Chain store operation monitoring management method and system based on cloud platform technology

Abstract

The invention discloses a chain store operation monitoring management method and system based on a cloud platform technology, and relates to the technical field of chain store management. A chain store operation monitoring management system based on a cloud platform technology comprises a data acquisition module, a data processing module, a time sequence analysis module, a sales prediction module and an inventory alarm module. According to the invention, the preprocessed data are used as the input of the sales volume prediction model, so that the model can comprehensively capture the comprehensive influence of various factors on the sales volume, thereby more accurately predicting the future sales volume of various commodities. Historical sales volume data is decomposed into periodic components, long-term trend components and short-term fluctuation components through a time sequence analysis method, and the periodic components, the long-term trend components and the short-term fluctuation components are input as models, so that clearer time sequence features can be provided for the models, and the models can focus on learning specific modes and other key features of each component. And the sales volume prediction precision of the chain store operation monitoring management method and system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of chain store management, and particularly relates to a method and system for monitoring and managing the operation of chain stores based on cloud platform technology. Background Art

[0002] A cloud platform is a computing service platform built based on cloud computing technology, which is used to provide elastic expansion, highly available computing, network, and storage resources. According to functions, cloud platforms can be divided into three categories: storage types that focus on data storage, computing types that focus on data processing, and comprehensive types that balance computing and storage. The method and system for monitoring and managing the operation of chain stores based on cloud platform technology aim to help chain stores achieve efficient operation and scientific management through real-time monitoring, intelligent analysis, automated adjustment, and real-time feedback. Currently, when using the existing methods and systems for monitoring and managing the operation of chain stores, historical sales data is usually input into a prediction model to predict the sales situation of the stores, and the inventory threshold of goods is adjusted according to the prediction results to avoid situations such as out-of-stock and overstocking.

[0003] However, the existing methods and systems for monitoring and managing the operation of chain stores only use historical sales data as the model input and do not consider the impact of holidays, macroeconomic data, and promotional activity data on the sales situation. At the same time, the existing prediction models are also difficult to capture the periodic characteristics, long-term trend characteristics, and short-term fluctuation characteristics in historical sales data well, which easily leads to a relatively low overall prediction accuracy of the model. Moreover, only adjusting the inventory threshold of goods according to the prediction results lacks comprehensive consideration of the attributes of the goods themselves, the change rate of sales trends, and logistics stability, etc., resulting in poor rationality in setting the dynamic inventory threshold.

[0004] Based on the above situation, the present invention proposes a method and system for monitoring and managing the operation of chain stores based on cloud platform technology with high prediction accuracy and reasonable threshold setting. Summary of the Invention

[0005] In order to overcome the shortcomings that the existing methods and systems for monitoring and managing the operation of chain stores only use historical sales data as the model input and do not consider the impact of holidays, macroeconomic data, and promotional activity data on the sales situation. At the same time, the existing prediction models are also difficult to capture the periodic characteristics, long-term trend characteristics, and short-term fluctuation characteristics in historical sales data well, which easily leads to a relatively low overall prediction accuracy of the model. Moreover, only adjusting the inventory threshold of goods according to the prediction results lacks comprehensive consideration of the attributes of the goods themselves, the change rate of sales trends, and logistics stability, etc., resulting in poor rationality in setting the dynamic inventory threshold, the present invention proposes a method and system for monitoring and managing the operation of chain stores based on cloud platform technology with high prediction accuracy and reasonable threshold setting.

[0006] A method for monitoring and managing the operation of chain stores based on cloud platform technology, comprising the following steps: Obtain the historical sales data and promotional activity data of various commodities in each chain store through the cloud platform, and obtain holiday data and macroeconomic data through Internet services and data platforms; Preprocess the obtained historical sales data, promotional activity data, holiday data and macroeconomic data to obtain preprocessed historical sales data, promotional activity data, holiday data and macroeconomic data; Perform time series analysis on the historical sales data of various commodities through wavelet transform method and moving average method to obtain the periodic component, long-term trend component and short-term fluctuation component in the historical sales data; Construct a sales volume prediction model based on the optimized bidirectional LSTM model, and input the preprocessed historical sales data, promotional activity data, holiday data and macroeconomic data, as well as the periodic component, long-term trend component and short-term fluctuation component into the trained sales volume prediction model to obtain the predicted sales volume of various commodities; Based on the predicted sales volume, self-attributes, sales trend change rate and logistics stability of various commodities, calculate the dynamic inventory threshold of various commodities, and trigger an alarm and start the corresponding replenishment process when the actual inventory is lower than the dynamic inventory threshold.

[0007] As a preferred aspect of the invention, the historical sales data includes the sales date, commodity type, sales quantity and sales amount; the promotional activity data includes the promotion time, promotion type, promotion intensity and promotion effect; the holiday data includes the name, date of the holiday and whether it is a long holiday; the macroeconomic data includes the GDP growth rate, inflation rate and consumer confidence index.

[0008] As a preferred aspect of the invention, the specific steps for obtaining the periodic component are: Select the corresponding wavelet basis function based on the data characteristics of the sales quantity in the historical sales data; Analyze the sales quantity data of various commodities to determine the main cycle of various commodities , based on the main cycle And obtain the corresponding wavelet transform scale through calculation , the calculation formula is: ; Decompose the sales quantity data into wavelet coefficients of different scales through the wavelet basis function and the scaling function, and the calculation formula is:

[0009] where is the scale The wavelet coefficients at the position are the observed values of the sales quantity data at time ; is the scaling factor, and the translated wavelet basis function is the translation parameter of the wavelet function; Reconstruct the wavelet coefficients at a specific scale to obtain the periodic component, and the calculation formula is:

[0010] where is the value of the periodic component at the scale at time .

[0011] As a preferred aspect of the invention, the specific steps for obtaining the long-term trend component and the short-term fluctuation component are as follows: Determine an appropriate window size and calculate the long-term trend component and the short-term trend component in the historical sales data through the formula. The general calculation formula is:

[0012] where is the observed value of the sales quantity data in the historical sales data at time ; represents the size of the moving window, is the trend component of the sales quantity data at time . When takes the value of 2, represents the short-term trend component, and when takes the value of 5, represents the long-term trend component; Based on the short-term trend component in the historical sales data, calculate the short-term fluctuation component through the formula. The calculation formula is: where represents the observed value of the sales quantity data at time ; represents the short-term trend component of the sales quantity data at time ; represents the short-term fluctuation component of the sales quantity data at time .

[0013] As a preferred aspect of the invention, the sales volume prediction model includes an input layer, a bidirectional LSTM layer, an attention mechanism layer, a Dropout layer, a fully connected layer, and an output layer. The input layer is used to receive input data. The bidirectional LSTM layer is used to receive the data from the input layer and capture the bidirectional dependencies of the time series. The attention mechanism layer is used to receive the output of the bidirectional LSTM layer and weight the importance of the hidden states at each time step. The Dropout layer is used to receive the output of the attention mechanism layer and randomly discard the outputs of some neurons. The fully connected layer is used to receive the output of the Dropout layer and map the output of the Dropout layer to the output space. The output layer is used to receive the output of the fully connected layer and generate the final prediction result.

[0014] As a preferred aspect of the invention, the calculation formula of the dynamic inventory threshold is: where is the dynamic inventory threshold; is the predicted sales volume; is the time length corresponding to the predicted sales volume; is the sales trend change rate; is the shelf life adjustment coefficient; is the logistics stability ratio; is the service coefficient; is the demand standard deviation; is the replenishment lead time; is the inventory holding cost; and is the stockout cost.

[0015] A chain store operation monitoring and management system based on cloud platform technology includes a data acquisition module, a data processing module, a time series analysis module, a sales volume prediction module, and an inventory alarm module. Data acquisition module: used to acquire historical sales data, promotion activity data, holiday data, and macroeconomic data of various commodities in each store of the chain store. Data processing module: used to preprocess the acquired historical sales data, promotion activity data, holiday data, and macroeconomic data. Time series analysis module: used to perform time series analysis on the historical sales data of various commodities by means of wavelet transform method and moving average method, and obtain the periodic component, long-term trend component, and short-term fluctuation component in the historical sales data. Sales volume prediction module: used to construct a sales volume prediction model, and predict the future sales volume of various commodities through the sales volume prediction model to obtain the predicted sales volume of various commodities. Inventory Alarm Module: It is used to set dynamic inventory thresholds for various types of goods based on multiple factors. When the actual inventory is lower than the dynamic inventory threshold, it triggers an alarm and initiates the corresponding replenishment process.

[0016] The present invention has the following advantages: 1. By using the preprocessed historical sales data, promotion activity data, holiday data, and macroeconomic data as the input of the sales volume prediction model, the present invention not only enables the model to comprehensively capture the comprehensive influence of various factors on the sales volume, thereby more accurately predicting the future sales volume of various types of goods, but also enables the model to capture the lagged or indirect influence of the economic cycle on sales, and accurately predict the sales volume trend under the economic change trend, thus improving the prediction accuracy of the model. By decomposing the historical sales volume data into periodic components, long-term trend components, and short-term fluctuation components through time series analysis methods and using them as model inputs, it can not only provide clearer time series characteristics for the model, thereby reducing the amount of information that the model needs to process, but also enables the model to more focus on and efficiently learn the specific patterns and other key features of each component, improving the sales volume prediction accuracy of this chain store operation monitoring and management method and system.

[0017] 2. By introducing an attention mechanism into the sales volume prediction model and weighting the hidden states output by the bidirectional LSTM layer, the present invention not only enables the model to automatically learn and focus on the time steps that are significant for prediction in the time series, accurately capture key features and the patterns and laws of sales volume changes, thereby improving the stability and generalization ability of the model and reducing the risk of model overfitting, but also the attention mechanism can provide richer information and more flexible feature capture ability for the model by directly associating elements at different positions in the sequence and alleviate the gradient disappearance problem, enabling the model to better capture the complex dependencies and patterns in the long sequence, thereby improving the prediction accuracy of the model for long sequence sales volume data and enhancing the sales volume prediction performance of this chain store operation monitoring and management method and system.

[0018] 3. By calculating the dynamic inventory thresholds for various types of goods based on the predicted sales volume, self-attributes, sales trend change rate, and logistics stability of various types of goods, the present invention can comprehensively consider factors such as the shelf life of goods, sales prediction, and inventory turnover rate. Thus, it can not only effectively avoid excessive inventory and resulting in the expiration and scrapping of goods, reduce loss costs, reduce capital occupation and warehousing costs, but also ensure the timeliness and stability of commodity supply, and reduce the occurrence of out-of-stock situations. Moreover, dynamically setting the inventory threshold can also make the store more scientific and accurate in inventory management, improving the rationality of inventory threshold setting in this chain store operation monitoring and management method and system. Description of the Drawings

[0019] Figure 1It is a flowchart of an operation monitoring and management method for chain stores based on cloud platform technology adopted in the embodiments of the present invention.

[0020] Figure 2 It is a structural schematic diagram of an operation monitoring and management system for chain stores based on cloud platform technology adopted in the embodiments of the present invention. Detailed implementation manners

[0021] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0022] Embodiment 1, an operation monitoring and management method for chain stores based on cloud platform technology, as Figure 1 shown, includes the following steps: Obtain the historical sales data and promotional activity data of various commodities in each chain store through the cloud platform, and obtain holiday data and macroeconomic data through Internet services and data platforms; Preprocess the obtained historical sales data, promotional activity data, holiday data and macroeconomic data to obtain the preprocessed historical sales data, promotional activity data, holiday data and macroeconomic data; Perform time series analysis on the historical sales data of various commodities through the wavelet transform method to obtain the periodic components in the historical sales data, and process the historical sales data of various commodities through the moving average method to obtain the long-term trend components and short-term fluctuation components in the historical sales data; Construct a sales volume prediction model based on the optimized bidirectional LSTM model, and input the preprocessed historical sales data, promotional activity data, holiday data and macroeconomic data, as well as the periodic components, long-term trend components and short-term fluctuation components into the trained sales volume prediction model to obtain the predicted sales volume of various commodities; Based on the predicted sales volume, own attributes, sales trend change rate and logistics stability of various commodities, calculate the dynamic inventory threshold of various commodities, and trigger an alarm and start the corresponding replenishment process when the actual inventory is lower than the dynamic inventory threshold.

[0023] The historical sales data includes the sales date, commodity type, sales quantity and sales amount; the promotional activity data includes the promotion time, promotion type (such as discount, full reduction, gift, etc.), promotion intensity and promotion effect; the holiday data includes the name, date of the holiday and whether it is a long holiday; the macroeconomic data includes the GDP growth rate, inflation rate and consumer confidence index.

[0024] It should be noted that the specific steps for preprocessing the obtained historical sales data, promotion activity data, holiday data, and macroeconomic data are as follows: Data conversion: Convert the obtained data into a time series format, usually with dates as the index and sales quantity or sales amount as the values. Time alignment: Unify the starting reference time points of all time series and the time frequencies of all time series. Fill in the missing values at new time points for time series with low time frequencies through interpolation methods such as linear interpolation or polynomial interpolation to ensure that all time series data have values at the same time points. Missing value handling: Handle the null or missing values in the time series. Fill in the missing values through interpolation methods such as linear interpolation or polynomial interpolation, or directly fill in the missing values with the mean or median of the time series to ensure data integrity and model stability. Outlier handling: Identify outliers that do not conform to the expected pattern in the time series through the Z-Score method or the IQR method. Avoid the negative impact of these outliers on model training by deleting the outliers and replacing them with the mean or median of the time series or using interpolation methods to repair the outliers. Data standardization: Standardize data with different dimensions through min-max standardization. The calculation formula for min-max standardization is: , where represents the original data value, represents the minimum value in the time series, represents the maximum value in the time series, represents the standardized data value.

[0025] The specific steps for performing time series analysis on the historical sales data of various commodities through the wavelet transform method to obtain the periodic components in the historical sales data are as follows: Select the corresponding wavelet basis function based on the data characteristics of the sales quantity in the historical sales data. If the sales quantity data is smooth and has a long-term trend, select the Daubechies function as the wavelet basis function. If the sales quantity data fluctuates violently, select the Haar function as the wavelet basis function. Analyze the sales quantity data of various commodities to determine the main periods of various commodities , based on the main period and calculate the corresponding wavelet transform scale . The specific calculation formula is: ; Decompose the sales quantity data into wavelet coefficients of different scales through the wavelet basis function and the scaling function. The specific calculation formula is:

[0026] where is the wavelet coefficient at the scale and the position, is the observed value of the sales quantity data at time , is the scale and the translated wavelet basis function, and is the translation parameter of the wavelet function, which represents the translation position of the wavelet basis function on the time axis and is used to capture the local characteristics of the time series at different positions. takes integer values ranging from to , but in practical applications, due to the limited data length, it only needs to take values within the valid range; Reconstruct the wavelet coefficients at a specific scale to obtain the periodic component. The specific calculation formula is:

[0027] where is the value of the periodic component at the scale at time , is the scale and the position wavelet coefficient, is the scale and the translated wavelet basis function.

[0028] It should be noted that the wavelet transform scales for different types of commodities may be different. The main basis for selection is the periodic characteristics of the commodity sales data. The specific steps for scale selection are as follows: Determine the commodity sales cycle. Analyze the historical sales data of the commodity to determine its main cycle. For example, some commodities have an obvious weekly cycle (cycle of 7 days), while others may have a monthly cycle (cycle of about 30 days) or a quarterly cycle (cycle of about 90 days), etc.; Select the scale according to the cycle. The relationship between the wavelet transform scale and the cycle is usually . Suppose the main sales cycle of a certain commodity is 12 months, then the corresponding wavelet scale is about 8.5, and the integer 9 is taken. For commodities with a weekly cycle (7 days), assuming that sales data is collected once a day, the corresponding wavelet scale is about 2.8 (the integer 3 is taken).

[0029] The specific steps of processing the historical sales data of various commodities by the moving average method to obtain the long-term trend component and short-term fluctuation component in the historical sales data are as follows: Determine an appropriate window size and calculate the long-term trend component and short-term trend component in the historical sales data through a formula. The general calculation formula is specifically:

[0030] Where represents the observed value of the sales quantity data in the historical sales data at time , represents the half-width of the moving average window, then represents the size of the moving window, represents the trend component of the sales quantity data at time . When takes the value of 2, represents the short-term trend component, and when takes the value of 5, represents the long-term trend component; Based on the short-term trend component in the historical sales data and through calculation, obtain the short-term fluctuation component. The calculation formula is specifically: Where represents the observed value of the sales quantity data at time , represents the short-term trend component of the sales quantity data at time , represents the short-term fluctuation component of the sales quantity data at time .

[0031] By using the preprocessed historical sales data, promotional activity data, holiday data, and macroeconomic data as the input of the sales volume prediction model, the model can not only comprehensively capture the comprehensive impact of various factors on the sales volume, so as to more accurately predict the future sales volume of various commodities, but also enable the model to capture the lagged or indirect impact of the economic cycle on sales, and accurately predict the sales volume trend under the economic change trend, thereby improving the prediction accuracy of the model. And by decomposing the historical sales volume data into periodic components, long-term trend components, and short-term fluctuation components through time series analysis methods and using them as model inputs, it can not only provide clearer time series characteristics for the model, thereby reducing the amount of information that the model needs to process, but also enable the model to more focus and efficiently learn the specific patterns and other key characteristics of each component, improving the sales volume prediction accuracy of this chain store operation monitoring and management method and system.

[0032] The sales volume prediction model based on the optimized bidirectional LSTM model includes an input layer, a bidirectional LSTM layer, an attention mechanism layer, a Dropout layer, a fully connected layer, and an output layer. The input layer is used to receive input data; the bidirectional LSTM layer is used to receive the data from the input layer and capture the bidirectional dependencies of the time series; the attention mechanism layer is used to receive the output of the bidirectional LSTM layer and weight the importance of the hidden states at each time step; the Dropout layer is used to receive the output of the attention mechanism layer and randomly discard the outputs of some neurons to prevent overfitting; the fully connected layer is used to receive the output of the Dropout layer and map the output of the Dropout layer to the output space; the output layer is used to receive the output of the fully connected layer and generate the final prediction result.

[0033] It should be noted that the specific content and implementation steps of the attention mechanism layer are as follows: Receive the hidden states from the bidirectional LSTM layer and calculate the attention scores for each time step through a feed-forward neural network, which usually contains a fully connected layer that can first map the hidden states to a lower-dimensional space and then introduce non-linearity using an activation function. The specific calculation formula is: where is the attention score at time step t, is the attention score, is the hidden state of the bidirectional LSTM layer at time step t, W is the weight matrix, b is the bias term; Convert the attention scores to attention weights through the softmax function and ensure that the sum of all weights is 1. The specific calculation formula is: where is the attention weight at time step t, T is the length of the time series; Perform a weighted sum of the attention weights and the hidden states of the bidirectional LSTM layer to obtain the weighted feature representation. The specific calculation formula is:

[0034] where is the weighted feature representation.

[0035] By introducing the attention mechanism into the sales volume prediction model and weighting the hidden states output by the bidirectional LSTM layer, the above steps not only enable the model to automatically learn and focus on the time steps that are significant for prediction in the time series, so as to accurately capture the key features and the patterns and laws of sales volume changes, thereby improving the stability and generalization ability of the model and reducing the risk of model overfitting, but also the attention mechanism can provide the model with richer information and more flexible feature capture ability and alleviate the vanishing gradient problem by directly correlating the elements at different positions in the sequence, enabling the model to better capture the complex dependencies and patterns in the long sequence, thus improving the prediction accuracy of the model for the long sequence sales volume data and enhancing the sales volume prediction performance of this chain store operation monitoring and management method and system.

[0036] It should be noted that the training steps of the sales volume prediction model are specifically as follows: Data preparation and preprocessing: Obtain the past sales data, past promotion activity data, past holiday data, and past macroeconomic data of various commodities in each store of the chain store, perform time alignment, missing value processing, outlier processing, and data standardization on all the obtained data, and divide all the preprocessed data into a training set, a validation set, and a test set according to the ratios of 70%, 15%, and 15% respectively. Forward propagation and loss calculation: Input the sample data in the training set into the sales volume prediction model to obtain the predicted sales volume of various commodities, and use the mean squared error loss function to calculate the loss between the predicted sales volume and the actual sales volume of various commodities. Backward propagation and parameter update: Calculate the gradients of the parameters of each layer from the loss in the reverse direction through the chain rule, and use gradient descent algorithms such as SGD and Adam to adjust the parameters to minimize the loss. Iterative optimization: Divide the training set into multiple small batches, repeat the previous two steps batch by batch, traverse the entire training set multiple times until the model converges or reaches the preset number of iterations. Validation and testing: Periodically evaluate the losses and metrics of the sales volume prediction model through the sample data in the validation set to prevent overfitting, and evaluate the final performance of this sales volume prediction model through the sample data in the test set after training.

[0037] The specific calculation formula for the dynamic inventory threshold of various commodities based on the predicted sales volume, its own attributes, sales trend change rate, and logistics stability is as follows: Where represents the dynamic inventory threshold; represents the predicted sales volume; represents the time length corresponding to the predicted sales volume; Represents the sales trend change rate, obtained through sales trend analysis; Represents the shelf life adjustment coefficient, determined according to the shelf life of various commodities; Represents the logistics stability ratio, determined according to the on-time rate of freight logistics; Represents the service coefficient, determined based on the service level expected by the store; Represents the demand standard deviation, calculated from historical sales data; Represents the replenishment lead time, specifically the time from when the store places an order to when the supplier delivers the goods; Represents the inventory holding cost; and Represents the stockout cost.

[0038] For example, assume that the predicted sales volume of a certain commodity is 10 pieces per day, the prediction period is 7 days, the service coefficient is 1.65, the demand standard deviation is 2 pieces per day, the replenishment lead time is 5 days, the sales trend change rate is 0.05, the shelf life adjustment coefficient is 0.1, the logistics stability ratio is 0.2, the inventory holding cost is 10% per year, and the stockout cost is 50 yuan per time. Then the dynamic inventory threshold is approximately: After rounding up, the dynamic inventory threshold of this commodity at this time is 81 pieces.

[0039] The above steps calculate the dynamic inventory thresholds of various commodities based on the predicted sales volume, self-attributes, sales trend change rate, and logistics stability of various commodities, etc., and can comprehensively consider factors such as the shelf life of commodities, sales forecasts, and inventory turnover rates. Thus, not only can it effectively avoid excessive inventory and resulting in the expiration and scrapping of commodities, reduce loss costs, reduce capital occupation and warehousing costs, but also ensure the timeliness and stability of commodity supply, and reduce the occurrence of stockout situations. Moreover, dynamically setting the inventory threshold can also make the store's inventory management more scientific and accurate, improving the rationality of the inventory threshold setting in this chain store operation monitoring and management method and system.

[0040] Embodiment 2, a chain store operation monitoring and management system based on cloud platform technology, as Figure 2 shown, includes a data acquisition module, a data processing module, a time series analysis module, a sales volume prediction module, and an inventory alarm module, Data acquisition module: Used to obtain the historical sales data and promotional activity data of various commodities in each store of the chain store through the cloud platform, and obtain holiday data and macroeconomic data through Internet services and data platforms; Data processing module: Used to preprocess the obtained historical sales data, promotional activity data, holiday data, and macroeconomic data to obtain the preprocessed historical sales data, promotional activity data, holiday data, and macroeconomic data; Time series analysis module: It is used to perform time series analysis on the historical sales data of various commodities through the wavelet transform method to obtain the periodic components in the historical sales data, and process the historical sales data of various commodities through the moving average method to obtain the long-term trend components and short-term fluctuation components in the historical sales data; Sales volume prediction module: It is used to construct a sales volume prediction model based on the optimized bidirectional LSTM model, and input the preprocessed historical sales data, promotion activity data, holiday data, and macroeconomic data, as well as the periodic components, long-term trend components, and short-term fluctuation components into the trained sales volume prediction model to obtain the predicted sales volume of various commodities; Inventory alarm module: It is used to calculate the dynamic inventory threshold of various commodities based on the predicted sales volume, self-attributes, sales trend change rate, and logistics stability of various commodities, and trigger an alarm and start the corresponding replenishment process when the actual inventory is lower than the dynamic inventory threshold.

[0041] It should be understood that those of ordinary skill in the art can make improvements or transformations according to the above description, and all such improvements and transformations should fall within the protection scope of the appended claims of the present invention. The parts not described in detail in this specification belong to the prior art well-known to those skilled in the art.

Claims

1. A chain store operation monitoring and management method based on cloud platform technology, characterized in that: The following steps are involved: Obtain historical sales data and promotion activity data of various commodities in each store of the chain store through the cloud platform, and obtain holiday data and macroeconomic data through Internet services and data platforms; Preprocessing the acquired historical sales data, promotion activity data, holiday data and macroeconomic data to obtain preprocessed historical sales data, promotion activity data, holiday data and macroeconomic data; Through wavelet transform and moving average method, the time series analysis of historical sales data of various commodities is carried out to obtain the cyclical components, long-term trend components and short-term fluctuation components in the historical sales data; Build a sales forecasting model based on the optimized bidirectional LSTM model, input the pre-processed historical sales data, promotional activity data, holiday data and macroeconomic data, as well as cyclical components, long-term trend components and short-term fluctuation components into the trained sales forecasting model, and obtain the predicted sales of various commodities; Based on the predicted sales volume, its own attributes, sales trend change rate and logistics stability of each type of commodity, the dynamic inventory threshold of each type of commodity is obtained through calculation. When the actual inventory is lower than the dynamic inventory threshold, an early warning is triggered and the corresponding replenishment process is started.

2. According to the cloud platform technology-based chain store operation monitoring and management method of claim 1, it is characterized in that: The historical sales data includes sales date, product type, sales quantity and sales amount; the promotion activity data includes promotion time, promotion type, promotion intensity and promotion effect; the holiday data includes the name and date of the holiday and whether it is a long holiday; the macroeconomic data includes GDP growth rate, inflation rate and consumer confidence index.

3. According to the cloud platform technology-based chain store operation monitoring and management method of claim 1, it is characterized in that: The specific steps of obtaining the periodic component are: Selecting a corresponding wavelet basis function based on the data features of the sales quantity in the historical sales data; Analyze the sales volume data of various commodities and determine the main cycles of various commodities , based on the main cycle And the corresponding wavelet transform scale is obtained by calculation , the calculation formula is: ; The sales quantity data is decomposed into wavelet coefficients of different scales through wavelet basis functions and scale functions. The calculation formula is: ， in It is Scale The wavelet coefficients of the position, is the sales quantity data at time The observed value of It is Scale translation The wavelet basis function is is the translation parameter of the wavelet function; The wavelet coefficients of a specific scale are reconstructed to obtain the periodic components. The calculation formula is: ， in It is The periodic components of the scale are in time The value of .

4. The chain store operation monitoring and management method based on cloud platform technology according to claim 1 is characterized in that: The specific steps of obtaining the long-term trend component and the short-term volatility component are: Determine the appropriate window size and calculate the long-term trend component and short-term trend component in the historical sales data through the formula. The general calculation formula is: ， in The sales quantity data in the historical sales data at time The observed value of represents the size of the moving window, is the sales quantity data at time The trend component of When the value of is 2, Represents a short-term trend component, while When the value of is 5, Represents the long-term trend component; Based on the short-term trend component in the historical sales data, the short-term volatility component is obtained by calculation. The calculation formula is: ， in Represents sales quantity data at time The observed value of Represents sales quantity data at time The short-term trend component of Represents sales quantity data at time short-term volatility component.

5. The chain store operation monitoring and management method based on cloud platform technology according to claim 1 is characterized in that: The sales prediction model includes an input layer, a bidirectional LSTM layer, an attention mechanism layer, a Dropout layer, a fully connected layer and an output layer, wherein the input layer is used to receive input data; the bidirectional LSTM layer is used to receive data from the input layer and capture the bidirectional dependency of the time series; the attention mechanism layer is used to receive the output of the bidirectional LSTM layer and weight the importance of the hidden state of each time step; the Dropout layer is used to receive the output of the attention mechanism layer and randomly discard the output of some neurons; the fully connected layer is used to receive the output of the Dropout layer and map the output of the Dropout layer to the output space; the output layer is used to receive the output of the fully connected layer and generate the final prediction result.

6. A chain store operation monitoring and management method based on cloud platform technology according to claim 5, characterized in that: The calculation formula of the dynamic inventory threshold is: ， in is the dynamic inventory threshold; is the forecast sales volume; is the length of time corresponding to the predicted sales volume; is the sales trend change rate; is the shelf life adjustment factor; is the logistics stability ratio; is the service factor; is the standard deviation of demand; is the replenishment lead time; is the inventory holding cost; and is the stock-out cost.

7. A chain store operation monitoring and management system based on cloud platform technology, applied to a chain store operation monitoring and management method based on cloud platform technology as described in any one of claims 1 to 6, characterized in that: It includes data acquisition module, data processing module, time series analysis module, sales forecast module and inventory alarm module. Data acquisition module: used to obtain historical sales data and promotion activity data of various commodities in each store in the chain store, as well as holiday data and macroeconomic data; Data processing module: used to pre-process the acquired historical sales data, promotion activity data, holiday data and macroeconomic data; Time series analysis module: used to perform time series analysis on the historical sales data of various commodities through wavelet transform method and moving average method, and obtain the cyclical components, long-term trend components and short-term fluctuation components in the historical sales data; Sales volume forecasting module: used to build a sales volume forecasting model and use the sales volume forecasting model to forecast the future sales volume of various commodities and obtain the forecast sales volume of various commodities; Inventory alarm module: used to set dynamic inventory thresholds for various commodities based on multiple factors. When the actual inventory is lower than the dynamic inventory threshold, an early warning is triggered and the corresponding replenishment process is started.

Images