An automobile parts sales forecasting method and system

Abstract

This invention discloses a method and system for predicting sales of automotive parts. This invention not only relies on historical sales data but also incorporates market environment, user behavior, and external economic factors into the prediction model, enhancing its sensitivity to demand fluctuations. By normalizing these data, the dimensional differences between different data sources are eliminated, ensuring the prediction model remains efficient and accurate even with multidimensional and complex data. A graph structure model is used to model the dependencies between parts, capturing historical data, correlations, and the influence of external factors on different parts to more accurately predict future sales demand, thereby effectively improving the efficiency of enterprise production and supply chain management. The graph structure and node characteristics are dynamically updated based on new data and environmental changes, maintaining the real-time nature and forward-looking nature of the prediction model.

Description

Technical Field

[0001] This invention relates to the field of automotive parts sales technology, and in particular to a method and system for predicting automotive parts sales. Background Technology

[0002] In the current automotive industry, supply chain management of auto parts has become a crucial factor affecting production efficiency and sales performance. Due to the volatility of market demand, changes in product lifecycles, and uncertainties in the external environment (such as the economy, climate, and policies), accurately predicting the sales volume of auto parts is essential for optimizing production scheduling, reducing inventory costs, and improving supply chain responsiveness. However, traditional sales forecasting methods, such as time series analysis, linear regression models, and simple machine learning models, often fail to fully capture the complex relationships between auto parts and their impact on overall demand. These methods largely rely on historical data and simple regression models, lacking a comprehensive consideration of multi-dimensional factors, especially the dependencies between different auto parts.

[0003] The shortcomings of existing technologies are mainly reflected in:

[0004] Single data dimension: Traditional forecasting methods rely solely on historical sales data, ignoring other factors such as market trends, usage environment, and user behavior. This makes forecast results susceptible to interference from changes in a single data point, making it difficult to accurately reflect fluctuations in future demand.

[0005] Ignoring interrelationships: There are complex commonalities and dependencies among spare parts. Existing forecasting models have failed to effectively capture and quantify these interrelationships, resulting in low forecasting accuracy and an inability to accurately adjust production plans and inventory.

[0006] Unable to dynamically adapt to market changes: The market environment and component technology are updated frequently, and traditional methods cannot update and adapt to new data and changes in demand in a timely manner, resulting in a lag in the prediction of future demand.

[0007] To address the aforementioned issues, this invention provides a method for predicting automotive parts sales based on Graph Neural Network (GNN). By using a graph structure model to model the dependencies between parts, and by capturing historical data, correlations, and the influence of external factors on different parts, it can more accurately predict future sales demand, thereby effectively improving the efficiency of enterprise production and supply chain management. Summary of the Invention

[0008] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0009] Therefore, to solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for predicting the sales volume of automotive parts, comprising the following specific steps:

[0010] Step 1: Collect historical sales data, market data, usage environment data, and user behavior data for auto parts;

[0011] Step 2: Clean the collected data, remove outliers, and standardize the data;

[0012] Step 3: Construct a graph data structure, with each type of auto parts as a node, and the edges between nodes representing the relationships between the parts. The weight of the edges is determined based on the historical sharing relationships and usage frequency of the parts.

[0013] Step 4: Train the constructed graph structure using the Graph Attention Network (GAT) model in graph neural networks to capture the interdependencies between parts;

[0014] Step 5: Use the trained GAT model to predict future sales of auto parts;

[0015] Step 6: Output the forecast results and adjust the production plan and inventory strategy based on the forecast results.

[0016] As a preferred embodiment of the automotive parts sales forecasting method of the present invention, in step one, market data such as oil prices, economic growth rates, and climate change, environmental data including climate conditions, geographical location, and road conditions, and user behavior data including vehicle usage frequency and maintenance records are included.

[0017] As a preferred embodiment of the method for predicting the sales volume of auto parts described in this invention, step one further includes collecting data on external influencing factors related to auto parts and normalizing these data.

[0018] As a preferred embodiment of the automotive parts sales forecasting method of the present invention, in the graph data structure constructed in step three, the node features include historical sales data, market trends and environmental influencing factors, and the edge features include edge weights and time intervals.

[0019] As a preferred embodiment of the automotive parts sales forecasting method of the present invention, in step four, an adaptive attention mechanism is used to give higher weights to edges with stronger relevance, thereby improving the prediction accuracy of the model.

[0020] As a preferred embodiment of the automotive parts sales forecasting method of the present invention, in step five, by capturing the complex relationship between parts i and other parts i, as well as external influencing factors, its sales at future moments are calculated.

[0021] Define Y j (τ) represents the sales data of auto parts (i.e., node feature data), A ij R represents the strength of the association between parts (i.e., the weight of the edge). k (t) represents external factors that influence the sales volume of auto parts; the predicted sales volume of a certain auto part i at time t is S. i (t), then its calculation formula is as follows:

[0022]

[0023] In this formula, Integrating over the time range [0, T] is used to consider the cumulative impact of changes in spare parts sales over time; the upper and lower limits of the integration can be determined based on the time range of historical data.

[0024] Summing is performed on other parts j that are related to part i, where N is the number of parts related to part i; this term captures the correlation between different parts.

[0025] A ij : Indicates the strength of the association between parts i and j, determined based on the number of times they were replaced simultaneously in history or the frequency of use;

[0026] e -λ(t-τ) : Describes the decay of influence over time, where λ is the decay coefficient, reflecting the diminishing impact of past events on current sales.

[0027] Used for influencing factor X j (τ) (such as market trends, usage environment, or user behavior) is normalized, β is the adjustment parameter, and μ is the normalization parameter. j This is for mean offset, controlling the steepness of the normalization curve;

[0028] Y j (τ) represents the characteristic value of spare part j at time t (such as sales volume or loss rate), which can be obtained through historical data;

[0029] To adjust the forecast results, considering multiple external factors R k (t), such as climate change; here W k For the weight, f k It describes R k Nonlinear functions of (t) (such as sin or cos, while h) k (u) is a time-dependent adjustment function.

[0030] As a preferred embodiment of the automotive parts sales forecasting method of the present invention, step five further includes periodically updating the graph structure and node features to reflect changes in new parts and the market environment, and to maintain the real-time performance of the forecasting model.

[0031] In a preferred embodiment of the automotive parts sales forecasting method of the present invention, step six outputs the sales forecast results of the parts to form a forecast report, including the predicted value of future sales.

[0032] In a preferred embodiment of the automotive parts sales forecasting method of the present invention, in step six, if the sales forecast shows an increase in demand, market promotion can be carried out in advance to increase market share; if the sales forecast shows a decrease in demand, promotional activities can be formulated or prices can be adjusted.

[0033] The present invention also provides a sales forecasting system for automotive parts, characterized in that: the system includes:

[0034] The data collection module is used to collect historical sales data, market data, usage environment data, and user behavior data for automotive parts.

[0035] The data processing module is used to clean and standardize the collected data;

[0036] The graph structure construction module is used to construct a graph data structure for automotive parts and their relationships, where nodes represent parts and edges represent relationships.

[0037] The graph neural network module uses a graph attention network (GAT) model to train graph data and capture the interdependencies between nodes.

[0038] The prediction module is used to predict future sales of automotive parts using a trained GAT model.

[0039] The output module is used to output the forecast results and generate a forecast report, which includes the predicted value of future sales, the confidence interval, and the fluctuation trend.

[0040] The above modules work together to achieve the entire process of automotive parts sales forecasting, provide accurate sales forecast results, and provide a basis for production and inventory management.

[0041] The beneficial effects of this invention are:

[0042] 1. This invention not only relies on historical sales data, but also incorporates market environment, user behavior, and external economic factors into the predictive model, enhancing its sensitivity to demand fluctuations. By normalizing this data, the dimensional differences between different data sources are eliminated, ensuring the predictive model remains efficient and accurate even when dealing with complex multidimensional data.

[0043] 2. Traditional prediction models are mostly based on linear or nonlinear regression analysis. This invention, however, introduces a graph structure, treating components as nodes in the graph. The dependencies between different components are represented by edges, each with a weight. This method effectively captures the complex interdependencies between components, solving the problem of traditional methods failing to identify these relationships, thereby improving prediction accuracy.

[0044] 3. This invention employs a Graph Attention Network (GAT) for model training. Compared to existing machine learning models based on time series or feature extraction, GAT can adaptively learn the correlations between different components, assigning higher weights to more important correlations, thereby enhancing the accuracy of sales forecasting. The multi-layered attention mechanism of the GAT model enables it to quickly adjust its prediction strategy when faced with various complex factors, achieving a dynamic response to market changes.

[0045] 4. This invention utilizes a GNN model to capture the dynamic dependencies between different nodes, enabling more accurate prediction of future demand fluctuations for spare parts. It dynamically updates the graph structure and node features based on new data and environmental changes, thus maintaining the real-time performance and forward-looking nature of the prediction model. Traditional static prediction models cannot cope with rapidly changing market environments, while the dynamic update mechanism of this invention ensures that the prediction results continuously track market trends, offering greater flexibility and accuracy. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0047] Figure 1 This is a flowchart of the process of the present invention.

[0048] Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0050] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0051] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0052] Example 1

[0053] Reference Figures 1-2 The first embodiment of the present invention provides a method for predicting the sales volume of automotive parts, including the following specific steps:

[0054] Step 1: Collect historical sales data, market data (such as oil prices, economic growth rate, climate change, etc.), usage environment data (such as climate conditions, geographical location, road conditions, etc.), user behavior data (such as vehicle usage frequency, maintenance records, etc.), and data on external influencing factors related to auto parts, and normalize these data.

[0055] Step 2: Clean the collected data, remove outliers, and standardize the data;

[0056] Step 3: Construct a graph data structure, with each type of auto parts as a node. The edges between nodes represent the relationships between the parts, and the weights of the edges are determined based on the historical sharing relationships and usage frequency of the parts. Node features include historical sales data, market trends, and environmental influencing factors, while edge features include edge weights and time intervals.

[0057] Step 4: Train the constructed graph structure using the Graph Attention Network (GAT) model in graph neural networks to capture the interdependencies between parts; give higher weights to edges with stronger correlations through an adaptive attention mechanism, thereby improving the prediction accuracy of the model.

[0058] Step 5: Use the trained GAT model to predict future sales of auto parts; by capturing the complex relationships between part i and other parts, as well as external influencing factors, calculate its sales at future moments, and regularly update the graph structure and node features to reflect changes in new parts and the market environment, and maintain the real-time performance of the prediction model.

[0059] Define Y j (τ) represents the sales data of auto parts (i.e., node feature data), A ij R represents the strength of the association between parts (i.e., the weight of the edge). k (t) represents external factors that influence the sales volume of auto parts; the predicted sales volume of a certain auto part i at time t is S. i (t), then its calculation formula is as follows:

[0060]

[0061] In this formula, Integrating over the time range [0, T] is used to consider the cumulative impact of changes in spare parts sales over time; the upper and lower limits of the integration can be determined based on the time range of historical data.

[0062] Summing is performed on other parts j that are related to part i, where N is the number of parts related to part i; this term captures the correlation between different parts.

[0063] A ij : Indicates the strength of the association between parts i and j, determined based on the number of times they were replaced simultaneously in history or the frequency of use;

[0064] e -λ(t-τ) : Describes the decay of influence over time, where λ is the decay coefficient, reflecting the diminishing impact of past events on current sales.

[0065] Used for influencing factor X j (τ) (such as market trends, usage environment, or user behavior) is normalized, β is the adjustment parameter, and μ is the normalization parameter. j This is for mean offset, controlling the steepness of the normalization curve;

[0066] Y j (τ) represents the characteristic value of spare part j at time t (such as sales volume or loss rate), which can be obtained through historical data;

[0067] To adjust the forecast results, considering multiple external factors R k (t), such as climate change; here W k For the weight, f kIt describes R k Nonlinear functions of (t) (such as sin or cos, while h) k (u) is a time-dependent adjustment function.

[0068] Step Six: Output the forecast results and adjust the production plan and inventory strategy based on the forecast results; output the sales forecast results of spare parts to form a forecast report, including the predicted value of future sales; in Step Six, if the sales forecast shows that demand is rising, market promotion can be carried out in advance to increase market share; if the sales forecast shows that demand is falling, promotional activities can be formulated or prices can be adjusted.

[0069] The present invention also provides a sales forecasting system for automotive parts, characterized in that: the system includes:

[0070] The data collection module is used to collect historical sales data, market data, usage environment data, and user behavior data for automotive parts.

[0071] The data processing module is used to clean and standardize the collected data;

[0072] The graph structure construction module is used to construct a graph data structure for automotive parts and their relationships, where nodes represent parts and edges represent relationships.

[0073] The graph neural network module uses a graph attention network (GAT) model to train graph data and capture the interdependencies between nodes.

[0074] The prediction module is used to predict future sales of automotive parts using a trained GAT model.

[0075] The output module is used to output the forecast results and generate a forecast report, which includes the predicted value of future sales, the confidence interval, and the fluctuation trend.

[0076] The above modules work together to achieve the entire process of automotive parts sales forecasting, provide accurate sales forecast results, and provide a basis for production and inventory management.

[0077] In summary, the method of this invention not only solves the problems of single data dimension, neglect of correlation, and insufficient response to market changes in existing technologies, but also significantly improves the accuracy of automotive parts sales forecasting through innovative graph structure modeling and graph neural network technology. Compared with existing technologies, this invention achieves technological innovation through the organic combination of deep learning and graph modeling techniques, and has high market application prospects and academic value.

[0078] Example 2

[0079] Reference Figure 2This is the second embodiment of the present invention, which differs from the first embodiment in that: this embodiment aims to verify the effectiveness of using a graph neural network (GNN) model for predicting the sales volume of automotive parts through experiments; the specific experimental steps are as follows:

[0080] I. Data Collection:

[0081] We collect sales data for spare parts over the past 12 months, including monthly sales of brake pads, brake discs, tires, and axles.

[0082] Collect market data related to spare parts, such as influencing factors like oil prices, economic growth rates, and climate change.

[0083] II. Data Preprocessing:

[0084] The sales data is cleaned to remove outliers and then standardized to fall within the range of 0 to 1.

[0085] Establish relationships between components and define relationship weights. For example, the relationship strength between brake pads and brake discs is set to 0.8, and the relationship strength between brake pads and tires is set to 0.3.

[0086] III. Graph Data Construction:

[0087] Each component is treated as a node, and the edges between nodes represent the relationships between different components.

[0088] Using the above data, a graph structure is constructed, defining node features as historical sales volume, market trends, etc., and edge features as association weight and time interval.

[0089] IV. Model Design and Training:

[0090] The model is trained using Graph Attention Networks (GAT) from the Graph Neural Networks framework, and appropriate hyperparameters (such as learning rate, number of layers, etc.) are selected.

[0091] Historical sales data is used for training, and model parameters are optimized to minimize prediction error.

[0092] V. Sales Forecast:

[0093] The trained model is used to predict sales for the next three months. By inputting the latest node features and graph structure, the predicted sales for each component are calculated.

[0094] VI. Results Evaluation and Comparison:

[0095] The predicted results are compared with actual sales to evaluate the model's prediction accuracy, and compared with traditional sales forecasting methods.

[0096] During implementation, special attention was paid to the model's ability to capture complex relationships between components. Through the model's adaptive characteristics, multi-level features can be effectively extracted, improving the accuracy of sales forecasting.

[0097] The specific experimental data are as follows:

[0098]

[0099] Analysis of the data in the table above shows that the sales forecasting results using the graph neural network model in this embodiment have a prediction error of 5%-10% compared to actual sales, while the error of traditional methods is typically between 15%-25%. This indicates that the model can more accurately capture the complex relationships between parts, thereby significantly improving the accuracy of sales forecasting.

[0100] Specifically, as shown in the table, the gap between actual and predicted sales of brake pads is decreasing month by month. For example, the actual sales in January were 200, while the predicted sales were 195, with an error of only 2.5%. In contrast, the error of traditional methods could reach 20 units or more in the same period, and this advantage continues in subsequent months.

[0101] The model successfully reflected the impact of market fluctuations. For example, in May, as the sales peak approached, sales increased from 240 to 290. The model also captured this, predicting a value of 285, demonstrating its sensitivity to market trends.

[0102] The model excels at capturing the correlation between parts; a clear interrelationship exists between the sales volumes of brake pads and brake discs. This trend is reflected in the table, and the model accurately captures the interdependence of their sales during peak periods.

[0103] The model has the ability to dynamically adjust according to the market environment. For example, when climate change affects sales, the model can quickly adapt and make reasonable predictions, which is reflected in the tabular data.

[0104] In summary, the sales forecasting method in this embodiment has significant advantages over traditional methods in terms of accuracy, market trend responsiveness, and dynamic adjustment capabilities, and provides a scientific basis for subsequent production and sales decisions.

[0105] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting sales of auto parts, characterized in that: The specific steps include the following: Step 1: Collect historical sales data, market data, usage environment data, and user behavior data for auto parts; Step 2: Clean the collected data, remove outliers, and standardize the data; Step 3: Construct a graph data structure, with each type of auto parts as a node, and the edges between nodes representing the relationships between the parts. The weight of the edges is determined based on the historical sharing relationships and usage frequency of the parts. Step 4: Use the GAT model in graph neural networks to train the constructed graph structure and capture the interdependencies between parts; Step 5: Use the trained GAT model to predict future sales of auto parts; Step Six: Output the forecast results and adjust the production plan and inventory strategy based on the forecast results; In step five, by capturing the complex relationships between component i and other components, as well as external influencing factors, its sales volume at future moments is calculated; definition This represents sales data for auto parts. Indicates the strength of the connection between parts. Data representing external factors that influence the sales of auto parts; Predict the sales volume of a certain spare part i at time t. The calculation formula is as follows: In this formula, Integrating over a time range [0, T] is used to consider the cumulative impact of changes in spare parts sales over time; the upper and lower limits of the integration are determined based on the time range of historical data. : Sum the other parts j that are related to part i, where N is the number of parts related to part i; This captures the interrelationships between different components; : Indicates the strength of the association between parts i and j, determined based on the number of times they were replaced simultaneously in history or the frequency of use; : Describes the decay of the effect over time. This is the decay coefficient, reflecting the diminishing impact of past events on current sales. Used for influencing factors Normalization is performed. To adjust the parameters, This is for mean offset, controlling the steepness of the normalization curve; This represents the characteristic value of component j at time t, obtained through historical data; Used to adjust forecast results, taking into account multiple external factors. ; As weight, It is a description The nonlinear function, and This is a time-dependent adjustment function.

2. The method for predicting sales of auto parts as described in claim 1, characterized in that: In step one, market data includes oil prices, economic growth rate, and climate change; environmental data includes climate conditions, geographical location, and road conditions; and user behavior data includes vehicle usage frequency and maintenance records.

3. The method for predicting sales of auto parts as described in claim 1, characterized in that: Step one also includes collecting data on external influencing factors related to automotive parts and normalizing that data.

4. The method for predicting sales of auto parts as described in claim 1, characterized in that: In the graph data structure constructed in step three, node features include historical sales data, market trends, and environmental influencing factors, while edge features include edge weights and time intervals.

5. The method for predicting sales of auto parts as described in claim 1, characterized in that: In step four, an adaptive attention mechanism is used to give higher weights to edges with stronger relevance, thereby improving the model's prediction accuracy.

6. The method for predicting sales of auto parts as described in claim 1, characterized in that: Step five also includes periodically updating the graph structure and node characteristics to reflect changes in new parts and the market environment, keeping the predictive model real-time.

7. The method for predicting sales of auto parts as described in claim 1, characterized in that: Step six outputs the sales forecast results for spare parts, forming a forecast report that includes the predicted value of future sales.

8. The method for predicting sales of auto parts as described in claim 1, characterized in that: In step six, if sales forecasts indicate rising demand, marketing efforts can be initiated in advance to increase market share; if sales forecasts indicate declining demand, promotional activities can be developed or prices adjusted.

9. A forecasting system using the sales forecasting method for automotive parts according to any one of claims 1 to 8, characterized in that: The system includes: The data collection module is used to collect historical sales data, market data, usage environment data, and user behavior data for automotive parts. The data processing module is used to clean and standardize the collected data; The graph structure construction module is used to construct a graph data structure for automotive parts and their relationships, where nodes represent parts and edges represent relationships. The graph neural network module uses a graph attention network model to train graph data and capture the interdependencies between nodes. The prediction module is used to predict future sales of automotive parts using a trained GAT model. The output module is used to output the forecast results and generate a forecast report, which includes the predicted value of future sales, the confidence interval, and the fluctuation trend. The above modules work together to achieve the entire process of automotive parts sales forecasting, provide accurate sales forecast results, and provide a basis for production and inventory management.

Images