the output vector of the first hidden layer.

[0089] The second hidden layer: contains 32 nodes and uses the ReLU activation function h 2 =ReLU(W 2 ·X+b 2 ), where W 2 is the weight matrix from the first hidden layer to the second hidden layer, b 2 is the bias, X represents the output vector of the first hidden layer, h 2 is the output vector of the second hidden layer.

[0090] (3) Output layer: contains 6 nodes, corresponding to the first predicted feed amount of 6 additives (including heat stabilizer, plasticizer, lubricant, filler, antioxidant, and ultraviolet absorber). The output value of each node is the feed amount prediction result. The output calculation formula is Y = W 3 ·h 2 +b 3 , where W 3 is the weight matrix from the second hidden layer to the output layer, b 3 is the bias, and Y is the final predicted feed output. Since the specific feed amount needs to be predicted, the output layer does not use an activation function and directly outputs a continuous value.

[0091] In the process of predictive model training, the mean squared error (MSE) is used as the loss function to measure the difference between the model's prediction results for the sample and the true label. The loss function formula is shown in the following formula (3):

[0092]

[0093] Among them, y i is the true label of the sample, is the prediction result of the model for the sample, and n is the number of samples.

[0094] 140. Input the real-time feature vector and the first predicted feed amount of each type of additive into a pre-trained reinforcement learning model to obtain the second predicted feed amount of each type of additive and the cumulative reward; the cumulative reward refers to the cumulative reward value of taking the second predicted feed amount under the current state represented by the real-time feature vector.

[0095] The reinforcement learning model is used to adjust the feeding strategy, and the first predicted feeding amount of each type of additive output based on the prediction model is further optimized to obtain the second predicted feeding amount of each type of additive. Preferably, the reinforcement learning model can use a deep deterministic policy gradient network (Deep Deterministic Policy Gradient, DDPG). Specifically, in the reinforcement learning task, a Markov decision process (Markov Decision Process, MDP) is defined to model the problem. MDP includes at least state (State, S), action (Action, A) and reward (Reward, R), where,

[0096] The state S describes the environment or situation of the current plastic production line. For example, the conditions in the current production process include the temperature of the current mixed material (including raw materials and additives), the viscosity of the current mixed material, the pressure of the production equipment, the type of additives, the amount of additives, the speed of additives and the processing time of additives, that is, the state S = [x 1 , x 2 , ..., x 7 ] is consistent with the real-time feature vector input into the pre-model.

[0097] Action A corresponds to a specific feeding decision, and each action A represents how much to feed a certain type of additive. For example, the feeding amount of different additives such as heat stabilizers, plasticizers, lubricants, fillers, antioxidants, and ultraviolet absorbers is adjusted separately. Reinforcement learning needs to output multiple actions, that is, A = [a 1 , a 2 , ..., a n ], each a i It represents the second predicted feeding amount of the additive of the i-th type.

[0098] The reward function is used to evaluate the effect of each feed adjustment. Specifically, the reward can be calculated based on the product quality test results fed back by resource modules such as the Quality Management System (QMS) or Enterprise Resource Planning (ERP). The reward function can be set to give positive or negative feedback based on the degree of improvement in product quality. The closer the product quality is to the target standard, the higher the reward value, and vice versa.

[0099] In addition, the DDPG network framework combines the policy network and the deep Q network, which is mainly used to solve the problem of continuous action control. The present invention improves production quality by optimizing the feed amount, and the specific composition is as follows:

[0100] The policy network accepts state 5 = [x1 , x 2 , ..., x 7 ] as input, output an action A = [a 1 , a 2 , ..., a n ], that is, the second predicted feeding amount A after adjusting the first predicted feeding amount of each additive = [a 1 , a 2 , ..., a 6 ], these feed amounts are continuous values, representing the precise amount of different additives at a specific time point. The calculation formula of the strategy network is shown in the following formula (4):

[0101] A=μ(S;θ μ )=f(W n .f(W n-1 ·...f(W 1 ·S+b 1 )+b n-1 )+b n ) (4)

[0102] Among them, W 1 , W 2 , ...W n is the weight matrix of the policy network, representing the connection between neurons in each layer, b 1 , b 2 , ..., b n represents the bias vector of each layer, f is the ReLU activation function, A is the output action vector, represents the second predicted feed amount of the additive, S is the input state (i.e., the real-time feature vector), θ μ Refers to the weight parameters of the policy network, including all weight matrices and bias vectors in the policy network.

[0103] The deep Q network accepts both state S and action A as input, and is used to estimate the expected cumulative reward, i.e., Q value, after executing feeding action A under the current state S. The calculation formula is shown in the following formula (5):

[0104]

[0105] Among them, S represents the state, A represents the action, r represents the immediate reward after executing action A, S′ represents the next state after executing the action, A, represents the action in the next state S′, γ represents the discount factor, which is used to measure the importance of future rewards and has a value range of [0, 1], and θ Q Represents the parameters of the deep Q network, which defines how the deep Q network calculates Q values ​​based on states and actions.

[0106] In summary, the main idea of ​​DDPG is to output actions through the policy network, evaluate these actions through the deep Q network, and then use the gradient descent method to optimize the policy network so that the actions output by the policy network can maximize the Q value (i.e., cumulative reward). The specific training process of DDPG is as follows:

[0107] In the early stage of model training, the policy network will explore, that is, try different feeding adjustment strategies to find the optimal feeding action under different conditions. As training progresses, the policy network will tend to use the best strategy that has been learned, that is, output those feeding actions that can produce the highest reward. In order to maintain the stability of the learning process, DDPG introduces a target network, including a target policy network and a target deep Q network. The parameters of the target network are soft-updated from the parameters of the main network in order to alleviate the instability in the policy update. After each training, the goal of the deep Q network is to minimize the following loss function, as shown in equation (6):

[0108]

[0109] Among them, the target Q value y i Represented as y i =r i +γ·Q(S i+1 ,μ(S i+1 θ μ′ );θ Q′ ), Q(S i+1 ,μ(S i+1 θ μ′ );θ Q′ ) refers to the future Q value calculated by the Q function, θ μ′ and θ Q′ are the parameters of the target policy network and the target Q network, respectively, which are used to improve stability.

[0110] 150. When the accumulated reward reaches a preset threshold, the feeding of various additives is controlled according to the second predicted feeding amount.

[0111] In some other possible embodiments, when the accumulated reward does not reach a preset threshold, the feeding of various types of additives may be controlled according to the first predicted feeding amount.

[0112] As an optional implementation, after controlling the feeding of various additives, the following steps 171 to 175 (not shown) may also be performed:

[0113] 171. Obtain multiple actual quality indicators of the finished plastic product after feeding.

[0114] The multiple quality indicators of plastic finished products include physical characteristic data and quality inspection data. The physical characteristic data include strength, hardness, and density. The quality inspection data include the test results of relevant quality inspection standards (such as impact performance, tensile performance, wear resistance, friction coefficient, tear performance, thermal stability, moisture content, vibration test, salt spray test, aging test, surface resistivity, volume resistivity, breakdown voltage, dielectric constant, composition analysis and other test items).

[0115] 172. Calculate an average value of deviation values ​​between multiple actual quality indicators and their corresponding preset standard indicators, and mark it as average deviation.

[0116] Based on the production plan and process standards, these actual measured quality indicators are compared with the preset standard indicator standards of the product. Finally, based on the difference between the two, the average value is calculated to obtain the average deviation.

[0117] As an optional implementation, after obtaining the actual quality index of the finished plastic product after feeding, the reward function index of the reinforcement learning model can also be calculated based on the actual quality index. When the reward function index is less than the preset index value, the reinforcement learning model is optimized and trained.

[0118] Specifically, the reward function index R can be calculated by the following formula (7):

[0119]

[0120] Among them, y i is the ith actual quality indicator, is the preset standard indicator corresponding to the ith actual quality indicator, and n is the number of actual quality indicators. When R tends to a higher value, it indicates that the feeding strategy optimized by the reinforcement learning model is more effective, otherwise it needs to be adjusted.

[0121] 173. When the average deviation is greater than the specified difference, a covariance analysis is performed on the real-time feature vector to obtain a target feature vector, and the target feature vector is input into a pre-trained prediction model to obtain a third predicted feed amount of each type of additive.

[0122] Among them, the characteristic variables included in the target characteristic vector are part or all of the characteristic variables included in the real-time characteristic vector. Through covariance analysis, it is possible to identify the characteristic variable pairs in the real-time characteristic vector whose mutual influence is greater than the specified degree, and to select and discard these characteristic variable pairs as appropriate, such as deleting the entire characteristic variable pair, or deleting only any one of the characteristic variables, and finally leaving the characteristic variables whose mutual influence is less than the specified degree, and re-optimizing the feeding strategy.

[0123] Specifically, the covariance analysis of the real-time feature vector can be performed using the following formula (8):

[0124]

[0125] Where X i , Y i are the i-th values ​​of the feature variables X and Y respectively; are the means of the feature variables X and Y respectively, and n is the number of feature variables.

[0126] 174. Input the target feature vector and the third predicted feeding amount of each type of additive into a pre-trained reinforcement learning model to obtain the fourth predicted feeding amount of each type of additive.

[0127] 175. When feeding next time, control the feeding of various additives according to the fourth predicted feeding amount.

[0128] Further optionally, in the process of controlling the feeding of various additives according to the fourth predicted feeding amount, the following steps S21 to S22 (not shown) may also be performed:

[0129] S21. Determine whether the actual quality index of the finished plastic product is abnormal. If abnormal, mark it as an abnormal quality index.

[0130] Specifically, the outlier value of the actual quality indicator can be calculated based on the Z-score anomaly detection formula, as shown in the following formula (9):

[0131]

[0132] Where x is the actual quality indicator, μ is the historical data mean of the actual quality indicator, and σ is the historical data standard deviation of the actual quality indicator. If the abnormal value Z exceeds the predetermined threshold, it means that the actual quality indicator is abnormal and is marked as an abnormal quality indicator.

[0133] S22. Determine a corresponding target additive according to the abnormal quality index, adjust the fourth predicted feeding amount of the target additive to obtain a fifth predicted feeding amount, and feed the target additive according to the fifth predicted feeding amount.

[0134] The fourth predicted feed amount of the target additive is adjusted to obtain the fifth predicted feed amount, which may specifically be: identifying the abnormal cause according to the abnormal quality index, and then adjusting the fourth predicted feed amount of the target additive according to the abnormal cause to obtain the fifth predicted feed amount.

[0135] For example, considering that in PVC production, plasticizers are used to improve the flexibility and ductility of the material, but excessive amounts will cause the product to soften excessively, affecting physical properties such as hardness or strength. Therefore, if the abnormal quality indicator is the hardness or strength of the product, the corresponding target additive is the plasticizer. After determining that the hardness or strength of the product is abnormal, the cause of the abnormality can be further identified, such as determining whether the hardness of the finished product exceeds the hardness range required for product production (for example, 65-75ShoreA). Assuming that the hardness of the product is 80Shore A, it exceeds the upper limit of the standard value, that is, the hardness is judged to be too high, and it is inferred that the amount of plasticizer added is too little. Based on this feedback on the cause of the abnormality, the amount of plasticizer added can be automatically adjusted, increasing by 10%-12% from the current fourth predicted amount; conversely, if the hardness is judged to be too low, it is inferred that the amount of plasticizer added is too much, which can be reduced by 10%-12% based on the current fourth predicted amount.

[0136] In some embodiments, if the hardness is still detected to be too high, it may be considered to increase the temperature control during feeding, thereby increasing the diffusion rate of the plasticizer and ensuring a more uniform mixing of the materials.

[0137] For another example, considering that heat stabilizers are used to prevent thermal degradation of PVC during processing, excessive addition will affect the decomposition temperature and processing performance of the material, which may affect the thermal stability of the product. Therefore, if the abnormal quality indicator is the decomposition temperature of the processed PVC material, the corresponding target additive is a heat stabilizer. Further, the cause of the abnormality can be identified, for example, to determine whether the decomposition temperature of the processed PVC material is higher than the standard decomposition temperature. If it is higher than the standard decomposition temperature, it is inferred that the amount of heat stabilizer added is insufficient. For example, during the production process, it is detected that the decomposition temperature of the processed PVC material is low (actually 170°C, while the standard requirement is 180°C). Based on this abnormal cause feedback, the amount of heat stabilizer added can be increased by 6%-7% from the existing fourth predicted amount; conversely, if it is lower than the standard decomposition temperature, it is inferred that the amount of heat stabilizer added is too much, then it is reduced by 6%-7% from the existing fourth predicted amount.

[0138] In some embodiments, if it is analyzed that the change in decomposition temperature during the production process is related to the processing time period, the heat stabilizer can be added in stages. For example, 70% is added in the first 30 minutes of processing, and the remaining 30% is gradually added in the subsequent time to ensure that the decomposition temperature reaches the optimal range.

[0139] For another example, consider that lubricants are used to reduce frictional heat during processing, optimize processing performance, and improve production efficiency. Therefore, if the abnormal quality indicator is frictional heat during processing, the corresponding target additive is lubricant. Further, the cause of the abnormality can be identified, for example, to determine whether the flow performance of the processed product is lower than the standard performance, that is, the fluidity is poor, resulting in excessive frictional heat during the extrusion molding process, affecting the uniformity of the material. If it is lower than the standard performance, the lubricant can be increased by 0.5% from the fourth predicted feed amount in the subsequent feeding; if it is higher than the standard performance, no adjustment is required.

[0140] In some embodiments, the lubricant delivery time can be adjusted based on the friction heat detection data, from delivery at the beginning of processing to delivery 15 minutes before the material begins to be extruded, so as to reduce the friction heat between the material and the machine.

[0141] In summary, the implementation of the embodiments of the present invention can achieve accurate additive placement through real-time analysis and learning of multivariate data in the production process, while having adaptive adjustment capabilities to effectively cope with process fluctuations in production.

[0142] like Figure 2 As shown, the embodiment of the present invention discloses an additive feeding control device for a plastic production line, including a collection unit 201, a preprocessing unit 202, an extraction unit 203, a first prediction unit 204, a first optimization unit 205, and a first control unit 206, wherein:

[0143] The collection unit 201 is used to collect material sensor data of the plastic production line in real time during the plastic production process;

[0144] A preprocessing unit 202, used for preprocessing the material sensing data to obtain preprocessed data;

[0145] An extraction unit 203 is used to extract a plurality of key feature data from the preprocessed data, and perform standardization and normalization processing on the plurality of key feature data to obtain a real-time feature vector;

[0146] The first prediction unit 204 is used to input the real-time feature vector into the pre-trained prediction model to obtain the first predicted feeding amount of each type of additive; wherein the real-time feature vector includes the temperature of the current mixed material, the viscosity of the current mixed material, the pressure of the production equipment, the type of the added additive, the feeding amount of the added additive, the feeding speed of the added additive and the processing time of the added additive; the current mixed material includes the raw material and the added additive;

[0147] The first optimization unit 205 is used to input the real-time feature vector and the first predicted feeding amount of each type of additive into a pre-trained reinforcement learning model to obtain the second predicted feeding amount of each type of additive and the accumulated reward; wherein the accumulated reward refers to the accumulated reward value of taking the second predicted feeding amount under the current state represented by the real-time feature vector;

[0148] The first control unit 206 is used to control the feeding of various additives according to the second predicted feeding amount when the accumulated reward reaches a preset threshold.

[0149] Optionally, the additive feeding control device for the plastic production line further includes the following units not shown in the figure:

[0150] An acquisition unit, used for acquiring a plurality of actual quality indicators of the finished plastic product after the first control unit 206 controls various additives to be added;

[0151] A calculation unit, used to calculate an average value of deviation values ​​between a plurality of actual quality indicators and their corresponding preset standard indicators, marked as an average deviation;

[0152] An analysis unit, used for performing covariance analysis on the real-time feature vector to obtain a target feature vector when the average deviation is greater than a specified difference;

[0153] The second prediction unit is also used to input the target feature vector into the prediction model to obtain the third predicted feeding amount of each type of additive;

[0154] The second optimization unit is further used to input the target feature vector and the third predicted feeding amount of each type of additive into the reinforcement learning model to obtain the fourth predicted feeding amount of each type of additive;

[0155] The second control unit is used to control the feeding of various additives according to the fourth predicted feeding amount during the next feeding.

[0156] Optionally, the additive feeding control device for the plastic production line further includes the following units not shown in the figure:

[0157] A measuring unit is used to calculate a reward function indicator of a reinforcement learning model according to the actual quality indicators after the obtaining unit obtains a plurality of actual quality indicators of the finished plastic product after feeding;

[0158] The optimization training unit is used to optimize the reinforcement learning model when the reward function index is less than the preset index value.

[0159] Optional, measurement unit, specifically used to calculate the reward function indicator of the reinforcement learning model through the following formula:

[0160]

[0161] Among them, R represents the reward function indicator, y i is the ith actual quality indicator, is the preset standard indicator corresponding to the ith actual quality indicator, and n is the number of actual quality indicators.

[0162] Optionally, the additive feeding control device for the plastic production line further includes the following units not shown in the figure:

[0163] A marking unit, used for judging whether the actual quality index of the finished plastic product is abnormal during the process in which the second control unit controls the feeding of various additives according to the fourth predicted feeding amount, and if abnormal, marking it as an abnormal quality index;

[0164] a third optimization unit, configured to determine a corresponding target additive according to the abnormal quality index, and adjust a fourth predicted feed amount of the target additive to obtain a fifth predicted feed amount;

[0165] The third control unit is used to feed the target additive according to a fifth predicted feeding amount.

[0166] like Figure 3 As shown, an embodiment of the present invention discloses an electronic device, including a memory 301 storing executable program codes and a processor 302 coupled to the memory 301;

[0167] The processor 302 calls the executable program code stored in the memory 301 to execute the additive feeding control method for the plastic production line described in the above embodiments.

[0168] An embodiment of the present invention further discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the additive dosing control method for a plastic production line described in the above embodiments.

[0169] The purpose of the above embodiments is to exemplarily reproduce and deduce the technical solution of the present invention, and to fully describe the technical solution, purpose and effect of the present invention. Its purpose is to make the public understand the disclosed content of the present invention more thoroughly and comprehensively, and it does not limit the scope of protection of the present invention.

[0170] The above embodiments are not exhaustive enumerations of the present invention, and there may be multiple other implementations not listed. Any replacement and improvement made without violating the concept of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for controlling the feeding of additives in a plastic production line, characterized in that: include: In the plastic production process, material sensor data of the plastic production line is collected in real time, and the material sensor data is preprocessed to obtain preprocessed data; Extracting a plurality of key feature data from the preprocessed data, and performing standardization and normalization processing on the plurality of key feature data to obtain a real-time feature vector; The real-time feature vector is input into a pre-trained prediction model to obtain a first predicted feed amount of each type of additive; wherein the real-time feature vector includes the temperature of the current mixed material, the viscosity of the current mixed material, the pressure of the production equipment, the type of the added additive, the feed amount of the added additive, the feed speed of the added additive and the processing time of the added additive; the current mixed material includes raw materials and the added additive; Input the real-time feature vector and the first predicted feed amount of each type of additive into a pre-trained reinforcement learning model to obtain the second predicted feed amount of each type of additive and the accumulated reward; wherein the accumulated reward refers to the accumulated reward value of taking the second predicted feed amount under the current state represented by the real-time feature vector; When the accumulated reward reaches a preset threshold, the feeding of various additives is controlled according to the second predicted feeding amount.

2. The additive feeding control method for a plastic production line according to claim 1, characterized in that: After controlling various additives to be fed, the method further comprises: Obtain multiple actual quality indicators of finished plastic products after feeding; Calculate an average value of deviation values ​​between the plurality of actual quality indicators and their corresponding preset standard indicators, marking the average deviation; When the average deviation is greater than a specified difference, a covariance analysis is performed on the real-time feature vector to obtain a target feature vector, and the target feature vector is input into the prediction model to obtain a third predicted feed amount of each type of additive; Inputting the target feature vector and the third predicted feeding amount of each type of additive into the reinforcement learning model to obtain a fourth predicted feeding amount of each type of additive; When feeding next time, various additives are fed according to the fourth predicted feeding amount.

3. The additive feeding control method for a plastic production line according to claim 2, characterized in that: After obtaining the actual quality index of the finished plastic product after feeding, the method further includes: Calculating a reward function indicator of a reinforcement learning model according to the actual quality indicator; When the reward function indicator is less than a preset indicator value, the reinforcement learning model is optimized and trained.

4. The additive feeding control method for a plastic production line according to claim 3, characterized in that: According to the actual quality index, the reward function index of the reinforcement learning model is calculated, including: The reward function indicator of the reinforcement learning model is calculated by the following formula: Among them, R represents the reward function indicator, y i is the ith actual quality indicator, is the preset standard indicator corresponding to the ith actual quality indicator, and n is the number of actual quality indicators.

5. The additive feeding control method for a plastic production line according to any one of claims 2 to 4, characterized in that: In the process of controlling the feeding of various additives according to the fourth predicted feeding amount, the method further includes: Determine whether the actual quality index of the finished plastic product is abnormal. If abnormal, mark it as an abnormal quality index; The corresponding target additive is determined according to the abnormal quality index, and the fourth predicted feeding amount of the target additive is adjusted to obtain a fifth predicted feeding amount, and the target additive is fed according to the fifth predicted feeding amount.

6. An additive feeding control device for a plastic production line, characterized in that: include: The collection unit is used to collect material sensor data of the plastic production line in real time during the plastic production process; A preprocessing unit, used for preprocessing the material sensing data to obtain preprocessing data; An extraction unit, used for extracting a plurality of key feature data from the preprocessed data, and performing standardization and normalization processing on the plurality of key feature data to obtain a real-time feature vector; A first prediction unit is used to input the real-time feature vector into a pre-trained prediction model to obtain a first predicted feed amount of each type of additive; wherein the real-time feature vector includes the temperature of the current mixed material, the viscosity of the current mixed material, the pressure of the production equipment, the type of the added additive, the feed amount of the added additive, the feed speed of the added additive and the processing time of the added additive; the current mixed material includes raw materials and added additives; A first optimization unit is used to input the real-time feature vector and the first predicted feed amount of each type of additive into a pre-trained reinforcement learning model to obtain a second predicted feed amount of each type of additive and a cumulative reward; wherein the cumulative reward refers to the cumulative reward value of taking the second predicted feed amount under the current state represented by the real-time feature vector; The first control unit is used to control the feeding of various additives according to the second predicted feeding amount when the accumulated reward reaches a preset threshold.

7. The additive feeding control device for a plastic production line according to claim 6, characterized in that: Also includes: An acquisition unit, used for acquiring a plurality of actual quality indicators of the finished plastic product after the first control unit controls various additives to be added; A calculation unit, used to calculate an average value of deviation values ​​between the plurality of actual quality indicators and their corresponding preset standard indicators, marked as an average deviation; An analysis unit, configured to perform covariance analysis on the real-time feature vector to obtain a target feature vector when the average deviation is greater than a specified difference; The second prediction unit is further used to input the target feature vector into the prediction model to obtain a third predicted feed amount of each type of additive; The second optimization unit is further used to input the target feature vector and the third predicted feeding amount of each type of additive into the reinforcement learning model to obtain a fourth predicted feeding amount of each type of additive; The second control unit is used to control the feeding of various additives according to the fourth predicted feeding amount during the next feeding.

8. The additive feeding control device for a plastic production line according to claim 7, characterized in that: Also includes: A measuring unit, configured to calculate a reward function indicator of a reinforcement learning model according to a plurality of actual quality indicators of the finished plastic product after the obtaining unit obtains the actual quality indicators; The optimization training unit is used to perform optimization training on the reinforcement learning model when the reward function indicator is less than a preset indicator value.

9. An electronic device, characterized in that: It comprises a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the additive feeding control method for the plastic production line according to any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the additive dosing control method for a plastic production line according to any one of claims 1 to 5.

Images