A training method, system and device of a personalized prediction model and a storage medium

Abstract

The application discloses a kind of training method, system, device and storage medium of personalized prediction model.Therein, training method includes: obtaining the running data of equipment, and according to artificial experience, candidate factor possibly influencing target result is selected;From the historical running data of equipment, candidate factor data and historical result data corresponding to candidate factor and target result are respectively called;From candidate factor, the influencing factor in linear relationship with target result is screened;Candidate factor data corresponding to influencing factor is used as historical factor data, and historical factor data, and historical result data corresponding thereto are used as training data;According to training data and preset training algorithm, prediction model capable of reflecting the relationship between influencing factor and target result is constructed and trained.The training method can quickly convert artificial experience into professional prediction model, so that non-algorithm professionals can also conduct algorithm training according to their own needs to obtain personalized prediction model.

Description

Technical Field

[0001] This invention belongs to the field of industrial control technology, and specifically relates to a training method, system, device, and storage medium for a personalized prediction model. Background Technology

[0002] Linear regression is a statistical technique that makes predictions by establishing a linear relationship between variables. Its core is to fit the best linear model by minimizing the difference between the predicted value and the actual value. It is characterized by its simple logic and the direct use of standard function libraries, and is widely used in fields such as image processing and prediction systems.

[0003] However, when using linear regression techniques to build and train predictive models for machine learning applications, a specialized model training system (such as a personal computer or enterprise server) is essential. The model training process involves several stages: data collection, data analysis, model selection, model training, model validation, and implementation. Each stage requires professionals to process data within this specialized system. Therefore, to use a specific predictive model on a non-specialized system (such as a PLC or Programmable Logic Controller), the equipment manufacturer's engineers must create and train the model on a specialized training system before shipping, and then transfer the trained model to the non-specialized system. However, this presents several challenges. First, due to inherent errors between different devices of the same model, the operating data from one device cannot be perfectly replicated on another. Second, even the same device often produces different products, processes, and requirements, resulting in significant differences in production parameters. Therefore, pre-setting a universally usable model before shipping is impractical.

[0004] Once equipment is put into production, users can immediately access its operational data and have a greater say in its daily operation. However, due to their own technical limitations, they cannot create and train models independently, nor can they translate personal experience into production technology. This leaves users with two choices: either lower their standards, relying on manual experience and regular maintenance for troubleshooting, sometimes only realizing the problem when product defects occur; or incur additional manpower and financial costs to have the manufacturer's engineers come to the site and train models tailored to the specific needs of users, products, and processes. In either case, user acceptance is low, and for equipment manufacturers, compromising user experience is tantamount to self-destruction—a lose-lose situation. Summary of the Invention

[0005] To address the aforementioned problems, this invention discloses a training method, system, apparatus, and storage medium for a personalized prediction model, thereby overcoming these issues.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention discloses a method for training a personalized prediction model, the method comprising: Acquire equipment operating data and, based on human experience, select candidate factors that may affect the target outcome. Based on the results of manual selection, candidate factor data corresponding to the candidate factors and historical result data corresponding to the target result are retrieved from the historical operating data of the equipment. The candidate factor data and the historical result data are presented in a corresponding manner, and the desired influencing factors that have a linear relationship with the target result are manually selected from the candidate factors; The candidate factor data corresponding to the influencing factors are used as historical factor data, and the historical factor data and the corresponding historical result data are used as training data. Based on the training data and the preset training algorithm, a predictive model that can reflect the relationship between the influencing factors and the target result is constructed and trained.

[0007] Furthermore, the process of mapping and presenting the candidate factor data and the historical result data includes: Scatter plots of the candidate factor data and the corresponding historical result data are drawn on the human-computer interaction interface.

[0008] Furthermore, the execution of constructing and training a predictive model that reflects the relationship between the influencing factors and the target result based on the training data and a preset training algorithm includes: Construct a linear regression model for the target outcome and the influencing factors:

[0009] In the formula, Represents the target result; x1 to x n Representing different influencing factors; x n Let ω1 be the nth influencing factor; b is the bias constant; ω1 to ω n Represents the weights corresponding to different influencing factors; The training data is input into the linear regression model to obtain the initial prediction model; The training data is input into the initial prediction model, and the influencing factors and their weights in the initial prediction model are filtered according to the goodness of fit to obtain the prediction model.

[0010] Further, the step of inputting the training data into the linear regression model to obtain the initial prediction model includes: The training data is grouped and input into the linear regression model to obtain multiple sets of weights and bias constants. Each set of weights and bias constants is used to create a linear prediction formula. Multiple linear prediction formulas are constructed based on the multiple sets of weights and bias constants. A loss function is used to minimize the loss function to determine the values ​​of each weight and the bias constant. From multiple linear prediction formulas, one linear prediction formula is selected as the initial prediction model. The loss function is: ,

[0011] Where x represents the influencing factor; ω represents the weight of the influencing factor; b is the bias constant in the linear regression model; h i H represents the target result calculated using a linear prediction formula. i It is the historical result data corresponding to the historical factor data used in the linear prediction formula; m is the number of comparison groups between the target result and the historical result data.

[0012] Further, the step of filtering the influencing factors and their weights in the initial prediction model based on goodness of fit to obtain the execution of the prediction model includes: Calculate the goodness of fit of the initial prediction model using the following formula:

[0013] in, For the initial prediction model The goodness of fit; The true value of the target result corresponding to the training data used in the initial prediction model; The predicted value of the target result calculated by the initial prediction model; The mean of the true values ​​of the target results corresponding to the training data used in the initial prediction model; Determine whether the goodness of fit is greater than or equal to a preset value; if yes, determine the initial prediction model as the prediction model; if not, proceed with the following steps: Delete the influencing factors corresponding to weights less than the set weight threshold and / or delete one or more corresponding influencing factors in ascending order of weight to generate a new round of linear regression model; The training data is input into a new round of linear regression model for training until the goodness of fit is greater than or equal to the preset value.

[0014] Furthermore, the prediction model training method also includes: The accuracy of the prediction model is presented using the training data.

[0015] Furthermore, the execution of presenting the accuracy of the prediction model using the training data includes: The historical factor data is input into the prediction model to calculate the predicted value of the prediction model; The historical results data are compared with the calculated predicted values, a comparison curve is plotted, and the comparison curve is displayed on the human-computer interface. The accuracy of the prediction model is visually presented using the comparison curves.

[0016] Another aspect of the present invention discloses a training system for a prediction model, the training system being capable of executing the aforementioned training method for a personalized prediction model, the training system comprising: The acquisition and selection unit is used to acquire the equipment's operating data and, based on human experience, select candidate factors that may affect the target result. The retrieval unit is used to retrieve candidate factor data corresponding to the candidate factors and historical result data corresponding to the target result from the historical operating data of the equipment based on the manual selection results. A filtering unit is used to present the candidate factor data and the historical result data in a correspondence manner, and to manually filter out the desired influencing factors that have a linear relationship with the target result from the candidate factors based on the presentation results. The training data construction unit is used to take the candidate factor data corresponding to the influencing factor as historical factor data, and the historical factor data and the corresponding historical result data as training data. A construction and training unit is used to construct and train a predictive model that reflects the relationship between the influencing factors and the target result based on the training data and a preset training algorithm.

[0017] Another aspect of the present invention discloses a prediction model training apparatus, the prediction model training apparatus comprising: Processor; and Memory for storing the executable instructions of the processor; The processor executes the executable instructions to enable the prediction model training device to implement the personalized prediction model training method described above.

[0018] The final aspect of this invention discloses a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implements the training method for the personalized prediction model described above.

[0019] The advantages and beneficial effects of this invention are: The predictive model training method in this invention allows non-algorithm professionals to simply screen candidate influencing factors based on human experience and rely on the predictive algorithms built into non-professional systems (including offline control systems such as PLCs) to automatically create complex predictive models. This quickly transforms human experience into a professional predictive model, enabling non-algorithm professionals to train algorithms according to their own needs and obtain personalized predictive models, greatly improving the intelligence level of equipment and user experience. Furthermore, the predictive model trained using this method can achieve relatively accurate result predictions. Attached Figure Description

[0020] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a diagram illustrating the implementation steps of a training method for a personalized prediction model in one embodiment of the present invention. Figure 2 This is a schematic diagram of the predictive model training device in another embodiment of the present invention; Figure 3 This is a scatter plot of injection time and pad position in another embodiment of the present invention; Figure 4 This is a scatter plot of plasticizing time and pad position in another embodiment of the present invention; Figure 5 This is a scatter plot of the plasticizing process and the position of the material pad in another embodiment of the present invention; Figure 6 This is a scatter plot of the pressure holding switching pressure and the pad position in another embodiment of the present invention; Figure 7 This is a scatter plot showing the maximum pressure and pad position in another embodiment of the present invention; Figure 8 This is a scatter plot of the maximum back pressure and pad position in another embodiment of the present invention; Figure 9 This is a scatter plot of the maximum injection speed and pad position in another embodiment of the present invention; Figure 10This is a scatter plot showing the switching position and the pad position in another embodiment of the present invention; Figure 11 This is a comparison curve of historical pad position and predicted pad position in another embodiment of the present invention. Detailed Implementation

[0021] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0022] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or sets.

[0023] To keep the drawings concise, only the parts relevant to the invention are shown schematically in each figure, and they do not represent the actual structure of the product. Furthermore, for ease of understanding, in some figures, components with the same structure or function are shown only schematically, or only one is labeled. In this document, "a" not only means "only one," but can also mean "more than one."

[0024] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without any creative effort.

[0026] One embodiment of the present invention provides a training method for a personalized prediction model, such as... Figure 1 As shown, the training method includes the following steps: Step 01: Obtain the equipment's operating data and, based on human experience, select candidate factors that may affect the target outcome. The target outcome is the result that needs to be predicted.

[0027] Understandably, the process involves first determining the desired prediction result, then acquiring extensive operational data from the equipment, and finally, based on the desired prediction result, selecting potential influencing factors from the extensive operational data using human experience. This allows human experience to be incorporated into a reproducible algorithm, simplifying the model training process and lowering the barrier to entry. The selection of potential influencing factors from the equipment's operational data can be performed on the control system's human-machine interface, further simplifying the model training process.

[0028] Step 02: Based on the results of manual selection, retrieve the candidate factor data corresponding to the candidate factors and the historical result data corresponding to the target result from the historical operation data of the equipment.

[0029] Understandably, based on the equipment's operating data over a period of time, such as the equipment's operating data over the past month, two months, one year, or two years, candidate factor data associated with the candidate factors and historical result data corresponding to the target result can be selected from the historical operating data. This selection process can also be performed on the human-machine interface of the control system, which can also simplify the operation steps of model training.

[0030] Step 03: The candidate factor data and historical result data are presented in a corresponding manner. Based on the presentation results, the desired influencing factors that have a linear relationship with the target result are manually selected from the candidate factors.

[0031] After obtaining candidate factor data that may affect the target result from the historical operation data of the equipment, as well as the historical result data corresponding to the target result, the influencing factors that have a linear relationship with the target result are further screened from the candidate factors. The screened influencing factors and the corresponding target result are in a linear relationship, that is, the influencing factors and the corresponding target result are positively correlated or negatively correlated. Candidate factors that do not have a linear relationship with the corresponding target result are deleted.

[0032] Understandably, users manually select candidate factors and their corresponding target results that exhibit linear correlation on the human-computer interface. This means that as a candidate factor increases or decreases along the X-axis, the corresponding target result also increases or decreases along the Y-axis. Based on the user's selection, the system retains the selected candidate factors and deletes those that do not have a linear relationship with the target result.

[0033] Step 04: Use the candidate factor data corresponding to the influencing factors as historical factor data, and use the historical factor data and the corresponding historical result data as training data.

[0034] Specifically, sample data consisting of historical factor data and historical result data that have a linear relationship are used as training data for subsequent model training.

[0035] Step 05: Based on the training data and the preset training algorithm, construct and train a predictive model that reflects the relationship between influencing factors and the target outcome. The preset training algorithm can be built into non-professional systems. This part is automatically trained based on user selection, allowing users (especially those without algorithm expertise) to train models according to their needs, outputting personalized predictive models and improving user experience.

[0036] Understandably, historical factor data from the training data is used as input to the x-value in the training algorithm, and historical result data from the training data is used as input to the y-value in the training algorithm. The weight and bias constant before each historical factor (i.e. x-value) are calculated, and then the prediction model is obtained.

[0037] In this way, model training, which should have been done by professionals using professional systems, can now be done automatically by embedding the prediction algorithm into non-professional systems (such as offline control systems like PLCs). Non-professionals only need to perform simple data filtering in the non-professional system to complete the personalized prediction model, and personalized settings can be made according to the user's choices during the process.

[0038] In summary, the prediction model training method in this embodiment allows non-algorithm professionals to simply screen candidate influencing factors based on human experience and rely on the prediction algorithms built into non-professional systems (including offline control systems such as PLCs) to automatically create complex prediction models. This quickly transforms human experience into professional prediction models, enabling non-algorithm professionals to train algorithms according to their own needs and obtain personalized prediction models, greatly improving the intelligence level of the equipment and the user experience. Furthermore, the prediction model trained using this method can achieve relatively accurate result predictions.

[0039] In this embodiment, the execution of the correspondence presentation of candidate factor data and historical result data in step 03 includes: Scatter plots of each candidate factor and its corresponding historical results are displayed on the human-computer interaction interface. Users can filter the candidate factor data themselves based on their experience, or directly select from the system's preset schemes. The system preset schemes are those set up after factory testing, and there can be one or more schemes.

[0040] Furthermore, step 05, which involves constructing and training a predictive model that reflects the relationship between influencing factors and the target outcome based on training data and a preset training algorithm, includes: Step 51, construct a linear regression model for the target outcome and influencing factors:

[0041] In the formula, Represents the target result; x1 to x n Representing different influencing factors; x n Let ω1 be the nth influencing factor; b is the bias constant; ω1 to ω n This represents the weights corresponding to different influencing factors. It's important to note that n is a positive integer greater than or equal to 1. When n is 1, the linear regression model is considered to have only one influencing factor specified, i.e. When n is 2, the linear regression model is considered to have two influencing factors, namely And so on. It's important to note that the calculation of a linear regression model involves dimensionless data processing, specifically a "preprocessing-fusion-inversion" process: first, the original parameters are normalized using known methods such as standardization, transforming them into dimensionless values; then, weighted summation is used to quantitatively aggregate multiple influencing factors; finally, inverse normalization maps the dimensionless fusion result back to the target result. Given that this is a common technique in the field, it will only be briefly explained here.

[0042] Step 52: Input the training data into the linear regression model to obtain the initial prediction model.

[0043] Understandably, training data is input into a linear regression model, resulting in multiple sets of weights ω and bias constants b. One set of weights ω and bias constants b is used to create a linear prediction formula, and multiple sets of weights ω and bias constants b result in multiple linear prediction formulas. Based on the loss function, one linear prediction formula is selected from these multiple formulas as the initial prediction model. Specifically, the loss function is minimized to determine the values ​​of weights ω and bias constants b, thereby selecting one from multiple linear prediction formulas as the initial prediction model.

[0044] The loss function is: ,

[0045] Where x represents the influencing factor; ω represents the weight of the influencing factor; b is the bias constant in the linear regression model; h i This represents the target result calculated using a linear prediction formula; a linear prediction formula yields a predicted value for the target result. iThese are the historical result data corresponding to the historical factor data used in the linear prediction formula, i.e., the detected values ​​of the target result. The number of historical result data points corresponds to the number of sets of historical factor data used in a linear prediction formula. m represents the number of comparison sets between the target result and the historical result data. In short, each linear prediction formula outputs a predicted value for the target result, and there are multiple historical result data points corresponding to each linear prediction formula. Therefore, the number of m is equal to the number of sets of historical factor data used in this linear prediction formula.

[0046] It is important to note that the number of historical factor data sets required for a linear prediction formula is related to the number of selected influencing factors. A linear prediction formula with n influencing factors requires at least n sets of historical factor data input. Taking the prediction of the material pad position in an injection molding machine as an example, the predicted value of the material pad position is the target result. In one injection cycle, the injection parameters obtained include, but are not limited to, injection time, plasticizing time, plasticizing stroke, switching pressure, maximum system pressure, maximum back pressure, switching position, the detected value of the material pad position, and the temperature of each area in the barrel (i.e., including candidate influencing factors and historical result data). After user screening, some or all of the candidate factors will be used as influencing factors. This set of various influencing factors in the same injection cycle constitutes a set of historical factor data. At the same time, the corresponding historical result data needs to be retained. In this embodiment, the historical material pad position is the historical result data, and the various influencing factors and historical result data in the same injection cycle constitute a set of training data.

[0047] Step 53: Input the training data into the initial prediction model, and filter the influencing factors and their weights in the initial prediction model according to the goodness of fit to obtain the prediction model.

[0048] Specifically, step 53 involves filtering the influencing factors and their weights in the initial prediction model based on the goodness of fit, resulting in the execution of the prediction model, which includes: Calculate the goodness of fit of the initial prediction model using the following formula:

[0049] in, For the initial prediction model The goodness of fit; The true value of the target result corresponding to the training data used in the initial prediction model; The predicted value of the target result is obtained by calculating the target result through the initial prediction model; This represents the mean of the true values ​​of the target results corresponding to the training data used in the initial prediction model. It should be noted that... It is a value less than 1, and the closer the value is to 1, the better the model's predictive performance.

[0050] Determine if the goodness of fit is greater than or equal to the preset value; if yes, determine the initial prediction model as the prediction model; if not, proceed with the following steps: Obtain the weight values ​​of each influencing factor in the initial prediction model where the goodness of fit is less than the preset value. Delete influencing factors with weights less than the set weight threshold and / or delete one or more corresponding influencing factors in ascending order of weight. Generate a new linear regression model and input the training data into the new linear regression model for training, i.e., perform steps 52 and 53, until the goodness of fit of the newly established initial prediction model is greater than or equal to the preset value. The preset value is determined based on the specific scenario.

[0051] Understandably, when the goodness of fit is less than a preset value, there are three ways to handle the subsequent processing. The first method is to delete the influencing factors corresponding to weights less than a set weight threshold. This involves setting a weight threshold, comparing each weight in the linear regression model to this threshold, filtering out weights less than the threshold, and deleting the influencing factors corresponding to these selected weights. The weight threshold is determined based on the specific scenario. The second method is to delete one or more corresponding influencing factors in ascending order of weight. This involves taking all weights in the linear regression model, determining one or more weights from the smallest side, and deleting the influencing factors corresponding to these determined weights. The third method is a combination of the first and second methods. Generally, the first method is preferred.

[0052] In this embodiment, the training method further includes: Step 06: Use training data to present the accuracy of the prediction model, so that staff can understand the model's performance based on the presented accuracy, and thus determine whether it can be directly used to predict future target results.

[0053] Specifically, step 06 includes: Step 61: Input the historical factor data from the training data into the prediction model to calculate the prediction result data. That is, take the historical factors from the training data as x terms, input them into the prediction model, and obtain the corresponding prediction result through the prediction model.

[0054] Step 62: Compare the historical results data in the training data with the predicted results data, plot the comparison curve, and display the comparison curve on the human-computer interface.

[0055] This allows for an intuitive understanding of the accuracy of the prediction model through a human-computer interaction interface, facilitating subsequent assessment of the model's predictive accuracy.

[0056] Step 63: Visually present the accuracy of the prediction model based on the comparison curve.

[0057] Understandably, the higher the overlap between historical and predicted data curves, the more accurate the prediction. Once the predicted curve reaches the expected accuracy, it can be used; otherwise, the model should be rebuilt or retrained.

[0058] Another embodiment of the present invention discloses a training system for a prediction model, the training system being capable of executing the above-described training method for a personalized prediction model, the training system comprising: The acquisition and selection unit is used to acquire the equipment's operating data and, based on human experience, select candidate factors that may affect the target result.

[0059] The retrieval unit is used to retrieve candidate factor data corresponding to the candidate factors and historical result data corresponding to the target result from the historical operating data of the equipment based on the results of manual selection.

[0060] The filtering unit is used to present the candidate factor data and historical result data in a corresponding manner, and to manually filter out the desired influencing factors that have a linear relationship with the target result from the candidate factors.

[0061] The training data construction unit is used to take the candidate factor data corresponding to the influencing factors as historical factor data, and the historical factor data and their corresponding historical result data as training data.

[0062] The construction and training unit is used to build and train a predictive model that reflects the relationship between influencing factors and target results based on training data and a preset training algorithm.

[0063] The training system in this embodiment not only simplifies the model training operation steps and lowers the threshold for model training, but also reduces the equipment requirements and professional requirements for personnel in training personalized prediction models. This allows users to quickly train personalized models according to their needs after only simple system operation training, greatly improving the user experience.

[0064] In some embodiments, the filtering unit is specifically used to: draw scatter plots of candidate factor data and corresponding historical result data on the human-computer interaction interface.

[0065] In some embodiments, the construction and training unit is used for: First, construct a linear regression model of the target outcome and influencing factors:

[0066] In the formula, Represents the target result; x1 to xn Representing different influencing factors; x n Let ω1 be the nth influencing factor; b is the bias constant; ω1 to ω n This represents the weights corresponding to different influencing factors.

[0067] The training data is then input into the linear regression model to obtain the initial prediction model.

[0068] Specifically, training data is input into a linear regression model to obtain multiple sets of weights ω and bias constants b. One set of weights ω and bias constants b is used to create a linear prediction formula. Multiple sets of weights ω and bias constants b result in multiple linear prediction formulas. Based on the loss function, one linear prediction formula is selected from these formulas as the initial prediction model. Specifically, the loss function is minimized to determine the values ​​of weights ω and bias constants b, thereby selecting one from multiple linear prediction formulas as the initial prediction model.

[0069] The loss function is: ,

[0070] Where x represents the influencing factor; ω represents the weight of the influencing factor; b is the bias constant in the linear regression model; h i H represents the target result calculated using a linear prediction formula. i It is the historical result data corresponding to the historical factor data used in the linear prediction formula; m is the number of comparison groups between the target result and the historical result data.

[0071] Finally, the training data is input into the initial prediction model, and the influencing factors and their weights in the initial prediction model are filtered according to the goodness of fit to obtain the prediction model.

[0072] Specifically, the influencing factors and their weights in the initial prediction model are screened based on the goodness of fit, resulting in the execution of the prediction model, which includes: Calculate the goodness of fit of the initial prediction model using the following formula:

[0073] in, For the initial prediction model The goodness of fit; The true value of the target result corresponding to the training data used in the initial prediction model; The predicted value of the target result is obtained by calculating the target result through the initial prediction model; It is the average of the true values ​​of the target results corresponding to the training data used in the initial prediction model.

[0074] Determine if the goodness of fit is greater than or equal to the preset value; if yes, determine the initial prediction model as the prediction model; if not, proceed with the following steps: Delete the influencing factors corresponding to weights less than the set weight threshold and / or delete one or more corresponding influencing factors in ascending order of weight to generate a new linear regression model; input the training data into the new linear regression model for training until the goodness of fit is greater than or equal to the preset value.

[0075] In some embodiments, the training system further includes: The accuracy presentation unit is used to present the accuracy of the prediction model using training data. Specifically, the accuracy presentation unit is used to input historical factor data into the prediction model, calculate the predicted value of the prediction model; compare the historical result data with the calculated predicted value, draw a comparison curve, and display the comparison curve on the human-computer interface; and intuitively present the accuracy of the prediction model based on the comparison curve.

[0076] It is understood that the training system of the above-mentioned prediction model can implement each step of the training method of the personalized prediction model provided in the foregoing embodiments. The relevant explanations of the training method of the personalized prediction model are applicable to the training system of the prediction model, and will not be repeated here.

[0077] In another embodiment of the present invention, a predictive model training device is provided, such as... Figure 2 As shown, the prediction model training device includes: The device includes one or more processors (or processing units); it may also include one or more memories coupled to the processor for storing executable instructions of the processor, and may further include a communication module coupled to the processor. The processor executes the executable instructions to enable the predictive model training apparatus to implement the personalized predictive model training method as described in the above embodiments.

[0078] A communication module can be used to communicate with other devices or apparatuses, such as sending or receiving data and / or signals. A communication module may have at least one communication module for communication. A communication module may include any interface necessary for communicating with other devices. Exemplarily, a communication module may be a transceiver, circuit, bus, module, or other type of communication module.

[0079] The processor may include, but is not limited to, one or more of the following: a general-purpose computer, a special-purpose computer, a microcontroller, a digital signal processor (DSP), or a controller-based multi-core controller architecture. The device may have multiple processors, such as application-specific integrated circuit (ASIC) chips, which are time-dependent on a clock synchronized with the main processor.

[0080] The memory may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, at least one of the following: read-only memory (ROM), electrically programmable read-only memory (EPROM), flash memory, hard disk, compact disc (CD), digital video disc (DVD), or other magnetic and / or optical storage. Examples of volatile memories include, but are not limited to, at least one of the following: random access memory (RAM), or other volatile memories that do not persist during the duration of a power outage.

[0081] A computer program consists of computer-executable instructions that are executed by an associated processor. Programs can be stored in ROM. A processor can perform any appropriate action and processing by loading the program into RAM.

[0082] Possible implementations of this application can be achieved through a program, enabling the communication device to execute any of the processes discussed in the foregoing embodiments. Possible implementations of this application can also be achieved through hardware or a combination of software and hardware.

[0083] In some implementations, the program may be tangibly contained in a computer-readable storage medium, which may include in a device (such as in memory) or other storage device accessible by the device. The program may be loaded from the computer-readable storage medium into RAM for execution. The computer-readable storage medium may include any type of tangible non-volatile memory, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc.

[0084] Embodiments of the present invention also provide a computer-readable storage medium having computer instructions or program code stored thereon, which, when executed by a processor, cause the processor to perform the methods and functions involved in any of the above embodiments. The computer-readable medium can be any tangible medium that contains or stores a program for or relating to an instruction execution system, apparatus, or device. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. More detailed examples of computer-readable storage media include electrical connections with one or more wires, magnetic media (e.g., disks, floppy disks, hard disks, magnetic tapes, magnetic storage devices), optical media (e.g., optical storage devices, DVDs), semiconductor media (e.g., solid-state drives), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), or any suitable combination thereof.

[0085] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. Embodiments of this application also provide at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. This computer program product includes one or more computer-executable instructions, such as instructions included in a program module, which execute in a device on a target's real or virtual processor to perform the processes, methods, and functions involved in any of the above embodiments. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0086] Embodiments of the present invention also provide a computer program product, including a computer program or instructions that, when executed on a computer, cause the computer to perform the processes, methods, and functions described in the above embodiments. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided as needed. Machine-executable instructions for program modules can be executed locally or in a distributed device. In a distributed device, program modules can reside in both local and remote storage media.

[0087] Generally, various embodiments of the present invention can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software, which can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are shown and described as block diagrams, flowcharts, or represented using some other illustration, it should be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0088] In injection molding machines, the screw check ring is a consumable component. Check ring failure leads to material shortages, unstable pressure control within the injection molding machine barrel, and ultimately, product scrap. Currently, check ring failure relies primarily on manual judgment, and it's often only detected after complete failure and product defects, resulting in irreparable losses. Therefore, early warning or accurate alerts for check ring failure can effectively reduce equipment downtime and product scrap rates, thereby lowering production costs. The inventors discovered that check ring failure directly causes instability in the material pad position; therefore, monitoring the stability of the material pad position is equivalent to monitoring check ring failure.

[0089] It should be noted that the material pad position refers to the axial position of the screw corresponding to the section of molten plastic remaining at the screw tip after the melting process is complete; that is, the minimum screw position. Furthermore, changes in the material pad position are closely related to multiple independent variables (i.e., influencing factors) during injection molding. These influencing factors include, but are not limited to, plasticizing stroke, holding pressure switching pressure, maximum pressure, and switching position. Changes in these influencing factors ultimately affect the material pad position.

[0090] Based on the above requirements, in a specific embodiment of the present invention, a method for training a prediction model for the position of the material pad on an injection molding machine is provided. The trained prediction model can predict the position of the material pad on the injection molding machine, and thus determine whether the check ring has failed.

[0091] The training method for this prediction model includes the following steps: Step 01: Obtain the operating data of the injection molding machine. Based on manual experience, select candidate operating parameters that may affect the change of the material pad position from the operating data of the injection molding machine. Changes in these candidate operating parameters ultimately lead to changes in the material pad position. Candidate operating parameters include injection time, plasticizing time, plasticizing stroke, holding pressure switching pressure, maximum pressure, maximum back pressure, and switching position. As the equipment runs for a longer time, more and more data will be generated.

[0092] It should be noted that injection time refers to the duration of the molten plastic being injected into the mold cavity, measured in seconds; plasticizing time refers to the time required to melt the resin and prepare it for the next injection; plasticizing stroke refers to the distance the screw retracts during the resin melting process, usually measured in millimeters; holding pressure switching pressure refers to the material pressure at the moment the injection stage transitions to the holding / pressure-holding stage; maximum system pressure refers to the peak pressure reached by the hydraulic press during injection, measured in bar or psi (pounds per square inch); maximum back pressure refers to the material thrust applied to the screw during plasticizing, measured in bar or psi; and switching position refers to the specific position of the screw at the moment it switches from the injection stage (high-speed, high-pressure filling) to the holding stage (low-speed, shrinkage compensation), usually measured in millimeters.

[0093] Step 02: Based on the manual selection results, retrieve the candidate operating parameter data corresponding to the candidate operating parameters and the historical material pad position data corresponding to the pad position from the historical operating data of the injection molding machine.

[0094] Step 03: The candidate operating parameter data and historical material pad position data are presented in a corresponding manner. Based on the presentation results, the required influencing factors that have a linear relationship with the pad position are manually selected from the candidate operating parameters.

[0095] Step 04: Use the candidate operating parameter data corresponding to the influencing factors as historical operating parameter data, and use the historical operating parameter data and the corresponding historical material pad position data as training data.

[0096] Step 05: Based on the training data and the preset training algorithm, construct and train a predictive model that can reflect the relationship between influencing factors and the location of the bedding material.

[0097] Furthermore, step 03, which involves mapping and presenting the candidate operating parameter data and historical material pad position data, includes: Scatter plots of candidate operating parameter data and corresponding historical material pad position data are drawn on the human-computer interaction interface.

[0098] Specifically, a scatter plot is drawn with injection time, plasticizing time, plasticizing stroke, holding pressure switching pressure, maximum system pressure, maximum back pressure, and switching position as x, and material pad position as y.

[0099] In addition, the execution of step 05 includes: First, construct a linear regression model regarding the location of the pad and influencing factors:

[0100] In the formula, The x represents the material pad position, x represents the injection parameters, x1 is the plasticizing stroke, x2 is the holding pressure switching pressure, x3 is the maximum pressure, and x4 is the switching position. n Let ω be the nth influencing factor in the injection molding process, b be the bias constant, ω represent the weight of the influencing factor, ω1 be the weight of the plasticizing stroke, ω2 be the weight of the holding pressure switching pressure, ω3 be the weight of the maximum pressure, and ω4 be the weight of the switching position. n The weights of other influencing factors.

[0101] It is important to note that n is a positive integer greater than or equal to 1. When n is 1, the linear regression model is considered to have only one influencing factor; in this case, x1 is not limited to the plasticizing process, and ω1 is not limited to the weight of the plasticizing process. When n is 2, the linear regression model is considered to have two influencing factors; in this case, x1 is not limited to the plasticizing process, x2 is not limited to the holding pressure switching pressure, ω1 is not limited to the weight of the plasticizing process, ω2 is not limited to the weight of the holding pressure switching pressure, and so on. In this embodiment, based on the experimental analysis results, x1-x4 are the most closely related influencing factors. Therefore, x1-x4 are listed separately, with n≥5.

[0102] Then, the training data is input into the linear regression model to obtain the initial prediction model.

[0103] Specifically, training data is input into a linear regression model to obtain multiple sets of weights ω and bias constants b. One set of weights ω and bias constants b is used to create a linear prediction formula. Multiple sets of weights ω and bias constants b result in multiple linear prediction formulas. Based on the loss function, one linear prediction formula is selected from these formulas as the initial prediction model. Specifically, the loss function is minimized to determine the values ​​of weights ω and bias constants b, thereby selecting one from multiple linear prediction formulas as the initial prediction model.

[0104] The loss function is: ,

[0105] Where x represents the influencing factor; ω represents the weight of the influencing factor; b is the bias constant in the linear regression model; hi H represents the predicted value of the pad position calculated using a linear prediction formula. i It is the detected value (actual value) of the pad position corresponding to the historical operating parameter data used in the linear prediction formula; m is the number of comparison groups between the predicted value and the detected value of the pad position.

[0106] Finally, the training data is input into the initial prediction model, and the influencing factors and their weights in the initial prediction model are filtered according to the goodness of fit to obtain the prediction model.

[0107] Specifically, the influencing factors and their weights in the initial prediction model are screened based on the goodness of fit, resulting in the execution of the prediction model, which includes: Calculate the goodness of fit of the initial prediction model using the following formula:

[0108] in, For the initial prediction model The goodness of fit; The true value of the pad position corresponding to the training data used in the initial prediction model; The predicted value of the pad position is obtained by calculating the initial prediction model; This represents the mean of the true values ​​of the pad positions corresponding to the training data used in the initial prediction model. It should be noted that... It is a value less than 1, and the closer the value is to 1, the better the model's predictive performance.

[0109] Determine if the goodness of fit is greater than or equal to a preset value, where the preset value ranges from 0.92 to 1; if yes, determine the initial prediction model as the prediction model; if not, proceed with the following steps: Obtain the weight values ​​of each influencing factor in the initial prediction model whose goodness of fit is less than the preset value, delete the influencing factors corresponding to weights less than the set weight threshold and / or delete one or more corresponding influencing factors in ascending order of weight, generate a new round of linear regression model, and input the training data into the new round of linear regression model for training until the goodness of fit of the established new initial prediction model is greater than or equal to the preset value.

[0110] Furthermore, training methods also include using training data to present the accuracy of the prediction model. Specifically: First, the historical operating parameter data from the training data is input into the prediction model to calculate the predicted material pad position data.

[0111] Then, the historical material pad position data in the training data is compared with the predicted material pad position data, a comparison curve is plotted, and the comparison curve and the bias constant in the prediction model are displayed on the human-machine interface.

[0112] Finally, the accuracy of the prediction model is presented intuitively using the comparison curves.

[0113] In this specific embodiment, the historical operating data refers to the data from the injection molding machine's operation over the past two years.

[0114] After manually analyzing the scatter plots of each dependent variable (historical candidate operating parameters) and the target result (material pad position), the following conclusions can be drawn for each dependent variable and result: like Figure 3 As shown in the scatter plot of injection time versus pad position, it can be seen that when the pad position remains constant at y = 20, the pad position does not increase or decrease with the increase of injection time (x-axis); however, when the injection time (x = 5s) remains constant, there is a change in pad position. Therefore, it can be concluded that there is no linear relationship between injection time and pad position.

[0115] like Figure 4 As shown in the scatter plot of plasticizing time versus pad position, it can be seen that when the pad position remains constant at y = 20, the pad position does not increase or decrease with the increase of plasticizing time (x-axis); however, when the plasticizing time (x = 10s ~ 15s) remains constant, there are also instances where the pad position changes. Therefore, it can be concluded that there is no linear relationship between plasticizing time and pad position.

[0116] like Figure 5 As shown in the scatter plot of plasticizing stroke versus pad position, when the plasticizing stroke is within the range of 140 mm to 150 mm, the change in pad position is represented by two clustered blocks. Therefore, it can be concluded that there is no linear relationship between plasticizing stroke and pad position.

[0117] like Figure 6 As shown in the scatter plot of the holding pressure switching pressure and the material pad position, it can be seen that when the holding pressure switching pressure varies from 30 bar to 100 bar, the material pad position does not change significantly most of the time, but a cluster appears in the middle of the plot. Therefore, it can be concluded that there is no linear relationship between the holding pressure switching pressure and the material pad position.

[0118] like Figure 7As shown in the scatter plot of maximum pressure and pad position, it can be seen that the judgment logic of maximum pressure is similar to that of pressure holding and switching pressure. When the maximum pressure varies in the range of 10 bar to 100 bar, there is no obvious linear relationship between pad position and maximum pressure.

[0119] like Figure 8 As shown in the scatter plot of maximum back pressure and pad position, it can be seen that when the maximum back pressure varies in the range of 200 bar to 1000 bar, there is no obvious linear relationship between the pad position and the maximum pressure.

[0120] like Figure 9 As shown in the scatter plot of maximum injection speed and pad position, it can be seen that when the injection speed varies in the range of 50 mm / s to 60 mm / s, the pad position is mainly divided into two clusters; it can be seen that there is no linear relationship between the two.

[0121] like Figure 10 As shown in the scatter plot of the switching position and the pad position, it can be seen that when the set switching position changes within the range of 10mm to 50mm, the pad position increases as the switching position increases; it can be clearly seen that there is a clear linear relationship between the set switching position and the pad position.

[0122] Therefore, historical data on injection time, plasticizing time, plasticizing stroke, holding pressure switching pressure, maximum pressure, and maximum back pressure are deleted, while historical data on switching positions are retained.

[0123] The training data was input into the linear regression algorithm, and the linear relationship was calculated as follows:

[0124] Input the historical data of the switching positions from the training data into the above formula to calculate the corresponding predicted pad position. Compare the historical pad positions in the training data with the predicted pad positions and plot a comparison curve. Figure 11 The comparison curve is then displayed on the human-computer interface. Through... Figure 11 It can be clearly seen that the predicted values ​​and the detected values ​​almost overlap in most cases.

[0125] Based on the comparison curve between the actual and predicted values ​​displayed on the human-machine interface of the control system, the user evaluates the accuracy of the prediction model as high. Therefore, the user selects to use this model for monitoring the future position of the material pad. The deviation between the predicted and actual material pad positions calculated by the model is used to determine whether the check ring on the injection molding machine has failed. If there is no deviation between the predicted and actual material pad positions, it indicates that the system is stable or the switching position has changed. If there is a significant deviation between the predicted and actual material pad positions, the check ring can be determined to have failed. In this case, the injection molding machine system needs to issue an alarm and prompt the customer to stop the machine for maintenance.

[0126] The above description is merely a specific embodiment of the present invention. Under the teachings of the present invention, those skilled in the art can make other improvements or modifications based on the above embodiments. Those skilled in the art should understand that the above specific description is only to better explain the purpose of the present invention, and the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A training method for a personalized prediction model, characterized in that, The training method includes: Acquire equipment operating data and, based on human experience, select candidate factors that may affect the target outcome. Based on the results of manual selection, candidate factor data corresponding to the candidate factors and historical result data corresponding to the target result are retrieved from the historical operating data of the equipment. The candidate factor data and the historical result data are presented in a corresponding manner, and the desired influencing factors that have a linear relationship with the target result are manually selected from the candidate factors; The candidate factor data corresponding to the influencing factors are used as historical factor data, and the historical factor data and the corresponding historical result data are used as training data. Based on the training data and the preset training algorithm, a predictive model that can reflect the relationship between the influencing factors and the target result is constructed and trained.

2. The training method for the personalized prediction model according to claim 1, characterized in that, The process of mapping and presenting the candidate factor data and the historical result data includes: Scatter plots of the candidate factor data and the corresponding historical result data are drawn on the human-computer interaction interface.

3. The training method for the personalized prediction model according to claim 1, characterized in that, The execution of constructing and training a predictive model that reflects the relationship between the influencing factors and the target result based on the training data and a preset training algorithm includes: Construct a linear regression model for the target outcome and the influencing factors: In the formula, Represents the target result; x1 to x n Representing different influencing factors; x n Let ω1 be the nth influencing factor; b is the bias constant; ω1 to ω n The weights corresponding to different influencing factors; The training data is input into the linear regression model to obtain the initial prediction model; The training data is input into the initial prediction model, and the influencing factors and their weights in the initial prediction model are filtered according to the goodness of fit to obtain the prediction model.

4. The training method for the personalized prediction model according to claim 3, characterized in that, The step of inputting the training data into the linear regression model to obtain the initial prediction model includes: The training data is grouped and input into the linear regression model to obtain multiple sets of weights and bias constants. One set of weights and bias constants is used to create a linear prediction formula, and multiple linear prediction formulas are constructed based on multiple sets of weights and bias constants. A loss function is used to minimize the loss function to determine the values ​​of each weight and the bias constant. From multiple linear prediction formulas, one linear prediction formula is selected as the initial prediction model. The loss function is: ， Where x represents the influencing factor; ω represents the weight of the influencing factor; b is the bias constant in the linear regression model; h i H represents the target result calculated using a linear prediction formula. i It is the historical result data corresponding to the historical factor data used in the linear prediction formula; m is the number of comparison groups between the target result and the historical result data.

5. The training method for the personalized prediction model according to claim 4, characterized in that, The step of filtering the influencing factors and their weights in the initial prediction model based on the goodness of fit to obtain the execution of the prediction model includes: Calculate the goodness of fit of the initial prediction model using the following formula: in, For the initial prediction model The goodness of fit; The true value of the target result corresponding to the training data used in the initial prediction model; The predicted value of the target result calculated by the initial prediction model; The mean of the true values ​​of the target results corresponding to the training data used in the initial prediction model; Determine whether the goodness of fit is greater than or equal to a preset value; if yes, determine the initial prediction model as the prediction model; if not, proceed with the following steps: Delete the influencing factors corresponding to weights less than the set weight threshold and / or delete one or more corresponding influencing factors in ascending order of weight to generate a new round of linear regression model; The training data is input into a new round of linear regression model for training until the goodness of fit is greater than or equal to the preset value.

6. The training method for the personalized prediction model according to claim 1, characterized in that, The prediction model training method further includes: The accuracy of the prediction model is presented using the training data.

7. The training method for the personalized prediction model according to claim 6, characterized in that, The execution of presenting the accuracy of the prediction model using the training data includes: The historical factor data is input into the prediction model to calculate the predicted value of the prediction model; The historical results data are compared with the calculated predicted values, a comparison curve is plotted, and the comparison curve is displayed on the human-computer interface. The accuracy of the prediction model is visually presented using the comparison curves.

8. A training system for a prediction model, characterized in that, The training system is capable of executing the training method for the personalized prediction model according to any one of claims 1 to 7, and the training system comprises: The acquisition and selection unit is used to acquire the equipment's operating data and, based on human experience, select candidate factors that may affect the target result. The retrieval unit is used to retrieve candidate factor data corresponding to the candidate factors and historical result data corresponding to the target result from the historical operating data of the equipment based on the manual selection results. A filtering unit is used to present the candidate factor data and the historical result data in a correspondence, and to manually filter out the desired influencing factors that have a linear relationship with the target result from the candidate factors. The training data construction unit is used to take the candidate factor data corresponding to the influencing factor as historical factor data, and the historical factor data and the corresponding historical result data as training data. A construction and training unit is used to construct and train a predictive model that reflects the relationship between the influencing factors and the target result based on the training data and a preset training algorithm.

9. A predictive model training device, characterized in that, The prediction model training device includes: Processor; and Memory for storing the executable instructions of the processor; The processor executes the executable instructions to enable the prediction model training device to implement the training method of the personalized prediction model as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instruction is executed by the processor, it implements the training method of the personalized prediction model according to any one of claims 1 to 7.

Images