A resin batch quality prediction method and device based on ensemble learning and a medium

Abstract

This invention discloses a method, device, and medium for predicting resin batch quality based on ensemble learning, relating to the field of machine learning technology. The method includes: collecting completed resin batch records, calculating the cumulative temperature during the production stage to form the current stage's production data; inputting three types of data into three basic learners to obtain predicted values ​​for the quality index to be predicted and the predicted value for the cumulative temperature of the next production stage; when it is not the first production stage, forming the predicted difference of the three cumulative temperature values, and updating the outputs of the three basic learners according to their sequence numbers; inputting the predicted value of the quality index to be predicted into the ensemble learner, and when it is not the first production stage, adding the integrated predicted value of the quality index to be predicted from the previous production stage and the predicted difference of the three cumulative temperature values ​​to obtain the integrated predicted value of the quality index to be predicted and forming the current resin batch quality prediction result. This invention achieves differentiated prediction of resin batch quality indicators based on raw material feeding status and temperature process status.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a method, device and medium for predicting the quality of resin batches based on ensemble learning. Background Technology

[0002] Resin production processes are characterized by a wide variety of raw materials, a clearly defined feeding sequence, continuous variations in reaction temperature across different process stages, and diverse finished product quality indicators. Quality indicators such as viscosity, acid value, solid content, epoxy equivalent, and gel time are typically influenced by factors including raw material properties, feeding status, reaction temperature, and production duration. With increasing digitalization of production processes, resin manufacturers are gradually utilizing historical batch production records and quality inspection results to establish data prediction relationships. They are also employing methods such as regression analysis, decision trees, or ensemble learning to predict the quality indicators of resin batches in production, providing data references for production quality analysis.

[0003] However, existing methods still have two limitations: First, heterogeneous data such as raw material quality, feeding execution, and temperature process are usually combined in the prediction, making it difficult to retain independent prediction information corresponding to different production states; second, the predictions at each stage are mostly relatively independent results, and the deviation of the previous stage's quality prediction and its prediction of the subsequent temperature process is not continuously fed back to the current stage, making it difficult to form a prediction relationship that is inherited and corrected stage by stage as production progresses. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, this invention provides a resin batch quality prediction method based on ensemble learning, which solves the problems of difficulty in differentiating and utilizing information on different production states in resin batch quality prediction, and difficulty in combining prediction results with actual temperature process deviations for stage-by-stage acceptance and correction.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a resin batch quality prediction method based on ensemble learning, comprising: collecting completed resin batch records; dividing the production process into stages according to production process nodes; calculating the cumulative temperature of each production stage based on the reactor temperature records and sampling times of each production stage to form the current stage production data for each production stage; dividing the current stage production data according to the raw material feeding status and temperature process status to obtain a raw material feeding data combination and a temperature process data combination, and merging them to obtain a comprehensive production data combination; inputting the three types of data combinations into a first basic learner, a second basic learner, and a third basic learner respectively; using the detection value of the quality index to be predicted and the cumulative temperature of the next production stage as training control data for training, to obtain the predicted value of the quality index to be predicted and the cumulative temperature of the next production stage. For non-first production stages, the cumulative temperature of the current production stage is compared with the predicted cumulative temperature of the next production stage output by the three basic learners of the previous production stage to form a temperature cumulative prediction difference. This difference is then combined with the corresponding data combination of the current production stage and fed back into the basic learner sequence number to update the predicted quality indicators output by the three basic learners, as well as the predicted cumulative temperature of the next production stage (excluding the last production stage). The predicted quality indicators output by the three basic learners of the current production stage are input into the integrated learner. For non-first production stages, the integrated predicted quality indicators of the previous production stage and the temperature cumulative prediction difference are added to obtain the integrated predicted quality indicators of the current production stage, forming the quality prediction result for the current resin batch.

[0008] As a preferred embodiment of the resin batch quality prediction method based on ensemble learning described in this invention, the step of dividing the production process into stages according to production process nodes and calculating the cumulative temperature of each production stage based on the reactor temperature records and sampling times of each production stage includes: arranging each production process node in chronological order, taking the production process between two adjacent production process nodes as a production stage, taking the node time of the previous production process node as the start time of the production stage, taking the node time of the next production process node as the end time of the production stage, and taking the time difference between the end time and the start time of the production stage as the stage duration. The production stages are arranged in chronological order of their start times to obtain a production stage sequence. For each production stage in the sequence, the start and end times of the production stage are used as the temperature accumulation boundaries. When the temperature accumulation boundary coincides with the sampling time, the reactor temperature value at the coincident sampling time is used. When the temperature accumulation boundary is between two adjacent sampling times, linear interpolation is performed according to the time position of the two adjacent sampling times and the reactor temperature value. The average value of the reactor temperature at each adjacent time interval is accumulated in chronological order, and the product of the average value of the reactor temperature at each time interval and the time length between the two times is added to obtain the cumulative temperature of the production stage.

[0009] As a preferred embodiment of the resin batch quality prediction method based on ensemble learning described in this invention, the step of dividing the current stage production data according to the raw material feeding status and temperature process status includes: arranging the raw material inspection values, actual feeding amount, and feeding elapsed time according to the feeding sequence in the same production process route to obtain a raw material feeding data combination, wherein the feeding elapsed time is the time difference between the corresponding feeding time and the first feeding time; arranging the stage duration from the first production stage to the current production stage and the cumulative temperature of the production stage according to the production stage sequence to obtain a temperature process data combination; and arranging the raw material feeding data combination and the temperature process data combination in the order of raw material feeding data combination first and temperature process data combination second to obtain a comprehensive production data combination.

[0010] As a preferred embodiment of the resin batch quality prediction method based on ensemble learning described in this invention, the step of inputting three types of data combinations into a first basic learner, a second basic learner, and a third basic learner, respectively, and using the detected value of the quality index to be predicted and the cumulative temperature of the next production stage as training control data, includes: using raw material input data combinations as training input for the first basic learner, temperature process data combinations as training input for the second basic learner, and comprehensive production data combinations as training input for the third basic learner; when the current production stage is not the first production stage in the production stage sequence, adding the predicted value of the quality index to be predicted and the predicted difference of the cumulative temperature output by the same basic learner in the previous production stage to the training input of the same basic learner; using the detected value of the quality index to be predicted for each completed resin batch as training control data for predicting the quality index to be predicted; when the current production stage is not the last production stage in the production stage sequence, using the cumulative temperature of the production stage immediately following the current production stage in the same completed resin batch as the cumulative temperature of the next production stage, and as training control data for predicting the cumulative temperature; for any basic learner, training C based on the corresponding training input and training control data. CART regression tree; when the current production stage is not the last production stage in the production stage sequence, two CART regression trees are trained with the same training input. One CART regression tree uses the detected value of the quality indicator to be predicted as the training control data, and the other CART regression tree uses the cumulative temperature of the next production stage as the training control data. When the current production stage is the last production stage in the production stage sequence, a CART regression tree is trained using the detected value of the quality indicator to be predicted as the training control data. When predicting the first production stage of the current resin batch, the raw material input data, temperature process data, and comprehensive data are combined. The production data combinations are input into the first, second, and third basic learners, respectively. Leaf nodes are found according to the node splitting rules of each CART regression tree. The leaf node outputs of the CART regression trees of the three basic learners, which use the detection values ​​of the quality indicators to be predicted as the training control data, are used as the predicted values ​​of the first, second, and third quality indicators to be predicted, respectively. The leaf node outputs of the CART regression trees of the next production stage, which use the cumulative temperature of the next production stage as the training control data, are used as the predicted values ​​of the cumulative temperature of the next production stage output by the corresponding basic learner.

[0011] As a preferred embodiment of the resin batch quality prediction method based on ensemble learning described in this invention, the step of training a CART regression tree based on corresponding training input and training control data includes: placing training samples from the same production stage at the root node, traversing the input variables one by one, arranging the training samples according to their values ​​on the current input variables from smallest to largest, and using the median value of two adjacent different values ​​as candidate split points; for each candidate split point, assigning training samples whose current input variable values ​​are less than or equal to the candidate split point to the left child node, and assigning training samples whose current input variable values ​​are greater than the candidate split point to the right child node, and calculating the values ​​of the left child nodes respectively. The sum of the squared differences between the training control data and the average value of the corresponding training control data within the right child node is used as the sum of the squared errors of the candidate split points. The input variable and candidate split point with the smallest sum of squared errors are selected to complete the binary split. The selection of candidate split points and binary splitting are repeated for the child nodes formed by the binary split until only one training sample remains in the child node, there are no candidate split points, or all candidate split points cannot reduce the sum of the squared differences between the training control data within the current node. The child node that stops splitting is designated as the leaf node, and the average value of the training control data of the training samples in the leaf node is used as the output of the leaf node.

[0012] As a preferred embodiment of the resin batch quality prediction method based on ensemble learning described in this invention, the step of forming the temperature cumulative prediction difference and combining it with the data combination corresponding to the current production stage according to the basic learner sequence number includes: when the current production stage is not the first production stage in the production stage sequence, subtracting the temperature cumulative prediction value of the next production stage output by the first, second, and third basic learners of the previous production stage from the temperature cumulative value of the current production stage, respectively, to obtain the first temperature cumulative prediction difference, the second temperature cumulative prediction difference, and the third temperature cumulative prediction difference; inputting the raw material input data combination of the current production stage, the predicted value of the quality index to be predicted output by the first basic learner of the previous production stage, and the first temperature cumulative prediction difference into the first basic learner; and inputting the temperature process data combination of the current production stage, the predicted value of the quality index to be predicted output by the second basic learner of the previous production stage, and the predicted value of the quality index to be predicted into the first basic learner. The predicted values ​​of the quantity indicators and the difference between the predicted values ​​of the second cumulative temperature are input into the second basic learner. The comprehensive production data combination of the current production stage, the predicted values ​​of the quality indicators to be predicted output by the third basic learner of the previous production stage, and the difference between the predicted values ​​of the third cumulative temperature are input into the third basic learner. When the current production stage is not the last production stage in the production stage sequence, the first, second, and third basic learners output the predicted values ​​of the first, second, and third predicted quality indicators to be predicted for the current production stage, respectively, and output the predicted value of the cumulative temperature for the next production stage. When the current production stage is the last production stage in the production stage sequence, the first, second, and third basic learners output the predicted values ​​of the first, second, and third predicted quality indicators to be predicted for the current production stage, respectively.

[0013] As a preferred embodiment of the resin batch quality prediction method based on ensemble learning described in this invention, the step of inputting the predicted quality index values ​​output by the three basic learners of the current production stage into the ensemble learner, and adding the ensemble predicted quality index values ​​of the previous production stage and the predicted differences of the three cumulative temperature values ​​when it is not the first production stage, to obtain the ensemble predicted quality index values ​​of the current production stage, includes: training the ensemble learners of each production stage sequentially according to the production stage sequence, reading the predicted quality index values ​​of each completed resin batch output by the first basic learner, the second basic learner, and the third basic learner in the current production stage, and then... The basic prediction combinations for each completed resin batch are formed in the order of the first, second, and third basic learners. When the current production stage is the first production stage in the production stage sequence, the basic prediction combinations are used as the training input for the integrated learner. When the current production stage is not the first production stage in the production stage sequence, the basic prediction combinations, the integrated prediction values ​​of the quality indicators to be predicted for the same completed resin batch in the previous production stage, the predicted difference of the first cumulative temperature, the predicted difference of the second cumulative temperature, and the predicted difference of the third cumulative temperature are used together as the training input for the integrated learner. The training input for the integrated learner corresponding to each completed resin batch is then used. Each value in the data is used as an input variable, and the predicted quality index values ​​of the corresponding completed resin batches are used as training control data. An ensemble learner is trained using a CART regression tree. After training, the training inputs for each completed resin batch are input into the ensemble learner to obtain the ensemble predicted values ​​of the predicted quality indexes for each completed resin batch at the current production stage. The ensemble predicted values ​​of the predicted quality indexes from the previous production stage are obtained by the ensemble learner from the previous production stage based on the training inputs of the ensemble learner corresponding to the same completed resin batch. When predicting the current resin batch, the first, second, and third base learners are used. The learners sequentially arrange the predicted values ​​of the quality indicators to be predicted output by the three basic learners in the current production stage to form the current basic prediction combination. When the current production stage is the first production stage in the production stage sequence, the current basic prediction combination is input into the ensemble learner of the current production stage. When the current production stage is not the first production stage in the production stage sequence, the current basic prediction combination, the integrated predicted value of the quality indicators to be predicted from the previous production stage, the prediction difference of the first cumulative temperature, the prediction difference of the second cumulative temperature, and the prediction difference of the third cumulative temperature are input into the ensemble learner of the current production stage to obtain the integrated predicted value of the quality indicators to be predicted in the current production stage.

[0014] As a preferred embodiment of the resin batch quality prediction method based on ensemble learning described in this invention, the step of forming the current resin batch quality prediction result includes: recording the integrated predicted value of the quality indicator to be predicted in the current production stage together with the batch identifier, resin product model, current production stage, and name of the quality indicator to be predicted to form the current resin batch quality prediction result. The current resin batch quality prediction result includes the batch identifier, resin product model, current production stage, name of the quality indicator to be predicted, and quality prediction value. The quality prediction value is the integrated predicted value of the quality indicator to be predicted in the current production stage. After the next production stage ends and the current stage production data of the next production stage is formed, the integrated predicted value of the quality indicator to be predicted formed in the next production stage is used as the new quality prediction value, and the next production stage is used as the new current production stage, and written into the current resin batch quality prediction result.

[0015] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein the computer program, when executed by the processor, implements any step of the resin batch quality prediction method based on ensemble learning as described in the first aspect of the present invention.

[0016] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the resin batch quality prediction method based on ensemble learning as described in the first aspect of the present invention.

[0017] The beneficial effects of this invention are as follows: by inputting the raw material input data combination, temperature process data combination, and comprehensive production data combination into the corresponding basic learners, the differential formation and integration of the predicted values ​​of the quality indicators to be predicted under different production states are realized; by feeding back the predicted difference of the cumulative temperature according to the basic learner sequence number, and adding the integrated predicted value of the quality indicators to be predicted in the previous production stage to the integrated learner, the prediction of the current production stage is updated to take over the deviation of the actual temperature process and the prediction results of the previous production stage, forming the current resin batch quality prediction result that is continuously corrected as the production stage progresses. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a resin batch quality prediction method based on ensemble learning.

[0020] Figure 2 This is a flowchart for calculating the cumulative temperature during the production stage and dividing the production data for the current stage.

[0021] Figure 3 This is a flowchart of the feedback between the training of the basic learner and the prediction of the temperature accumulation.

[0022] Figure 4 A flowchart for updating the predictions of the integrated learner with the current resin batch quality prediction results. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] Reference Figures 1-4 This is one embodiment of the present invention, which provides a resin batch quality prediction method based on ensemble learning, comprising the following steps:

[0027] S1. Collect completed resin batch records, divide the production into stages according to the production process nodes, calculate the cumulative temperature of each production stage by using the reactor temperature records and sampling time of each production stage, and form the current stage production data of each production stage.

[0028] Furthermore, a single production run of the same resin product using the same production process is considered a resin batch. Residual samples from the same resin batch are subjected to resin quality testing, and a quality testing record is generated based on the resin quality testing results. Resin quality testing refers to the factory quality inspection process that tests whether the finished resin product meets the production quality requirements. The testing content includes one or more of viscosity, acid value, solid content, epoxy equivalent, and gel time. The testing content that needs to be predicted is taken as the quality indicator to be predicted, and the actual test values ​​recorded in the quality testing record are taken as the test values ​​of the quality indicator to be predicted.

[0029] Furthermore, resin batches that have completed production and formed quality inspection records under the same resin product and the same production process route are selected as completed resin batches. The batch identifier, resin product model, raw material inspection record, actual feeding record, production process node record, reactor temperature record and quality inspection record of the completed resin batch are read to form a completed resin batch record.

[0030] Among them, batch identifiers are used to distinguish different resin batches, resin product models are used to screen the same resin products, raw material inspection records include raw material batch number and raw material inspection values, actual feeding records include raw material name, actual feeding amount and feeding time, production process node records include node times of switching between various process states during resin production, reactor temperature records include reactor temperature values ​​at each sampling time, and quality inspection records record the detection values ​​of the quality indicators to be predicted.

[0031] Furthermore, the resin production process includes processes such as feeding, heating, holding, cooling, and discharging. The production process nodes are selected based on the location of process state switching in the actual resin process route. In this embodiment, the production process nodes include, but are not limited to, the feeding completion node, the entry into the process temperature range node, the holding start node, the holding end node, the cooling start node, and the discharging completion node. The node time represents the moment when the process state of the same resin batch changes at the production process node. The process temperature range is determined by the lower limit and upper limit of the process temperature recorded in the production process route corresponding to the same resin product.

[0032] Specifically, the production process nodes are arranged in chronological order according to their time. The production process between two adjacent production process nodes is considered as a production stage. The time of the preceding production process node is taken as the start time of the production stage, and the time of the following production process node is taken as the end time of the production stage. The time difference between the end time and the start time of the production stage is taken as the stage duration. The production stages are then arranged in chronological order according to their start times to obtain the production stage sequence.

[0033] It should be noted that the production stage sequence refers to the production stages arranged in the order of process execution from the completion of feeding to the completion of discharging of the same resin batch. The start time and end time of the production stage are used as the time boundary for trunculating the temperature record of the reactor when calculating the cumulative temperature of the production stage.

[0034] Furthermore, for each production stage in the production stage sequence, the start and end times of the production stage are used as the temperature accumulation boundaries. The trapezoidal integral method is used to calculate the cumulative temperature of the production stage. When the temperature accumulation boundary coincides with the sampling time, the reactor temperature value at the coincident sampling time is directly used. When the temperature accumulation boundary is located between two adjacent sampling times, linear interpolation is performed according to the time position of the two adjacent sampling times and the reactor temperature value to obtain the reactor temperature value at the temperature accumulation boundary. The start time of the production stage, the sampling time within the production stage, and the end time of the production stage are arranged in chronological order, and the product of the average value of the reactor temperature at two adjacent time intervals and the time length between the two time intervals is accumulated to obtain the cumulative temperature of the production stage.

[0035] It should be noted that the trapezoidal integral method refers to using the reactor temperature values ​​at two adjacent moments as the two parallel sides of a trapezoid, and the time length between the two moments as the height of the trapezoid. The cumulative result of the reactor temperature over time is represented by the sum of the trapezoidal surfaces of each adjacent time interval. The cumulative temperature of a production stage represents the result of the reactor temperature accumulating over the production duration within a production stage. The value increases as the reactor temperature rises or the production duration extends. When calculating the cumulative temperature of each production stage, the reactor temperature value uses the same temperature unit, and the time length uses the same time unit. The unit of measurement for the cumulative temperature of a production stage is the product of the temperature unit and the time unit used.

[0036] Furthermore, for each completed resin batch, each production stage in the production stage sequence is taken as the current production stage. The raw material inspection records, actual material input records, the node time of the production process node that has been reached, the duration of the completed production stage, and the cumulative temperature of the production stage that have been reached up to the end of the current production stage are summarized to form the current stage production data of each production stage.

[0037] It should be noted that the current stage production data refers to the production process records obtained by the resin batch production up to the end of the current production stage. Production process records formed after the end of the current production stage are not included in the current stage production data, and the test values ​​of the predicted quality indicators in the quality inspection records are not included in the current stage production data.

[0038] S2. Divide the current stage production data according to the raw material feeding status and temperature process status to obtain the raw material feeding data combination and the temperature process data combination, and merge them to obtain the comprehensive production data combination. Input the three types of data combinations into the first basic learner, the second basic learner and the third basic learner respectively. Use the detection value of the quality index to be predicted and the cumulative temperature of the next production stage as training control data to obtain the predicted value of the quality index to be predicted and the predicted value of the cumulative temperature of the next production stage.

[0039] Furthermore, in resin batch production, the predicted quality indicators are simultaneously affected by differences in raw material quality, differences in feeding execution, and differences in the temperature processes already experienced within the reactor. Raw material quality differences are reflected in raw material inspection values, while differences in feeding execution are reflected in the actual amount of material fed and the time elapsed during feeding. The production records representing these raw material inspection values, actual amount of material fed, and feeding time are considered the raw material feeding status. Temperature process differences are reflected in the duration of completed production stages and the cumulative temperature of those stages. The production records representing these durations and cumulative temperature are considered the temperature process status. Therefore, raw material feeding data is compiled from the current stage production data according to the raw material feeding status, and temperature process data is compiled from the current stage production data according to the temperature process status. Finally, the raw material feeding data and temperature process data are merged into a comprehensive production data set.

[0040] Specifically, the raw material input data combination includes raw material inspection values, actual input quantity, and input time. Each raw material is arranged according to the input order in the same production process route. When there are multiple raw material inspection items for the same raw material, for each completed resin batch under the same resin product and production process route, the corresponding raw material inspection values ​​are arranged according to the same raw material inspection item order. The input time is the time difference between the corresponding input time and the first input time. The temperature process data combination includes the stage duration from the first production stage to the current production stage and the cumulative temperature of the production stage, and is arranged according to the production stage sequence. The comprehensive production data combination includes raw material input data combination and temperature process data combination, and is arranged in the order of raw material input data combination first and temperature process data combination second.

[0041] It should be noted that, for the same production stage, the raw material input data combination, temperature process data combination, and comprehensive production data combination for each completed resin batch adopt the same arrangement rules. In the raw material input data combination, the same arrangement position corresponds to the same raw material inspection item, actual input quantity, or input time of the same raw material. In the temperature process data combination, the same arrangement position corresponds to the stage duration or cumulative temperature of the same production stage. Batch identifier, resin product model, raw material batch number, raw material name, and production process node name are used for recording, screening, association, and sorting, and are not written into the raw material input data combination, temperature process data combination, and comprehensive production data combination.

[0042] Furthermore, each production stage is processed sequentially according to the production stage sequence. For the first production stage, the raw material input data combination, temperature process data combination, and comprehensive production data combination of each completed resin batch are used as training inputs. For subsequent production stages, the temperature cumulative prediction difference is the difference between the current production stage's temperature cumulative value and the predicted temperature cumulative value for the next production stage output by the same basic learner from the previous production stage. The corresponding data combination of the current production stage, the predicted quality index value output by the same basic learner from the previous production stage, and the temperature cumulative prediction difference are all used as training inputs for the same basic learner. The corresponding data combination for the first basic learner is the raw material input. The data combination for the material input is as follows: the corresponding data combination for the second basic learner is the temperature process data combination, and the corresponding data combination for the third basic learner is the comprehensive production data combination. Based on the batch identifier, the predicted quality index detection value of the same completed resin batch is read, and this value is used as the training control data for predicting the predicted quality index. When the current production stage is not the last production stage in the production stage sequence, the cumulative temperature of the production stage immediately following the current production stage in the production stage sequence of the same completed resin batch is read. This cumulative temperature is used as the cumulative temperature of the next production stage, and the cumulative temperature of the next production stage is used as the training control data for predicting the cumulative temperature.

[0043] It should be noted that, for each production stage in the production stage sequence, after the training inputs for each completed resin batch are formed, for any completed resin batch, the first, second, and third basic learners corresponding to the current production stage are trained using the corresponding training inputs and corresponding training control data of the other completed resin batches excluding the current completed resin batch. Then, the trained first, second, and third basic learners, based on the corresponding training inputs of the current completed resin batch, respectively generate the predicted values ​​of the first, second, and third quality indicators to be predicted for the current completed resin batch in the current production stage. When the current production stage is not the last production stage in the production stage sequence, the corresponding predicted value of the cumulative temperature for the next production stage is also generated. Each completed resin batch is processed one by one in the same way to generate the corresponding predicted value for each completed resin batch in the current production stage.

[0044] Furthermore, for each production stage in the production stage sequence, a first, second, and third basic learner are trained respectively, combining the corresponding training input and corresponding training control data. The first, second, and third basic learners all employ the CART (Classification and Regression Tree) algorithm to construct regression trees, achieving continuous numerical prediction. The training input for the first basic learner is a combination of raw material input data, and when the current production stage is not the first stage in the production stage sequence, the predicted values ​​of the quality indicators to be predicted and the predicted difference of the cumulative temperature output from the first basic learner of the previous production stage are added. The training input for the second basic learner is a combination of temperature process data, and when the current production stage is not the first stage in the production stage sequence, the predicted values ​​of the quality indicators to be predicted and the predicted difference of the cumulative temperature output from the second basic learner of the previous production stage are added. The training input for the third basic learner is a combination of comprehensive production data, and when the current production stage is not the first stage in the production stage sequence, the predicted values ​​of the quality indicators to be predicted and the predicted difference of the cumulative temperature output from the third basic learner of the previous production stage are added.

[0045] Specifically, each raw material inspection value, each actual input quantity, and each input time in the raw material input data combination are used as input variables for the first basic learner; the duration of each stage and the cumulative temperature of each production stage in the temperature process data combination are used as input variables for the second basic learner; and each value in the comprehensive production data combination is used as input variables for the third basic learner. When the current production stage is not the first production stage in the production stage sequence, the predicted value of the quality indicator to be predicted and the predicted difference of the cumulative temperature output by the same basic learner in the previous production stage are also used as input variables for the same basic learner. When the current production stage is not the last production stage in the production stage sequence, for any basic learner, two CART regression trees are trained with the same training input. One CART regression tree uses the detected value of the quality indicator to be predicted as the training control data, and the other CART regression tree uses the cumulative temperature of the next production stage as the training control data. When the current production stage is the last production stage in the production stage sequence, for any basic learner, only the CART regression tree using the detected value of the quality indicator to be predicted as the training control data is trained.

[0046] When training a CART regression tree, training samples from the same production stage are placed at the root node. Input variables are iterated one by one, and the training samples are arranged in ascending order of their values ​​on the current input variable. The median value of two adjacent different values ​​is used as a candidate split point. For each candidate split point, training samples whose current input variable value is less than or equal to the candidate split point are assigned to the left child node, and training samples whose current input variable value is greater than the candidate split point are assigned to the right child node. The sum of the squared differences between the training control data in the left child node and the average value of the training control data in the left child node, and the sum of the squared differences between the training control data in the right child node and the average value of the training control data in the right child node are calculated. The sum of the two sums of squared differences is used as the sum of squared errors of the candidate split point. The input variable and the candidate split point with the smallest sum of squared errors are selected to complete a binary split.

[0047] Continue to select candidate split points and perform binary splits on the child nodes formed by the binary split until there is only one training sample left in the child node, or there are no candidate split points, or all candidate split points cannot reduce the sum of the squared differences between the training and control data in the current node. The child node that stops splitting is taken as the leaf node, and the average value of the training and control data of the training samples in the leaf node is taken as the output of the leaf node.

[0048] It should be noted that the total squared difference of the training and control data within the current node refers to the total squared difference between each training and control data point within the current node and the average value of the training and control data within the current node; the ability of a candidate split point to reduce the total squared difference of the training and control data within the current node means that after dividing the training samples into left and right child nodes according to the candidate split point, the sum of the total squared differences between the left and right child nodes is less than the total squared difference before the split at the current node; when the sum of the squared errors formed by all candidate split points is greater than or equal to the total squared difference before the split at the current node, it indicates that none of the candidate split points can reduce the total squared difference of the training and control data within the current node.

[0049] When predicting the first production stage of the current resin batch, the same production stage division rules, data processing rules, and arrangement rules as those for completed resin batches are used to obtain the raw material input data combination, temperature process data combination, and comprehensive production data combination for the current resin batch in the current production stage. The raw material input data combination is input into the first basic learner, the temperature process data combination into the second basic learner, and the comprehensive production data combination into the third basic learner. For any trained CART regression tree, starting from the root node, the input variables and candidate split points recorded in the nodes are read. The values ​​of the current resin batch on the input variables are compared with the candidate split points, and... The search proceeds layer by layer according to the rule that if the value is less than or equal to the candidate split point, it enters the left child node; if the value is greater than the candidate split point, it enters the right child node. This process continues until a leaf node is reached. When the CART regression tree uses the detection value of the quality indicator to be predicted as the training control data, the leaf node outputs of the corresponding CART regression trees in the first, second, and third basic learners are used as the predicted values ​​of the first, second, and third quality indicators to be predicted, respectively. When the CART regression tree uses the cumulative temperature of the next production stage as the training control data, the output of the leaf node read is used as the predicted value of the cumulative temperature of the next production stage.

[0050] It should be noted that, in this embodiment, the basic learner refers to a learner that takes a combination of data as its basic input and outputs continuous predicted values ​​along the production stages. When the current production stage is not the last production stage in the production stage sequence, the same training input corresponds to two training control data: the detected value of the quality indicator to be predicted and the cumulative temperature of the next production stage. The production meaning and measurement unit of the two training control data are different. Two CART regression trees are trained separately, and quality prediction output and temperature cumulative prediction output are formed respectively. The leaf node output represents the average level of the completed resin batch falling into the same leaf node on the corresponding training control data. After the current resin batch falls into the leaf node according to the segmentation rules of the same CART regression tree, the leaf node output is read as the corresponding predicted value. When the quality indicator to be predicted is any one of viscosity, acid value, solid content, epoxy equivalent, or gel time, the CART regression tree used to obtain the predicted value of the quality indicator to be predicted is trained with the corresponding detected value. When the same resin batch needs to predict multiple quality indicators, the corresponding CART regression tree is trained according to each quality indicator to be predicted, and the predicted value of each quality indicator to be predicted is obtained respectively.

[0051] S3. When it is not the first production stage, the temperature cumulative amount of the current production stage is compared with the temperature cumulative amount prediction value of the next production stage output by the three basic learners of the previous production stage to form the temperature cumulative amount prediction difference. Combined with the corresponding data combination of the current production stage and the feedback input according to the basic learner number, the predicted value of the quality index to be predicted output by the three basic learners, as well as the temperature cumulative amount prediction value of the next production stage that is not the last production stage, are updated.

[0052] Furthermore, when predicting the current resin batch, if the current production stage is not the first production stage in the production stage sequence, the cumulative temperature of the current production stage and the predicted cumulative temperature of the next production stage output by the first, second, and third basic learners of the previous production stage are read. The difference between the cumulative temperature of the current production stage and the predicted cumulative temperature of the next production stage output by the first basic learner of the previous production stage is taken as the first cumulative temperature prediction difference, and the difference between the cumulative temperature of the current production stage and the predicted cumulative temperature of the next production stage output by the second basic learner of the previous production stage is taken as the second cumulative temperature prediction difference. The temperature cumulative prediction difference is the difference between the current production stage's temperature cumulative value and the predicted temperature cumulative value for the next production stage output by the third basic learner from the previous production stage. A positive temperature cumulative prediction difference indicates that the current production stage's temperature cumulative value is higher than the predicted temperature cumulative value for the next production stage output by the same basic learner from the previous production stage. A negative temperature cumulative prediction difference indicates that the current production stage's temperature cumulative value is lower than the predicted temperature cumulative value for the next production stage output by the same basic learner from the previous production stage. A zero temperature cumulative prediction difference indicates that the two values ​​are the same.

[0053] Specifically, the raw material input data combination of the current production stage, the predicted value of the quality index to be predicted output by the first basic learner of the previous production stage, and the predicted difference of the first cumulative temperature are input into the first basic learner. The temperature process data combination of the current production stage, the predicted value of the quality index to be predicted output by the second basic learner of the previous production stage, and the predicted difference of the second cumulative temperature are input into the second basic learner. The comprehensive production data combination of the current production stage, the predicted value of the quality index to be predicted output by the third basic learner of the previous production stage, and the predicted difference of the third cumulative temperature are input into the third basic learner.

[0054] After receiving feedback input, when the current production stage is not the last production stage in the production stage sequence, the first basic learner outputs the predicted value of the first quality indicator to be predicted for the current production stage and the predicted value of the cumulative temperature for the next production stage; the second basic learner outputs the predicted value of the second quality indicator to be predicted for the current production stage and the predicted value of the cumulative temperature for the next production stage; and the third basic learner outputs the predicted value of the third quality indicator to be predicted for the current production stage and the predicted value of the cumulative temperature for the next production stage. When the current production stage is the last production stage in the production stage sequence, the first basic learner outputs the predicted value of the first quality indicator to be predicted for the current production stage; the second basic learner outputs the predicted value of the second quality indicator to be predicted for the current production stage; and the third basic learner outputs the predicted value of the third quality indicator to be predicted for the current production stage.

[0055] It should be noted that the predicted cumulative temperature value for the next production stage output by the same basic learner from the previous production stage represents the value obtained by predicting the cumulative temperature of the current production stage based on the already formed production data from the previous production stage; the cumulative temperature of the current production stage represents the actual cumulative temperature formed in the current production stage; the predicted difference in cumulative temperature represents the deviation between the actual cumulative temperature formed in the current production stage and the cumulative temperature predicted by the same basic learner from the previous production stage. Feeding the predicted difference in cumulative temperature back into the basic learner of the current production stage allows the prediction process of the current production stage to simultaneously refer to the deviation information between the current production stage's production status and the prediction results of the previous production stage.

[0056] It should also be noted that during feedback input, the predicted difference of the cumulative temperature, combined with the data corresponding to the current production stage and the predicted value of the quality index to be predicted output by the same basic learner in the previous production stage, are jointly input into the basic learner of the current production stage. This ensures that the prediction result of the current production stage inherits the quality prediction result of the previous production stage based on the current actual production status, and continues to be updated in combination with the deviation of the cumulative temperature. The first, second, and third predicted differences of the cumulative temperature are fed back to the first, second, and third basic learners, respectively, to avoid the mixed use of prediction deviations formed under different basic learners.

[0057] S4. Input the predicted values ​​of the quality indicators to be predicted from the three basic learners in the current production stage into the integrated learner. When it is not the first production stage, add the integrated predicted value of the quality indicators to be predicted from the previous production stage and the predicted difference of the three cumulative temperature values ​​to obtain the integrated predicted value of the quality indicators to be predicted in the current production stage and form the quality prediction result of the current resin batch.

[0058] Furthermore, the ensemble learner for each production stage is trained sequentially according to the production stage sequence. For each production stage, the predicted values ​​of the first, second, and third quality indicators of each completed resin batch are read to form the basic prediction combination corresponding to each completed resin batch. The training input of the ensemble learner corresponding to each completed resin batch is formed according to the position of the current production stage in the production stage sequence.

[0059] Specifically, for any completed resin batch, if the current production stage is the first production stage in the production stage sequence, the corresponding basic prediction combination is used as the training input for the ensemble learner. If the current production stage is not the first production stage in the production stage sequence, the ensemble prediction values ​​of the quality indicators to be predicted, the first cumulative temperature prediction difference, the second cumulative temperature prediction difference, and the third cumulative temperature prediction difference of the same completed resin batch in the previous production stage are read. The basic prediction combination, the ensemble prediction values ​​of the quality indicators to be predicted in the previous production stage, the first cumulative temperature prediction difference, the second cumulative temperature prediction difference, and the third cumulative temperature prediction difference are used together as the training input for the ensemble learner. Among them, the ensemble prediction values ​​of the quality indicators to be predicted in the previous production stage are formed by the ensemble learner trained in the previous production stage based on the ensemble learner training inputs corresponding to the other completed resin batches (excluding the current completed resin batch) and the detection values ​​of the quality indicators to be predicted, and based on the ensemble learner training inputs of the current completed resin batch in the previous production stage.

[0060] When training the ensemble learner, for any completed resin batch, the values ​​in the ensemble learner training inputs for all other completed resin batches (excluding the current batch) are used as input variables, and the predicted quality index detection values ​​for the other completed resin batches are used as training control data. The ensemble learner is implemented using a CART regression tree, and its training process is the same as that of the CART regression tree in the basic learner. After training, the trained ensemble learner obtains the ensemble predicted value of the predicted quality index for the current completed resin batch at the current production stage based on the ensemble learner training inputs for the current completed resin batch. Each completed resin batch is processed in the same way to form the ensemble predicted value of the predicted quality index for each completed resin batch at the current production stage. After the ensemble predicted values ​​of the predicted quality index for each completed resin batch are formed, the values ​​in the ensemble learner training inputs for all completed resin batches are used as input variables, and the predicted quality index detection values ​​for the corresponding completed resin batches are used as training control data to train the ensemble learner for the current production stage.

[0061] After the ensemble learner is trained, when predicting the current resin batch, it reads the predicted values ​​of the first, second, and third quality indicators to be predicted for the current production stage to form the current basic prediction combination. If the current production stage is the first production stage in the production stage sequence, the current basic prediction combination is input into the ensemble learner for the current production stage to obtain the ensemble predicted value of the quality indicators to be predicted for the current production stage. If the current production stage is not the first production stage in the production stage sequence, it reads the ensemble predicted value of the quality indicators to be predicted from the previous production stage, the predicted difference of the first cumulative temperature, the predicted difference of the second cumulative temperature, and the predicted difference of the third cumulative temperature, and inputs the current basic prediction combination, the ensemble predicted value of the quality indicators to be predicted from the previous production stage, the predicted difference of the first cumulative temperature, the predicted difference of the second cumulative temperature, and the predicted difference of the third cumulative temperature together into the ensemble learner for the current production stage to obtain the ensemble predicted value of the quality indicators to be predicted for the current production stage.

[0062] The predicted values ​​of the quality indicators to be predicted at the current production stage are integrated and recorded together with the batch identifier, resin product model, current production stage, and name of the quality indicator to be predicted for the current resin batch to form the quality prediction result for the current resin batch. The quality prediction result for the current resin batch includes the batch identifier, resin product model, current production stage, name of the quality indicator to be predicted, and quality prediction value. The quality prediction value is the integrated predicted value of the quality indicators to be predicted for the current production stage. After the next production stage ends and the current stage production data for the next production stage is formed, the integrated predicted value of the quality indicators to be predicted formed in the next production stage is used as the new quality prediction value, and the next production stage is used as the new current production stage and written into the quality prediction result for the current resin batch.

[0063] It should be noted that the predicted values ​​of the first, second, and third quality indicators to be predicted are the basic prediction results output by the first, second, and third basic learners, respectively. The integrated learner receives the basic prediction combination in the first production stage, and in non-first production stages, it receives the basic prediction combination, the integrated prediction value of the quality indicators to be predicted in the previous production stage, and the prediction difference of the three cumulative temperature values, and outputs the integrated prediction value of the quality indicators to be predicted. The integrated prediction value of the quality indicators to be predicted is used as the quality prediction value in the current resin batch quality prediction result.

[0064] It should also be noted that when the quality indicator to be predicted is viscosity, the predicted quality value represents the viscosity value predicted for the current resin batch at the current production stage; when the quality indicator to be predicted is acid value, solid content, epoxy equivalent, or gel time, the predicted quality value represents the acid value, solid content, epoxy equivalent, or gel time value predicted for the current resin batch at the current production stage, respectively.

[0065] This embodiment also provides a computer device applicable to the resin batch quality prediction method based on ensemble learning, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the resin batch quality prediction method based on ensemble learning as proposed in the above embodiment.

[0066] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0067] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the resin batch quality prediction method based on ensemble learning as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0068] In summary, this invention achieves differentiated formation and integration of predicted values ​​for quality indicators under different production states by inputting raw material input data combinations, temperature process data combinations, and comprehensive production data combinations into corresponding basic learners; by feeding back the predicted difference of cumulative temperature according to the basic learner sequence number and adding the integrated predicted value of the quality indicators to be predicted in the previous production stage to the integrated learner, the current production stage prediction is updated to reflect the deviation of the actual temperature process and the prediction results of the previous production stage, forming a continuously corrected quality prediction result for the current resin batch as the production stage progresses.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting resin batch quality based on ensemble learning, characterized in that, include: Collect completed resin batch records, divide the production into stages according to the production process nodes, calculate the cumulative temperature of each production stage by using the reactor temperature records and sampling time of each production stage, and form the current stage production data of each production stage. The current stage of production data is divided into raw material input status and temperature process status to obtain raw material input data combination and temperature process data combination, which are then merged to obtain comprehensive production data combination. The three types of data combinations are input into the first basic learner, the second basic learner and the third basic learner respectively. The detection value of the quality index to be predicted and the cumulative temperature of the next production stage are used as training control data to obtain the predicted value of the quality index to be predicted and the predicted value of the cumulative temperature of the next production stage. When it is not the first production stage, the temperature cumulative amount of the current production stage is compared with the temperature cumulative amount prediction values ​​of the next production stage output by the three basic learners of the previous production stage to form the temperature cumulative amount prediction difference. Combined with the corresponding data combination of the current production stage and the feedback input according to the basic learner number, the predicted values ​​of the quality indicators to be predicted output by the three basic learners, as well as the temperature cumulative amount prediction values ​​of the next production stage that are not the last production stage, are updated. The predicted values ​​of the quality indicators to be predicted output by the three basic learners in the current production stage are input into the integrated learner. When it is not the first production stage, the integrated predicted value of the quality indicators to be predicted in the previous production stage and the predicted difference of the three cumulative temperature values ​​are added to obtain the integrated predicted value of the quality indicators to be predicted in the current production stage, and the quality prediction result of the current resin batch is formed.

2. The resin batch quality prediction method based on ensemble learning as described in claim 1, characterized in that, The process of dividing production into stages according to process nodes, and calculating the cumulative temperature of each production stage based on the reactor temperature records and sampling times of each stage, includes: The production process nodes are arranged in chronological order according to their time. The production process between two adjacent production process nodes is considered as a production stage. The time of the previous production process node is taken as the start time of the production stage, and the time of the next production process node is taken as the end time of the production stage. The time difference between the end time and the start time of the production stage is taken as the stage duration. The production stages are arranged in chronological order according to their start times to obtain the production stage sequence. For each production stage in the production stage sequence, the start time and end time of the production stage are used as the temperature accumulation boundary. When the temperature accumulation boundary coincides with the sampling time, the reactor temperature value at the coincident sampling time is used. When the temperature accumulation boundary is located between two adjacent sampling times, linear interpolation is performed according to the time position of the two adjacent sampling times and the reactor temperature value. The average value of the reactor temperature at the two times is accumulated in chronological order for each adjacent time interval, and the product of the average value of the reactor temperature at the two times and the time length between the two times is accumulated to obtain the temperature accumulation of the production stage.

3. The resin batch quality prediction method based on ensemble learning as described in claim 2, characterized in that, The current stage of production data is divided according to the raw material feeding status and temperature process status, including: The raw material inspection values, actual feeding amount and feeding time are arranged in the feeding sequence according to the same production process route to obtain the raw material feeding data combination. The feeding time is the time difference between the corresponding feeding time and the first feeding time. The duration of each production stage from the first stage to the current stage and the cumulative temperature of each production stage are arranged according to the production stage sequence to obtain a combination of temperature process data. The raw material input data combination and the temperature process data combination are arranged in the order of raw material input data combination first and temperature process data combination second to obtain the comprehensive production data combination.

4. The resin batch quality prediction method based on ensemble learning as described in claim 2 or 3, characterized in that, The process of inputting the three types of data into the first, second, and third basic learners respectively, and using the detected value of the quality indicator to be predicted and the cumulative temperature of the next production stage as training control data, includes: The raw material input data is used as the training input for the first basic learner, the temperature process data is used as the training input for the second basic learner, and the comprehensive production data is used as the training input for the third basic learner. When the current production stage is not the first production stage in the production stage sequence, the predicted value of the quality index to be predicted and the predicted difference of the cumulative temperature output by the same basic learner in the previous production stage are added to the training input of the same basic learner. The predicted quality index detection values ​​of each completed resin batch are used as training control data for predicting the predicted quality index; when the current production stage is not the last production stage in the production stage sequence, the cumulative temperature of the production stage immediately following the current production stage in the same completed resin batch is used as the cumulative temperature of the next production stage, and is used as training control data for predicting the cumulative temperature. For any base learner, train a CART regression tree based on the corresponding training input and training control data. When the current production stage is not the last production stage in the production stage sequence, train two CART regression trees with the same training input. One CART regression tree uses the detection value of the quality indicator to be predicted as the training control data, and the other CART regression tree uses the cumulative temperature of the next production stage as the training control data. When the current production stage is the last production stage in the production stage sequence, train a CART regression tree using the detection value of the quality indicator to be predicted as the training control data. When predicting the first production stage of the current resin batch, the raw material input data combination, temperature process data combination, and comprehensive production data combination are input into the first, second, and third basic learners, respectively. Leaf nodes are found according to the node splitting rules of each CART regression tree. The leaf node outputs of the CART regression trees of the three basic learners, which use the detection values ​​of the quality indicators to be predicted as training control data, are used as the predicted values ​​of the first, second, and third quality indicators to be predicted, respectively. The leaf node outputs of the CART regression trees of the next production stage, which use the cumulative temperature as training control data, are used as the predicted values ​​of the cumulative temperature of the next production stage output by the corresponding basic learner.

5. The resin batch quality prediction method based on ensemble learning as described in claim 4, characterized in that, The step of training a CART regression tree based on the corresponding training input and training control data includes: Place the training samples from the same production stage at the root node, traverse the input variables one by one, arrange the training samples according to their values ​​on the current input variables from smallest to largest, and use the median value of two adjacent different values ​​as candidate split points. For each candidate split point, training samples whose current input variable value is less than or equal to the candidate split point are assigned to the left child node, and training samples whose current input variable value is greater than the candidate split point are assigned to the right child node. The sum of the squared differences between the training control data and the average value of the corresponding training control data in the left and right child nodes is calculated respectively. The sum of the two sums of squared differences is taken as the sum of the squared errors of the candidate split points. The input variable and the candidate split point with the smallest sum of squared errors are selected to complete the binary split. Repeatedly select candidate split points and perform binary splits on the child nodes formed by the binary split until only one training sample remains in the child node, there are no candidate split points, or all candidate split points cannot reduce the sum of the squared differences between the training and control data in the current node. The child node that stops splitting is taken as the leaf node, and the average value of the training and control data of the training samples in the leaf node is taken as the output of the leaf node.

6. The resin batch quality prediction method based on ensemble learning as described in claim 3, characterized in that, The predicted difference in the cumulative temperature formation, combined with the corresponding data combination of the current production stage, is fed back as input according to the basic learner sequence number, including: When the current production stage is not the first production stage in the production stage sequence, the temperature accumulation of the current production stage is subtracted from the predicted temperature accumulation of the next production stage output by the first, second, and third basic learners of the previous production stage, respectively, to obtain the first temperature accumulation prediction difference, the second temperature accumulation prediction difference, and the third temperature accumulation prediction difference. The raw material input data combination of the current production stage, the predicted value of the quality index to be predicted output by the first basic learner of the previous production stage, and the predicted difference of the first temperature cumulative amount are input into the first basic learner. The temperature process data combination of the current production stage, the predicted value of the quality index to be predicted output by the second basic learner of the previous production stage, and the predicted difference of the second temperature cumulative amount are input into the second basic learner. The comprehensive production data combination of the current production stage, the predicted value of the quality index to be predicted output by the third basic learner of the previous production stage, and the predicted difference of the third temperature cumulative amount are input into the third basic learner. When the current production stage is not the last production stage in the production stage sequence, the first basic learner, the second basic learner, and the third basic learner output the predicted values ​​of the first, second, and third unpredictable quality indicators for the current production stage, respectively, and also output the predicted value of the cumulative temperature for the next production stage. When the current production stage is the last production stage in the production stage sequence, the first basic learner, the second basic learner, and the third basic learner output the predicted values ​​of the first, second, and third unpredictable quality indicators for the current production stage, respectively.

7. The resin batch quality prediction method based on ensemble learning as described in claim 6, characterized in that, The process involves inputting the predicted quality indicators output by the three basic learners of the current production stage into the ensemble learner, and, in non-first production stages, adding the ensemble predicted quality indicators from the previous production stage and the predicted differences of the three cumulative temperature values ​​to obtain the ensemble predicted quality indicators for the current production stage, including: The integrated learners for each production stage are trained sequentially according to the production stage sequence. The predicted values ​​of the quality indicators to be predicted for each completed resin batch in the current production stage are read from the first basic learner, the second basic learner and the third basic learner. The basic prediction combination corresponding to each completed resin batch is formed in the order of the first basic learner, the second basic learner and the third basic learner. When the current production stage is the first production stage in the production stage sequence, the basic prediction combination is used as the training input of the ensemble learner; when the current production stage is not the first production stage in the production stage sequence, the basic prediction combination, the integrated prediction value of the quality index to be predicted of the same completed resin batch in the previous production stage, the prediction difference of the first temperature cumulative amount, the prediction difference of the second temperature cumulative amount, and the prediction difference of the third temperature cumulative amount are used together as the training input of the ensemble learner. Each value in the training input of the ensemble learner corresponding to each completed resin batch is used as an input variable, and the detection value of the quality index to be predicted for the corresponding completed resin batch is used as training control data. The ensemble learner is trained using a CART regression tree. After training, the training input of the ensemble learner corresponding to each completed resin batch is input into the ensemble learner to obtain the ensemble predicted value of the quality index to be predicted for each completed resin batch in the current production stage. The ensemble predicted value of the quality index to be predicted in the previous production stage is obtained by the ensemble learner of the previous production stage based on the training input of the ensemble learner corresponding to the same completed resin batch. When predicting the current resin batch, the predicted values ​​of the quality indicators to be predicted output by the three basic learners of the current production stage are arranged in the order of the first basic learner, the second basic learner, and the third basic learner to form the current basic prediction combination. When the current production stage is the first production stage in the production stage sequence, the current basic prediction combination is input into the integrated learner of the current production stage. When the current production stage is not the first production stage in the production stage sequence, the current basic prediction combination, the integrated predicted value of the quality indicators to be predicted from the previous production stage, the predicted difference of the first cumulative temperature, the predicted difference of the second cumulative temperature, and the predicted difference of the third cumulative temperature are input into the integrated learner of the current production stage to obtain the integrated predicted value of the quality indicators to be predicted for the current production stage.

8. The resin batch quality prediction method based on ensemble learning as described in claim 7, characterized in that, The formation of the current resin batch quality prediction result includes: The predicted value of the quality indicators to be predicted at the current production stage is recorded together with the batch identifier, resin product model, current production stage and name of the quality indicator to be predicted for the current resin batch to form the quality prediction result of the current resin batch. The quality prediction result of the current resin batch includes the batch identifier, resin product model, current production stage, name of the quality indicator to be predicted and quality prediction value. The quality prediction value is the integrated predicted value of the quality indicators to be predicted at the current production stage. Once the next production stage ends and the current stage production data for the next production stage is generated, the predicted values ​​of the quality indicators to be predicted generated in the next production stage are integrated as new quality prediction values, and the next production stage is taken as the new current production stage and written into the current resin batch quality prediction results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the resin batch quality prediction method based on ensemble learning as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the resin batch quality prediction method based on ensemble learning as described in any one of claims 1 to 8.

Images