Intelligent production management system and method for dry polyacrylamide powder

Abstract

The invention relates to an intelligent production management system and method for polyacrylamide dry powder, and the system comprises a data collection and preprocessing module, an intelligent production control module, and a quality monitoring and optimization module. And continuously optimizing the quality prediction model and the production control strategy. The feedback optimization mechanism enables the production management system to continuously learn and improve, adapts to various changes and challenges in the production process, and improves the adaptability and intelligence of the production management system.

Description

Technical Field

[0001] The present invention relates to the technical field of polyacrylamide dry powder production, and in particular to an intelligent production management system and method for polyacrylamide dry powder, which aims to optimize the production process of polyacrylamide dry powder through innovative algorithm models and intelligent control means, improve product quality stability, reduce production costs, and achieve high efficiency, safety and sustainability of the production process. Background Art

[0002] As an important high molecular polymer, polyacrylamide powder is widely used in many fields such as water treatment, oil extraction, papermaking, etc. Its production process involves complex chemical reactions, material mixing, polymerization reaction control, drying, crushing, packaging and other processes. It has the characteristics of complex process, strict requirements on production conditions, and difficult quality control.

[0003] Traditional polyacrylamide dry powder production management mainly relies on manual experience and simple automatic control systems. Production operators set process parameters such as reaction temperature, pressure, material flow rate, etc. based on experience, and conduct manual inspections and adjustments during the production process. This approach has many disadvantages: First, manual experience is difficult to accurately grasp the complex chemical reaction process and the relationship between multiple variables, resulting in inaccurate process parameter settings and large fluctuations in product quality, such as uneven molecular weight distribution and poor solubility; second, manual inspections have defects such as long time intervals and incomplete detection, making it difficult to detect abnormal conditions in the production process in a timely manner, such as abnormal temperature rise in the reactor, material leakage, etc., which can easily cause safety accidents and production interruptions; third, traditional management methods lack the ability to deeply mine and analyze production data, and cannot optimize production processes and predict equipment failures based on historical production data, making it difficult to achieve intelligent decision-making and refined management of the production process.

[0004] In recent years, with the development of intelligent manufacturing technology, some enterprises have begun to try to apply automation and information technology to the production of polyacrylamide dry powder, but they still face many challenges. The existing intelligent production management system has deficiencies in dealing with the nonlinear, time-varying characteristics and multi-source heterogeneous data fusion in the production process of polyacrylamide, resulting in limited accuracy and adaptability of the control model; some studies only focus on the intelligent improvement of a single production link (such as polymerization reaction control or drying process control), and fail to optimize and integrate from the global perspective of the entire production process, making it difficult to achieve an overall improvement in production efficiency and product quality; in addition, when dealing with uncertain factors such as raw material quality fluctuations and changes in market demand, the current production management method lacks effective robustness and flexibility, and cannot adjust the production strategy in time to ensure the economic benefits and market competitiveness of the enterprise. Therefore, there is an urgent need for an innovative intelligent production management system and method for polyacrylamide dry powder to overcome the above difficulties and promote the intelligent transformation and upgrading of the polyacrylamide dry powder production industry. Summary of the invention

[0005] The present invention provides a polyacrylamide dry powder intelligent production management system and method, which effectively solves the problems in the prior art by designing a unique algorithm model and an innovative data processing flow.

[0006] A polyacrylamide dry powder intelligent production management system comprises: a data acquisition and preprocessing module, which is used to deploy high-precision sensors at various key links on the polyacrylamide dry powder production line to comprehensively collect production data, and transmit the collected data to a data processing center through a high-speed industrial Ethernet for preprocessing operations, wherein the preprocessing operations include using an outlier detection algorithm based on deep learning to detect and repair outliers and standardizing all sensor data; an intelligent production control module, which is used to control the production process based on the preprocessed data using an algorithm combining multivariable predictive control with reinforcement learning, and enhance the robustness of the control strategy through a robust optimization method; a quality monitoring and optimization module, which is used to monitor and analyze the product quality of the polyacrylamide dry powder in real time, optimize and adjust the production process based on quality prediction results and actual detection data, and continuously optimize the quality prediction model and control strategy using quality data feedback.

[0007] In the data acquisition and preprocessing module, the sensors installed in the raw material storage and transportation link include mass flowmeter, liquid level sensor and moisture sensor, which are respectively used to accurately measure the flow rate, tank liquid level and water content of the raw materials, wherein the flow rate accuracy is ±0.1kg / h, the liquid level accuracy is ±0.05m, and the water content accuracy is ±0.1%; the sensors installed in the polymerization reactor include temperature sensor, pressure sensor, viscosity sensor and online molecular weight meter, which are respectively used to monitor the key parameters in the reaction process in real time, with temperature accuracy of ±0.1°C, pressure accuracy of ±0.01MPa, viscosity accuracy of ±1mPa·s, and molecular weight meter accuracy of ±5%; the sensors set in the drying equipment include temperature sensor, humidity sensor and material flow sensor, which are used to monitor the state of the drying process, with temperature accuracy of ±1°C, humidity accuracy of ±2% RH, and material flow accuracy of ±0.2kg / h; the sensors equipped in the crushing and packaging link include particle size analyzer and weight sensor, which are used to ensure that the product particle size and packaging weight meet the requirements, with the particle size analyzer accuracy of ±5μm and the weight sensor accuracy of ±0.05kg.

[0008] The outlier detection algorithm based on deep learning adopted by the data acquisition and preprocessing module is to construct a deep autoencoder network. The deep autoencoder network includes 3 hidden layers, which contain 64, 32 and 16 neurons respectively. ReLU is used as the activation function. The feature representation of the training data is trained by inputting normal production data into the network, and then the data to be detected is input into the trained network for decoding and reconstruction, and the reconstruction error is calculated. If the reconstruction error exceeds the set threshold, the data is judged as an outlier, and a repair method based on data distribution characteristics is used to repair the outlier.

[0009] In the intelligent production control module, the multivariable predictive control (MPC) algorithm is used to model and control the core process of polyacrylamide dry powder production, taking into account the mutual coupling relationship between multiple variables such as reaction temperature, pressure, material flow, etc., and establishing a prediction model with product quality indicators as the control target, and solving the optimal control sequence in the future period through rolling optimization; the reinforcement learning algorithm is used to dynamically optimize and adjust the MPC control strategy, construct an intelligent agent based on a deep Q network (DQN), and use the real-time status information in the production process as the input of the intelligent agent. The intelligent agent learns the optimal control strategy adjustment rules through interaction with the production environment; the robust optimization method is used to introduce the interval representation of uncertain parameters in the MPC model, and obtain a control strategy that can still maintain good performance within a certain uncertainty range by solving the robust optimization problem.

[0010] When the multivariable predictive control (MPC) algorithm is applied to a polymerization reaction process, the reaction temperature, monomer concentration, and initiator dosage are selected as control variables, and the product molecular weight is selected as the controlled variable. A state space model is established, and the upper and lower limits and change rate limits of the control variables, as well as the target value and allowable deviation range of the controlled variables are set. The prediction time domain of the MPC algorithm is set to 10 minutes, and the control time domain is set to 3 minutes. By solving the quadratic programming problem, the optimal control sequence for a period of time in the future is obtained.

[0011] In the reinforcement learning algorithm, the real-time status information in the production process is digitized and characterized to construct a state space, including expressing the raw material quality fluctuation as the percentage of raw material purity deviation, and the equipment operation status change as the failure rate of key equipment, etc. The intelligent agent selects actions through the deep Q network according to the current state. The actions include adjusting the control parameters in the MPC model, switching backup equipment, etc. According to the execution results of the actions, the environment gives the intelligent agent corresponding reward feedback. The reward function comprehensively considers factors such as the achievement of product quality indicators, the improvement of production efficiency, and the reduction of energy consumption.

[0012] In the quality monitoring and optimization module, a quality prediction model based on support vector machine regression (SVR) is adopted, and the polymerization reaction temperature, initiator dosage, raw material purity, etc. are selected as input variables, and the characteristic parameters of the product molecular weight distribution are used as output variables. The SVR model is trained by a large amount of historical data, and the radial basis function (RBF) is selected as the kernel function. The kernel function parameters and penalty factors are optimized by the cross-validation method to accurately predict the changing trend of the product molecular weight distribution according to the current polymerization reaction conditions and raw material information, and the prediction error is within ±10%.

[0013] When the quality monitoring and optimization module optimizes and adjusts the production process according to the quality prediction results and the actual detection data, when it is predicted that the molecular weight distribution of the product may exceed the target range, the intelligent production control module adjusts the process parameters such as the temperature and initiator dosage of the polymerization reaction; when the quality data feedback is used to continuously optimize the control strategy of the SVR model and the intelligent agent, if it is found that there is a large deviation between the actual product quality and the predicted result, the cause of the deviation is analyzed, the parameters of the SVR model are adjusted in a targeted manner, and the control strategy of the intelligent agent is updated.

[0014] On the other hand, the present invention also provides a polyacrylamide dry powder intelligent production management method, comprising the following steps:

[0015] In the data collection and preprocessing step, multi-source heterogeneous data are comprehensively collected from the raw material supply, polymerization reaction, drying, crushing and packaging of polyacrylamide dry powder production, and the collected data are preprocessed using an outlier detection algorithm based on deep learning;

[0016] In the intelligent production control step, the multivariable predictive control (MPC) algorithm is used to model and control the core process of polyacrylamide dry powder production, the reinforcement learning algorithm is introduced to dynamically optimize the MPC control strategy, and the robust optimization method is used to enhance the robustness of the control strategy; in the quality monitoring and optimization step, a real-time monitoring system for the quality of polyacrylamide dry powder products is established, and a quality prediction model based on support vector machine regression (SVR) is used for quality prediction. The production process is optimized and adjusted according to the quality prediction results and actual test data, and the quality data feedback is used to continuously optimize the SVR model and the control strategy of the intelligent agent.

[0017] In the data collection and preprocessing steps, the flow, liquid level and water content data of the raw materials are collected every 10 minutes in the raw material supply link; in the polymerization reaction link, the temperature, pressure, viscosity, molecular weight and other data in the reaction process are collected once per second; in the drying link, the temperature, humidity and material flow data are collected once per second; in the crushing and packaging link, the particle size and weight data are collected once for each batch of packaging; a deep autoencoder network is constructed using an outlier detection algorithm based on deep learning, and the network structure includes 3 hidden layers, which contain 64, 32, and 16 neurons respectively. ReLU is used as the activation function. The feature representation of the learning data is trained by inputting normal production data into the network, and then the data to be detected is input into the trained network for decoding and reconstruction, and the reconstruction error is calculated. If the reconstruction error exceeds the set threshold, the data is judged to be an outlier. The outlier is repaired by a repair method based on data distribution characteristics, and all data are standardized.

[0018] Beneficial technical effects:

[0019] A comprehensive data collection system was built to obtain multi-source heterogeneous data from all aspects of polyacrylamide dry powder production. An outlier detection and repair algorithm based on deep learning was used to effectively identify and process outliers in the data, avoiding the interference of abnormal data on production control and quality prediction model training. At the same time, the multi-source data was standardized so that it could be integrated and analyzed under the same mathematical framework, providing high-quality data input for subsequent model training, significantly improving the prediction accuracy and control effect of the model. An algorithm combining multivariable predictive control (MPC) with reinforcement learning was proposed for polyacrylamide dry powder production process control. MPC can accurately model and control the core links of the production process, consider the coupling relationship between multiple variables, and achieve effective control of product quality indicators; reinforcement learning can optimize and adjust the MPC control strategy according to the real-time dynamic information in the production process, and achieve effective response to uncertain factors and dynamic decision-making. This hybrid algorithm gives full play to the advantages of the two methods, overcomes the limitations of traditional production control methods in dealing with complex processes and dynamic problems, and improves the accuracy and flexibility of production. The robust optimization method is introduced to enhance the robustness of the control strategy, and the influence of uncertain factors such as raw material supply and equipment performance in the production process of polyacrylamide dry powder is considered. By solving the robust optimization problem, a control strategy that can still maintain good performance within a certain uncertainty range is obtained, which effectively reduces the interference of uncertain factors on the production process and improves the stability and reliability of production. A quality prediction model based on machine learning is established, which can accurately predict the changing trend of product quality based on production process data and detect potential quality problems in advance. Through the combination of online quality detection equipment and quality prediction models, real-time monitoring and analysis of product quality are realized, providing a timely and accurate basis for the optimization and adjustment of production processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 :It is the working principle flow chart of the present invention DETAILED DESCRIPTION

[0021] Example 1

[0022] 1. System Architecture

[0023] Data acquisition and preprocessing module:

[0024] High-precision sensors are deployed at each key link of the polyacrylamide dry powder production line to comprehensively collect production data. In the raw material storage and transportation link, mass flow meters, liquid level sensors and moisture sensors are installed to accurately measure the flow rate (accuracy of ±0.1kg / h), tank liquid level (accuracy of ±0.05m) and water content (accuracy of ±0.1%) of raw materials (acrylamide monomer, initiator, cross-linking agent, etc.); in the polymerization reactor, temperature sensors (accuracy of ±0.1℃), pressure sensors (accuracy of ±0.01MPa), viscosity sensors (accuracy of ±1mPa·s) and online molecular weight determination instruments (accuracy of ±5%) are installed to monitor the key parameters in the reaction process in real time; in the drying equipment, temperature sensors (accuracy of ±1℃), humidity sensors (accuracy of ±2%RH) and material flow sensors (accuracy of ±0.2kg / h) are set to monitor the state of the drying process; in the crushing and packaging link, particle size analyzers (accuracy of ±5μm) and weight sensors (accuracy of ±0.05kg) are equipped to ensure that the product particle size and packaging weight meet the requirements.

[0025] The collected data is transmitted to the data processing center via high-speed industrial Ethernet, where it is preprocessed. An outlier detection algorithm based on deep learning is used, such as building a deep autoencoder network to learn and encode the features of normal data, and then the data to be detected is input into the network for decoding and reconstruction, and the reconstruction error is calculated. If the reconstruction error exceeds the set threshold, the data is judged as an outlier, and a repair method based on data distribution characteristics is used, such as replacing it with the weighted average of neighboring normal data points. All sensor data are standardized so that the mean of the data is 0 and the variance is 1, which is convenient for subsequent machine learning model processing.

[0026] Intelligent production control module:

[0027] Based on the preprocessed data, an innovative algorithm combining multivariable predictive control and reinforcement learning is used to control the production process. First, the multivariable predictive control (MPC) algorithm is used to model and control the core processes of polyacrylamide dry powder production (such as polymerization reaction and drying process). The MPC algorithm considers the mutual coupling relationship between multiple variables such as reaction temperature, pressure, material flow, etc., takes product quality indicators (such as molecular weight, solubility, solid content, etc.) as control targets, establishes a prediction model, and solves the optimal control sequence for a period of time in the future through rolling optimization. For example, during the polymerization reaction, the MPC model predicts the change trend of the product molecular weight in the next few minutes based on the current reaction temperature, monomer concentration, initiator dosage and other parameters, and calculates the optimal adjustment strategy for the corresponding temperature, initiator flow and other control variables to ensure that the molecular weight is stable within the target range.

[0028] Reinforcement learning algorithms are introduced to dynamically optimize and adjust the MPC control strategy. An intelligent agent based on a deep Q network (DQN) is constructed, and the real-time status information of the production process (such as raw material quality fluctuations, changes in equipment operating status, market demand adjustments, etc.) is used as the input of the intelligent agent. The intelligent agent learns the optimal control strategy adjustment rules through interaction with the production environment. For example, when the purity of acrylamide monomer in the raw materials decreases, the intelligent agent decides whether to adjust the polymerization reaction temperature, initiator dosage, and temperature and time of the drying process based on the current production status and quality index deviation to ensure that the product quality is not greatly affected. During the training process, a technology based on priority experience playback is used to prioritize the storage and learning of experience data that is of great value to the improvement of the intelligent agent strategy, thereby improving learning efficiency and convergence speed.

[0029] In addition, considering the uncertainty factors in the production process of polyacrylamide dry powder, the robust optimization method is used to enhance the robustness of the control strategy. The interval representation of uncertain parameters is introduced into the MPC model, such as the fluctuation range of raw material supply time and the uncertainty of equipment performance parameters. By solving the robust optimization problem, a control strategy that can still maintain good performance within a certain uncertainty range is obtained, thereby improving the anti-interference ability of the production process to uncertain factors.

[0030] Quality monitoring and optimization module:

[0031] Monitor and analyze the product quality of polyacrylamide dry powder in real time. Use online quality detection equipment (such as gel permeation chromatograph, ion chromatograph, etc.) to obtain key quality indicator data such as product molecular weight distribution and ionicity, and conduct comprehensive analysis in combination with production process data (such as reaction conditions, raw material information, etc.). Use a quality prediction model based on machine learning, such as building a support vector machine regression (SVR) model, taking process parameters, raw material characteristics and other data in the production process as input, and product quality indicators as output. The model is trained with a large amount of historical data to enable it to accurately predict the changing trend of product quality. For example, based on the current polymerization reaction temperature, initiator dosage, and raw material purity information, the molecular weight distribution range and solubility index of the product are predicted.

[0032] Based on the quality prediction results and actual test data, the production process is optimized and adjusted. When it is predicted that the product quality may deviate, the corresponding process parameters are adjusted through the intelligent production control module, such as optimizing the polymerization reaction conditions, adjusting the drying process parameters, etc., to improve the stability of product quality. At the same time, the quality data feedback is used to continuously optimize the quality prediction model and control strategy. For example, if it is found that there is a large deviation between the actual product quality and the predicted result, the cause of the deviation is analyzed (such as changes in the impurity content in the raw materials, equipment aging, etc.), the parameters of the SVR model are adjusted in a targeted manner, and the control strategy of the intelligent agent is updated to improve the accuracy and adaptability of the production management system.

[0033] 2. Methodology

[0034] Data collection and preprocessing steps:

[0035] Comprehensively collect multi-source heterogeneous data from the raw material supply, polymerization reaction, drying, crushing and packaging of polyacrylamide dry powder production. In the raw material supply link, the flow, liquid level and water content data of the raw materials are collected every 10 minutes; in the polymerization reaction link, the temperature, pressure, viscosity, molecular weight and other data during the reaction process are collected once per second; in the drying link, the temperature, humidity and material flow data are collected once per second; in the crushing and packaging link, the particle size and weight data are collected once for each batch of packaging.

[0036] The collected data is preprocessed using an outlier detection algorithm based on deep learning. A deep autoencoder network is constructed. The network structure includes three hidden layers, which contain 64, 32, and 16 neurons respectively, and ReLU is used as the activation function. Normal production data is input into the network for training to learn the feature representation of the data. Then the data to be detected is input into the trained network for decoding and reconstruction, and the reconstruction error is calculated. If the reconstruction error exceeds the set threshold (such as 0.1), the data is judged to be an outlier. For outliers, a repair method based on data distribution characteristics is used. For example, for traffic data, it is replaced by the weighted average of the five normal data points before and after. The weight is determined according to the distance between the data point and the outlier. The closer the distance, the greater the weight. All data are standardized so that the data can be analyzed at the same scale.

[0037] Intelligent production control steps:

[0038] The core process of polyacrylamide dry powder production is modeled and controlled using the multivariable predictive control (MPC) algorithm. Taking the polymerization process as an example, the reaction temperature, monomer concentration, and initiator dosage are selected as control variables, and the product molecular weight is selected as the controlled variable to establish a state space model. According to the production process requirements and equipment limitations, the upper and lower limits and change rate limits of the control variables, as well as the target value and allowable deviation range of the controlled variables are set. The prediction time domain of the MPC algorithm is set to 10 minutes, and the control time domain is set to 3 minutes. By solving the quadratic programming problem, the optimal control sequence in the future period is obtained, that is, the adjustment strategy of the control variables such as reaction temperature and initiator dosage.

[0039] The reinforcement learning algorithm is introduced to dynamically optimize the MPC control strategy. The real-time state information in the production process is digitized and characterized to construct the state space. For example, the raw material quality fluctuation is expressed as the percentage of raw material purity deviation, and the equipment operation state change is expressed as the failure rate of key equipment (such as reactors, dryers, etc.). According to the current state, the agent selects actions through the deep Q network. The actions include adjusting the control parameters in the MPC model and switching backup equipment. According to the execution results of the actions, the environment gives the agent corresponding reward feedback. The reward function comprehensively considers factors such as the achievement of product quality indicators, the improvement of production efficiency, and the reduction of energy consumption. For example, if the agent's actions stabilize the product molecular weight within the target range, and the production efficiency is increased by 10%, and the energy consumption is reduced by 5%, a higher reward is given. Through a large number of training iterations, the agent learns the optimal control strategy adjustment rules to achieve dynamic optimization of the production process.

[0040] The robust optimization method is used to enhance the robustness of the control strategy. In the MPC model, uncertain parameters such as raw material supply time and equipment performance parameters are expressed as interval numbers. For example, the raw material supply time is expected to be between 2-4 hours, which is expressed as an interval [2,4]. By solving the robust optimization problem, a control strategy that can maintain good performance within the fluctuation range of uncertain parameters is obtained. For example, even if the raw material supply time is delayed to 4 hours, the control strategy can still ensure stable product quality and continuous production process. Quality monitoring and optimization steps:

[0041] A real-time monitoring system for the quality of polyacrylamide dry powder products was established. The key quality indicator data such as the molecular weight distribution and ionicity of the product were collected every 30 minutes using online quality detection equipment, and comprehensive analysis was performed in combination with the production process data. A quality prediction model based on support vector machine regression (SVR) was adopted, and the polymerization reaction temperature, initiator dosage, raw material purity, etc. were selected as input variables, and the characteristic parameters of the product molecular weight distribution (such as weight average molecular weight, number average molecular weight, molecular weight dispersion coefficient) were used as output variables. The SVR model was trained using a large amount of historical data, and the radial basis function (RBF) was selected as the kernel function, and the kernel function parameters and penalty factors were optimized using the cross-validation method. For example, after training, the SVR model can accurately predict the changing trend of the product molecular weight distribution based on the current polymerization reaction conditions and raw material information, with a prediction error within ±10%.

[0042] According to the quality prediction results and actual test data, the production process is optimized and adjusted. When it is predicted that the molecular weight distribution of the product may exceed the target range, the process parameters such as the temperature and initiator dosage of the polymerization reaction are adjusted through the intelligent production control module. For example, if it is predicted that the molecular weight of the product will be too high, the initiator dosage is appropriately reduced and the reaction time is extended to adjust the molecular weight to the target range. At the same time, the control strategy of the SVR model and the intelligent agent is continuously optimized using quality data feedback. For example, if it is found that there is a large deviation between the actual product quality and the predicted result, the cause of the deviation is analyzed, such as the change in the impurity content in the raw materials leading to a change in the reaction activity, and the parameters of the SVR model are adjusted in a targeted manner, such as re-optimizing the kernel function parameters and the penalty factor, and updating the control strategy of the intelligent agent, such as adjusting the response rules to fluctuations in raw material quality, to improve the accuracy and adaptability of the production management system.

[0043] 3. Innovation

[0044] Innovative data collection and processing methods:

[0045] A comprehensive data collection system has been constructed, which can obtain multi-source heterogeneous data from all aspects of polyacrylamide dry powder production, including raw material data, reaction process data, product quality data, etc., providing a rich data foundation for intelligent production management. Compared with traditional data collection methods, the data collection scope of the present invention is wider, the data types are more complete, and the actual situation of the production process can be more comprehensively reflected.

[0046] The outlier detection and repair algorithm based on deep learning is used to effectively identify and process outliers in the data, avoiding the interference of abnormal data on production control and quality prediction model training. At the same time, multi-source data is standardized so that it can be integrated and analyzed under the same mathematical framework, providing high-quality data input for subsequent model training, significantly improving the prediction accuracy and control effect of the model.

[0047] Intelligent control strategy innovation based on MPC and reinforcement learning:

[0048] An algorithm combining multivariable predictive control (MPC) and reinforcement learning is proposed for the control of the production process of polyacrylamide dry powder. MPC can accurately model and control the core links of the production process, consider the coupling relationship between multiple variables, and achieve effective control of product quality indicators; reinforcement learning can optimize and adjust the MPC control strategy according to the real-time dynamic information in the production process, and achieve effective response and dynamic decision-making to uncertain factors. This hybrid algorithm gives full play to the advantages of the two methods, overcomes the limitations of traditional production control methods in dealing with complex processes and dynamic problems, and improves the accuracy and flexibility of production. The robust optimization method is introduced to enhance the robustness of the control strategy, and the influence of uncertain factors such as raw material supply and equipment performance in the production process of polyacrylamide dry powder is considered. By solving the robust optimization problem, a control strategy that can still maintain good performance within a certain uncertainty range is obtained, which effectively reduces the interference of uncertain factors on the production process and improves the stability and reliability of production.

[0049] Innovation of intelligent mechanism for quality monitoring and optimization:

[0050] A quality prediction model based on machine learning has been established, which can accurately predict the changing trend of product quality based on production process data and discover potential quality problems in advance. By combining online quality inspection equipment with the quality prediction model, real-time monitoring and analysis of product quality is achieved, providing a timely and accurate basis for the optimization and adjustment of production processes.

[0051] Using quality data feedback, the quality prediction model and production control strategy are continuously optimized. This feedback optimization mechanism enables the production management system to continuously learn and improve, adapt to various changes and challenges in the production process, and improve the adaptability and intelligence of the production management system.

[0052] Example 2

[0053] (I) Data acquisition module

[0054] 1. Production process data collection

[0055] Sensors are installed at various key links on the production line, including temperature sensors, pressure sensors, flow sensors, concentration sensors, etc., to collect real-time data such as temperature, pressure, material flow, monomer concentration in the polymerization reactor, as well as temperature, humidity, material flow rate and other data during the drying process.

[0056] 2. Raw material data collection

[0057] Carry out quality inspection on raw materials (acrylamide monomer, initiator, additives, etc.) entering the production line, collect data such as purity, impurity content, particle size, etc., and record information such as batch and supplier of raw materials.

[0058] 3. Product quality data collection

[0059] Through online quality testing equipment, the quality indicators of the produced polyacrylamide dry powder, such as molecular weight, hydrolysis degree, solubility, solid content, etc., are tested in real time.

[0060] (II) Data preprocessing module

[0061] 1. Data cleaning and calibration

[0062] The collected sensor data and quality inspection data are cleaned to remove abnormal values ​​(such as erroneous data caused by sensor failure or detection error). At the same time, the data is calibrated according to the sensor calibration parameters and quality inspection standards to ensure the accuracy of the data.

[0063] 2. Data normalization and feature extraction

[0064] Normalize production data and quality data of different magnitudes to put them in the same numerical range. Through data analysis and domain knowledge, extract key features related to product quality and production efficiency from production process data, such as the correlation between reaction temperature and molecular weight, the effect of drying temperature on product solubility, etc.

[0065] (III) Intelligent control and optimization module

[0066] 1. Control model construction

[0067] According to the characteristics and requirements of polyacrylamide dry powder production, an intelligent control model is constructed to achieve the goal of stable production of high-quality products, while taking into account multiple factors such as production cost, energy consumption, as well as complex constraints such as raw material quality changes and equipment status.

[0068] 2. New algorithm model driven control and optimization strategy generation

[0069] The following three unreported algorithm models are used to realize the generation of intelligent control and optimization strategies. Appropriate control instructions are calculated based on real-time collected and preprocessed data to adjust various parameters in the production process.

[0070] (IV) Execution module

[0071] 1. Production equipment control

[0072] The control instructions generated by the intelligent control module are sent to the corresponding production equipment control system, such as adjusting the heating and cooling systems of the polymerization reactor, controlling the flow of the material conveying pump, adjusting the temperature and wind speed of the drying equipment, etc., so as to achieve precise control of the production process.

[0073] 2. Feedback and Adjustment

[0074] Monitor the operating status of production equipment and changes in product quality in real time, and pass feedback information to the intelligent control module. The intelligent control module adjusts the control strategy in time according to the feedback information to ensure the stability of the production process and the consistency of product quality.

[0075] 4. Management Solutions

[0076] (I) Neural network control model based on quantum particle swarm optimization

[0077] 1. Model Principle

[0078] The quantum particle swarm optimization algorithm (QPSO) combines the advantages of quantum theory and particle swarm optimization algorithm. The quantum bits in quantum theory have superposition and entanglement characteristics, which make the position representation of particles in the search space richer and increase the diversity of the search.

[0079] The particle swarm optimization algorithm finds the optimal solution by sharing information and updating positions between particles. Applying it to the training process of neural networks can optimize the weights and thresholds of neural networks, enabling neural networks to better handle the complex nonlinear relationships in the production of polyacrylamide dry powder.

[0080] 2. Algorithm Implementation

[0081] First, a neural network structure suitable for the production control of polyacrylamide dry powder is constructed. The input layer nodes include production process data (such as temperature, pressure, concentration, etc.) and raw material data, and the output layer nodes are the control parameters of the production equipment (such as reactor temperature setting value, material flow setting value, etc.). The quantum particle swarm optimization algorithm is used for neural network training to initialize the particle position (corresponding to the weight and threshold of the neural network) and speed and other parameters. In each iteration, the particle updates its position based on its own experience and the optimal information of the group, and uses operations such as quantum revolving doors to update the quantum state of the particle to improve the search ability. After multiple iterations, the optimized neural network model is obtained for the control of the production process.

[0082] 3. Advantages

[0083] This model can effectively overcome the problem that traditional neural network training is prone to falling into local optimality, and use the global search capability of quantum particle swarm optimization algorithm to find better neural network parameters. In the production of polyacrylamide dry powder, it can more accurately predict and adjust control parameters based on input data, improve product quality stability and production efficiency, and have better adaptability to nonlinear changes and multivariable coupling in the production process.

[0084] (II) Production parameter optimization model integrating echo state network and genetic algorithm

[0085] 1. Model Principle

[0086] Echo state network (ESN) is a special recursive neural network that has the ability to quickly model dynamic systems. In the production of polyacrylamide dry powder, the production process is regarded as a dynamic system, and ESN can learn the dynamic relationship between production process data (such as temperature, pressure, quality indicators at different times, etc.). Genetic algorithm (GA) optimizes the parameters of ESN (such as the connection weight of the reserve pool, the weight of the output layer, etc.) by simulating the selection, crossover and mutation operations in the biological evolution process to improve the modeling accuracy and prediction ability of ESN for the production process.

[0087] 2. Algorithm Implementation

[0088] First, an echo state network is constructed, the connection weights of the reserve pool are randomly initialized, and the connection structure between the input layer, reserve pool, and output layer is determined. The historical data of the production process is used as the input of the ESN, and the ESN is trained to learn the dynamic characteristics of the production process and output the prediction of the future production status. Then, the parameters of the ESN are optimized using a genetic algorithm. The fitness function is defined, such as the error between the predicted product quality and the actual quality, production efficiency, and other indicators to evaluate the quality of individuals (a set of ESN parameters). Through the iterative operation of the genetic algorithm, excellent individuals are selected, crossed, and mutated, and the parameters of the ESN are updated to improve its performance. Finally, based on the prediction results of the optimized ESN model, a production parameter optimization strategy is generated.

[0089] 3. Advantages

[0090] This fusion model can make full use of the modeling ability of the echo state network for dynamic systems and the global optimization ability of the genetic algorithm. In the production of polyacrylamide dry powder, it can better handle the dynamic changes and uncertainties in the production process, such as raw material quality fluctuations, equipment performance changes, etc. By optimizing production parameters, it can effectively improve product quality and production efficiency and reduce production costs.

[0091] (III) Quality control and management model based on deep forest and fuzzy logic

[0092] 1. Model Principle

[0093] Deep forest is a new type of deep model based on decision trees, which has the advantages of fast training speed and strong adaptability to small data volume and unbalanced data. In the production of polyacrylamide dry powder, deep forest is used to learn and classify the features of production process data and product quality data, and to explore the impact patterns of different factors on product quality. Fuzzy logic is used to deal with ambiguity and uncertainty in the production process, such as fuzzy concepts such as "good quality" and "poor quality", as well as the fuzzy relationship between factors such as raw material quality and production conditions and product quality. By combining deep forest and fuzzy logic, accurate control and management of product quality can be achieved.

[0094] 2. Mathematical Model

[0095] Construct a deep forest model, take the collected production data and quality data as input, and perform feature extraction and classification through multi-layer random forest and completely random forest structures. For the fuzzy logic part, define fuzzy sets and membership functions, such as dividing product quality indicators (molecular weight, degree of hydrolysis, etc.) into different fuzzy sets (such as "high", "medium" and "low"), and determine the membership of each quality indicator in different fuzzy sets. According to the output of the deep forest and the reasoning rules of fuzzy logic, establish a quality control and management model. For example, if the deep forest determines that a certain production link may lead to a decline in product quality, the fuzzy logic determines the degree of quality decline and possible causes based on the current production data, and then formulates corresponding control measures.

[0096] This model combines the powerful feature learning ability of deep forest and the advantage of fuzzy logic in handling uncertainty. In the production of polyacrylamide dry powder, it can effectively control quality with limited data and accurately handle fuzzy information and complex relationships in the production process. It can timely discover product quality problems and take targeted measures to improve the stability and pass rate of product quality.

[0097] The present invention has been described in detail above in conjunction with the embodiments. However, those skilled in the art will appreciate that, without departing from the spirit of the present invention, the specific parameters in the above embodiments may be changed to form a plurality of specific embodiments, which are all within the common variation range of the present invention and will not be described in detail herein.

Claims

1. An intelligent production management system for polyacrylamide dry powder, characterized in that: include: The data acquisition and preprocessing module is used to deploy high-precision sensors at various key links in the polyacrylamide dry powder production line to comprehensively collect production data, and transmit the collected data to the data processing center via high-speed industrial Ethernet for preprocessing operations. The preprocessing operations include using an outlier detection algorithm based on deep learning to detect and repair outliers and standardize all sensor data; Intelligent production control module, which is used to control the production process based on preprocessed data using an algorithm that combines multivariable predictive control with reinforcement learning, and enhances the robustness of the control strategy through robust optimization methods; The quality monitoring and optimization module is used to monitor and analyze the product quality of polyacrylamide dry powder in real time, optimize and adjust the production process based on the quality prediction results and actual test data, and continuously optimize the quality prediction model and control strategy using quality data feedback.

2. The intelligent production management system for polyacrylamide dry powder according to claim 1, characterized in that: In the data acquisition and preprocessing module, the sensors installed in the raw material storage and transportation links include mass flow meters, liquid level sensors and moisture sensors, which are used to accurately measure the flow rate, tank liquid level and moisture content of the raw materials respectively.

3. The intelligent production management system for polyacrylamide dry powder according to claim 1, characterized in that: The outlier detection algorithm based on deep learning adopted by the data acquisition and preprocessing module is to construct a deep autoencoder network. The deep autoencoder network includes 3 hidden layers, which contain 64, 32 and 16 neurons respectively. ReLU is used as the activation function. The feature representation of the training data is trained by inputting normal production data into the network, and then the data to be detected is input into the trained network for decoding and reconstruction, and the reconstruction error is calculated. If the reconstruction error exceeds the set threshold, the data is judged as an outlier, and a repair method based on data distribution characteristics is used to repair the outlier.

4. The intelligent production management system for polyacrylamide dry powder according to claim 1, characterized in that: In the intelligent production control module, the multivariable predictive control (MPC) algorithm is used to model and control the core process of polyacrylamide dry powder production, considering the mutual coupling relationship between multiple variables such as reaction temperature, pressure, and material flow, establishing a prediction model with product quality indicators as the control target, and solving the optimal control sequence in the future through rolling optimization; the reinforcement learning algorithm is used to dynamically optimize and adjust the MPC control strategy, construct an intelligent agent based on a deep Q network (DQN), and use the real-time status information in the production process as the input of the intelligent agent. The intelligent agent learns the optimal control strategy adjustment rules through interaction with the production environment; the robust optimization method is used to introduce the interval representation of uncertain parameters in the MPC model, and obtain a control strategy that can still maintain good performance within a certain uncertainty range by solving the robust optimization problem.

5. The intelligent production management system for polyacrylamide dry powder according to claim 4, characterized in that: When the multivariable predictive control (MPC) algorithm is applied to the polymerization process, the reaction temperature, monomer concentration, and initiator dosage are selected as control variables, the product molecular weight is selected as the controlled variable, a state space model is established, and the upper and lower limits and change rate limits of the control variables are set.

6. The intelligent production management system for polyacrylamide dry powder according to claim 4, characterized in that: In the reinforcement learning algorithm, the real-time status information in the production process is digitized and characterized to construct a state space, including expressing the raw material quality fluctuation as the percentage of raw material purity deviation, and the equipment operation status change as the failure rate of key equipment. The intelligent agent selects actions through a deep Q network according to the current state. The actions include adjusting the control parameters in the MPC model and switching to spare equipment. According to the execution results of the actions, the environment gives the intelligent agent corresponding reward feedback. The reward function comprehensively considers factors such as the achievement of product quality indicators, the improvement of production efficiency, and the reduction of energy consumption.

7. The intelligent production management system for polyacrylamide dry powder according to claim 1, characterized in that: In the quality monitoring and optimization module, a quality prediction model based on support vector machine regression (SVR) is adopted, the polymerization reaction temperature, the amount of initiator, and the purity of raw materials are selected as input variables, and the characteristic parameters of the product molecular weight distribution are selected as output variables. The SVR model is trained by a large amount of historical data, the radial basis function (RBF) is selected as the kernel function, and the cross-validation method is used to optimize the kernel function parameters and penalty factors, so as to accurately predict the changing trend of the product molecular weight distribution according to the current polymerization reaction conditions and raw material information, and the prediction error is within ±10%.

8. The intelligent production management system for polyacrylamide dry powder according to claim 1, characterized in that: The quality monitoring and optimization module optimizes and adjusts the production process according to the quality prediction results and actual test data, and continuously optimizes the control strategy of the SVR model and the intelligent agent by using the quality data feedback.

9. A method for intelligent production management of polyacrylamide dry powder, characterized in that: The following steps are involved: In the data collection and preprocessing step, multi-source heterogeneous data are comprehensively collected from the raw material supply, polymerization reaction, drying, crushing and packaging of polyacrylamide dry powder production, and the collected data are preprocessed using an outlier detection algorithm based on deep learning; Intelligent production control steps, using multivariable predictive control (MPC) algorithm to model and control the core process of polyacrylamide dry powder production, introducing reinforcement learning algorithm to dynamically optimize the MPC control strategy, and using robust optimization method to enhance the robustness of the control strategy; In the quality monitoring and optimization steps, a real-time monitoring system for the quality of polyacrylamide dry powder products is established. A quality prediction model based on support vector machine regression (SVR) is used for quality prediction. The production process is optimized and adjusted according to the quality prediction results and actual test data. Quality data feedback is used to continuously optimize the control strategy of the SVR model and intelligent agent.

10. The intelligent production management method of polyacrylamide dry powder according to claim 9, characterized in that: In the data collection and preprocessing steps, a deep autoencoder network is constructed using an outlier detection algorithm based on deep learning. The network structure includes three hidden layers, which contain 64, 32, and 16 neurons respectively. ReLU is used as the activation function. The feature representation of the training data is trained by inputting normal production data into the network, and then the data to be detected is input into the trained network for decoding and reconstruction, and the reconstruction error is calculated. If the reconstruction error exceeds the set threshold, the data is judged to be an outlier. The outlier is repaired by a repair method based on data distribution characteristics, and all data are standardized.

Images