Intelligent simulation system for Lyocell fiber production line based on digital twinning

Abstract

The invention relates to the technical field of digital twinning simulation, and discloses a digital twinning-based intelligent simulation system for a lyocell fiber production line, and the system comprises a 3D modeling and digital twinning module which constructs a three-dimensional model of the lyocell fiber production line, associates the three-dimensional model with real-time production data, and finally forms a three-dimensional digital twinning model of the production line; the key equipment state prediction module predicts the residual life of the film evaporator according to the vibration data of the film evaporator; the energy consumption prediction module is used for obtaining key parameters of the glue making process and predicting an energy consumption value according to the process list of the lyocell fiber glue making process; the early warning module is used for early warning in an equipment model corresponding to the target film evaporator in the three-dimensional digital twin model according to the residual life prediction result; and the process management module is responsible for compiling, managing and issuing process files. According to the system disclosed by the invention, all the functional modules are fused and combined together to form a complete system, so that the system is of great significance in guiding lyocell production and can help enterprises to carry out better cost control.

Description

Technical Field

[0001] The present invention relates to the technical field of digital twin simulation, and more specifically to an intelligent simulation system for a lyocell fiber production line based on digital twin. Background Art

[0002] As an important component of new fiber materials, Lyocell fiber is derived from nature, the product is naturally degradable, and the production process is green and environmentally friendly. It has broad application prospects and belongs to the national strategic emerging industry. In terms of economic construction, promoting the production of Lyocell fiber will help optimize resource allocation, improve the overall benefits of the industrial chain, ease the pressure on resources and the environment, promote employment and local economic development, and meet the multiple needs of social development.

[0003] The entire lyocell production system is complex, and the production process belongs to the process industry, with many production processes and long processes. Among them, the glue making process is the key link in the production process of lyocell fiber, which directly affects the quality and performance of the fiber, and the steam and electricity energy consumption account for a large proportion of the cost of the lyocell glue making process. Accurate prediction of them will help enterprises optimize production processes, reduce costs and improve market competitiveness. At present, there are few studies on energy consumption prediction in the production process of lyocell fiber, and no perfect technical method has been formed for the direction of energy consumption prediction. For the current energy consumption prediction direction, it is mostly concentrated in the fields of construction and equipment, and the prediction object is mainly electric energy. Moreover, these energy consumption prediction methods often regard energy consumption data as time series data, combined with environmental parameters, and through time prediction algorithms, real-time dynamic prediction is performed, but lack of attention to the static prediction of energy consumption under a given process order. For the production of lyocell fiber, the lack of energy consumption prediction obtained under a given process order will cause the power supply and gas supply related departments of the enterprise to be unable to provide future electricity and gas consumption plan reports, and it will be difficult to cope with sudden power outages and gas outages during peak electricity and gas consumption in the future, which will require more energy consumption. Objectively, it affects the rational allocation of corporate resources and long-term energy efficiency planning.

[0004] Furthermore, the status of key equipment on the Lyocell production line, including its remaining life and health, has a significant impact on the entire production. The maintenance and overhaul of key equipment on the production line requires the production line to be shut down, which leads to large fluctuations in production costs, increases the complexity and cost of production management, and affects production planning and resource allocation; and the status of key equipment will also affect product quality.

[0005] Therefore, it is necessary to simulate the key equipment and key processes of Lyocell production and establish a series of models to predict, analyze and monitor the various production factors in the Lyocell fiber production process. Among them, the most important difficulty lies in prediction, including energy consumption prediction, key equipment status prediction, etc. The common point of these predictions is cost control, which is of great significance for guiding production, and can significantly improve the economic benefits of Lyocell production. Summary of the invention

[0006] The present invention proposes a lyocell fiber production line intelligent simulation system based on digital twins. The present invention predicts, analyzes and monitors various production factors under a given parameter combination, including energy consumption prediction, key equipment status prediction, etc., to provide a basis and support for the company's production guidance; and can help the company to better control costs and significantly improve the economic benefits of lyocell production.

[0007] In order to achieve the above-mentioned object of the invention, the technical solution of the present invention is as follows: A lyocell fiber production line intelligent simulation system based on digital twin, the system includes: 3D modeling and digital twin module, key equipment status prediction module, energy consumption prediction module, early warning module and process management module; wherein, 3D modeling and digital twin module, which is used to build a 3D model of the Lyocell fiber production line, associate the real-time production data of the production line with the 3D model, and generate a 3D digital twin model of the Lyocell fiber production line based on digital twin technology; A key equipment status prediction module includes a remaining life prediction submodule for a thin film evaporator, wherein the remaining life prediction submodule for the thin film evaporator obtains real-time vibration data collected by a vibration sensor on a target thin film evaporator and real-time spindle speed transmitted by a central control platform of the target thin film evaporator; the real-time vibration data is segmented according to a set time interval to obtain a plurality of data segments at a set time interval, and then each data segment is analyzed in the time domain and frequency domain to obtain the root mean square, kurtosis and frequency center of gravity of each data segment; the spindle fault characteristic frequency is obtained according to the obtained real-time spindle speed, and then the spindle bearing fault characteristic frequency is obtained according to the real-time spindle speed and the size parameters of the spindle bearing; finally, the obtained root mean square, kurtosis, frequency center of gravity, spindle fault characteristic frequency and spindle bearing fault characteristic frequency are input into a remaining life prediction model, and finally the remaining life prediction result of the target thin film evaporator is output to the early warning module; The energy consumption prediction module, according to the process sheet of the glue making process provided by the process management module, extracts the porridge weight, process parameter group 1, process parameter group 2 and process parameter group 3 corresponding to the current process sheet; then obtains the glue liquid weight based on the porridge weight and process parameter group 1, obtains the glue making consumption time based on the porridge weight and process parameter group 3, and obtains the glue liquid refractive index based on process parameter group 2; finally, the porridge weight, glue liquid weight, glue liquid refractive index and glue making consumption time are input into the trained energy consumption prediction model, and the energy consumption prediction result corresponding to the glue making process is output; wherein, the process parameter group 1 includes the motor speed of the thin film evaporator, the temperature of the four zones of the thin film evaporator, the vacuum system pressure and the vacuum pump speed; the process parameter group 2 includes the motor speed of the thin film evaporator, the temperature of the four zones of the thin film evaporator, the vacuum pump speed and the outlet temperature of the insulation water circulation pump; the process parameter group 3 includes the motor speed of the thin film evaporator and the vacuum pump speed; The early warning module is used to issue an early warning in a device model corresponding to the target thin film evaporator in the three-dimensional digital twin model according to the remaining life prediction result of the target thin film evaporator; The process management module is used for the preparation, management and issuance of process documents.

[0008] Preferably, the real-time vibration data includes acceleration signals in three directions respectively collected by vibration sensors at the upper and lower ends and the middle area of ​​the target thin film evaporator cylinder; wherein the vibration sensor at the upper end of the cylinder is used to monitor the fault condition of the upper end bearing of the main shaft, the vibration sensor at the lower end of the cylinder is used to monitor the fault condition of the lower end bearing of the main shaft, and the vibration sensor in the middle area of ​​the cylinder is used to monitor the fault condition of the main shaft.

[0009] Preferably, the construction process of the remaining life prediction model of the thin film evaporator is as follows: Obtain historical vibration data and historical spindle speed data of the entire process from the normal operation of the thin film evaporator to the failure of the spindle and / or spindle bearing resulting in the failure of the thin film evaporator to operate normally; The historical vibration data includes acceleration signals in three directions respectively collected by vibration sensors at the upper and lower ends and the middle area of ​​the thin film evaporator cylinder; wherein the vibration sensor at the upper end of the cylinder is used to monitor the fault condition of the upper end bearing of the main shaft, the vibration sensor at the lower end of the cylinder is used to monitor the fault condition of the lower end bearing of the main shaft, and the vibration sensor in the middle area of ​​the cylinder is used to monitor the fault condition of the main shaft; The acceleration signal is segmented at set time intervals to obtain multiple data segments, and then each data segment is analyzed in the time domain and the frequency domain to extract the root mean square, kurtosis and frequency center of gravity of each data segment; the fault characteristic frequency of the spindle is obtained according to the historical spindle speed; the fault characteristic frequency of the spindle bearing is obtained according to the historical spindle speed and the spindle bearing size parameters; The acceleration signals of the three directions of each vibration sensor are vector-added to obtain multiple vibration vector data, and a one-dimensional convolutional neural network model including convolution, pooling, activation function and full connection layer is built for each vibration vector data; the response trend of the vibration response of each vibration vector data changing with time is extracted through the one-dimensional convolutional neural network model of each vibration vector data, and the vibration vector data characteristic curves of the upper and lower end bearings and the main shaft are obtained; Multiple high-order exponential functions are used to fit the vibration vector data characteristic curves of the upper and lower end bearings and the main shaft, respectively, to form the remaining life characteristic curves of the upper end bearing of the main shaft, the remaining life characteristic curves of the lower end bearing of the main shaft and the remaining life characteristic curves of the main shaft, and then expressed in percentage form; Taking the set time interval as the time resolution, the data set of the remaining life prediction model is formed according to the root mean square, kurtosis, frequency centroid, fault characteristic frequency of the spindle bearing, fault characteristic frequency of the spindle, and the data on the remaining life characteristic curve of the upper end bearing of the spindle, the remaining life characteristic curve of the lower end bearing of the spindle and the remaining life characteristic curve of the spindle; wherein the data on the remaining life characteristic curve is the label data; The obtained data set is divided into a training set, a validation set and a test set; the constructed remaining life prediction deep learning model is trained, validated and tested using the divided data sets, and finally a trained remaining life prediction model is obtained; the remaining life prediction model is a one-dimensional convolutional neural network model.

[0010] Preferably, the main shaft bearing fault characteristic frequencies include: Inner race fault characteristic frequency: ; Outer race fault characteristic frequency: ; Rolling element failure characteristic frequency: ; Cage failure characteristic frequency: ; Where: Q is the number of rolling elements, d The diameter of the rolling element, D Bearing pitch diameter, β is the contact angle, f r The rotation frequency of the main axis.

[0011] Preferably, the main shaft fault characteristic frequency is: ; Where: n 1 is the multiple number,f r is the rotation frequency of the main axis; is the main axis disturbance frequency.

[0012] As a preferred embodiment, the calculation formula expressed in percentage form after forming the remaining life characteristic curve of the upper end bearing of the main shaft, the remaining life characteristic curve of the lower end bearing of the main shaft and the remaining life characteristic curve of the main shaft is as follows: ; Where: , , , , is an adjustable parameter; is the horizontal coordinate of the vibration data, i.e., the system time of the vibration data; The time limit for the equipment to go from fully normal to abnormal is defined; is the remaining life percentage; e Is the natural base.

[0013] Preferably, the energy consumption prediction model is constructed as follows: Collect the historical raw data of evaporation system parameter data, glue and sealing system parameter data, insulation water and vacuum system parameter data, slurry weight, glue weight, glue refractive index, glue consumption time and energy consumption data during the Lyocell fiber glue making process, and perform data preprocessing to obtain historical data; Calculate the process similarity of the corresponding historical data according to the process sheet of the Lyocell fiber glue making process, and rearrange the historical data according to the process similarity to obtain the historical sorted data; Select a reference group sequence from the historical sorting data, calculate the process similarity of each reference group sequence, obtain the historical sorting data sequence, and find the threshold interval of the process similarity, and then use the historical sorting data sequence within the threshold interval to calculate the energy consumption prediction value sequence; Compare the energy consumption forecast value sequence with the historical energy consumption data of the corresponding reference group sequence, and record their absolute errors and X; Increase the threshold interval percentage P of process similarity to obtain the threshold interval of the current process similarity until the threshold interval of the current process similarity reaches the set maximum value, and obtain the threshold interval percentage P of process similarity corresponding to the minimum absolute error and X, and mark P as P s ; Build an energy consumption prediction model and use P sThe historical screening and sorting data are obtained by screening, and the porridge weight, glue liquid weight, glue liquid refractive index and glue making time in the historical screening and sorting data are used as the data set of the energy consumption prediction model. After the energy consumption prediction model is trained using the data set, a trained energy consumption prediction model is finally obtained.

[0014] Preferably, the historical original data includes historical original process parameter data, historical original energy consumption data and historical original intermediate parameter data; the historical data includes historical process parameter data, historical energy consumption data and historical intermediate parameter data; the historical sorted data includes historical sorted process parameter data, historical sorted intermediate parameter data and historical sorted energy consumption data; the historical filtered and sorted data includes historical filtered and sorted process parameter data, historical filtered and sorted intermediate parameter data and historical filtered and sorted energy consumption data.

[0015] Preferably, the process sheet includes process parameter data; the process parameter data include evaporation system parameter data, glue liquid and sealing system parameter data, insulation water and vacuum system parameter data and porridge weight; the evaporation system parameter data include evaporation heating system parameter data, thin film evaporator parameter data and evaporation condensation water system parameter data; the glue liquid and sealing system parameter data include sealing liquid system parameter data and glue liquid transport parameter data; the insulation water and vacuum system parameter data include insulation water system parameter data and vacuum system parameter data.

[0016] Preferably, the evaporation heating system parameter data includes the heating system temperature, the heating system liquid level and the heating system pressure; the thin film evaporator parameter data includes the four-zone temperature of the thin film evaporator, the thin film evaporator motor speed, the thin film evaporator bearing lubricating oil circulation temperature, the reducer bearing speed, the cooling fan speed, the thin film evaporator liquid level, the thin film evaporator outlet pressure and the thin film evaporator bottom outlet temperature; the four-zone temperature of the thin film evaporator includes the jacket temperature of the four evaporation zones and the glue temperature of the four evaporation zones; the glue delivery parameter data includes the temperature of the insulation water heat exchanger in the pump, the flow rate of the insulation water heat exchanger in the pump, the insulation water heat exchanger outside the pump The data include the temperature of the water heat exchanger, the flow rate of the insulation water heat exchanger outside the pump, the speed of the glue delivery pump, the outlet pressure of the glue delivery pump and the outlet temperature of the glue delivery pump; the parameter data of the sealing liquid system include the temperature of the sealing liquid tank, the sealing liquid flow rate and the liquid level of the sealing liquid tank; the parameter data of the insulation water system include the outlet temperature of the insulation water circulation pump and the hot water pressure of the insulation system; the parameter data of the evaporative condensing water system include the temperature of the evaporative condenser, the liquid level of the evaporative condenser, the pressure of the evaporative condenser, the flow rate of cooling water to the evaporative condenser and the outlet flow rate of the evaporative condensing liquid pump; the parameter data of the vacuum system include the liquid level of the vacuum system, the pressure of the vacuum system and the speed of the vacuum pump.

[0017] Preferably, the intermediate parameter data include glue liquid weight, glue liquid refractive index and glue making time.

[0018] As a preferred embodiment, data preprocessing is specifically as follows: based on 3 σ In principle, outliers are removed; after outliers are removed, the missing values ​​are filled by taking the weighted average of the K neighboring data values ​​near the missing value.

[0019] As a preferred method, the specific method of selecting the reference group from the historical sorting data is: Z The row data constitutes the reference group sequence ; By reference group sequence For each sequence in the reference group, we obtain the historical sorting data corresponding to each sequence in the reference group. , thus forming the historical sorting data sequence of the reference group sequence , and record the maximum process similarity value in the process ; Among them, the threshold interval of process similarity is , is the minimum process similarity value.

[0020] Preferably, the energy consumption prediction value sequence is calculated using the historical sorted data sequence within the threshold interval of process similarity, and the specific method is as follows: The porridge weight of the historically sorted intermediate parameter data and the historically sorted process parameter data in the historically sorted data sequence of the reference group sequence is used as the training set of the neural network model to obtain the energy consumption prediction value sequence of the reference group sequence; the process similarity corresponding to the historically sorted intermediate parameter data belongs to the threshold interval range of the process similarity.

[0021] Preferably, the energy consumption prediction value sequence of the reference group sequence is compared with the historical energy consumption data of the corresponding reference group sequence, and their absolute errors and X are recorded. The specific method is: Compare the energy consumption prediction value sequence of the reference group sequence with the historical energy consumption data of the corresponding reference group sequence, record their absolute error values, and traverse the historical sorting data sequence. Z Sorting historical data to obtain the absolute error value sequence of energy consumption prediction , and then accumulate the energy consumption prediction absolute error value sequence to obtain the absolute error and X and record it.

[0022] Preferably, the porridge weight, glue weight, glue refractive index and consumption time in the historical screening and sorting data are used as the data set of the energy consumption prediction model. After the energy consumption prediction model is trained using the data set, a trained energy consumption prediction model is finally obtained, including: The glue liquid weight, glue liquid refractive index, consumption time and porridge weight in the historical screening and sorting of intermediate parameter data are used as input features of the training set of the energy consumption prediction model, and the energy consumption data of historical screening and sorting are used as the target output of the energy consumption prediction model. The model is trained using the above data to obtain a trained energy consumption prediction model.

[0023] Preferably, the energy consumption data is the sum of electricity energy consumption and steam energy consumption.

[0024] Preferably, the key equipment status prediction module also includes a health prediction submodule of a thin film evaporator; the health prediction submodule of the thin film evaporator obtains the current production parameters of the target thin film evaporator and the current production parameters of the upstream pulper, and then preprocesses the collected data, including the step of eliminating outliers and normalization processing; then based on the expert scoring method, the collected parameters are weighted from three dimensions: the degree of parameter fluctuation, the degree of influence of the parameters on the health status of the equipment, and the severity of the deviation of the parameters from the normal range, and the sum of the products of each dimension of each parameter and the corresponding weight is calculated to obtain the influence weight of each parameter in the health prediction; finally, the weighted data of each parameter is input into the trained health prediction model to predict the health of the target thin film evaporator, and output the health index of the target thin film evaporator; wherein, the real-time production parameters of the target thin film evaporator include evaporation chamber pressure, stirring motor current, feed flow rate, feed pipeline pressure, dissolving liquid temperature, solution viscosity and solvent recovery rate; the real-time production parameters of the upstream pulper include discharge flow rate, slurry concentration and slurry temperature.

[0025] Preferably, when collecting the real-time production parameters of the thin film evaporator and the real-time production parameters of the pulper, time synchronization is performed on each parameter to ensure timing alignment of different data sources; wherein the parameters of the pulper are compensated in advance according to the material conveying time to match the operating status of the thin film evaporator.

[0026] Preferably, the step of eliminating outliers is performed by combining physical constraint method and statistical method. The physical constraint method refers to setting parameter thresholds based on the equipment operating range, and exceeding the range is regarded as an abnormality; the statistical method refers to using the 3σ principle to eliminate abnormal points based on the distribution of historical data.

[0027] Preferably, the parameter fluctuation degree is measured by normalized standard deviation, which is calculated as follows: ; In the formula, Indicates i Normalized standard deviation of production parameters; For the i The standard deviation of the production parameters, is the sum of the standard deviations of all production parameters.

[0028] Preferably, the influence of production parameters on equipment health status is measured by normalized expert score, which is calculated as follows: ; In the formula, Indicates i Normalized expert scores for production parameters; For the i Expert ratings of production parameters, For the total score.

[0029] Preferably, the severity of the deviation of the production parameters from the normal range is measured by the physical constraint weight, which is calculated as follows: ; In the formula, Indicates i The severity of the deviation of a production parameter from the normal range; For production parameters i The number of times beyond the normal range, is the total number of samples of production parameters.

[0030] Preferably, the health index is calculated as follows: ; In the formula, is the health index of the thin film evaporator; is the predicted failure probability of the thin film evaporator.

[0031] Preferably, the health prediction model is a long short-term memory artificial neural network, which includes an input layer, an LSTM layer, a fully connected layer and an output layer. The input layer is used to accept time series feature data, and its shape is determined by the step size and the total number of features at each step size; the LSTM layer is a single layer with 64 hidden units, which is used to capture long-term dependencies in the time series; the fully connected layer is used to generate prediction results; and the output layer is used to output the failure probability.

[0032] Preferably, the system also includes a cost prediction module, which includes a raw material cost prediction sub-module. The raw material cost prediction sub-module obtains and pre-processes the raw material data of the lyocell production process, inputs the pre-processed raw material data into the raw material consumption prediction overall model, and outputs the raw material consumption prediction values ​​corresponding to different machine learning algorithm models, and then obtains the final raw material consumption prediction value by weighted fusion of the prediction values ​​of each machine learning algorithm model; finally, the raw material prediction cost is obtained in combination with the raw material price; wherein the raw material data includes wood pulp parameter data, NMMO solution parameter data, and chemical additive parameter data; the wood pulp parameter data includes wood pulp viscosity, wood pulp polymerization degree and wood pulp methyl fiber content; the NMMO solution parameter data includes NMMO solution concentration; the chemical additive parameter data includes propylene glycol content and hydroxylamine content.

[0033] Preferably, the predicted raw material consumption value includes a predicted value of wood pulp consumption, a predicted value of NMMO solution consumption, a predicted value of propylene glycol consumption and a predicted value of hydroxylamine consumption; the predicted raw material cost includes a predicted cost of wood pulp, a predicted cost of NMMO solution, a predicted cost of propylene glycol and a predicted cost of hydroxylamine.

[0034] Preferably, the overall raw material consumption prediction model includes a first raw material consumption prediction sub-model constructed based on random forest, a second raw material consumption prediction sub-model constructed based on support vector machine, and a third raw material consumption prediction sub-model constructed based on long short-term memory network.

[0035] Preferably, the raw material data is preprocessed, specifically: Adoption 3 σ In principle, outlier detection is performed on the numerical data in the raw material data, and the interquartile range method is used to detect outliers on the non-normal distribution data in the raw material data, and the outliers are removed; then missing value processing and duplicate data removal are performed; Min-Max was used to normalize the raw material data; Pearson correlation analysis and principal component analysis were used to extract key features of raw material data; K-means cluster analysis and information entropy screening methods were used to conduct final screening of raw material data.

[0036] Preferably, the specific method for processing missing values ​​is: time series data is filled by linear interpolation method; if the missing ratio of numerical data is less than 20%, k-nearest neighbor interpolation is used to fill it, if the missing ratio exceeds 20%, the sample is eliminated; 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 exceeds 20%, the sample is eliminated.

[0037] Preferably, the final raw material consumption prediction value is obtained by weighted fusion of the prediction values ​​of each machine learning algorithm model, specifically: setting the initial weight distribution of the total raw material consumption prediction model, and calculating the mean square error of each machine learning algorithm model; then updating the Beta distribution parameters according to the mean square error, and introducing Gaussian process regression to calculate the uncertainty of each machine learning algorithm model within the total model, and then updating the Beta distribution parameters again according to the uncertainty; finally, according to the updated Beta distribution parameters combined with variational Bayesian inference calculation, the fusion weights of each machine learning algorithm model are obtained respectively.

[0038] Preferably, the system further comprises a production management module, and the production management module is used to formulate, adjust, approve, query and issue annual production plans, monthly production plans and daily production plans.

[0039] Preferably, the system further comprises an equipment management module, which is used to create and manage equipment information of production line, configure and issue inspection plans, and configure and issue equipment maintenance strategies; the equipment management module comprises an equipment information maintenance submodule, an equipment inspection submodule, and an equipment repair and maintenance submodule; wherein, The equipment information maintenance submodule is used to create and manage production line equipment information, including basic data such as the name, type and number of the configured equipment. The equipment inspection submodule is used to configure the inspection plan and send it to the employee terminal, and collect the inspection results to generate the corresponding inspection log; The equipment inspection and maintenance submodule is used to configure the maintenance strategy and send it to the employee terminal, and collect equipment inspection and maintenance to generate corresponding inspection and maintenance logs.

[0040] Beneficial effects of the present invention: 1. The key equipment status prediction module of the system of the present invention can realize the remaining life prediction and / or health prediction of the lyocell thin film evaporator, can discover potential problems of the equipment in advance, avoid sudden failure and shutdown of the equipment during operation, and cause interruption of the production process, thereby improving the reliability of the equipment, and can reasonably arrange the maintenance, overhaul and replacement of the equipment and production tasks according to the prediction results, reduce unnecessary equipment maintenance and replacement, reduce costs and improve production efficiency. Further, the key equipment status prediction module can also cooperate with the early warning module, 3D modeling and digital twin module. When there is a problem with the equipment status, the location of the problem equipment can be quickly located through the three-dimensional digital twin model of the production line. Furthermore, the energy consumption prediction module of the present invention can realize the energy consumption prediction in the production process of the lyocell fiber glue making process under the given process single conditions, and the obtained energy consumption prediction value helps the power supply and gas supply related departments of the enterprise to provide future power and gas consumption plan reports to cope with sudden power outages and gas outages during peak power and gas consumption in the future, and more energy consumption is required. Furthermore, the cost prediction module of the present invention considers the two aspects of raw material consumption and equipment maintenance, which helps to realize refined cost management in the production and manufacturing process of lyocell. In summary, the system of the present invention integrates 3D modeling with digital twin modules, key equipment status prediction modules, energy consumption prediction modules, cost prediction modules, early warning modules and other functional modules to form a complete system. This system is of great significance for guiding the production of lyocell fibers, and can help enterprises to better control costs and arrange production tasks, significantly improving the economic benefits of lyocell fiber production.

[0041] 2. The system of the present invention integrates the data of MES, DCS and other systems to establish an intelligent simulation system for lyocell fiber production line based on digital twins, which can break the data barriers between systems, achieve data consistency and accuracy, and improve the efficiency of information flow. The system of the present invention can provide a more comprehensive view of production data. During the production process, various departments can work together based on unified data to reduce errors and delays caused by inconsistent data.

[0042] 3. The system of the present invention considers the prediction of thin-film evaporators from two aspects: remaining life and health. Among them, the remaining life prediction focuses on the physical state changes of key components of the equipment, such as the vibration data of the upper and lower ends and the middle area of ​​the thin-film evaporator cylinder. Through the analysis and processing of these data, a remaining life prediction model is constructed to determine how long the equipment can still operate normally. The health prediction comprehensively considers the process production parameters of the equipment itself and related equipment, such as the temperature of the dissolved liquid, the pressure of the evaporation chamber, the slurry concentration, etc., to evaluate the health status of the equipment from two aspects: the operating status of the equipment and the process environment. The combination of the two can capture fault information from multiple dimensions of physical level and operating status, avoiding the limitations of a single prediction method.

[0043] 4. The system of the present invention is aimed at predicting the remaining life of the thin film evaporator. It uses multi-component characteristic frequency data and big data feature extraction data to form a data matrix, and combines the vibration data collected from the bearings at the upper and lower ends of the thin film evaporator cylinder and the main shaft in the middle area with the characteristic frequencies of the key components of the main shaft and bearings. In practical applications, it can more accurately capture the equipment degradation information, make the correlation between the characteristic data and the life label higher, and more effectively build the relationship model between the fault characteristics and the life label, thereby forming a more accurate remaining life prediction model.

[0044] 5. Aiming at the remaining life prediction of the thin film evaporator, the present invention collects vibration data of four areas, namely the upper and lower ends and the middle area of ​​the thin film evaporator cylinder, within a single life cycle from normal to abnormal, and designs a remaining life characteristic curve by coupling and driving multiple high-order exponential function empirical mathematical models with big data feature extraction, thereby reducing the complexity of the feature extraction network model, improving the real-time performance of the model operation, meeting the real-time life prediction requirements, and providing data labels more accurately, which can effectively improve the accuracy of the remaining life prediction.

[0045] 6. The present invention aims at predicting the remaining life of a thin film evaporator. The average value of four characteristic curves about the main shaft in the middle area of ​​the thin film evaporator is taken to obtain the remaining life characteristic curve of the main shaft. The characteristic curves of the upper bearing, the lower bearing and the main shaft of the thin film evaporator that independently reflect the health status are constructed respectively, thereby simplifying the calculation process.

[0046] 7. The present invention aims at predicting the remaining life of a thin film evaporator. By integrating the characteristic frequency data of multiple components in historical data with the feature extraction data of big data and the characteristic curve data of remaining life, a three-dimensional training data set is formed, and a remaining life prediction model based on deep learning model design is trained. The remaining life prediction model is trained based on the data set, and an association mapping between the characteristic data and the life label is established to improve the prediction accuracy, thereby obtaining a remaining life prediction model with satisfactory accuracy.

[0047] 8. The present invention improves the prediction accuracy of the health prediction of thin film evaporators by integrating the equipment-level parameters and process-level parameters of thin film evaporators. Based on time series analysis and LSTM networks, by collecting the equipment parameters of thin film evaporators and the intrinsic product parameters of the equipment, combined with the equipment parameters of the upstream pulper and the intrinsic product parameters of the equipment, a cross-equipment and cross-process data coupling relationship is established. This multi-level data fusion method can more comprehensively reflect the production conditions, so that the prediction model can more accurately identify faults, thereby reducing the false alarm and missed alarm rates and improving the prediction reliability.

[0048] 9. For the health prediction of thin film evaporator, the present invention adopts a method combining standard deviation + expert experience scoring + physical constraints to assign weights to various parameters to ensure that the importance of key parameters in the prediction model is reasonably reflected.

[0049] 10. The health prediction model of the present invention adopts the LSTM time series model to improve the accuracy of equipment health prediction. LSTM (Long Short-Term Memory Network) is used to learn time series data to improve the long-term dependence ability of the prediction model. LSTM can capture long-term trends and avoid misjudgments caused by short-term fluctuations.

[0050] 11. The present invention improves the intelligent management of equipment through the maintenance suggestion mechanism. HI ), provides maintenance suggestions under different health conditions: HI >0.8, it means the equipment is operating normally and only requires daily monitoring; HI In the range of 0.6 ~ 0.8, it means that the equipment is slightly abnormal and the process parameters can be adjusted; HI In the range of 0.4 to 0.6, planned maintenance is required; when HI In the range of 0.2 ~ 0.4, it indicates that the equipment is seriously abnormal and it is recommended to check it immediately. HI <0.2, indicating equipment failure, it is recommended to shut down for repair or replace parts.

[0051] 12. Since steam energy consumption and electricity energy consumption account for a large proportion of the cost of the Lyocell fiber glue making process, the present invention specifically integrates and constructs an energy consumption prediction module in the system to realize the energy consumption prediction of the Lyocell fiber glue making process, which helps enterprises to assist in optimizing the production process of Lyocell fibers and reduce the cost of the Lyocell fiber glue making process.

[0052] 13. For the energy consumption prediction module, the method adopted by the present invention controls the size of the training set by calculating the most suitable threshold interval percentage of process similarity, thereby improving the credibility and accuracy of the model prediction.

[0053] 14. The present invention predicts the energy consumption of the lyocell glue making process by screening out historical screening and sorting data, and using the historical screening and sorting intermediate parameter data and the historical screening and sorting energy consumption parameter data in the historical screening and sorting data as training sets of the neural network. An energy consumption prediction model is obtained through neural network training. Compared with the prior art that uses a large amount of data as the training set of the neural network model, the present method only uses the screened historical screening and sorting data, which effectively reduces the complexity of the model and improves the fitting ability and generalization ability of the model.

[0054] 15. The present invention constructs a glue liquid weight calculation model, a glue liquid refractive index calculation model and a glue production consumption time calculation model to extract and process the glue liquid weight, glue liquid refractive index and consumption time from the current process sheet as the input results of the energy consumption prediction model, so that the energy consumption prediction model can achieve the technical effect of energy consumption prediction. 16. The present invention predicts the energy consumption of the Lyocell glue making process by solving the process similarity to adjust the parameters of each model, thereby reducing the risk of model failure due to process changes and enhancing the reliability of the model.

[0055] 17. The present invention predicts the energy consumption of the lyocell glue making process by selecting a feature set for training based on the analysis of the mechanism that affects the output of each model. On the one hand, the model has strong interpretability, and on the other hand, it reduces the input features, which facilitates the implementation of the energy consumption prediction model.

[0056] 18. The present invention constructs a glue liquid weight calculation model, a glue liquid refractive index calculation model and a glue making time calculation model based on a BP neural network algorithm and a least squares method-based nonlinear parameter fitting algorithm, thereby ensuring the accuracy of the final glue liquid weight, glue liquid refractive index and consumption time, making the final energy consumption prediction result more accurate.

[0057] 19. The energy consumption forecast value of the glue making process output by the energy consumption forecast module of the present invention helps the power supply and gas supply related departments of the enterprise to provide future power and gas consumption plan reports to cope with sudden power and gas outages during peak power and gas consumption times in the future, which will require more energy consumption.

[0058] 20. For the energy consumption prediction value of the glue making process output by the energy consumption prediction module of the present invention, the relevant staff can judge whether the process and process equipment are unstable based on the obtained energy consumption prediction value. If unstable, the relevant staff can check and judge the specific problems based on the unstable situation, thereby alerting the relevant staff to pay attention to the technical effects of the process and process equipment and ensuring the normal operation of the process and process equipment.

[0059] 21. The cost prediction module of the present invention optimizes the weight coefficients, first accurately predicts the raw material consumption and equipment maintenance, and then accurately calculates and predicts the total cost of Lyocell production and manufacturing through the obtained accurate prediction values, providing a scientific basis for Lyocell process optimization and resource scheduling, and assisting enterprises to achieve refined cost management in the Lyocell production and manufacturing process.

[0060] 22. The present invention adopts three different machine learning methods based on random forest, support vector machine and long short-term memory network to construct a raw material consumption prediction model, an energy consumption prediction model and an equipment maintenance prediction model, and then the three raw material consumption prediction values, three energy consumption prediction values ​​and three equipment maintenance prediction values ​​are respectively obtained by weighted voting to accurately obtain the final raw material consumption prediction value, the final energy consumption prediction value and the final equipment maintenance prediction value, thereby improving the accuracy of the final predicted total production and manufacturing cost.

[0061] 23. The present invention aims at predicting the manufacturing cost of lyocell fiber, combining variational Bayesian inference, Gaussian process regression and Beta distribution update to optimize the model fusion weight calculation, adapt it to different data distributions, and improve the stability and accuracy of system cost prediction.

[0062] 24. Aiming at the prediction of the manufacturing cost of lyocell fiber, the present invention introduces Gaussian process regression to calculate the uncertainty of the three sub-models within each overall model, and incorporates it into the weight update rule, thereby achieving the technical effect of reducing the weight of the high uncertainty model and improving the system's stability in cost prediction.

[0063] 25. The present invention predicts the manufacturing cost of lyocell fibers and adopts Beta distribution update, which is different from the traditional Beta distribution update. It adopts variational inference to dynamically associate the parameter update of Beta distribution with data errors and uncertainties, thereby achieving the technical effect of improving the speed and adaptability of weight adjustment.

[0064] 26. The present invention predicts the manufacturing cost of lyocell fibers and obtains the fusion weights of the three sub-models in each overall model through variational Bayesian inference calculation, thereby realizing dynamic optimization of model weights, ensuring that the weight calculation is not only based on error distribution, but also can adapt to different data characteristics and improve the technical effect of prediction accuracy.

[0065] 27. The cost prediction module of the present invention predicts the raw material costs and equipment maintenance costs, which is helpful to guide the enterprise in future production and sales, effectively reduces the cost of the enterprise in production and manufacturing, and increases the profit of the enterprise in future sales.

[0066] 28. The cost prediction value obtained by the cost prediction module of the present invention can assist in determining the process consumption cost involved in the Lyocell manufacturing process. Relevant staff can specifically determine the recovery cost involved in the process consumption cost, thereby finely managing the costs involved in the entire process production and manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] The foregoing and following detailed description of the present invention will become more apparent when read in conjunction with the following drawings, in which: Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a diagram showing the arrangement of vibration sensor measuring points on the cylinder of the thin film evaporator of the present invention; Figure 3 It is the vibration vector data diagram of the present invention; Figure 4 is the vibration vector data characteristic curve of the present invention; Figure 5 This is a data processing architecture diagram of the energy consumption prediction module of the present invention; Figure 6 This is a diagram of the data processing architecture of the cost prediction module of the present invention. DETAILED DESCRIPTION

[0068] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions for achieving the purpose of the present invention will be further described below through specific embodiments. It should be noted that the technical solutions claimed for protection in the present invention include but are not limited to the following embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0069] Example 1 The embodiment of the present invention proposes a lyocell fiber production line intelligent simulation system based on digital twin, Figure 1 This is the system architecture diagram of the present invention, please refer to the attached manual Figure 1 The system mainly includes the following functional modules connected to the central command and dispatch system: 3D modeling and digital twin module, key equipment status prediction module, energy consumption prediction module, early warning module and process management module.

[0070] The following will provide a detailed explanation and description of each functional model.

[0071] First, for the 3D modeling and digital twin module, the 3D modeling and digital twin module uses SolidWorks software to build a three-dimensional model of the Lyocell production line based on the actual site information (spatial layout information) of the Lyocell production line scanned and obtained by the data acquisition end. The three-dimensional model includes an equipment model and a fixed building model (the model covers the plane geometry of the production line and the equipment in detail, as well as the assembly structure and other information, including accurate modeling of key production equipment such as pulpers, thin-film evaporators and spinning equipment); then use 3Ds Max three-dimensional rendering software to convert the constructed three-dimensional model into a format conversion and lightweight processing, and convert it into an FBX format file; then import the model into Unity 3D software, and build a virtual simulation model based on the actual environment of the factory production line; finally, through the connection with sensors, MES (manufacturing execution system), ERP (enterprise resource planning system), DCS (distributed control system) and OPC servers, real-time data such as physical equipment, work-in-progress and environment, such as the operating temperature, rotation speed, material flow, etc. of the equipment. By binding these real-time data with the equipment in the corresponding three-dimensional scene, and driven by the Unity 3D virtual engine, the virtual model and the actual production process are synchronized, and finally a three-dimensional digital twin model of the Lyocell production line is constructed. Through the constructed three-dimensional digital twin model of the Lyocell production line, operators can intuitively observe the real-time operating status of the production line, such as the operation of the equipment and the flow trajectory of materials, which provides convenience for production monitoring and management.

[0072] The key equipment status prediction module includes a remaining life prediction submodule for a thin film evaporator; the remaining life prediction submodule for the thin film evaporator processes the real-time vibration data collected and transmitted by the vibration sensor on the target thin film evaporator and the real-time spindle speed data transmitted by the central control platform of the target thin film evaporator, and inputs the real-time vibration data and the real-time spindle speed data into a trained remaining life prediction model to obtain a remaining life prediction result for the target thin film evaporator. Specifically, the remaining life prediction submodule of the thin film evaporator segments the vibration data according to the set time interval to obtain multiple vibration data segments with set time intervals, and then performs time domain and frequency domain analysis on each vibration data segment to obtain the root mean square, kurtosis and frequency centroid of each vibration data segment; obtains the spindle fault characteristic frequency according to the acquired real-time spindle speed, and then obtains the spindle bearing fault characteristic frequency according to the real-time spindle speed and the size parameters of the spindle bearing; finally, the obtained root mean square, kurtosis, frequency centroid, spindle fault characteristic frequency, and spindle bearing fault characteristic frequency are input into the trained remaining life prediction model, and the remaining life prediction result of the target thin film evaporator is output; the remaining life prediction of the target thin film evaporator includes the remaining life of the spindle and / or the remaining life of the spindle bearing.

[0073] The energy consumption prediction module extracts the pulp porridge weight, process parameter group 1, process parameter group 2 and process parameter group 3 corresponding to the current process list according to the process list provided by the process management module; then the pulp porridge weight and process parameter group 1 are integrated into the glue liquid weight through the glue liquid weight calculation model constructed based on the BP neural network algorithm to obtain the glue liquid weight; the pulp porridge weight and process parameter group 3 are integrated into the glue making consumption time through the glue making consumption time calculation model constructed based on the least squares method fitting nonlinear parameter algorithm to obtain the glue liquid refractive index; the process parameter group 2 is passed through the glue liquid refractive index calculation model constructed based on the BP neural network algorithm to obtain the glue liquid refractive index; finally, the pulp porridge weight, glue liquid weight, glue liquid refractive index and glue making consumption time are input into the trained energy consumption prediction model, and the energy consumption prediction result corresponding to the glue making process is output.

[0074] In this embodiment, for the remaining life prediction submodule of the thin film evaporator, the influence of the equipment structure characteristics and vibration sensitivity of the thin film evaporator on production is mainly considered. By collecting the vibration data of the thin film evaporator during operation and combining it with the physical characteristic frequency for analysis, the remaining life of the equipment is finally predicted by the deep learning model. The main shaft of the large thin film evaporator has a large axial dimension and is sensitive to vibration. The main shaft is connected to a complex structure, and the system structure mode is complex. The vibration of the main shaft can easily excite the system structure mode, affecting the normal operation of the equipment and even damaging the equipment. At the same time, the main shaft locking / stuck is a common fault in the working process of the large thin film evaporator, and the excessive vibration displacement of the main shaft is an important cause of the fault. The causes of abnormal vibration of the main shaft include imperfect equipment design, easy resonance during operation, damage to the main shaft bearing, unbalanced main shaft wear, and main shaft bending failure. These faults can lead to a decrease in the evaporation efficiency of the thin film evaporator, unstable product quality, and even equipment shutdown, seriously affecting the continuity and economy of spinning production.

[0075] Specifically, the construction process of the remaining life prediction submodule of the thin film evaporator is as follows: (1) Constructing a thin film evaporator vibration data acquisition unit for acquiring thin film evaporator vibration data and spindle speed; The vibration data are acceleration signals in three directions collected by a total of 6 vibration sensors arranged at the upper and lower ends and the middle area of ​​the thin film evaporator cylinder, a total of 18 acceleration signals; among them, 1 vibration sensor at the upper end of the cylinder is used to monitor the fault condition of the upper end bearing of the main shaft, 1 vibration sensor at the lower end of the cylinder is used to monitor the fault condition of the lower end bearing of the main shaft, and 4 vibration sensors in the middle area of ​​the cylinder are used to monitor the fault condition of the main shaft; (2) Writing a vibration data preprocessing calculation template into a vibration data preprocessing unit to construct a vibration data preprocessing unit; the vibration data preprocessing calculation template is used to process the acquired vibration data and spindle speed data of the thin film evaporator, and the data processing process involved in the preprocessing calculation template is as follows: The acquired vibration data is segmented according to set time intervals to obtain multiple data segments of set time intervals; time domain analysis and frequency domain analysis are performed on each data segment to obtain the root mean square, kurtosis and frequency center of gravity of each vibration data segment; the fault characteristic frequency of the spindle is obtained according to the spindle speed data, and the fault characteristic frequency of the spindle bearing is obtained according to the spindle speed data and the size parameters of the spindle bearing; For the root mean square, kurtosis and frequency centroid of each acceleration signal, as well as the fault characteristic frequency of the bearing and the fault characteristic frequency of the main shaft, these data can be integrated into a data matrix, and the root mean square, kurtosis and frequency centroid of each acceleration signal, the fault characteristic frequency of the bearing and the fault characteristic frequency of the main shaft in the data matrix are normalized and then output; The normalization process used in the vibration data of the present invention adopts the most conventional method, and the calculation method is: the difference between the size of the current data and the minimum value of the data is divided by the difference between the maximum value of the data and the minimum value, and a number between 0 and 1 is obtained; (3) Using the historical vibration data of the thin film evaporator to train the constructed remaining life prediction model, obtaining a trained remaining life prediction model, and finally constructing a remaining life prediction output unit; (4) The vibration data acquisition unit, the vibration data preprocessing unit and the remaining life prediction output unit are associated according to the signal relationship to form a remaining life prediction submodule of the thin film evaporator.

[0076] The remaining life prediction model is constructed and trained in the following way: Step S1, acquiring historical vibration data and historical spindle speed data of the whole process from normal operation to failure of the spindle and / or spindle bearing resulting in failure of the thin film evaporator to operate normally.

[0077] The vibration data includes acceleration signals in three directions respectively collected by vibration sensors at the upper and lower ends and the middle area of ​​the thin film evaporator cylinder; wherein, one vibration sensor is respectively arranged at the upper and lower ends to monitor the fault conditions of the upper and lower end bearings of the main shaft, and four vibration sensors are arranged in the middle area of ​​the cylinder to monitor the fault conditions of the main shaft.

[0078] Step S2, segmenting the acquired historical vibration data according to set time intervals to obtain multiple data segments of set time intervals; performing time domain analysis and frequency domain analysis on each data segment, wherein the time domain analysis is used to extract the root mean square and kurtosis of each data segment, and the frequency domain analysis is used to convert each data segment from the time domain to the frequency domain, and extract the frequency center of gravity; then obtaining the fault characteristic frequency of the spindle according to the historical spindle speed data obtained in step S1, and finally obtaining the fault characteristic frequency of the spindle bearing according to the spindle speed data and the size parameters of the spindle bearing; It is understandable that the data processing process involved in step S2 is the same as the data processing process involved in the vibration data preprocessing calculation template. In addition, during implementation, the time interval is set to 0.5s.

[0079] Step S3, performing vector addition on the acceleration signals in three directions of each vibration sensor obtained in step S1, and obtaining a total of 6 vibration vector data, and building a one-dimensional convolutional neural network model including convolution, pooling, activation function and full connection layer for each vibration vector data; extracting the response trend of the vibration response of each vibration vector data changing with time through the one-dimensional convolutional neural network model of each vibration vector data, and obtaining the vibration vector data characteristic curve of the upper and lower end bearings and the main shaft; using the model to capture the changing trend of the vibration response of the vibration vector data over time, this trend reflects the characteristic data of the equipment life degradation, and presents as extended over time such as Figure 4 A curve is shown in the figure. In this figure, the horizontal axis is time, and the data on the curve reflects the relationship between vibration data and life at a certain point in time.

[0080] In step S3, the one-dimensional convolutional neural network model output changes with time to obtain Figure 4 The vibration vector data shown shows the changing trend of vibration response over time.

[0081] Step S4, using multiple high-order exponential functions to fit the vibration vector data characteristic curves of the upper and lower end bearings and the main shaft obtained in step S3, respectively, to form the remaining life characteristic curve of the upper end bearing of the main shaft, the remaining life characteristic curve of the lower end bearing of the main shaft and the remaining life characteristic curve of the main shaft, and then express them in percentage form.

[0082] In step S4, the vibration vector data characteristic curve of the main shaft is obtained by taking the average value of the vibration vector data characteristic curves corresponding to the vibration sensors in the middle area of ​​the cylinder.

[0083] In step S4, the calculation formula expressed in percentage form after forming the remaining life characteristic curve of the upper end bearing of the main shaft, the remaining life characteristic curve of the lower end bearing of the main shaft and the remaining life characteristic curve of the main shaft is as follows: ; Where: , , , , It is an adjustable parameter. The specific debugging process is as follows: first set the initial value to 1, further observe the fit between the vibration vector data characteristic curve and the remaining life characteristic curve obtained after fitting, and then adjust , , , , until the vibration vector data characteristic curve is well matched with the remaining life characteristic curve obtained after fitting; is the system time of vibration data; The time limit for the equipment to go from completely normal to abnormal can be obtained from the horizontal axis of the vibration data; is the remaining life percentage; e Is the natural base.

[0084] Step S5, taking the time interval set in step S2 as the time resolution, the root mean square, kurtosis, frequency centroid, fault characteristic frequency of the spindle bearing and the fault characteristic frequency of the spindle obtained in step S2, and the data on the remaining life characteristic curve of the upper end bearing of the spindle, the remaining life characteristic curve of the lower end bearing of the spindle and the remaining life characteristic curve of the spindle obtained in step S4, these data together constitute the data set of the remaining life prediction model; among which, the data on the remaining life characteristic curve is label data.

[0085] In practical applications, for ease of understanding, the root mean square, kurtosis, frequency centroid, fault characteristic frequency of the spindle bearing, fault characteristic frequency of the spindle obtained in step S2, and the remaining life characteristic curve of the upper end bearing of the spindle, the remaining life characteristic curve of the lower end bearing of the spindle and the remaining life characteristic curve data of the spindle obtained in step S4 are drawn on the same two-dimensional coordinate graph. Each moment on the two-dimensional coordinate graph corresponds to a series of data, and these data together constitute a data set.

[0086] Step S6, constructing a remaining life prediction model based on the one-dimensional convolutional neural network model, dividing the data set obtained in step S5 into a training set, a validation set and a test set in a ratio of 8:1:1, and using the divided data sets to train, validate and test the constructed remaining life prediction model, and finally obtaining a trained remaining life prediction model with accuracy that meets the requirements.

[0087] Since the mechanical system will vibrate during operation, the vibration will change when a component fails, and the vibration response will be different for different fault levels. Therefore, the time of component failure or even failure can be predicted by analyzing the characteristic changes of the vibration response, that is, life prediction. The root mean square, kurtosis and frequency center of gravity, the fault characteristic frequency of the spindle and the fault characteristic frequency of the spindle bearing reflect the characteristics of the vibration response, and these characteristics can be used to describe the vibration situation. The remaining life prediction model is established based on the deep learning model. The establishment process uses the characteristic data of the vibration response of the equipment parts throughout the life cycle to train the deep learning model. The deep learning model learns the characteristics of the component degradation process through training with a large amount of data, thereby realizing the prediction of component life.

[0088] In the embodiment described in the present invention, the design method of the remaining life label data in the remaining life prediction is as follows: first, the label data set is designed. First, in multiple experiments, the data acquisition module is used to obtain 18 acceleration signals in three directions of six vibration sensors during the whole process from the equipment running intact to the failure of the main shaft and the main shaft bearing causing the equipment to fail to operate normally. Then, the vibration acceleration in three directions of each sensor is vector-added to obtain 6 vibration vector data, and a schematic diagram of one vibration vector data is shown as follows: Figure 3 As shown, the amplitude represents the magnitude of the vibration acceleration collected by the vibration sensor, and the sample point represents the number of sampling points of the vibration acceleration collected by the vibration sensor.

[0089] For each vibration vector data collected by the vibration sensor, a one-dimensional convolutional neural network deep learning mathematical model including convolution, pooling, activation function and fully connected layer is built. The response trend of the vibration response over time is extracted through the one-dimensional convolutional neural network deep learning mathematical model, and finally the vibration vector data characteristic curve formed by the response trend of the vibration vector data of the upper and lower end bearings and the main shaft of the six main shafts over time is obtained. Figure 4 This is a schematic diagram of one of the vibration vector data characteristic curves. The amplitude represents the magnitude of the vibration acceleration collected by the vibration sensor, and the sample point represents the number of sampling points of the vibration acceleration collected by the vibration sensor. The curve has obvious characteristics. As time goes by, the damage of the spindle and the spindle bearing gradually accumulates, and when it accumulates to a certain level, it collapses rapidly. This phenomenon is the same as the damage phenomenon of mechanical parts in real situations. Therefore, this curve can be used as a reference for estimating the remaining life of the spindle and the spindle bearing.

[0090] Among the six vibration vector data characteristic curves, two vibration vector data characteristic curves are derived from the data of the upper and lower sensors of the thin film evaporator, which respectively reflect the health status of the upper and lower main shaft bearings. Another vibration vector data characteristic curve is derived from the data of the four vibration sensors on the cylinder of the thin film evaporator, which reflects the health status of the main shaft. For the convenience of calculation, the data of these four vibration vector data characteristic curves are averaged to obtain a vibration vector data characteristic curve of the main shaft. Finally, a total of three vibration vector data characteristic curves reflecting the health status of the main shaft and main shaft bearings of the thin film evaporator will be obtained, namely, one vibration vector data characteristic curve of the upper end bearing of the main shaft, one vibration vector data characteristic curve of the lower end bearing of the main shaft, and one vibration vector data characteristic curve of the main shaft.

[0091] However, the obtained vibration vector data characteristic curve still has defects, such as obvious fluctuations, which are inconsistent with the actual damage caused by the accumulation of thin film evaporator parts. This is because there are many interferences in the measured vibration signal. In view of this, this embodiment uses a method that combines empirical mathematical models with big data feature extraction to design the remaining life characteristic curve. This embodiment uses multiple high-order exponential functions to fit the vibration vector data characteristic curve, and forms the corresponding remaining life characteristic curve of the upper end bearing of the main shaft, the remaining life characteristic curve of the lower end bearing of the main shaft, and the remaining life characteristic curve of the main shaft, and then expresses it in percentage form.

[0092] At this point, with a time resolution of 0.5s, the response time of the vibration data can be used as the horizontal axis, and then the root mean square, kurtosis, frequency center of gravity, bearing fault characteristic frequency, shaft fault characteristic frequency and remaining life characteristic curve data of the vibration data can be plotted on the same two-dimensional coordinate graph. Each moment in the two-dimensional coordinate graph corresponds to a series of data, and these data together constitute the data set, among which the data on the remaining life characteristic curve is the label data.

[0093] In the embodiment described in the present invention, the vibration data segment is analyzed in the time domain and the frequency domain to obtain the root mean square, kurtosis and frequency center of gravity of the vibration data segment, including: Perform time domain analysis on the data segments and extract the root mean square and kurtosis of each data segment. The calculation method is as follows: ; ; Where: is the RMS of the data segment; is the number of samples in the vibration data segment; The first i Sample values: is the kurtosis of the data segment; u is the average value of the data segment; Use frequency domain analysis to convert the data fragment from time domain to frequency domain, and then extract the frequency center of gravity; The data fragment is converted from time domain to frequency domain, and the calculation method is: ; Where: is the frequency domain value after the data segment is transformed; N 1 is the number of discrete sampling points of the data segment; is the sequence number of the time domain discrete value of the data segment; The first n data samples; J is an imaginary unit; Furthermore, the calculation method for extracting the frequency center of gravity is: ; Where: is the frequency center of gravity; For the k The frequency value of the spectral line; is the total number of spectral lines.

[0094] Further, in step A2, the main shaft bearing fault characteristic frequency includes: Inner race fault characteristic frequency: ; Outer race fault characteristic frequency: ; Rolling element failure characteristic frequency: ; Cage failure characteristic frequency: ; Where: Q is the number of rolling elements, d The diameter of the rolling element, D Bearing pitch diameter, β is the contact angle, f r is the rotation frequency of the spindle, obtained according to the spindle speed; For the spindle fault characteristic frequency, the calculation method is as follows: ; Where: n 1 is the multiple number, f r is the rotation frequency of the main axis; is the main axis disturbance frequency.

[0095] In the present invention, the main shaft fault characteristic frequency and the main shaft bearing fault characteristic frequency together constitute the characteristic frequency of the key components of the thin film evaporator.

[0096] The construction process of the energy consumption prediction module is as follows: (5) Constructing a process sheet receiving unit for obtaining the process sheet of the rubber making stage of Lyocell fiber production; (6) Constructing a process data extraction unit that internally stores a calculation model for glue liquid weight, a calculation model for glue consumption time, and a calculation model for glue liquid refractive index; (7) Constructing an energy consumption prediction output unit that internally stores a trained energy consumption prediction model; (8) The process sheet receiving unit, the process sheet data extraction unit and the energy consumption prediction output unit are associated according to signal relationships to form the energy consumption prediction module.

[0097] The construction and training process of the energy consumption prediction model is as follows: Step B1, collecting the evaporation system parameter data, glue and sealing system parameter data, insulation water and vacuum system parameter data, porridge weight, glue weight, glue refractive index, consumption time and energy consumption data of the historical original data in the Lyocell fiber glue making process, and performing data preprocessing to obtain historical data; the data preprocessing technology is a mature technical method in this field in the prior art, and these methods can be used to eliminate abnormal values ​​in the historical original data and fill in the missing values ​​in the historical original data.

[0098] In this embodiment, the data preprocessing in step B1 adopts 3 σ In principle, outliers are removed; after outliers are removed, the missing values ​​are filled by taking the weighted average of the K neighboring data values ​​near the missing value.

[0099] Step B2: Calculate the process similarity of the historical process parameter data in the corresponding historical data according to the process sheet of the Lyocell fiber glue making process, and rearrange the historical data in descending order of process similarity to obtain historical sorted data.

[0100] In this embodiment, in step B2, the process similarity is calculated using cosine similarity, and the calculation method is: ; In the formula, θ i is the process similarity; X i For the i A vector consisting of the process parameter data of the row history; X now A vector consisting of the process sheet data for the Lyocell fiber glue making process.

[0101] Step B3, select a reference group sequence from the historical sorting data, calculate the process similarity of each reference group sequence, thereby obtaining the historical sorting data sequence of the reference group sequence, and in the process of obtaining the historical sorting data sequence of the reference group sequence, obtain the threshold interval of process similarity, and then use the historical sorting data sequence of the reference group sequence within the threshold interval of process similarity to calculate the energy consumption prediction value sequence of the reference group sequence; the method for solving the process similarity in step B3 is the same as that in step B2.

[0102] In this embodiment, the specific method of selecting a reference group from the historical sorting data is: Z The row data constitutes the reference group sequence , Z The value of should not be too large, otherwise it will increase the amount of calculation. Z You can take an even random number in [10,20]; the reference group sequence Each sequence in is used as the current process list, and the historical sorting data corresponding to each reference group sequence is obtained respectively. , thus forming the historical sorting data sequence of the reference group sequence , and record the maximum process similarity value in the process ; The threshold interval of process similarity is , the minimum process similarity value The calculation method is as follows: ; In the formula, is the initial value of the threshold interval percentage of the set process similarity; Furthermore, in step B3, the energy consumption prediction value sequence of the reference group sequence is calculated using the historical sorted data sequence of the reference group sequence in the following specific manner: The porridge weight of the historically sorted intermediate parameter data and the historically sorted process parameter data in the historically sorted data sequence of the reference group sequence is used as a training set of the neural network model, thereby obtaining an energy consumption prediction value sequence of the reference group sequence; the process similarity corresponding to the historically sorted intermediate parameter data belongs to the threshold interval range of the process similarity; Furthermore, in step B3, before using the porridge weight of the historically sorted intermediate parameter data and the historically sorted process parameter data as the training set of the neural network model for training, it is necessary to eliminate the difference between the dimensions of the input data through homogenization processing, and the calculation method is: ; In the formula, For the training set i OK j The result after column homogenization; For the training set i OK j Column original data; For the training set j The minimum value of the process parameters in the column; For the training set j The maximum value among the process parameters.

[0103] It can be understood that the neural network model in step B3 specifically selects a three-layer BP neural network, with 4 neurons in the input layer; 4 neurons in the output layer; 1 hidden layer, and 4 neurons; the activation function selects the Tanh activation function; the loss function selects the MSE; the initial weight takes a value between -1 and 1; the initial bias value is zero; the learning rate is set to 0.001; and the precision is set to five decimal places.

[0104] Step B4: Compare the energy consumption prediction value sequence of the reference group sequence with the historical energy consumption data of the corresponding reference group sequence respectively, and record their absolute errors and X.

[0105] In this embodiment, the energy consumption prediction value sequences of the reference group sequences are compared with the historical energy consumption data of the corresponding reference group sequences, and the specific method of recording their absolute errors and X is as follows: Compare the energy consumption prediction value sequence of the reference group sequence with the historical energy consumption data of the corresponding reference group sequence, record their absolute error values, and traverse the historical sorting data sequence. Z Sorting historical data to obtain the absolute error value sequence of energy consumption prediction , accumulate the energy consumption prediction absolute error value sequence to obtain the absolute error and X and record it.

[0106] Step B5: gradually increase the threshold interval percentage P of process similarity to obtain the threshold interval of current process similarity; repeat steps B3 to B4 until the threshold interval of current process similarity reaches the set maximum value, and obtain the threshold interval percentage P of process similarity corresponding to the minimum absolute error and X in the process of repeating steps B3 to B4, and mark P as P s ; Among them, the maximum value of the threshold interval setting is artificially set and should not be too large. If the maximum value of the threshold interval setting is too large, it will lead to too much training data, causing overfitting and extending the calculation time. In practice, the maximum value of the threshold interval setting is half of the maximum similarity.

[0107] In this embodiment, the threshold interval percentage of the process similarity is increased step by step. The specific calculation formula is: ; In the formula, P represents the threshold interval percentage of the current process similarity; Indicates the initial value of the threshold interval percentage of the set process similarity; S Indicates the percentage step value of the threshold interval of the set process similarity; Indicates the current number of steps; The threshold interval of the current process similarity is : ; In the formula, Indicates the current minimum process similarity value; Indicates the current maximum process similarity value that is set.

[0108] Step B6: Build an energy consumption prediction model based on BP neural network, and then use the historical sorting data to calculate the threshold interval percentage P of process similarity. s The set range filtering is used to obtain the historical filtering and sorting data, and the porridge weight, glue weight, glue refractive index, consumption time and energy consumption data in the historical filtering and sorting data are used as the data set of the energy consumption prediction model. The data set is divided into training set, test set and validation set, and then the divided data set is used to train, test and validate the constructed energy consumption prediction model, and finally the trained energy consumption prediction model is obtained.

[0109] In this embodiment, step B6 is specifically as follows: the glue liquid weight, glue liquid refractive index and consumption time in the intermediate parameter data of historical screening and sorting, and the porridge weight in the process parameters of historical screening and sorting are used as the input features of the training set of the energy consumption prediction model constructed above, and the energy consumption data of historical screening and sorting are used as the target output of the energy consumption prediction model, and the model is trained using the above data to obtain a trained energy consumption prediction model; the energy consumption prediction model uses a three-layer BP neural network, the number of neurons in the input layer is 4, and the number of neurons in the output layer is 1; the hidden layer is 1, and the number of neurons in this layer is 4.

[0110] In this embodiment, the training set of the energy consumption prediction model needs to be homogenized to eliminate the differences between data dimensions, and then the model is trained using the training set.

[0111] In the embodiment described in the present invention, the energy consumption data is the sum of the electricity energy consumption and the steam energy consumption; wherein the specific calculation formula for converting the steam energy consumption into the electricity energy consumption is as follows: ; In the formula, The electricity consumption for steam; is the price of steam, in yuan / m 3 ; is the price of electric energy, in Yuan / KW·h; is the amount of steam used, in m 3 .

[0112] In the embodiment depicted in the present invention, historical original data include historical original process parameter data, historical original energy consumption data and historical original intermediate parameter data; historical data include historical process parameter data, historical energy consumption data and historical intermediate parameter data; historical sorted data include historical sorted process parameter data, historical sorted intermediate parameter data and historical sorted energy consumption data; historical screened and sorted data include historical screened and sorted process parameter data, historical screened and sorted intermediate parameter data and historical screened and sorted energy consumption data; intermediate parameter data include glue liquid weight, glue liquid refractive index and consumption time; energy consumption data include electricity energy consumption and steam energy consumption.

[0113] In the embodiment depicted in the present invention, the process sheet includes process parameter data; the process parameter data include evaporation system parameter data, glue liquid and sealing system parameter data, insulation water and vacuum system parameter data and porridge weight; the evaporation system parameter data include evaporation heating system parameter data, thin film evaporator parameter data and evaporation condensation water system parameter data; the glue liquid and sealing system parameter data include sealing liquid system parameter data and glue liquid transport parameter data; the insulation water and vacuum system parameter data include insulation water system parameter data and vacuum system parameter data.

[0114] In the embodiment depicted in the present invention, the evaporation heating system parameter data includes the heating system temperature, the heating system liquid level and the heating system pressure; the thin film evaporator parameter data includes the four-zone temperature of the thin film evaporator, the thin film evaporator motor speed, the thin film evaporator bearing lubricating oil circulation temperature, the reducer bearing speed, the cooling fan speed, the thin film evaporator liquid level, the thin film evaporator outlet pressure and the thin film evaporator bottom outlet temperature; the four-zone temperature of the thin film evaporator includes the jacket temperature of the four evaporation zones and the glue temperature of the four evaporation zones; the glue delivery parameter data includes the temperature of the insulation water heat exchanger in the pump, the flow rate of the insulation water heat exchanger in the pump, The temperature of the insulation water heat exchanger outside the pump, the flow rate of the insulation water heat exchanger outside the pump, the speed of the glue delivery pump, the outlet pressure of the glue delivery pump and the outlet temperature of the glue delivery pump; the sealing liquid system parameter data include the sealing liquid tank temperature, the sealing liquid flow and the sealing liquid tank liquid level; the insulation water system parameter data include the insulation water circulation pump outlet temperature and the insulation system hot water pressure; the evaporative condensate water system parameter data include the evaporative condenser temperature, the evaporative condenser liquid level, the evaporative condenser pressure, the flow rate of cooling water to the evaporative condenser and the outlet flow rate of the evaporative condensate pump; the vacuum system parameter data include the vacuum system liquid level, the vacuum system pressure and the vacuum pump speed.

[0115] In this embodiment, the glue liquid weight calculation model is constructed based on the BP neural network algorithm. Specifically, firstly, a glue liquid weight calculation model is constructed based on the BP neural network algorithm, and then the slurry weight, thin film evaporator motor speed, thin film evaporator four-zone temperature, vacuum system pressure and vacuum pump speed in the historically screened and sorted process parameter data are used as input features of the glue liquid weight calculation model training set, and the glue liquid weight in the historically screened and sorted intermediate parameter data is used as the target output of the glue liquid weight calculation model training set, and the above data are used to train, test and verify the model, and finally a trained glue liquid weight calculation model is obtained.

[0116] It can be understood that the number of input layer neurons of the glue weight calculation model is set to 8, and the number of output layer neurons is 1; there are two hidden layers, the number of neurons in the first layer is 5, and the number of neurons in the second layer is 4; the activation function uses the tanh activation function; the loss function uses MSE; the initial weight is randomly generated using a Gaussian distribution with a mean of 0 and a standard deviation of 0.01; the initial bias value is zero; the learning rate is set to 0.001; and the precision is set to three decimal places.

[0117] In this embodiment, the glue liquid refractive index calculation model is constructed based on the BP neural network algorithm. Specifically, firstly, a glue liquid refractive index calculation model is constructed based on the BP neural network algorithm, and then the thin film evaporator motor speed, the four-zone temperature of the thin film evaporator, the vacuum pump speed and the insulation water circulation pump outlet temperature in the historical screening and sorting process parameter data are used as the input features of the glue liquid refractive index calculation model training set, and the glue liquid refractive index in the historical screening and sorting intermediate parameter data is used as the target output of the glue liquid refractive index calculation model training set; the above data are used to train, test and verify the model, and finally a trained glue liquid refractive index calculation model is obtained.

[0118] It can be understood that the number of input layer neurons of the glue refractive index calculation model is set to 8, and the number of output layer neurons is set to 1; there are two hidden layers, the number of neurons in the first layer is 6, and the number of neurons in the second layer is 4; the activation function is the tanh activation function; the loss function is MSE; the initial weight is in the range of -1 to 1; the initial bias value is zero; the learning rate is set to 0.001; and the precision is set to four decimal places.

[0119] In this embodiment, the glue-making time consumption calculation model is constructed based on the least squares method to fit the nonlinear parameter algorithm. Specifically: first, a glue-making time consumption calculation model is constructed based on the least squares method to fit the nonlinear parameter algorithm, and then the porridge weight, thin film evaporator motor speed and vacuum pump speed in the historically screened and sorted process parameter data are used as the input features of the training set of the glue-making time consumption calculation model, and the consumption time in the historically screened and sorted intermediate parameter data is used as the target output of the training set of the glue-making time consumption calculation model; finally, the above data is used to train, test and verify the constructed model, and finally a trained glue-making time consumption calculation model is obtained. The glue-making time consumption calculation model is as follows: ; In the formula, It takes time to make glue; and These are the parameters to be fitted; is the motor speed of the thin film evaporator; is the vacuum pump speed; It is the weight of porridge.

[0120] It is understandable that each time a new process sheet is input, fitting needs to be performed once. The fitting method is a well-known technique in the art and will not be described in detail here. and The initial value of is 1, otherwise it is initialized to the parameters of the previous fitting.

[0121] In this embodiment, for the construction of the early warning module: the early warning module is constructed based on the early warning threshold information of the key equipment status. That is, the early warning module stores the early warning threshold information of the key equipment status internally, and compares and warns according to the equipment status prediction value transmitted by the key equipment status prediction module and the internally stored threshold information. For example, when the early warning module determines that the remaining life of the current target thin film evaporator is lower than the threshold, it is displayed in red in the device model corresponding to the target thin film evaporator in the three-dimensional digital twin model, and the predicted remaining life value is displayed synchronously to achieve the effect of early warning; if it is determined that the remaining life of the current target thin film evaporator is higher than the threshold, it is displayed in green in the device model corresponding to the target thin film evaporator in the three-dimensional digital twin model. The above early warning method facilitates operators to quickly locate the location of faulty equipment and implement equipment replacement and maintenance plans as soon as possible.

[0122] In this embodiment, a process management module is constructed based on the process database, and the process management module can be used for writing, managing and issuing process documents.

[0123] Based on the various functional modules constructed in the above manner, the 3D modeling is finally associated with the digital twin module, the key equipment status prediction module, the energy consumption prediction module, the fault prediction module and the process management module according to the signal relationship, and the connection relationship and interaction logic between the functional modules are established to finally construct the intelligent simulation system of the present invention.

[0124] After the intelligent simulation system is constructed, the coordination among the above-mentioned functional modules can realize intelligent control of the whole process of Lyocell fiber production, which can be used to guide the production of Lyocell fibers, help enterprises to better control costs, and significantly improve the economic benefits of Lyocell production.

[0125] Example 2 On the basis of Example 1, the remaining life prediction submodule of the thin film evaporator can be used to predict the remaining life of the target thin film evaporator. The overall working principle and process are as follows.

[0126] For the remaining life prediction of thin film evaporator, the specific steps are as follows: Step A1: collect real-time vibration data collected by 6 vibration sensors at the upper and lower ends and the middle area of ​​the target thin film evaporator cylinder through a vibration data acquisition unit, and obtain real-time spindle speed from the central control platform of the target thin film evaporator.

[0127] The real-time vibration data includes acceleration signals in three directions collected by 6 vibration sensors arranged at the upper and lower ends and the middle area of ​​the thin film evaporator cylinder, totaling 18 acceleration signals, where the three directions refer to the x-axis, y-axis and z-axis directions; among them, the vibration sensor arranged at the upper end of the cylinder is used to monitor the fault condition of the upper end bearing of the main shaft, the vibration sensor arranged at the lower end of the cylinder is used to monitor the fault condition of the lower end bearing of the main shaft, and the 4 vibration sensors arranged in the middle area of ​​the cylinder are used to monitor the fault condition of the main shaft.

[0128] Step A2, the vibration data acquisition unit inputs the above 18 acceleration signals into the vibration data preprocessing unit, processes the collected data based on the internally stored vibration data preprocessing calculation template, and obtains the root mean square, kurtosis and frequency center of gravity of each real-time vibration data segment, as well as the fault characteristic frequency of the main shaft and the fault characteristic frequency of the main shaft bearing (the specific data processing process can refer to step S2).

[0129] Step A3, then the vibration data preprocessing unit inputs the root mean square, kurtosis, frequency center of gravity, main shaft fault characteristic frequency and main shaft bearing fault characteristic frequency obtained in step A2 into the remaining life prediction output unit, predicts the remaining life of the target thin film evaporator through the internally constructed and trained remaining life prediction model, and finally outputs the remaining life of the target thin film evaporator, and sends the life prediction result to the early warning module of the system. The specific predicted remaining life includes the remaining life of the main shaft and / or the remaining life of the main shaft bearing.

[0130] like Figure 2 As shown, multiple vibration sensors distributed in preset areas on the surface of the thin film evaporator cylinder collect vibration data of the upper and lower bearings of the main shaft and the vibration data of the main shaft in real time.

[0131] Figure 2 The six sensors are respectively the first vibration sensor 1 for collecting vibration data of the upper bearing in the upper end area of ​​the thin film evaporator, the sixth vibration sensor 6 for collecting vibration data of the lower bearing in the lower end area, and the second vibration sensor 2, the third vibration sensor 3, the fourth vibration sensor 4 and the fifth vibration sensor 5 for collecting vibration data of the main shaft in the middle area. In implementation, as a preferred embodiment, the first vibration sensor 1 is installed on the cylinder in the area corresponding to the upper bearing and the upper mounting plate, and the second vibration sensor 2 is installed on the cylinder in the area corresponding to the lower bearing and the lower mounting plate. The vibration data includes 18 acceleration signals in three directions collected by a total of 6 vibration sensors at the upper and lower ends and the middle area of ​​the cylinder of the thin film evaporator; among them, the vibration sensor at the upper end of the cylinder is used to monitor the fault condition of the upper end bearing of the main shaft, the vibration sensor at the lower end of the cylinder is used to monitor the fault condition of the lower end bearing of the main shaft, and the 4 vibration sensors in the middle area of ​​the cylinder are used to monitor the fault condition of the main shaft.

[0132] The vibration data obtained by the vibration sensor is preprocessed through the thin film evaporator vibration data preprocessing calculation template to obtain the root mean square, kurtosis and frequency center of gravity, and the fault characteristic frequency of the bearing and the fault characteristic frequency of the main shaft are calculated. After normalization, these data are input into the trained remaining life prediction model, and the remaining life prediction model predicts the main shaft and the main shaft bearing, and outputs the remaining life of the main shaft and the main shaft bearing; the remaining life prediction result is the remaining life percentage of the equipment.

[0133] The remaining life prediction model of this embodiment predicts the remaining life percentage of the target thin film evaporator, and this prediction result is of great significance for equipment management and production scheduling. In terms of equipment management, the maintenance team can prepare the corresponding maintenance tools, parts and components in advance and deploy professional maintenance personnel based on the remaining life prediction results. In this way, when equipment failure occurs, it can respond quickly, reduce equipment downtime, and reduce production interruption losses caused by equipment failure. For production scheduling, production management personnel can reasonably adjust the production plan according to the remaining life of the equipment to avoid affecting product delivery due to equipment failure.

[0134] In addition, through the long-term accumulation and analysis of equipment remaining life prediction data, we can also discover the potential patterns of equipment failure. For example, in a specific production environment or operating conditions, the remaining life of the equipment decays faster and the frequency of failures is higher. Based on these findings, we can further optimize the equipment maintenance strategy and production process, extend the service life of the equipment, and improve the stability and reliability of production.

[0135] Example 3 On the basis of Example 1 or Example 2, the key equipment status prediction module of this embodiment may also include a health prediction submodule of a thin film evaporator, wherein the health prediction submodule of the thin film evaporator is used to obtain the current real-time production parameters of the target thin film evaporator and the current real-time production parameters of the upstream pulper, and then pre-process the collected data, including the step of removing outliers and normalization processing; then based on the expert scoring method, weight the collected parameters from three dimensions: the degree of parameter fluctuation, the degree of influence of the parameters on the health status of the equipment, and the severity of the deviation of the parameters from the normal range, and calculate the sum of the products of each dimension of each parameter and the corresponding weight to obtain the influence weight of each parameter in the health prediction; finally, the weighted data of each parameter is input into the trained health prediction model, the health of the target thin film evaporator is predicted, and the health index of the target thin film evaporator is output; wherein the real-time production parameters of the thin film evaporator include evaporation chamber pressure, stirring motor current, feed flow rate, feed pipeline pressure, dissolving liquid temperature, solution viscosity and solvent recovery rate; the real-time production parameters of the pulper include discharge flow rate, slurry concentration and slurry temperature.

[0136] For the health prediction submodule of the above-mentioned thin film evaporator, the working principle of the thin film evaporator is to make the pulp flow in a film along the wall of the heating tube, evaporate the water therein, dissolve the cellulose, and thus prepare the spinning dope. The factors affecting the operation status of the thin film evaporator not only depend on the equipment parameters itself, but also are closely related to the internal product parameters of the equipment. For example, from the perspective of equipment parameters, the current fluctuation of the stirring motor of the thin film evaporator can reflect the load condition of the stirring motor, and abnormalities may indicate motor overload or failure; the feed pipeline pressure reflects the fluid flow state, and abnormalities may indicate blockage or pressure loss, etc. From the perspective of the internal product parameters of the equipment, the temperature of the dissolving liquid of the thin film evaporator directly affects the dissolution efficiency of cellulose. When the temperature is abnormal, it may cause uneven slurry, thereby affecting the entire process; the pressure of the evaporation chamber affects the evaporation rate of the solvent, and abnormal pressure may cause evaporation failure or vacuum system failure, etc. In addition, as a downstream device of the pulper, the operation status of the thin film evaporator is also affected by the operation status of the pulper. For example, the pulper's slurry concentration will affect the dissolution uniformity, and too high a concentration may cause blockage; the pulper's slurry temperature directly affects the feed temperature of the thin film evaporator; the pulper's discharge flow rate reflects the pulper's discharge stability, affecting the continuity of subsequent processes, etc.

[0137] However, the current conventional equipment health prediction methods are mostly limited to the equipment level, relying only on the equipment's own operating data (such as vibration, current, pressure, etc.) for health prediction, and lacking process-level health correlation analysis. For the equipment on the Lyocell fiber production line, since its operating status is affected by many factors, it is particularly easy for short-term instantaneous process changes to cause certain parameters to exceed the limit warning, resulting in false alarms, causing unnecessary downtime, and delaying production progress. For example, assuming that the operating temperature of the thin film evaporator equipment exceeds the normal range, the equipment data alone may be judged as a fault, but combined with the process data analysis, it is found that it is a normal phenomenon caused by the current production formula adjustment, and there is no need for a false alarm warning. Therefore, due to the limited prediction accuracy, the prediction system that lacks process-level analysis is prone to misjudge normal conditions as faults, or fail to capture real potential faults.

[0138] Therefore, the thin film evaporator health prediction submodule of the present invention mainly predicts the health of the thin film evaporator by combining the data at the equipment level and the process level for fault correlation analysis, and finally outputs the health index of the equipment, which improves the prediction accuracy and reliability, optimizes production scheduling, and improves the overall production efficiency.

[0139] Therefore, the construction and training process of the health prediction model involved in the health prediction submodule has the following steps: Step D1, collecting production parameters related to the operating status of the thin film evaporator, including the equipment parameters of the thin film evaporator and the internal product parameters of the equipment; specifically, the collected equipment parameters of the thin film evaporator include the evaporation chamber pressure, the stirring motor current, the feed flow rate, and the feed pipeline pressure, and the internal product parameters of the thin film evaporator include the solution temperature, the solution viscosity, and the solvent recovery rate.

[0140] The influence of the above parameters on the operating status of the thin film evaporator is as follows: Evaporation chamber pressure (Pa) - The evaporation chamber pressure affects the evaporation rate. Too low pressure will lead to too high solution concentration, affecting NMMO recovery; too high pressure may cause equipment leakage.

[0141] Stirring motor current (A) - The stirring motor current affects the uniformity of the slurry. Too high a current may cause motor failure due to increased load, while a current below the normal range may indicate motor aging or low viscosity.

[0142] Feed flow rate (m³ / h)——The feed flow rate directly affects the stability of the evaporation and dissolution process. Low flow rate may mean that the pipeline is blocked or the pump is operating abnormally, and too high flow rate may cause solution dilution and affect evaporation efficiency.

[0143] Feed pipeline pressure (Pa) - Feed pipeline pressure reflects the fluid flow state. Abnormalities may indicate blockage or pressure loss. It has a more indirect impact on thin film evaporators, but can be used to monitor abnormalities.

[0144] Dissolving liquid temperature (°C) - The dissolving liquid temperature directly affects the dissolving efficiency of cellulose. Abnormal temperature may lead to uneven slurry, which in turn affects the entire process.

[0145] Solution viscosity (Pa·s)——The solution viscosity directly affects the spinning quality. If it is too low, it may mean that the solution concentration is insufficient, affecting the quality of the finished product; if it is too high, it may cause equipment blockage.

[0146] Solvent (NMMO) Recovery (%) - Solvent concentration affects the final dissolution result. Low recovery may mean a leak or vacuum system failure.

[0147] Furthermore, as a downstream device of the pulper, the operation state of the thin film evaporator is also affected by the operation state of the pulper. For example, when the viscosity of the pulper slurry is abnormal, it may affect the dissolution efficiency of the thin film evaporator, causing the viscosity of the solution to fluctuate, thereby affecting the subsequent spinning quality. In order to facilitate data collection, no additional collection equipment is added to the feed end of the thin film evaporator. In this embodiment, the production parameters of the pulper are directly included in the data collection stage. The production parameters of the pulper also include the equipment parameters of the pulper and the internal product parameters of the equipment.

[0148] Specifically, the equipment parameters of the pulper collected in this embodiment include the discharge flow rate, and the internal product parameters of the pulper include the pulp concentration and the pulp temperature. The influence of the above parameters on the operation state of the thin film evaporator is as follows: Discharge flow rate (m³ / h)——The discharge flow rate of the pulper reflects the discharge stability of the pulper and affects the continuity of subsequent processes.

[0149] Slurry concentration (%) - affects dissolution uniformity, too high a concentration may cause blockage.

[0150] Slurry Temperature (°C) – affects the feed temperature to the thin film evaporator.

[0151] Finally, the collected parameters related to the health prediction of the thin film evaporator are shown in Table 1 below.

[0152]

[0153] When collecting the above data, the timestamp synchronization method is used to uniformly add accurate timestamps to all data to ensure the timing alignment of different data sources. It should be noted that since the change in pulp quality of the pulper will not immediately affect the thin film evaporator, but there is a lag of about 10 minutes, the parameters of the pulper should be compensated in advance according to the material transportation time (using data about 10 minutes ahead of the thin film evaporator) to match the operating status of the thin film evaporator. As shown in Table 2 below.

[0154]

[0155] Step D2: preprocessing the collected original production parameter time series data, including removing outliers and normalizing.

[0156] In this embodiment, a combination of physical constraint method and statistical method is used to eliminate outliers. The physical constraint method refers to setting parameter thresholds based on the operating range of the equipment. For example, the pressure in the evaporation chamber of a thin film evaporator should be between 400Pa and 600Pa, and it is considered abnormal if it exceeds the range. The statistical method refers to eliminating outliers based on the distribution of historical data using the 3σ principle.

[0157] In this embodiment, the purpose of data normalization is to scale the data to a uniform range, such as [0, 1], to eliminate the influence of different dimensions. Normalization processing is a well-known technology in the art and will not be described in detail here.

[0158] The data in the original production parameter time series data table are processed by the above normalization formula to obtain the normalized production parameter data table shown in Table 3 below.

[0159]

[0160] Step D3: Combining expert knowledge and experience, weight the collected parameters from three dimensions: the degree of production parameter fluctuation, the degree of impact of production parameters on equipment health status, and the severity of parameter deviation from the normal range, so as to further improve the accuracy of the prediction results.

[0161] The degree of fluctuation of production parameters is determined according to the following method: Calculate the standard deviation of each production parameter ( σ ), the larger the standard deviation, the greater the fluctuation of production parameters, which may have a stronger impact on the equipment status.

[0162] In order to ensure that the sum of the volatility contribution weights of all production parameters is 1, the normalized standard deviation is used to measure the parameter volatility intensity. The calculation formula is as follows: ; In the formula, Indicates i Normalized standard deviation of production parameters; Indicates i The standard deviation of the production parameters, is the sum of the standard deviations of all production parameters.

[0163] The degree of influence of production parameters on equipment health status is determined according to the following method: In this embodiment, the influence of production parameters on the health status of equipment is measured by expert scoring. Combining the knowledge and experience of experts, each parameter is ranked and scored (0-10 points). The higher the score, the greater the influence of the parameter on the health status of the equipment.

[0164] The importance ranking and scores of the above production parameters can be found in Tables 4-1 and 4-2 below.

[0165]

[0166]

[0167] It should be noted that although the pulp concentration and feed flow rate parameters related to the pulper equipment in Table 4-1 and Table 4-2 are ranked at the bottom in importance, based on the physical mechanism analysis and the actual production process impact (combined with expert knowledge and experience), the scores set are higher than the parameters ranked higher in importance.

[0168] Based on the above scores, the normalized expert scores of each parameter are obtained, and the calculation formula is as follows: ; In the formula, Indicates i Normalized expert scores for production parameters; For the i Expert ratings of production parameters, For the total score.

[0169] The severity of a parameter deviation from the normal range is determined as follows: Set the physical constraint weight, taking into account the severity of the parameter deviation from the normal range. The calculation formula is as follows: ; In the formula, Indicates i The severity of the deviation of a production parameter from the normal range; For the i The number of times a production parameter exceeds the normal range, is the total number of samples of production parameters.

[0170] ① Determine the final weight of each production parameter under the influence of the above three dimensions.

[0171] The final weight of each production parameter considering the influence of the above three dimensions Determine according to the following formula: ; In the formula, Indicates i The weight of each production parameter; and

[0172] , , .

[0173] In this embodiment, it can be set .

[0174] In this step, the method of combining standard deviation + expert experience score + physical constraints is used to assign weights to each parameter, which can ensure that the importance of key parameters in the prediction model is reasonably reflected. When the early data is insufficient, more reliance is placed on expert experience to determine the weight value of each dimension (i.e. α 1. α 2. α 3); after data accumulation, parameter weight allocation can be more dependent on data-driven.

[0175] Step D4: construct a production parameter feature sequence based on the time series data after the weights of various production parameters to obtain training sample data for the health prediction model.

[0176] Determine the sliding window size and step size, and convert the above time series data into a sliding window of fixed length. The sliding window method divides the continuous data into subsequences of fixed length to facilitate the model to learn time dependencies. Each input sliding window includes time series features (solvent temperature, evaporation chamber pressure, etc.) and target labels (normal / faulty). The target label data is obtained from long-term historical data, mainly relying on equipment maintenance records.

[0177] By constructing a feature sequence in the above manner, training sample data for a health prediction model is obtained.

[0178] Step D5: Input the obtained training sample data into the constructed LSTM model architecture for training, use multiple rounds of iterations to optimize the model parameters, and use an early stopping strategy to prevent overfitting.

[0179] <1> Model architecture design Since the original data is a continuous time series that is continuously collected, LSTM needs to input a fixed-length time window for training. When training LSTM, each sample is a (w3, F n ) matrix, w3 represents the window size, F n The total number of features at each step size.

[0180] The model architecture includes: Input layer: accepts time series data with a shape of (step length, number of features).

[0181] LSTM layer: In this embodiment, a single-layer LSTM is used with 64 hidden units to capture long-term dependencies in the time series.

[0182] Fully connected layer: used to generate prediction results.

[0183] <2> Loss Function and Optimizer Classification task: Cross Entropy Loss (Binary Cross-Entropy).

[0184] Loss function: ; In the formula, is the real fault label, 0=normal, 1=fault; The failure probability predicted by the LSTM model; N 2 is the total number of samples.

[0185] For regression tasks: mean absolute error (MAE) or mean squared error (MSE).

[0186] Optimizer: Adam optimizer with a learning rate of 0.001.

[0187] The above loss function selected in this embodiment is suitable for the binary classification problem. P fault Deviation from the true label , the loss will become larger, guiding the gradient adjustment.

[0188] <3> Model Training The training sample data obtained in step D4 is input into the constructed model framework, and the training sample data is divided into a training set and a validation set in a ratio of 80: 20. The model parameters are optimized through multiple rounds of iteration (set to 100 rounds in this embodiment).

[0189] Specifically, in this embodiment, an early stopping strategy is used to prevent overfitting. The stopping training strategy is as follows: The training loss converges, that is, when the training loss changes for multiple consecutive rounds (set to 10 rounds in this embodiment) are less than a certain threshold ϵ (the threshold is set to 10 in this embodiment) -4 ), then stop training, the formula is as follows: ; in, Represents the loss of this round of training, is the loss of the previous round of training.

[0190] When the training loss tends to be stable, it means that LSTM has learned the optimal parameters and is no longer optimized significantly.

[0191] Step D6: Finally, the model accuracy is evaluated and verified using evaluation indicators to finally obtain a trained prediction model.

[0192] In this embodiment, the model evaluation index may be precision, recall, F1 score, mean square error (MSE) or mean absolute error (MAE), etc. The model validation method may be time series cross validation, or the model performance may be tested using actual device operation data.

[0193] After the health prediction model is constructed and trained based on the above steps, the health prediction submodule of the thin film evaporator can predict the health of the target thin film evaporator based on the trained health prediction model. The specific process is as follows: The real-time production parameters of the target thin-film evaporator and the real-time production parameters of the upstream pulper are obtained (the collected real-time production parameters are the same as those in Table 1), and then the collected real-time data are processed in the same data processing method as in the model training stage of steps D1 to D4. The processed data are then input into the trained health prediction model, and finally the current health index of the target thin-film evaporator can be obtained.

[0194] LSTM processes data through multiple gates (forget gate, input gate, output gate) and finally calculates the failure probability at the output layer P fault The last time step of the LSTM outputs the hidden state ℎ 𝑇 , and then calculate the failure probability through the fully connected layer (Dense Layer): P fault = σ ( W h* h T + b h ); In the formula, W h and b h Parameters learned for LSTM training; h T The hidden state after LSTM processes the last time step; σ ( W h* h T + b h ) represents the Sigmoid function.

[0195] In this embodiment, the Sigmoid function is selected as the activation function because the output range of the function is (0, 1), so it is suitable for binary classification problems and represents the probability of equipment failure.

[0196] In this embodiment, the failure probability of the target thin film evaporator is output through the LSTM model, and then the health prediction submodule calculates the health index (Health Index) of the target thin film evaporator according to the failure probability output by the model. If the health index is lower than the set safety threshold (in this embodiment, the safety threshold is set to 0.3), it is determined that the equipment has a potential failure. The health index formula is as follows: ; in, is the health index of the thin film evaporator; is the predicted failure probability of the thin film evaporator; if >0.7 (more than 70%), indicating that the equipment has a high probability of failure and requires maintenance; if <0.3 (less than 30%), indicating that the device is in good health; if Gradually increasing, indicating that the equipment is gradually degrading. Gradually decrease.

[0197] Based on the above fault probability prediction value and health index, the health prediction submodule of the thin film evaporator outputs the fault probability prediction value and health index to the early warning module, and the early warning module determines whether to trigger the early warning information according to the health index of the equipment. For example, when the early warning module determines that the health index of the current target thin film evaporator is lower than the threshold, it will be displayed in red and the predicted health index value will be displayed synchronously in the device model corresponding to the target thin film evaporator in the three-dimensional digital twin model to achieve the effect of early warning; if it is determined that the health index of the current target thin film evaporator is higher than the threshold, it will be displayed in green in the device model corresponding to the target thin film evaporator in the three-dimensional digital twin model.

[0198] It should be noted that the health index only reflects the current operating status of the equipment and does not indicate a specific fault.

[0199] Finally, the early warning module can also output the current equipment status and guiding maintenance suggestions on the operation display interface according to the health index output by the health prediction model, referring to the fault warning and maintenance suggestions, as shown in Table 5 below.

[0200]

[0201] The following is a comparative example to illustrate the health prediction effect of the health prediction submodule of the present invention on the thin film evaporator.

[0202] On a certain day, the DCS system monitored the real-time production parameters of the target thin film evaporator over a period of time as shown in Table 6 below: Table 6 Real-time production parameters of target thin film evaporator

[0203] The physical constraint ranges of various real-time production parameters are shown in Table 7 below:

[0204] According to the physical constraint range of the parameters in Table 7, the device-level data in Table 6 has the following anomalies: The stirring motor current exceeds the upper limit of 40A; The evaporation chamber pressure is close to the upper limit of 600Pa; The power of the stirring motor exceeds the upper limit of 3.5KW; The discharge flow rate of 30 m³ / h exceeds the limit.

[0205] At the same time, the DCS system gave a warning message, believing that the current mixing system may have an overload problem and recommended shutting down for inspection; the vacuum of the evaporation system was too high and there might be a risk of leakage; the discharge flow was too high and there might be problems with pump delivery.

[0206] At this time, the process level data is analyzed: The slurry concentration is in the normal range of 8 ~ 12%, indicating that the slurry concentration control is stable and not too concentrated or too diluted. There is no problem of excessive solvent volatilization, eliminating the risk of vacuum system leakage.

[0207] The solution viscosity is in the normal range of 4.5 ~ 6.5 Pa·s, indicating that the stirring load is normal and no additional maintenance is required.

[0208] The feed flow rate is in the normal range of 40 ~ 50 m³ / h, indicating that the feed is stable and the feed abnormality is eliminated.

[0209] The solvent recovery rate is within the normal range of 98-99%, indicating that the solvent recovery system is operating normally without obvious leakage or excessive volatilization. The pipeline is well sealed and the solvent loss is controlled within a reasonable range.

[0210] The feed pipeline pressure is in the normal range of 48000 ~ 56000 Pa, indicating that the fluid is transported normally without blockage or leakage. There is no need to adjust the flow or check the pipeline, and the system runs smoothly.

[0211] The slurry temperature is in the normal range of 55 ~ 65°C, which means that the temperature is stable and suitable for the evaporation dissolution process, and will not cause insufficient dissolution of the slurry.

[0212] The real-time production parameters of the target thin film evaporator in Table 6 above are input into the health prediction model constructed and trained in this embodiment, and the LSTM outputs the failure probability. P fault The device health index is 0.18. HI The corresponding maintenance recommendation is to maintain the current operation and monitor key parameters.

[0213] At the same time, according to the maintenance records of the day, the operation and maintenance team arrived at the site to troubleshoot the problem, and found no abnormalities. And 10 minutes after the fault alarm, the current of the stirring motor dropped to 36A, the pressure dropped to 550 Pa, and the stirring power recovered to 3.2kW, all returning to the normal range. By integrating the process-level data analysis, it was confirmed that the short-term fluctuations in the equipment-level parameters were caused by the feed adjustment, and no shutdown for maintenance was required. The correctness of the prediction results of this scheme was verified.

[0214] Example 4 Based on Example 1, this example further explains and illustrates the energy consumption prediction module. In this example, the energy consumption prediction module mainly predicts the energy consumption of the key rubber making process in the production of lyocell fibers. Figure 5 This is a data processing architecture diagram of the energy consumption prediction module of the present invention. The specific workflow of the energy consumption prediction module for energy consumption prediction is as follows: Step E1, the process sheet receiving unit in the energy consumption prediction module obtains the process sheet corresponding to the glue making stage through the process management module of the system, and then transmits the obtained process sheet to the process sheet data extraction unit; Step E2, the process sheet data extraction unit extracts the porridge weight, process parameter group 1, process parameter group 2 and process parameter group 3 corresponding to the current process sheet according to the current process sheet; then the current porridge weight and process parameter group 1 are input into the internally stored and trained glue weight calculation model to comprehensively obtain the glue weight; the process parameter group 2 is input into the internally stored and trained glue refractive index calculation model to calculate the glue refractive index; and the current porridge weight and process parameter group 3 are input into the internally stored and trained glue consumption time calculation model to comprehensively obtain the glue consumption time; Step E3, the process sheet data extraction unit inputs the obtained porridge weight, glue liquid weight, glue liquid refractive index and glue making time into the energy consumption prediction model of the energy consumption prediction unit, and the energy consumption prediction model outputs the energy consumption prediction value corresponding to the current glue making process sheet.

[0215] In the embodiment described in the present invention, the energy consumption data is the sum of electricity energy consumption and steam energy consumption; the specific calculation formula for converting steam energy consumption into electricity energy consumption is as follows: ; In the formula, The electricity consumption for steam; is the price of steam, in yuan / m 3 ; is the price of electric energy, in Yuan / KW·h; is the amount of steam used, in m 3 .

[0216] In the embodiment described in the present invention, the process parameter group 1 includes the motor speed of the thin film evaporator, the temperature of the four zones of the thin film evaporator, the vacuum system pressure and the vacuum pump speed; the process parameter group 2 includes the motor speed of the thin film evaporator, the temperature of the four zones of the thin film evaporator, the vacuum pump speed and the outlet temperature of the insulation water circulation pump; the process parameter group 3 includes the motor speed of the thin film evaporator and the vacuum pump speed.

[0217] After the above steps E1~E2, the porridge weight, glue liquid weight, glue liquid refractive index and glue making time corresponding to the current glue making process can be obtained. Based on the above four indicators, the energy consumption prediction model trained in advance and stored in the energy consumption prediction unit can finally output the energy consumption prediction result corresponding to the current glue making process.

[0218] Example 5 In the production and manufacturing of lyocell, the cost of raw materials and the cost of equipment maintenance are key factors affecting the profit margin and market competitiveness of enterprises. The fluctuation of raw material costs and equipment maintenance costs caused by the raw material consumption data involved in the raw material costs have led to excessive fluctuations in process consumption costs. If key factors such as raw material costs and equipment maintenance costs cannot be accurately predicted, and because the fluctuations in process consumption costs are too large, the environmental protection costs in the process consumption costs, such as recycling costs, cannot be confirmed, making it impossible for enterprises to carry out refined management of the costs involved in the entire process production and manufacturing.

[0219] Therefore, based on any of the above embodiments, this embodiment also proposes a cost prediction module in the intelligent simulation system to provide a scientific basis for process optimization and resource scheduling, and to assist enterprises in conducting refined cost management in the Lyocell production and manufacturing process.

[0220] The cost prediction module of this embodiment is composed of a raw material cost prediction submodule and an equipment maintenance cost prediction submodule. Figure 6 The data processing architecture diagram of the cost prediction module is shown in FIG. 1 . The two submodules implement cost prediction in a similar manner. This embodiment takes the raw material cost prediction submodule as an example to introduce the workflow of the raw material cost prediction submodule.

[0221] The specific steps for predicting the raw material cost of Lyocell fiber production process are as follows: Step F1, obtaining raw material data in the current lyocell production process, including wood pulp parameter data, NMMO solution parameter data, and chemical additive parameter data; the wood pulp parameter data includes wood pulp viscosity, wood pulp polymerization degree and wood pulp methyl fiber content; the NMMO solution parameter data includes NMMO solution concentration; the chemical additive parameter data includes propylene glycol content and hydroxylamine content.

[0222] In step F1, the raw material data also includes raw material purchase price data, raw material consumption data, and raw material inventory data; It can be understood that, for the above raw material data, the wood pulp viscosity, wood pulp polymerization degree and wood pulp methyl fiber content are monitored and collected in real time through online viscometers and optical measuring instruments; 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 raw material purchase price data, raw material inventory data, and raw material consumption data are collected through the enterprise resource planning system and synchronized with the external market database.

[0223] Step F2: preprocess the raw material data. The specific preprocessing process is as follows: Step a: Detect outliers in the raw material data and remove them; then process missing values ​​and remove duplicate data; for the outlier check of numerical data in the raw material data, 3 σ The principle is to detect outliers in the data; for outlier detection of non-normally distributed data in raw material data, the interquartile range method is used to detect outliers and calculate the first quartile Q 1 and the third quartile Q 3. The set exception range is as follows: ; ; Data outside this range are considered outliers and are removed; where: IQR The meaning of is the interquartile range; Furthermore, for the processing of missing values ​​of raw material data, the specific processing method is as follows: time series data is filled by linear interpolation method; if the missing ratio of numerical data is less than 20%, k-nearest neighbor interpolation is used to fill it, if the missing ratio exceeds 20%, the sample is eliminated; 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 exceeds 20%, the sample is eliminated.

[0224] Step b: Use Min-Max to normalize the raw material data. The calculation method is as follows: ; In the formula, x norm is the normalized value; x 原始 is the original data; 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, and the calculation method is as follows: ; In the formula, μ is the mean of the original data; σ is the standard deviation, x scaled It is the data value after Z-score standardization.

[0225] Step c: Pearson correlation analysis and principal component analysis are used to extract key features of raw material data, as follows: Step c1, calculate the relevant features between the same category data and different category data, analyze the linear correlation between variables, and then screen redundant features, calculate the Pearson correlation coefficient between features, and specifically calculate the Pearson correlation coefficient between the following features: within the raw material data, analyze the correlation between wood pulp viscosity, wood pulp polymerization degree, and wood pulp methyl fiber content to determine their linear dependence in the data set; at the same time, analyze the correlation between NMMO solution concentration, propylene glycol content, and hydroxylamine content to evaluate the impact of chemical additives on the production process; in addition, it is necessary to calculate the correlation between raw material inventory data, raw material purchase price data, and raw material consumption data to analyze the impact of inventory management on cost.

[0226] The specific calculation method is as follows: ; In the formula, r is the Pearson correlation coefficient, h i For variables h No. i Observations, H i For variables H No. i Observations, For variables h The mean of For variables H The mean of When r When |>0.8, the two features are considered highly correlated and one of them is removed; Step c2: Perform principal component analysis to reduce dimensionality. First, construct a standardized data matrix G , the dimension is m G × n G ,in m G is the number of samples, n G is the characteristic number, and then calculate the covariance matrix V , calculated as follows: ; In the formula,V is the covariance matrix; G T The standardized data matrix G The transposed matrix of Step c3, solve the covariance matrix V The eigenvalues ​​and eigenvectors of V The eigenvalue of λ 1, λ 2, ..., λn , sort by eigenvalue from large to small, and calculate the cumulative contribution rate; select the top ones with a cumulative contribution rate exceeding 95%. q The eigenvectors form a dimension reduction projection matrix W , through the dimension reduction projection matrix W Obtain the projection data, and finally reduce the dimension of the data after principal component analysis from n G The dimension is reduced to q Dimension; 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% is used as an empirical threshold to ensure that most information is retained, while significantly reducing the data dimension and improving the generalization ability and computational efficiency of the model; The specific formula for obtaining the projection data is as follows: ; In the formula, W is the dimension reduction projection matrix, G′ is the feature data after dimensionality reduction.

[0227] Step d: Use K-means cluster analysis and information entropy screening methods to perform final screening of raw material data to extract key features and remove redundant data, thereby improving the model's computational efficiency and prediction accuracy. In terms of raw material data, the viscosity of wood pulp, degree of polymerization of wood pulp, methyl fiber content of wood pulp, concentration of NMMO solution, propylene glycol content, hydroxylamine content, raw material purchase price data, raw material inventory data, and raw material consumption data are screened to determine which raw material parameters have the greatest impact on production costs and energy consumption.

[0228] The specific process is as follows: Step d1: Use K-means cluster analysis, initialize first, and set the number of clusters l =5, select l A random sample is used as the initial cluster center, and then the Euclidean distance from the sample to the cluster center is calculated as follows: ; In the formula,d ( x , c ) means sample point x With cluster center c The Euclidean distance between them is a measure of the similarity between the data point and the cluster center. The smaller the distance, the closer the data point is to the cluster center. x i For the i Point data samples, C i For the i The new cluster centers of the clusters, for n dimensional space; Step d2: Reassign clusters, calculate the distance between each sample point and all cluster centers, and assign the sample point to the nearest cluster; Step d3, 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 iterating; the calculation method of the new cluster center is: ; In the formula, C i For the i The new cluster centers of the clusters, C i ∣ is a cluster C i The number of internal sample points; Step d4, 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 d5: Although the most representative samples have been selected through the intra-cluster distance in step d4, 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. H ( C i ), calculated as follows: ; In the formula, P B Cluster C i Medium Category B probability; M is the number of categories within the cluster; Step d6: 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; finally, obtain the preprocessed raw material data.

[0229] Step F3: input the raw material data preprocessed in step F2 into the trained raw material consumption prediction model, and output the raw material consumption prediction values ​​corresponding to different machine learning algorithm models.

[0230] The overall raw material consumption prediction model of step F3 is composed of a first raw material consumption prediction sub-model based on random forest, a second raw material consumption prediction sub-model based on support vector machine, and a third raw material consumption prediction sub-model based on long short-term memory network.

[0231] Step F4: weighted fusion of the predicted values ​​of each machine learning algorithm model to obtain the final raw material consumption prediction value, which is calculated as follows: ; In the formula, Indicates the final forecast value of raw material consumption; represents the raw material consumption forecast value of the first raw material consumption forecast sub-model; represents the raw material consumption forecast value of the second raw material consumption forecast sub-model; represents the raw material consumption forecast value of the third raw material consumption forecast sub-model; The meaning of is the contribution of the first raw material consumption prediction sub-model in the weighted fusion, that is, the weight value of the first raw material consumption prediction sub-model; The meaning of is the contribution of the second raw material consumption prediction sub-model in the weighted fusion, that is, the weight value of the second raw material consumption prediction sub-model; The meaning of is the contribution of the third raw material consumption prediction sub-model in weighted fusion, that is, the weight value of the third raw material consumption prediction sub-model; Step F5: Finally, the raw material price and the final raw material consumption forecast are combined to obtain the raw material forecast cost. The calculation method is as follows: ; In the formula, is the total cost forecast of raw materials; The final consumption forecast of wood pulp raw materials; is the unit price of wood pulp raw materials; is the predicted value of raw material consumption of the final NMMO solution; is the unit price of raw materials of NMMO solution; is the final predicted value of propylene glycol raw material consumption; is the unit price of propylene glycol raw material; is the final hydroxylamine raw material consumption value; is the unit price of hydroxylamine raw material.

[0232] In the present invention, the consumption prediction of raw materials includes the final predicted value of wood pulp consumption, the final predicted value of NMMO solution consumption, the final predicted value of propylene glycol consumption and the final predicted value of hydroxylamine consumption.

[0233] Therefore, based on the above prediction process, it can be known that the raw material consumption prediction value finally predicted by the present invention includes the wood pulp consumption prediction value, the NMMO solution consumption prediction value and the chemical additive consumption prediction value; the chemical additive consumption prediction value includes the propylene glycol consumption prediction value and the hydroxylamine consumption prediction value.

[0234] The total raw material cost prediction value finally predicted by the present invention consists of the wood pulp raw material cost prediction value, the NMMO solution raw material cost prediction value, the propylene glycol raw material cost prediction value and the hydroxylamine raw material cost prediction value.

[0235] In the embodiment described in the present invention, the construction and training process of the above three raw material consumption prediction sub-models will be described as follows: (I) constructing a first raw material consumption prediction sub-model for raw material consumption prediction based on random forest, and then collecting historical raw material data in the production process of lyocell, including wood pulp viscosity, wood pulp polymerization degree, wood pulp methyl fiber content, NMMO solution concentration, propylene glycol content, hydroxylamine content, raw material purchase price, raw material inventory data, raw material consumption data, wood pulp consumption, NMMO solution consumption, propylene glycol consumption and hydroxylamine consumption; using the above collected data as the data set of the first raw material consumption prediction sub-model, and processing the data in the same way as the above step F2, dividing the processed data set into a training set, a test set and a validation set, training, testing and validating the constructed first raw material consumption prediction sub-model, and finally obtaining a trained first raw material consumption prediction sub-model; wherein, during model training, wood pulp viscosity, wood pulp polymerization degree, wood pulp methyl fiber content, NMMO solution concentration, propylene glycol content, hydroxylamine content, raw material purchase price, raw material inventory data, and raw material consumption data are used as input features of the training set, wood pulp consumption, NMMO solution concentration, propylene glycol content, hydroxylamine content, raw material purchase price, raw material inventory data, and raw material consumption data are used as input features of the training set, and wood pulp consumption, NMMO solution concentration, propylene glycol content, hydroxylamine content, raw material purchase price, raw material inventory data, and raw material consumption data are used as input features of the training set. The consumption of solution, propylene glycol, and hydroxylamine are used as output features of the training set.

[0236] (II) A second raw material consumption prediction sub-model for raw material consumption prediction is constructed based on a support vector machine. The collected historical wood pulp viscosity, wood pulp polymerization degree, wood pulp methyl fiber content, NMMO solution concentration, propylene glycol content, hydroxylamine content, raw material purchase price, raw material inventory data, raw material consumption data, wood pulp consumption, NMMO solution consumption, propylene glycol consumption and hydroxylamine consumption are processed in the same manner as step F2. The processed data set is divided into a training set, a test set and a validation set. The constructed second raw material consumption prediction sub-model is trained, tested and validated to finally obtain a trained second raw material consumption prediction sub-model. When training the model, wood pulp viscosity, wood pulp polymerization degree, wood pulp methyl fiber content, NMMO solution concentration, propylene glycol content, hydroxylamine content, raw material purchase price, raw material inventory data and raw material consumption data are used as input features of the training set, and wood pulp consumption, NMMO solution consumption, propylene glycol consumption and hydroxylamine consumption are used as output features of the training set.

[0237] (III) A third raw material consumption prediction sub-model for raw material consumption prediction is constructed based on a long short-term memory network. The collected historical wood pulp viscosity, wood pulp polymerization degree, wood pulp methyl fiber content, NMMO solution concentration, propylene glycol content, hydroxylamine content, raw material purchase price, raw material inventory data, raw material consumption data, wood pulp consumption, NMMO solution consumption, propylene glycol consumption and hydroxylamine consumption are processed in the same manner as step F2. The processed data set is divided into a training set, a test set and a validation set. The constructed third raw material consumption prediction sub-model is trained, tested and validated to finally obtain a trained third raw material consumption prediction sub-model. When training the model, wood pulp viscosity, wood pulp polymerization degree, wood pulp methyl fiber content, NMMO solution concentration, propylene glycol content, hydroxylamine content, raw material purchase price, raw material inventory data and raw material consumption data are used as input features of the training set. Wood pulp consumption, NMMO solution concentration, propylene glycol content, hydroxylamine content and raw material consumption data are used as input features of the training set. The consumption of solution, propylene glycol, and hydroxylamine are used as output features of the training set.

[0238] After the above steps (I) to (III), the first raw material consumption prediction sub-model, the second raw material consumption prediction sub-model and the third raw material consumption prediction sub-model can be finally obtained. The three trained sub-models can be used for raw material consumption prediction.

[0239] It should be noted that for the first raw material consumption prediction sub-model, in order to prevent overfitting and underfitting during the training process, the random forest model is composed of 100 decision trees based on experience, and the maximum depth of each tree is set to 15 to prevent overfitting; in the training stage, the Bootstrap sampling method is first used to randomly extract sub-samples from the training data set, that is, m samples are randomly extracted with replacement from the original data set as the training data set for each tree, and each sample may be extracted 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 a node, it randomly selects 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.

[0240] Furthermore, for regression tasks, such as raw material consumption prediction, energy consumption prediction, or equipment maintenance prediction, the mean square error (MSE) is used as the loss function to minimize the error between the predicted value and the true value.

[0241] The final output of the random forest is calculated by integrating multiple decision trees to obtain the final prediction value, which is the weighted average of all decision trees.

[0242] For the second raw material consumption prediction sub-model, during the training process, the optimization goal of support vector regression is to minimize the ε-insensitive loss function, that is, to ensure the generalization ability of the model through the combination of regularization terms and error penalty terms.

[0243] For the third raw material consumption prediction sub-model, during the training process, LSTM inputs the training data into the model for training. The LSTM model structure includes LSTM layer, fully connected layer, optimizer and loss function. The LSTM layer contains 64 LSTM units and uses Tanh activation function to enhance nonlinear expression ability; the fully connected layer includes two layers of neural networks, which are composed of 128 neurons and 64 neurons respectively, and uses ReLU activation function for nonlinear transformation to improve the model fitting ability. During the training process, each batch contains 32 samples (Batch Size=32); forward propagation is used to calculate the output, LSTM processes the input data in sequence through time steps, and finally the prediction value is calculated by the fully connected layer.

[0244] Furthermore, for step F4, the weighted fusion value of each machine learning algorithm model is determined as follows: Assume that the weights of the three prediction sub-models included in the raw material prediction model obey Beta distributed: ; In the formula, The weight coefficient of each prediction sub-model included in the overall model for raw material consumption prediction; α 0 and β 0 is set as the initial parameter, generally: ; In the formula, C 0 is a positive number; According to the mean square errors of the first raw material consumption prediction sub-model, the second raw material consumption prediction sub-model and the third raw material consumption prediction sub-model, the prediction errors of the three raw material consumption prediction sub-models are obtained, and the calculation method is as follows: ; ; ; In the formula, represents the mean square error of the first raw material consumption prediction sub-model; represents the mean square error of the second raw material consumption prediction sub-model; represents the mean square error of the third raw material consumption prediction sub-model; N 3 represents the total number of training samples of the first raw material consumption prediction sub-model, the second raw material consumption prediction sub-model and the third raw material consumption prediction sub-model; Indicates i The true value of raw material data samples; The first raw material consumption prediction sub-model i The predicted value of raw material data samples; The second raw material consumption prediction sub-model i The predicted value of raw material data samples; The third raw material consumption prediction sub-model i The predicted value of raw material data samples; Then update the mean square error Beta distributed: ; ; In the formula, It represents a distribution parameter in the fusion weight calculation of the overall model for raw material consumption forecasting; Represents another distribution parameter in the calculation of the fusion weight of the overall model for raw material consumption forecasting; represents the mean square error of the overall model for raw material consumption prediction; Gaussian process regression is introduced to calculate the uncertainty of the overall model for raw material consumption prediction. The calculation method is as follows: ; ; In the formula, It means the uncertainty of the overall model for raw material consumption prediction; The meaning of is the prediction variance of the total model for raw material consumption prediction on all samples; The total raw material consumption forecast model is i The predicted value of raw material data samples; It represents the average prediction value of all raw material data samples by the overall model for raw material consumption prediction; Introducing uncertainty into the overall model for raw material consumption forecasting Beta distributed: ; ; In the formula, It represents a distribution parameter in the calculation of the fusion weight of the updated raw material consumption forecasting overall model; represents another distribution parameter in the calculation of the fusion weight of the updated raw material consumption forecasting overall model; Finally, variational Bayesian inference is used to calculate the final weights of the overall model for raw material consumption prediction: ; In the formula, The overall model for raw material consumption prediction is t +Model weight after 1 round of iteration; Represents the contribution of actual data; Then, the fusion weights of the three raw material consumption prediction sub-models are calculated respectively: ; ; ; In the formula, is a distribution parameter in the calculation of the fusion weight of the updated first raw material consumption prediction sub-model; is another distribution parameter in the calculation of the fusion weight of the updated first raw material consumption prediction sub-model; It is a distribution parameter in the calculation of the fusion weight of the updated second raw material consumption prediction sub-model; is another distribution parameter in the calculation of the fusion weight of the updated second raw material consumption prediction sub-model; It is a distribution parameter in the calculation of the fusion weight of the updated third raw material consumption prediction sub-model; It is another distribution parameter in the fusion weight calculation of the updated third raw material consumption prediction sub-model.

[0245] In the embodiment depicted in the present invention, for raw material cost forecasting, the time series data includes raw material inventory data and raw material consumption data; 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 raw material purchase price data; The non-normal distribution data includes raw material purchase price data, raw material inventory data, and raw material consumption data.

[0246] In the embodiment described in the present invention, raw material data can be obtained through the material management module of the system.

[0247] Example 6 On the basis of Example 5, the cost prediction module of this embodiment also includes an equipment maintenance cost prediction submodule, which is used to predict the equipment maintenance cost involved in the production of lyocell fibers. The equipment maintenance cost prediction submodule is also composed of a first equipment maintenance prediction submodel constructed based on a random forest, a second equipment maintenance prediction submodel constructed based on a support vector machine, and a third equipment maintenance prediction submodel constructed based on a long short-term memory network. The specific data processing method and model construction and training method of each submodel are the same as the three prediction submodels in the raw material cost prediction submodule, and will not be described in detail here.

[0248] The difference lies in the different data used. Specifically, for equipment maintenance cost prediction, when training the three equipment maintenance prediction sub-models, the collected equipment maintenance data includes equipment operating time, maintenance cost, number of failures, equipment failure rate, equipment vibration data, and equipment maintenance history. Among them, equipment operating time, maintenance cost, number of failures, equipment failure rate, equipment vibration data, and equipment maintenance history data are used as input features of the model training set when training the equipment maintenance prediction sub-model, and equipment failure rate and maintenance number are used as output features of the model training set.

[0249] In the embodiment depicted in the present invention, after the equipment maintenance prediction sub-model is trained, the equipment operation time, maintenance cost, number of failures, equipment failure rate, equipment vibration data and equipment maintenance history in the production process are collected, and the above data are processed by the same preprocessing method as step F2, and then input into each trained equipment maintenance prediction sub-model. Finally, each equipment maintenance prediction sub-model outputs the equipment failure rate and maintenance number as the predicted value of the model; then the predicted values ​​of the three equipment maintenance prediction sub-models are fused in the same weighted fusion method to obtain the final equipment maintenance prediction value. The weight values ​​of the three equipment maintenance prediction sub-models can be determined by referring to the three raw material consumption prediction sub-models.

[0250] Furthermore, for the equipment maintenance prediction submodule, the time series data includes equipment operation time and number of failures; the numerical data includes maintenance cost and equipment failure rate; the non-normal distribution data includes maintenance cost, number of failures, equipment failure rate, equipment vibration data and equipment maintenance history; Pearson correlation analysis and principal component analysis are used to extract key features of equipment maintenance data, and the correlation between equipment operating time, number of failures, and equipment failure rate is calculated to evaluate the impact of equipment aging on the failure rate. At the same time, the relationship between maintenance cost and equipment operating time and number of failures is analyzed to optimize maintenance budgets and repair plans.

[0251] When the K-means cluster analysis and information entropy screening methods are used for the final screening of the equipment maintenance data, in terms of equipment maintenance data, the focus is on screening the equipment operating time, maintenance cost, number of failures and equipment failure rate to analyze the impact of equipment status on production efficiency and maintenance cost.

[0252] Finally, the final equipment maintenance cost forecast value can be calculated by the following formula: ; In the formula, Indicates the final equipment maintenance forecast value; represents the equipment maintenance prediction value of the first equipment maintenance prediction sub-model; represents the equipment maintenance prediction value of the second equipment maintenance prediction sub-model; represents the equipment maintenance prediction value of the third equipment maintenance prediction sub-model; The meaning of is the contribution of the first equipment maintenance prediction sub-model in the weighted fusion, that is, the weight value of the first equipment maintenance prediction sub-model; The meaning of is the contribution of the second equipment maintenance prediction sub-model in the weighted fusion, that is, the weight value of the second equipment maintenance prediction sub-model; The meaning of is the contribution of the third equipment maintenance prediction sub-model in the weighted fusion, that is, the weight value of the third equipment maintenance prediction sub-model; Furthermore, the final equipment maintenance forecast cost can be calculated using the following formula: ; C 设备维护 To provide a forecast for the final equipment maintenance cost; C 维护 is the fixed maintenance cost of the equipment; C 单次故障 is the cost of repairing a single fault; Y final-故障 is the final number of equipment failures.

[0253] Example 7 On the basis of any of the above embodiments, the intelligent simulation system of this embodiment may also include production operation management modules or production comprehensive management modules such as user management module, production management module, material management module, equipment management module, etc. These modules further improve the intelligent simulation system of the present invention, enhance the practicality and reliability of the system, and enable it to better meet the actual needs of enterprise production management. By realizing comprehensive digital management of the production process, production efficiency, product quality and resource utilization are also improved, while reducing production costs and risks.

[0254] Specifically, each of the above functional modules will be introduced and explained in detail below.

[0255] The user management module is mainly used to create, maintain, log in to user accounts, and set and maintain user account permissions; specifically, the user management module can be composed of the following sub-modules, including: The account management submodule is used by system administrators to maintain user accounts and create new users; creating new users usually includes entering basic user information and setting an initial login password.

[0256] The login management module is used to verify the user's login process, including verifying the input account and password; if the information is correct, login is allowed; if it is wrong, it prompts to re-enter the information and records the number of errors. When a certain number of errors is reached, the account is locked to enhance system security.

[0257] The permission management module is used to set and maintain permissions for user accounts. When a user attempts to access a function beyond the permission limit, the system will immediately block the user and prompt the user that the permission is insufficient.

[0258] It is understandable that the user management module can also communicate with other modules of the system, for example, passing the user's identity and authority information to the equipment management module, production management module, etc. If the current user's position is an inspector, then in the equipment management module, the user can only view the equipment information related to the inspection task; in the production management module, the user can only view the production plan and cannot modify it, so as to ensure the standardization of system operation and the security of data.

[0259] The production management module is used to formulate, adjust, approve, query and issue production plans, which are composed of annual production plans, monthly production plans and daily production plans. The production management module is usually used by the production management department of the enterprise. The production management department formulates production plans through the production management module based on market orders and enterprise production capacity, and clarifies the output targets and production schedules of each production line. After the plan is formulated, the reviewer will review the plan in the system. If approved, the production management module will issue the production plan to each relevant module.

[0260] For example, after receiving the production plan transmitted by the production management module, the material management module calculates the required raw material quantity and arranges raw material procurement and inventory allocation according to the planned production demand and raw material consumption quota. The equipment management module reasonably arranges the maintenance and repair time of the equipment according to the production plan to ensure the normal operation of the equipment during production. For example, arrange deep maintenance of the thin film evaporator during the period when the production task is light.

[0261] In addition, the production management module can also communicate with the mobile terminal device. When the production plan is formulated and approved, the formulated production plan is sent to the mobile terminal device of the corresponding staff member through the wireless transmission module.

[0262] For the material management module, in the raw material procurement process, the procurement department determines the procurement needs based on the production plan issued by the production management module and the inventory data of the material management module. After the procurement is completed, the raw materials arrive and are put into storage. The warehouse manager enters the batch, quantity, quality inspection report and other information of the raw materials through the warehousing function of the material management module. The system automatically updates the inventory ledger and increases the inventory quantity.

[0263] Furthermore, when the production workshop collects materials according to the production plan, the workshop staff submits a material collection application in the material management module, indicating the type, quantity and purpose of the required raw materials. The material management module reviews the application and checks whether the inventory is sufficient. If the inventory meets the demand, the material collection application is approved, and the inventory ledger is updated to reduce the inventory quantity; if the inventory is insufficient, the information is fed back to the production management module and the early warning module.

[0264] In general, during the production process, the material management module monitors the consumption of materials in real time. By linking with the equipment management module and the production management module, the equipment operation data and production progress data are obtained to calculate the actual consumption rate of the materials. When abnormal material consumption is found (such as consumption too fast or too slow), the abnormal information is transmitted to the early warning module, and the early warning module promptly notifies the relevant production management personnel to conduct inspections and verifications to ensure the normal production process and avoid material problems affecting product quality and production efficiency. At the same time, the material management module regularly counts the inventory and compares the inventory results with the system inventory data. If there are differences, the reasons are promptly found and adjusted to ensure the accuracy of the inventory data.

[0265] For the early warning module of this embodiment, during the production process of lyocell fiber, the material management module of the system monitors the raw material inventory in real time. When the inventory of a certain raw material (such as wood pulp) is lower than the set safety threshold, the material management module transmits the inventory data to the early warning module. After receiving the data, the early warning module determines that the inventory is insufficient and triggers the material inventory early warning.

[0266] At the same time, when the system is running, the key equipment status prediction module of the system continuously monitors the status of key equipment such as the thin film evaporator, and transmits the predicted remaining life or health index data of the target thin film evaporator to the early warning module; the early warning module determines whether the remaining life or health index of the target thin film evaporator is lower than the safety threshold and achieves the early warning effect.

[0267] In general, the early warning module issues an alarm in a variety of ways to notify relevant personnel, such as popping up a striking logo on the system interface, sending SMS notifications to relevant personnel (equipment maintenance personnel, production supervisors, etc.), and issuing early warnings in the three-dimensional digital twin model of the production line.

[0268] Furthermore, it is understandable that on the operation interface of the entire intelligent simulation system, operators can intuitively see detailed information on the above-mentioned insufficient material inventory and potential equipment failures, including inventory quantity, name and location of faulty equipment, etc. After receiving the warning information transmitted by the early warning module, production managers can adjust the production plan in time through the production management module to reduce dependence on insufficient raw materials, or suspend related production links to avoid production interruptions due to material shortages or equipment failures. In addition, the equipment management module can arrange maintenance personnel to prepare for maintenance work based on equipment failure warnings, prepare maintenance tools and spare parts in advance, and ensure that equipment failures can be handled in time to maintain production continuity.

[0269] Furthermore, for the equipment management module of this embodiment, the equipment management module is generally composed of an equipment information maintenance submodule, an equipment inspection submodule and an equipment repair and maintenance submodule. In the equipment information maintenance submodule, managers enter detailed information about newly purchased equipment (such as a new spinning machine), including basic data such as the equipment name, model, manufacturer, installation location, technical parameters, and assign a unique number to the equipment.

[0270] The equipment inspection submodule formulates an inspection plan based on the equipment type, operating status and maintenance requirements, such as requiring a comprehensive inspection of the spinning machine every week. The inspection plan is sent to the terminal device (such as a mobile phone or tablet) of the inspector through the system. The inspector performs the inspection task according to the plan, uses the terminal device to scan the equipment QR code, records the inspection time, equipment operating status (such as temperature, vibration, pressure and other parameters), whether any abnormalities are found, and uploads it to the equipment management module, finally forming an equipment inspection log.

[0271] If any equipment abnormality is found during the inspection process, the equipment management module will receive the data and transmit the abnormal information to the early warning module for early warning.

[0272] The equipment inspection and maintenance submodule is used to formulate equipment inspection and maintenance plans and arrange maintenance personnel to perform maintenance. After the maintenance is completed, the maintenance personnel will record the maintenance content, replaced parts, maintenance time and other information through the equipment management module to generate an inspection and maintenance log. In addition, the equipment management module also communicates with the production management module. When the equipment needs maintenance and may affect the production plan, it will promptly notify the production management module to adjust the production schedule to ensure that production and equipment maintenance work are coordinated. At the same time, the equipment management module provides the equipment's operating data and maintenance records to the cost prediction module as a basis for equipment maintenance cost prediction.

[0273] Finally, for the process management module of this embodiment, the process engineer compiles a new lyocell fiber production process file in the process management module, and records in detail the production process, process parameters of each process (such as dissolution temperature, spinning speed, drafting multiple, etc.), quality control standards and other information. After the process file is compiled, it is submitted for approval. After approval, the process management module publishes the process file to relevant modules such as the production management module, equipment management module and quality control module.

[0274] At the same time, the process management module is linked with the 3D modeling and digital twin modules to simulate the production process of the new process in a virtual environment, discover potential problems in advance, verify the feasibility and stability of the process, and reduce risks in actual production.

[0275] It is understandable that the process management module assists the energy consumption prediction module in realizing energy consumption prediction of the glue making process by providing the energy consumption prediction module with a process sheet of the glue making process of lyocell fiber production.

[0276] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A lyocell fiber production line intelligent simulation system based on digital twin, characterized in that: The system includes: 3D modeling and digital twin module, key equipment status prediction module, energy consumption prediction module and early warning module; 3D modeling and digital twin module, which is used to build a 3D model of the Lyocell fiber production line, associate the real-time production data of the production line with the 3D model, and generate a 3D digital twin model of the Lyocell fiber production line based on digital twin technology; A key equipment status prediction module includes a remaining life prediction submodule for a thin film evaporator, wherein the remaining life prediction submodule for the thin film evaporator obtains real-time vibration data collected by a vibration sensor on a target thin film evaporator and real-time spindle speed transmitted by a central control platform of the target thin film evaporator; the real-time vibration data is segmented according to a set time interval to obtain a plurality of data segments at set time intervals, and then each data segment is analyzed in the time domain and frequency domain to obtain the root mean square, kurtosis and frequency center of gravity of each data segment; the spindle fault characteristic frequency is obtained according to the obtained real-time spindle speed, and then the spindle bearing fault characteristic frequency is obtained according to the real-time spindle speed and the size parameters of the spindle bearing; finally, the obtained root mean square, kurtosis, frequency center of gravity, spindle fault characteristic frequency and spindle bearing fault characteristic frequency are input into a remaining life prediction model, and finally the remaining life prediction result is output to the early warning module; The energy consumption prediction module extracts the porridge weight, process parameter group 1, process parameter group 2 and process parameter group 3 corresponding to the current process list according to the process list of the glue making process provided by the process management module; then obtains the glue liquid weight based on the porridge weight and process parameter group 1, obtains the glue making consumption time based on the porridge weight and process parameter group 3, and obtains the glue liquid refractive index based on process parameter group 2; finally, the porridge weight, glue liquid weight, glue liquid refractive index and glue making consumption time are input into the trained energy consumption prediction model, and the energy consumption prediction value corresponding to the glue making process is output; wherein, the process parameter group 1 includes the motor speed of the thin film evaporator, the temperature of the four zones of the thin film evaporator, the vacuum system pressure and the vacuum pump speed; the process parameter group 2 includes the motor speed of the thin film evaporator, the temperature of the four zones of the thin film evaporator, the vacuum pump speed and the outlet temperature of the insulation water circulation pump; the process parameter group 3 includes the motor speed of the thin film evaporator and the vacuum pump speed; The early warning module is used to issue an early warning in a device model corresponding to the target thin film evaporator in the three-dimensional digital twin model according to the remaining life prediction result of the target thin film evaporator.

2. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 1, characterized in that: The real-time vibration data includes acceleration signals in three directions respectively collected by vibration sensors at the upper and lower ends and the middle area of ​​the target thin film evaporator cylinder; wherein the vibration sensor at the upper end of the cylinder is used to monitor the fault condition of the upper end bearing of the main shaft, the vibration sensor at the lower end of the cylinder is used to monitor the fault condition of the lower end bearing of the main shaft, and the vibration sensor in the middle area of ​​the cylinder is used to monitor the fault condition of the main shaft.

3. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 1, characterized in that: The construction process of the remaining life prediction model of the thin film evaporator is as follows: Obtain historical vibration data and historical spindle speed data of the entire process from the normal operation of the thin film evaporator to the failure of the spindle and / or spindle bearing resulting in the failure of the thin film evaporator to operate normally; The historical vibration data includes acceleration signals in three directions respectively collected by vibration sensors at the upper and lower ends and the middle area of ​​the thin film evaporator cylinder; wherein the vibration sensor at the upper end of the cylinder is used to monitor the fault condition of the upper end bearing of the main shaft, the vibration sensor at the lower end of the cylinder is used to monitor the fault condition of the lower end bearing of the main shaft, and the vibration sensor in the middle area of ​​the cylinder is used to monitor the fault condition of the main shaft; The acceleration signal is segmented at set time intervals to obtain multiple data segments, and then each data segment is analyzed in the time domain and the frequency domain to extract the root mean square, kurtosis and frequency center of gravity of each data segment; the fault characteristic frequency of the spindle is obtained according to the historical spindle speed; the fault characteristic frequency of the spindle bearing is obtained according to the historical spindle speed and the spindle bearing size parameters; After vector addition of acceleration signals in three directions of each vibration sensor, multiple vibration vector data are obtained. A one-dimensional convolutional neural network model including convolution, pooling, activation function and full connection layer is built for each vibration vector data. The response trend of the vibration response of each vibration vector data changing with time is extracted through the one-dimensional convolutional neural network model of each vibration vector data, and the vibration vector data characteristic curves of the upper and lower end bearings and the main shaft are obtained. Multiple high-order exponential functions are used to fit the vibration vector data characteristic curves of the upper and lower end bearings and the main shaft, respectively, to form the remaining life characteristic curves of the upper end bearing of the main shaft, the remaining life characteristic curves of the lower end bearing of the main shaft and the remaining life characteristic curves of the main shaft, and then expressed in percentage form; Taking the set time interval as the time resolution, the data set of the remaining life prediction model is formed based on the root mean square, kurtosis, frequency centroid, fault characteristic frequency of the spindle bearing, fault characteristic frequency of the spindle, and the data on the remaining life characteristic curve of the upper end bearing of the spindle, the remaining life characteristic curve of the lower end bearing of the spindle, and the remaining life characteristic curve of the spindle; wherein the data on the remaining life characteristic curve is the label data; The obtained data set is divided into a training set, a validation set and a test set; the constructed remaining life prediction model is trained, verified and tested respectively using the divided data sets, and finally a trained remaining life prediction model is obtained; the remaining life prediction model is a one-dimensional convolutional neural network model.

4. A lyocell fiber production line intelligent simulation system based on digital twin according to claim 1 or 3, characterized in that: The characteristic frequencies of spindle bearing faults include: Inner race fault characteristic frequency: ； Outer race fault characteristic frequency: ； Rolling element failure characteristic frequency: ； Cage failure characteristic frequency: ； Where: Q is the number of rolling elements, d The diameter of the rolling element, D Bearing pitch diameter, β is the contact angle, f r The rotation frequency of the main axis.

5. A Lyocell fiber production line intelligent simulation system based on digital twin according to claim 1 or 3, characterized in that: The characteristic frequency of spindle fault is: ； Where: n 1 is the multiple number, f r is the rotation frequency of the main axis; is the main axis disturbance frequency.

6. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 3, characterized in that: The calculation formula expressed in percentage form after forming the remaining life characteristic curve of the upper end bearing of the main shaft, the remaining life characteristic curve of the lower end bearing of the main shaft and the remaining life characteristic curve of the main shaft is as follows: ； Where: , , , , is an adjustable parameter; is the horizontal coordinate of the vibration data, i.e., the system time of the vibration data; The time limit for the equipment to go from fully normal to abnormal is defined; is the remaining life percentage; e Is the natural base.

7. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 1, characterized in that: The construction process of the energy consumption prediction model is as follows: Collect the historical raw data of evaporation system parameter data, glue and sealing system parameter data, insulation water and vacuum system parameter data, slurry weight, glue weight, glue refractive index, glue consumption time and energy consumption data during the Lyocell fiber glue making process, and perform data preprocessing to obtain historical data; Calculate the process similarity of the corresponding historical data according to the process sheet of the Lyocell fiber glue making process, and rearrange the historical data according to the process similarity to obtain the historical sorted data; A reference group sequence is selected from the historical sorting data, and the process similarity of each reference group sequence is calculated to obtain a historical sorting data sequence, and a threshold interval of process similarity is obtained; then, the historical sorting data sequence within the threshold interval is used to calculate the energy consumption prediction value sequence; Compare the energy consumption forecast value sequence with the historical energy consumption data of the corresponding reference group sequence, and record their absolute errors and X; Increase the threshold interval percentage P of process similarity to obtain the threshold interval of the current process similarity until the threshold interval of the current process similarity reaches the set maximum value, and obtain the threshold interval percentage P of process similarity corresponding to the minimum absolute error and X, and mark P as P s ; Build an energy consumption prediction model and use P s The historical screening and sorting data are obtained by screening, and the porridge weight, glue liquid weight, glue liquid refractive index and glue making time in the historical screening and sorting data are used as the data set of the energy consumption prediction model. The energy consumption prediction model is trained using the data set to finally obtain a trained energy consumption prediction model.

8. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 7, characterized in that: Historical original data include historical original process parameter data, historical original energy consumption data and historical original intermediate parameter data; historical data include historical process parameter data, historical energy consumption data and historical intermediate parameter data; historical sorted data include historical sorted process parameter data, historical sorted intermediate parameter data and historical sorted energy consumption data; historical filtered and sorted data include historical filtered and sorted process parameter data, historical filtered and sorted intermediate parameter data and historical filtered and sorted energy consumption data.

9. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 8, characterized in that: The process sheet includes process parameter data; the process parameter data include evaporation system parameter data, glue liquid and sealing system parameter data, insulation water and vacuum system parameter data and porridge weight; the evaporation system parameter data include evaporation heating system parameter data, thin film evaporator parameter data and evaporation condensation water system parameter data; the glue liquid and sealing system parameter data include sealing liquid system parameter data and glue liquid transportation parameter data; the insulation water and vacuum system parameter data include insulation water system parameter data and vacuum system parameter data.

10. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 9, characterized in that: The evaporation heating system parameter data includes the heating system temperature, the heating system liquid level and the heating system pressure; the thin film evaporator parameter data includes the four-zone temperature of the thin film evaporator, the thin film evaporator motor speed, the thin film evaporator bearing lubricating oil circulation temperature, the reducer bearing speed, the cooling fan speed, the thin film evaporator liquid level, the thin film evaporator outlet pressure and the thin film evaporator bottom outlet temperature; the four-zone temperature of the thin film evaporator includes the jacket temperature of the four evaporation zones and the glue temperature of the four evaporation zones; the glue delivery parameter data includes the temperature of the insulation water heat exchanger in the pump, the flow rate of the insulation water heat exchanger in the pump, and the insulation water heat exchanger outside the pump. The data include the temperature of the sealing liquid tank, the flow rate of the sealing liquid tank and the liquid level of the sealing liquid tank; the data include the outlet temperature of the insulation water circulation pump and the hot water pressure of the insulation system; the data include the temperature of the evaporative condenser, the liquid level of the evaporative condenser, the pressure of the evaporative condenser, the flow rate of the cooling water to the evaporative condenser and the outlet flow rate of the evaporative condensate pump; the data include the vacuum system liquid level, the pressure of the vacuum system and the speed of the vacuum pump.

11. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 8, characterized in that: The intermediate parameter data include glue liquid weight, glue liquid refractive index and glue making time.

12. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 7, characterized in that: The data preprocessing method is: based on 3 σ In principle, outliers are removed; after outliers are removed, the missing values ​​are filled by taking the weighted average of the K neighboring data values ​​near the missing value.

13. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 7, characterized in that: The specific method of selecting a reference group from historical sorting data is: Z The row data constitutes the reference group sequence ; By reference group sequence For each sequence in the reference group, we obtain the historical sorting data corresponding to each sequence in the reference group. , thus forming the historical sorting data sequence of the reference group sequence , and record the maximum process similarity value in the process ; Among them, the threshold interval of process similarity is , is the minimum process similarity value.

14. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 7, characterized in that: Using the historical sorted data sequence within the threshold range, the energy consumption forecast value sequence is calculated. The specific method is as follows: The porridge weight of the historically sorted intermediate parameter data and the historically sorted process parameter data in the historically sorted data sequence of the reference group sequence is used as the training set of the neural network model to obtain the energy consumption prediction value sequence of the reference group sequence; the process similarity corresponding to the historically sorted intermediate parameter data belongs to the threshold interval range of the process similarity.

15. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 7, characterized in that: The energy consumption prediction value sequence of the reference group sequence is compared with the historical energy consumption data of the corresponding reference group sequence, and their absolute errors and X are recorded. The specific method is: Compare the energy consumption prediction value sequence of the reference group sequence with the historical energy consumption data of the corresponding reference group sequence, record their absolute error values, and traverse the historical sorting data sequence. Z Sorting historical data to obtain the absolute error value sequence of energy consumption prediction , and then accumulate the energy consumption prediction absolute error value sequence to obtain the absolute error and X and record it.

16. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 7, characterized in that: The porridge weight, glue weight, glue refractive index and consumption time in the historical screening and sorting data are used as the data set of the energy consumption prediction model. The energy consumption prediction model is trained using the data set, and finally a trained energy consumption prediction model is obtained, including: The glue liquid weight, glue liquid refractive index, consumption time and porridge weight in the historical screening and sorting of intermediate parameter data are used as input features of the training set of the energy consumption prediction model, and the energy consumption data of historical screening and sorting are used as the target output of the energy consumption prediction model. The model is trained using the above data to obtain a trained energy consumption prediction model; the energy consumption prediction model uses a three-layer BP neural network.

17. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 7, characterized in that: The energy consumption data is the sum of electricity consumption and steam consumption.

18. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 1, characterized in that: The key equipment status prediction module also includes a health prediction submodule for a thin film evaporator; the health prediction submodule for the thin film evaporator obtains the current production parameters of the target thin film evaporator and the current production parameters of the upstream pulper, and then pre-processes the collected data, including the step of removing outliers and normalization processing; then based on the expert scoring method, the collected parameters are weighted from three dimensions: the degree of parameter fluctuation, the degree of influence of the parameters on the health status of the equipment, and the severity of the deviation of the parameters from the normal range, and the sum of the products of each dimension of each parameter and the corresponding weight is calculated to obtain the influence weight of each parameter in the health prediction; finally, the weighted data of each parameter are input into the trained health prediction model to predict the health of the target thin film evaporator, and the health index of the target thin film evaporator is output; wherein, the real-time production parameters of the thin film evaporator include evaporation chamber pressure, stirring motor current, feed flow rate, feed pipeline pressure, dissolving liquid temperature, solution viscosity and solvent recovery rate; the real-time production parameters of the pulper include discharge flow rate, slurry concentration and slurry temperature.

19. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 18, characterized in that: When collecting the real-time production parameters of the target thin-film evaporator and the real-time production parameters of the pulper, the parameters are synchronized in time to ensure the timing alignment of different data sources; among them, the parameters of the pulper are compensated in advance according to the material transportation time to match the operating status of the thin-film evaporator.

20. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 19, characterized in that: Outliers are eliminated by combining physical constraint methods and statistical methods. The physical constraint method refers to setting parameter thresholds based on the equipment operating range, and exceeding the range is considered an abnormality; the statistical method refers to eliminating abnormal points based on the distribution of historical data using the 3σ principle.

21. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 18, characterized in that: The degree of parameter fluctuation is measured by normalized standard deviation, which is calculated as follows: ； In the formula, Indicates i Normalized standard deviation of production parameters; For the i The standard deviation of the production parameters, is the sum of the standard deviations of all production parameters.

22. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 18, characterized in that: The degree of influence of parameters on the health status of equipment is measured by normalized expert score, which is calculated as follows: ； In the formula, Indicates i Normalized expert scores for production parameters; For the i Expert ratings of production parameters, For the total score.

23. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 18, characterized in that: The severity of a parameter deviation from the normal range is measured using the physical constraint weight, which is calculated as follows: ； In the formula, Indicates i The severity of the deviation of a production parameter from the normal range; For production parameters i The number of times beyond the normal range, is the total number of samples of production parameters.

24. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 18, characterized in that: The health index is calculated as follows: ； In the formula, I is the health index of the thin film evaporator; is the predicted failure probability of the thin film evaporator.

25. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 18, characterized in that: The health prediction model is a long short-term memory artificial neural network, which includes an input layer, an LSTM layer, a fully connected layer and an output layer. The input layer is used to accept time series feature data, and its shape is determined by the step size and the total number of features at each step size; the LSTM layer is a single layer with 64 hidden units, which is used to capture long-term dependencies in the time series; the fully connected layer is used to generate prediction results; and the output layer is used to output the failure probability.

26. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 1, characterized in that: The system also includes a cost prediction module, which includes a raw material cost prediction submodule. The raw material cost prediction submodule obtains raw material data of the lyocell fiber production process and preprocesses it, inputs the preprocessed raw material data into the trained raw material consumption prediction overall model, outputs the raw material consumption prediction values ​​corresponding to different machine learning algorithm models, and then obtains the final raw material consumption prediction value by weighted fusion of the prediction values ​​of each machine learning algorithm model; finally, the raw material prediction cost is obtained in combination with the raw material price; wherein the raw material data includes wood pulp parameter data, NMMO solution parameter data, and chemical additive parameter data; the wood pulp parameter data includes wood pulp viscosity, wood pulp polymerization degree and wood pulp methyl fiber content; the NMMO solution parameter data includes NMMO solution concentration; the chemical additive parameter data includes propylene glycol content and hydroxylamine content.

27. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 26, characterized in that: The raw material consumption forecast value includes the wood pulp consumption forecast value, the NMMO solution consumption forecast value, the propylene glycol consumption forecast value and the hydroxylamine consumption forecast value; the raw material forecast cost includes the wood pulp forecast cost, the NMMO solution forecast cost, the propylene glycol forecast cost and the hydroxylamine forecast cost.

28. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 26, characterized in that: The raw material consumption prediction overall model includes a first raw material consumption prediction sub-model constructed based on random forest, a second raw material consumption prediction sub-model constructed based on support vector machine, and a third raw material consumption prediction sub-model constructed based on long short-term memory network.

29. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 26, characterized in that: Preprocess the raw material data, specifically: Adoption 3 σ In principle, outlier detection is performed on the numerical data in the raw material data, and the interquartile range method is used to detect outliers on the non-normal distribution data in the raw material data, and the outliers are removed; then missing value processing and duplicate data removal are performed; Min-Max was used to normalize the raw material data; Pearson correlation analysis and principal component analysis were used to extract key features of raw material data; K-means cluster analysis and information entropy screening methods were used to conduct final screening of raw material data.

30. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 29, 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, and if the missing ratio exceeds 20%, the sample is eliminated; if the missing ratio of non-normal distribution data is less than 20%, k-nearest neighbor interpolation is used to fill it, and if the missing ratio exceeds 20%, the sample is eliminated.

31. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 26, characterized in that: The final raw material consumption prediction value is obtained by weighted fusion of the predicted values ​​of each machine learning algorithm model. Specifically, the initial weight distribution of the total raw material consumption prediction model is set, and the mean square error of each machine learning algorithm model is calculated; then the Beta distribution parameters are updated according to the mean square error, and Gaussian process regression is introduced to calculate the uncertainty of each machine learning algorithm model in the total model, and then the Beta distribution parameters are updated again according to the uncertainty; finally, the fusion weights of each machine learning algorithm model are obtained according to the updated Beta distribution parameters combined with variational Bayesian inference calculation.

32. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 1, characterized in that: The system also includes a production management module, which is used to formulate, adjust, approve, query and issue annual production plans, monthly production plans and daily production plans.

33. The digital twin-based lyocell fiber production line intelligent simulation system according to claim 1, characterized in that: The system also includes an equipment management module, which is used to create and manage the equipment information of the production line, configure and issue inspection plans, and configure and issue equipment maintenance strategies; the equipment management module includes an equipment information maintenance submodule, an equipment inspection submodule, and an equipment repair and maintenance submodule; wherein, The equipment information maintenance submodule is used to create and manage production line equipment information, including basic data such as the name, type and number of the configured equipment. The equipment inspection submodule is used to configure the inspection plan and send it to the employee terminal, and collect the inspection results to generate the corresponding inspection log; The equipment inspection and maintenance submodule is used to configure the maintenance strategy and send it to the employee terminal, and collect equipment inspection and maintenance to generate corresponding inspection and maintenance logs.

Images