Waste plastic regeneration production risk coping method based on multivariate innovation algorithm

Abstract

The invention belongs to the technical field of information, and particularly relates to a waste plastic regeneration production risk coping method based on a multivariate innovation algorithm, which is applied to waste plastic regeneration enterprises. Firstly, Monte Carlo simulation is used for analyzing the influence of uncertain factors such as raw material quality, equipment faults and market requirements on production; then, in the embodiment, a self-adaptive neural fuzzy inference system, a grey wolf pack algorithm optimized support vector machine, a variational mode decomposition and gating circulation unit and other algorithms are adopted, risks are accurately evaluated, and corresponding strategies are formulated, such as raw material detection standard establishment, equipment spare part inventory increase, flexible production plan formulation and the like; and the product quality stability and the economic benefits of enterprises are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of production management of waste plastic recycling enterprises, and specifically relates to a method for coping with risks in waste plastic recycling production based on a multi - innovation algorithm. Background Art

[0002] Waste plastic recycling production, as an important way to solve plastic pollution and achieve resource recycling, occupies an important position in the environmental protection industry. However, this production process faces many uncertain factors. In terms of raw material quality, due to the wide and complex sources of waste plastics, the composition, impurity content, etc. of raw materials in different batches vary greatly, directly affecting product quality. In terms of equipment failures, production equipment operates for a long time and is affected by factors such as wear and aging, and the occurrence of failures is random, which may lead to production interruptions and increase costs. In terms of market demand, affected by the macro - economic environment, changes in consumer preferences, etc., the demand for recycled plastic products in the market fluctuates continuously in terms of quantity and variety, posing challenges to enterprise production plans and sales.

[0003] Traditional coping methods mainly rely on experience and simple statistical analysis, and it is difficult to comprehensively and accurately evaluate the risks brought by uncertain factors. Although Monte Carlo simulation can simulate uncertainties to a certain extent, as the complexity of the production process increases, its limitations gradually emerge. Therefore, developing new innovative algorithms to more accurately analyze uncertain factors and formulate effective risk coping strategies is of great significance for the development of the waste plastic recycling production industry. Summary of the Invention

[0004] The present invention provides a method for coping with risks in waste plastic recycling production based on a multi - innovation algorithm, including the following analysis and coping steps based on different algorithms:

[0005] When processing waste plastic raw material quality data and constructing decision tree nodes for splitting, a feature selection strategy combining mutual information and principal component analysis is adopted. That is, first calculate the correlation between each raw material quality feature and the product quality index through mutual information, screen out a subset of features with higher correlations, and then perform principal component analysis on this subset to reduce the feature dimension and retain key information, so as to construct a decision tree and improve the prediction accuracy of product quality;

[0006] Utilize the raw material quality - product quality relationship model trained by the improved random forest algorithm, combined with the Markov chain, divide the raw material quality state into different grades, statistically calculate the transition probabilities between different states based on historical data, and simulate and predict the future change path of the raw material quality state and the corresponding product quality pass rate;

[0007] Risk response strategy: If it is predicted that the product quality pass rate may decline, establish a strict raw material quality traceability system, record in detail the source of each batch of raw materials, the collection process, the pre-treatment situation, etc., and at the same time optimize the raw material pre-treatment process, such as developing an efficient impurity removal technology for raw materials with high impurities.

[0008] Furthermore, in the analysis and response of raw material quality risks, when calculating the feature correlation through mutual information, a specific mutual information calculation method is used for calculation to ensure that the features truly related to the product quality indicators are screened out.

[0009] Furthermore, it includes equipment failure risk analysis and response based on quantum genetic - Bayesian network: Improve the traditional quantum genetic algorithm. In the quantum bit encoding, according to the problem complexity and the size of the search space, adopt an adaptive dynamic encoding strategy to adjust the encoding length. Introduce chaotic mapping in the quantum rotation gate update strategy to increase population diversity and avoid algorithm premature convergence;

[0010] Bayesian network construction: Use the improved quantum genetic algorithm to learn the equipment failure historical data, construct an equipment failure Bayesian network to graphically represent the causal relationship between variables, and optimize its structure and parameters through the quantum genetic algorithm to clarify the influence probability of component failures on other components and the entire equipment system failure;

[0011] Risk response strategy: Based on the Bayesian network failure probability prediction results, increase the inventory of spare parts for key equipment components and reasonably allocate the quantity according to the failure probability. Establish an equipment failure early warning system to monitor the operating state parameters of the equipment in real time. When the state is close to failure, give an early warning in time and arrange preventive maintenance.

[0012] Furthermore, it also includes market demand risk analysis and response based on deep reinforcement learning - wavelet analysis: Construct a market demand prediction model based on deep reinforcement learning, use the long short - term memory network (LSTM) as the main structure, model the market demand prediction as a reinforcement learning task, the agent selects prediction actions according to the current market state, and the environment gives rewards or punishments according to the prediction results, so that the agent learns to make the optimal prediction decision;

[0013] Introduce wavelet analysis to pre - process the market demand time - series data, decompose it into a low - frequency approximation part and a high - frequency detail part, analyze and process them separately, and then reconstruct the data and input it into the deep reinforcement learning model to improve the prediction accuracy;

[0014] Risk response strategy: According to the prediction results, when it is predicted that the market demand will rise, increase the production scale in advance and reasonably arrange production resources. When the demand drops, adjust the production plan in time and reduce production. At the same time, strengthen market research and analysis and develop new products that adapt to market demand changes.

[0015] Furthermore, it includes:

[0016] Collect quality data such as the plastic composition ratio, water content, and degree of aging of waste plastic raw materials, as well as performance indicators such as the tensile strength, elongation at break, and hardness of the corresponding recycled products, and normalize them to the interval [0, 1];

[0017] ANFIS model construction: Determine the structure of the fuzzy inference system, select the Gaussian membership function to describe the fuzzy set of raw material quality characteristics, use the neural network backpropagation algorithm and the least squares method to train the parameters of the fuzzy inference system, and adjust the membership function parameters and fuzzy rule weights to reflect the relationship between raw material quality and product performance;

[0018] Risk assessment and response strategies: Input new raw material quality data into the trained ANFIS model to predict product performance, set a risk threshold for product performance. If the predicted product performance is close to or lower than the threshold, strengthen the inspection of raw materials, strictly conduct quality inspections on each batch of raw materials, establish an evaluation system for raw material suppliers, and rectify or replace suppliers with unstable raw material quality.

[0019] Furthermore, the fuzzy sets are divided according to the actual range and characteristics of raw material quality characteristics. For example, the water content of raw materials is divided into three fuzzy sets: "low", "medium", and "high", and the Gaussian membership function parameters of each fuzzy set are determined through optimization.

[0020] Furthermore, it includes:

[0021] Install sensors on the waste plastic recycling production equipment to collect vibration, temperature, and current data in real time, and perform feature extraction on the original data by calculating the root mean square value, peak factor of the vibration signal, and temperature change rate;

[0022] Grey wolf optimization algorithm for SVM: Use the classification accuracy of SVM as the fitness function of the grey wolf optimization algorithm, and use the grey wolf optimization algorithm to search for the optimal SVM penalty factor C and kernel function parameter γ to construct a better-performing SVM fault prediction model;

[0023] Fault prediction and risk response: Input the extracted equipment characteristic data into the optimized SVM model to predict the operating state of the equipment in real time. When a high fault risk is predicted, start the emergency plan, arrange professional maintenance personnel to conduct a comprehensive inspection and maintenance of the equipment, allocate maintenance resources according to the severity of the fault and the difficulty of maintenance, and establish an equipment fault knowledge base to record the cause, treatment method, and preventive measures of each fault.

[0024] 8. The method for predicting and responding to faults of waste plastic recycling equipment based on the support vector machine (SVM) optimized by the grey wolf optimization algorithm according to claim 7, wherein in the optimization process of the grey wolf optimization algorithm, the wolf pack search strategy and parameter settings are adjusted through [specific adjustment method] to improve the convergence speed and global search ability of the algorithm, and ensure that the optimal SVM parameter combination is found.

[0025] Furthermore, it includes:

[0026] Collect market demand data on the monthly sales volume, market price, changes in relevant policies, and product information of competitors of waste plastic recycled products in the past five years, clean and organize it, and arrange it in a time series after removing outliers and missing values;

[0027] Use the VMD algorithm to decompose the market demand time series into multiple Intrinsic Mode Function (IMF) components with different central frequencies, and each IMF component represents the change characteristics of the market demand at different time scales;

[0028] Build a GRU model for each IMF component respectively, use the historical data of each IMF component as input to train the model, learn the change law of this component, and superimpose the prediction results of each GRU model to obtain the final predicted value of the market demand;

[0029] Risk response strategy: Based on the prediction results of the VMD-GRU model, formulate flexible production and inventory management strategies. When the market demand rises, increase production input in advance, arrange the progress reasonably, optimize the inventory management system at the same time, adjust the safety inventory level according to the change of the predicted demand, strengthen market research and customer relationship management, and adjust the production plan in a timely manner.

[0030] On the other hand, the present invention also provides an uncertainty analysis and risk response system for the waste plastic recycling production process, including:

[0031] A raw material quality risk analysis and response module based on an improved random forest-Markov chain, which is used for the raw material quality risk analysis and response steps;

[0032] A device failure risk analysis and response module based on quantum genetic-Bayesian network, which is used for the device failure risk analysis and response steps;

[0033] A market demand risk analysis and response module based on deep reinforcement learning-wavelet analysis, which is used for the market demand risk analysis and response steps;

[0034] A raw material quality and product performance correlation analysis and risk response module based on an adaptive neuro-fuzzy inference system, which is used to implement the raw material quality and product performance correlation analysis and risk response method;

[0035] A device failure prediction and risk response module based on a support vector machine optimized by a grey wolf pack algorithm, which is used to implement the waste plastic recycling device failure prediction and risk response method;

[0036] A market demand prediction and risk response module based on variational mode decomposition and gated recurrent unit, which is used to implement the market demand prediction and risk response method.

[0037] Beneficial effects

[0038] Analyze the correlation between raw material quality and product performance. The enterprise has successfully reduced the defective rate and improved the product quality stability. This not only reduces resource waste and production costs, but also enhances the market competitiveness of the products. Significantly improve the accuracy of equipment failure prediction and ensure production continuity. Facing the problem of production interruption caused by sudden equipment failures in large-scale waste plastic recycling enterprises, install sensors to collect equipment operation data and extract key features, use the grey wolf pack algorithm to optimize the SVM parameters, and construct a high-performance failure prediction model. Conduct market demand forecasting and response for enterprises focusing on specific products, effectively enhancing the enterprise's response ability to market demand changes. Due to market demand fluctuations, this enterprise faces inventory and production planning problems. Based on the forecasting results, formulate flexible production and inventory management strategies, and strengthen market research and customer relationship management. Enable the enterprise to better adapt to market changes, reasonably arrange production and inventory, enhance market competitiveness, and achieve sustainable development. Description of the Drawings

[0039] Figure 1 Flowchart of the improved random forest-Markov chain model

[0040] Figure 2 Diagram of the construction of the quantum genetic-Bayesian network model

[0041] Figure 3 Diagram of the market demand forecasting model based on deep reinforcement learning-wavelet analysis Detailed Implementation Manner

[0042] Example 1

[0043] (1) Risk analysis and response of raw material quality based on the improved random forest-Markov chain

[0044] Improved random forest algorithm: When dealing with high-dimensional and non-linear raw material quality data, the traditional random forest algorithm has problems such as inaccurate feature selection and limited model generalization ability. The improved random forest algorithm proposed in the present invention introduces a new feature selection metric method when constructing the decision tree node split. Traditional methods are mostly based on information gain or Gini coefficient, while this algorithm considers the complex non-linear relationship between features and product quality indicators, and adopts a feature selection strategy based on the combination of mutual information and principal component analysis. Specifically, first calculate the correlation between each feature and the product quality indicator through mutual information, and screen out the feature subset with higher correlation. Then, perform principal component analysis on this subset to further extract the principal components, reducing the feature dimension while retaining key information. In this way, when constructing the decision tree, it can more accurately select the features that have a significant impact on product quality and improve the model prediction accuracy.

[0045] Markov chain prediction: Use the improved random forest algorithm to train the historical raw material quality data and the corresponding product quality data to obtain a raw material quality - product quality relationship model. On this basis, combine the Markov chain to predict the future change trend of raw material quality and the corresponding product quality qualification rate. The Markov chain assumes that the future state of the system only depends on the current state and does not depend on the past state. Divide the raw material quality state into different grades (such as excellent, good, medium, poor), and statistically calculate the transition probabilities between different states according to historical data. For example, statistically calculate the probability of transitioning from the "good" state to the "medium" state. Through Markov chain simulation, predict the change path of the future raw material quality state, and then combine the improved random forest model to predict the product quality qualification rate.

[0046] Risk response strategy: According to the simulation results, if it is predicted that the product quality qualification rate may decline, formulate a risk response strategy for raw material quality. On the one hand, establish a strict raw material quality traceability system to control the raw material quality from the source. Keep detailed records of the source, collection process, pre - treatment situation, etc. of each batch of raw materials, so as to quickly locate the cause in case of quality problems. On the other hand, optimize the raw material pre - treatment process. For example, for raw materials with high impurity content, develop more efficient impurity removal technologies to improve the purity of raw materials and ensure the stability of product quality.

[0047] (2) Equipment failure risk analysis and response based on quantum genetic - Bayesian network

[0048] Optimization of quantum genetic algorithm: When the traditional genetic algorithm searches the equipment failure space, it is easy to fall into local optimal solutions. The quantum genetic algorithm, based on the superposition state and entanglement characteristics of quantum bits, has stronger global search ability. The present invention improves the quantum genetic algorithm. In terms of quantum bit encoding, an adaptive dynamic encoding strategy is adopted. According to the complexity of the problem and the size of the search space, dynamically adjust the encoding length of quantum bits to improve the encoding efficiency. In the quantum rotation gate update strategy, introduce chaotic mapping to increase the diversity of the population. Chaotic mapping has characteristics such as randomness and ergodicity, which can enable the quantum rotation gate to more effectively explore the search space when updating quantum bits and avoid algorithm prematurity.

[0049] Construction of Bayesian network: Use the improved quantum genetic algorithm to learn the historical equipment failure data and construct an equipment failure Bayesian network. The Bayesian network graphically represents the causal relationship between variables. Nodes represent each component or state of the equipment, and edges represent the fault propagation relationship between components. Optimize the structure and parameters of the Bayesian network through the quantum genetic algorithm to make it more accurately reflect the internal mechanism of equipment failure. For example, determine the influence probability of a certain component failure on other components and the failure of the entire equipment system.

[0050] Risk response strategy: Based on the failure probability prediction results of the Bayesian network, formulate a risk response strategy for equipment failures. Increase the inventory of spare parts for key equipment components and reasonably allocate the inventory quantity according to the magnitude of the failure probability. At the same time, establish an equipment failure warning system to monitor the operating state parameters of the equipment in real time (such as temperature, vibration, current, etc.). When the monitoring data shows that the equipment state is approaching the failure state, issue a warning in a timely manner, arrange maintenance personnel for preventive maintenance, reduce the probability of equipment failure, and reduce the losses caused by production interruptions.

[0051] (3) Market demand risk analysis and response based on deep reinforcement learning - wavelet analysis

[0052] Deep reinforcement learning model: Market demand has high dynamics and uncertainty, and traditional prediction methods are difficult to accurately capture its changing patterns. The present invention constructs a market demand prediction model based on deep reinforcement learning. A deep neural network (such as a long short-term memory network LSTM) is used as the main structure. LSTM can effectively handle the long-term dependence problem in time series data and capture the trend of market demand changing over time. The market demand prediction problem is modeled as a reinforcement learning task. The agent interacts with the market environment and selects a prediction action (predicting the market demand in a future period) according to the current market state (including historical demand data, macroeconomic indicators, industry policies, etc.). The environment gives the agent rewards or punishments according to the prediction results. Through continuous learning, the agent gradually learns to make optimal prediction decisions.

[0053] Wavelet analysis assistance: To further improve the prediction accuracy, wavelet analysis is introduced to preprocess the market demand time series data. Wavelet analysis can decompose the time series data into components of different frequencies, highlighting the local features and changing trends in the data. Through wavelet transform, the market demand data is decomposed into a low-frequency approximation part and a high-frequency detail part. The low-frequency part reflects the long-term trend of market demand, and the high-frequency part reflects the short-term fluctuations. Analyze and process the components of different frequencies separately, and then reconstruct the data and input it into the deep reinforcement learning model. This can enable the model to better capture the complex changes in market demand and improve the prediction accuracy.

[0054] Risk response strategy: According to the prediction results of the deep reinforcement learning - wavelet analysis model, formulate a risk response strategy for market demand. Formulate a flexible production plan. When it is predicted that the market demand will increase, increase the production scale in advance, reasonably arrange production resources, and ensure that the market demand is met on time. When it is predicted that the market demand will decrease, adjust the production plan in a timely manner, reduce the output, and avoid inventory backlogs. At the same time, strengthen market research and analysis, combine the prediction results, develop new products that adapt to the changes in market demand, expand the market share, and reduce the risks brought by market demand fluctuations.

[0055] (1) Raw material quality risk analysis and response implementation based on improved random forest - Markov chain

[0056] Data collection and preprocessing: Collect detailed quality data of the enterprise's waste plastic raw materials in the past three years, including information such as plastic types, impurity content, color, density, etc., as well as corresponding product quality data, such as product strength, toughness, purity, etc. Clean the data to remove outliers and missing values. For missing values, use a filling method based on the K - nearest neighbor algorithm to fill them. Then standardize the data to map data with different features to the same numerical range for subsequent analysis.

[0057] Implementation of the improved random forest algorithm: According to the steps of the improved random forest algorithm, first calculate the mutual information between each raw material quality feature and the product quality index, and select the feature subset with mutual information greater than a certain threshold (such as 0.3). Conduct principal component analysis on this subset to determine the number of principal components so that the cumulative contribution rate reaches more than 90%. Use the selected features to construct a random forest model, set the number of decision trees to 100, and the maximum depth to 10. Adjust the model parameters through the cross - validation method to improve the model prediction accuracy.

[0058] Implementation of Markov chain prediction: According to the statistical characteristics of the raw material quality data, divide the raw material quality state into four levels. Count the number of transitions between different states and calculate the transition probability matrix. Use the Markov chain to conduct 1000 simulations to predict the change path of the raw material quality state in the next 12 months. Combine with the improved random forest model to predict the product quality qualification rate under each state.

[0059] Implementation of risk response strategies: Establish a raw material quality traceability system, assign a unique identification code to each batch of raw materials, and record the whole - process information of the raw materials from collection to entry into the production link. Develop an efficient impurity removal process for raw materials with high impurities, such as using a new method combining physical adsorption and chemical separation, install corresponding impurity removal equipment on the production line, and pre - process the raw materials.

[0060] (2) Equipment failure risk analysis and response implementation based on quantum genetic - Bayesian network

[0061] Data collection and collation: Collect the failure data of the enterprise's main production equipment in the past five years, including information such as failure occurrence time, failure type, failure components, maintenance records, etc. At the same time, collect real - time monitoring data during the operation of the equipment, such as temperature, pressure, rotation speed, etc. Collate the data, classify and code the equipment failure types, and arrange the monitoring data in time series.

[0062] Implementation of Quantum Genetic Algorithm Optimization: An adaptive dynamic coding strategy is adopted. According to the dimension and complexity of equipment failure data, the initial encoding length of quantum bits is set to 8. During the algorithm iteration process, the encoding length is dynamically adjusted according to the change of the search space. A chaotic mapping (such as Logistic mapping) is introduced to update the quantum rotation gate, and the chaotic mapping parameters are set to keep the population diverse during the search process. The iteration number of the quantum genetic algorithm is set to 200, and the population size is set to 50.

[0063] Implementation of Bayesian Network Construction: Use the improved quantum genetic algorithm to learn the equipment failure data to determine the structure and parameters of the Bayesian network. The K2 algorithm is used for Bayesian network structure learning, with the minimum description length (MDL) as the scoring function. In the parameter learning stage, the maximum likelihood estimation method is used to calculate the conditional probability table of the nodes. After construction, the Bayesian network is verified to ensure that it can accurately reflect the causal relationship of equipment failures.

[0064] Implementation of Risk Response Strategy: Classify key equipment components according to the equipment failure probability predicted by the Bayesian network. For components with a high failure probability, increase the inventory of spare parts to ensure that the inventory quantity can meet the demand for the continuous operation of the equipment for three months. Establish an equipment failure early warning system, input the real-time monitoring data into the Bayesian network model, and when the predicted failure probability exceeds the set threshold (such as 0.7), notify the maintenance personnel in time by text message, email, etc. for preventive maintenance.

[0065] (III) Implementation of Market Demand Risk Analysis and Response Based on Deep Reinforcement Learning - Wavelet Analysis

[0066] Data Collection and Processing: Collect the market demand data for recycled plastic products in the past decade, including information such as the demand quantity, price, and sales area of different types of products. At the same time, collect macroeconomic indicator data, such as GDP growth rate, inflation rate, and industry policy information. Organize the market demand data in a time series, and quantify the macroeconomic indicator and industry policy information. Use wavelet analysis to decompose the market demand time series, select an appropriate wavelet basis function (such as Daubechies wavelet), decompose the data into low-frequency and high-frequency components. Normalize different frequency components and then reconstruct the data.

[0067] Implementation of Deep Reinforcement Learning Model: Construct a deep reinforcement learning model based on LSTM, set the number of layers of the LSTM network to 3, and the number of neurons in each layer to 64. Define the state space of the agent as a vector composed of the reconstructed market demand data, macroeconomic indicators, and industry policy information; the action space as different values for predicting the market demand in the next month; and the reward function as the reciprocal of the error between the predicted value and the actual value, where the smaller the error, the greater the reward. Set the learning rate of reinforcement learning to 0.001, the discount factor to 0.95, and perform 10,000 training iterations.

[0068] Implementation of Risk Response Strategies: Based on the prediction results of the deep reinforcement learning-wavelet analysis model, formulate production plan adjustment strategies. When the predicted market demand growth in the next three months exceeds 20%, increase the operating time of production equipment two months in advance and recruit temporary workers to ensure that the output can meet the market demand. When the predicted market demand decline exceeds 15%, reduce production shifts in a timely manner and adjust the raw material procurement plan to avoid inventory backlogs. At the same time, establish a market research team, conduct a market research every quarter in combination with the prediction results, analyze the changing trend of market demand, and develop new products. For example, in response to the market trend of enhanced environmental awareness, develop degradable recycled plastic products.

[0069] Example 4:

[0070] Correlation Analysis and Risk Response of Raw Material Quality and Product Performance Based on Adaptive Neuro-Fuzzy Inference System (ANFIS):

[0071] During the production process, there is a problem that the unstable quality of raw materials has a significant impact on product performance. Due to the lack of accurate analysis of the complex non-linear relationship between raw material quality and product performance, it is difficult to predict product quality risks in advance, resulting in a relatively high defective rate.

[0072] Data Collection and Preprocessing: Collect various quality data of waste plastic raw materials used by the enterprise in the past two years, such as the proportion of plastic components, water content, degree of aging, etc., and at the same time record the corresponding performance indicators of recycled products, such as tensile strength, elongation at break, hardness, etc. Normalize the data and map all data to the [0,1] interval to eliminate the influence of different feature dimensions.

[0073] ANFIS Model Construction: The Adaptive Neuro-Fuzzy Inference System combines the learning ability of neural networks and the linguistic expression ability of fuzzy logic. First, determine the structure of the fuzzy inference system and select appropriate membership functions (such as Gaussian membership functions) to describe the fuzzy sets of raw material quality characteristics. For example, for the water content of raw materials, it can be divided into three fuzzy sets: "low", "medium", and "high", and each set corresponds to a Gaussian membership function. Then, use the backpropagation algorithm and least squares method of neural networks to train the parameters of the fuzzy inference system, adjust the parameters of the membership functions and the weights of the fuzzy rules, so that the ANFIS model can accurately reflect the relationship between raw material quality and product performance.

[0074] Risk Assessment and Response Strategies: Input new raw material quality data through the trained ANFIS model to predict product performance. According to the prediction results, set the risk threshold of product performance. If the predicted product performance is close to or lower than the risk threshold, it is determined that there is a quality risk. In response to this situation, on the one hand, strengthen the raw material inspection link and conduct more strict quality inspections on each batch of raw materials; on the other hand, establish a raw material supplier evaluation system to rectify or replace suppliers with unstable raw material quality.

[0075] Implementation Effect: After implementation, the defective product rate of the product decreased by 15%, and by accurately analyzing the correlation between raw material quality and product performance, the quality stability of the product was effectively improved.

[0076] Example Five:

[0077] Equipment Fault Prediction and Risk Response Based on Support Vector Machine (SVM) Optimized by Grey Wolf Pack Algorithm:

[0078] Enterprise Background and Problem Description: Large-scale waste plastic recycling enterprises have many complex production equipment. The suddenness of equipment failures leads to production interruptions and causes significant economic losses. The accuracy of traditional equipment fault prediction methods is insufficient and cannot meet the enterprise's need for stable operation of equipment.

[0079] Data Collection and Feature Extraction: Install various sensors on the enterprise's production equipment to collect data such as vibration, temperature, and current during the operation of the equipment in real time. Extract features from the collected raw data, such as calculating the root mean square value and peak factor of the vibration signal, and the change rate of temperature, etc., to obtain key features that can characterize the operation state of the equipment.

[0080] Grey Wolf Optimizer Algorithm Optimized SVM: Support Vector Machine (SVM) is a commonly used classification and regression model, but the parameter selection of traditional SVM has a great impact on the model performance. Grey Wolf Optimizer Algorithm is a new intelligent optimization algorithm with the characteristics of fast convergence speed and strong global search ability. The penalty factor C and kernel function parameter γ of SVM are optimized by Grey Wolf Optimizer Algorithm. During the optimization process, the classification accuracy of SVM is used as the fitness function of Grey Wolf Optimizer Algorithm, and the optimal values of C and γ are found through the search behavior of the wolf pack. After multiple iterations, the optimal parameter combination is determined, so as to construct a SVM fault prediction model with better performance.

[0081] Fault Prediction and Risk Response: The extracted device feature data is input into the optimized SVM model to predict the running state of the device in real time and judge whether the device is about to fail. When a high fault risk of the device is predicted, the emergency plan is immediately activated. First, arrange professional maintenance personnel to conduct a comprehensive inspection and maintenance of the device; second, reasonably allocate maintenance resources according to the urgency and maintenance difficulty of the device fault to ensure that the device resumes normal operation as soon as possible. At the same time, establish a device fault knowledge base, and record in detail the causes, treatment methods and preventive measures of each fault, providing a reference for subsequent fault prediction and treatment.

[0082] Implementation Effect: The accuracy rate of device fault prediction is increased to more than 90%, and the number of production interruptions is reduced by 30%, effectively reducing the impact of device faults on production and ensuring the continuity of enterprise production.

[0083] Example 6: Market Demand Prediction and Risk Response Based on Variational Mode Decomposition and Gated Recurrent Unit (VMD-GRU)

[0084] Enterprise Background and Problem Description: An enterprise specializing in the production of specific types of recycled plastic products faces the risk of inventory backlog or out-of-stock due to the fluctuations in market demand. Existing market demand prediction methods cannot accurately capture the complex patterns of demand changes, resulting in a mismatch between the enterprise's production plan and market demand.

[0085] Specific Implementation

[0086] Data Collection and Processing: Collect the market demand data of the enterprise's products in the past five years, including monthly sales volume, market price, relevant policy changes, and competitor product information, etc. Clean and organize the data, remove outliers and missing values, and arrange them in time series.

[0087] Variational Mode Decomposition (VMD): Since the market demand time series data contains components with different frequencies, variational mode decomposition can decompose it into multiple Intrinsic Mode Functions (IMFs) with different central frequencies. Through the VMD algorithm, the market demand time series is decomposed into multiple IMF components, and each IMF component represents the change characteristics of market demand at different time scales. For example, the low-frequency IMF component may reflect the long-term trend of market demand, while the high-frequency IMF component reflects short-term fluctuations.

[0088] Gated Recurrent Unit (GRU) Model Construction: For each decomposed IMF component, a Gated Recurrent Unit (GRU) model is constructed for prediction. GRU is a special type of recurrent neural network that can effectively handle the long-term dependence problem in time series data. Using the historical data of each IMF component as input, the GRU model is trained to learn the change pattern of that component. Then, the prediction results of each GRU model are superimposed to obtain the final predicted value of market demand.

[0089] Risk Response Strategy: Based on the prediction results of the VMD-GRU model, flexible production and inventory management strategies are formulated. When it is predicted that the market demand will increase, production inputs are increased in advance, and the production schedule is reasonably arranged to ensure that products can be supplied to the market in a timely manner. At the same time, the inventory management system is optimized, and the safety inventory level is adjusted according to the predicted demand changes to avoid inventory backlogs or out-of-stock situations. In addition, market research and customer relationship management are strengthened to timely understand market dynamics and customer demand changes, so as to more accurately adjust the production plan.

[0090] Implementation Effect: The market demand prediction error is reduced to within 10%, the inventory backlog rate is reduced by 25%, and the out-of-stock rate is reduced by 20%, effectively improving the enterprise's response ability to market demand changes and enhancing the enterprise's market competitiveness.

Claims

1. A method for dealing with waste plastic recycling production risks based on a multivariate innovation algorithm, characterized in that: It includes the following analysis and response steps based on different algorithms: When processing the quality data of waste plastic raw materials, when constructing the decision tree node splitting, a feature selection strategy based on mutual information and principal component analysis is adopted, that is, the correlation between each raw material quality feature and the product quality index is first calculated through mutual information, and the feature subset with higher correlation is screened out. Then, the principal component analysis is performed on the subset to reduce the feature dimension and retain the key information, so as to construct a decision tree and improve the prediction accuracy of product quality. The raw material quality-product quality relationship model obtained by training with the improved random forest algorithm is combined with the Markov chain to divide the raw material quality status into different levels. The transition probability between different states is statistically analyzed based on historical data, and the future raw material quality status change path and the corresponding product quality pass rate are predicted through simulation; Risk response strategy: If it is predicted that the product quality qualification rate may decline, establish a strict raw material quality traceability system, record in detail the source, collection process, pretreatment of each batch of raw materials, etc., and optimize the raw material pretreatment process, such as developing efficient impurity removal technology for high-impurity raw materials.

2. The risk management method according to claim 1, characterized in that: In the raw material quality risk analysis and response, when calculating the feature correlation through mutual information, a specific mutual information calculation method is used to ensure that the features that are truly relevant to the product quality indicators are screened out.

3. The risk management method according to claim 1, characterized in that: The change includes equipment failure risk analysis and response based on quantum genetic-Bayesian network: improving the traditional quantum genetic algorithm, in quantum bit coding, using adaptive dynamic coding strategy to adjust the coding length according to the complexity of the problem and the size of the search space, introducing chaotic mapping in the quantum revolving door update strategy, increasing population diversity and avoiding premature algorithm maturation; Bayesian network construction: Use the improved quantum genetic algorithm to learn the historical data of equipment failures, build the equipment failure Bayesian network, graphically represent the causal relationship between variables, optimize its structure and parameters through the quantum genetic algorithm, and clarify the probability of component failure affecting other components and the entire equipment system failure; Risk response strategy: Based on the failure probability prediction results of the Bayesian network, increase the inventory of spare parts for key equipment components and reasonably allocate the quantity according to the failure probability, establish an equipment failure early warning system, monitor the equipment operating status parameters in real time, and issue timely warnings and arrange preventive maintenance when the status is close to failure.

4. The uncertainty analysis and risk response method for waste plastic recycling production process based on multivariate innovation algorithm according to claim 1 is characterized in that: It also includes market demand risk analysis and response based on deep reinforcement learning-wavelet analysis: building a market demand forecasting model based on deep reinforcement learning, using long short-term memory network (LSTM) as the main structure, modeling market demand forecasting as a reinforcement learning task, the agent selects the forecasting action according to the current market status, and the environment gives rewards or penalties according to the forecast results, so that the agent learns to make the best forecasting decision; Wavelet analysis is introduced to pre-process the market demand time series data, decomposing it into a low-frequency approximate part and a high-frequency detail part. After analysis and processing, the reconstructed data is input into the deep reinforcement learning model to improve the prediction accuracy. Risk response strategy: Based on the forecast results, when the market demand is expected to increase, increase the production scale in advance and arrange production resources reasonably; when the demand decreases, adjust the production plan in time and reduce output. At the same time, strengthen market research and analysis and develop new products that adapt to changes in market demand.

5. A method for correlation analysis and risk response between quality of recycled waste plastic raw materials and product performance based on adaptive neural fuzzy inference system (ANFIS), characterized in that: include: Collect quality data such as the plastic component ratio, water content, aging degree, etc. of waste plastic raw materials and corresponding performance indicators such as tensile strength, elongation at break, hardness, etc. of recycled products, and normalize them to the [0,1] interval; ANFIS model construction: determine the fuzzy inference system structure, select Gaussian membership function to describe the fuzzy set of raw material quality characteristics, use neural network back propagation algorithm and least squares method to train fuzzy inference system parameters, adjust membership function parameters and fuzzy rule weights to reflect the relationship between raw material quality and product performance; Risk assessment and response strategies: Use the trained ANFIS model to input new raw material quality data to predict product performance and set product performance risk thresholds. If the predicted product performance is close to or below the threshold, strengthen raw material inspection, conduct strict quality inspections on each batch of raw materials, establish a raw material supplier evaluation system, and rectify or replace suppliers with unstable raw material quality.

6. The method for correlation analysis and risk management of waste plastic recycled raw material quality and product performance based on adaptive neural fuzzy inference system (ANFIS) according to claim 5, characterized in that: The fuzzy sets are divided according to the actual range and characteristics of the raw material quality characteristics, such as dividing the raw material moisture content into three fuzzy sets of "low", "medium" and "high", and the Gaussian membership function parameters of each fuzzy set are determined by optimization.

7. A method for failure prediction and risk response of waste plastic recycling equipment based on support vector machine (SVM) optimized by grey wolf pack algorithm, characterized in that: include: Install sensors on waste plastic recycling production equipment to collect vibration, temperature, and current data in real time, and calculate the root mean square value, peak factor, and temperature change rate of the vibration signal from the original data for feature extraction; Grey wolf pack algorithm optimizes SVM: Taking the SVM classification accuracy as the fitness function of the grey wolf pack algorithm, the grey wolf pack algorithm is used to search for the optimal SVM penalty factor C and kernel function parameter γ to build an SVM fault prediction model with better performance; Fault prediction and risk response: The extracted equipment feature data is input into the optimized SVM model to predict the equipment operating status in real time. When a high fault risk is predicted, the emergency plan is activated and professional maintenance personnel are arranged to comprehensively inspect and maintain the equipment. Maintenance resources are deployed according to the urgency of the fault and the difficulty of maintenance. An equipment fault knowledge base is established to record the cause, treatment method and preventive measures of each fault.

8. The method for failure prediction and risk response of waste plastic recycling equipment based on support vector machine (SVM) optimized by grey wolf pack algorithm according to claim 7 is characterized in that: During the optimization process of the grey wolf pack algorithm, the wolf pack search strategy and parameter settings are adjusted by [specific adjustment method] to improve the algorithm convergence speed and global search capability, ensuring that the optimal SVM parameter combination is found.

9. A method for predicting market demand and coping with risks of waste plastic recycling based on variational mode decomposition and gated cyclic units, characterized in that: include: Collect monthly sales volume, market prices, relevant policy changes, and competitor product information market demand data of waste plastic recycled products over the past five years, clean and organize them, remove outliers and missing values, and arrange them in time series; The VMD algorithm is used to decompose the market demand time series into multiple intrinsic mode function (IMF) components with different central frequencies. Each IMF component represents the change characteristics of market demand at different time scales. A GRU model is constructed for each IMF component, and the historical data of each IMF component is used as input to train the model, learn the change law of the component, and superimpose the prediction results of each GRU model to obtain the final prediction value of market demand; Risk response strategy: Based on the prediction results of the VMD-GRU model, formulate flexible production and inventory management strategies, increase production input in advance and arrange the progress reasonably when market demand increases. At the same time, optimize the inventory management system, adjust the safety stock level according to the predicted demand changes, strengthen market research and customer relationship management, and adjust the production plan in time.

10. An uncertainty analysis and risk response system for waste plastic recycling production process, characterized in that: include: The raw material quality risk analysis and response module based on the improved random forest-Markov chain is used for the raw material quality risk analysis and response steps; Equipment failure risk analysis and response module based on quantum genetic-Bayesian network, used for equipment failure risk analysis and response steps; Market demand risk analysis and response module based on deep reinforcement learning-wavelet analysis, used for market demand risk analysis and response steps; The raw material quality and product performance correlation analysis and risk response module based on the adaptive neuro-fuzzy inference system is used to implement the raw material quality and product performance correlation analysis and risk response method; Equipment failure prediction and risk response module based on support vector machine optimized by grey wolf pack algorithm, used to implement waste plastic recycling equipment failure prediction and risk response method; The market demand forecasting and risk response module based on variational mode decomposition and gated recurrent unit is used to implement market demand forecasting and risk response methods.

Images