Machine learning based cost data determination method, apparatus, computer program product

Abstract

This application relates to a method, apparatus, and computer program product for determining cost data based on machine learning. The method includes: acquiring part data of a target part; constructing a feature vector corresponding to the target part using the part data; determining the similarity between the target part and historical parts by using the distance between the feature vector of the target part and the feature vectors of pre-stored historical parts; constructing learning data using the feature vectors of historical parts and historical cost data, and training a machine learning model using the learning data; and, when the similarity meets a preset numerical range, using the machine learning model to predict the cost using the feature vector of the target part, thereby obtaining the target cost data of the target part. This method can utilize historical parts that are moderately similar to the target part to construct learning data to train the machine learning model, ensuring the reliability of the model's prediction results and thus improving the accuracy of the target cost data.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and in particular to a method, apparatus, computer device, computer-readable storage medium, and computer program product for determining cost data based on machine learning. Background Technology

[0002] In the machining industry, mold parts and complex machined parts are typically non-standardized, custom-produced components. Because these parts have varied drawing requirements and complex manufacturing processes, it's difficult to achieve the same immediate cost estimates as standard parts. Currently, experienced employees primarily estimate the manufacturing time and costs for non-standardized parts. However, differences in analysis and estimation among different personnel for the same part can lead to fluctuations in the output cost data. Therefore, a more objective and accurate method for calculating the cost of non-standardized parts is urgently needed. Summary of the Invention

[0003] Therefore, it is necessary to provide a method, apparatus, computer device, computer-readable storage medium, and computer program product for determining cost data based on machine learning, which can improve the accuracy of data, in response to the above-mentioned technical problems.

[0004] Firstly, this application provides a method for determining cost data based on machine learning, including:

[0005] Obtain the part data of the target part, and use the part data to construct the feature vector corresponding to the target part;

[0006] The similarity between the target part and the historical parts is determined by using the distance between the feature vector of the target part and the feature vector of the pre-stored historical parts.

[0007] Learning data is constructed using the feature vectors of the historical parts and historical cost data, and a machine learning model is trained using the learning data.

[0008] If the similarity meets the preset numerical range, the machine learning model is used to predict the cost of the target part using its feature vector to obtain the target cost data of the target part.

[0009] In one embodiment, the machine learning model integrates a gradient boosting tree model and a neural network model; the step of using the machine learning model to perform cost prediction using the feature vector of the target part to obtain the target cost data of the target part includes:

[0010] The feature vector of the target part is input into the gradient boosting tree model for regression prediction to obtain the first prediction data;

[0011] The feature vector of the target part is input into the neural network model to predict the cost, and a second prediction data is obtained.

[0012] The target cost data is obtained by combining the first prediction data and the second prediction data.

[0013] In one embodiment, when the part data is two-dimensional data, constructing the feature vector corresponding to the target part using the part data includes:

[0014] Using the geometric parameters in the part data, construct the geometric feature parameters of the target part;

[0015] Using the layer parameters in the part data, construct the layer feature parameters of the target part;

[0016] Using the dimensional parameters in the part data, construct the dimensional feature parameters of the target part;

[0017] The geometric feature parameters, the layer feature parameters, and the size feature parameters are used as feature data of the target part in multiple dimensions.

[0018] The feature vector is constructed using the feature data across the multiple dimensions.

[0019] In one embodiment, when the part data is three-dimensional data, the step of obtaining the part data of the target part and constructing a feature vector corresponding to the target part using the part data includes:

[0020] Using the geometric parameters in the part data, construct the geometric feature parameters of the target part;

[0021] Using the topology parameters in the part data, construct the topology feature parameters of the target part;

[0022] Using the shape parameters in the part data, construct the shape descriptor parameters of the target part;

[0023] Using the geometric parameters, topological parameters, shape parameters, and material parameters in the part data, the part complexity parameters of the target part are constructed.

[0024] The part complexity parameter, the geometric feature parameter, the topological feature parameter, the shape descriptor parameter, and the part production quantity and part precision requirement parameter in the part data are used as the feature data of the target part in multiple dimensions.

[0025] The feature vector is constructed using the feature data across the multiple dimensions.

[0026] In one embodiment, constructing the feature vector using feature data from the multiple dimensions includes:

[0027] Read the initial feature weights for each dimension respectively;

[0028] Based on the part type of the target part, the initial feature weights are adjusted to obtain the target feature weights corresponding to each dimension;

[0029] The target feature weights corresponding to each dimension are normalized to obtain normalized weights;

[0030] The feature data under each dimension are normalized using the normalized weights corresponding to each dimension, and the normalized feature data are combined to obtain the feature vector of the target part.

[0031] In one embodiment, adjusting the initial feature weights according to the part type of the target part to obtain the target feature weights corresponding to each dimension includes:

[0032] When the part type is a small and simple part type, the initial feature weights corresponding to the size dimension and the basic geometric dimension are increased, and the initial feature weights of the topology dimension and the shape descriptor dimension are decreased, so as to obtain the target feature weights of the size dimension, the basic geometric dimension, the topology dimension and the shape descriptor dimension.

[0033] When the part type is a complex part type, the initial feature weights of the topology dimension, the shape descriptor dimension, and the convex hull dimension are increased, and the initial feature weights corresponding to the size dimension are decreased, so as to obtain the target feature weights of the topology dimension, the shape descriptor dimension, the convex hull dimension, and the size dimension.

[0034] When the part type is a large part type, the initial feature weights corresponding to the basic geometric dimension and the moment of inertia dimension are increased, and the initial feature weights of the curvature dimension and the symmetry dimension are decreased to obtain the target feature weights of the basic geometric dimension, the moment of inertia dimension, the curvature dimension and the symmetry dimension.

[0035] When the part type is a flat part type, the initial feature weights of the symmetry dimension and the shape descriptor dimension are increased to obtain the target feature weights of the symmetry dimension and the shape descriptor dimension.

[0036] In one embodiment, determining the similarity between the target part and the historical parts by utilizing the distance between the feature vector of the target part and the feature vector of pre-stored historical parts includes:

[0037] Multiple historical part clusters are obtained by iteratively clustering using the feature vector distances between the various historical parts.

[0038] By using the distances between the feature vector of the target part and the center feature vector of each of the historical part clusters, the target part cluster corresponding to the target part is determined from the multiple historical part clusters;

[0039] The similarity between the target part and each historical part in the target part cluster is determined by using the distance between the feature vector of the target part and the feature vector of each historical part in the target part cluster.

[0040] In one embodiment, the method further includes:

[0041] Determine the basic cost data of the target part based on the part data;

[0042] In the event that the machine learning model fails to predict the cost, the average similarity between the target part and the historical parts corresponding to the learning data is determined. The failure of the machine learning model to predict the cost includes at least one of the following: the machine learning model does not exist; the amount of the learning data is less than a preset data amount threshold; and the machine learning model reports an error during the cost prediction process.

[0043] The similarity weights of the historical parts corresponding to the learning data are determined based on the similarity.

[0044] Determine the average cost data corresponding to the learning data;

[0045] The target cost data for the target part is generated using the average cost data, the similarity weight, and the basic cost data.

[0046] Secondly, this application also provides a cost data determination apparatus based on machine learning, comprising:

[0047] The vector construction module is used to acquire the part data of the target part and construct the feature vector corresponding to the target part using the part data;

[0048] The similarity calculation module is used to determine the similarity between the target part and the historical parts by using the distance between the feature vector of the target part and the feature vector of the pre-stored historical parts;

[0049] The cost prediction module is used to construct learning data using the feature vectors of the historical parts and historical cost data, and to train a machine learning model using the learning data; when the similarity meets a preset numerical range, the machine learning model is used to perform cost prediction using the feature vectors of the target part to obtain the target cost data of the target part.

[0050] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the machine learning-based cost data determination method described in any of the embodiments of the first aspect.

[0051] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the machine learning-based cost data determination method described in any of the embodiments of the first aspect.

[0052] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the machine learning-based cost data determination method described in any of the embodiments of the first aspect.

[0053] The aforementioned method, apparatus, computer equipment, computer-readable storage medium, and computer program product for determining cost data based on machine learning construct feature vectors for target parts using part data. They then use the similarity between feature vectors and a preset numerical range to filter historical parts that are moderately similar to the target part. Furthermore, they construct learning data using the feature vectors of historical parts and historical cost data, and train a machine learning model using this learning data. When the similarity meets the preset numerical range, the machine learning model uses the feature vectors of the target part to predict the cost, thus obtaining the target cost data for the target part. This method can find historical parts that are moderately similar to the target part and use the historical parts to construct learning data to train the machine learning model, ensuring the reliability of the model's prediction results and thereby improving the accuracy of the target cost data. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1This is a flowchart illustrating a machine learning-based cost data determination method in one embodiment.

[0056] Figure 2 This is a flowchart illustrating the steps for constructing a vector from two-dimensional data for part data in one embodiment.

[0057] Figure 3 This is a flowchart illustrating the steps involved in constructing a vector from three-dimensional data for part data in one embodiment.

[0058] Figure 4 This is a flowchart illustrating the feature weight normalization process in one embodiment.

[0059] Figure 5 This is a flowchart illustrating the similarity calculation steps in one embodiment;

[0060] Figure 6 This is a flowchart illustrating the target cost data calculation steps in one embodiment;

[0061] Figure 7 A structural block diagram of a machine learning-based cost data determination device 700 in one embodiment;

[0062] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements / data objects, but these elements / data objects are not limited by these terms. These terms are only used to distinguish the first element / data object from the second element / data object. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the solutions, or any combination of multiple solutions. 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, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with relevant regulations.

[0065] In one embodiment, such as Figure 1As shown, a method for determining cost data based on machine learning is provided. This embodiment illustrates the application of this method to a server, but it is understood that the method can also be applied to a terminal, or to a system including both a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps S102 to S108, wherein:

[0066] Step S102: Obtain the part data of the target part and construct the feature vector corresponding to the target part using the part data.

[0067] The target part can be used to characterize non-standardized parts for which cost data is unknown. Part data can be data describing the target part's model shape, structure, color, material, manufacturing process, and other attributes.

[0068] For example, the server can receive part data of the target part uploaded by the user through a client interface. Alternatively, it can download part engineering drawings or part model files from a data communication network through an application programming interface (API), and obtain the part data of the target part from the part engineering drawings or part model data. The part data is then vectorized to obtain the feature vector of the target part.

[0069] Step S104: Determine the similarity between the target part and the historical parts by using the distance between the feature vector of the target part and the feature vector of the pre-stored historical parts.

[0070] Historical parts can be used to characterize parts with known costs. Optionally, in some implementations, after obtaining the target cost data of the target part, the target part can be stored as a historical part for subsequent use in the calculation of the target cost data of new parts.

[0071] For example, the server can pre-store historical part data and feature vectors constructed using this data in a database. The feature vectors of the target part and the historical parts are then processed using Euclidean distance, cosine distance, and other distance calculation formulas to obtain the distance between them. Based on the conversion relationship between distance and similarity, the distance between the feature vectors is converted into a corresponding similarity score, thus obtaining the similarity between the target part and the historical parts.

[0072] Step S106: Construct learning data using the feature vectors of historical parts and historical cost data, and use the learning data to train a machine learning model.

[0073] For example, the server can use the feature vectors of historical parts as sample data and the historical cost data of historical parts as sample labels, combining the corresponding sample data and sample labels to obtain learning data. The learning data is then input into an initial machine learning model for iterative training. Training stops when the loss value of the machine learning model is less than a preset loss threshold or the number of training iterations reaches a preset threshold, resulting in the trained machine learning model.

[0074] Step S108: If the similarity meets the preset numerical range, the target cost data of the target part is obtained by using the feature vector of the target part through a machine learning model to predict the cost.

[0075] For example, the server may store a preset numerical range. Optionally, in some embodiments, the preset numerical range may be used to characterize the numerical range corresponding to moderate similarity, such as 60%–79%. The server can compare the currently calculated similarity with the preset numerical range. If the similarity between the target part and historical parts meets the preset numerical range, the feature vector of the target part is input into the machine learning model. Based on the data relationship between the feature vectors of historical parts and historical cost data learned during the training phase, the machine learning model predicts and generates target cost data corresponding to the feature vector of the target part.

[0076] Optionally, in some implementations, the material, production quantity, and target cost data of the target part can be used for calculation to obtain the final production cost data of the target part.

[0077] In the aforementioned machine learning-based cost data determination method, feature vectors of target parts are constructed using part data. Historical parts that are moderately similar to the target part are filtered based on the similarity between feature vectors and a preset numerical range. Learning data is constructed using the feature vectors of historical parts and historical cost data, and a machine learning model is trained using this learning data. When the similarity meets the preset numerical range, the machine learning model uses the feature vectors of the target part to predict the cost, obtaining the target cost data for the target part. This method can find historical parts that are moderately similar to the target part and use the historical parts to construct learning data to train the machine learning model, ensuring the reliability of the model's prediction results and thus improving the accuracy of the target cost data.

[0078] In an exemplary embodiment, the machine learning model integrates a gradient boosting tree model and a neural network model. Step S108 further includes: inputting the feature vector of the target part into the gradient boosting tree model for regression prediction to obtain first prediction data; inputting the feature vector of the target part into the neural network model for cost prediction to obtain second prediction data; and fusing the first prediction data and the second prediction data to obtain target cost data.

[0079] Among them, the gradient boosting tree model (eXtreme Gradient Boosting, or XGBoost for short) is an ensemble learning algorithm model based on the gradient boosting framework. It achieves high-precision regression or classification prediction by constructing multiple decision trees and weighting and combining their results.

[0080] Neural network models can be used to characterize a machine learning model that simulates the structure of neurons in the human brain. Through the connection weights between multiple layers of neurons, they learn the complex mapping relationship between input features and output results.

[0081] For example, the server can input the feature vector of the target part into a gradient boosting tree model trained on learning data. Each decision tree in the gradient boosting tree model performs regression prediction on the feature vector of the target part. Finally, the outputs of each decision tree are weighted and combined to obtain the first predicted cost data for the target part. Alternatively, the server can input the feature vector of the target part into a neural network model trained on learning data. The neural network model uses model parameters trained based on the feature vectors and historical cost data of historical parts to process the feature vector of the target part and generate a second predicted cost data for the target part. The average of the first and second predicted data is calculated as the final target cost data. Alternatively, prediction weights can be set for the gradient boosting tree model and the neural network model respectively. The first and second predicted data are then weighted and summed using the corresponding prediction weights to obtain the target cost data.

[0082] Optionally, in some implementations, the number of decision trees can be between 100 and 1000. A larger number of decision trees results in greater complexity and stronger predictive power for the gradient boosting tree model. A smaller number of decision trees leads to higher predictive efficiency. In other implementations, the learning rate of the gradient boosting tree model can be set to 0.01-0.3 to control the contribution weight of each decision tree and prevent overfitting. In other implementations, the maximum depth of the decision tree can be set to 3-10 to control the complexity of a single decision tree. In other implementations, the minimum weight of the child nodes of the decision tree can be set to 1-10 to control the growth of the decision tree and prevent overfitting. In other implementations, the training sample sampling ratio and feature sampling ratio of the gradient boosting tree model can be set to 0.8-1.0 to control randomness and prevent overfitting. In other implementations, a splitting threshold of 0-5 can be set to control the splitting of the decision tree and prevent overfitting. In other implementations, a gradient boosting tree model can be used to combine the feature vectors, process information, and material information of the target part for prediction.

[0083] Optionally, in some embodiments, a neural network model can be used to combine geometric features, topological features, and process features from the feature vector of the target part for cost prediction. In other embodiments, the network structure of the neural network model can include 3-5 neuron layers. Each neuron layer can include 32-256 neurons. The network structure can determine the model complexity and learning ability of the neural network model. In other embodiments, the neural network model can use the ReLU function as the activation function to activate neurons. The neural network model can introduce nonlinear relationships through this activation function, improving the model's expressive power. In other embodiments, the neural network model can use the Adam algorithm as the weight update algorithm of the optimizer to accelerate model convergence and improve model training efficiency. In other embodiments, the learning rate used by the neural network model to determine the weight update step size can be set to 0.001-0.01 to control the model's convergence speed. In other embodiments, the number of samples for each training iteration of the neural network model can be set to 32-128 to balance training speed and model stability. In other embodiments, the regularization parameter of the neural network model can be set to Dropout (0.2-0.5) to prevent overfitting and improve the model's generalization ability.

[0084] In this embodiment, a gradient boosting tree model is used to initially predict the cost of the target part, which can handle complex nonlinear relationships and improve the prediction accuracy of the first prediction result. A neural network model is then used to perform a secondary cost prediction of the target part, which can capture the complex nonlinear relationships between part features, automatically learn effective features, better adapt to new part types and process changes, and improve the prediction accuracy of the second prediction data. Finally, the cost prediction results of the heterogeneous gradient boosting tree model and neural network model are fused to obtain the target cost data of the target part. This improves the generalization ability and adaptability of the machine learning model, reduces the bias of single-model predictions, and improves the prediction accuracy of the target cost data.

[0085] In one exemplary embodiment, such as Figure 2 As shown, when the part data of the target part is two-dimensional data, step S102 includes steps S202 to S210. Wherein:

[0086] Step S202: Construct the geometric feature parameters of the target part using the geometric parameters in the part data.

[0087] The geometric parameters may include, but are not limited to, entity distribution parameters, bounding box parameters, entity quantity parameters, shape complexity parameters, length parameters, and area parameters.

[0088] For example, the server can load a 2D vector graphics file (Drawing Exchange Format, DXF) of the target part to obtain part data belonging to 2D data. Entities such as lines, circles, arcs, and polylines are extracted from the part data, and the quantity of each type of entity is counted. Each type of entity is treated as a vector dimension, and the quantities of various entities are combined to form a vector-based entity distribution parameter. For example, the entity distribution parameter could be {(line: 3), (circle: 5), (arc: 7)}. The total quantity of each type of entity is calculated to obtain the entity quantity parameter. Bounding boxes in the part data are identified, and their aspect ratio, area, width, height, and minimum / maximum coordinates are calculated. The center point of the bounding box is calculated using the minimum / maximum coordinates. The aspect ratio, area, width, height, and center point are treated as vector dimensions and combined to form a vector-based bounding box parameter. The shape complexity parameter is calculated using the roundness of the circle and the aspect ratio of the bounding box. The total length of the target part is calculated using the part data as the length parameter, and the total area of ​​the target part is calculated as the area parameter. By combining entity distribution parameters, bounding box parameters, entity quantity parameters, shape complexity parameters, length parameters, and area parameters, the geometric feature parameters of the target part are obtained.

[0089] Optionally, in some implementations, during the subsequent calculation of the distance between feature vectors, the similarity between entity distribution parameters, bounding box parameters, entity quantity parameters, shape complexity parameters, length parameters, and area parameters can be calculated separately. Then, the corresponding coefficients are multiplied by the similarity between entity distribution parameters, bounding box parameters, entity quantity parameters, shape complexity parameters, length parameters, and area parameters, and summed to obtain the geometric feature similarity between the target part and the historical part. The coefficients range from 0 to 1.

[0090] Step S204: Construct the layer feature parameters of the target part using the layer parameters in the part data.

[0091] The layer parameters may include, but are not limited to, the number of layers and the color distribution parameters.

[0092] For example, the server can count the number of layers in the part data to obtain a layer count parameter. It can then extract the usage frequency of each color. Each color is treated as a vector dimension, and the usage frequencies of various colors are combined to form a vector-like color distribution parameter. For example, the color distribution parameter could be {(red: 13), (yellow: 15), (blue: 17)}. Combining the layer count parameter and the color distribution parameter yields the layer feature parameters of the target part.

[0093] Optionally, in some implementations, during the subsequent calculation of the distance between feature vectors, the similarity between the layer quantity parameter and the similarity between the color distribution parameter can be calculated separately. Then, the corresponding coefficients are multiplied by the similarity between the layer quantity parameter and the similarity between the color distribution parameter, and the results are summed to obtain the layer feature similarity between the target part and the historical part. The coefficients range from 0 to 1.

[0094] Step S206: Construct the dimensional feature parameters of the target part using the dimensional parameters in the part data.

[0095] The dimensional parameters may include, but are not limited to, quantity parameters, linear dimensional parameters, angular dimensional parameters, and radial dimensional parameters.

[0096] For example, the server can extract the total number of dimension annotations from the part data to obtain the dimension quantity parameter. It then categorizes and counts the number of linear, angular, and radial dimensions to obtain linear dimension parameters, angular dimension parameters, and radial dimension parameters. Finally, it combines the dimension quantity parameter, linear dimension parameter, angular dimension parameter, and radial dimension parameter to obtain the dimensional feature parameters of the target part.

[0097] Optionally, in some implementations, during the subsequent calculation of the distance between feature vectors, the similarity between quantity parameters, the similarity between linear dimension parameters, the similarity between angular dimension parameters, and the similarity between radial dimension parameters can be calculated separately. Then, the corresponding coefficients are multiplied by the similarity between quantity parameters, the similarity between linear dimension parameters, and the similarity between angular dimension parameters and radial dimension parameters, and summed to obtain the dimensional feature similarity between the target part and the historical part. The coefficients range from 0 to 1.

[0098] Step S208: Use geometric feature parameters, layer feature parameters, and dimensional feature parameters as feature data of the target part in multiple dimensions.

[0099] Step S210: Construct feature vectors using feature data from multiple dimensions.

[0100] For example, the server combines geometric feature parameters, layer feature parameters, and dimensional feature parameters as feature data of the target part in multiple dimensions to form a multi-dimensional feature vector.

[0101] Optionally, in some implementations, during the subsequent calculation of the distance between feature vectors, the corresponding coefficients can be multiplied by the geometric feature similarity, layer feature similarity, and size feature similarity, and then summed to obtain the similarity between the target part and the historical parts. The coefficients range from 0 to 1.

[0102] In this embodiment, by extracting geometric feature parameters, layer feature parameters, and dimensional feature parameters to construct the feature vector of the target part when the part data is two-dimensional data, it is possible to retain key features such as the dimensional constraints, geometric structure, and layer process requirements of the target part, thereby improving the data integrity of the feature vector.

[0103] In one example, such as Figure 3 As shown, when the part data of the target part is three-dimensional data, step S102 includes steps S302 to S312. Wherein:

[0104] Step S302: Construct the geometric feature parameters of the target part using the geometric parameters in the part data.

[0105] The geometric parameters may include, but are not limited to, length, volume, and surface area.

[0106] For example, the server can load the 3D solid model file (Standard for the Exchange of Product model data, STP) of the target part to obtain part data. The total length, total volume, and total surface area of ​​the target part are calculated using the part data. The total length, total volume, and total surface area are then combined as elements in a vector dimension to obtain geometric feature parameters.

[0107] Optionally, in some implementations, during the subsequent calculation of the distance between feature vectors, the similarity of length, volume, and surface area can be calculated separately, and the corresponding coefficients can be multiplied by the similarity of length, volume, and surface area respectively, and then summed to obtain the geometric feature similarity between the target part and the historical part. The coefficients range from 0 to 1.

[0108] Step S304: Construct the topological feature parameters of the target part using the topological parameters in the part data.

[0109] The topology parameters may include, but are not limited to, the number of faces, the number of edges, and the number of vertices.

[0110] For example, the server can identify faces, edges, and vertices in the part data, and count the number of faces, edges, and vertices. Using the number of faces, edges, and vertices as elements in a vector dimension, they are combined to obtain topological feature parameters.

[0111] Optionally, in some implementations, during the subsequent calculation of the distance between feature vectors, the similarity of the number of faces, edges, and vertices can be calculated separately, and the corresponding coefficients can be multiplied by the similarity of the number of faces, edges, and vertices respectively, and then summed to obtain the topological feature similarity between the target part and the historical part. The coefficients range from 0 to 1.

[0112] Step S306: Construct the shape descriptor parameters of the target part using the shape parameters in the part data.

[0113] The shape parameters may include, but are not limited to, sphericity, elongation, and compactness.

[0114] Sphericity can be used to characterize how similar the shape of a target part is to an ideal sphere. Optionally, in some embodiments, sphericity can be calculated with reference to the following formula:

[0115] Sphericity = (π^(1 / 3) × (6 × volume)^(2 / 3)) / surface area.

[0116] Elongation is an indicator that measures the ratio of the length to the width and height of a target part, characterizing the slenderness of the target part. Optionally, in some embodiments, elongation can be calculated using the following formula: Elongation = Maximum dimension / Minimum dimension.

[0117] Compactness can be used to characterize the relationship between the volume and surface area of ​​a target part. For the same volume, the smaller the surface area, the higher the compactness.

[0118] For example, the server can read the sphericity, elongation, and compactness from the part data. Using the sphericity, elongation, and compactness of the target part as elements in a vector dimension, they are combined to obtain the shape descriptor parameters.

[0119] Optionally, in some implementations, during the subsequent calculation of the distance between feature vectors, the similarity of sphericity, elongation, and compactness can be calculated separately, and the corresponding coefficients can be multiplied by the similarity of sphericity, elongation, and compactness, and then summed to obtain the shape descriptor similarity between the target part and the historical part. The coefficients range from 0 to 1.

[0120] Step S308: Construct the part complexity parameters of the target part using geometric parameters, topological parameters, shape parameters, and material parameters from the part data.

[0121] The material parameters may include the material complexity coefficient corresponding to the material of the target part.

[0122] For example, the server can calculate the basic complexity of the target part using topology parameters such as the number of faces, edges, and vertices. It can calculate the shape complexity using shape parameters such as sphericity and elongation. It can calculate convexity using geometric parameters such as volume, and then calculate the convex hull complexity using the convexity. It can calculate the dimensional complexity using geometric parameters such as dimensional variance. Finally, using the basic complexity, shape complexity, convex hull complexity, dimensional complexity, and material complexity coefficients, the server can calculate the part complexity parameters of the target part.

[0123] Alternatively, in some implementations, the basic complexity can be calculated using the following formula:

[0124] Basic complexity = number of faces × 0.5 + number of edges × 0.3 + number of vertices × 0.2.

[0125] Optionally, in some implementations, shape complexity may include:

[0126] Shape complexity based on sphericity: (1 - sphericity) × 100;

[0127] Shape complexity based on elongation: (elongation - 1) × 50.

[0128] Optionally, in some embodiments, convexity is used to characterize the ratio of the total volume of the target part to the volume of its convex hull portion, and its value ranges from (0,1). The convex hull complexity can be determined with reference to the following formula:

[0129] Convexity = Total volume / Volume of convex hull;

[0130] The complexity of convex hull = (1 - convexity) × 100.

[0131] Alternatively, in some implementations, the size complexity can be calculated using the following formula:

[0132] Size complexity = size variance × 50.

[0133] Optionally, in some implementations, the material complexity coefficient is 1.5 when the target part is made of a difficult-to-machine material (such as SKD11, SKD61, etc.). The material complexity coefficient is 0.8 when the target part is made of an easy-to-machine material (such as aluminum alloy). The material complexity coefficient is 1.0 when the target part is made of ordinary steel (such as S45C).

[0134] Alternatively, in some implementations, the part complexity coefficient can be obtained with reference to the following formula:

[0135] Part complexity coefficient = (basic complexity + shape complexity + convex hull complexity + dimensional complexity) × material complexity coefficient.

[0136] Step S310: Use the part complexity parameters, geometric feature parameters, topological feature parameters, shape descriptor parameters, part production quantity and part accuracy requirement parameters in the part data as feature data of the target part in multiple dimensions.

[0137] Step S312: Construct feature vectors using feature data from multiple dimensions.

[0138] For example, the server can refer to the feature vector construction method provided in the above embodiments to combine the part complexity parameters, geometric feature parameters, topological feature parameters, shape descriptor parameters, part production quantity and part accuracy requirement parameters in the part data as feature data of the target part in multiple dimensions to obtain a feature vector.

[0139] In this embodiment, when the part data of the target part is three-dimensional, feature vectors are constructed by extracting part complexity parameters, geometric feature parameters, topological feature parameters, shape descriptor parameters, part production quantity and part precision requirement parameters from the part data, which can improve the data integrity of the feature vectors.

[0140] In one exemplary embodiment, such as Figure 4 As shown, the feature vector of the target part can be constructed by referring to the following steps S402 to S408, wherein:

[0141] Step S402: Read the initial feature weights corresponding to each dimension.

[0142] Step S404: Adjust the initial feature weights according to the part type of the target part to obtain the target feature weights corresponding to each dimension.

[0143] The part types can include, but are not limited to, small simple part types, complex part types, large part types, and flat part types. Small simple part types can be used to characterize parts with a volume less than 1,000,000 cubic millimeters and fewer than 100 faces. Complex part types can be used to characterize parts with more than 500 faces. Large part types can be used to characterize parts with a volume greater than 1,000,000 cubic millimeters. Flat part types can be used to characterize parts with a flatness greater than 5.0.

[0144] For example, when the target part is a small and simple part, it is necessary to increase the initial feature weights corresponding to the size dimension (e.g., target feature weight = initial feature weight × 1.2) and the initial feature weights corresponding to the basic geometric dimension (e.g., target feature weight = initial feature weight × 1.1), and decrease the initial feature weights corresponding to the topology dimension (e.g., target feature weight = initial feature weight × 0.8) and the initial feature weights of the shape descriptor dimension (e.g., target feature weight = initial feature weight × 0.7).

[0145] When the target part is a complex part type, it is necessary to increase the initial feature weights corresponding to the topology dimension (e.g., target feature weight = initial feature weight × 1.3), the initial feature weights corresponding to the shape descriptor dimension (e.g., target feature weight = initial feature weight × 1.2), and the initial feature weights corresponding to the convex hull dimension (e.g., target feature weight = initial feature weight × 1.1), and decrease the initial feature weights corresponding to the size dimension (e.g., target feature weight = initial feature weight × 0.8).

[0146] When the target part is a large part, it is necessary to increase the initial feature weights corresponding to the basic geometric dimension (e.g., target feature weight = initial feature weight × 1.2) and the initial feature weights corresponding to the moment of inertia dimension (e.g., target feature weight = initial feature weight × 1.1), and decrease the initial feature weights corresponding to the curvature dimension (e.g., target feature weight = initial feature weight × 0.7) and the initial feature weights of the symmetry dimension (e.g., target feature weight = initial feature weight × 0.8).

[0147] When the target part is a flat part, it is necessary to increase the initial feature weight of the symmetry dimension (e.g., target feature weight = initial feature weight × 1.2) and the initial feature weight of the shape descriptor dimension (e.g., target feature weight = initial feature weight × 1.1).

[0148] Step S406: Normalize the target feature weights corresponding to each dimension to obtain normalized weights.

[0149] Step S408: Normalize the feature data under each dimension using the normalized weights corresponding to each dimension, and combine the normalized feature data to obtain the feature vector.

[0150] For example, the target feature weights for each dimension are normalized so that the sum of the normalized weights for each dimension is 1. The feature data for each dimension are then weighted using the normalized weights for each dimension to obtain the normalized feature data for each dimension. Finally, the normalized feature data for each dimension are combined to obtain the feature vector.

[0151] In this embodiment, by automatically adjusting the feature weights according to the part type and normalizing the adjusted feature weights, the feature vector is obtained by combining the normalized feature data of each dimension, which can improve the accuracy of the feature vector and thus help improve the accuracy of subsequent cost prediction.

[0152] In one exemplary embodiment, such as Figure 5 As shown, step S104 may further include steps S502 to S506. Wherein:

[0153] Step S502: Using the feature vector distance between multiple historical parts, iterative clustering is performed to obtain multiple historical part clusters.

[0154] For example, the server can calculate the distance between feature vectors of multiple historical parts. The distance between feature vectors is used to measure the similarity between historical parts. The K-means clustering algorithm is used to iteratively cluster the multiple historical parts to ensure that the similarity between historical parts within a cluster is high and the similarity between historical parts between clusters is low, thus forming multiple clusters of historical parts.

[0155] Optionally, in some implementations, the number of historical part clusters (i.e., the K value of the K-means clustering algorithm) can be set to 5-20. In other implementations, the maximum number of iterations for iterative clustering can be set to 100-500 to control the convergence speed of iterative clustering. In still other implementations, K-means++ can be used to select the initial cluster centers, accelerating algorithm convergence and improving clustering quality.

[0156] Step S504: Using the distance between the feature vector of the target part and the center feature vector of each historical part cluster, determine the target part cluster corresponding to the target part from multiple historical part clusters.

[0157] Step S506: Determine the similarity between the target part and each historical part in the target part cluster by using the distance between the feature vector of the target part and the feature vector of each historical part in the target part cluster.

[0158] For example, the server can calculate the center feature vector of each historical part cluster using the feature vectors of each historical part within each cluster. The distance between the feature vector of the target part and the center feature vector of each historical part cluster is then calculated. The historical part cluster with the closest distance, or the historical part cluster with a distance less than a preset distance threshold, is selected as the target part cluster. The distance between the feature vector of the target part and the feature vector of each historical part within the target part cluster is calculated, and this distance is converted into a corresponding similarity score to obtain the similarity score between the target part and each historical part within the target part cluster.

[0159] In this embodiment, similar parts are grouped by performing cluster analysis on historical parts. The similarity between these grouped historical part clusters and the target part is calculated. Target part clusters that are moderately similar to the target part are then selected. The similarity between the target part and the historical parts within these target part clusters is then calculated. This reduces the number of similarity calculations and improves the efficiency of similarity calculation. Furthermore, it allows for the identification and differentiation of significant differences between historical parts based on historical part clusters, facilitating a more intuitive understanding of the part distribution.

[0160] In one exemplary embodiment, such as Figure 6 As shown, it may also include the following steps S602 to S610. Wherein:

[0161] Step S602: Determine the basic cost data of the target part based on the part data.

[0162] For example, the server can store standard cost calculation methods for standard parts. The standard cost calculation methods are used to process the part data to obtain the basic cost data of the target part.

[0163] Step S604: In the event that the machine learning model fails to predict the cost, determine the average similarity between the target part and the historical parts corresponding to the learning data.

[0164] The failure of a machine learning model in cost prediction may include, but is not limited to, at least one of the following: the machine learning model does not exist; the amount of learning data is less than a preset data amount threshold; or the machine learning model reports an error during the cost prediction process.

[0165] For example, when the machine learning model fails to predict costs, it returns a message indicating that the cost prediction failed. Upon receiving this message, the average similarity can be obtained by averaging the similarities between the target part and each historical part in the learning data.

[0166] Step S606: Determine the similarity weight of the historical parts corresponding to the learning data based on the similarity.

[0167] For example, the server can store the mapping relationship between the numerical values ​​of similarity and the similarity weights. Based on this mapping relationship, the similarity weight corresponding to the learning data is determined. For instance, the higher the similarity, the higher the similarity weight.

[0168] Step S608: Determine the average cost data corresponding to the learning data.

[0169] Step S610: Using average cost data, similarity weights, and basic cost data, generate target cost data for the target part.

[0170] For example, the server can calculate the average cost data using the historical cost data of each historical part corresponding to the learning data. The average cost data, similarity weights, and basic cost data are then processed to generate the target cost data for the target part.

[0171] Alternatively, in some implementations, the target cost data can be calculated with reference to the following formula:

[0172] Target cost data = average cost data × similarity weight + basic cost data × (1 - similarity weight).

[0173] In this embodiment, when the machine learning model fails to predict costs, the average similarity between the target part and the learning data is calculated, and the weight is calculated based on the similarity. The target cost data of the target part is generated using the average cost data, similarity weight, and basic cost data, which can improve the generalization of the cost data determination method.

[0174] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0175] Based on the same inventive concept, this application also provides a machine learning-based cost data determination apparatus for implementing the aforementioned machine learning-based cost data determination method. The solution provided by this apparatus is similar to the implementation described in the above-described method; therefore, the specific limitations in one or more machine learning-based cost data determination apparatus embodiments provided below can be found in the limitations of the machine learning-based cost data determination method described above, and will not be repeated here.

[0176] In one exemplary embodiment, such as Figure 7 As shown, a cost data determination device 700 based on machine learning is provided, including: a vector construction module 702, a similarity calculation module 704, and a cost prediction module 706, wherein:

[0177] The vector construction module 702 is used to obtain the part data of the target part and construct the feature vector corresponding to the target part using the part data.

[0178] The similarity calculation module 704 is used to determine the similarity between the target part and the historical parts by using the distance between the feature vector of the target part and the feature vector of the pre-stored historical parts.

[0179] The cost prediction module 706 is used to construct learning data using the feature vectors of historical parts and historical cost data, and to train a machine learning model using the learning data. When the similarity meets the preset numerical range, the machine learning model uses the feature vectors of the target part to predict the cost and obtain the target cost data of the target part.

[0180] In an exemplary embodiment, the machine learning model integrates a gradient boosting tree model and a neural network model. The cost prediction module 706 includes: a decision tree prediction unit, used to input the feature vector of the target part into the gradient boosting tree model for regression prediction to obtain first prediction data; a neural network prediction unit, used to input the feature vector of the target part into the neural network model for cost prediction to obtain second prediction data; and a data fusion unit, used to fuse the first prediction data and the second prediction data to obtain target cost data.

[0181] In an exemplary embodiment, when the part data is two-dimensional data, the vector construction module 702 is further configured to construct geometric feature parameters of the target part using geometric parameters in the part data; construct layer feature parameters of the target part using layer parameters in the part data; construct dimensional feature parameters of the target part using dimensional parameters in the part data; use the geometric feature parameters, layer feature parameters, and dimensional feature parameters as feature data of the target part in multiple dimensions; and construct feature vectors using the feature data in multiple dimensions.

[0182] In an exemplary embodiment, when the part data is three-dimensional data, the vector construction module 702 is further configured to: construct geometric feature parameters of the target part using geometric parameters in the part data; construct topological feature parameters of the target part using topological parameters in the part data; construct shape descriptor parameters of the target part using shape parameters in the part data; construct part complexity parameters of the target part using geometric parameters, topological parameters, shape parameters, and material parameters in the part data; use the part complexity parameters, geometric feature parameters, topological feature parameters, shape descriptor parameters, part production quantity, and part precision requirement parameters in the part data as feature data of the target part in multiple dimensions; and construct feature vectors using feature data in multiple dimensions.

[0183] In an exemplary embodiment, the vector construction module 702 is further configured to read the initial feature weights corresponding to each dimension; adjust the initial feature weights according to the part type of the target part to obtain the target feature weights corresponding to each dimension; normalize the target feature weights corresponding to each dimension to obtain normalized weights; normalize the feature data under each dimension using the normalized weights corresponding to each dimension, and combine the normalized feature data to obtain the feature vector of the target part.

[0184] In an exemplary embodiment, the vector construction module 702 is further configured to: increase the initial feature weights corresponding to the size dimension and the basic geometric dimension, and decrease the initial feature weights corresponding to the topology dimension and the shape descriptor dimension when the part type is a small, simple part type, to obtain target feature weights for the size dimension, the basic geometric dimension, the topology dimension, and the shape descriptor dimension; increase the initial feature weights corresponding to the topology dimension, the shape descriptor dimension, and the convex hull dimension when the part type is a complex part type, and decrease the initial feature weights corresponding to the size dimension, to obtain target feature weights for the topology dimension, the shape descriptor dimension, the convex hull dimension, and the size dimension; increase the initial feature weights corresponding to the basic geometric dimension and the moment of inertia dimension when the part type is a large part type, and decrease the initial feature weights corresponding to the curvature dimension and the symmetry dimension, to obtain target feature weights for the basic geometric dimension, the moment of inertia dimension, the curvature dimension, and the symmetry dimension; and increase the initial feature weights corresponding to the symmetry dimension and the shape descriptor dimension when the part type is a flat part type, to obtain target feature weights for the symmetry dimension and the shape descriptor dimension.

[0185] In an exemplary embodiment, the similarity calculation module 704 is further configured to iteratively cluster multiple historical part clusters using the feature vector distance between multiple historical parts; determine the target part cluster corresponding to the target part from the multiple historical part clusters using the distance between the feature vector of the target part and the central feature vector of each historical part cluster; and determine the similarity between the target part and each historical part in the target part cluster using the distance between the feature vector of the target part and the feature vector of each historical part in the target part cluster.

[0186] In an exemplary embodiment, the machine learning-based cost data determination device 700 further includes a cost calculation module, used to determine the basic cost data of the target part based on the part data; in the event that the machine learning model fails to predict the cost, determine the average similarity between the target part and the historical parts corresponding to the learning data, wherein the machine learning model fails to predict the cost in at least one of the following situations: the machine learning model does not exist, the amount of learning data is less than a preset data amount threshold, and the machine learning model reports an error during the cost prediction process; determine the similarity weight of the historical parts corresponding to the learning data based on the similarity; determine the average cost data corresponding to the learning data; and generate the target cost data of the target part using the average cost data, the similarity weight, and the basic cost data.

[0187] The modules in the aforementioned machine learning-based cost data determination device 700 can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can invoke and execute the operations corresponding to each module.

[0188] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores part data, feature vectors, historical cost data, target cost data, etc. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a machine learning-based cost data determination method.

[0189] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0190] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0191] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0192] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0193] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0194] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0195] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for determining cost data based on machine learning, characterized in that, The method includes: Obtain the part data of the target part, and use the part data to construct the feature vector corresponding to the target part; The similarity between the target part and the historical parts is determined by using the distance between the feature vector of the target part and the feature vector of the pre-stored historical parts. Learning data is constructed using the feature vectors of the historical parts and historical cost data, and a machine learning model is trained using the learning data. If the similarity meets the preset numerical range, the machine learning model is used to predict the cost of the target part using its feature vector to obtain the target cost data of the target part.

2. The method according to claim 1, characterized in that, The machine learning model integrates a gradient boosting tree model and a neural network model; the step of using the feature vector of the target part to predict the cost through the machine learning model, and obtaining the target cost data of the target part, includes: The feature vector of the target part is input into the gradient boosting tree model for regression prediction to obtain the first prediction data; The feature vector of the target part is input into the neural network model to predict the cost, and a second prediction data is obtained. The target cost data is obtained by combining the first prediction data and the second prediction data.

3. The method according to claim 1, characterized in that, When the part data is two-dimensional data, constructing the feature vector corresponding to the target part using the part data includes: Using the geometric parameters in the part data, construct the geometric feature parameters of the target part; Using the layer parameters in the part data, construct the layer feature parameters of the target part; Using the dimensional parameters in the part data, construct the dimensional feature parameters of the target part; The geometric feature parameters, the layer feature parameters, and the size feature parameters are used as feature data of the target part in multiple dimensions. The feature vector is constructed using the feature data across the multiple dimensions.

4. The method according to claim 1, characterized in that, When the part data is three-dimensional data, the step of acquiring the part data of the target part and constructing the feature vector corresponding to the target part using the part data includes: Using the geometric parameters in the part data, construct the geometric feature parameters of the target part; Using the topology parameters in the part data, construct the topology feature parameters of the target part; Using the shape parameters in the part data, construct the shape descriptor parameters of the target part; Using the geometric parameters, topological parameters, shape parameters, and material parameters in the part data, the part complexity parameters of the target part are constructed. The part complexity parameter, the geometric feature parameter, the topological feature parameter, the shape descriptor parameter, and the part production quantity and part precision requirement parameter in the part data are used as the feature data of the target part in multiple dimensions. The feature vector is constructed using the feature data across the multiple dimensions.

5. The method according to claim 3 or 4, characterized in that, The step of constructing the feature vector using feature data from the multiple dimensions includes: Read the initial feature weights for each dimension respectively; Based on the part type of the target part, the initial feature weights are adjusted to obtain the target feature weights corresponding to each dimension; The target feature weights corresponding to each dimension are normalized to obtain normalized weights; The feature data under each dimension are normalized using the normalized weights corresponding to each dimension, and the normalized feature data are combined to obtain the feature vector of the target part.

6. The method according to claim 5, characterized in that, The step of adjusting the initial feature weights according to the part type of the target part to obtain the target feature weights corresponding to each dimension includes: When the part type is a small and simple part type, the initial feature weights corresponding to the size dimension and the basic geometric dimension are increased, and the initial feature weights of the topology dimension and the shape descriptor dimension are decreased, so as to obtain the target feature weights of the size dimension, the basic geometric dimension, the topology dimension and the shape descriptor dimension. When the part type is a complex part type, the initial feature weights of the topology dimension, the shape descriptor dimension, and the convex hull dimension are increased, and the initial feature weights corresponding to the size dimension are decreased, so as to obtain the target feature weights of the topology dimension, the shape descriptor dimension, the convex hull dimension, and the size dimension. When the part type is a large part type, the initial feature weights corresponding to the basic geometric dimension and the moment of inertia dimension are increased, and the initial feature weights of the curvature dimension and the symmetry dimension are decreased to obtain the target feature weights of the basic geometric dimension, the moment of inertia dimension, the curvature dimension and the symmetry dimension. When the part type is a flat part type, the initial feature weights of the symmetry dimension and the shape descriptor dimension are increased to obtain the target feature weights of the symmetry dimension and the shape descriptor dimension.

7. The method according to claim 1, characterized in that, The step of determining the similarity between the target part and the historical parts by utilizing the distance between the feature vector of the target part and the feature vector of pre-stored historical parts includes: Multiple historical part clusters are obtained by iteratively clustering using the feature vector distances between the various historical parts. By using the distances between the feature vector of the target part and the center feature vector of each of the historical part clusters, the target part cluster corresponding to the target part is determined from the multiple historical part clusters; The similarity between the target part and each historical part in the target part cluster is determined by using the distance between the feature vector of the target part and the feature vector of each historical part in the target part cluster.

8. The method according to claim 1, characterized in that, The method further includes: Determine the basic cost data of the target part based on the part data; In the event that the machine learning model fails to predict the cost, the average similarity between the target part and the historical parts corresponding to the learning data is determined. The failure of the machine learning model to predict the cost includes at least one of the following: the machine learning model does not exist; the amount of the learning data is less than a preset data amount threshold; and the machine learning model reports an error during the cost prediction process. The similarity weights of the historical parts corresponding to the learning data are determined based on the similarity. Determine the average cost data corresponding to the learning data; The target cost data for the target part is generated using the average cost data, the similarity weight, and the basic cost data.

9. A cost data determination device based on machine learning, characterized in that, The device includes: The vector construction module is used to acquire the part data of the target part and construct the feature vector corresponding to the target part using the part data; The similarity calculation module is used to determine the similarity between the target part and the historical parts by using the distance between the feature vector of the target part and the feature vector of the pre-stored historical parts; The cost prediction module is used to construct learning data using the feature vectors of the historical parts and historical cost data, and to train a machine learning model using the learning data; when the similarity meets a preset numerical range, the machine learning model is used to perform cost prediction using the feature vectors of the target part to obtain the target cost data of the target part.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images