An intelligent approach to hotel real-time demand forecasting and inventory management

Abstract

The present application belongs to the field of hotel management technology, and more specifically, relates to an intelligent method for hotel real-time demand forecasting and inventory management. The present invention ensures the accuracy and validity of data through data acquisition and preprocessing. The feature extraction link extracts key factors, such as the impact of favorable comment rate and regional activities on the hotel, and provides rich input data for subsequent deep learning models. Deep learning demand forecasting uses real-time data to quickly adapt to market changes and improve the accuracy of forecasts. In competitor analysis, by setting fluctuation thresholds of distance and favorable comment rate, an accurate grasp of the competitive environment is ensured, thereby providing a basis for hotels to formulate more targeted strategies. Finally, reinforcement learning is used to optimize inventory management and pricing strategies, so that hotels can respond flexibly in a dynamic market and improve operating efficiency.

Description

Technical Field

[0001] The present application belongs to the field of hotel management technology, and more specifically, relates to an intelligent method for real-time hotel demand forecasting and inventory management. Background Art

[0002] As an important part of the tourism industry, the hotel industry has developed rapidly in recent years with the economic globalization and the improvement of people's living standards. Hotels not only provide accommodation services, but also undertake multiple functions such as culture, entertainment, and leisure, becoming an important part of the tourism experience. With the popularization of information technology and the Internet, consumers have more diverse ways to obtain hotel information and more convenient booking methods, which has brought new development opportunities to the hotel industry, but also unprecedented challenges.

[0003] In this context, effective inventory management has become a key link in hotel operations. Hotel inventory management mainly refers to the reservation, allocation and scheduling of hotel rooms and other resources. Reasonable inventory management is crucial for hotels, which is directly related to hotel revenue management, including pricing strategy, room allocation, etc.

[0004] Currently, the hotel industry usually uses revenue management systems (RMS) for inventory management. These systems generally analyze historical data and combine market demand, seasonal factors, holidays, etc. to predict room demand and adjust room rates and inventory accordingly. However, these systems have the following limitations:

[0005] Prediction models are often based on historical data and are difficult to adapt to real-time market changes; historical data reflects past trends and patterns, but the market environment is dynamically changing. New market events, such as emergencies, major events, natural disasters, etc., may have a significant impact on current and future demand, and models based on historical data often cannot capture these changes in real time. And they cannot fully consider the impact of competitors' pricing strategies and customer reviews on demand, because: competitors may change prices in real time based on market demand, their own inventory conditions, and strategy adjustments. This dynamic pricing strategy has a direct impact on hotel demand, but traditional prediction models often cannot consider changes in competitors' prices in real time. Summary of the invention

[0006] The present invention provides an intelligent method for hotel real-time demand forecasting and inventory management, which is intended to solve the technical problems that the current forecasting model is based on historical data, is difficult to adapt to real-time market changes, and cannot consider the price changes of competitors in real time.

[0007] An intelligent method for real-time hotel demand forecasting and inventory management includes the following steps:

[0008] Data acquisition and preprocessing: collect holiday data, historical check-in data, customer evaluation data, regional activity data, and scenic spot data, and preprocess the collected data;

[0009] Feature extraction: Extract the praise rate of the hotel to be predicted based on customer evaluation data, extract the distance between regional activities and attractions and the hotel to be predicted based on regional activity data and attraction data, and calculate the impact of regional activities and attractions on the hotel to be predicted based on the distance;

[0010] Deep learning demand forecasting: Based on the pre-processed holiday data, historical check-in data and feature-extracted data as input to the deep learning model, the deep learning model is trained, and based on real-time demand forecasting, the real-time data is input into the trained deep learning model to obtain the hotel's demand forecast results;

[0011] Competitor determination: construct distance fluctuation thresholds and favorable comment rate fluctuation thresholds; filter out hotels within the threshold range based on the distance between regional activities and attractions and the hotel to be predicted and the distance fluctuation thresholds to obtain preliminary selected competitors; then filter out the preliminary selected competitors based on favorable comment rates and favorable comment rate fluctuation thresholds to obtain final competitor information;

[0012] Inventory management and scheduling: Based on the pricing information corresponding to the finalized competitor information and demand forecasts, reinforcement learning is used to optimize the hotel's inventory management and pricing strategy.

[0013] The present invention ensures the accuracy and validity of data through data acquisition and preprocessing. Then, the feature extraction link extracts key factors, such as the impact of favorable reviews and regional activities on the hotel, providing rich input data for subsequent deep learning models. Deep learning demand forecasting uses real-time data to quickly adapt to market changes and improve the accuracy of forecasts. In competitor analysis, by setting fluctuation thresholds for distance and favorable reviews, an accurate grasp of the competitive environment is ensured, thereby providing a basis for hotels to formulate more targeted strategies. Finally, reinforcement learning is used to optimize inventory management and pricing strategies, allowing hotels to respond flexibly in a dynamic market and improve operating efficiency.

[0014] Preferably, the specific steps of calculating the impact of regional activities on the hotels to be predicted are as follows:

[0015] Calculate the distance from each regional activity to the hotel:

[0016] ;

[0017] Where: R represents the radius of the earth; represents the latitude difference between hotel and activity j; represents the longitude difference between hotel and activity j; , Respectively represent the latitude and longitude of the hotel; , denote the latitude and longitude of activity j, respectively; represents the distance between hotel and activity j;

[0018] Calculate the impact of each activity:

[0019] ;

[0020] ;

[0021] ;

[0022] Where: represents the number of participants or scale of activity j; Indicates the number of days or time that activity j lasts; and represents the weight coefficient; represents the impact value of activity j on demand; represents the intensity of activity j; represents the geographical distance between hotel and activity j; The maximum distance parameter representing distance attenuation; represents the total impact of all regional activities on demand; J represents the total number of activities.

[0023] Preferably, the step of calculating the influence of scenic spots on the predicted hotels is as follows:

[0024] Calculate the geographical distance between each attraction and the hotel:

[0025] ;

[0026] Where: R represents the radius of the earth; represents the latitude difference between the hotel and the attraction k; represents the longitude difference between the hotel and the attraction k; , Respectively represent the latitude and longitude of the hotel; , Respectively represent the latitude and longitude of scenic spot k; represents the distance between the hotel and attraction k;

[0027] Calculate the impact of each attraction:

[0028] ;

[0029] ;

[0030] ;

[0031] Where: represents the average tourist flow of scenic spot k in the past n years; represents the weight coefficient; represents the impact of attraction k on hotel demand; represents the intensity of scenic spot k; represents the geographical distance between the hotel and attraction k; The maximum distance parameter representing distance attenuation; represents the total impact of all attractions on demand; K represents the number of attractions.

[0032] Preferably, the structure of the deep learning model is as follows:

[0033] Input layer: It includes four input channels, which respectively input holiday data, historical occupancy data, regional activity impact and scenic spot impact, and praise rate;

[0034] LSTM layer: LSTM is used to process the time dependency of holiday data, historical check-in data, and favorable review rates, and to extract time series patterns;

[0035] Convolutional layer: Apply convolutional layers to the regional activity impact and scenic spot impact data to extract local features and capture information at different scales through multiple convolution kernels;

[0036] Feature fusion layer: The features extracted by the LSTM layer and the features extracted by the convolution layer are combined into feature vectors through splicing operations to obtain fused features;

[0037] Fully connected layer: Based on the fusion features, feature mapping and fusion are performed through multiple fully connected layers, and each fully connected layer uses the ReLU activation function;

[0038] Output layer: includes a fully connected layer to output the prediction results, using a linear activation function.

[0039] Preferably, the loss function of the deep learning model is as follows:

[0040] ;

[0041] Where: represents the hotel demand forecast value given by the deep learning model; represents the actual demand at time step t; N represents the total number of samples; represents a hyperparameter used to control the penalty intensity of the weighted error term; Represents the weight coefficient at time step t, which is dynamically adjusted according to time; represents the hyperparameter of the balanced loss term;

[0042] ;

[0043] Where: Indicates the adjustment weight of the impact of holidays; Indicates the impact adjustment weights of working days and non-working days; Indicates the historical demand trend adjustment weight; Represents a holiday marking function. If it is a holiday, If it is not a holiday, ; Represents the weekday marking function. If it is a weekday, then If it is a non-working day, ; It represents the average demand value before time step tk.

[0044] Preferably, the competitor determination comprises the following steps:

[0045] Filter hotels that meet the distance fluctuation threshold:

[0046] ;

[0047] Where: Indicates the distance between the hotel to be predicted and the scenic spot; Indicates the distance fluctuation threshold; Indicates the distance from the candidate competitor to the attraction;

[0048] ;

[0049] Where: Indicates the distance between the hotel to be predicted and the activity area; represents the distance from the candidate competitor to the active area;

[0050] Filter out hotels that meet the load distance fluctuation threshold and obtain preliminary selected competitors;

[0051] The finalized competitors are screened based on the positive review rate and the positive review rate fluctuation threshold:

[0052] ;

[0053] Where: Indicates the praise rate of the hotel to be predicted; Indicates the fluctuation threshold of the favorable comment rate; Indicates the favorable rating of candidate competitors;

[0054] Competitors that meet the fluctuation threshold of favorable review rate are screened out from the initially screened competitors and are selected as the final competitors.

[0055] Preferably, the reinforcement learning in inventory management and scheduling is specifically as follows:

[0056] Optimization goal: Maximize the hotel's total revenue, which is achieved by optimizing pricing and inventory management strategies:

[0057] ;

[0058] Where: Represents the room rate of the hotel to be predicted at time t; represents the inventory of the hotel to be predicted at time t; represents the unit room cost; represents the profit at time t;

[0059] Adopt reinforcement learning framework to make optimal decisions, including state representation, action representation, reward function and Q learning update formula;

[0060] The state representation includes: current inventory levels, current demand forecasts, competitor pricing, and current historical revenue;

[0061] The actions are expressed as follows:

[0062] ;

[0063] in, To indicate an action;

[0064] Reward function:

[0065] ;

[0066] Where: represents the value of the reward function; represents the weight coefficient, which is used to control the impact of competitor price differences on hotel revenue; represents the house price of competitor i at time t;

[0067] Q function update formula:

[0068] ;

[0069] Where: represents the updated Q value; represents the Q value before updating; Indicates that from the next state Start and choose the Q value of the optimal action; represents the learning rate; represents the discount factor; represents the optimal action.

[0070] The beneficial effects of the present invention include:

[0071] The present invention ensures the accuracy and validity of data through data acquisition and preprocessing. Then, the feature extraction link extracts key factors, such as the impact of favorable reviews and regional activities on the hotel, providing rich input data for subsequent deep learning models. Deep learning demand forecasting uses real-time data to quickly adapt to market changes and improve the accuracy of forecasts. In competitor analysis, by setting fluctuation thresholds for distance and favorable reviews, an accurate grasp of the competitive environment is ensured, thereby providing a basis for hotels to formulate more targeted strategies. Finally, reinforcement learning is used to optimize inventory management and pricing strategies, allowing hotels to respond flexibly in a dynamic market and improve operating efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0073] Figure 1 An overall step block diagram provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0075] See also Figure 1 As shown, an intelligent method for hotel real-time demand forecasting and inventory management includes the following steps:

[0076] Data acquisition and preprocessing: collect holiday data, historical check-in data, customer evaluation data, regional activity data, and scenic spot data, and preprocess the collected data;

[0077] In the preprocessing stage, we mark the holiday data, that is, use and Mark the function. For details, refer to the marking rules mentioned in the subsequent steps. Please note that the marking is performed in this step.

[0078] As for the historical occupancy data, we extract the occupancy status of each day, i.e., the occupancy rate;

[0079] For customer evaluation data, we extracted the number of favorable comments and the number of reviews, so that in the subsequent stage, the favorable rate can be calculated by the number of favorable comments and the number of reviews;

[0080] For regional activity data and scenic spot data, we extract the regional activity location, number of people or scale, duration, etc.; for scenic spot data, we extract the location and the average tourist flow over the past n years to facilitate corresponding calculations during subsequent feature extraction.

[0081] Feature extraction: Extract the praise rate of the hotel to be predicted based on customer evaluation data, extract the distance between regional activities and attractions and the hotel to be predicted based on regional activity data and attraction data, and calculate the impact of regional activities and attractions on the hotel to be predicted based on the distance;

[0082] The specific steps of calculating the impact of regional activities on the predicted hotels are as follows:

[0083] Calculate the distance from each regional activity to the hotel:

[0084] ;

[0085] Where: R represents the radius of the earth; represents the latitude difference between hotel and activity j; represents the longitude difference between hotel and activity j; , Respectively represent the latitude and longitude of the hotel; , denote the latitude and longitude of activity j, respectively; represents the distance between hotel and activity j;

[0086] Calculate the impact of each activity:

[0087] ;

[0088] ;

[0089] ;

[0090] Where: represents the number of participants or scale of activity j; Indicates the number of days or time that activity j lasts; and represents the weight coefficient; represents the impact value of activity j on demand; represents the intensity of activity j; represents the geographical distance between hotel and activity j; The maximum distance parameter representing distance attenuation; represents the total impact of all regional activities on demand; J represents the total number of activities.

[0091] The steps for calculating the influence of attractions on the predicted hotels are as follows:

[0092] Calculate the geographical distance between each attraction and the hotel:

[0093] ;

[0094] Where: R represents the radius of the earth; represents the latitude difference between the hotel and the attraction k; represents the longitude difference between the hotel and the attraction k; , Respectively represent the latitude and longitude of the hotel; , Respectively represent the latitude and longitude of scenic spot k; represents the distance between the hotel and attraction k;

[0095] Calculate the impact of each attraction:

[0096] ;

[0097] ;

[0098] ;

[0099] Where: represents the average tourist flow of scenic spot k in the past n years; represents the weight coefficient; represents the impact of attraction k on hotel demand; represents the intensity of scenic spot k; represents the geographical distance between the hotel and attraction k; The maximum distance parameter representing distance attenuation; Represents the total impact of all attractions on demand.

[0100] In this embodiment, by calculating the impact of regional activities and attractions in detail, the impact of these factors on hotel demand can be more accurately evaluated. This analysis method based on geographical distance and activity intensity can help hotel managers better understand market dynamics, thereby formulating more accurate marketing strategies and pricing plans to enhance the competitiveness of hotels.

[0101] Deep learning demand forecasting: Based on the pre-processed holiday data, historical check-in data and feature-extracted data as input to the deep learning model, the deep learning model is trained, and based on real-time demand forecasting, the real-time data is input into the trained deep learning model to obtain the hotel's demand forecast results;

[0102] The structure of the deep learning model is as follows:

[0103] Input layer: It includes four input channels, which respectively input holiday data, historical occupancy data, regional activity impact and scenic spot impact, and praise rate;

[0104] LSTM layer: LSTM is used to process the time dependency of holiday data, historical check-in data, and favorable review rates, and to extract time series patterns;

[0105] The specific configuration of the LSTM layer is as follows:

[0106] Input: The three input channels (holiday data, historical occupancy data, and favorable review rate) are input into separate LSTM layers.

[0107] Layer configuration:

[0108] Input dimension: Assuming that the input data at each time step is d-dimensional, the input dimension is d;

[0109] The output dimension (hidden state dimension) of the LSTM layer: h, that is, the hidden state dimension of each time step in the LSTM, is a hyperparameter selected by cross-validation.

[0110] Number of LSTM layers: There can be one or more LSTM layers to increase the expressiveness and complexity of the model.

[0111] The LSTM layer captures the long-term dependencies of the data through gating mechanisms (forget gate, input gate, and output gate), especially the trends and patterns of time series features such as holidays, historical occupancy rates, and positive review rates.

[0112] Convolutional layer: Apply convolutional layers to the regional activity impact and scenic spot impact data to extract local features and capture information at different scales through multiple convolution kernels;

[0113] The configuration of the convolutional layer is as follows:

[0114] Input: Regional activities and attractions impact data, the dimension of each data is usually a two-dimensional matrix (number of time steps T × number of features);

[0115] Convolutional layer configuration:

[0116] Number of convolution kernels: Set k convolution kernels to extract local features of different scales.

[0117] Convolution kernel size: Set the size of the convolution kernel to 1×m, where m is the feature dimension.

[0118] Activation function: The output of the convolutional layer is nonlinearly mapped through the ReLU activation function.

[0119] The convolutional layer can effectively extract the local impact of regional and scenic spot activities on demand and capture the impact patterns of different activities in different time periods.

[0120] Feature fusion layer: The features extracted by the LSTM layer and the features extracted by the convolution layer are combined into feature vectors through splicing operations to obtain fused features;

[0121] Operation: The outputs of the LSTM and convolutional layers are combined through concatenation. For example, the output feature of the LSTM layer is , the feature of the convolutional layer output is , then the fused features are: ; The fused feature vector contains the comprehensive information of time series features and local features.

[0122] Fully connected layer: Based on the fusion features, feature mapping and fusion are performed through multiple fully connected layers, and each fully connected layer uses the ReLU activation function;

[0123] After feature fusion, it enters multiple fully connected layers for further feature mapping. Each fully connected layer uses the ReLU activation function;

[0124] Fully connected layer configuration:

[0125] Input dimension: equal to the fused feature dimension ;

[0126] Output dimension: dimension 64;

[0127] Activation function: Use ReLU activation function for nonlinear mapping;

[0128] Through multiple fully connected layers, the model is able to learn higher-order abstract representations from the fused features.

[0129] Output layer: includes a fully connected layer to output the prediction results, using a linear activation function.

[0130] The loss function of the deep learning model is as follows:

[0131] ;

[0132] Where: represents the hotel demand forecast value given by the deep learning model; represents the actual demand at time step t; N represents the total number of samples; represents a hyperparameter used to control the penalty intensity of the weighted error term; Represents the weight coefficient at time step t, which is dynamically adjusted according to time; represents the hyperparameter of the balanced loss term;

[0133] ;

[0134] Where: Indicates the adjustment weight of the impact of holidays; Indicates the impact adjustment weights of working days and non-working days; Indicates the historical demand trend adjustment weight; Represents a holiday marking function. If it is a holiday, If it is not a holiday, ; Represents the weekday marking function. If it is a weekday, then If it is a non-working day, ; It represents the average demand value before time step tk.

[0135] In this embodiment, the advantages of LSTM and convolutional layers are combined to effectively extract time series data and local features. This multi-level feature extraction method can improve the prediction accuracy of the model, thereby helping hotels make more scientific decisions in a complex market environment; and can fully consider the impact between activities and attractions through the convolutional layer. Since some attractions also have activity problems, which will lead to an increase in the number of tourists, we fully consider this impact through the convolutional layer. When processing regional activity impact and attraction impact data, the convolutional layer can extract local features through the convolution kernel to capture information of different scales, especially the spatial correlation of geography and activities. The convolutional layer can effectively identify the distance and influence pattern between activities and attractions and hotels, providing more valuable features for hotel demand forecasting.

[0136] In this embodiment, the loss function takes into account the impact of holidays, working days and historical demand trends, so that the model can dynamically adjust the weights during the training process. This flexibility not only improves the adaptability of the model, but also makes the prediction results more in line with actual demand changes, thereby improving the hotel's operational efficiency.

[0137] Competitor determination: construct distance fluctuation thresholds and favorable comment rate fluctuation thresholds; filter out hotels within the threshold range based on the distance between regional activities and attractions and the hotel to be predicted and the distance fluctuation thresholds to obtain preliminary selected competitors; then filter out the preliminary selected competitors based on favorable comment rates and favorable comment rate fluctuation thresholds to obtain final competitor information;

[0138] The competitor determination comprises the following steps:

[0139] Filter hotels that meet the distance fluctuation threshold:

[0140] ;

[0141] Where: Indicates the distance between the hotel to be predicted and the scenic spot; Indicates the distance fluctuation threshold; Indicates the distance from the candidate competitor to the attraction;

[0142] ;

[0143] Where: Indicates the distance between the hotel to be predicted and the activity area; represents the distance from the candidate competitor to the active area;

[0144] Filter out hotels that meet the load distance fluctuation threshold and obtain preliminary selected competitors;

[0145] The finalized competitors are screened based on the positive review rate and the positive review rate fluctuation threshold:

[0146] ;

[0147] Where: Indicates the praise rate of the hotel to be predicted; Indicates the fluctuation threshold of the favorable comment rate; Indicates the favorable rating of candidate competitors;

[0148] Competitors that meet the fluctuation threshold of favorable review rate are screened out from the initially screened competitors and are selected as the final competitors.

[0149] In this embodiment, when screening competitors, the above three conditions should be met at the same time, so as to screen out the most similar hotels; of course, the above embodiment is not fixed, and can be adjusted based on the actual application scenario. For example, any two or one of the above conditions can be satisfied as the final competitor; and in this embodiment, the above three conditions are satisfied at the same time, so it can be adjusted based on the actual situation.

[0150] In this embodiment, by setting the distance fluctuation threshold and the favorable review rate fluctuation threshold, it is possible to accurately identify competitors that are similar to the hotel to be predicted in terms of geographical location and customer reviews. This method not only improves the accuracy of competitor selection, but also effectively reduces the complexity of market analysis, enabling hotels to better grasp market dynamics and formulate corresponding marketing strategies, thereby improving customer satisfaction and occupancy rate, and promoting the sustainable development of hotels.

[0151] Inventory management and scheduling: Based on the pricing information corresponding to the finalized competitor information and demand forecasts, reinforcement learning is used to optimize the hotel's inventory management and pricing strategy.

[0152] The specific details of reinforcement learning in inventory management and scheduling are as follows:

[0153] Optimization goal: Maximize the hotel's total revenue, which is achieved by optimizing pricing and inventory management strategies:

[0154] ;

[0155] Where: Represents the room rate of the hotel to be predicted at time t; represents the inventory of the hotel to be predicted at time t; represents the unit room cost; represents the profit at time t;

[0156] Adopt reinforcement learning framework to make optimal decisions, including state representation, action representation, reward function and Q learning update formula;

[0157] The state representation includes: current inventory levels, current demand forecasts, competitor pricing, and current historical revenue;

[0158] The actions are expressed as follows:

[0159] ;

[0160] in, To indicate an action;

[0161] Reward function:

[0162] ;

[0163] Where: represents the value of the reward function; represents the penalty factor, which is used to control the impact of competitor price differences on hotel revenue; represents the house price of competitor i at time t;

[0164] Q function update formula:

[0165] ;

[0166] Where: represents the updated Q value; represents the Q value before updating; Indicates that from the next state Start and choose the Q value of the optimal action; represents the learning rate; represents the discount factor; represents the optimal action.

[0167] The specific steps to use reinforcement learning to optimize the hotel's inventory management and pricing strategy are as follows:

[0168] Initialization: At the beginning of learning, the Q function is initialized and the Q value is initialized to zero or a small random value. The Q function is gradually updated through learning to reflect the expected long-term benefits of each state-action pair;

[0169] Parameter initialization: learning rate , Discount Factor , Penalty Factor wait;

[0170] Get input data: collect state information about the current time t ,in represents the inventory level at the current time t; Represents the demand for deep learning model prediction; represents the house price set by competitor i at time t; Indicates the hotel's revenue data for the past period of time;

[0171] Action selection: At each time t, choose an action ; Based on the current strategy selection, the ε-greedy strategy is adopted, which includes the following steps:

[0172] Exploration: Randomly select an action with a certain probability ε to explore new strategies;

[0173] Utilization: Select the action with the largest current Q value with a probability of 1-ε, that is, select the pricing and inventory combination that is currently considered optimal;

[0174] Perform actions and get rewards: Perform the selected actions , and calculate the reward according to the reward function ;

[0175] Update Q value: adjust Q based on the Q value update formula;

[0176] Repeat training: Repeat training until the Q value converges or the preset number of training times is reached;

[0177] Final strategy: Once training is completed, the optimal strategy is generated by the learned Q function: , select the corresponding optimal action , that is, choosing the pricing and inventory mix that maximizes long-term returns.

[0178] In this embodiment, by analyzing the current inventory level, demand forecast and competitor pricing in real time, the room rate and inventory can be adjusted dynamically to ensure the optimization of revenue in different market environments. The application of the reinforcement learning framework enables the hotel to quickly adapt and make the best decision in the ever-changing market conditions, thus improving operational efficiency and economic benefits.

[0179] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. An intelligent method for hotel real-time demand forecasting and inventory management, characterized in that: The following steps are involved: Data acquisition and preprocessing: collect holiday data, historical check-in data, customer evaluation data, regional activity data, and scenic spot data, and preprocess the collected data; Feature extraction: Extract the praise rate of the hotel to be predicted based on customer evaluation data, extract the distance between regional activities and attractions and the hotel to be predicted based on regional activity data and attraction data, and calculate the impact of regional activities and attractions on the hotel to be predicted based on the distance; Deep learning demand forecasting: Based on the pre-processed holiday data, historical check-in data and feature-extracted data as input to the deep learning model, the deep learning model is trained, and based on real-time forecast demand, the real-time data is input into the trained deep learning model to obtain the hotel's demand forecast results; Competitor determination: construct distance fluctuation thresholds and favorable review rate fluctuation thresholds; filter out hotels within the threshold range based on the distance between regional activities and attractions and the hotel to be predicted and the distance fluctuation thresholds, and obtain preliminary selected competitors; Then, based on the favorable review rate and favorable review rate fluctuation threshold, the competitors that have been initially screened are screened to obtain the final competitor information; The competitor determination comprises the following steps: Filter hotels that meet the distance fluctuation threshold: Where: Indicates the distance between the hotel to be predicted and the scenic spot; Indicates the distance fluctuation threshold; Indicates the distance from the candidate competitor to the attraction; Where: Indicates the distance between the hotel to be predicted and the activity area; represents the distance from the candidate competitor to the active area; Filter out hotels that meet the distance fluctuation threshold and obtain preliminary selected competitors; The finalized competitors are screened based on the positive review rate and the positive review rate fluctuation threshold: Where: Indicates the praise rate of the hotel to be predicted; Indicates the fluctuation threshold of the favorable comment rate; Indicates the favorable review rate of candidate competitors; competitors that meet the favorable review rate fluctuation threshold are screened out from the initially screened competitors and are used as the final competitors; Inventory management and scheduling: Based on the pricing information corresponding to the finalized competitor information and demand forecasts, reinforcement learning is used to optimize the hotel's inventory management and pricing strategy.

2. The intelligent method for hotel real-time demand forecasting and inventory management according to claim 1, characterized in that: The specific steps of calculating the impact of regional activities on the predicted hotels are as follows: Calculate the distance from each regional activity to the hotel: ; Where: R represents the radius of the earth; represents the latitude difference between hotel and activity j; represents the longitude difference between hotel and activity j; , Respectively represent the latitude and longitude of the hotel; , denote the latitude and longitude of activity j, respectively; represents the distance between hotel and activity j; Calculate the impact of each activity: ; ; ; Where: represents the number of participants or scale of activity j; Indicates the number of days or time that activity j lasts; and represents the weight coefficient; represents the impact value of activity j on demand; represents the intensity of activity j; represents the geographical distance between hotel and activity j; The maximum distance parameter representing distance attenuation; represents the total impact of all regional activities on demand; J represents the total number of activities.

3. The intelligent method for hotel real-time demand forecasting and inventory management according to claim 1, characterized in that: The steps to calculate the impact of attractions on the predicted hotels are as follows: Calculate the geographical distance between each attraction and the hotel: ; Where: R represents the radius of the earth; represents the latitude difference between the hotel and the attraction k; represents the longitude difference between the hotel and the attraction k; , Respectively represent the latitude and longitude of the hotel; , Respectively represent the latitude and longitude of scenic spot k; represents the distance between the hotel and attraction k; Calculate the impact of each attraction: ; ; ; Where: represents the average tourist flow of scenic spot k in the past n years; represents the weight coefficient; represents the impact of attraction k on hotel demand; represents the intensity of scenic spot k; represents the geographical distance between the hotel and attraction k; The maximum distance parameter representing distance attenuation; represents the total impact of all attractions on demand; K represents the number of attractions.

4. The intelligent method for hotel real-time demand forecasting and inventory management according to claim 1, characterized in that: The structure of the deep learning model is as follows: Input layer: It includes four input channels, which respectively input holiday data, historical occupancy data, regional activity impact and scenic spot impact, and praise rate; LSTM layer: LSTM is used to process the time dependency of holiday data, historical check-in data, and favorable review rates, and to extract time series patterns; Convolutional layer: Apply convolutional layers to the regional activity impact and scenic spot impact data to extract local features and capture information at different scales through multiple convolution kernels; Feature fusion layer: The features extracted by the LSTM layer and the features extracted by the convolution layer are combined into feature vectors through splicing operations to obtain fused features; Fully connected layer: Based on the fusion features, feature mapping and fusion are performed through multiple fully connected layers, and each fully connected layer uses the ReLU activation function; Output layer: includes a fully connected layer to output the prediction results, using a linear activation function.

5. The intelligent method for hotel real-time demand forecasting and inventory management according to claim 4 is characterized in that: The loss function of the deep learning model is as follows: ; Where: represents the hotel demand forecast value given by the deep learning model; represents the actual demand at time step t; N represents the total number of samples; represents a hyperparameter used to control the penalty intensity of the weighted error term; Represents the weight coefficient at time step t, which is dynamically adjusted according to time; represents the hyperparameter of the balanced loss term; ; Where: Indicates the adjustment weight of the impact of holidays; Indicates the impact adjustment weights of working days and non-working days; Indicates the historical demand trend adjustment weight; Represents a holiday marking function. If it is a holiday, If it is not a holiday, ; Represents the weekday marking function. If it is a weekday, then ; If it is a non-working day, ; It represents the average demand value before time step tk.

6. The intelligent method for hotel real-time demand forecasting and inventory management according to claim 1, characterized in that: The specific details of reinforcement learning in inventory management and scheduling are as follows: Optimization goal: Maximize the hotel's total revenue, which is achieved by optimizing pricing and inventory management strategies: ; Where: represents the room rate of the hotel to be predicted at time t; represents the inventory of the hotel to be predicted at time t; represents the unit room cost; represents the profit at time t; Adopt reinforcement learning framework to make optimal decisions, including state representation, action representation, reward function and Q learning update formula; The state representation includes: current inventory levels, current demand forecasts, competitor pricing, and current historical revenue; The actions are expressed as follows: ;in, To indicate an action; Reward function: ; Where: represents the value of the reward function; represents the weight coefficient, which is used to control the impact of competitor price differences on hotel revenue; represents the house price of competitor i at time t; Q function update formula: ; Where: represents the updated Q value; represents the Q value before updating; Indicates that from the next state Start and choose the Q value of the optimal action; represents the learning rate; represents the discount factor; represents the optimal action.

Images