Catering material preparation prediction method and system
By adopting order prediction and granularity participation fusion methods with different time granularity in the catering industry, combining the LSTM model and dish characteristic parameters, the accuracy and comprehensiveness of dish order prediction in the catering industry are solved, and the accuracy and adaptability of prediction are improved , suitable for a variety of catering scenarios.
Patent Information
- Application Number
- CN202411287854.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2024-12-31
AI Technical Summary
The prior art is difficult to accurately predict dish orders in the catering industry, especially due to the difficulty in selecting time granularity, which leads to difficulty in fusion of local and global features, affecting the accuracy and comprehensiveness of the prediction.
In the catering preparation prediction method, order quantity prediction is carried out using different time particle sizes (such as days, weeks, and months), the granularity participation degree of each time particle size is obtained, the prediction results of different time particle sizes are fused, and the LSTM model is used to predict the dish order quantity. And adjust the weight according to the dish discount window and the new parameters to obtain the final order prediction results.
It improves the accuracy and adaptability of dish order prediction, can better capture different levels of characteristics in order volume, enhances the accuracy and applicability of prediction, and is suitable for catering institutions of different types and conditions.
Smart Images
Figure CN119228489A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of order data prediction based on neural network, and in particular to a catering material preparation prediction method and system. Background Art
[0002] The catering industry needs to face a large number of food orders every day, and preparing ingredients is a daily task.
[0003] It is challenging to prepare ingredients accurately based on predicted orders. It is also difficult to accurately capture the characteristics of sequence changes by using the collected data sequence for state prediction based on a simple neural network that can predict the state. In particular, the selection of the time granularity for state prediction is a difficult problem. Smaller time granularity can capture more local data change characteristics, but the data noise is also greater and the global performance is poor. Larger data granularity reduces the impact of noise and is easier to capture global fluctuation trends, but the ability to capture local fluctuations becomes weaker. Fusion of local and global results helps to obtain better results, but because the features contained in local and global predictions are more complex, such as time periods including seasons, months, holidays, etc., as well as food discounts and new tastes, it is challenging to fuse them in actual calculations to better retain the implicit features captured at different time granularities.
[0004] Therefore, the present invention needs to optimize the prediction of catering orders to achieve more accurate and comprehensive prediction of dish orders through local and global fusion for better catering preparation. Summary of the invention
[0005] The purpose of the present invention is to provide a catering preparation prediction method and system, which can achieve dish order prediction through local and global fusion to more accurately and comprehensively integrate global and local features for better catering preparation.
[0006] The present invention is achieved through the following technical solutions:
[0007] The present invention first provides a method for predicting catering preparation, comprising the following steps:
[0008] The order quantity of the next stage based on different time granularities is predicted by the prediction model, and a first order prediction result based on the first time granularity prediction, a second order prediction result based on the second time granularity prediction, and a third order prediction result based on the third time granularity prediction are obtained. The larger the time granularity, the more total days corresponding to the data sampled for predicting the order quantity of the next stage through the prediction model;
[0009] Obtaining the granularity participation π1, π2 and π3 of the first time granularity, the second time granularity and the third time granularity respectively, wherein the granularity participation is used to measure the numerical reliability of the prediction result of the corresponding time granularity in the current prediction, and the higher the granularity participation, the higher the numerical reliability;
[0010] Obtain the final order prediction result, and obtain the material preparation value according to the final order prediction result.
[0011] Preferably, the method for predicting the order volume of the next stage by using the prediction model is:
[0012] Obtain the actual order volume of each dish in multiple consecutive stages based on corresponding time granularity;
[0013] Based on the LSTM model, the predicted order volume for each dish in the next stage is output respectively.
[0014] Preferably, the method for respectively obtaining the granularity participations π1, π2 and π3 of the first time granularity, the second time granularity and the third time granularity is:
[0015] Get an array of the actual order volume for each dish at multiple consecutive stages based on the corresponding time granularity:
[0016]
[0017] in, represents the actual order quantity array of the dish numbered j at multiple consecutive stages based on the i-th time granularity, represents the actual order quantity of dish numbered j in the st-th stage based on the i-th time granularity, st=1,2,…,n, n is the total number of stages, and the value of n is different in each time granularity;
[0018] Get order influencing factor parameters;
[0019] Based on the and the order influencing factor parameters to obtain the granular participation.
[0020] Preferably, the method for obtaining order influencing factor parameters is:
[0021] Get the menu discount window parameters:
[0022]
[0023] in, represents the discount window parameter of the dish numbered j in multiple consecutive stages based on the i-th time granularity, pr j is the original price of dish numbered j, i,j,d% is the discount value of the dish numbered j on the dth day in the multiple consecutive stages based on the i-th time granularity, D i is the total number of days in multiple consecutive stages based on the i-th time granularity, dis i,d2now is the number of days from the dth day in the continuous multiple stages based on the i-th time granularity to the present, and τ is the empirical time coefficient;
[0024] Get new parameters of dishes:
[0025]
[0026] 2≤α≤2.5;
[0027] in, represents the new parameters of the dish numbered j in multiple consecutive stages based on the i-th time granularity, ag i,j,d is the number of days from the first launch of the dish numbered j on the dth day in the multiple consecutive stages based on the i-th time granularity, α is the first adjustment coefficient, β is the second adjustment coefficient, and e is a natural constant.
[0028] Preferably, the method for obtaining the second adjustment coefficient β is:
[0029] Record the number of orders for dish numbered j every day since the dish numbered j was first launched;
[0030] Build the order volume equation:
[0031]
[0032] Among them, na j,u is the number of orders for the dish numbered j on the uth day since its first launch, na j,v is the order volume of the dish numbered j on the vth day since its first launch, U is the number of days to be solved, and δ is the empirical threshold;
[0033] Solve to obtain the range of U values and take its minimum value U min ;
[0034] Get the value range of the second adjustment coefficient β:
[0035]
[0036] Preferably, based on the The method for obtaining the granular participation degree by using the order influencing factor parameters is as follows:
[0037]
[0038] Among them, π i ′is the intermediate parameter, ω i is the sequence number of the i-th time granularity among the three time granularities arranged from the smallest to the smallest, and J is the total number of dishes.
[0039] Preferably, the method for obtaining the final order prediction result is:
[0040]
[0041] Among them, PR j is the final order prediction result of the dish numbered j, is the prediction result of the i-th order of dish numbered j, alld i The number of days contained in a phase of the i-th time granularity.
[0042] Preferably, the first time granularity is day, the second time granularity is week, and the third time granularity is month.
[0043] The present invention also provides a catering material preparation prediction system, which is applied to any one of the catering material preparation prediction methods described above, comprising:
[0044] A first prediction module, used to predict the order quantity of the next stage through a prediction model based on the first time granularity to obtain a first order prediction result;
[0045] A second prediction module is used to predict the order quantity of the next stage through the prediction model based on the second time granularity to obtain a second order prediction result;
[0046] A third prediction module is used to predict the order quantity of the next stage through the prediction model based on the third time granularity to obtain a third order prediction result;
[0047] A granularity participation acquisition module is used to respectively acquire the granularity participation π3 of the first time granularity π1, the second time granularity π2 and the third time granularity, wherein the granularity participation is used to measure the numerical reliability of the prediction result of the corresponding time granularity in the current prediction, and the higher the granularity participation, the higher the numerical reliability;
[0048] The final prediction module is used to obtain the final order prediction result and obtain the material preparation value according to the final order prediction result.
[0049] The technical solution of the embodiment of the present invention has at least the following advantages and beneficial effects:
[0050] The present invention integrates order forecasting results of different time granularities, can capture different levels of features in order volume through data of different time granularities, and improves the accuracy of overall forecasting by using comprehensive information of short-term fluctuations and long-term trends;
[0051] The present invention assigns different weights to different time granularities to fuse the prediction results, which can effectively integrate the advantages of different time granularities, thereby further improving the accuracy and adaptability of the overall prediction;
[0052] The present invention determines the weights based on the data floating situation and the factors affecting the data, which can make the fusion adjustment more flexible, determine the reliability differences of different time granularities under different scenarios and times, and further improve the accuracy of the fusion prediction results;
[0053] The present invention is reasonably designed and can use existing neural networks to make predictions at different time granularities and then process them. The steps are easy to build and implement, and it is easy to apply and implement to catering institutions of different types and conditions, and has strong applicability and scalability. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] Figure 1 A schematic flow chart of a method for predicting food preparation in accordance with Embodiment 1 of the present invention. DETAILED DESCRIPTION
[0055] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0056] Example 1
[0057] The present invention first provides a method for predicting food preparation. Figure 1 , including the following steps:
[0058] The order quantity of the next stage based on different time granularities is predicted by the prediction model, and a first order prediction result based on the first time granularity prediction, a second order prediction result based on the second time granularity prediction, and a third order prediction result based on the third time granularity prediction are obtained. The larger the time granularity, the more total days corresponding to the data sampled for predicting the order quantity of the next stage through the prediction model;
[0059] Obtaining the granularity participation π1, π2 and π3 of the first time granularity, the second time granularity and the third time granularity respectively, wherein the granularity participation is used to measure the numerical reliability of the prediction result of the corresponding time granularity in the current prediction, and the higher the granularity participation, the higher the numerical reliability;
[0060] Obtain the final order prediction result, and obtain the material preparation value according to the final order prediction result.
[0061] In this embodiment, the first time granularity is day, the second time granularity is week, and the third time granularity is month.
[0062] As a preferred solution, the method for predicting the order volume of the next stage by using the prediction model is:
[0063] Obtain the actual order volume of each dish in multiple consecutive stages based on corresponding time granularity;
[0064] Based on the LSTM model, the predicted order volume for each dish in the next stage is output respectively. In the specific implementation, the length of each stage corresponds to the size of the time granularity. For example, the first time granularity is day, one stage is one day, the input is the actual order volume of daily dishes for multiple days, and the output is the predicted order volume for the next day. The second time granularity is week, one stage is one week, the input is the actual order volume of weekly dishes, and the output is the predicted order volume for the next week. The same is true for the third time granularity.
[0065] Furthermore, the method of respectively obtaining the granularity participations π1, π2 and π3 of the first time granularity, the second time granularity and the third time granularity is:
[0066] Get an array of the actual order volume for each dish at multiple consecutive stages based on the corresponding time granularity:
[0067]
[0068] in, represents the actual order quantity array of the dish numbered j at multiple consecutive stages based on the i-th time granularity, represents the actual order quantity of dish numbered j in the st-th stage based on the i-th time granularity, st=1,2,…,n, n is the total number of stages, and the value of n is different in each time granularity;
[0069] Get order influencing factor parameters;
[0070] Based on the and the order influencing factor parameters to obtain the granular participation.
[0071] Specifically, the method for obtaining the order influencing factor parameters is preferably:
[0072] Get the menu discount window parameters:
[0073]
[0074] in, represents the discount window parameter of the dish numbered j in multiple consecutive stages based on the i-th time granularity, prj is the original price of dish numbered j, i,j,d % is the discount value of the dish numbered j on the dth day in the multiple consecutive stages based on the i-th time granularity, D i is the total number of days in multiple consecutive stages based on the i-th time granularity, dis i,d2now is the number of days from the dth day in the continuous multiple stages based on the i-th time granularity to the present, and τ is the empirical time coefficient;
[0075] The parameters of the menu discount window actually obtain the proportion of discount periods in the parameters used in the prediction of each time granularity and the comprehensive situation of the distance of the discount window from now. The higher the discount proportion and the closer the time is to now, the more comprehensive the impact of the prediction of the nearest time granularity on the discount will be. Therefore The larger it is, the more convenient it will be for subsequent calculations to give the corresponding time granularity a greater weight.
[0076] Get new parameters of dishes:
[0077]
[0078] 2≤α≤2.5;
[0079] in, represents the new parameters of the dish numbered j in multiple consecutive stages based on the i-th time granularity, ag i,j,d is the number of days from the first launch of the dish numbered j on the dth day in the multiple consecutive stages based on the i-th time granularity, α is the first adjustment coefficient, β is the second adjustment coefficient, and e is a natural constant.
[0080] here The main consideration is that the sales volume of new dishes is not in the stable period after the new dishes are launched, which mainly represents the proportion of the new dishes window. i,j,d ) is an intermediate parameter obtained by the number of days the dish is launched. Its value is less than or equal to 1 and the larger the number of days, the closer it is to 1. The purpose of β is to adjust the number of days to start outputting the value 1. In other words, the fewer the stable period of the corresponding time granularity, the The larger the value of , the more features it captures are needed to obtain prediction results. For example, if new product promotion is mainly concentrated in the early stage of the most important third granularity, the second and first granularities may not capture its features at all, while the neural network based on the third granularity can capture the state changes of this part.
[0081] Based on the above solution, the method for obtaining the second adjustment coefficient β may be:
[0082] Record the number of orders for dish numbered j every day since the dish numbered j was first launched;
[0083] Build the order volume equation:
[0084]
[0085] Among them, na j,u is the number of orders for the dish numbered j on the uth day since its first launch, na j,v is the order volume of the dish numbered j on the vth day since its first launch, U is the number of days to be solved, and δ is the empirical threshold;
[0086] Solve to obtain the range of U values and take its minimum value U min ;
[0087] Get the value range of the second adjustment coefficient β:
[0088]
[0089] The main idea of the above method for obtaining the second adjustment coefficient β is to obtain the number of days when the sales volume begins to stabilize after the new dish is launched, and set the second adjustment coefficient β according to this number of days.
[0090] Next, based on the The method for obtaining the granular participation degree by using the order influencing factor parameters is as follows:
[0091]
[0092] Among them, π i ′ is the intermediate parameter, ω i is the sequence number of the i-th time granularity in the three time granularities arranged from the smallest to the smallest, and J is the total number of dishes. It is particularly noted that when obtaining the fusion weight, that is, the participation degree of the granularity π i When π i ′ Then normalize to get the actual weight of each time granularity. Calculate π i ′ The initial value of each time granularity is This value here becomes larger as the time granularity becomes smaller, that is, when fusion is performed, it is preset to pay more attention to local features, because local features usually have more details, and then a part of the global features are fused through the weights calculated in this embodiment, and each time granularity has certain global features, so that the prediction results that consider more details are integrated into a part of the results with more global features for refinement. It represents the comprehensive fluctuation of the corresponding granularity prediction. The larger the fluctuation, the larger the value. Therefore, if the fluctuation is large, it means that the data is more affected in the prediction of the time granularity. In other words, it is assumed that the prediction will capture more features, and more attention should be paid to the prediction results.
[0093] Finally, the method for obtaining the final order prediction result is:
[0094]
[0095] Among them, PR j is the final order prediction result of the dish numbered j, is the prediction result of the i-th order of dish numbered j, alld i is the number of days in a stage of the i-th time granularity. In other words, based on the prediction results of each time granularity, the prediction results are integrated with the weights to obtain the specific predicted sales of each dish in the future predicted day. After obtaining the predicted sales of each dish, the corresponding quantity of materials can be purchased in combination with the recipe requirements. The demand for ingredients can be calculated strictly based on sales, or adaptive adjustments can be made, such as purchasing 5%-10% more of each ingredient based on the predicted results.
[0096] To summarize, this embodiment fuses the prediction results based on different time granularities, which helps to capture global and local features and more accurately assign different weights to the results of each time granularity in different situations, better integrate the advantages of the predictions of each time granularity, and retain the impact of external factors on the sales of dishes as much as possible during the fusion, which helps to assist catering institutions in preparing ingredients more optimally and in a targeted manner.
[0097] Example 2
[0098] This embodiment provides a catering material preparation prediction system, which is applied to a catering material preparation prediction method described in the above embodiment, including:
[0099] A first prediction module, used to predict the order quantity of the next stage through a prediction model based on the first time granularity to obtain a first order prediction result;
[0100] A second prediction module is used to predict the order quantity of the next stage through the prediction model based on the second time granularity to obtain a second order prediction result;
[0101] A third prediction module is used to predict the order quantity of the next stage through the prediction model based on the third time granularity to obtain a third order prediction result;
[0102] A granularity participation acquisition module is used to respectively acquire the granularity participation π3 of the first time granularity π1, the second time granularity π2 and the third time granularity, wherein the granularity participation is used to measure the numerical reliability of the prediction result of the corresponding time granularity in the current prediction, and the higher the granularity participation, the higher the numerical reliability;
[0103] The final prediction module is used to obtain the final order prediction result and obtain the material preparation value according to the final order prediction result.
[0104] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for predicting food preparation, characterized in that: The following steps are involved: The order quantity of the next stage based on different time granularities is predicted by the prediction model, and a first order prediction result based on the first time granularity prediction, a second order prediction result based on the second time granularity prediction, and a third order prediction result based on the third time granularity prediction are obtained. The larger the time granularity, the more total days corresponding to the data sampled for predicting the order quantity of the next stage through the prediction model; Obtaining the granularity participation π1, π2 and π3 of the first time granularity, the second time granularity and the third time granularity respectively, wherein the granularity participation is used to measure the numerical reliability of the prediction result of the corresponding time granularity in the current prediction, and the higher the granularity participation, the higher the numerical reliability; Obtain the final order prediction result, and obtain the material preparation value according to the final order prediction result.
2. A catering preparation prediction method according to claim 1, characterized in that: The method for predicting the order volume in the next stage by using the prediction model is: Obtain the actual order volume of each dish in multiple consecutive stages based on corresponding time granularity; Based on the LSTM model, the predicted order volume for each dish in the next stage is output respectively.
3. A catering preparation prediction method according to claim 1, characterized in that: The method of respectively obtaining the granularity participations π1, π2 and π3 of the first time granularity, the second time granularity and the third time granularity is: Get an array of the actual order volume for each dish at multiple consecutive stages based on the corresponding time granularity: in, represents the actual order quantity array of the dish numbered j at multiple consecutive stages based on the i-th time granularity, represents the actual order quantity of dish numbered j in the st-th stage based on the i-th time granularity, st=1,2,…,n, n is the total number of stages, and the value of n is different in each time granularity; Get order influencing factor parameters; Based on the and the order influencing factor parameters to obtain the granular participation.
4. A catering preparation prediction method according to claim 3, characterized in that: The method for obtaining order influencing factor parameters is: Get the menu discount window parameters: in, represents the discount window parameter of the dish numbered j in multiple consecutive stages based on the i-th time granularity, pr j is the original price of dish numbered j, i,j,d % is the discount value of the dish numbered j on the dth day in the multiple consecutive stages based on the i-th time granularity, D i is the total number of days in multiple consecutive stages based on the i-th time granularity, dis i,d2now is the number of days from the dth day in the continuous multiple stages based on the i-th time granularity to the present, and τ is the empirical time coefficient; Get new parameters of dishes: 2≤α≤2.5； in, represents the new parameters of the dish numbered j in multiple consecutive stages based on the i-th time granularity, ag i,j,d is the number of days from the first launch of the dish numbered j on the dth day in the multiple consecutive stages based on the i-th time granularity, α is the first adjustment coefficient, β is the second adjustment coefficient, and e is a natural constant.
5. A catering preparation prediction method according to claim 4, characterized in that: The method for obtaining the second adjustment coefficient β is: Record the number of orders for dish numbered j every day since the dish numbered j was first launched; Build the order volume equation: Among them, na j,u is the number of orders for the dish numbered j on the uth day since its first launch, na j,v is the order volume of the dish numbered j on the vth day since its first launch, U is the number of days to be solved, and δ is the empirical threshold; Solve to obtain the range of U values and take its minimum value U min ; Get the value range of the second adjustment coefficient β:
6. A method for predicting food preparation according to claim 5, characterized in that: Based on the The method for obtaining the granular participation degree by using the order influencing factor parameters is as follows: Among them, π i ′ is the intermediate parameter, ω i is the sequence number of the i-th time granularity among the three time granularities arranged from the smallest to the smallest, and J is the total number of dishes.
7. A method for predicting food preparation according to claim 1, characterized in that: The method for obtaining the final order prediction result is: Among them, PR j is the final order prediction result of the dish numbered j, is the prediction result of the i-th order of dish numbered j, alld i The number of days contained in a phase of the i-th time granularity.
8. A method for predicting food preparation according to claim 1, characterized in that: The first time granularity is day, the second time granularity is week, and the third time granularity is month.
9. A catering material preparation prediction system, applied to a catering material preparation prediction method according to any one of claims 1 to 8, characterized in that: include: A first prediction module, used to predict the order quantity of the next stage through a prediction model based on the first time granularity to obtain a first order prediction result; A second prediction module is used to predict the order quantity of the next stage through the prediction model based on the second time granularity to obtain a second order prediction result; A third prediction module is used to predict the order quantity of the next stage through the prediction model based on the third time granularity to obtain a third order prediction result; A granularity participation acquisition module is used to respectively acquire the granularity participation π3 of the first time granularity π1, the second time granularity π2 and the third time granularity, wherein the granularity participation is used to measure the numerical reliability of the prediction result of the corresponding time granularity in the current prediction, and the higher the granularity participation, the higher the numerical reliability; The final prediction module is used to obtain the final order prediction result and obtain the material preparation value according to the final order prediction result.