A prediction method for the consumption in the production and manufacturing of Lyocell fibers

Abstract

The present invention relates to the technical fields of intelligent manufacturing and cost management, and particularly relates to a method for predicting the production manufacturing cost of Lyocell fiber and its application; collecting data in the production manufacturing process of Lyocell fiber; performing data preprocessing on the data; constructing a total prediction model; extracting three groups of process parameter sets from the data after data preprocessing and inputting them into the total prediction model to output the predicted values of raw material consumption, energy consumption, and equipment maintenance; setting the initial weight distribution of the total prediction model, calculating the mean square error of the three models in each total model, introducing Gaussian process regression, and obtaining the weights of the three models within each total model; calculating the predicted values of equipment maintenance, raw material consumption, and energy consumption after weighted fusion; obtaining the predicted total production manufacturing cost; the present invention assists enterprises in realizing refined cost management in the production manufacturing process of Lyocell fiber.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent manufacturing and cost management, and particularly relates to a method for predicting the production consumption of Lyocell fiber. Background Art

[0002] In the production of Lyocell fiber, raw material cost, energy consumption cost, and equipment maintenance cost are key factors affecting the profit rate and market competitiveness of enterprises. The sum of raw material cost, energy consumption cost, and equipment maintenance cost is the total production cost of enterprises in the production of Lyocell fiber. Among them, the fluctuations in raw material cost, energy consumption cost, and equipment maintenance cost caused by the consumption of raw materials involved in raw material cost lead to excessive fluctuations in process consumption cost. If the total production cost cannot be accurately predicted, and due to the excessive fluctuations in process consumption cost, the environmental protection cost in process consumption cost, such as recycling cost, cannot be confirmed, making it impossible for enterprises to conduct refined management of the costs involved in the entire process of production and manufacturing.

[0003] In the prior art, the methods for predicting the production cost of Lyocell fiber are mostly based on traditional empirical formulas or static analysis of a single cost module, and have the following limitations:

[0004] 1. Insufficient accuracy in predicting raw material cost: The main raw materials for the production of Lyocell fiber are cellulose materials such as dissolving pulp, and their prices are significantly affected by the supply and demand relationship in the international market and policy regulation. Existing methods usually use the historical price average for prediction, without fully considering the dynamic fluctuations of the raw material supply chain and the impact of process optimization on raw material utilization rate, resulting in large prediction deviations.

[0005] 2. Imperfect modeling of energy consumption cost: The production of Lyocell fiber requires solvent spinning under high temperature and high pressure conditions, and the energy consumption accounts for a relatively high proportion. Existing energy consumption predictions mostly rely on the rated power of equipment or fixed unit energy consumption coefficients, lacking prediction and analysis in combination with specific process parameters, and it is difficult to accurately reflect the energy consumption fluctuations in actual production.

[0006] 3. The equipment maintenance cost has not been systematically quantified: The maintenance cost of the equipment on the Lyocell fiber production line is closely related to the degree of equipment aging, failure rate, and preventive maintenance strategy; traditional methods usually adopt regular maintenance plans or simple statistical models based on failure records, and fail to predict in combination with real-time operation data, resulting in lag or redundancy in the estimation of maintenance cost.

[0007] Therefore, enterprises are in urgent need of a method that can predict the production cost of Lyocell fiber, provide a scientific basis for process optimization and resource scheduling, and assist enterprises in conducting refined cost management during the production and manufacturing process of Lyocell fiber.

[0008] For example, a Chinese patent with the publication number CN117788037A, publication date of January 2, 2024, and invention title of "A Multi-Value Chain Cost Prediction Method for Electric Power Cost" has the following specific technical solution: This invention utilizes network data mining technology to collect various cost data of electric power equipment manufacturing enterprises in multi-value chain activities, and constructs an influencing factor library under the coordination of multi-value chains of electric power equipment manufacturing enterprises; uses the Pearson correlation coefficient and grey correlation method to analyze and screen the influencing factors in the influencing factor library, selects the indicators with a relatively high correlation with cost prediction as the main influencing factors for prediction model analysis, and optimizes the BP neural network model using a combined algorithm of firefly perturbation and sparrow search algorithm to construct an FA-SSA-BP cost intelligent prediction model that can combine multi-value chain data such as product production, sales, supply, and service for coordination.

[0009] The above patent optimizes the model prediction through the BP neural network and the sparrow algorithm to achieve the prediction result of the final cost of electric power equipment manufacturing enterprises. However, the method for optimizing the weight coefficient in the above patent is too single, and the method in the above patent cannot be applied to process the parameters consumed in the production and manufacturing process of Lyocell fiber, and predict the consumption involved in the production and manufacturing process of Lyocell fiber. Summary of the Invention

[0010] In order to solve the problems existing in the above-mentioned prior art, the present application provides a way of algorithm model fusion and optimizes the model fusion weight, so as to achieve an accurate prediction of the consumption in the production and manufacturing of Lyocell fiber, a method for predicting the consumption in the production and manufacturing of Lyocell fiber.

[0011] In order to achieve the above technical effects, the technical solution of the present application is as follows:

[0012] In the first aspect, a method for predicting the consumption in the production and manufacturing of Lyocell fiber includes the following specific steps:

[0013] Step 1: Collect historical raw material data, energy consumption data, and equipment maintenance data during the production and manufacturing of Lyocell fiber; the raw material data includes wood pulp parameter data, NMMO solution parameter data, and chemical additive parameter data; the energy consumption data includes temperature data, pressure data, and flow data; the equipment maintenance data includes equipment operation time, number of failures, equipment failure rate, equipment vibration data, and equipment maintenance history.

[0014] Step 2: Perform data preprocessing on the raw material data, energy consumption data, and equipment maintenance data.

[0015] Step 3: Construct and train the total model for predicting raw material consumption, the total model for predicting energy consumption, and the total model for predicting equipment maintenance.

[0016] Step 4: Input Process Parameter Set 1, Process Parameter Set 2, and Process Parameter Set 3 in the current Lyocell fiber production and manufacturing process into the three overall models in Step 3 to obtain three raw material consumption prediction values, three energy consumption prediction values, and three equipment maintenance prediction values;

[0017] Step 5: Set the initial weight distribution of the overall models for raw material consumption prediction, energy consumption prediction, and equipment maintenance prediction, calculate the mean squared error of the three models within each overall model, then update the Beta distribution parameters of the overall model according to the mean squared error. At the same time, introduce Gaussian process regression to calculate the uncertainty of the three models within each overall model, then update the Beta distribution parameters again through the uncertainty, and then calculate the fusion weights of the three models within each overall model respectively by combining variational Bayesian inference based on the Beta distribution parameters updated again;

[0018] Step 6: Calculate the final prediction values after weighted fusion according to the three types of prediction values in Step 4 and the fusion weights of the three models within their respective overall models.

[0019] Further, the overall models for raw material consumption prediction, energy consumption prediction, and equipment maintenance prediction in Step 3 are constructed and trained based on random forest, support vector machine, and long short-term memory network; Process Parameter Set 1 in Step 4 is the raw material data in the current Lyocell fiber production and manufacturing process, Process Parameter Set 2 is the equipment running time in the energy consumption data and equipment maintenance data in the current Lyocell fiber production and manufacturing process, and Process Parameter Set 3 is the equipment maintenance data in the current Lyocell fiber production and manufacturing process; The specific method of Step 4 is: Input Process Parameter Set 1 into the overall model for raw material consumption prediction to output three raw material consumption prediction values, input Process Parameter Set 2 into the overall model for energy consumption prediction to output three energy consumption prediction values, and input Process Parameter Set 3 into the overall model for equipment maintenance prediction to output three equipment maintenance prediction values; The three types of prediction values in Step 6 include three raw material consumption prediction values, three energy consumption prediction values, and three equipment maintenance prediction values; The final prediction values include the final raw material consumption prediction value, the final energy consumption prediction value, and the final equipment maintenance prediction value; The specific method of Step 6 is: For the three obtained raw material consumption prediction values, calculate the final raw material consumption prediction value after weighted fusion according to the fusion weights of the three models within the overall model for raw material consumption prediction; For the three obtained energy consumption prediction values, calculate the final energy consumption prediction value after weighted fusion according to the fusion weights of the three models within the overall model for energy consumption prediction; For the three obtained equipment maintenance prediction values, calculate the final equipment maintenance prediction value after weighted fusion according to the fusion weights of the three models within the overall model for equipment maintenance prediction.

[0020] Furthermore, the pulp parameter data in Step 1 includes pulp viscosity, pulp degree of polymerization, and pulp methyl cellulose content; the NMMO solution parameter data includes NMMO solution concentration; the chemical additive parameter data includes propylene glycol content and hydroxylamine content; the raw material data also includes the inventory data and consumption data of raw materials; the energy consumption data also includes the equipment operation time; the temperature data includes dissolution temperature, reactor temperature, temperature of the heat preservation water system, air temperature of the spinning air system, coagulation bath temperature, and ambient temperature; the pressure data includes solvent delivery pressure, vacuum pressure, and ambient air pressure; the flow data includes NMMO solvent flow, water consumption, and steam consumption.

[0021] Even further, the pulp viscosity, pulp degree of polymerization, and pulp methyl cellulose content are collected by real-time monitoring with an online viscometer and an optical measuring instrument; the NMMO solution concentration is collected by a liquid-phase concentration detection sensor; the propylene glycol content and hydroxylamine content are detected and collected by chemical sensors; the inventory data and consumption data of raw materials are collected through an enterprise resource planning system and synchronized with an external market database.

[0022] Even further, the dissolution temperature, reactor temperature, temperature of the heat preservation water system, air temperature of the spinning air system, coagulation bath temperature, and ambient temperature are measured by PT100 platinum resistance temperature sensors or thermocouples; the solvent delivery pressure, vacuum pressure, and ambient air pressure are collected by MEMS piezoresistive pressure sensors; the NMMO solvent flow, water consumption, and steam consumption are collected by electromagnetic flowmeters and ultrasonic flowmeters, and the flow cumulative value is calculated in combination with a PLC; the reactor temperature, temperature of the heat preservation water system, pressure of the heat preservation water system, air temperature of the spinning air system, and air pressure of the spinning air system are collected through a SCADA system.

[0023] Even further, the equipment operation time, number of faults, equipment failure rate, equipment vibration data, and equipment maintenance history are recorded and collected through a MES system.

[0024] Furthermore, the specific method steps for data preprocessing of raw material data, energy consumption data, and equipment maintenance data are as follows:

[0025] Step a: Perform outlier detection on the time series data, numerical data, and non-normal distribution data in the raw material data, energy consumption data, and equipment maintenance data, remove the outliers, and finally perform missing value processing and duplicate data removal.

[0026] Step b: Normalize the raw material data, energy consumption data, and equipment maintenance data.

[0027] Step c: Extract key features from the raw material data, energy consumption data, and equipment maintenance data using Pearson correlation analysis and principal component analysis.

[0028] Step d: Use the K-means clustering analysis and information entropy screening method to perform final screening on the raw material data, energy consumption data, and equipment maintenance data.

[0029] Furthermore, the specific method for detecting outliers in the numerical data in step a is as follows: Use the 3σ principle to detect outliers. For each data point f i , calculate the mean μ and standard deviation σ of the original data. If it satisfies:

[0030] ;

[0031] then it is determined as abnormal data and is excluded; in the formula, is the i-th data point.

[0032] Furthermore, the specific method for detecting outliers in the non-normal distribution data in step a is as follows: Use the interquartile range method to detect outliers. Calculate the first quartile Q1 and the third quartile Q3. The set abnormal range is as follows:

[0033] ;

[0034] ;

[0035] Data outside this range is regarded as an outlier and is excluded; in the formula, the meaning of IQR is the interquartile range.

[0036] Furthermore, the specific method for handling missing values is as follows: For time series data, use the linear interpolation method to fill; for numerical data, if the missing ratio is less than 20%, use k-nearest neighbor interpolation to fill, and if the missing ratio exceeds 20%, then exclude this numerical data; for non-normal distribution data, if the missing ratio is less than 20%, use k-nearest neighbor interpolation to fill, and if the missing ratio exceeds 20%, then exclude this non-normal distribution data.

[0037] Furthermore, the time series data includes the inventory data of raw materials, the consumption data of raw materials, the dissolution temperature, the reactor temperature, the temperature of the heat preservation water system, the air temperature of the spinning air system, the temperature of the coagulation bath, the ambient temperature, the solvent delivery pressure, the vacuum pressure, the pressure of the heat preservation water system, the air pressure of the spinning air system, the ambient air pressure, the NMMO solvent flow rate, the water consumption, the steam consumption, the equipment operation time, and the number of failures.

[0038] Furthermore, the numerical data includes wood pulp viscosity, wood pulp polymerization degree, wood pulp methyl cellulose content, NMMO solution concentration, propylene glycol content, hydroxylamine content, and equipment failure incidence rate; the non-normal distribution data includes raw material inventory data, raw material consumption data, number of failures, equipment failure incidence rate, equipment vibration data, and equipment maintenance history.

[0039] Furthermore, for the raw material data, energy consumption data, and equipment maintenance data, data normalization is specifically performed using Min-Max normalization, and the specific formula is as follows:

[0040] ;

[0041] In the formula, x is the original data value; x min is the minimum value of the original data; x max is the maximum value of the original data;

[0042] For abnormally distributed data, Z-score standardization is used:

[0043] ;

[0044] In the formula, μ is the mean of the original data; σ is the standard deviation, is the data value after Z-score standardization, x is the original data value.

[0045] Furthermore, the specific method steps for extracting key features from the raw material data, energy consumption data, and equipment maintenance data using Pearson correlation analysis and principal component analysis in step c are as follows:

[0046] Step a1: Calculate the Pearson correlation coefficient between features, and the specific formula is as follows:

[0047] ;

[0048] When |r| > 0.8, it is determined that two features are highly correlated, and one of them is removed; in the formula, r is the Pearson correlation coefficient, h i is the i-th observation value of variable h, y i is the i-th observation value of variable y, is the mean of variable h, μ y is the mean of variable y;

[0049] Step b1: Perform principal component analysis for dimensionality reduction. First, construct a standardized data matrix X with a dimension of m×n, where m is the number of samples and n is the number of features, and then calculate the covariance matrix , and the specific calculation formula is as follows:

[0050] ;

[0051] In the formula, X T is the transpose matrix of the standardized data matrix X;

[0052] Step c1: Solve the eigenvalues and eigenvectors of the covariance matrix First, calculate the eigenvalues of the covariance matrix , sort them in descending order of eigenvalues, and calculate the cumulative contribution rate; select the first k eigenvectors with a cumulative contribution rate exceeding 95% to form the dimensionality reduction projection matrix W, obtain the projection data through the dimensionality reduction projection matrix W, and finally reduce the dimensionality of the data after principal component analysis from n dimensions to k dimensions; the specific formula for obtaining the projection data is as follows:

[0053] ;

[0054] In the formula, W is the dimensionality reduction projection matrix, and X′ is the eigen data after dimensionality reduction.

[0055] Furthermore, the specific method steps for finally screening the raw material data, energy consumption data, and equipment maintenance data by using the K-means clustering analysis and information entropy screening method are as follows:

[0056] Step a2: Use K-means clustering analysis. First, perform initialization, set the number of clusters k = 5, select k random samples as the initial cluster centers, and then calculate the Euclidean distance from the samples to the cluster centers. The specific formula is as follows:

[0057] ;

[0058] In the formula, represents the Euclidean distance between the sample point s and the cluster center c; is the i-th data sample, is the new cluster center of the i-th cluster, and n is the n-dimensional space;

[0059] Step b2: Reassign the clusters, calculate the distances from each sample point to all cluster centers, and assign the sample points to the nearest cluster;

[0060] Step c2: Update the cluster centers. First, calculate the new cluster centers and continue to iterate until convergence. When the change in the cluster centers is less than the set threshold, stop the iteration; the specific calculation formula for the new cluster centers is:

[0061] ;

[0062] In the formula, is the new cluster center of the i-th cluster, is the cluster C i the number of sample points within, and s is the data sample point;

[0063] Step d2: Filter out representative data, calculate the average distance from each cluster to the cluster center, and select 10% of the data around the cluster center in each cluster as training samples;

[0064] Step e2: Through information entropy screening, further evaluate the information content of each cluster and calculate the information entropy of each cluster , the specific formula is as follows:

[0065] ;

[0066] In the formula, Cluster The probability of category j in ; is the number of categories within the cluster;

[0067] Step f2: Filter the cluster with the highest information entropy, calculate the information entropy values ​​of all clusters, and select the top 50% clusters with the highest information entropy. Among the selected clusters, retain the samples around the center of the cluster and remove outliers to form the final training data set.

[0068] Furthermore, in the step three, the total model for raw material consumption prediction, the total model for energy consumption prediction and the total model for equipment maintenance prediction constructed and trained based on random forest, support vector machine and long short-term memory network are specifically as follows: the total model for raw material consumption prediction constructed and trained based on random forest, support vector machine and long short-term memory network includes a first raw material consumption prediction model constructed based on random forest, a second raw material consumption prediction model constructed based on support vector machine and a third raw material consumption prediction model constructed based on long short-term memory network; the total model for energy consumption prediction constructed and trained based on random forest, support vector machine and long short-term memory network includes a first energy consumption prediction model constructed based on random forest, a second energy consumption prediction model constructed based on support vector machine and a third energy consumption prediction model constructed based on long short-term memory network; the total model for equipment maintenance prediction constructed and trained based on random forest, support vector machine and long short-term memory network includes a first equipment maintenance prediction model constructed based on random forest, a second equipment maintenance prediction model constructed based on support vector machine and a third equipment maintenance prediction model constructed based on long short-term memory network.

[0069] Furthermore, the specific method steps for constructing the first raw material consumption prediction model, the first energy consumption prediction model and the first equipment maintenance prediction model based on random forest are as follows:

[0070] Step a3: Construct a first raw material consumption prediction model based on random forest; the specific construction method of the first raw material consumption prediction model is as follows: Use pulp viscosity, pulp degree of polymerization, NMMO solution concentration, pulp methyl cellulose content, propylene glycol content, hydroxylamine content, raw material inventory data, and raw material consumption data as the input features of the training set, and use the consumption of pulp, the consumption of NMMO solution, and the consumption of chemical additives as the output features of the training set to train and obtain the first raw material consumption prediction model; the consumption of chemical additives includes the consumption of propylene glycol and the consumption of hydroxylamine;

[0071] Step b3: Construct a first energy consumption prediction model based on random forest; the specific construction method of the first energy consumption prediction model is as follows: Use temperature data, pressure data, flow data, and equipment operation time as the input features of the training set, and use power consumption, water consumption, and steam consumption as the output features to train and obtain the first energy consumption prediction model;

[0072] Step c3: Construct a first equipment maintenance prediction model based on random forest; the specific construction method of the first equipment maintenance prediction model is as follows: Use equipment operation time, number of failures, equipment failure rate, equipment vibration data, and equipment maintenance history as the input features of the training set, and use equipment failure rate and number of maintenance times as the output features to train and obtain the first equipment maintenance prediction model.

[0073] Furthermore, the specific method steps for constructing a second raw material consumption prediction model, a second energy consumption prediction model, and a second equipment maintenance prediction model based on support vector machine are as follows:

[0074] Step a4: Construct a second raw material consumption prediction model based on support vector machine; the specific construction method of the second raw material consumption prediction model is as follows: Use pulp viscosity, pulp degree of polymerization, NMMO solution concentration, pulp methyl cellulose content, propylene glycol content, hydroxylamine content, raw material inventory data, and raw material consumption data as the input features of the training set, and use the consumption of pulp, the consumption of NMMO solution, and the consumption of chemical additives as the output features of the training set to train and obtain the second raw material consumption prediction model; the consumption of chemical additives includes the consumption of propylene glycol and the consumption of hydroxylamine;

[0075] Step b4: Construct a second energy consumption prediction model based on support vector machine; the specific construction method of the second energy consumption prediction model is as follows: Use temperature data, pressure data, flow data, and equipment operation time as the input features of the training set, and use power consumption, water consumption, and steam consumption as the output features to train and obtain the second energy consumption prediction model;

[0076] Step c4: Build the second equipment maintenance prediction model based on the support vector machine; the specific construction method of the second equipment maintenance prediction model is as follows: Take the equipment operation time, the number of failures, the equipment failure rate, the equipment vibration data, and the equipment maintenance history as the input features of the training set, and take the equipment failure rate and the number of maintenance times as the output features, and train to obtain the second equipment maintenance prediction model.

[0077] Furthermore, the specific method steps for building the third raw material consumption prediction model, the third energy consumption prediction model, and the third equipment maintenance prediction model based on the long short-term memory network are as follows:

[0078] Step a5: Build the third raw material consumption prediction model based on the long short-term memory network; the specific construction method of the third raw material consumption prediction model is as follows: Take the viscosity of wood pulp, the degree of polymerization of wood pulp, the concentration of NMMO solution, the methyl cellulose content of wood pulp, the content of propylene glycol, the content of hydroxylamine, the inventory data of raw materials, and the consumption data of raw materials as the input features of the training set, and take the consumption of wood pulp, the consumption of NMMO solution, and the consumption of chemical additives as the output features of the training set, and train to obtain the third raw material consumption prediction model; the consumption of chemical additives includes the consumption of propylene glycol and the consumption of hydroxylamine;

[0079] Step b5: Build the third energy consumption prediction model based on the long short-term memory network; the specific construction method of the third energy consumption prediction model is as follows: Take the temperature data, the pressure data, the flow data, and the equipment operation time as the input features of the training set, and take the power consumption, the water consumption, and the steam consumption as the output features, and train to obtain the third energy consumption prediction model;

[0080] Step c5: Build the third equipment maintenance prediction model based on the long short-term memory network; the specific construction method of the third equipment maintenance prediction model is as follows: Take the equipment operation time, the number of failures, the equipment failure rate, the equipment vibration data, and the equipment maintenance history as the input features of the training set, and take the equipment failure rate and the number of maintenance times as the output features, and train to obtain the third equipment maintenance prediction model.

[0081] Further, set the initial weight distribution of the total models for raw material consumption prediction, energy consumption prediction, and equipment maintenance prediction in step five, calculate the mean square error of the three models within each total model, then introduce Gaussian process regression to calculate the uncertainty of the three models within each total model, and finally calculate the fusion weights of the three models within each total model respectively according to variational Bayesian inference. Specifically: Assume that the weights of the three models within each total model follow a Beta distribution:

[0082] ;

[0083] In the formula, are the weight coefficients of the three models inside the total model for raw material consumption prediction, the weight coefficients of the three models inside the total model for energy consumption prediction, or the weight coefficients of the three models inside the total model for equipment maintenance prediction, and are set as initial parameters, generally taking:

[0084] ;

[0085] In the formula, is a positive number to ensure uniform distribution of weights in the initial state;

[0086] For the prediction errors of the three models inside the total model for raw material consumption prediction, the prediction errors of the three models inside the total model for energy consumption prediction, or the prediction errors of the three models inside the total model for equipment maintenance prediction, they are evaluated by calculating the mean squared errors of the three models inside the total model for raw material consumption prediction, the mean squared errors of the three models inside the total model for energy consumption prediction, or the mean squared errors of the three models inside the total model for equipment maintenance prediction. The specific calculation formula is as follows:

[0087] ;

[0088] ;

[0089] ;

[0090] In the formula, represents the mean squared error of the first raw material consumption prediction model, the first energy consumption prediction model, or the first equipment maintenance prediction model based on random forest; represents the mean squared error of the second raw material consumption prediction model, the second energy consumption prediction model, or the second equipment maintenance prediction model based on support vector machine; represents the mean squared error of the third raw material consumption prediction model, the third energy consumption prediction model, or the third equipment maintenance prediction model based on long short-term memory network; N represents the total number of training samples; represents the true value of the i-th sample; represents the predicted value of the i-th sample by the first raw material consumption prediction model, the first energy consumption prediction model, or the first equipment maintenance prediction model based on random forest; represents the predicted value of the i-th sample by the second raw material consumption prediction model, the second energy consumption prediction model, or the second equipment maintenance prediction model based on support vector machine; denotes the predicted value of the $i$-th sample by the third raw material consumption prediction model, the third energy consumption prediction model, or the third equipment maintenance prediction model based on the long short-term memory network;

[0091] Then update the Beta distribution parameters according to the mean square error:

[0092] ;

[0093] ;

[0094] In the formula, denotes a distribution parameter in the calculation of the fusion weight of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction, denotes another distribution parameter in the calculation of the fusion weight of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction; represents the mean square error of the total model for raw material consumption prediction, the mean square error of the total model for energy consumption prediction, or the mean square error of the total model for equipment maintenance prediction;

[0095] Introduce Gaussian process regression to calculate the uncertainty of each model. The specific formula is as follows:

[0096] ;

[0097] In the formula, means the uncertainty of the model; $N$ is the total number of training samples; denotes the prediction variance of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction on all samples. The specific solution formula is as follows:

[0098] ;

[0099] In the formula, denotes the predicted value of the $i$-th sample by the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction; denotes the average predicted value of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction on all samples;

[0100] Then introduce the uncertainty into the Beta distribution update:

[0101] ;

[0102] ;

[0103] In the formula, represents a distribution parameter in the calculation of the fusion weight of the updated total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction; represents another distribution parameter in the calculation of the fusion weight of the updated total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction;

[0104] Finally, variational Bayesian inference is used to calculate the final weight:

[0105] ;

[0106] In the formula, is the model weight after the (t + 1)-th iteration of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction; represents the contribution of the actual data;

[0107] Then, the fusion weights of the three models within each model are obtained respectively:

[0108] ;

[0109] ;

[0110] ;

[0111] In the formula, The meaning of is the weight coefficient of the first raw material consumption prediction model, the first energy consumption prediction model, or the first equipment maintenance prediction model; The meaning of is the weight coefficient of the second raw material consumption prediction model, the second energy consumption prediction model, or the second equipment maintenance prediction model; The meaning of is the weight coefficient of the third raw material consumption prediction model, the third energy consumption prediction model, or the third equipment maintenance prediction model; The meaning of is a distribution parameter in the calculation of the fusion weight of the updated first raw material consumption prediction model, the first energy consumption prediction model, or the first equipment maintenance prediction model; The meaning of is a distribution parameter in the calculation of the fusion weight of the updated second raw material consumption prediction model, the second energy consumption prediction model, or the second equipment maintenance prediction model; The meaning of is a distribution parameter in the calculation of the fusion weight of the updated third raw material consumption prediction model, the third energy consumption prediction model, or the third equipment maintenance prediction model; The meaning of is another distribution parameter in the calculation of the fusion weight of the updated first raw material consumption prediction model, the first energy consumption prediction model, or the first equipment maintenance prediction model; It refers to another distribution parameter in the calculation of the fusion weight of the updated second raw material consumption prediction model, the second energy consumption prediction model, or the second equipment maintenance prediction model; It refers to another distribution parameter in the calculation of the fusion weight of the updated third raw material consumption prediction model, the third energy consumption prediction model, or the third equipment maintenance prediction model; It refers to the sum of the predicted output values of all training samples of the first raw material consumption prediction model, the first energy consumption prediction model, or the first equipment maintenance prediction model based on a random forest; It refers to the sum of the predicted output values of all training samples of the second raw material consumption prediction model, the second energy consumption prediction model, or the second equipment maintenance prediction model based on a support vector machine; It refers to the sum of the predicted output values of all training samples of the third raw material consumption prediction model, the third energy consumption prediction model, or the third equipment maintenance prediction model based on a long short-term memory network.

[0112] Furthermore, the process parameter group 1 includes wood pulp viscosity, wood pulp degree of polymerization, wood pulp methyl cellulose content, NMMO solution concentration, propylene glycol content, hydroxylamine content, inventory data of raw materials, and consumption data of raw materials; the predicted raw material consumption values include wood pulp consumption, NMMO solution consumption, and chemical additive consumption; the chemical additive consumption includes propylene glycol consumption and hydroxylamine consumption; the process parameter group 2 includes temperature data, pressure data, flow data, and equipment operation time; the predicted energy consumption values include electricity consumption, water consumption, and steam consumption; the process parameter group 3 includes equipment operation time, number of failures, equipment failure rate, equipment vibration data, and equipment maintenance history; the predicted equipment maintenance values include equipment failure rate and number of equipment maintenance times.

[0113] Furthermore, the specific formulas for calculating the final predicted raw material consumption value, the final predicted energy consumption value, and the final predicted equipment maintenance value after final weighted fusion are as follows:

[0114] ;

[0115] In the formula, represents the final predicted raw material consumption value, the final predicted energy consumption value, or the final predicted equipment maintenance value; when represents the predicted raw material consumption value of the first raw material consumption prediction model based on a random forest, represents the predicted raw material consumption value of the second raw material consumption prediction model based on a support vector machine, represents the predicted raw material consumption value of the third raw material consumption prediction model based on a long short-term memory network, then representing the weight coefficients of the first raw material consumption prediction model, representing the weight coefficients of the second raw material consumption prediction model, and representing the weight coefficients of the third raw material consumption prediction model; when representing the energy consumption prediction value of the first energy consumption prediction model based on a random forest, representing the energy consumption prediction value of the second energy consumption prediction model based on a support vector machine, representing the energy consumption prediction value of the third energy consumption prediction model based on a long short-term memory network, then representing the weight coefficients of the first energy consumption prediction model, representing the weight coefficients of the second energy consumption prediction model, and representing the weight coefficients of the third energy consumption prediction model; when representing the equipment maintenance prediction value of the first equipment maintenance prediction model based on a random forest, representing the equipment maintenance prediction value of the second equipment maintenance prediction model based on a support vector machine, representing the equipment maintenance prediction value of the third equipment maintenance prediction model based on a long short-term memory network, then representing the weight coefficients of the first equipment maintenance prediction model, representing the weight coefficients of the second equipment maintenance prediction model, and representing the weight coefficients of the third equipment maintenance prediction model.

[0116] In a second aspect, an application of a consumption prediction method for Lyocell fiber production and manufacturing is applicable to the consumption prediction of Lyocell fiber production and manufacturing.

[0117] According to the above technical solution, the beneficial effects of the present application are as follows:

[0118] 1. The method adopted by the present invention optimizes the weight coefficients, accurately predicts the raw material consumption, energy consumption, and equipment maintenance, provides a scientific basis for the optimization of the Lyocell process and resource scheduling, and assists the enterprise to achieve refined management in the process of Lyocell fiber production and manufacturing.

[0119] 2. The method adopted by the present invention constructs and trains the total model for raw material consumption prediction, the total model for energy consumption prediction, and the total model for equipment maintenance prediction based on a random forest, a support vector machine, and a long short-term memory network, and then accurately obtains the final raw material consumption prediction value, the final energy consumption prediction value, and the final equipment maintenance prediction value respectively through weighted fusion of the three obtained raw material consumption prediction values, three energy consumption prediction values, and three equipment maintenance prediction values, improving the accuracy of the final prediction.

[0120] 3. The method of the present invention combines variational Bayesian inference, Gaussian process regression, and Beta distribution update to optimize the calculation of model fusion weights, making it adaptable to different data distributions and improving the stability and accuracy of the final prediction.

[0121] 4. The method of the present invention uses Gaussian process regression to calculate the uncertainties of the three models within each total model and incorporates them into the weight update rule, achieving the technical effect of reducing the weights of models with high uncertainties and improving prediction stability.

[0122] 5. Compared with the traditional Beta distribution update, the Beta distribution update adopted by the method of the present invention uses variational inference to achieve the dynamic correlation between the parameter update of the Beta distribution and data errors and uncertainties, achieving the technical effect of improving the speed and adaptability of weight adjustment.

[0123] 6. The method adopted by the present invention calculates the fusion weights of the three models within each total model through variational Bayesian inference, achieving the dynamic optimization of model weights, ensuring that the weight calculation is not only based on the error distribution but also adaptable to different data characteristics, and improving the technical effect of prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0124] Figure 1 is the structural block diagram of the present invention.

[0125] Figure 2 is the process schematic diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0126] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments.

[0127] Embodiment 1

[0128] A method for predicting the consumption in the production and manufacturing of Lyocell fibers includes the following specific steps:

[0129] Step 1: Collect historical raw material data, energy consumption data, and equipment maintenance data during the production and manufacturing of Lyocell fibers; the raw material data includes wood pulp parameter data, NMMO solution parameter data, and chemical additive parameter data; the energy consumption data includes temperature data, pressure data, and flow data; the equipment maintenance data includes equipment operation time, number of failures, equipment failure rate, equipment vibration data, and equipment maintenance history;

[0130] Step 2: Perform data preprocessing on the raw material data, energy consumption data, and equipment maintenance data; data preprocessing is a mature technology in the field, and its purpose is to remove outliers in the data and handle missing values;

[0131] Step 3: Construct and train the total model for raw material consumption prediction, the total model for energy consumption prediction, and the total model for equipment maintenance prediction based on random forest, support vector machine, and long short-term memory network; random forest, support vector machine, and long short-term memory network are all existing technologies in the field;

[0132] Step 4: Take the raw material data in the current Lyocell fiber production and manufacturing process as process parameter group 1 and input it into the total model for raw material consumption prediction to obtain three raw material consumption prediction values. Take the energy consumption data and the equipment operation time in the equipment maintenance data as process parameter group 2 and input it into the total model for energy consumption prediction to obtain three energy consumption prediction values. Take the equipment maintenance data as process parameter group 3 and input it into the total model for equipment maintenance prediction to obtain three equipment maintenance prediction values;

[0133] Step 5: Set the initial weight distribution of the total model for raw material consumption prediction, the total model for energy consumption prediction, and the total model for equipment maintenance prediction, and calculate the mean square error of the three models within each total model. Then update the Beta distribution parameters according to the mean square error. At the same time, introduce Gaussian process regression to calculate the uncertainty of the three models within each total model, and then update the Beta distribution parameters again through the uncertainty. Calculate the fusion weights of the three models within each total model respectively according to the Beta distribution parameters updated again combined with variational Bayesian inference;

[0134] Step 6: For the three obtained raw material consumption prediction values, calculate the final weighted fusion raw material consumption prediction value according to the fusion weights of the three models within the total model for raw material consumption prediction; for the three obtained energy consumption prediction values, calculate the final weighted fusion energy consumption prediction value according to the fusion weights of the three models within the total model for energy consumption prediction; for the three obtained equipment maintenance prediction values, calculate the final weighted fusion equipment maintenance prediction value according to the fusion weights of the three models within the total model for equipment maintenance prediction.

[0135] Example 2

[0136] A method for predicting the consumption in the production and manufacturing of Lyocell fiber, including the following specific steps:

[0137] Step 1: Collect historical raw material data, energy consumption data, and equipment maintenance data during the production and manufacturing process of Lyocell fiber; the raw material data includes wood pulp parameter data, NMMO solution parameter data, and chemical additive parameter data; the energy consumption data includes temperature data, pressure data, and flow rate data; the equipment maintenance data includes equipment operation time, number of failures, equipment failure rate, equipment vibration data, and equipment maintenance history;

[0138] Step 2: Perform data preprocessing on the raw material data, energy consumption data, and equipment maintenance data;

[0139] Step 3: Construct and train the overall model for raw material consumption prediction, the overall model for energy consumption prediction, and the overall model for equipment maintenance prediction based on random forest, support vector machine, and long short-term memory network;

[0140] Step 4: Input the raw material data during the current Lyocell fiber production and manufacturing process as process parameter group 1 into the overall model for raw material consumption prediction to obtain three raw material consumption prediction values, input the energy consumption data and the equipment operation time in the equipment maintenance data as process parameter group 2 into the overall model for energy consumption prediction to obtain three energy consumption prediction values, and input the equipment maintenance data as process parameter group 3 into the overall model for equipment maintenance prediction to obtain three equipment maintenance prediction values;

[0141] Step 5: Set the initial weight distribution of the overall model for raw material consumption prediction, the overall model for energy consumption prediction, and the overall model for equipment maintenance prediction, calculate the mean squared error of the three models within each overall model, then update the Beta distribution parameters according to the mean squared error, at the same time introduce Gaussian process regression to calculate the uncertainty of the three models within each overall model, and then update the Beta distribution parameters again through the uncertainty, and calculate the fusion weights of the three models within each overall model respectively according to the Beta distribution parameters updated again combined with variational Bayesian inference;

[0142] Step 6: Calculate the final raw material consumption prediction value after final weighted fusion for the three obtained raw material consumption prediction values according to the fusion weights of the three models within the overall model for raw material consumption prediction; calculate the final energy consumption prediction value after final weighted fusion for the three obtained energy consumption prediction values according to the fusion weights of the three models within the overall model for energy consumption prediction; calculate the final equipment maintenance prediction value after final weighted fusion for the three obtained equipment maintenance prediction values according to the fusion weights of the three models within the overall model for equipment maintenance prediction;

[0143] In Step 1, the pulp parameter data includes pulp viscosity, pulp degree of polymerization, and pulp methyl cellulose content; the NMMO solution parameter data includes NMMO solution concentration; the chemical additive parameter data includes propylene glycol content and hydroxylamine content; the raw material data also includes the inventory data and consumption data of raw materials; the energy consumption data also includes the equipment operation time; the temperature data includes dissolution temperature, reactor temperature, temperature of the heat preservation water system, air temperature of the spinning air system, coagulation bath temperature, and ambient temperature; the pressure data includes solvent delivery pressure, vacuum pressure, and ambient air pressure; the flow data includes NMMO solvent flow rate, water consumption, and steam consumption.

[0144] The pulp viscosity, pulp degree of polymerization, and pulp methyl cellulose content are monitored and collected in real time by an on-line viscometer and an optical measuring instrument; the NMMO solution concentration is collected by a liquid-phase concentration detection sensor; the propylene glycol content and hydroxylamine content are detected and collected by chemical sensors; the inventory data and consumption data of raw materials are collected through an enterprise resource planning system and synchronized with an external market database.

[0145] The dissolution temperature, reactor temperature, temperature of the heat preservation water system, air temperature of the spinning air system, coagulation bath temperature, and ambient temperature are measured by PT100 platinum resistance temperature sensors or thermocouples; the solvent delivery pressure, vacuum pressure, and ambient air pressure are collected by MEMS piezoresistive pressure sensors; the NMMO solvent flow rate, water consumption, and steam consumption are collected by electromagnetic flow meters and ultrasonic flow meters, and the flow cumulative value is calculated in combination with a PLC; the reactor temperature, temperature of the heat preservation water system, pressure of the heat preservation water system, air temperature of the spinning air system, and air pressure of the spinning air system are collected by a SCADA system.

[0146] The equipment operation time, number of faults, equipment failure rate, equipment vibration data, and equipment maintenance history are recorded and collected by a MES system.

[0147] Example 3

[0148] On the basis of Example 2, the specific method steps for data preprocessing of raw material data, energy consumption data, and equipment maintenance data are as follows:

[0149] Step a: Perform outlier detection on the time series data, numerical data, and non-normal distribution data in the raw material data, energy consumption data, and equipment maintenance data, remove the outliers, and finally perform missing value processing and duplicate data removal.

[0150] Step b: Normalize the raw material data, energy consumption data, and equipment maintenance data.

[0151] Step c: Extract key features from the raw material data, energy consumption data, and equipment maintenance data by using Pearson correlation analysis and principal component analysis.

[0152] Step d: Use the K-means clustering analysis and information entropy screening method to perform final screening on the raw material data, energy consumption data, and equipment maintenance data.

[0153] The specific method for detecting outliers in numerical data in step a is: Use the 3σ principle to detect outliers. For each data point f i , calculate the mean μ and standard deviation σ of the original data. If it satisfies:

[0154] ;

[0155] Then it is determined as abnormal data and eliminated; where is the i-th data point.

[0156] For outliers in non-normal distributions, use Z-score standardization to make them satisfy the standard normal distribution. The specific method for detecting outliers in non-normal distribution data in step a is: Use the interquartile range method to detect outliers. Calculate the first quartile Q1 and the third quartile Q3. The set abnormal range is as follows:

[0157] ;

[0158] ;

[0159] Data outside this range is regarded as an outlier and eliminated; where the meaning of IQR is the interquartile range.

[0160] The specific method for handling missing values is: For time series data, use the linear interpolation method to fill; for numerical data, if the missing ratio is less than twenty percent, use k-nearest neighbor interpolation to fill. If the missing exceeds twenty percent, then eliminate this numerical data; for non-normal distribution data, if the missing ratio is less than twenty percent, use k-nearest neighbor interpolation to fill. If the missing exceeds twenty percent, then eliminate this non-normal distribution data.

[0161] Time series data includes the inventory data of raw materials, the consumption data of raw materials, the dissolution temperature, the reactor temperature, the temperature of the heat preservation water system, the air temperature of the spinning air system, the temperature of the coagulation bath, the ambient temperature, the solvent delivery pressure, the vacuum pressure, the pressure of the heat preservation water system, the air pressure of the spinning air system, the ambient air pressure, the NMMO solvent flow rate, the water consumption, the steam consumption, the equipment operation time, and the number of faults.

[0162] Numerical data includes wood pulp viscosity, wood pulp degree of polymerization, wood pulp methyl cellulose content, NMMO solution concentration, propylene glycol content, hydroxylamine content, and equipment failure rate; non-normal distribution data includes the inventory data of raw materials, the consumption data of raw materials, the number of faults, the equipment failure rate, the equipment vibration data, and the equipment maintenance history.

[0163] For the raw material data, energy consumption data, and equipment maintenance data, data normalization is specifically performed using Min-Max normalization. The specific formula is as follows:

[0164] ;

[0165] In the formula, x is the original data value; x min is the minimum value of the original data; x max is the maximum value of the original data;

[0166] For data with abnormal distributions, Z-score standardization is used:

[0167] ;

[0168] In the formula, μ is the mean of the original data; σ is the standard deviation, is the data value after Z-score standardization, x is the original data value.

[0169] In step c, the specific method steps for extracting key features from the raw material data, energy consumption data, and equipment maintenance data using Pearson correlation analysis and principal component analysis are as follows:

[0170] Step a1: Calculate the correlation features between data of the same category and different categories, analyze the linear correlation between variables, and then screen out redundant features. Calculate the Pearson correlation coefficients between features. Specifically, calculate the Pearson correlation coefficients between the following features: Inside the raw material data, analyze the correlation between the viscosity of wood pulp, the degree of polymerization of wood pulp, and the methyl cellulose content of wood pulp to judge their linear dependence relationship in the dataset; at the same time, analyze the correlation between the concentration of NMMO solution, the content of propylene glycol, and the content of hydroxylamine to evaluate the impact of chemical additives on the production process;

[0171] Inside the energy consumption data, calculate the correlation between temperature data to judge the temperature control relationship in different production stages. At the same time, analyze the interaction between pressure data to optimize the pressure regulation mechanism; in addition, calculate the correlation between flow rate data to optimize the solvent and energy consumption strategy.

[0172] Inside the equipment maintenance data, calculate the correlation between the equipment operation time, the number of failures, and the equipment failure rate to evaluate the impact of equipment aging on the failure rate; at the same time, analyze the relationship between the equipment operation time and the number of failures to optimize the maintenance budget and repair plan; in addition, conduct a correlation analysis of cross-category features. For example: Calculate the correlation between the consumption data of raw materials and the energy consumption data to evaluate the relationship between raw material consumption and energy use. Calculate the correlation between the equipment operation time and the energy consumption data to analyze the impact of the equipment operation status on energy consumption;

[0173] The specific calculation formula is as follows:

[0174] ;

[0175] When ∣r∣ > 0.8, it is determined that two features are highly correlated, and one of them is removed; where r is the Pearson correlation coefficient, h i is the i-th observation of variable h, and y i is the i-th observation of variable y, is the mean value of variable h, and μ y is the mean value of variable y;

[0176] Step b1: Perform principal component analysis for dimensionality reduction. First, construct a standardized data matrix X with dimensions m×n, where m is the number of samples and n is the number of features. Then calculate the covariance matrix , and the specific calculation formula is as follows:

[0177] ;

[0178] where X T is the transpose matrix of the standardized data matrix X;

[0179] Step c1: Solve the eigenvalues and eigenvectors of the covariance matrix . First, calculate the eigenvalues λ1, λ2,..., λn of the covariance matrix , sort them in descending order of eigenvalues, and calculate the cumulative contribution rate; select the first k eigenvectors with a cumulative contribution rate exceeding 95% to form a dimensionality reduction projection matrix W. Obtain the projection data through the dimensionality reduction projection matrix W, and finally reduce the dimension of the data after principal component analysis from n dimensions to k dimensions; the reason for choosing 95% as the cumulative contribution rate threshold is that it can achieve a good balance between information retention and dimensionality reduction effect. If the cumulative contribution rate is too low, it may lead to the loss of important information and affect the accuracy of data analysis; if it is too high, the dimensionality reduction effect is limited and the computational complexity cannot be effectively reduced; therefore, 95% as an empirical threshold can ensure that most information is retained, while significantly reducing the data dimension, improving the generalization ability and computational efficiency of the model; subsequently, perform a linear transformation on the original data through the dimensionality reduction projection matrix W to obtain the projection data, and finally reduce the dimension of the data after principal component analysis from n dimensions to k dimensions; the specific formula for obtaining the projection data is as follows:

[0180] ;

[0181] where W is the dimensionality reduction projection matrix and X′ is the feature data after dimensionality reduction.

[0182] The K-means clustering analysis and information entropy screening method are used to finally screen the raw material data, energy consumption data, and equipment maintenance data to extract key features and remove redundant data, improving the calculation efficiency and prediction accuracy of the model; in terms of raw material data, the viscosity of wood pulp, degree of polymerization of wood pulp, methyl cellulose content of wood pulp, NMMO solution concentration, propylene glycol content, hydroxylamine content, inventory data of raw materials, and consumption data of raw materials are screened to determine which raw material parameters have the greatest impact on energy consumption; in terms of energy consumption data, the dissolution temperature, reactor temperature, temperature of the heat preservation water system, air temperature of the spinning air system, coagulation bath temperature, ambient temperature, solvent delivery pressure, vacuum pressure, pressure of the heat preservation water system, air pressure of the spinning air system, ambient air pressure, NMMO solvent flow rate, water consumption, and steam consumption are screened to extract key energy consumption features and optimize energy utilization efficiency; in terms of equipment maintenance data, the equipment operation time, number of failures, and equipment failure rate are mainly screened to analyze the impact of equipment status on production efficiency and optimize the maintenance strategy. The specific method steps for the final screening are as follows:

[0183] Step a2: Use K-means clustering analysis. First, perform initialization. Set the number of clusters k = 5, and select k random samples as the initial cluster centers. Then calculate the Euclidean distance from the samples to the cluster centers. The specific formula is as follows:

[0184] ;

[0185] In the formula, The meaning of represents the Euclidean distance between the sample point s and the cluster center c, that is, it measures the similarity between the data point and the cluster center. The smaller the distance, the closer the data point is to the cluster center; is the i-th data sample, is the new cluster center of the i-th cluster, and n is the n-dimensional space;

[0186] Step b2: Reassign the clusters. Calculate the distance from each sample point to all cluster centers and assign the sample point to the nearest cluster;

[0187] Step c2: Update the cluster centers. First, calculate the new cluster centers and continue to iterate until convergence. When the change in the cluster centers is less than the set threshold, stop the iteration. The specific calculation formula for the new cluster centers is:

[0188] ;

[0189] In the formula, is the new cluster center of the i-th cluster, is the cluster C i The number of sample points within, and s is the data sample point;

[0190] Step d2: Filter out representative data, calculate the average distance from each cluster to the cluster center, and select 10% of the data around the cluster center in each cluster as training samples;

[0191] Step e2: Although the most representative samples have been selected through the intra-cluster distance in step d2, the importance and information content of different clusters may still vary greatly; some clusters may contain too much redundant information, while some clusters may contribute more to the data distribution; therefore, it is necessary to further evaluate the information content of each cluster through information entropy screening and calculate the information entropy of each cluster. , the specific formula is as follows:

[0192] ;

[0193] In the formula, Cluster The probability of category j in ; is the number of categories within the cluster;

[0194] Step f2: Filter the cluster with the highest information entropy, calculate the information entropy values ​​of all clusters, and select the top 50% clusters with the highest information entropy. Among the selected clusters, retain the samples around the center of the cluster and remove outliers to form the final training data set.

[0195] Example 4

[0196] Based on Example 3, in step three, the total model for raw material consumption prediction, the total model for energy consumption prediction and the total model for equipment maintenance prediction are constructed and trained based on random forest, support vector machine and long short-term memory network. Specifically, the total model for raw material consumption prediction constructed and trained based on random forest, support vector machine and long short-term memory network includes a first raw material consumption prediction model constructed based on random forest, a second raw material consumption prediction model constructed based on support vector machine and a third raw material consumption prediction model constructed based on long short-term memory network; the total model for energy consumption prediction constructed and trained based on random forest, support vector machine and long short-term memory network includes a first energy consumption prediction model constructed based on random forest, a second energy consumption prediction model constructed based on support vector machine and a third energy consumption prediction model constructed based on long short-term memory network; the total model for equipment maintenance prediction constructed and trained based on random forest, support vector machine and long short-term memory network includes a first equipment maintenance prediction model constructed based on random forest, a second equipment maintenance prediction model constructed based on support vector machine and a third equipment maintenance prediction model constructed based on long short-term memory network.

[0197] The specific method steps for constructing the first raw material consumption prediction model, the first energy consumption prediction model, and the first equipment maintenance prediction model based on random forest are as follows:

[0198] Step a3: Construct the first raw material consumption prediction model based on random forest; the specific construction method of the first raw material consumption prediction model is: taking the viscosity of wood pulp, the degree of polymerization of wood pulp, the concentration of NMMO solution, the methyl cellulose content of wood pulp, the content of propylene glycol, the content of hydroxylamine, the inventory data of raw materials, and the consumption data of raw materials as the input features of the training set, and taking the consumption of wood pulp, the consumption of NMMO solution, and the consumption of chemical additives as the output features of the training set, and training to obtain the first raw material consumption prediction model; the consumption of chemical additives includes the consumption of propylene glycol and the consumption of hydroxylamine;

[0199] Step b3: Construct the first energy consumption prediction model based on random forest; the specific construction method of the first energy consumption prediction model is: taking the temperature data, pressure data, flow data, and equipment operation time as the input features of the training set, and taking the power consumption, water consumption, and steam consumption as the output features, and training to obtain the first energy consumption prediction model;

[0200] Step c3: Construct the first equipment maintenance prediction model based on random forest; the specific construction method of the first equipment maintenance prediction model is: taking the equipment operation time, the number of failures, the equipment failure rate, the equipment vibration data, and the equipment maintenance history as the input features of the training set, and taking the equipment failure rate and the number of maintenance times as the output features, and training to obtain the first equipment maintenance prediction model.

[0201] During the training process, in order to prevent overfitting and underfitting, according to experience, the random forest model consists of 100 decision trees, and the maximum depth of each tree is set to 15 to prevent overfitting; in the training stage, first, the Bootstrap sampling method is adopted to randomly extract subsamples from the training dataset, that is, randomly draw m samples with replacement from the original dataset as the training dataset for each tree, and each sample may be drawn multiple times or not selected, thus forming multiple different data subsets; the main purpose of doing this is to improve the generalization ability of the model, reduce overfitting, and enhance the stability of the model; in addition, when each decision tree splits at a node, randomly select some features for decision-making to ensure the diversity of the model, further reduce the dependence on a single feature, and improve the overall prediction effect.

[0202] For regression tasks, such as raw material consumption prediction, energy consumption prediction, or equipment maintenance prediction, the mean squared error (MSE) is used as the loss function to minimize the error between the predicted value and the true value, and the calculation formula is as follows:

[0203] ;

[0204] Among them, is the true value, is the predicted value, and N is the number of samples; for classification tasks (such as equipment failure prediction), the Gini index is used as the feature splitting criterion to measure the impurity of the data, and the calculation formula is as follows:

[0205] ;

[0206] Among them, p i is the sample proportion of class k in the current node, k is the total number of classes.

[0207] After training is completed, the test data set is used for prediction. The final output result of the random forest is calculated through the integration of multiple decision trees to obtain the final predicted value. The final predicted value is the weighted average of all decision trees:

[0208] ;

[0209] Among them, T is the number of decision trees, is the t th tree's predicted value.

[0210] The construction process of the random forest based on training is a mature technology existing in the art. The innovation point of this part lies in the matching of input data features and output features, so as to obtain a predictable model through training and finally obtain the predicted value.

[0211] The specific method steps for constructing the second raw material consumption prediction model, the second energy consumption prediction model, and the second equipment maintenance prediction model based on the support vector machine are as follows:

[0212] Step a4: Construct the second raw material consumption prediction model based on the support vector machine; the specific construction method of the second raw material consumption prediction model is: taking the viscosity of wood pulp, the degree of polymerization of wood pulp, the concentration of NMMO solution, the methyl cellulose content of wood pulp, the content of propylene glycol, the content of hydroxylamine, the inventory data of raw materials, and the consumption data of raw materials as the input features of the training set, and taking the consumption of wood pulp, the consumption of NMMO solution, and the consumption of chemical additives as the output features of the training set, and training to obtain the second raw material consumption prediction model; the consumption of chemical additives includes the consumption of propylene glycol and the consumption of hydroxylamine;

[0213] Step b4: Construct a second energy consumption prediction model based on support vector machine; the specific construction method of the second energy consumption prediction model is as follows: Use temperature data, pressure data, flow data, and equipment operation time as input features of the training set, and use power consumption, water consumption, and steam consumption as output features to train and obtain the second energy consumption prediction model;

[0214] Step c4: Construct a second equipment maintenance prediction model based on support vector machine; the specific construction method of the second equipment maintenance prediction model is as follows: Use equipment operation time, number of failures, equipment failure rate, equipment vibration data, and equipment maintenance history as input features of the training set, and use equipment failure rate and number of maintenance times as output features to train and obtain the second equipment maintenance prediction model.

[0215] During the training process, the optimization objective of support vector regression is to minimize the ε-insensitive loss function, that is, through the combination of the regularization term and the error penalty term, to ensure the generalization ability of the model. Its objective function is as follows:

[0216] ;

[0217] In w1 is the model weight vector, b is the bias term, N is the number of samples, is the regularization parameter (the optimization range is set to [0.1, 100]), which is used to control the balance between the error and the model complexity; is a slack variable, is another slack variable, dealing with samples that exceed the error tolerance interval ε ; ε is the error tolerance interval (the optimization range is set to [0.01, 1]), that is, when the error is less than ε , it is not included in the loss.

[0218] To enhance the model's learning ability for non-linear data, the present invention uses a radial basis kernel function for high-dimensional mapping, and its calculation formula is as follows:

[0219] ;

[0220] Among them, is the kernel parameter, which controls the influence range of data mapping to the high-dimensional space. The optimization range is set to [10 −3 , 10], and the optimal value is selected through cross-validation; is the input data point, is the center point of the radial basis kernel function.

[0221] At the same time, cross-validation is used to optimize the hyperparameter , γ, and ε to ensure the optimal model configuration; secondly, the model is optimized using stochastic gradient descent, and the maximum number of iterations is set to 1000 to ensure that the model converges within a reasonable calculation time. Its update formula is:

[0222] ;

[0223] where, is the weight of the t+ 1st round of iteration, η is the learning rate, is the gradient of the loss function; finally, the model predicts using the trained weight parameters w2 and the bias term b for prediction:

[0224] ;

[0225] where, is the feature vector after kernel mapping, is the final predicted value, and the transpose of the weight parameter w2 .

[0226] The training construction process based on the support vector machine is a mature technology existing in the art. The innovation point of this part lies in the matching of input data features and output features, so as to obtain a predictable model through training and finally obtain the predicted value.

[0227] The specific method steps for constructing the third raw material consumption prediction model, the third energy consumption prediction model, and the third equipment maintenance prediction model based on the long short-term memory network are as follows:

[0228] Step a5: Construct the third raw material consumption prediction model based on the long short-term memory network; the specific construction method of the third raw material consumption prediction model is: using the viscosity of wood pulp, the degree of polymerization of wood pulp, the concentration of NMMO solution, the methyl cellulose content of wood pulp, the content of propylene glycol, the content of hydroxylamine, the inventory data of raw materials, and the consumption data of raw materials as the input features of the training set, and using the consumption of wood pulp, the consumption of NMMO solution, and the consumption of chemical additives as the output features of the training set to train and obtain the third raw material consumption prediction model; the consumption of chemical additives includes the consumption of propylene glycol and the consumption of hydroxylamine;

[0229] Step b5: Construct the third energy consumption prediction model based on the long short-term memory network; the specific construction method of the third energy consumption prediction model is: using the temperature data, pressure data, flow data, and equipment operation time as the input features of the training set, and using the power consumption, water consumption, and steam consumption as the output features to train and obtain the third energy consumption prediction model;

[0230] Step c5: Construct a third equipment maintenance prediction model based on the long short-term memory network; the specific construction method of the third equipment maintenance prediction model is as follows: Use the equipment operation time, the number of failures, the equipment failure rate, the equipment vibration data, and the equipment maintenance history as the input features of the training set, and use the equipment failure rate and the number of maintenance times as the output features to train and obtain the third equipment maintenance prediction model.

[0231] During the training process, the LSTM inputs the training data into the model for training. The LSTM model structure includes an LSTM layer, a fully connected layer, an optimizer, and a loss function. Among them, the LSTM layer contains 64 LSTM units and uses the Tanh activation function to enhance the non-linear expression ability; the fully connected layer includes two layers of neural networks, consisting of 128 neurons and 64 neurons respectively, and uses the ReLU activation function for non-linear transformation to improve the model fitting ability.

[0232] During the optimization process, the LSTM uses the Adam optimizer, and its parameter update method is as follows:

[0233] ;

[0234] ;

[0235] where, m t represents the exponential moving average of the first moment estimate in the Adam optimizer, g t is the gradient, and v t represents the unbiased estimate of the second moment; and are momentum parameters, enabling the model to have a faster convergence speed in the high-dimensional optimization space. The loss function uses the mean squared error MSE, and the calculation formula is as follows:

[0236] ;

[0237] where, is the true value, is the predicted value, N is the number of samples, and the mean squared error is used to measure the error between the predicted value and the true value;

[0238] During the training process, each batch contains 32 samples (Batch Size = 32); the forward propagation is used to calculate the output. The LSTM processes the input data step by step through time steps, and finally the fully connected layer calculates the predicted value:

[0239] ;

[0240] where, W 1 and W2 is the weight matrix of the fully connected layer, b1 and b2 are the bias terms, and h is the hidden state vector of the LSTM layer; then calculate the loss function, calculate the gradient through backpropagation, and use Adam for optimization. The maximum number of training iterations is set to 200 to ensure training convergence;

[0241] The training construction process based on the long short-term memory network is a mature technology existing in the art. The innovation point of this part lies in the matching of input data features and output features, so as to obtain a predictable model through training and finally obtain the predicted value.

[0242] Example 5

[0243] Based on Example 4, in step 5, set the initial weight distribution of the total model for raw material consumption prediction, the total model for energy consumption prediction, and the total model for equipment maintenance prediction, and calculate the mean square error of the three models inside each total model. Then introduce Gaussian process regression to calculate the uncertainty of the three models inside each total model. Finally, calculate the fusion weights of the three models inside each total model respectively according to variational Bayesian inference. Specifically: Assume that the weights of the three models inside each total model follow a Beta distribution:

[0244] ;

[0245] In the formula, is the weight coefficient of the three models inside the total model for raw material consumption prediction, the weight coefficient of the three models inside the total model for energy consumption prediction, or the weight coefficient of the three models inside the total model for equipment maintenance prediction, and are set as initial parameters, generally taking:

[0246] ;

[0247] In the formula, is a positive number to ensure uniform weight distribution in the initial state;

[0248] For the prediction errors of the three models inside the total model for raw material consumption prediction, the prediction errors of the three models inside the total model for energy consumption prediction, or the prediction errors of the three models inside the total model for equipment maintenance prediction, evaluate by calculating the mean square error of the three models inside the total model for raw material consumption prediction, the mean square error of the three models inside the total model for energy consumption prediction, or the mean square error of the three models inside the total model for equipment maintenance prediction. The specific calculation formula is as follows:

[0249] ;

[0250] ;

[0251] ;

[0252] In the formula, represents the mean square error of the first raw material consumption prediction model, the first energy consumption prediction model, or the first equipment maintenance prediction model based on the random forest; represents the mean square error of the second raw material consumption prediction model, the second energy consumption prediction model, or the second equipment maintenance prediction model based on the support vector machine; represents the mean square error of the third raw material consumption prediction model, the third energy consumption prediction model, or the third equipment maintenance prediction model based on the long short-term memory network; N represents the total number of training samples; represents the true value of the i-th sample; represents the predicted value of the i-th sample by the first raw material consumption prediction model, the first energy consumption prediction model, or the first equipment maintenance prediction model based on the random forest; represents the predicted value of the i-th sample by the second raw material consumption prediction model, the second energy consumption prediction model, or the second equipment maintenance prediction model based on the support vector machine; represents the predicted value of the i-th sample by the third raw material consumption prediction model, the third energy consumption prediction model, or the third equipment maintenance prediction model based on the long short-term memory network;

[0253] Then, update the Beta distribution parameters according to the mean square error:

[0254] ;

[0255] ;

[0256] In the formula, represents a distribution parameter in the calculation of the fusion weight of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction, represents another distribution parameter in the calculation of the fusion weight of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction; represents the mean square error of the total model for raw material consumption prediction, the mean square error of the total model for energy consumption prediction, or the mean square error of the total model for equipment maintenance prediction;

[0257] Introduce Gaussian process regression to calculate the uncertainty of each model. The specific formula is as follows:

[0258] ;

[0259] In the formula, It represents the uncertainty of the model and is used to measure the degree of uncertainty of the total models for raw material consumption prediction, energy consumption prediction, and equipment maintenance prediction across all samples; N is the total number of training samples; It is expressed as the prediction variance of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction across all samples. This variance is calculated based on Gaussian process regression and represents the degree of uncertainty in the model output under the same input conditions. If the variance of the model prediction values is large across different training sets, it indicates that the confidence level of the model in this input region is low. The specific solution formula is as follows:

[0260] ;

[0261] In the formula, represents the predicted value of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction for the i-th sample; represents the average predicted value of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction across all samples;

[0262] Then, the uncertainty is introduced into the Beta distribution for update:

[0263] ;

[0264] ;

[0265] In the formula, represents a distribution parameter in the calculation of the fusion weights of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction after update; represents another distribution parameter in the calculation of the fusion weights of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction after update; these distribution parameters are updated in each iteration to reflect the uncertainty of the model;

[0266] Finally, variational Bayesian inference is used to calculate the final weights:

[0267] ;

[0268] In the formula, is the model weight after the (t + 1)-th iteration of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction; represents the contribution of the actual data;

[0269] Then, the fusion weights of the three models within each model are obtained respectively:

[0270] ;

[0271] ;

[0272] ;

[0273] In the formula, means the weight coefficient of the first raw material consumption prediction model, the first energy consumption prediction model or the first equipment maintenance prediction model; means the weight coefficient of the second raw material consumption prediction model, the second energy consumption prediction model or the second equipment maintenance prediction model; means the weight coefficient of the third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model; means a distribution parameter in the calculation of the fusion weight of the updated first raw material consumption prediction model, the first energy consumption prediction model or the first equipment maintenance prediction model; means a distribution parameter in the calculation of the fusion weight of the updated second raw material consumption prediction model, the second energy consumption prediction model or the second equipment maintenance prediction model; means a distribution parameter in the calculation of the fusion weight of the updated third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model; means another distribution parameter in the calculation of the fusion weight of the updated first raw material consumption prediction model, the first energy consumption prediction model or the first equipment maintenance prediction model; means another distribution parameter in the calculation of the fusion weight of the updated second raw material consumption prediction model, the second energy consumption prediction model or the second equipment maintenance prediction model; means another distribution parameter in the calculation of the fusion weight of the updated third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model; means the sum of the predicted output values of all training samples of the first raw material consumption prediction model, the first energy consumption prediction model or the first equipment maintenance prediction model based on the random forest; means the sum of the predicted output values of all training samples of the second raw material consumption prediction model, the second energy consumption prediction model or the second equipment maintenance prediction model based on the support vector machine; means the sum of the predicted output values of all training samples of the third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model based on the long short-term memory network.

[0274] Process parameter group 1 includes wood pulp viscosity, wood pulp degree of polymerization, wood pulp methyl cellulose content, NMMO solution concentration, propylene glycol content, hydroxylamine content, inventory data of raw materials, and consumption data of raw materials; the physical properties and chemical composition of wood pulp determine its dissolution performance during the production process, the NMMO solution concentration and chemical additive content affect the stability of the solvent system, and the inventory data of raw materials reflects the impact of the supply chain on consumption. By comprehensively considering these factors, the model can accurately predict the consumption of raw materials during the production process, providing decision-making support for raw material procurement and inventory management; the predicted values of raw material consumption include wood pulp consumption, NMMO solution consumption, and chemical additive consumption; the chemical additive consumption includes propylene glycol consumption and hydroxylamine consumption; Process parameter group 2 includes temperature data, pressure data, flow data, and equipment operation time; the predicted values of energy consumption include electricity consumption, water consumption, and steam consumption; temperature data and pressure data reflect the load status of the energy system, flow data is directly related to energy consumption, and equipment operation time determines the total energy consumption of the entire production cycle; based on these input features, the model can effectively predict electricity consumption, water consumption, and steam consumption, thereby optimizing energy management and improving production efficiency; Process parameter group 3 includes equipment operation time, number of failures, equipment failure rate, equipment vibration data, and equipment maintenance history; the predicted values of equipment maintenance include equipment failure rate and number of equipment maintenance; equipment operation time and number of failures directly reflect the usage status of the equipment, the maintenance history provides past maintenance conditions and repair frequencies, and equipment vibration data is an important indicator of the equipment's health status, which can early warn of possible mechanical failures. Through these input features, the model can accurately predict the equipment failure rate and number of equipment maintenance, thereby optimizing the equipment maintenance plan, reducing unexpected shutdowns, and improving production stability.

[0275] The specific formulas for calculating the final predicted values of raw material consumption, final predicted values of energy consumption, and final predicted values of equipment maintenance after final weighted fusion are as follows:

[0276] ;

[0277] In the formula, represents the final predicted value of raw material consumption, the final predicted value of energy consumption, or the final predicted value of equipment maintenance; when represents the predicted value of raw material consumption of the first raw material consumption prediction model based on random forest, represents the predicted value of raw material consumption of the second raw material consumption prediction model based on support vector machine, represents the predicted value of raw material consumption of the third raw material consumption prediction model based on long short-term memory network, then represents the weight coefficient of the first raw material consumption prediction model, represent the weight coefficients of the second raw material consumption prediction model and represent the weight coefficients of the third raw material consumption prediction model; when represent the energy consumption prediction value of the first energy consumption prediction model based on the random forest, represent the energy consumption prediction value of the second energy consumption prediction model based on the support vector machine, represent the energy consumption prediction value of the third energy consumption prediction model based on the long short-term memory network, then represent the weight coefficients of the first energy consumption prediction model, represent the weight coefficients of the second energy consumption prediction model and represent the weight coefficients of the third energy consumption prediction model; when represent the equipment maintenance prediction value of the first equipment maintenance prediction model based on the random forest, represent the equipment maintenance prediction value of the second equipment maintenance prediction model based on the support vector machine, represent the equipment maintenance prediction value of the third equipment maintenance prediction model based on the long short-term memory network, then represent the weight coefficients of the first equipment maintenance prediction model, represent the weight coefficients of the second equipment maintenance prediction model and represent the weight coefficients of the third equipment maintenance prediction model.

[0278] Example 7

[0279] As Figure 1 shown, input the process parameter group 1 into the first raw material consumption prediction model constructed based on the random forest, the second raw material consumption prediction model constructed based on the support vector machine, and the third raw material consumption prediction model constructed based on the long short-term memory network, so as to obtain three raw material consumption prediction values, and the three raw material consumption prediction values are weighted and fused to obtain the final raw material consumption prediction value; input the process parameter group 2 into the first energy consumption prediction model constructed based on the random forest, the second energy consumption prediction model constructed based on the support vector machine, and the third energy consumption prediction model constructed based on the long short-term memory network, so as to obtain three energy consumption prediction values, and the three energy consumption prediction values are weighted and fused to obtain the final energy consumption prediction value; input the process parameter group 3 into the first equipment maintenance prediction model constructed based on the random forest, the second equipment maintenance prediction model constructed based on the support vector machine, and the third equipment maintenance prediction model constructed based on the long short-term memory network, so as to obtain three equipment maintenance prediction values, and the three equipment maintenance prediction values are weighted and fused to obtain the final equipment maintenance prediction value; the final equipment maintenance prediction value, the final energy consumption prediction value, and the final raw material consumption prediction value.

[0280] Example 8

[0281] As Figure 2 shown, the process of the method of the present invention is as follows: First, data collection and preprocessing are carried out. After data cleaning, data normalization, and principal component analysis plus clustering information processing, the data is used for model construction and training. The methods of model construction and training are based on random forest trees, support vector machines, and long short-term memory networks; predictions are made according to the constructed model, and the predicted values are weighted and fused to obtain the final predicted value.

[0282] Example 9

[0283] Based on Example 6, an application of a consumption prediction method for Lyocell fiber production and manufacturing is applicable to the consumption prediction of Lyocell fiber production and manufacturing.

[0284] The above description is a detailed description of the preferred feasible embodiments of the present application. However, the embodiments are not intended to limit the scope of the patent application of the present application. Any equivalent changes or modifications made under the technical spirit disclosed in the present application shall fall within the scope of the patent covered by the present application.

Claims

1. A method for predicting consumption of lyocell fiber production, characterized in that: The specific steps include: Step 1: Collect historical raw material data, energy consumption data and equipment maintenance data in the production process of lyocell fiber; raw material data include wood pulp parameter data, NMMO solution parameter data and chemical additive parameter data; energy consumption data include temperature data, pressure data and flow data; the flow data include NMMO solvent flow, water consumption and steam consumption; equipment maintenance data include equipment operating time, number of failures, equipment failure rate, equipment vibration data and equipment maintenance history; Step 2: Preprocess raw material data, energy consumption data, and equipment maintenance data; Step 3: Build and train the overall model for raw material consumption prediction, the overall model for energy consumption prediction, and the overall model for equipment maintenance prediction; The overall model for raw material consumption prediction, the overall model for energy consumption prediction and the overall model for equipment maintenance prediction in step 3 are constructed and trained based on random forest, support vector machine and long short-term memory network; Step 4: Input the process parameter group 1, process parameter group 2 and process parameter group 3 in the current lyocell fiber production process into the three total models in step 3 respectively, and obtain three raw material consumption prediction values, three energy consumption prediction values ​​and three equipment maintenance prediction values; In step 4, process parameter group 1 is raw material data in the current lyocell fiber production process, process parameter group 2 is energy consumption data in the current lyocell fiber production process and equipment operation time in equipment maintenance data, and process parameter group 3 is equipment maintenance data in the current lyocell fiber production process; Step 5: Set the initial weight distribution of the total model for raw material consumption prediction, the total model for energy consumption prediction, and the total model for equipment maintenance prediction, and calculate the mean square error of the three models within each total model. Then update the Beta distribution parameters of the total model based on the mean square error. At the same time, introduce Gaussian process regression to calculate the uncertainty of the three models within each total model, and then update the Beta distribution parameters again based on the uncertainty. Then, based on the updated Beta distribution parameters combined with variational Bayesian inference, calculate the fusion weights of the three models within each total model. Step 6: Based on the three types of prediction values ​​in step 4 and the fusion weights of the three models within the corresponding total model, the final prediction value after weighted fusion is calculated.

2. A method for predicting consumption of lyocell fiber production according to claim 1, characterized in that: The specific method of step 4 is: inputting process parameter group 1 into the total model output of raw material consumption prediction to obtain three raw material consumption prediction values, inputting process parameter group 2 into the total model output of energy consumption prediction to obtain three energy consumption prediction values, and inputting process parameter group 3 into the total model output of equipment maintenance prediction to obtain three equipment maintenance prediction values; The three types of prediction values ​​in step 6 include three raw material consumption prediction values, three energy consumption prediction values ​​and three equipment maintenance prediction values; the final prediction value includes the final raw material consumption prediction value, the final energy consumption prediction value and the final equipment maintenance prediction value; the specific method of step 6 is: the three raw material consumption prediction values ​​are obtained, according to the fusion weights of the three models within the total model of raw material consumption prediction, to calculate the final raw material consumption prediction value after weighted fusion; The three energy consumption prediction values ​​obtained are calculated according to the fusion weights of the three models within the total model of energy consumption prediction to obtain the final energy consumption prediction value after weighted fusion; The three equipment maintenance prediction values ​​obtained are calculated according to the fusion weights of the three models within the total model of equipment maintenance prediction to obtain the final equipment maintenance prediction value after weighted fusion.

3. A method for predicting consumption of lyocell fiber production according to claim 1, characterized in that: The wood pulp parameter data in step one include wood pulp viscosity, wood pulp polymerization degree and wood pulp methyl fiber content; the NMMO solution parameter data include NMMO solution concentration; the chemical additive parameter data include propylene glycol content and hydroxylamine content; the raw material data also include raw material inventory data and raw material consumption data; the temperature data include dissolution temperature, reactor temperature, insulation water system temperature, spinning wind system temperature, coagulation bath temperature and ambient temperature; the pressure data include solvent delivery pressure, vacuum pressure and ambient air pressure.

4. A method for predicting consumption of lyocell fiber production according to claim 3, characterized in that: The viscosity of wood pulp, degree of polymerization of wood pulp and methyl cellulose content of wood pulp are monitored and collected in real time through online viscometers and optical measuring instruments; the concentration of NMMO solution is collected by liquid phase concentration detection sensors; the content of propylene glycol and the content of hydroxylamine are detected and collected by chemical sensors; the inventory data of raw materials and the consumption data of raw materials are collected through the enterprise resource planning system and synchronized with the external market database.

5. A method for predicting consumption of lyocell fiber production according to claim 3, characterized in that: The dissolution temperature, reactor temperature, insulation water system temperature, spinning wind system temperature, coagulation bath temperature and ambient temperature are measured by PT100 platinum resistance temperature sensor or thermocouple; the solvent delivery pressure, vacuum pressure and ambient air pressure are collected by MEMS piezoresistive pressure sensor; NMMO solvent flow, water consumption and steam consumption are collected by electromagnetic flowmeter and ultrasonic flowmeter, and the accumulated flow value is calculated in combination with PLC; the reactor temperature, insulation water system temperature and spinning wind system temperature are collected by SCADA system.

6. A method for predicting consumption of lyocell fiber production according to claim 1, characterized in that: Equipment operating time, number of failures, equipment failure rate, equipment vibration data and equipment maintenance history are recorded and collected through the MES system.

7. A method for predicting consumption of lyocell fiber production according to claim 1, characterized in that: The specific steps for data preprocessing of raw material data, energy consumption data and equipment maintenance data are as follows: Step a: Perform outlier detection on time series data, numerical data, and non-normal distribution data in raw material data, energy consumption data, and equipment maintenance data, and remove outliers. Finally, perform missing value processing and duplicate data removal. Step b: normalizing raw material data, energy consumption data and equipment maintenance data; Step c: Use Pearson correlation analysis and principal component analysis to extract key features of raw material data, energy consumption data, and equipment maintenance data; Step d: Use K-means cluster analysis and information entropy screening methods to conduct final screening of raw material data, energy consumption data, and equipment maintenance data.

8. A method for predicting consumption of lyocell fiber production according to claim 7, characterized in that: The specific method of performing outlier detection on numerical data in step a is: using the 3σ principle to detect outliers, for each data point f i , calculate the mean μ and standard deviation σ of the original data, if it satisfies: ； It is judged as abnormal data and is removed; where, is the ith data point.

9. A method for predicting consumption of lyocell fiber production according to claim 7, characterized in that: The specific method of performing outlier detection on non-normal distribution data in step a is: using the interquartile range method to detect outliers, calculating the first quartile Q1 and the third quartile Q3, and setting the outlier range as follows: ； ； Data outside this range are considered outliers and are removed; where IQR means interquartile range.

10. A method for predicting consumption of lyocell fiber production according to claim 7, characterized in that: The specific method for handling missing values ​​is as follows: time series data is filled using linear interpolation; if the missing ratio of numerical data is less than 20%, k-nearest neighbor interpolation is used to fill it; if the missing ratio of non-normal distribution data is less than 20%, k-nearest neighbor interpolation is used to fill it; if the missing ratio of non-normal distribution data is more than 20%, the non-normal distribution data is excluded.

11. A method for predicting consumption of lyocell fiber production according to claim 10, characterized in that: The time series data include inventory data of raw materials, consumption data of raw materials, dissolution temperature, reactor temperature, insulation water system temperature, spinning wind system temperature, coagulation bath temperature, ambient temperature, solvent delivery pressure, vacuum pressure, insulation water system pressure, spinning wind system pressure, ambient air pressure, NMMO solvent flow, water consumption, steam consumption, equipment operating time and number of failures.

12. A method for predicting consumption of lyocell fiber production according to claim 10, characterized in that: The numerical data include wood pulp viscosity, wood pulp polymerization degree, wood pulp methyl fiber content, NMMO solution concentration, propylene glycol content, hydroxylamine content and equipment failure rate; the non-normal distribution data include raw material inventory data, raw material consumption data, number of failures, equipment failure rate, equipment vibration data and equipment maintenance history.

13. A method for predicting consumption of lyocell fiber production according to claim 7, characterized in that: Min-Max normalization is used to normalize raw material data, energy consumption data, and equipment maintenance data. The specific formula is as follows: ； In the formula, x is the original data value; x min is the minimum value of the original data; x max is the maximum value of the original data; For data with abnormal distribution, Z-score standardization is used: ； In the formula, μ is the mean of the original data; σ is the standard deviation, is the data value after Z-score standardization, x is the original data value.

14. A method for predicting consumption of lyocell fiber production according to claim 7, characterized in that: The specific method steps for extracting key features of raw material data, energy consumption data and equipment maintenance data using Pearson correlation analysis and principal component analysis in step c are as follows: Step a1: Calculate the Pearson correlation coefficient between features. The specific formula is as follows: ； When |r|>0.8, the two features are determined to be highly correlated and one of them is removed; where r is the Pearson correlation coefficient, h i is the i-th observation value of variable h, y i is the ith observation value of variable y, is the mean of variable h, μ y is the mean of variable y; Step b1: Perform principal component analysis to reduce the dimension. First, construct a standardized data matrix X with a dimension of m×n, where m is the number of samples and n is the number of features. Then calculate the covariance matrix , the specific calculation formula is as follows: ； Where, X T is the transposed matrix of the standardized data matrix X; Step c1: Solve the covariance matrix The eigenvalues ​​and eigenvectors of , and sort them from large to small according to the eigenvalues, and calculate the cumulative contribution rate; select the first k eigenvectors with cumulative contribution rates exceeding 95% to form a dimensionality reduction projection matrix W, and obtain projection data through the dimensionality reduction projection matrix W, and finally reduce the data dimension after principal component analysis from n dimensions to k dimensions; the specific formula for obtaining the projection data is as follows: ； Where W is the dimension reduction projection matrix, and X′ is the feature data after dimension reduction.

15. A method for predicting consumption of lyocell fiber production according to claim 7, characterized in that: The specific steps for final screening of raw material data, energy consumption data and equipment maintenance data using K-means cluster analysis and information entropy screening methods are as follows: Step a2: Use K-means cluster analysis, first initialize, set the number of clusters k=5, select k random samples as the initial cluster centers, and then calculate the Euclidean distance from the sample to the cluster center. The specific formula is as follows: ； In the formula, The meaning of is expressed as the Euclidean distance between the sample point s and the cluster center c; is the i-th data sample, is the new cluster center of the i-th cluster, and n is the n-dimensional space; Step b2: Reassign clusters, calculate the distance between each sample point and all cluster centers, and assign the sample point to the nearest cluster; Step c2: Update the cluster center. First, calculate the new cluster center and continue to iterate until convergence. When the cluster center change is less than the set threshold, stop the iteration. The specific calculation formula for calculating the new cluster center is: ； In the formula, is the new cluster center of the ith cluster, Cluster C i The number of internal sample points, s is the data sample point; Step d2: Filter out representative data, calculate the average distance from each cluster to the cluster center, and select 10% of the data around the cluster center in each cluster as training samples; Step e2: Through information entropy screening, further evaluate the information content of each cluster and calculate the information entropy of each cluster , the specific formula is as follows: ； In the formula, Cluster The probability of category j in ; is the number of categories within the cluster; Step f2: Filter the cluster with the highest information entropy, calculate the information entropy values ​​of all clusters, and select the top 50% clusters with the highest information entropy. Among the selected clusters, retain the samples around the center of the cluster and remove outliers to form the final training data set.

16. A method for predicting consumption of lyocell fiber production according to claim 2, characterized in that: The overall model for raw material consumption prediction constructed and trained based on random forest, support vector machine and long short-term memory network includes a first raw material consumption prediction model constructed based on random forest, a second raw material consumption prediction model constructed based on support vector machine and a third raw material consumption prediction model constructed based on long short-term memory network; the overall model for energy consumption prediction constructed and trained based on random forest, support vector machine and long short-term memory network includes a first energy consumption prediction model constructed based on random forest, a second energy consumption prediction model constructed based on support vector machine and a third energy consumption prediction model constructed based on long short-term memory network; the overall model for equipment maintenance prediction constructed and trained based on random forest, support vector machine and long short-term memory network includes a first equipment maintenance prediction model constructed based on random forest, a second equipment maintenance prediction model constructed based on support vector machine and a third equipment maintenance prediction model constructed based on long short-term memory network.

17. A method for predicting consumption of lyocell fiber production according to claim 16, characterized in that: The specific method steps for constructing the first raw material consumption prediction model, the first energy consumption prediction model and the first equipment maintenance prediction model based on random forest are as follows: Step a3: constructing a first raw material consumption prediction model based on random forest; the specific construction method of the first raw material consumption prediction model is: using wood pulp viscosity, wood pulp polymerization degree, NMMO solution concentration, wood pulp methyl fiber content, propylene glycol content, hydroxylamine content, raw material inventory data and raw material consumption data as input features of the training set, and using wood pulp consumption, NMMO solution consumption and chemical additive consumption as output features of the training set, training to obtain the first raw material consumption prediction model; the consumption of the chemical additives includes propylene glycol consumption and hydroxylamine consumption; Step b3: constructing a first energy consumption prediction model based on random forest; the specific construction method of the first energy consumption prediction model is: taking temperature data, pressure data, flow data and equipment operation time as input features of the training set, and taking power consumption, water consumption and steam consumption as output features, training to obtain the first energy consumption prediction model; Step c3: Construct a first equipment maintenance prediction model based on random forest; the specific construction method of the first equipment maintenance prediction model is: use the equipment operation time, number of failures, equipment failure rate, equipment vibration data and equipment maintenance history as input features of the training set, and use the equipment failure rate and maintenance number as output features, and train to obtain the first equipment maintenance prediction model.

18. A method for predicting consumption of lyocell fiber production according to claim 16, characterized in that: The specific method steps for constructing the second raw material consumption prediction model, the second energy consumption prediction model and the second equipment maintenance prediction model based on the support vector machine are as follows: Step a4: constructing a second raw material consumption prediction model based on a support vector machine; the specific construction method of the second raw material consumption prediction model is: using wood pulp viscosity, wood pulp polymerization degree, NMMO solution concentration, wood pulp methyl fiber content, propylene glycol content, hydroxylamine content, raw material inventory data and raw material consumption data as input features of the training set, and using wood pulp consumption, NMMO solution consumption and chemical additive consumption as output features of the training set, and training to obtain the second raw material consumption prediction model; the consumption of the chemical additives includes propylene glycol consumption and hydroxylamine consumption; Step b4: constructing a second energy consumption prediction model based on a support vector machine; the specific construction method of the second energy consumption prediction model is: taking temperature data, pressure data, flow data and equipment operation time as input features of a training set, and taking power consumption, water consumption and steam consumption as output features, and training to obtain the second energy consumption prediction model; Step c4: Construct a second equipment maintenance prediction model based on support vector machine; the specific construction method of the second equipment maintenance prediction model is: use the equipment operation time, number of failures, equipment failure rate, equipment vibration data and equipment maintenance history as input features of the training set, and use the equipment failure rate and maintenance number as output features, and train to obtain the second equipment maintenance prediction model.

19. A method for predicting consumption of lyocell fiber production according to claim 16, characterized in that: The specific method steps for constructing the third raw material consumption prediction model, the third energy consumption prediction model and the third equipment maintenance prediction model based on the long short-term memory network are as follows: Step a5: constructing a third raw material consumption prediction model based on the long short-term memory network; the specific construction method of the third raw material consumption prediction model is: using wood pulp viscosity, wood pulp polymerization degree, NMMO solution concentration, wood pulp methyl fiber content, propylene glycol content, hydroxylamine content, raw material inventory data and raw material consumption data as input features of the training set, and using wood pulp consumption, NMMO solution consumption and chemical additive consumption as output features of the training set, and training to obtain the third raw material consumption prediction model; the consumption of the chemical additives includes propylene glycol consumption and hydroxylamine consumption; Step b5: constructing a third energy consumption prediction model based on the long short-term memory network; the specific construction method of the third energy consumption prediction model is: taking temperature data, pressure data, flow data and equipment operation time as input features of the training set, and taking power consumption, water consumption and steam consumption as output features, training to obtain the third energy consumption prediction model; Step c5: construct a third equipment maintenance prediction model based on the long short-term memory network; the specific construction method of the third equipment maintenance prediction model is: use the equipment operating time, number of failures, equipment failure rate, equipment vibration data and equipment maintenance history as input features of the training set, and use the equipment failure rate and maintenance number as output features, and train to obtain the third equipment maintenance prediction model.

20. The method for predicting consumption of lyocell fiber production according to claim 1, characterized in that: In the step 5, the initial weight distribution of the total model for raw material consumption prediction, the total model for energy consumption prediction and the total model for equipment maintenance prediction is set, and the mean square error of the three models within each total model is calculated, and then Gaussian process regression is introduced to calculate the uncertainty of the three models within each total model. Finally, the fusion weights of the three models within each total model are obtained according to the variational Bayesian inference calculation. Specifically, it is assumed that the weights of the three models within each total model obey the Beta distribution: ； In the formula, The weight coefficients of the three models within the total model for raw material consumption prediction, the weight coefficients of the three models within the total model for energy consumption prediction, or the weight coefficients of the three models within the total model for equipment maintenance prediction, and Set as initial parameters, take: ； In the formula, is a positive number, ensuring that the weights are evenly distributed in the initial state; The prediction errors of the three models within the total model for raw material consumption prediction, the prediction errors of the three models within the total model for energy consumption prediction, or the prediction errors of the three models within the total model for equipment maintenance prediction are evaluated by calculating the mean square error of the three models within the total model for raw material consumption prediction, the mean square error of the three models within the total model for energy consumption prediction, or the mean square error of the three models within the total model for equipment maintenance prediction. The specific calculation formula is as follows: ； ； ； In the formula, represents a mean square error of the first raw material consumption prediction model, the first energy consumption prediction model, or the first equipment maintenance prediction model based on random forest; represents a mean square error of a second raw material consumption prediction model, a second energy consumption prediction model, or a second equipment maintenance prediction model based on a support vector machine; represents the mean square error of the third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model based on the long short-term memory network; N represents the total number of training samples; Represents the true value of the i-th sample; represents the predicted value of the first raw material consumption prediction model, the first energy consumption prediction model or the first equipment maintenance prediction model based on random forest for the i-th sample; represents the predicted value of the second raw material consumption prediction model, the second energy consumption prediction model or the second equipment maintenance prediction model based on the support vector machine for the i-th sample; represents the predicted value of the third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model based on the long short-term memory network for the i-th sample; Then update the Beta distribution parameters according to the mean square error: ； ； In the formula, It represents a distribution parameter in the calculation of the fusion weight of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction. represents another distribution parameter in the calculation of the fusion weight of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction; represents the mean square error of the total model for raw material consumption prediction, the mean square error of the total model for energy consumption prediction, or the mean square error of the total model for equipment maintenance prediction; Gaussian process regression is introduced to calculate the uncertainty of each model. The specific formula is as follows: ； In the formula, It means the uncertainty of the model; N is the total number of training samples; It is expressed as the prediction variance of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction on all samples. The specific solution formula is as follows: ； In the formula, represents the predicted value of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction for the i-th sample; It represents the average prediction value of the total model for raw material consumption prediction, the total model for energy consumption prediction or the total model for equipment maintenance prediction on all samples; Then introduce uncertainty into the Beta distribution update: ； ； In the formula, represents a distribution parameter in the calculation of the fusion weight of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction after update; Another distribution parameter representing an updated fusion weight calculation of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction; Finally, the final weights are calculated using variational Bayesian inference: ； In the formula, The model weight after the t+1th iteration of the total model for raw material consumption prediction, the total model for energy consumption prediction, or the total model for equipment maintenance prediction; Represents the contribution of actual data; Then, the fusion weights of the three models within each model are obtained: ； ； ； In the formula, The meaning of is the weight coefficient of the first raw material consumption prediction model, the first energy consumption prediction model or the first equipment maintenance prediction model; Meaning is the weight coefficient of the second raw material consumption prediction model, the second energy consumption prediction model or the second equipment maintenance prediction model; The meaning is the weight coefficient of the third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model; It means a distribution parameter in the fusion weight calculation of the updated first raw material consumption prediction model, the first energy consumption prediction model or the first equipment maintenance prediction model; It means a distribution parameter in the fusion weight calculation of the updated second raw material consumption prediction model, the second energy consumption prediction model or the second equipment maintenance prediction model; It means a distribution parameter in the fusion weight calculation of the updated third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model; Meaning is another distribution parameter in the fusion weight calculation of the updated first raw material consumption prediction model, the first energy consumption prediction model or the first equipment maintenance prediction model; Meaning is another distribution parameter in the fusion weight calculation of the updated second raw material consumption prediction model, the second energy consumption prediction model or the second equipment maintenance prediction model; Meaning is another distribution parameter in the calculation of the fusion weight of the updated third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model; It means the sum of the predicted output values ​​of all training samples of the first raw material consumption prediction model, the first energy consumption prediction model or the first equipment maintenance prediction model based on random forest; The meaning is the sum of the predicted output values ​​of all training samples of the second raw material consumption prediction model, the second energy consumption prediction model or the second equipment maintenance prediction model based on the support vector machine; It means the sum of the predicted output values ​​of all training samples of the third raw material consumption prediction model, the third energy consumption prediction model or the third equipment maintenance prediction model based on the long short-term memory network.

21. A method for predicting consumption of lyocell fiber production according to claim 2, characterized in that: The process parameter group 1 includes wood pulp viscosity, wood pulp polymerization degree, wood pulp methyl fiber content, NMMO solution concentration, propylene glycol content, hydroxylamine content, raw material inventory data and raw material consumption data; the raw material consumption forecast value includes wood pulp consumption, NMMO solution consumption and chemical additive consumption; the chemical additive consumption includes propylene glycol consumption and hydroxylamine consumption; the process parameter group 2 includes temperature data, pressure data, flow data and equipment operation time; the energy consumption forecast value includes electricity consumption, water consumption and steam consumption; the process parameter group 3 includes equipment operation time, number of failures, equipment failure rate, equipment vibration data and equipment maintenance history; the equipment maintenance forecast value includes equipment failure rate and equipment maintenance number.

22. A method for predicting consumption of lyocell fiber production according to claim 2, characterized in that: The specific formulas for calculating the final weighted fusion forecast value of raw material consumption, the final energy consumption forecast value, and the final equipment maintenance forecast value are as follows: ； In the formula, Indicates the final raw material consumption forecast value, the final energy consumption forecast value or the final equipment maintenance forecast value; when represents the raw material consumption prediction value of the first raw material consumption prediction model based on random forest, represents the raw material consumption prediction value of the second raw material consumption prediction model based on support vector machine, When represents the raw material consumption forecast value of the third raw material consumption forecast model based on the long short-term memory network, then represents the weight coefficient of the first raw material consumption prediction model, represents the weight coefficient and represents the weight coefficient of the third raw material consumption forecast model; when represents the energy consumption prediction value of the first energy consumption prediction model based on random forest, represents the energy consumption prediction value of the second energy consumption prediction model based on the support vector machine, When represents the energy consumption prediction value of the third energy consumption prediction model based on the long short-term memory network, represents the weight coefficient of the first energy consumption prediction model, represents the weight coefficient of the second energy consumption prediction model and represents the weight coefficient of the third energy consumption prediction model; when represents the equipment maintenance prediction value of the first equipment maintenance prediction model based on random forest, represents the equipment maintenance prediction value of the second equipment maintenance prediction model based on support vector machine, When represents the equipment maintenance prediction value of the third equipment maintenance prediction model based on the long short-term memory network, then represents the weight coefficient of the first equipment maintenance prediction model, represents the weight coefficient and Represents the weight coefficient of the third equipment maintenance prediction model.

Images