Parameter prediction device and method
Through Bayesian optimization model training architecture, screening and predicting key parameter projects in complex processes, the problems of difficulty in parameter design and low prediction accuracy in the existing technology are solved, and more efficient parameter prediction and production assistance are achieved.
Patent Information
- Application Number
- CN202311515373.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-10-31
- Filing Date
- 2023-11-14
- Publication Date
- 2025-05-02
AI Technical Summary
In complex processes, parameter design is not easy, R&D progress is slow, and the existing technology lacks an effective model training architecture, resulting in low accuracy of parameter prediction and requires multiple corrections and simulations.
The model training architecture of Bayesian optimization preprocessing model and prediction model is adopted, and the key input parameter items are filtered through preliminary screening model and the first Bayesian optimization model, and then parameter prediction combinations and adjustment suggestions are generated through the parameter optimization model and the second Bayesian optimization model.
Directly generate parameter prediction combinations and adjustment suggestions before actual manufacturing, shorten parameter correction and mold test time, improve parameter prediction accuracy, reduce manual intervention, and improve production efficiency.
Smart Images
Figure CN119918381A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a parameter prediction device and method. Specifically, the present invention relates to a parameter prediction device and method that can improve the accuracy of parameter prediction. Background Art
[0002] In traditional industries, for highly complex processes (e.g. casting technology, sand casting, precision lost wax, metal mold casting, etc.), due to the large number and complexity of components (e.g. bicycle components), it is difficult to design process parameters, the R&D progress is slow, and experience is not easy to pass on. In addition, it is still necessary to rely on experienced professionals to design and adjust parameters.
[0003] Even though computer-aided engineering analysis has been introduced in recent years to analyze and simulate parameters, there is a lack of efficient model training architecture. Relying solely on simple linear models or basic machine learning models usually still requires multiple corrections and time-consuming multiple simulations, and is unable to efficiently and accurately predict parameter combinations, often leading to prediction errors.
[0004] In view of this, how to provide a parameter prediction technology that can improve the accuracy of parameter prediction is a goal that the industry urgently needs to work hard on. Summary of the invention
[0005] One object of the present invention is to provide a parameter prediction device. The parameter prediction device includes a transceiver interface, a storage and a processor, and the processor is electrically connected to the transceiver interface and the storage. The processor receives a plurality of input data corresponding to a prediction target, wherein each of the plurality of input data includes a plurality of input parameter items, and each of the plurality of input parameter items corresponds to a parameter value. The processor inputs the plurality of input data into a pre-processing model to filter out a plurality of first input parameter items from the plurality of input parameter items, wherein the pre-processing model is based on a preliminary screening model and a first Bayesian optimization model. The processor inputs the plurality of input data corresponding to the plurality of first input parameter items into a prediction model to generate a plurality of parameter prediction combinations, wherein the prediction model is based on a parameter optimization model and a second Bayesian optimization model.
[0006] Another object of the present invention is to provide a parameter prediction method for an electronic device, wherein the parameter prediction method is executed by the electronic device and includes the following steps: receiving multiple input data corresponding to a prediction target, wherein each of the multiple input data includes multiple input parameter items, and each of the multiple input parameter items corresponds to a parameter value; inputting the multiple input data into a pre-processing model to filter out multiple first input parameter items from the multiple input parameter items, wherein the pre-processing model is based on a preliminary screening model and a first Bayesian optimization model; and inputting the multiple input data corresponding to the multiple first input parameter items into a prediction model to generate multiple parameter prediction combinations, wherein the prediction model is based on a parameter optimization model and a second Bayesian optimization model.
[0007] In one embodiment of the present invention, each of the plurality of parameter prediction combinations includes the plurality of first input parameter items and each of the plurality of first input parameter items corresponds to a parameter prediction value.
[0008] In one embodiment of the present invention, the pre-processing model further includes performing the following operations: inputting the multiple input data into the preliminary screening model to generate corresponding multiple first model hyperparameter information; and inputting the multiple input data and the multiple first model hyperparameter information into the first Bayesian optimization model to perform a first Bayesian optimization operation to determine the multiple first input parameter items among the multiple input parameter items.
[0009] In one embodiment of the present invention, the plurality of first model hyperparameter information includes at least one of the number of decision trees, a loss function, and a neural network depth, or a combination thereof.
[0010] In one embodiment of the present invention, the first Bayesian optimization model is generated based on the following operations: defining an objective function and multiple parameter combinations based on the multiple input data; based on the objective function, inputting the multiple parameter combinations into an agent model to generate an objective function evaluation result; based on the objective function evaluation result and the multiple parameter combinations, updating the agent model; and setting the trained agent model as the first Bayesian optimization model.
[0011] In one embodiment of the present invention, the prediction model further includes performing the following operations: inputting a plurality of historical time series data into the parameter optimization model to generate a corresponding plurality of second model hyperparameter information, wherein the plurality of historical time series data include the input data corresponding to the plurality of first input parameter items; and inputting the input data and the plurality of second model hyperparameter information into the second Bayesian optimization model, performing a second Bayesian optimization operation to generate the plurality of parameter prediction combinations.
[0012] In one embodiment of the present invention, the plurality of second model hyperparameter information includes at least one of the number of decision trees, a loss function, and a neural network depth, or a combination thereof.
[0013] In one embodiment of the present invention, the processor further performs the following operations: receiving an input data to be predicted; and based on the prediction model, generating a result evaluation and a parameter adjustment suggestion corresponding to the input data to be predicted, wherein the parameter adjustment suggestion is generated based on the plurality of parameter prediction combinations.
[0014] In one embodiment of the present invention, the processor further performs the following operation: inputting the input data to be predicted and a feedback data corresponding to the input data to be predicted into the prediction model to update the prediction model, wherein the feedback data is generated based on the result evaluation and the parameter adjustment suggestion.
[0015] The parameter prediction technology (at least including the device and method) provided by the present invention can directly generate parameter prediction combinations and parameter adjustment suggestions before actual manufacturing through the Bayesian optimization pre-processing model and the model training framework of the prediction model, and assist the manufacturer in setting and adjusting parameters, thereby improving production quality. Since the parameter prediction technology provided by the present invention can generate parameter prediction results without actual manufacturing, it can shorten the time for repeatedly correcting casting parameters and trial molds, improve the accuracy of parameter prediction, and solve the problem that the conventional technology requires manual intervention to judge the results.
[0016] The detailed technology and implementation methods of the present invention are described below in conjunction with the accompanying drawings so that a person having ordinary knowledge in the technical field to which the present invention belongs can understand the technical features of the invention for which protection is sought. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 is a schematic diagram depicting the architecture of a parameter prediction device according to a first embodiment;
[0018] Figure 2 is a schematic diagram depicting the operation of certain embodiments;
[0019] Figure 3is a schematic diagram depicting a data pre-processing flow chart of certain embodiments;
[0020] Figure 4 is a schematic diagram of a Bayesian optimization process depicting certain embodiments;
[0021] Figure 5 is a schematic diagram of a training process depicting certain implementations;
[0022] Figure 6 is a schematic diagram depicting output results of certain embodiments; and
[0023] Figure 7 is a partial flow chart depicting the parameter prediction method of the second embodiment.
[0024]
Explanation of symbols
[0025] 1: Parameter prediction device
[0026] 11: Transceiver interface
[0027] 13: Storage
[0028] 15: Processor
[0029] 200: Operational flow chart
[0030] P1, P2: Path
[0031] S21, S22, S23, S24, S25, S26, S27: Operation
[0032] 300: Schematic diagram of data pre-processing process
[0033] S32, S33, S34, S35, S36, S37, S38, S39: Operation
[0034] 400: Bayesian Optimization Process Diagram
[0035] S40, S41, S42, S43, S44, S45, S46, S47, S48, S49: Operation
[0036] 500: Diagram of training process
[0037] PPM: Pre-processing model
[0038] PM: Predictive Model
[0039] S51, S52, S53, S54, S55, S56, S57, S58: Operation
[0040] 600: Output result diagram
[0041] S60: Operation
[0042] 700: Parameter Prediction Methods
[0043] S701, S703, S705: Steps DETAILED DESCRIPTION
[0044] A parameter prediction device and method provided by the present invention will be explained below through implementation examples. However, the multiple implementation examples are not intended to limit the present invention to any environment, application or method as described in the multiple implementation examples. Therefore, the description of the implementation examples is only for the purpose of explaining the present invention, and is not intended to limit the scope of the present invention. It should be understood that in the following implementation examples and drawings, elements that are not directly related to the present invention have been omitted and are not illustrated, and the size of each element and the size ratio between the elements are only for illustration, and are not intended to limit the scope of the present invention.
[0045] The first embodiment of the present invention is a parameter prediction device 1, the structure diagram of which is depicted in FIG. Figure 1 The parameter prediction device 1 comprises a transceiver interface 11 , a storage 13 and a processor 15 . The processor 15 is electrically connected to the transceiver interface 11 and the storage 13 .
[0046] It should be noted that the transceiver interface 11 is an interface capable of receiving and transmitting data or other interfaces capable of receiving and transmitting data known to a person skilled in the art in the art to which the present invention belongs. The storage 13 may be a memory, a USB disk, a hard disk, an optical disk, a flash drive, or any other storage medium or circuit with the same function known to a person skilled in the art in the art to which the present invention belongs. The processor 15 may be various processing units, a central processing unit (CPU), a microprocessor, or other computing devices known to a person skilled in the art in the art to which the present invention belongs.
[0047] To understand the operation process of this disclosure, please refer to Figure 2 200. Figure 2 As shown, the formulation of process parameters generally requires first performing model design S21 and computer-aided engineering simulation S22 to generate data output S23.
[0048] Next, in the existing architecture, since the actual production of products is required and the judgment of the quality of the products by experienced professionals is still relied upon, the production line production S25 in path P2 and the manual judgment of the quality S26 will be executed in the existing technology, and finally the optimization parameters S27 can be re-formulated / adjusted based on the results of the manual judgment.
[0049] However, in the parameter prediction technology proposed in the present disclosure, since the parameter prediction device 1 of the present disclosure can pre-train and Bayesian optimize the model by designing a model training architecture, the parameter prediction S24 in the path P1 can be directly executed, and the optimization parameters S27 can be directly re-formulated / adjusted based on the prediction results. The details of the execution of the parameter prediction device 1 of the present disclosure will be described in detail below.
[0050] First, in this embodiment, the processor 15 receives a plurality of input data corresponding to a prediction target, wherein each of the plurality of input data includes a plurality of input parameter items, and each of the plurality of input parameter items corresponds to a parameter value.
[0051] It should be noted that the parameter prediction device 1 may first pre-process the multiple input data corresponding to the prediction target to conform to the format received by the model during machine learning.
[0052] For easier understanding, please refer to Figure 3 Schematic diagram 300 of the data pre-processing process. Figure 3 As shown, the parameter prediction device 1 may firstly execute input parameter data S32, wherein the parameter data may be a plurality of combinations, for example, a plurality of casting schemes including casting temperature, injection pressure, casting gate position, and shell mold processing thickness. Then, the data is checked S33 to check whether there are missing values, abnormal values, noise, etc.
[0053] Next, data cleaning S34 is performed to fill missing values, delete outliers, and process noise by smoothing or filtering. Subsequently, in order to enable the model to understand the scale and meaning of the original data, feature selection and extraction S35 is performed, such as: numerical feature scaling, discrete feature encoding, text feature vectorization, etc.
[0054] Next, data conversion S36 is performed, for example, data conversion such as standardization, normalization, encoding or vectorization is performed on the data. Subsequently, data segmentation S37 is performed, for example, the data is segmented into a training set, a validation set, a test set, and the like.
[0055] Next, in order to process the data into the input format required by the model, data format processing S38 is performed, such as: tensor, sequence, other data formats CSV, json, etc. Finally, the data can be input into the subsequent model for machine learning S39.
[0056] Next, in this embodiment, the processor 15 determines relatively critical input parameter items from the plurality of input data. Specifically, the processor 15 inputs the plurality of input data into a pre-processing model to filter out a plurality of first input parameter items from the plurality of input parameter items, wherein the pre-processing model is based on a preliminary screening model and a first Bayesian optimization model.
[0057] In some embodiments, the pre-processing model further includes performing the following operations: the processor 15 inputs the plurality of input data into the preliminary screening model to generate corresponding plurality of first model hyperparameter information. Then, the processor 15 inputs the plurality of input data and the plurality of first model hyperparameter information into the first Bayesian optimization model to perform a first Bayesian optimization operation to determine the plurality of first input parameter items among the plurality of input parameter items.
[0058] For easier understanding, please refer to Figure 5 A training flow diagram 500 is shown. Figure 5 As shown, after executing data format processing S38, the processor 15 inputs the multiple input data into the preliminary screening model S51 in the pre-processing model PPM, and inputs the generated hyperparameter information into the first Bayesian optimization model S52, and performs multiple optimization iterations S53 to determine which input parameter items among the multiple input parameter items are important.
[0059] For example, if there are originally 13 casting parameters, the most critical casting parameters (eg, 5 of them) and the best hyperparameter combination can be found through the above operation.
[0060] It should be noted that in the present disclosure, the preliminary screening model is a model that executes the parameter preliminary screening algorithm (i.e., the first model referred to in the present disclosure). For example, the first model can be implemented under the algorithm framework of XGBoost. In addition, the first model hyperparameter information is used to control the parameters of the first model in the learning process. For example, the plurality of first model hyperparameter information may include at least one of the number of decision trees, a loss function, and a neural network depth, or a combination thereof.
[0061] It should be noted that in the present disclosure, Bayesian optimization can quickly select a better hyperparameter combination from multiple parameters by optimizing iteratively.
[0062] Specifically, the first Bayesian optimization model is generated based on the following operations: the processor 15 defines an objective function and a plurality of parameter combinations based on the plurality of input data. Then, based on the objective function, the processor 15 inputs the plurality of parameter combinations into an agent model to generate an objective function evaluation result. Then, the processor 15 updates the agent model based on the objective function evaluation result and the plurality of parameter combinations. Finally, the processor 15 sets the trained agent model as the first Bayesian optimization model.
[0063] For easier understanding, please refer to Figure 4 Schematic diagram 400 of the Bayesian optimization process. Figure 4 As shown, after inputting parameters S40 (e.g., casting temperature, injection pressure, filling speed), define the objective function S41 (e.g., model prediction accuracy, model effectiveness) and define the parameter space S42 (e.g., casting parameter combination) according to the input data. Then, establish the initial sample set S43 and establish the proxy model and associate the parameters with the objective function S44.
[0064] Subsequently, a Bayesian optimization iteration S45 (e.g., optimization, search) is performed to evaluate the objective function S46 and update the proxy model S47 (e.g., new sample points, the value of the objective function). If the result of determining whether the termination condition S48 is reached is no, the aforementioned operation is continued. If the result is yes, the best solution is selected S49.
[0065] Next, in this embodiment, the processor 15 performs parameter prediction on the input data generated by the pre-processing model PPM. Specifically, in this embodiment, the processor 15 inputs the plurality of input data corresponding to the plurality of first input parameter items into a prediction model to generate a plurality of parameter prediction combinations, wherein the prediction model is based on a parameter optimization model and a second Bayesian optimization model.
[0066] In some embodiments, the prediction model further includes performing the following operations: the processor 15 inputs a plurality of historical time series data into the parameter optimization model to generate a plurality of corresponding second model hyperparameter information, wherein the plurality of historical time series data includes the input data corresponding to the plurality of first input parameter items. Then, the processor 15 inputs the input data and the plurality of second model hyperparameter information into the second Bayesian optimization model to perform a second Bayesian optimization operation to generate the plurality of parameter prediction combinations.
[0067] In some embodiments, each of the plurality of parameter prediction combinations includes the plurality of first input parameter items and each of the plurality of first input parameter items corresponds to a parameter prediction value.
[0068] For easier understanding, please continue to refer to Figure 5 A training flow diagram 500 is shown. Figure 5 As shown, after the pre-processing model PPM generates the multiple input data corresponding to the input parameter items, the processor 15 inputs the multiple input data into the parameter optimization model S54 in the prediction model PM, and inputs the generated hyperparameter information into the second Bayesian optimization model S55, and after executing multiple optimization iterations S56, generates a parameter prediction combination S57.
[0069] It should be noted that in the present disclosure, the parameter optimization model is a model that executes the data optimization algorithm (i.e., the second model referred to in the present disclosure). For example, the second model can be implemented under the algorithm framework of Long Short-Term Memory (LSTM). In addition, the second model hyperparameter information is used to control the parameters of the second model in the learning process. For example, the plurality of first model hyperparameter information may include at least one of the number of decision trees, a loss function, and a neural network depth, or a combination thereof.
[0070] In some embodiments, after the prediction model training is completed, the processor 15 may further perform a result evaluation on the new input data. Specifically, the processor 15 receives an input data to be predicted. Then, based on the prediction model, the processor 15 generates a result evaluation and a parameter adjustment suggestion corresponding to the input data to be predicted, wherein the parameter adjustment suggestion is generated based on the plurality of parameter prediction combinations.
[0071] For easier understanding, please refer to Figure 6 Schematic diagram 600 of the output result. Figure 6 As shown, the output result OR may also include information such as parameter correctness, abnormality reasons and parameter modification suggestions.
[0072] For example, when the prediction model determines that the input data corresponds to an abnormal cause of overflow / insufficient, the prediction model further provides its recommended parameter combination A (i.e., parameter adjustment suggestion). In addition, when the prediction model determines that the input data corresponds to an abnormal cause of size mismatch, the prediction model further provides its recommended parameter combination B. In addition, when the prediction model determines that the input data corresponds to an abnormal cause of mold damage, the prediction model further provides its recommended parameter combination C. Accordingly, since the prediction model of the present disclosure directly provides parameter modification suggestions, the parameter settings can be directly adjusted based on these modification suggestions to perform production operation S60.
[0073] In some embodiments, the processor 15 may further update the prediction model PM based on the feedback data (e.g. Figure 5Specifically, the processor 15 inputs the input data to be predicted and a feedback data corresponding to the input data to be predicted into the prediction model to update the prediction model, wherein the feedback data is generated based on the result evaluation and the parameter adjustment suggestion.
[0074] From the above description, it can be seen that the parameter prediction device 1 provided by the present invention can directly generate parameter prediction combinations and parameter adjustment suggestions before actual manufacturing through the Bayesian optimization pre-processing model and the model training architecture of the prediction model, and assist the manufacturer in setting and adjusting parameters, thereby improving production quality. Since the parameter prediction device 1 provided by the present invention can generate parameter prediction results without actual manufacturing, it can shorten the time for repeatedly correcting casting parameters and trial molds, improve the accuracy of parameter prediction, and solve the problem that the conventional technology requires manual intervention to judge the results.
[0075] The second embodiment of the present invention is a parameter prediction method, the flow chart of which is depicted in Figure 7 The parameter prediction method 700 is applicable to an electronic device, such as the parameter prediction device 1 described in the first embodiment. The parameter prediction method 700 generates a plurality of parameter prediction combinations through steps S701 to S705.
[0076] In step S701, a plurality of input data corresponding to a prediction target is received by an electronic device, wherein each of the plurality of input data includes a plurality of input parameter items, and each of the plurality of input parameter items corresponds to a parameter value. Then, in step S703, the electronic device inputs the plurality of input data into a pre-processing model to filter out a plurality of first input parameter items from the plurality of input parameter items, wherein the pre-processing model is based on a preliminary screening model and a first Bayesian optimization model.
[0077] Finally, in step S705, the electronic device inputs the plurality of input data corresponding to the plurality of first input parameter items into a prediction model to generate a plurality of parameter prediction combinations, wherein the prediction model is based on a parameter optimization model and a second Bayesian optimization model.
[0078] In addition to the above steps, the second embodiment can also perform all operations and steps of the parameter prediction device 1 described in the first embodiment, has the same functions, and achieves the same technical effects. A person with ordinary knowledge in the technical field to which the present invention belongs can directly understand how the second embodiment performs these operations and steps based on the above-mentioned first embodiment, has the same functions, and achieves the same technical effects, so it will not be repeated.
[0079] It should be noted that in the patent specification and the scope of the patent application of the present invention, certain terms (including: input parameter items, Bayesian optimization models, model hyperparameter information, Bayesian optimization operations, etc.) are preceded by "first" or "second", and the multiple "first" or "second" are only used to distinguish different terms. For example: the "first" and "second" in the first Bayesian optimization operation and the second Bayesian optimization operation are only used to indicate the Bayesian optimization operations used in different operations.
[0080] In summary, the parameter prediction technology (at least including the device and method) provided by the present invention can directly generate parameter prediction combinations and parameter adjustment suggestions before actual manufacturing through the Bayesian optimization pre-processing model and the model training framework of the prediction model, and assist the manufacturer in setting and adjusting the parameters, thereby improving the production quality. Since the parameter prediction technology provided by the present invention can generate parameter prediction results without actual manufacturing, it can shorten the time for repeatedly correcting casting parameters and trial molds, improve the accuracy of parameter prediction, and solve the problem that the conventional technology requires manual intervention to judge the results.
[0081] The above embodiments are only used to illustrate some embodiments of the present invention and to explain the technical features of the present invention, and are not used to limit the protection category and scope of the present invention. Any changes or equivalent arrangements that can be easily completed by a person with ordinary knowledge in the technical field to which the present invention belongs are within the scope claimed by the present invention, and the scope of protection of the present invention is subject to the claims.
Claims
1. A parameter prediction device, characterized in that: Include: A transceiver interface; a storage device; as well as A processor is electrically connected to the transceiver interface and the storage, and is used to perform the following steps: Receiving a plurality of input data corresponding to a prediction target, wherein each of the plurality of input data comprises a plurality of input parameter items, and each of the plurality of input parameter items corresponds to a parameter value; Inputting the plurality of input data into a pre-processing model to select a plurality of first input parameter items from the plurality of input parameter items, wherein the pre-processing model is based on a preliminary screening model and a first Bayesian optimization model; as well as The plurality of input data corresponding to the plurality of first input parameter items are input into a prediction model to generate a plurality of parameter prediction combinations, wherein the prediction model is based on a parameter optimization model and a second Bayesian optimization model.
2. The parameter prediction device according to claim 1, characterized in that: Each of the plurality of parameter prediction combinations includes the plurality of first input parameter items and each of the plurality of first input parameter items corresponds to a parameter prediction value.
3. The parameter prediction device according to claim 1, characterized in that: The pre-processing model further includes performing the following operations: Inputting the plurality of input data into the preliminary screening model to generate corresponding plurality of first model hyperparameter information; as well as The plurality of input data and the plurality of first model hyperparameter information are input into the first Bayesian optimization model, and a first Bayesian optimization operation is performed to determine the plurality of first input parameter items among the plurality of input parameter items.
4. The parameter prediction device according to claim 3, characterized in that: The plurality of first model hyperparameter information includes at least one of the number of decision trees, a loss function, and a neural network depth, or a combination thereof.
5. The parameter prediction device according to claim 3, characterized in that: The first Bayesian optimization model is generated based on the following operation: Defining an objective function and a plurality of parameter combinations based on the plurality of input data; Based on the objective function, inputting the plurality of parameter combinations into an agent model to generate an objective function evaluation result; Based on the objective function evaluation result and the plurality of parameter combinations, updating the proxy model; and The trained proxy model is set as the first Bayesian optimization model.
6. The parameter prediction device according to claim 1, characterized in that: The prediction model further comprises performing the following operations: Inputting a plurality of historical time series data into the parameter optimization model to generate a plurality of corresponding second model hyperparameter information, wherein the plurality of historical time series data includes the input data corresponding to the plurality of first input parameter items; as well as The input data and the plurality of second model hyperparameter information are input into the second Bayesian optimization model, and a second Bayesian optimization operation is performed to generate the plurality of parameter prediction combinations.
7. The parameter prediction device according to claim 6, characterized in that: The plurality of second model hyperparameter information includes at least one of the number of decision trees, a loss function, and a neural network depth, or a combination thereof.
8. The parameter prediction device according to claim 1, characterized in that: The processor further performs the following operations: receiving input data to be predicted; and Based on the prediction model, a result evaluation and a parameter adjustment suggestion corresponding to the input data to be predicted are generated, wherein the parameter adjustment suggestion is generated based on the plurality of parameter prediction combinations.
9. The parameter prediction device according to claim 8, characterized in that: The processor further performs the following operations: Inputting the input data to be predicted and a feedback data corresponding to the input data to be predicted into the prediction model to update the prediction model, The feedback data is generated based on the result evaluation and the parameter adjustment suggestion.
10. A parameter prediction method, characterized in that: For an electronic device, the parameter prediction method is executed by the electronic device and comprises the following steps: Receiving a plurality of input data corresponding to a prediction target, wherein each of the plurality of input data comprises a plurality of input parameter items, and each of the plurality of input parameter items corresponds to a parameter value; Inputting the plurality of input data into a pre-processing model to select a plurality of first input parameter items from the plurality of input parameter items, wherein the pre-processing model is based on a preliminary screening model and a first Bayesian optimization model; as well as The plurality of input data corresponding to the plurality of first input parameter items are input into a prediction model to generate a plurality of parameter prediction combinations, wherein the prediction model is based on a parameter optimization model and a second Bayesian optimization model.