Retail industry inventory management and demand prediction system

Abstract

The invention relates to a retail industry inventory management and demand prediction system, and the system comprises a data collection subsystem which is used for collecting the multi-dimensional data of the retail industry; the data preprocessing subsystem is used for carrying out cleaning, standardization and feature engineering on the collected data; the model training subsystem is used for generating a commodity demand prediction model by adopting big data and through a machine learning algorithm; the demand prediction subsystem is used for obtaining a prediction result based on the commodity demand prediction model; and the inventory optimization subsystem is used for dynamically adjusting the inventory level according to the prediction result so as to realize inventory optimization. According to the invention, by integrating the big data technology and the machine learning algorithm, automation and intelligence of inventory management and demand prediction are realized, and the management efficiency and the prediction accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the field of big data analysis and prediction, and in particular to a retail industry inventory management and demand prediction system. Background Art

[0002] In the retail industry, inventory management and demand forecasting are two core links, which are directly related to the operational efficiency and cost control of the enterprise. The existing inventory management and demand forecasting technologies mainly include the following:

[0003] 1. Experience-based inventory management: Many retailers rely on experienced managers to judge and adjust inventory levels. This approach lacks data support and is easily affected by subjective judgment.

[0004] 2. Simple statistical forecasting methods: Some companies use simple statistical methods such as moving average or exponential smoothing to forecast demand, but these methods often fail to accurately capture the complexity of market dynamics and consumer behavior.

[0005] 3. ERP systems: Many retail companies use enterprise resource planning (ERP) systems to manage inventory, but these systems typically focus on process automation rather than forecasting and optimization based on advanced data analytics.

[0006] 4. Supply chain management software: Some supply chain management software provide inventory optimization tools, but they usually require users to manually enter parameters and adjust settings, and lack automation and intelligence.

[0007] Although existing technologies support inventory management and demand forecasting in the retail industry to some extent, they have the following limitations and shortcomings:

[0008] 1. Reliance on subjective judgment: Experience-based management methods rely too much on personal experience and lack objective data support, which can easily lead to wrong decisions.

[0009] 2. Lack of flexibility: Simple statistical methods cannot adapt to rapid changes in the market and cannot respond promptly to fluctuations in consumer demand.

[0010] 3. Insufficient data utilization: Existing technologies fail to fully utilize the advantages of big data and ignore the possibility of extracting useful information from multi-dimensional data for in-depth analysis.

[0011] 4. Limited prediction accuracy: Due to the lack of advanced algorithm support, existing technologies have limitations in prediction accuracy and cannot accurately predict future market demand.

[0012] 5. Low degree of automation: Many systems and software require manual intervention, and the degree of automation is not high, resulting in inefficiency and waste of human resources.

[0013] 6. Lack of real-time performance: Existing systems often fail to provide real-time inventory and demand data, making it difficult for companies to respond quickly.

[0014] The information disclosed in this background technology section is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0015] In view of the defects existing in the prior art, the purpose of the present invention is to provide a retail industry inventory management and demand forecasting system, which realizes the automation and intelligence of inventory management and demand forecasting by integrating big data technology and machine learning algorithms, thereby improving management efficiency and forecasting accuracy.

[0016] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0017] A retail industry inventory management and demand forecasting system, characterized by comprising:

[0018] Data collection subsystem, used to collect multi-dimensional data of the retail industry;

[0019] Data preprocessing subsystem, used to clean, standardize and feature engineer the collected data;

[0020] A model training subsystem, which is used to use big data and generate a commodity demand forecasting model through machine learning algorithms;

[0021] Demand forecasting subsystem, used to obtain forecasting results based on commodity demand forecasting model;

[0022] The inventory optimization subsystem is used to dynamically adjust inventory levels based on forecast results to achieve inventory optimization.

[0023] Based on the above technical solution, the data collection subsystem connects with the following systems to extract data, and then stores it in the big data platform to form a data warehouse:

[0024] POS system, including information related to transaction processing and fee settlement;

[0025] Ordering systems, including information related to order receipt and order fulfillment;

[0026] Supply chain systems, including logistics and inventory-related information;

[0027] Cloud commerce system, including information related to online sales and promotions;

[0028] Social systems, including information about customer interactions and market trends.

[0029] Based on the above technical solution, the source data in the data warehouse is parsed and labeled according to the following dimensions to form multi-dimensional data:

[0030] Historical sales data;

[0031] Inventory level data;

[0032] Promotional activity data;

[0033] Holiday data;

[0034] Seasonal data;

[0035] Economic indicator data;

[0036] Market trend data.

[0037] On the basis of the above technical solution, the collected data is cleaned, standardized and feature engineered, wherein the data cleaning uses a distributed architecture to parallelly process the data cleaning tasks, specifically including:

[0038] Analyze missing values ​​and fill or delete them based on actual conditions;

[0039] Analyze outliers and mark or remove them;

[0040] Data deduplication processing;

[0041] The data standardization specifically includes:

[0042] Convert all date data to a uniform format;

[0043] Convert numerical data to a consistent scale for comparison and analysis;

[0044] The feature engineering of data includes:

[0045] Feature extraction, extracting useful information from raw data as features;

[0046] Feature selection: select the features that have the greatest impact on the model’s prediction performance from the existing features to reduce the model’s complexity and the risk of overfitting;

[0047] Feature construction: creating feature sets based on raw data to improve model performance; methods of feature construction include but are not limited to:

[0048] Feature combination: combine two or more original features to form a new feature;

[0049] Feature derivation: new features are calculated based on the original features;

[0050] Feature aggregation: Group or segment the original data, and then perform aggregation operations on each group or segment to form new features;

[0051] Convert text data into numerical features;

[0052] Automatic extraction of image features based on deep learning model is performed through convolutional neural network (CNN).

[0053] On the basis of the above technical solutions, in the feature construction stage, dimensionality reduction processing is performed based on the historical feature construction results to reduce the number of features to reduce the data dimension, while retaining important information in the original data as much as possible.

[0054] Based on the above technical solution, the model training subsystem obtains the characterization data from the data preprocessing subsystem, determines the model architecture according to the characterization data, and performs training;

[0055] The training results are evaluated through the model evaluation module. If the performance does not meet the standard, the model parameters are adjusted and retrained, or the model architecture is changed and retrained.

[0056] The model architecture includes: random forest, gradient boosting machine GBM or neural network;

[0057] The performance evaluation was performed using cross validation to estimate model performance.

[0058] Based on the above technical solution, the model training subsystem performs multi-point intervention in the training process according to user instructions, and the intervention is any one of the following:

[0059] Adjust tunable parameters in the data preprocessing subsystem, specify model architecture, and optimize model parameters.

[0060] On the basis of the above technical solution, the demand forecasting subsystem predicts future demand according to a preset time span. When forecasting, the corresponding forecasting results are presented by selecting reference factors. The reference factors include: historical sales trends, seasonal changes, market activities, and economic indicators.

[0061] On the basis of the above technical solution, the inventory optimization subsystem presets several inventory strategies, adapts each inventory strategy and generates an optimal inventory level configuration report based on the forecast results of the demand forecasting subsystem, to assist enterprises in formulating procurement and sales plans.

[0062] Based on the above technical solution, the inventory optimization subsystem obtains data from the data collection subsystem and compares it with the optimal inventory level configuration report to determine the specific inventory execution status and generate corresponding inventory strategies based on the specific execution status for auditing and inventory strategy adjustment.

[0063] The retail industry inventory management and demand forecasting system described in the present invention has the following features:

[0064] Beneficial effects:

[0065] 1. Through a unified data interface and distributed cleaning engine, the system can efficiently process and analyze data from multiple sources such as supply chain, social, ordering, cloud commerce and POS systems, ensuring data consistency and availability;

[0066] The system can process and analyze larger data sets, fully exploit the advantages of big data, and tap into the potential value in the data;

[0067] Through big data and machine learning algorithms, the system can automatically process and analyze multi-dimensional data, reduce interference from human factors, and improve the intelligence level of inventory management;

[0068] 2. Through feature engineering and data set construction, the system can build an accurate demand forecasting model, thereby optimizing inventory strategies, reducing inventory backlogs and out-of-stock risks, and ultimately improving the company's operational efficiency and market responsiveness;

[0069] Through data preprocessing and feature engineering, the system can improve data quality and provide more accurate data support for model training;

[0070] The system can provide inventory and demand data in real time, enabling enterprises to respond quickly to market changes and reduce inventory backlogs and out-of-stock risks;

[0071] The system can dynamically adjust inventory strategies based on forecast results, achieve optimal inventory level configuration, reduce inventory costs, flexibly adjust inventory strategies, and improve the market competitiveness of enterprises;

[0072] Based on accurate forecasting results, the system can provide scientific decision-making support for enterprises and help them formulate more reasonable procurement and sales plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] The present invention has the following accompanying drawings:

[0074] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0075] Figure 1 System architecture diagram of embodiment 1 of a retail industry inventory management and demand forecasting system described in the present invention. DETAILED DESCRIPTION

[0076] The present invention is further described in detail below in conjunction with the accompanying drawings. The detailed description is an explanation made in conjunction with the exemplary embodiments of the present invention, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0077] like Figure 1 As shown, the retail industry inventory management and demand forecasting system described in the present invention includes:

[0078] The data collection subsystem is used to collect multi-dimensional data of the retail industry. The data collection subsystem connects with other systems through interfaces to collect multi-dimensional data of the retail industry.

[0079] Data preprocessing subsystem, used to clean, standardize and feature engineer the collected data;

[0080] A model training subsystem, which is used to use big data and generate a commodity demand forecasting model through machine learning algorithms;

[0081] Demand forecasting subsystem, used to obtain forecasting results based on commodity demand forecasting model;

[0082] The inventory optimization subsystem is used to dynamically adjust inventory levels based on forecast results to achieve inventory optimization.

[0083] This embodiment improves the accuracy of demand forecasting through the collection of multi-dimensional data and feature engineering. The model can be adjusted according to actual conditions, which improves the adaptability of the system. Assisted decision-making can dynamically adjust inventory levels and reduce inventory backlogs and out-of-stock risks.

[0084] Based on the above technical solution, the data collection subsystem connects with the following systems to extract data, and then stores it in the big data platform to form a data warehouse:

[0085] POS system, including information related to transaction processing and fee settlement;

[0086] Ordering systems, including information related to order receipt and order fulfillment;

[0087] Supply chain systems, including logistics and inventory-related information;

[0088] Cloud commerce system, including information related to online sales and promotions;

[0089] Social systems, including information about customer interactions and market trends.

[0090] When extracting data, the data is classified according to importance, where:

[0091] Key data is actively submitted by each system to the data collection subsystem, and only each system can decide the data content and whether the data needs to be updated;

[0092] Common data is collected regularly by the data collection subsystem. When each system updates the common data, it initiates a data synchronization request to the data collection subsystem. After the request is approved, only the changed common data is submitted.

[0093] Privacy data adopts a real-time subscription mechanism, and each system and data collection subsystem subscribes to each other and updates synchronously;

[0094] Fixed data uses an authorized access mechanism and is only saved by the system where the data was originally generated. Other systems and data collection subsystems can temporarily copy and set a destruction deadline after authentication.

[0095] Based on the above technical solution, the source data in the data warehouse is parsed and labeled according to the following dimensions to form multi-dimensional data:

[0096] Historical sales data;

[0097] Inventory level data;

[0098] Promotional activity data;

[0099] Holiday data;

[0100] Seasonal data;

[0101] Economic indicator data;

[0102] Market trend data.

[0103] Going a step further, geographical information can be parsed and inventory levels can be divided according to the communities they belong to.

[0104] On the basis of the above technical solution, the collected data is cleaned, standardized and feature engineered, wherein the data cleaning uses a distributed architecture to parallelly process the data cleaning tasks, specifically including:

[0105] Analyze missing values ​​and fill or delete them according to the actual situation; for example, fill them with mean, median, mode, or fill them according to business logic;

[0106] Analyze outliers and mark or remove them;

[0107] Data deduplication processing;

[0108] The data standardization specifically includes:

[0109] Convert all date data to a uniform format;

[0110] Convert numerical data to a uniform scale for comparison and analysis; select appropriate standardization methods based on the distribution characteristics of the data, such as minimum-maximum standardization, Z-score standardization, etc.;

[0111] The feature engineering of data includes:

[0112] Feature extraction, extracting useful information from raw data as features;

[0113] Feature selection: select the features that have the greatest impact on the model’s prediction performance from the existing features to reduce the model’s complexity and the risk of overfitting;

[0114] Feature construction: creating feature sets based on raw data to improve model performance; methods of feature construction include but are not limited to:

[0115] Feature combination: combine two or more original features to form a new feature;

[0116] Feature derivation: new features are calculated based on the original features;

[0117] Feature aggregation: Group or segment the original data, and then perform aggregation operations on each group or segment to form new features;

[0118] Convert text data into numerical features;

[0119] Automatic extraction of image features based on deep learning model is performed through convolutional neural network (CNN).

[0120] Data cleaning is an important step in data preprocessing, which aims to identify and correct errors, inconsistencies and outliers in the data to improve data quality and ensure the accuracy of subsequent analysis and modeling;

[0121] Through data standardization, the data is scaled so that it falls into a small specific interval (such as 0 to 1 or -1 to 1) to eliminate the influence of different dimensions on data analysis and improve the convergence speed and accuracy of the model;

[0122] Feature engineering is the process of extracting, selecting, and constructing features from raw data for model training, aiming to improve the predictive performance of the model.

[0123] On the basis of the above technical solutions, in the feature construction stage, based on the historical feature construction results, dimensionality reduction processing is performed to reduce the number of features to reduce the data dimension, while retaining important information in the original data as much as possible. The main purpose of dimensionality reduction is to simplify the data model, improve computational efficiency, and reduce the risk of overfitting.

[0124] For example, dimensionality reduction is an optional step in the data preparation stage. Through dimensionality reduction, noise and redundant information in the data can be removed, the generalization ability of the model can be improved, and the consumption of computing resources can be reduced. When performing dimensionality reduction, it is necessary to weigh the relationship between information retention and computational efficiency and select appropriate dimensionality reduction methods and parameters. At the same time, it is also necessary to evaluate the impact of dimensionality reduction on model performance through methods such as cross-validation to ensure that the final selected feature subset or new feature set can maximize the predictive performance of the model.

[0125] Feature selection is the process of selecting the most useful feature subset from the original feature set to predict the target. This is usually based on the importance, relevance, or redundancy of the feature with other features. Feature selection methods include filtering methods (such as chi-square test, mutual information), packaging methods (such as recursive feature elimination), and embedding methods (such as model-based feature selection).

[0126] Feature extraction is to convert the original features into new, fewer features through some transformation. These new features are usually some combination or abstract representation of the original features. Common feature extraction methods include principal component analysis (PCA), linear discriminant analysis (LDA), singular value decomposition (SVD), and autoencoders in deep learning.

[0127] Based on the above technical solution, the model training subsystem obtains the characterization data from the data preprocessing subsystem, determines the model architecture according to the characterization data, and performs training;

[0128] The training results are evaluated through the model evaluation module. If the performance does not meet the standard, the model parameters are adjusted and retrained, or the model architecture is changed and retrained.

[0129] The model architecture includes: random forest, gradient boosting machine GBM or neural network;

[0130] The performance evaluation was performed using cross validation to estimate model performance.

[0131] Random Forest is an ensemble learning algorithm that improves prediction accuracy by building multiple decision trees and aggregating their prediction results. Specifically, when building each decision tree, Random Forest randomly selects a portion of samples (row sampling) and a portion of features (column sampling) from the training data set. This randomness helps reduce the risk of overfitting the model and improves the generalization ability of the model. The output of Random Forest is usually the average of the outputs of all decision trees (for regression problems) or the majority voting result (for classification problems).

[0132] Gradient Boosting Machine (GBM) is an ensemble learning algorithm based on the idea of ​​boosting. It iteratively trains a series of weak learners (such as decision trees) and combines their prediction results to form a strong learner. In each iteration, GBM tries to correct the prediction error of the previous learner, that is, to train a new learner based on the residual (the difference between the true value and the predicted value) of the previous learner. This step-by-step optimization method makes GBM perform well in processing complex data and solving problems such as regression and classification.

[0133] Neural Network is a computational model that simulates the structure and function of biological neural networks. It is composed of a large number of neurons (or nodes) connected to each other. Each neuron receives input signals from other neurons and generates output signals after processing through a certain activation function. Neural networks learn the feature representation and mapping relationship of data by adjusting the connection weights between neurons. With the development of deep learning technology, neural networks have achieved remarkable results in image recognition, speech recognition, natural language processing and other fields. Neural networks can automatically extract features from raw data and learn complex nonlinear mapping relationships, which makes them powerful in processing high-dimensional and nonlinear data.

[0134] Based on the above technical solution, the model training subsystem performs multi-point intervention in the training process according to user instructions, and the intervention is any one of the following:

[0135] Adjust tunable parameters in the data preprocessing subsystem, specify model architecture, and optimize model parameters.

[0136] On the basis of the above technical solution, the demand forecasting subsystem predicts future demand according to a preset time span. When forecasting, the corresponding forecasting results are presented by selecting reference factors. The reference factors include: historical sales trends, seasonal changes, market activities, and economic indicators.

[0137] On the basis of the above technical solution, the inventory optimization subsystem presets several inventory strategies, adapts each inventory strategy and generates an optimal inventory level configuration report based on the forecast results of the demand forecasting subsystem, to assist enterprises in formulating procurement and sales plans.

[0138] Based on the above technical solution, the inventory optimization subsystem modifies the weight and importance ranking of each inventory strategy according to user instructions, and then optimizes the inventory strategy as needed.

[0139] Furthermore, the inventory optimization subsystem modifies the inventory strategy based on the inventory level according to the user's instructions, including:

[0140] The corresponding relationship between low-level inventory and medium-level inventory is set, and a first threshold is set. The medium-level inventory and low-level inventory within the first threshold ratio can be borrowed from each other;

[0141] A corresponding relationship between the middle-level inventory and the high-level inventory is set, and a second threshold is set. The high-level inventory and the middle-level inventory within the second threshold ratio can be borrowed from each other.

[0142] Based on the above technical solution, the inventory optimization subsystem obtains data from the data collection subsystem and compares it with the optimal inventory level configuration report to determine the specific inventory execution status and generate corresponding inventory strategies based on the specific execution status for auditing and inventory strategy adjustment.

[0143] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

[0144] The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. Any equivalent modifications or changes made by those skilled in the art based on the contents disclosed by the present invention should be included in the protection scope recorded in the claims.

Claims

1. A retail industry inventory management and demand forecasting system, characterized in that: include: Data collection subsystem, used to collect multi-dimensional data of the retail industry; Data preprocessing subsystem, used to clean, standardize and feature engineer the collected data; A model training subsystem, which is used to use big data and generate a commodity demand forecasting model through machine learning algorithms; Demand forecasting subsystem, used to obtain forecasting results based on commodity demand forecasting model; The inventory optimization subsystem is used to dynamically adjust inventory levels based on forecast results to achieve inventory optimization.

2. A retail industry inventory management and demand forecasting system as claimed in claim 1, characterized in that: The data collection subsystem connects with the following systems to extract data, and then stores it in the big data platform to form a data warehouse: POS system, including information related to transaction processing and fee settlement; Ordering systems, including information related to order receipt and order fulfillment; Supply chain systems, including logistics and inventory-related information; Cloud commerce system, including information related to online sales and promotions; Social systems, including information about customer interactions and market trends.

3. A retail industry inventory management and demand forecasting system as claimed in claim 2, characterized in that: The source data in the data warehouse is parsed and labeled according to the following dimensions to form multi-dimensional data: Historical sales data; Inventory level data; Promotional activity data; Holiday data; Seasonal data; Economic indicator data; Market trend data.

4. A retail industry inventory management and demand forecasting system as claimed in claim 1, characterized in that: The collected data is cleaned, standardized and feature engineered, wherein data cleaning utilizes a distributed architecture to parallelize data cleaning tasks, specifically including: Analyze missing values ​​and fill or delete them based on actual conditions; Analyze outliers and mark or remove them; Data deduplication processing; The data standardization specifically includes: Convert all date data to a uniform format; Convert numerical data to a consistent scale for comparison and analysis; The feature engineering of data includes: Feature extraction, extracting useful information from raw data as features; Feature selection: select the features that have the greatest impact on the model’s prediction performance from the existing features to reduce the model’s complexity and the risk of overfitting; Feature construction: creating feature sets based on raw data to improve model performance; methods of feature construction include but are not limited to: Feature combination: combine two or more original features to form a new feature; Feature derivation: new features are calculated based on the original features; Feature aggregation: Group or segment the original data, and then perform aggregation operations on each group or segment to form new features; Convert text data into numerical features; Automatic extraction of image features based on deep learning model is performed through convolutional neural network (CNN).

5. A retail industry inventory management and demand forecasting system as claimed in claim 4, characterized in that: In the feature construction stage, dimensionality reduction processing is performed based on the historical feature construction results to reduce the number of features to reduce the data dimension, while retaining important information in the original data as much as possible.

6. A retail industry inventory management and demand forecasting system as claimed in claim 1, characterized in that: The model training subsystem obtains the characterized data from the data preprocessing subsystem, determines the model architecture based on the characterized data, and performs training; The training results are evaluated through the model evaluation module. If the performance does not meet the standard, the model parameters are adjusted and retrained, or the model architecture is changed and retrained. The model architecture includes: random forest, gradient boosting machine GBM or neural network; The performance evaluation was performed using cross validation to estimate model performance.

7. A retail industry inventory management and demand forecasting system as claimed in claim 6, characterized in that: The model training subsystem performs multiple interventions in the training process according to user instructions. The interventions are any of the following: Adjust tunable parameters in the data preprocessing subsystem, specify model architecture, and optimize model parameters.

8. A retail industry inventory management and demand forecasting system as claimed in claim 1, characterized in that: The demand forecasting subsystem predicts future demand based on a preset time span and presents corresponding forecast results by selecting reference factors during the forecasting process. The reference factors include: historical sales trends, seasonal changes, market activities, and economic indicators.

9. A retail industry inventory management and demand forecasting system as claimed in claim 1, characterized in that: The inventory optimization subsystem presets several inventory strategies, adapts each inventory strategy and generates an optimal inventory level configuration report based on the forecast results of the demand forecasting subsystem, to assist enterprises in formulating procurement and sales plans.

10. A retail industry inventory management and demand forecasting system as claimed in claim 9, characterized in that: The inventory optimization subsystem obtains data from the data collection subsystem and compares it with the optimal inventory level configuration report to determine the specific inventory execution status and generate corresponding inventory strategies based on the specific execution status for auditing and inventory strategy adjustment.

Images