Production yield prediction method, apparatus, storage medium, and program product for a product

Abstract

The application provides a production yield prediction method, device, storage medium and program product, relates to the technical field of data processing, and comprises the following steps: on the basis of obtaining reference product information of each reference product and target product information of a target product, performing clustering processing on each reference product to obtain at least two reference product clustering sets; for each reference product clustering set: performing multi-head self-attention fusion processing to generate a corresponding product relationship degree matrix. According to the target product information of the target product, a corresponding target reference product clustering set is matched, and a target reference product is determined. Under each production time window: an initial yield prediction model is trained to obtain a target yield prediction model; the product relationship degree matrix corresponding to the target reference product clustering set, the target product information of the target product and a future time feature vector are input into a fully connected neural network layer, and a target prediction yield is output; and the problem that yield prediction of a newly developed product is inaccurate is solved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, device, storage medium, and program product for predicting the output of manufactured products. Background Technology

[0002] Currently, based on cost considerations, enterprises are adopting refined management of the production and storage stages of their products. The production stage is in a state of capacity balance, that is, there is no overcapacity or undercapacity. The storage stage is in a state of storage balance, that is, there is no inventory backlog or inventory shortage.

[0003] In existing technologies, for newly developed products, companies often use data augmentation to determine the predicted output of the new products, and then determine the production plan for the new products based on the predicted output, so as to achieve refined management of the new products in the production and storage stages.

[0004] However, due to insufficient historical data on newly developed products, existing technologies lead to inaccurate production forecasts for these products. Summary of the Invention

[0005] This application provides a method, apparatus, storage medium, and program product for predicting the output of manufactured products, in order to solve the problem of inaccurate output prediction for newly developed products caused by existing solutions.

[0006] In a first aspect, embodiments of this application provide a method for predicting the output of manufactured products, comprising: acquiring reference product information for each reference product and target product information for a target product; clustering the reference products according to the reference product information to obtain at least two reference product cluster sets; for each reference product cluster set: generating a production output feature vector for each reference product under each production time window based on the reference product information of each reference product in the reference product cluster set; performing multi-head self-attention fusion processing on the production output feature vectors of each reference product under each production time window to generate a product relationship degree matrix for the corresponding reference product cluster set; and filtering the at least two reference product cluster sets according to the target product information of the target product to obtain a target reference product cluster set. The process involves: 1. Determining a target reference product from among the reference products in the target reference product cluster set, based on the target product information and the reference product information of each reference product in the target reference product cluster set; 2. Training an initial production prediction model within each production time window, based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product, to obtain a target production prediction model; 3. Inputting the product relationship degree matrix corresponding to the target reference product cluster set and the target product information of the target product into the target production prediction model, and outputting a predicted production feature vector; 4. Inputting the predicted production feature vector and the future time feature vector into a fully connected neural network layer, and outputting the target predicted production.

[0007] In one possible implementation, before training the initial production prediction model based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain the target production prediction model, the process of determining the initial production prediction model includes: inputting the product relationship degree matrix corresponding to each reference product cluster set into the graph convolutional neural network model of the original production prediction model, and outputting the corresponding product spatial structure feature vector; inputting the product spatial structure feature vector corresponding to each reference product cluster set and the reference product information of each reference product in each reference product cluster set into the long short-term memory network model of the original production prediction model, and outputting the corresponding predicted production output feature vector; inputting the production output feature vector of each reference product in each reference product cluster set and the predicted production output feature vector corresponding to each reference product cluster set into a preset loss function to obtain a loss value; and training the original production prediction model based on the loss value to obtain the initial production prediction model.

[0008] In one possible implementation, the step of performing multi-head self-attention fusion processing on the production output feature vectors of each reference product under each production time window to generate a product relationship matrix for the corresponding reference product cluster set includes: under each production time window: calculating the similarity of the production output feature vectors of each reference product to obtain a product similarity matrix between each reference product; normalizing the product similarity matrix between each reference product to obtain a normalized product similarity matrix between each reference product; and performing multi-head self-attention fusion processing on the normalized product similarity matrix between each reference product to generate a product relationship matrix for the corresponding reference product cluster set.

[0009] In one possible implementation, the step of performing multi-head self-attention fusion processing on the normalized product similarity matrix among the reference products to generate the product relationship degree matrix of the corresponding reference product cluster set includes: mapping the normalized product similarity matrix among the reference products based on a self-attention mechanism to obtain a query vector, a key vector, and a value vector; generating a one-head attention vector corresponding to each reference product based on the query vector, the key vector, and the value vector; fusing the one-head attention vectors corresponding to each reference product to obtain a multi-head attention matrix; and reshaping the dimensions of the multi-head attention matrix to generate the product relationship degree matrix of the corresponding reference product cluster set.

[0010] In one possible implementation, determining the target reference product from the reference products in the target reference product cluster set based on the target product information and the reference product information of each reference product in the target reference product cluster set includes: for each reference product in the target reference product cluster set: obtaining multiple corresponding distances based on the target product information and the reference product information of each reference product; performing a weighted calculation on the multiple distances to obtain a comprehensive distance; and determining the target reference product from the reference products in the target reference product cluster set based on the comprehensive distance between the target product and each reference product.

[0011] In one possible implementation, generating production output feature vectors for each reference product under each production time window based on the reference product information of each reference product in the reference product cluster set includes: obtaining the production time window based on the reference product information of each reference product in the reference product cluster set; and, under the production time sequence of each reference product, extracting features from the reference product information of the corresponding reference product based on the production time window to generate production output feature vectors for each reference product under each production time window.

[0012] Secondly, embodiments of this application provide a production volume prediction device for manufactured products, comprising:

[0013] The acquisition module is used to acquire reference product information for each reference product and target product information for the target product.

[0014] The processing module is configured to: cluster the reference products according to the reference product information to obtain at least two reference product cluster sets; for each reference product cluster set: generate production output feature vectors for each reference product under each production time window based on the reference product information of each reference product in the cluster set; perform multi-head self-attention fusion processing on the production output feature vectors of each reference product under each production time window to generate a product relationship degree matrix for the corresponding reference product cluster set; filter the at least two reference product cluster sets according to the target product information to obtain a target reference product cluster set; determine the target reference product from the reference products in the target reference product cluster set based on the target product information and the reference product information of each reference product in the target reference product cluster set; and, under each production time window: train an initial output prediction model based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain a target output prediction model.

[0015] The output module is used to input the product relationship degree matrix corresponding to the cluster set of the target reference product and the target product information of the target product into the target output prediction model, and output the predicted output feature vector; input the predicted output feature vector and the future time feature vector into the fully connected neural network layer, and output the target predicted output.

[0016] In one possible implementation, before training the initial production prediction model based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain the target production prediction model, the production prediction device specifically performs the following steps in determining the initial production prediction model: inputting the product relationship degree matrix corresponding to each reference product cluster set into the graph convolutional neural network model of the original production prediction model, and outputting the corresponding product spatial structure feature vector; inputting the product spatial structure feature vector corresponding to each reference product cluster set and the reference product information of each reference product in each reference product cluster set into the long short-term memory network model of the original production prediction model, and outputting the corresponding predicted production output feature vector; inputting the production output feature vector of each reference product in each reference product cluster set and the predicted production output feature vector corresponding to each reference product cluster set into a preset loss function to obtain a loss value; and training the original production prediction model based on the loss value to obtain the initial production prediction model.

[0017] In one possible implementation, when the processing module performs multi-head self-attention fusion processing on the production output feature vectors of each reference product under each production time window to generate the product relationship degree matrix of the corresponding reference product cluster set, it specifically performs the following: under each production time window: calculates the similarity of the production output feature vectors of each reference product to obtain the product similarity matrix between each reference product; normalizes the product similarity matrix between each reference product to obtain the normalized product similarity matrix between each reference product; and performs multi-head self-attention fusion processing on the normalized product similarity matrix between each reference product to generate the product relationship degree matrix of the corresponding reference product cluster set.

[0018] In one possible implementation, when the processing module performs multi-head self-attention fusion processing on the normalized product similarity matrix between each reference product to generate the product relationship degree matrix of the corresponding reference product cluster set, it specifically performs the following steps: For the normalized product similarity matrix between each reference product: mapping is performed based on the self-attention mechanism to obtain query vector, key vector, and value vector; based on the query vector, key vector, and value vector, a one-head attention vector corresponding to each reference product is generated; the one-head attention vectors corresponding to each reference product are fused to obtain a multi-head attention matrix; and the multi-head attention matrix is ​​dimensionally reshaped to generate the product relationship degree matrix of the corresponding reference product cluster set.

[0019] In one possible implementation, when the processing module determines the target reference product from the reference products in the target reference product cluster set based on the target product information and the reference product information of each reference product in the target reference product cluster set, it specifically performs the following: for each reference product in the target reference product cluster set: obtain multiple corresponding distances based on the target product information and the reference product information of each reference product; perform a weighted calculation on the multiple distances to obtain a comprehensive distance; and determine the target reference product from the reference products in the target reference product cluster set based on the comprehensive distance between the target product and each reference product.

[0020] In one possible implementation, when the processing module generates the production output feature vector of each reference product under each production time window based on the reference product information of each reference product in the reference product cluster set, it is specifically used to: obtain the production time window based on the reference product information of each reference product in the reference product cluster set; and extract features from the reference product information of the corresponding reference product under the production time sequence of each reference product, thereby generating the production output feature vector of each reference product under each production time window.

[0021] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0022] The memory stores computer-executed instructions;

[0023] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0024] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0025] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0026] The production output prediction method, equipment, storage medium, and program products provided in this application, based on obtaining reference product information for each reference product and target product information for the target product, perform clustering processing on each reference product to obtain at least two reference product cluster sets. Then, for each reference product cluster set: based on the reference product information of each reference product in the cluster set, a production output feature vector for each reference product under each production time window is generated, and multi-head self-attention fusion processing is performed to generate a product relationship degree matrix for the corresponding reference product cluster set. Further, based on the target product information of the target product, the corresponding target reference product cluster set is matched, and the target reference product is determined within the target reference product cluster set. Furthermore, within each production time window: based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product, the initial output prediction model is trained to obtain the target output prediction model; the product relationship degree matrix corresponding to the target reference product cluster set and the target product information of the target product are input into the target output prediction model, and the predicted output feature vector is output; the predicted output feature vector and the future time feature vector are input into the fully connected neural network layer, and the target predicted output is output; this solves the problem of inaccurate output prediction for newly developed products caused by the existing technology. Attached Figure Description

[0027] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0028] Figure 1 A flowchart illustrating a production output prediction method for a product manufactured according to an embodiment of this application;

[0029] Figure 2 A schematic diagram of the system architecture of a production volume prediction system for a product manufactured according to an embodiment of this application;

[0030] Figure 3 A schematic diagram of the structure of a production output prediction device for a product manufactured according to an embodiment of this application;

[0031] Figure 4 A schematic diagram of the structure of the electronic device provided in this application.

[0032] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0033] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0034] The technical solution of this application involves the collection, storage, use, processing, transmission, provision and disclosure of user personal information and data, which comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0035] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0036] Currently, driven by cost considerations, enterprises employ refined management in both the production and storage stages of their products. Production aims for a balanced capacity, preventing both overcapacity and undercapacity, while storage aims for a balanced inventory, avoiding both overstocking and understocking. In existing technologies, for newly developed products, enterprises often use data augmentation to determine the projected output, and then use this forecast to formulate production plans, thus achieving refined management of the new products in both production and storage. However, due to insufficient historical data on newly developed products, existing solutions suffer from inaccurate output forecasts.

[0037] To address the aforementioned technical problems, the embodiments of this application propose the following technical concepts:

[0038] Based on the acquisition of reference product information for each reference product and target product information for the target product, at least two reference product cluster sets are obtained by clustering each reference product. Then, for each reference product cluster set: based on the reference product information of each reference product in the cluster set, a production output feature vector for each reference product under each production time window is generated, and multi-head self-attention fusion processing is performed to generate the product relationship degree matrix of the corresponding reference product cluster set. Further, based on the target product information of the target product, the corresponding target reference product cluster set is matched, and the target reference product is determined within the target reference product cluster set. Furthermore, within each production time window: based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product, the initial output prediction model is trained to obtain the target output prediction model; the product relationship degree matrix corresponding to the target reference product cluster set and the target product information of the target product are input into the target output prediction model, and the predicted output feature vector is output; the predicted output feature vector and the future time feature vector are input into the fully connected neural network layer, and the target predicted output is output; this solves the problem of inaccurate output prediction for newly developed products caused by the existing technology.

[0039] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0040] Figure 1 A flowchart of a production output prediction method for a product provided in one embodiment of this application is shown below. Figure 1As shown, the production volume prediction method for manufactured products provided in this embodiment can be executed by any form of electronic device. For example, this embodiment uses a computer device as the execution subject for the method. The production volume prediction method for manufactured products provided in this embodiment includes the following steps:

[0041] Step S101: Obtain reference product information for each reference product and target product information for the target product.

[0042] For example, in this embodiment, the target product corresponds to the newly developed product mentioned in the prior art.

[0043] Step S102: Cluster each reference product according to the reference product information to obtain at least two reference product cluster sets.

[0044] For example, based on the reference product information of each reference product, the computer device generates a corresponding reference product feature vector according to the reference product information of each reference product; then, based on the vector distance between the reference product feature vectors of each reference product, clustering processing is performed to obtain at least two reference product cluster sets; specifically, the representation of the reference product cluster set is shown in equation (1).

[0045] (1)

[0046] in, Indicates the first Reference product feature vectors of each reference product; Indicates the first A set of reference product clusters; express A set of reference product clusters.

[0047] Step S103: For each reference product cluster set, execute steps S104-S105.

[0048] Step S104: Based on the reference product information of each reference product in the reference product cluster set, generate the production output feature vector of each reference product under each production time window.

[0049] For example, for each reference product cluster set, the reference product information of each reference product in the reference product cluster set is extracted by means of a sliding window, and then the production output feature vector of each reference product under each production time window is generated.

[0050] Specifically, the specific implementation steps of step S104 include:

[0051] Step S1041: Obtain the production time window based on the reference product information of each reference product in the reference product cluster set.

[0052] For example, for each reference product cluster set, the common production time characteristics of each reference product are determined based on the reference product information of each reference product in the reference product cluster set; then, the production time window is determined based on the production time characteristics, and the production time window represents the size of the sliding window.

[0053] Step S1042: Under the production time series of each reference product, feature extraction is performed on the reference product information of the corresponding reference product according to the production time window to generate the production output feature vector of each reference product under each production time window.

[0054] For example, for each reference product, feature extraction is performed on the reference product information based on the production time window to generate the production output feature vector of the reference product under each production time window; wherein, the representation of the generated production output feature vector is shown in Equation (2).

[0055] (2)

[0056] in, Indicates the first Production time series of each product; Indicates the size of the production time window; Used to indicate the One production time window; Indicates the first The product in the first Production output feature vector extracted within each production time window.

[0057] Step S105: Perform multi-head self-attention fusion processing on the production output feature vectors of each reference product under each production time window to generate the product relationship degree matrix of the corresponding reference product cluster set.

[0058] Specifically, under each production time window, the specific implementation steps of step S105 include:

[0059] Step S1051: Calculate the similarity of the production output feature vectors of each reference product to obtain the product similarity matrix between each reference product.

[0060] For example, the formula for calculating the similarity of the production output feature vectors of each reference product is shown in equation (3).

[0061] (3)

[0062] in, In the first Under a production time window, the product With products The product similarity value represents the product similarity score on the 1st... Products within a production time window With products Similarity; Indicates the first The product in the first Production output feature vector extracted within each production time window; Indicates the first The product in the first Production output feature vector extracted within each production time window.

[0063] Furthermore, based on the product similarity values ​​between each reference product, a product similarity matrix between each reference product is obtained.

[0064] Step S1052: Normalize the product similarity matrix between each reference product to obtain the normalized product similarity matrix between each reference product.

[0065] For example, the product similarity matrix between each reference product includes the product similarity value between each reference product. The calculation formula for normalizing the product similarity value between each reference product is shown in equation (4).

[0066] (4)

[0067] in, In the first Under a production time window, the product With products The normalized product similarity value.

[0068] Furthermore, based on the normalized product similarity values ​​among the reference products... The normalized product similarity matrix among the reference products is obtained. , That is to say, the first The normalized product similarity matrix corresponding to each reference product cluster set.

[0069] Furthermore, The set of normalized product similarity matrices corresponding to each reference product cluster set. As shown in equation (5).

[0070] (5)

[0071] Step S1053: Perform multi-head self-attention fusion processing on the normalized product similarity matrix among the reference products to generate the product relationship matrix of the corresponding reference product cluster set.

[0072] Specifically, for the normalized product similarity matrix among the reference products, that is, for the normalized product similarity matrix corresponding to the cluster set of each reference product, the specific implementation steps of step S1053 include:

[0073] Step S10531: Mapping is performed based on the self-attention mechanism to obtain the query vector, key vector, and value vector.

[0074] For example, the calculation formula for the self-attention mechanism mapping processing of the normalized product similarity matrix is ​​shown in Equation (6).

[0075] (6)

[0076] in, Indicates the first The normalized product similarity matrix corresponding to each reference product cluster set; Represents the query weight matrix; Represents the key weight matrix; Represents the value weight matrix; Indicates the first The query vector corresponding to each reference product cluster set; Indicates the first The key vectors corresponding to each reference product cluster set; Indicates the first The value vector corresponding to each reference product cluster set.

[0077] Step S10532: Generate a single attention vector for each reference product based on the query vector, key vector, and value vector.

[0078] For example, the formula for calculating the attention vector corresponding to each reference product is shown in equation (7).

[0079] (7)

[0080] in, Indicates the first One-head attention vector corresponding to each reference product cluster set; The vector dimension is represented by ; where the attention vector corresponding to the cluster set of reference products is the attention vector corresponding to each reference product.

[0081] Step S10533: Fuse the single-head attention vectors corresponding to each reference product to obtain a multi-head attention matrix.

[0082] For example, the one-head attention vectors corresponding to each reference product are fused, that is, the one-head attention vectors corresponding to each cluster set of reference products are fused, and then the calculation formula of the multi-head attention matrix is ​​obtained, as shown in Equation (8).

[0083] (8)

[0084] in, Represents a multi-head attention matrix; This represents the total number of heads receiving multi-head attention. This represents the output mapping matrix after multi-head attention fusion.

[0085] Step S10534: Reshape the dimensions of the multi-head attention matrix to generate a product relationship degree matrix for the corresponding reference product cluster set.

[0086] For example, the multi-head attention matrix is ​​dimensionally reshaped to generate a product relationship degree matrix for the corresponding reference product cluster set. The calculation formula is shown in equation (9).

[0087] (9)

[0088] Furthermore, the graph structure obtained based on the product relationship degree matrix As shown in equation (10).

[0089] (10)

[0090] in, Indicates the product node of each reference product; Represents the relationship edges between product nodes of each reference product; This represents the product relationship matrix.

[0091] Step S106: Based on the target product information, filter at least two reference product cluster sets to obtain the target reference product cluster set.

[0092] For example, based on the target product information of the target product, the target product feature vector of the target product is calculated; based on the reference product feature vectors of each reference product in the reference product cluster set, the category feature vector of the reference product cluster set is calculated; then, the vector similarity between the target product feature vector of the target product and the category feature vectors of each reference product cluster set is calculated, and the reference product cluster set corresponding to the maximum vector similarity is determined as the target reference product cluster set.

[0093] Step S107: Based on the target product information and the reference product information of each reference product in the target reference product cluster set, determine the target reference product from each reference product in the target reference product cluster set.

[0094] For example, based on the target product information, the target product feature vector is obtained; in the target reference product cluster set, based on the reference product information of each reference product, the corresponding reference product feature vector is obtained; then, by calculating the vector distance between the target product feature vector and the feature vectors of each reference product, the reference product corresponding to the smallest vector distance is determined as the target reference product.

[0095] In another possible implementation, for each reference product in the target reference product cluster set, the specific implementation steps of step S107 include:

[0096] Step S1071: Based on the target product information and the reference product information of each reference product, obtain the corresponding multiple distances.

[0097] For example, for each reference product, based on the target product information and the reference product information of the reference products, multiple distance calculation formulas are obtained, that is, the formulas for calculating the vector distance between the feature vector of the target product and the feature vector of the reference product are shown in equation (11).

[0098] (11)

[0099] Among them, the target product feature vector Indicates the target product In the feature vector Each feature element; reference product feature vector Indicates reference product In the feature vector The number of feature elements in the target product feature vector is equal to the number of feature elements in the reference product feature vector. This represents the Euclidean distance between the feature vectors of the target product and the feature vectors of the reference product. This represents the Manhattan distance between the feature vector of the target product and the feature vector of the reference product; This represents the dynamic time-warped distance between the feature vector of the target product and the feature vector of the reference product; Represents the feature vector of the target product The first in Each feature element; Represents the feature vector of the reference product The first in Each feature element; Represents the feature vector of the reference product The first in Each feature element; In and In The values ​​can be the same or different, for use in calculations. .

[0100] Step S1072: Perform a weighted calculation on multiple distances to obtain the comprehensive distance.

[0101] Step S1073: Based on the comprehensive distance between the target product and each reference product, determine the target reference product from each reference product in the target reference product cluster set.

[0102] For example, multiple distances are weighted and calculated to obtain the formula for calculating the comprehensive distance, as shown in equation (12).

[0103] (12)

[0104] in, Represents the target product feature vector and the first The combined distance between the feature vectors of each reference product; Indicates the first Weighting coefficients for different distances; Indicates the total number of distances; Represents the target product feature vector and the first The reference product feature vector corresponding to the first reference product Vector distance.

[0105] Therefore, the reference product corresponding to the minimum integrated distance is determined as the target reference product.

[0106] In steps S106-S107 of this embodiment, by first determining the target reference product cluster set from each reference product cluster set, and then determining the target reference product from the target reference product cluster set, the similarity between the determined target reference product and the target product is improved, thereby improving the matching degree between the target output prediction model trained in subsequent steps and the target product.

[0107] Step S108: Under each production time window: Train the initial output prediction model based on the product relationship degree matrix corresponding to the cluster set of the target reference product, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain the target output prediction model.

[0108] For example, under each production time window: the product relationship degree matrix corresponding to the cluster set of the target reference product is input into the graph convolutional neural network model of the initial production prediction model, and the corresponding product spatial structure feature vector is output; the product spatial structure feature vector corresponding to the cluster set of the target reference product and the reference product information of the target reference product are input into the long short-term memory network model of the initial production prediction model, and the corresponding predicted production output feature vector is output; the production output feature vector of the target reference product and the predicted production output feature vector of the target reference product are input into a preset loss function to obtain the loss value; based on the loss value, the initial production prediction model is trained to obtain the initial production prediction model; wherein, the reference product information of the target reference product includes the time characteristics corresponding to the production of the target reference product, such as the quarter, month, and week when the demand for the target reference product is high.

[0109] Furthermore, before training the initial production prediction model based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain the target production prediction model, the steps provided in this application embodiment also include a process for determining the initial production prediction model; specifically, the process for determining the initial production prediction model includes:

[0110] Step S1001: Input the product relationship degree matrix corresponding to each reference product cluster set into the graph convolutional neural network model of the original production prediction model, and output the corresponding product spatial structure feature vector.

[0111] Step S1002: Input the product spatial structure feature vector corresponding to each reference product cluster set and the reference product information of each reference product in each reference product cluster set into the long short-term memory network model of the original production prediction model, and output the corresponding predicted production output feature vector.

[0112] Step S1003: Input the production output feature vector of each reference product in each reference product cluster set and the predicted production output feature vector corresponding to each reference product cluster set into the preset loss function to obtain the loss value.

[0113] Step S1004: Train the original production prediction model based on the loss value to obtain the initial production prediction model.

[0114] In this embodiment, the initial production prediction model is first trained using the product relationship degree matrix corresponding to each reference product cluster set, the reference product information of each reference product in each reference product cluster set, and the production output feature vector of each reference product in each reference product cluster set to obtain an initial production prediction model. Then, the initial production prediction model is trained using the product relationship degree matrix corresponding to the target reference product cluster set determined based on the target product, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain a target production prediction model. This improves the matching degree between the target production prediction model and the target product.

[0115] Step S109: Input the product relationship degree matrix corresponding to the cluster set of target reference products and the target product information of the target products into the target output prediction model, and output the predicted output feature vector.

[0116] Step S110: Input the predicted output feature vector and the future time feature vector into the fully connected neural network layer and output the target predicted output.

[0117] For example, after obtaining the predicted output feature vector, the predicted output feature vector and the future time feature vector are concatenated, and then the concatenated comprehensive feature vector is input into a two-layer fully connected neural network layer to output the target predicted output.

[0118] Specifically, the predicted output feature vector and the future time feature vector are concatenated to obtain the expression for the comprehensive feature vector, as shown in equation (13).

[0119] (13)

[0120] in, Indicates the first The comprehensive feature vector at each time step; Indicates the first Predicted output feature vector at each time step; Indicates the first Future time feature vectors at each time step.

[0121] Furthermore, the integrated feature vector The input is fed into a two-layer fully connected neural network, and the output is an expression for the target predicted output, as shown in Equation (14).

[0122] (14)

[0123] in, This represents the weights of the first fully connected neural network layer; This represents the bias term of the first fully connected neural network layer; Indicates the activation function; Indicates the first Intermediate output values ​​of the hidden layer at each time step; This represents the weights of the second fully connected neural network layer; This represents the bias term of the second fully connected neural network layer; Indicates the first The target predicted output at each time step.

[0124] Furthermore, after outputting the target predicted output, the method provided in this application embodiment further includes: constructing a loss function based on the target predicted output and the actual output; and retraining the target output prediction model and the fully connected neural network layer based on the loss value corresponding to the loss function.

[0125] For example, a loss function is constructed based on the target predicted output and the actual output. The calculation formula is shown in equation (15).

[0126] (15)

[0127] in, Indicates the first Target predicted output at each time step; Indicates the first Actual output at each time step; This indicates the total number of time steps.

[0128] In this embodiment, based on obtaining the reference product information of each reference product and the target product information of the target product, at least two reference product cluster sets are obtained by clustering each reference product. Then, for each reference product cluster set: based on the reference product information of each reference product in the cluster set, a production output feature vector for each reference product under each production time window is generated, and then multi-head self-attention fusion processing is performed to generate the product relationship degree matrix of the corresponding reference product cluster set. Further, based on the target product information of the target product, the corresponding target reference product cluster set is matched, and the target reference product is determined within the target reference product cluster set. Furthermore, within each production time window: based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product, the initial output prediction model is trained to obtain the target output prediction model; the product relationship degree matrix corresponding to the target reference product cluster set and the target product information of the target product are input into the target output prediction model, and the predicted output feature vector is output; the predicted output feature vector and the future time feature vector are input into the fully connected neural network layer, and the target predicted output is output; this solves the problem of inaccurate output prediction for newly developed products caused by the existing technology.

[0129] Figure 2 This is a schematic diagram of the system architecture of a production volume prediction system for manufactured products provided in one embodiment of this application, as shown below. Figure 2 As shown, the production output prediction system includes a data acquisition module, a product relationship determination module, a similar product determination module, a model training module, and a production output prediction module.

[0130] The data acquisition module is used to acquire reference product information for each reference product and target product information for the target product.

[0131] The product relationship determination module is used to cluster each reference product based on reference product information to obtain at least two reference product cluster sets. For each reference product cluster set: based on the reference product information of each reference product in the reference product cluster set, the module generates the production output feature vector of each reference product under each production time window; and performs multi-head self-attention fusion processing on the production output feature vector of each reference product under each production time window to generate the product relationship degree matrix of the corresponding reference product cluster set.

[0132] The similar product determination module is used to filter at least two reference product cluster sets based on the target product information of the target product to obtain a target reference product cluster set; and to determine the target reference product from each reference product in the target reference product cluster set based on the target product information and the reference product information of each reference product in the target reference product cluster set.

[0133] The model training module is used to train the initial output prediction model under each production time window based on the product relationship degree matrix corresponding to the cluster set of the target reference product, the production output feature vector of the target reference product, and the reference product information of the target reference product, so as to obtain the target output prediction model.

[0134] The production forecasting module is used to input the product relationship degree matrix corresponding to the cluster set of the target reference product and the target product information into the target production forecasting model, and output the predicted production feature vector; the predicted production feature vector and the future time feature vector are input into the fully connected neural network layer, and the target predicted production is output.

[0135] Furthermore, before training the initial production prediction model based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product, the model training module is also used to input the product relationship degree matrix corresponding to each reference product cluster set into the graph convolutional neural network model of the original production prediction model, and output the corresponding product spatial structure feature vector; input the product spatial structure feature vector corresponding to each reference product cluster set and the reference product information of each reference product in each reference product cluster set into the long short-term memory network model of the original production prediction model, and output the corresponding predicted production output feature vector; input the production output feature vector of each reference product in each reference product cluster set and the predicted production output feature vector corresponding to each reference product cluster set into a preset loss function to obtain the loss value; and train the original production prediction model based on the loss value to obtain the initial production prediction model.

[0136] The production output prediction system for manufactured products provided in this embodiment can perform the following... Figure 1 The technical solutions of the method embodiments shown are similar in principle and in effect, and will not be described again here.

[0137] Figure 3 This is a schematic diagram of the structure of a production output prediction device for a product provided in one embodiment of this application, as shown below. Figure 3 As shown, the production output prediction device 3 for manufactured products provided in this embodiment includes:

[0138] Module 31 is used to acquire reference product information of each reference product and target product information of the target product;

[0139] Processing module 32 is used to cluster each reference product according to the reference product information to obtain at least two reference product cluster sets; for each reference product cluster set: based on the reference product information of each reference product in the reference product cluster set, generate the production output feature vector of each reference product under each production time window; perform multi-head self-attention fusion processing on the production output feature vector of each reference product under each production time window to generate the product relationship degree matrix of the corresponding reference product cluster set; filter at least two reference product cluster sets according to the target product information to obtain the target reference product cluster set; determine the target reference product from each reference product in the target reference product cluster set according to the target product information and the reference product information of each reference product in the target reference product cluster set; under each production time window: train the initial output prediction model according to the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain the target output prediction model;

[0140] Output module 33 is used to input the product relationship degree matrix corresponding to the cluster set of target reference products and the target product information of the target products into the target output prediction model and output the predicted output feature vector; input the predicted output feature vector and the future time feature vector into the fully connected neural network layer and output the target predicted output.

[0141] In one possible implementation, before training the initial production prediction model based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain the target production prediction model, the production prediction device 3, in the process of determining the initial production prediction model, specifically performs the following: inputting the product relationship degree matrix corresponding to each reference product cluster set into the graph convolutional neural network model of the original production prediction model, and outputting the corresponding product spatial structure feature vector; inputting the product spatial structure feature vector corresponding to each reference product cluster set and the reference product information of each reference product in each reference product cluster set into the long short-term memory network model of the original production prediction model, and outputting the corresponding predicted production output feature vector; inputting the production output feature vector of each reference product in each reference product cluster set and the predicted production output feature vector corresponding to each reference product cluster set into a preset loss function to obtain the loss value; and training the original production prediction model based on the loss value to obtain the initial production prediction model.

[0142] In one possible implementation, when processing module 32 performs multi-head self-attention fusion processing on the production output feature vectors of each reference product under each production time window to generate the product relationship degree matrix of the corresponding reference product cluster set, it specifically performs the following: under each production time window: calculates the similarity of the production output feature vectors of each reference product to obtain the product similarity matrix between each reference product; normalizes the product similarity matrix between each reference product to obtain the normalized product similarity matrix between each reference product; and performs multi-head self-attention fusion processing on the normalized product similarity matrix between each reference product to generate the product relationship degree matrix of the corresponding reference product cluster set.

[0143] In one possible implementation, when processing module 32 performs multi-head self-attention fusion processing on the normalized product similarity matrix between reference products to generate the product relationship degree matrix of the corresponding reference product cluster set, it specifically performs the following: For the normalized product similarity matrix between reference products: mapping based on the self-attention mechanism to obtain query vector, key vector, and value vector; generating a one-head attention vector corresponding to each reference product based on the query vector, key vector, and value vector; fusing the one-head attention vectors corresponding to each reference product to obtain a multi-head attention matrix; and reshaping the dimensions of the multi-head attention matrix to generate the product relationship degree matrix of the corresponding reference product cluster set.

[0144] In one possible implementation, when the processing module 32 determines the target reference product from the reference products in the target reference product cluster set based on the target product information and the reference product information of each reference product in the target reference product cluster set, it specifically performs the following: for each reference product in the target reference product cluster set: obtain multiple corresponding distances based on the target product information and the reference product information of each reference product; perform weighted calculation on the multiple distances to obtain a comprehensive distance; and determine the target reference product from the reference products in the target reference product cluster set based on the comprehensive distance between the target product and each reference product.

[0145] In one possible implementation, when the processing module 32 generates the production output feature vector of each reference product under each production time window based on the reference product information of each reference product in the reference product cluster set, it is specifically used to: obtain the production time window based on the reference product information of each reference product in the reference product cluster set; and extract features from the reference product information of the corresponding reference product under each production time window according to the production time sequence of each reference product, thereby generating the production output feature vector of each reference product under each production time window.

[0146] The acquisition module 31, processing module 32, and output module 33 are connected sequentially. The production output prediction device 3 for manufactured products provided in this embodiment can perform the following... Figure 1 The technical solutions of the method embodiments shown are similar in principle and in effect, and will not be described again here.

[0147] Figure 4 A schematic diagram of the structure of the electronic device provided in this application. Figure 4 As shown, the electronic device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus.

[0148] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0149] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0150] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0151] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0152] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0153] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0154] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0155] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0156] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0157] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0158] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0159] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0160] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0161] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0162] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for predicting the output of a manufactured product, characterized in that, The method includes: Obtain reference product information for each reference product and target product information for the target product; Based on the reference product information, the reference products are clustered to obtain at least two reference product cluster sets; For each reference product cluster set: Based on the reference product information of each reference product in the reference product cluster set, generate the production output feature vector of each reference product under each production time window; Multi-head self-attention fusion processing is performed on the production output feature vectors of each reference product under each production time window to generate the product relationship degree matrix of the corresponding reference product cluster set. Based on the target product information of the target product, the at least two reference product cluster sets are filtered to obtain the target reference product cluster set; Based on the target product information and the reference product information of each reference product in the target reference product cluster set, the target reference product is determined from each reference product in the target reference product cluster set; Under each production time window: Based on the product relationship degree matrix corresponding to the cluster set of the target reference product, the production output feature vector of the target reference product, and the reference product information of the target reference product, the initial output prediction model is trained to obtain the target output prediction model. The product relationship matrix corresponding to the cluster set of the target reference products and the target product information of the target products are input into the target output prediction model, and the predicted output feature vector is output. The predicted output feature vector and the future time feature vector are input into a fully connected neural network layer to output the target predicted output. Before training the initial production prediction model based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain the target production prediction model, the process of determining the initial production prediction model includes: The product relationship degree matrix corresponding to each reference product cluster set is input into the graph convolutional neural network model of the original output prediction model, and the corresponding product spatial structure feature vector is output. The product spatial structure feature vector corresponding to each reference product cluster set and the reference product information of each reference product in each reference product cluster set are input into the long short-term memory network model of the original production prediction model, and the corresponding predicted production output feature vector is output. The production output feature vector of each reference product in each reference product cluster set and the predicted production output feature vector corresponding to each reference product cluster set are input into the preset loss function to obtain the loss value; Based on the loss value, the original output prediction model is trained to obtain the initial output prediction model; The step of performing multi-head self-attention fusion processing on the production output feature vectors of each reference product under each production time window to generate a product relationship degree matrix of the corresponding reference product cluster set includes: Under each production time window: The similarity of the production output feature vectors of each reference product is calculated to obtain the product similarity matrix between the reference products. The product similarity matrix between each reference product is normalized to obtain the normalized product similarity matrix between each reference product. Multi-head self-attention fusion processing is performed on the normalized product similarity matrix among the reference products to generate the product relationship matrix of the corresponding reference product cluster set.

2. The method according to claim 1, characterized in that, The step of performing multi-head self-attention fusion processing on the normalized product similarity matrix among each reference product to generate the product relationship degree matrix of the corresponding reference product cluster set includes: For the normalized product similarity matrix among the reference products: The mapping is based on a self-attention mechanism to obtain query vector, key vector and value vector; Based on the query vector, the key vector, and the value vector, generate a single-headed attention vector for each reference product; The single-head attention vectors corresponding to each reference product are fused to obtain a multi-head attention matrix; The multi-head attention matrix is ​​dimensionally reshaped to generate a product relationship degree matrix for the corresponding reference product cluster set.

3. The method according to claim 1, characterized in that, The step of determining the target reference product from the reference products in the target reference product cluster set based on the target product information and the reference product information of each reference product in the target reference product cluster set includes: For each reference product in the target reference product cluster set: Based on the target product information and the reference product information of each reference product, multiple corresponding distances are obtained; The multiple distances are weighted and calculated to obtain the comprehensive distance; The target reference product is determined from the reference products in the target reference product cluster set based on the comprehensive distance between the target product and each reference product.

4. The method according to any one of claims 1-3, characterized in that, The step of generating a production output feature vector for each reference product under each production time window based on the reference product information of each reference product in the reference product cluster set includes: The production time window is obtained based on the reference product information of each reference product in the reference product cluster set; Under the production time series of each reference product, feature extraction is performed on the reference product information of the corresponding reference product according to the production time window to generate the production output feature vector of each reference product under each production time window.

5. A production output prediction device for a manufactured product, characterized in that, include: The acquisition module is used to acquire reference product information for each reference product and target product information for the target product. The processing module is configured to: cluster the reference products according to the reference product information to obtain at least two reference product cluster sets; for each reference product cluster set: generate production output feature vectors for each reference product under each production time window based on the reference product information of each reference product in the reference product cluster set; perform multi-head self-attention fusion processing on the production output feature vectors of each reference product under each production time window to generate a product relationship degree matrix for the corresponding reference product cluster set; filter the at least two reference product cluster sets according to the target product information to obtain a target reference product cluster set; and determine the target reference product from the reference products in the target reference product cluster set based on the target product information and the reference product information of each reference product in the target reference product cluster set. Under each production time window: the initial output prediction model is trained based on the product relationship degree matrix corresponding to the cluster set of the target reference product, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain the target output prediction model; The output module is used to input the product relationship degree matrix corresponding to the cluster set of the target reference products and the target product information of the target products into the target output prediction model, and output the predicted output feature vector; the predicted output feature vector and the future time feature vector are input into the fully connected neural network layer to output the target predicted output. Before training the initial production prediction model based on the product relationship degree matrix corresponding to the target reference product cluster set, the production output feature vector of the target reference product, and the reference product information of the target reference product to obtain the target production prediction model, the production prediction device specifically performs the following steps in determining the initial production prediction model: Inputting the product relationship degree matrix corresponding to each reference product cluster set into the graph convolutional neural network model of the original production prediction model, and outputting the corresponding product spatial structure feature vector; inputting the product spatial structure feature vector corresponding to each reference product cluster set and the reference product information of each reference product in each reference product cluster set into the long short-term memory network model of the original production prediction model, and outputting the corresponding predicted production output feature vector; inputting the production output feature vector of each reference product in each reference product cluster set and the predicted production output feature vector corresponding to each reference product cluster set into a preset loss function to obtain a loss value; and training the original production prediction model based on the loss value to obtain the initial production prediction model. When the processing module performs multi-head self-attention fusion processing on the production output feature vectors of each reference product under each production time window to generate the product relationship degree matrix of the corresponding reference product cluster set, it specifically performs the following: Under each production time window: calculate the similarity of the production output feature vectors of each reference product to obtain the product similarity matrix between each reference product; normalize the product similarity matrix between each reference product to obtain the normalized product similarity matrix between each reference product; and perform multi-head self-attention fusion processing on the normalized product similarity matrix between each reference product to generate the product relationship degree matrix of the corresponding reference product cluster set.

6. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 4.

8. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 4.

Images