Self-adaptive optimization and algorithm fused catering store turnover prediction system

Abstract

The invention discloses an adaptive optimization and algorithm fused catering store turnover prediction system, which relates to the technical field of turnover prediction and specifically comprises a data acquisition and synchronization module, a data preprocessing module, a store adaptive optimization module, an algorithm fusion module and an integration and application module. Wherein the data acquisition and synchronization module regularly and automatically processes and stores data to a system database; the data preprocessing module is used for detecting and processing abnormal data; the store adaptive optimization module identifies the business states of different stores and matches the prediction model to train the stores; the algorithm fusion module compares the prediction models, selects high-precision models for fusion, and generates an optimal prediction model; and the integration and application module stores the prediction results in a database, and records and manages the prediction results of different stores. According to the method, the limitation of prediction precision of a single algorithm is broken through, advantage complementary prediction is realized, and the stability of a prediction result is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of turnover forecasting, and in particular to a restaurant store turnover forecasting system integrating adaptive optimization and algorithms. Background Art

[0002] Revenue forecasting remains a core and crucial component of daily operations and management for restaurant companies. Accurate revenue forecasts provide a scientific and effective basis for a variety of key operational decisions, including procurement planning, inventory control, staff scheduling, and promotional strategy development. These forecasts play a decisive role in reducing operating costs and increasing profit margins.

[0003] Currently, restaurant sales forecasts rely primarily on two mainstream methods: one is subjective estimation based on the experience of restaurant managers, and the other is mechanized calculation using simple statistical algorithms. Specifically, the former relies heavily on the personal historical experience and subjective judgment of restaurant managers, lacking an objective scientific basis. The forecast results are highly subjective and inaccurate, which in practice manifests as large forecast errors, making it difficult to accurately and effectively guide actual business decisions. While the latter introduces the concept of data analysis, it often remains based on basic traditional statistical models such as linear regression, moving average, and exponential smoothing. These models are overly simple and cannot effectively capture the nonlinear characteristics, cyclical fluctuations, and seasonal trends commonly found in restaurant sales data. This results in unreliable forecast results and an inability to meet the actual needs of refined restaurant management in complex scenarios.

[0004] Therefore, from the above analysis, it can be seen that the existing technology has the following obvious deficiencies: 1. Lack of adaptive store optimization mechanism: There are significant differences between catering stores in different regions, different sizes, and different consumer group characteristics. The patterns of turnover changes and influencing factors are also different. Most current prediction methods generally adopt a unified prediction model or parameter configuration, and fail to perform personalized modeling and optimization based on the unique operating characteristics of each store. This leads to poor generalization ability of the model. The prediction accuracy is high in some stores, but the prediction effect drops significantly in other stores, making it difficult to meet the wide and diverse needs in actual application scenarios.

[0005] 2. Insufficient robustness of a single algorithm: Existing methods often rely on a single statistical algorithm or a simple machine learning model, and fail to comprehensively consider the complementary advantages of multiple algorithms. When faced with abnormal situations such as sudden changes in the market environment and the impact of unexpected events (such as holiday promotions, large-scale public events or sudden weather disasters), the prediction accuracy often drops sharply. The model is not robust enough and lacks adaptability and fault tolerance to complex and changing business environments, making it difficult to provide long-term, stable and reliable data support for management decisions.

[0006] 3. Insufficient integration of multi-dimensional data: The accuracy of turnover forecasts is closely related to the multi-dimensional integration of data, but the input features used in existing technical solutions are usually relatively simple, and they fail to fully utilize external environmental data such as weather factors (such as temperature and rainfall), holiday factors (such as public holidays and promotional activity dates), and different business formats (fast food and group meals have inconsistent performance characteristics). The lack of data dimensions makes it impossible for the model to effectively explore and characterize the complex correlations and dynamic interactions between various factors, which seriously restricts the room for improvement of the model's predictive capabilities.

[0007] 4. Lack of dynamic adjustment mechanism: Existing turnover forecasting models generally exhibit static characteristics in actual applications. That is, once the model is built, it remains fixed for a long time. There is no effective real-time or regular update and iteration mechanism. The model parameters cannot be updated in time with the continuous accumulation of new historical data. It is difficult to adapt to the ever-changing market environment and consumption trends, resulting in the gradual failure of the forecasting model, thereby reducing the accuracy and effectiveness of the company's long-term operation and management decisions.

[0008] Based on the above problems, there is an urgent need to research and develop a catering store turnover prediction method that automatically adapts to the characteristics of different stores, integrates the advantages of multiple algorithms, has high robustness and supports the effective integration of multi-dimensional data, so as to significantly improve the accuracy and reliability of turnover prediction and better serve the refined and intelligent management decision-making needs of catering enterprises. Summary of the Invention

[0009] There are currently two methods for predicting restaurant sales. One is too subjective, and the other is that even if sales are predicted through data analysis, the model used is relatively simple and cannot meet the actual needs of refined management of restaurant stores in complex scenarios.

[0010] In order to solve the above technical problems, the technical solution of the present invention is: A restaurant store turnover prediction system that integrates adaptive optimization and algorithms, including Data acquisition and synchronization module, which is used to access the data source and automatically process and store data in the system database on a regular basis; A data preprocessing module, which is used for detecting abnormal data and processing abnormal data and missing data; A store adaptive optimization module, which automatically identifies the store's business format and matches it with several prediction models for each store, and trains the prediction models using an optimization algorithm; An algorithm fusion module compares the errors of several optimized prediction models through a test set, selects several high-precision and non-outlier models for fusion, and generates an optimal prediction model; The integration and application module stores the prediction results in the database, records and manages the prediction results of different stores, and is connected to the external API interface.

[0011] Preferably, the data source includes turnover data, which is automatically synchronized and obtained from the API of the business system every day, or the actual turnover data is uploaded by the user every day. The data source also includes weather historical data and future forecast data obtained through weather web pages, as well as holiday, adjusted holiday and working day information.

[0012] Preferably, the data preprocessing module includes an abnormal data detection module and an abnormal data processing module. The abnormal data detection module uses a sliding window to calculate the short-term mean and standard deviation, and locates extreme anomalies that deviate from the mean by more than 3σ through the interquartile range method. The abnormal data processing module performs a time series continuity check on the data. If there are short-term anomalies in the data, the average of the sales on the same day in the four weeks before the missing date is calculated as the sales on that day to supplement the data. If there are large missing data, the data are truncated.

[0013] Preferably, the business formats identified by the store adaptive optimization module include fast food business formats, formal dining and group dining business formats and casual dining business formats.

[0014] Preferably, the optimization algorithm includes random search, particle swarm optimization and Bayesian optimization.

[0015] Preferably, the algorithm fusion module filters abnormal prediction results, selects the top three models with the lowest errors, and generates store sales prediction results through a fusion method of a learning method or a weighted average method.

[0016] Preferably, the algorithm fusion module stores the prediction results and manages the store prediction records.

[0017] Preferably, the integration and application module is connected to the user end, the third-party production scheduling system, the supply chain system and the store system.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The present application provides a catering store turnover prediction system that integrates adaptive optimization and algorithms. It works together through multiple heterogeneous algorithms and dynamically adjusts the weights of each algorithm based on historical accuracy. It not only breaks through the limitations of the prediction accuracy of a single algorithm and avoids the limitations and deviations of a single model, but also realizes complementary prediction and significantly improves the stability of the prediction results.

[0019] In this application, an optimization algorithm is applied to tune the personalized parameters of stores, a more matching model is established, and the optimal parameter combination is automatically selected for the historical sales training model of different stores, achieving a breakthrough from "general model" to "store-specific model", and improving adaptability and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram of the structure of a restaurant store sales prediction system that integrates adaptive optimization and algorithms; Figure 2 This is a working diagram of a restaurant store sales forecasting system that integrates adaptive optimization and algorithms; Figure 3 A schematic diagram of the model training and fusion process in a restaurant store sales prediction system that integrates adaptive optimization and algorithms; Figure 4 This is a line chart of sales forecasts for different models in a restaurant store sales forecasting system that integrates adaptive optimization and algorithms. DETAILED DESCRIPTION

[0021] The following is a further description of specific embodiments of the present invention in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is intended to facilitate understanding of the present invention and does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0022] like Figures 1 to 4 As shown, the present invention discloses a restaurant store turnover prediction system that integrates adaptive optimization and algorithm, which specifically includes the following modules: The data collection and synchronization module is used to access data sources and automatically process and store data in the system database on a regular basis. Preferably, the data source includes sales data, which is automatically synchronized daily from the business system API or uploaded daily by the user. The data source also includes historical weather data and future forecast data obtained from the weather website, as well as information on holidays, adjusted holidays, and weekdays.

[0023] The data preprocessing module is used to detect abnormal data and process abnormal data and missing data. Preferably, the data preprocessing module includes an abnormal data detection module and an abnormal data processing module. The abnormal data detection module uses a sliding window to calculate the short-term mean and standard deviation, and uses the interquartile range method to locate extreme anomalies that deviate from the mean by more than 3σ. The appearance of abnormal turnover values ​​may be caused by the presence of rest, holiday closures, system or manual entry errors, etc., which cause the data to appear negative or extreme values. The abnormal data processing module checks the continuity of the time series of the data. If there are short-term anomalies in the data, the average turnover of the same day in the four weeks before the missing date is calculated as the turnover of the day to supplement it. If there are large gaps in the data (business is suspended for more than 30 consecutive days), the data is truncated. It also includes a model feature engineering module used as the input of the AI ​​data-driven model. After data processing, it selects historical sales (as the overall trend for the model to learn), time, day of the week (the weight during the week is lower than that of the weekend, and it will also be adjusted dynamically with the business type), holidays (the weight of holidays is higher than that of daily life, and it will also be adjusted dynamically with the business type), season (different business types will have peaks in different seasons, such as hot pot and barbecue), city, temperature (temperature will affect restaurant sales), wind direction and wind speed (bad weather will have an impact on sales), store business types and other data as features for the model to refer to in order to predict future sales.

[0024] The store adaptive optimization module automatically identifies the store's business type and matches it to several prediction models, such as BiLSTM, LSTM, four-week, Hlot-winters, GBDT, grey model, sarima, and prophet. It then trains the prediction models using optimization algorithms. Preferably, the store adaptive optimization module identifies business types including fast food, formal dining, group dining, and casual dining. Preferably, the optimization algorithms used include random search, particle swarm optimization, and Bayesian optimization.

[0025] For example, common business formats: Fast food business formats: in office areas, the characteristic is that daily sales are greater than holidays and weekends; in supermarkets and residential areas, the characteristic is that daily sales are less than holidays and weekends; formal dining and group dining business formats: common characteristics are that daily sales are lower than holidays and weekends; casual dining business formats: sales are more stable than other business formats, etc., and so on. The store adaptive optimization module first identifies the business format and then matches the corresponding feature weights. Use efficient hyperparameter optimization algorithms to optimize the parameters of the basic model, and choose to use efficient optimization algorithms to optimize the model, such as Random Search, Particle Swarm Optimization, Bayesian Optimization, etc. Figure 2 and Figure 3 As shown in the figure, the adaptive optimization of door-to-door stores is to select all the sales data of each store, combine all the prediction models, and use the optimization algorithm to conduct targeted training on each model, so that the optimized model hyperparameters are most suitable for the current store, and the optimal model training of the store is realized. At the same time, all models are compared based on the error to select several optimal models for application algorithm fusion.

[0026] The algorithm fusion module compares the errors of several optimized prediction models using a test set, selects several high-precision, non-outlier models for fusion, and generates the optimal prediction model. Preferably, the algorithm fusion module filters out abnormal prediction results, selects the top three models, and generates store sales forecasts through a learning method or weighted average fusion method. Preferably, the algorithm fusion module stores the prediction results and manages store prediction records.

[0027] The algorithm fusion module compares the training errors of multiple prediction algorithms and selects the model training results with high accuracy and no outliers (the models with excellent training performance among the above models are generally selected in the top three, and the specific models can be referred to the model library in the figure); the estimated effects of multiple excellent algorithms are fused by weighted or average methods to generate the sales forecast results of the store. The advantage of model fusion is that it can achieve the complementary advantages of different models and make the model results the most stable. At the same time, it filters out abnormal prediction results (by comparing with surrounding values, and at the same time, upper and lower limits cannot be set to prevent extreme errors). Figure 4 As shown in the figure below, the example fusion effect is shown (the prediction result of a randomly selected store). The three models with the smallest errors are selected for fusion. The prediction value after fusion has the highest stability and the smallest abnormal fluctuation.

[0028] The integration and application module stores the forecast results in a database, records and manages the forecast results for different stores, and is connected to the external API interface. Preferably, the integration and application module is connected to the user end, third-party production scheduling system, supply chain system, and store system, allowing users to directly view the forecast results. It not only supports multi-scenario calls, but also allows the option to directly connect data to the supply chain system to optimize supply chain efficiency. It can also connect to the store manager's service assistant to assist in manpower scheduling and improve labor efficiency.

[0029] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It is apparent to those skilled in the art that various changes, modifications, substitutions, and variations to these embodiments may be made without departing from the principles and spirit of the present invention, and the changes still fall within the scope of protection of the present invention.

Claims

1. A restaurant store sales forecasting system integrating adaptive optimization and algorithms, characterized by: include Data acquisition and synchronization module, which is used to access the data source and automatically process and store data in the system database on a regular basis; A data preprocessing module, which is used for detecting abnormal data and processing abnormal data and missing data; A store adaptive optimization module, which automatically identifies the store's business format and matches it with several prediction models for each store, and trains the prediction models using an optimization algorithm; An algorithm fusion module compares the errors of several optimized prediction models through a test set, selects several high-precision and non-outlier models for fusion, and generates an optimal prediction model; The integration and application module stores the prediction results in the database, records and manages the prediction results of different stores, and is connected to the external API interface.

2. The restaurant store turnover prediction system integrating adaptive optimization and algorithm according to claim 1 is characterized by: The data source includes turnover data, which is automatically synchronized and obtained from the API of the business system every day, or the actual turnover data is uploaded by the user every day. The data source also includes historical weather data and future forecast data obtained through weather web pages, as well as holiday, adjusted holiday and working day information.

3. The restaurant store turnover prediction system integrating adaptive optimization and algorithm according to claim 1 is characterized by: The data preprocessing module includes an abnormal data detection module and an abnormal data processing module. The abnormal data detection module uses a sliding window to calculate the short-term mean and standard deviation, and uses the interquartile range method to locate extreme anomalies that deviate from the mean by more than 3σ. The abnormal data processing module performs a time series continuity check on the data. If there are short-term anomalies in the data, the average of the sales on the same day in the four weeks before the missing date is calculated as the sales on that day to supplement the data. If there are large missing data, the data is truncated.

4. The restaurant store turnover prediction system integrating adaptive optimization and algorithm according to claim 1 is characterized by: The business formats identified by the store adaptive optimization module include fast food formats, formal dining and group dining formats, and casual dining formats.

5. The restaurant store turnover prediction system integrating adaptive optimization and algorithm according to claim 4 is characterized by: The optimization algorithms include random search, particle swarm optimization, and Bayesian optimization.

6. The restaurant store sales forecasting system integrating adaptive optimization and algorithm according to claim 1 is characterized by: The algorithm fusion module filters abnormal prediction results, selects the top three models with the lowest errors, and generates store sales prediction results through a fusion method of learning method or weighted average method.

7. The restaurant store turnover prediction system integrating adaptive optimization and algorithm according to claim 6 is characterized by: The algorithm fusion module stores the prediction results and manages the store prediction records.

8. The restaurant store sales forecasting system integrating adaptive optimization and algorithm according to claim 1, characterized in that: The integration and application module is connected to the user end, the third-party production scheduling system, the supply chain system and the store system.

Images