Multi-channel room reservation method and system for hotel

Abstract

The invention provides a multi-channel room reservation method and system for a hotel, and the method comprises the steps: obtaining the room reservation historical data of different reservation channels, the room reservation historical data comprises the room price level, price fluctuation and room type inventory data, segmenting the data according to the customer reservation behavior, the check-in duration, and the average room price key attributes of historical orders, and storing the segmented data in a database; obtaining a room type data set; removing abnormal values and missing values in the room type data set, extracting customer age groups, preferred room types, price sensitivity and historical check-in frequency as key features, and performing normalization processing on the key features to form standardized feature vectors for describing demand features and price sensitivity of each subdivided customer group; and in combination with the real-time room state data of the hotel, the room type demand quantity prediction result and the price sensitivity threshold, dynamically shutting down sales of part of room types.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a multi-channel room booking method and system for a hotel. Background Art

[0002] In the hotel's multi-channel booking management, there is a problem of inconsistent customer behavior data from different channels. How to standardize these multi-dimensional data and accurately extract customer demand characteristics and price sensitivity is an important issue. Hotels need to classify customers based on these characteristics and predict the room type needs of various customer groups in different time periods in the future. However, how to combine these prediction results with real-time room status management to maximize revenue while avoiding the loss of potential customers due to incorrect room closing operations is still a difficulty. Building a dynamic pricing mechanism so that price adjustments can meet customer price sensitivity, while forming a closed-loop feedback with demand forecasts to achieve continuous optimization, is an important challenge to improve hotel management efficiency. Summary of the invention

[0003] To solve the above problems, the present invention provides a multi-channel room booking method for a hotel, which mainly includes:

[0004] Obtain historical booking data from different booking channels. The historical booking data includes room rate levels, price fluctuations, and room inventory data. The data is segmented based on key attributes such as customer booking behavior, length of stay, and average room rate of historical orders to obtain a room data set.

[0005] Remove outliers and missing values ​​from the room type data set, extract customer age, preferred room type, price sensitivity, and historical check-in frequency as key features, normalize the key features, and form a standardized feature vector that describes the demand characteristics and price sensitivity of each customer segment;

[0006] Based on the standardized feature vectors describing the demand characteristics and price sensitivity of each customer segment, a clustering algorithm is used to divide customers into different demand preference groups. For each demand preference group, a standardized feature vector dataset and a price sensitivity threshold dataset of room type demand under different channels and time granularity are established.

[0007] Based on the room demand feature vector dataset of each demand preference group, a time series prediction model is used to dynamically adjust the time granularity of the training data to predict the room demand of different channels and different groups in the future. At the same time, the upper limit of the price adjustment range for different groups is determined based on the price sensitivity threshold dataset.

[0008] Combine the hotel's real-time room status data, room demand forecast results and price sensitivity thresholds to dynamically adjust prices and shut down sales of some room types;

[0009] Based on the price information and price sensitivity threshold after dynamic price adjustment, combined with the key attributes of historical check-in frequency, evaluation score, and repurchase rate of different demand preference groups, a collaborative filtering algorithm is used to match the best room type combination from the hotel room type library, and personalized room type recommendations are pushed to target customers through different room status release channels and different time nodes;

[0010] If a customer books a preferred room type and the booking price is within its price sensitivity threshold, the booking of that room type will be locked, and the room status association strategy for that room type will be dynamically adjusted based on the customer's booking time and stay duration attributes. If a customer books a non-preferred room type, or the booking price exceeds its price sensitivity threshold, an alternative room type will be dynamically recommended within the price sensitivity threshold based on the customer's demand characteristics and the hotel's room inventory.

[0011] The present invention also provides a multi-channel room reservation system for a hotel, comprising:

[0012] The data collection and preprocessing module is used to obtain historical booking data from different booking channels. The historical booking data includes room rate levels, price fluctuations, and room inventory data. The data is segmented according to the key attributes of customer booking behavior, length of stay, and average room rate of historical orders to obtain a room data set. Outliers and missing values ​​in the room data set are removed, and customer age, preferred room type, price sensitivity, and historical check-in frequency are extracted as key features. The key features are normalized to form a standardized feature vector that describes the demand characteristics and price sensitivity of each customer segment.

[0013] The customer clustering module is used to classify customers into different demand preference groups based on the standardized feature vectors that describe the demand characteristics and price sensitivity of each customer segment, and to establish a standardized feature vector dataset and price sensitivity threshold dataset for each demand preference group under different channels and time granularities;

[0014] The demand forecasting module is used to predict the future demand for housing types in different channels and for different groups based on the housing demand feature vector dataset of each demand preference group, using a time series forecasting model and dynamically adjusting the time granularity of the training data. At the same time, based on the price sensitivity threshold dataset, the upper limit of the price adjustment range for different groups is determined;

[0015] Dynamic pricing module, which is used to dynamically adjust prices and stop the sale of some room types by combining the hotel's real-time room status data, room demand forecast results and price sensitivity thresholds;

[0016] The personalized recommendation module is used to match the optimal room type combination from the hotel room type library based on the price information and price sensitivity threshold after dynamic price adjustment, combined with the key attributes of historical check-in frequency, evaluation score, and repurchase rate of different demand preference groups, using a collaborative filtering algorithm, and push personalized room type recommendations to target customers through different room status release channels and at different time nodes.

[0017] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0018] The present invention discloses a multi-channel room booking method. Aiming at the problems of inaccurate demand prediction and inflexible pricing strategy in hotel room booking, the present invention obtains the room price level, price fluctuation and room inventory data under multi-granularity time dimensions, and combines key attributes such as customer booking behavior, stay duration, and average room price of historical orders to segment and clean the data, extract customer age group, preferred room type, price sensitivity, and historical stay frequency as key features, and adopts a clustering algorithm to divide customers into different demand preference groups. On this basis, the present invention establishes a standardized feature vector describing the demand characteristics and price sensitivity of each segmented customer group, and adopts a time series prediction model to predict the room demand of different channels and different groups in the future, and at the same time determines the upper limit of the price adjustment range of different groups. Finally, the present invention dynamically adjusts the room price and shuts down the room sales according to the room demand prediction result and the price sensitivity threshold, combined with the hotel's real-time room status data, and adopts a collaborative filtering algorithm to push personalized room recommendations to target customers. The present invention can improve the accuracy of hotel room demand prediction, realize dynamic pricing and personalized recommendations, thereby optimizing hotel room inventory management, and improving hotel operation efficiency and revenue level. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flowchart of a multi-channel hotel reservation method.

[0020] Figure 2 The present invention is a block diagram of a multi-channel room reservation system for hotels. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] like Figure 1 In this embodiment, a multi-channel hotel reservation method may specifically include:

[0023] Step S101, obtaining historical booking data of different booking channels, the historical booking data including room rate level, price fluctuation and room type inventory data, and segmenting the data according to key attributes of customer booking behavior, length of stay and average room rate of historical orders to obtain a room type data set.

[0024] The method comprises the following steps: obtaining raw data carrying a booking channel identifier, wherein the raw data is sent from different booking channels; preprocessing the raw data to obtain a feature matrix, wherein the feature matrix includes features of multiple time granularities; clustering the feature matrix to obtain clustering results of different room types; constructing a room type sub-dataset for each clustering result; executing a time series decomposition instruction on the room type sub-dataset, wherein the time series decomposition instruction is used to separate trend, seasonality and random fluctuation components; calculating a price fluctuation index according to the time series decomposition result, wherein the price fluctuation index includes multiple time granularities; and merging the price fluctuation index with historical data to obtain a room type data set.

[0025] Specifically, we obtain the original data of different booking channels, build a separate data set for each booking channel, use data cleaning methods to remove outliers and missing values, unify the data formats of different channels through data standardization, and aggregate the data at multiple time granularities such as hours, days, weeks, and months according to the timestamp to obtain multi-dimensional time series data including housing price levels, price fluctuations, and room inventory. We perform feature engineering on the aggregated multi-dimensional time series data, use the groupby and agg functions of the pandas library to calculate the average booking lead time, booking frequency, and length of stay for each customer, extract customer booking behavior features, calculate the average length of stay for each customer, and count the average room price of historical orders. For categorical features such as room types, we use the get_dummies function of pandas for one-hot encoding, and use the PCA function of the sklearn library for principal component analysis, set the 95% variance as the threshold, and generate a feature matrix containing multiple time granularities. Based on the generated feature matrix, the KMeans function of the sklearn library is used to cluster the room types. The optimal number of clusters is selected by the silhouette coefficient. The Euclidean distance from the sample to each cluster center is calculated. The iterative optimization is performed until the center point is stable to obtain the clustering results of different room types. For each clustering result, a sub-dataset containing the characteristics of this type of room type is constructed separately. The seasonal_decompose function of the statsmodels library is used to decompose the time series of each room type sub-dataset to separate the trend, seasonality and random fluctuation components. The SimpleExpSmoothing function is used to predict the trend component. The prediction period is set to 30 days. The price standard deviation of the past 30 days is calculated as the price fluctuation index. The price fluctuation index at each time granularity is calculated by combining the prediction results and historical data. The merge function of pandas is used to merge all the processing results to generate a final room type data set containing multiple time granularities. When obtaining booking channel data, the original data is collected from multiple online travel platforms and hotel direct sales channels, including booking date, check-in date, check-out date, room type, price and other information. During data cleaning, records with negative or abnormally high prices are detected and deleted, and missing room type information is filled. Data standardization unifies the date format of different channels into YYYY-MM-DD, and the price unit into RMB. When aggregating time, the average room rate and remaining inventory are calculated by hour, the booking volume and total revenue are summarized by day, and the occupancy rate is calculated by week and month. In the feature engineering stage, the average booking lead time (the mean of the difference between the booking date and the check-in date), booking frequency (number of bookings per month), and average stay time (the mean of the difference between the check-out date and the check-in date) of each customer are calculated. One-hot encoding is performed on room types, and "standard room" and "double bed room" are converted into binary features. Principal component analysis retains 95% of the variance, which may reduce the original 20+ features to about 10.When clustering rooms, the K-means algorithm is used to calculate the silhouette coefficients under different K values ​​from 2 to 10, and the K value with the largest silhouette coefficient is selected as the optimal number of clusters. Assuming that the optimal number of clusters is 4, the clustering results of four types of rooms, namely economy, business, luxury and suite, are obtained. Time series decomposition decomposes the average daily price of each room type into trend, seasonality and random components. For example, the trend of economy rooms may show a slow increase, and the seasonality is a 15% increase in weekend prices. The SimpleExpSmoothing model is used for prediction, with α set to 0.3, to predict the trend for the next 30 days. The price fluctuation index is calculated as the standard deviation of the price in the past 30 days. For example, the standard deviation of a certain type of room is 50 yuan, indicating that the price fluctuation is large. The multi-time granularity room data set finally generated contains the average price, predicted price, price fluctuation index, inventory and other information of each room type at the hour, day, week and month scales.

[0026] Step S102, remove outliers and missing values ​​in the room type data set, extract customer age group, preferred room type, price sensitivity, and historical check-in frequency as key features, normalize the key features, and form a standardized feature vector that describes the demand characteristics and price sensitivity of each customer segment group.

[0027] The room type data set obtained is preprocessed, including deleting outliers and filling missing values, to obtain a complete data set. Customer age, preferred room type, price sensitivity, and historical check-in frequency are extracted from the complete data set as key features. Price sensitivity is determined by calculating the standard deviation of the customer's booking price in the past predefined time period and dividing it by the average price. The extracted key features are normalized, and the minimum-maximum scaling method is used to map the feature values ​​to the predefined interval for continuous numerical features. The processed features are merged to obtain a standardized feature vector that describes the demand characteristics and price sensitivity of each customer segment.

[0028] Specifically, after obtaining the multi-time granularity housing data set, the interquartile range method is used to detect outliers, and the interquartile range of each feature is calculated. If the data point exceeds the upper quartile plus 1.5 times the interquartile range or is lower than the lower quartile minus 1.5 times the interquartile range, it is determined to be an outlier and the record is deleted to obtain the cleaned data set. For the missing values ​​in the cleaned data set, the missing rate of each feature is calculated, and different processing methods are selected according to the data type: for numerical features, if the missing rate is less than 5%, the median is used for filling, and if the missing rate is between 5% and 20%, the MICE algorithm in the multiple interpolation method is used for filling; for categorical features, the mode is used for filling. If the missing rate exceeds 20%, the feature is deleted to obtain a complete data set. From the complete data set, we extract customer age, preferred room type, price sensitivity, and historical check-in frequency as key features. We determine price sensitivity by calculating the standard deviation of customer booking prices in the past year and dividing it by the average price. We calculate the proportion of bookings for each room type and select the room type with the highest proportion to determine the preferred room type. We calculate the age of the customer based on his / her date of birth and divide it into predefined age groups of 18-25, 26-35, 36-45, etc. We count the number of check-ins of the customer in the past year to get the historical check-in frequency. We normalize the extracted key features. For continuous numerical features such as price sensitivity and historical check-in frequency, we use the minimum-maximum scaling method to map the feature values ​​to the [0,1] interval. The calculation method is (x-min(x)) / (max(x)-min(x)), where x is the original feature value, min(x) and max(x) are the minimum and maximum values ​​of the feature respectively. For discrete features such as age group and preferred room type, we use one-hot encoding to convert them into binary feature vectors. Finally, we merge all the processed features to obtain standardized feature vectors that describe the demand characteristics and price sensitivity of each customer segment. When processing a multi-time granularity room type dataset, first apply the interquartile range method to each feature to detect outliers. Taking the room price as an example, the first quartile Q1 is calculated to be 300 yuan, the third quartile Q3 is 800 yuan, and the interquartile range IQR is 500 yuan. Set the outlier thresholds to Q1-1.5IQR (-450 yuan) and Q3+1.5IQR (1550 yuan), and delete records with room prices lower than -450 yuan or higher than 1550 yuan. For missing values, calculate the missing rate of each feature. For example, if the missing rate of customer age is 3%, the median of 35 years old is used for filling; if the missing rate of room type preference is 15%, the MICE algorithm is used for multiple interpolation. When extracting key features, calculate price sensitivity; if the standard deviation of a customer's booking price in the past year is 200 yuan and the average price is 1000 yuan, then its price sensitivity is 0.2. The preferred room type is determined by counting the proportion of booking times. For example, if a customer's booking proportions for a standard room, a king-size bed room, and a suite are 60%, 30%, and 10% respectively, the preferred room type is a standard room.The customer's age is calculated based on his / her date of birth. For example, a customer born in 1990 is 34 years old and is classified as 26-35 years old. The historical check-in frequency directly counts the number of check-ins in the past year, such as 12 times. When normalizing, the price sensitivity is scaled min-max. Assuming the minimum value is 0.1 and the maximum value is 0.5, the original value of 0.2 is normalized to 0.25. The age group and preferred room type are converted through one-hot encoding, such as the 26-35 age group is converted to [0,1,0,0], and the preferred standard room is converted to [1,0,0]. Finally, all processed features are merged to form a standardized feature vector to describe the demand characteristics and price sensitivity of the segmented customer groups.

[0029] Step S103, based on the standardized feature vectors describing the demand characteristics and price sensitivity of each customer segment group, a clustering algorithm is used to divide the customers into different demand preference groups, and for each demand preference group, a standardized feature vector dataset and a price sensitivity threshold dataset of room type demand under different channels and different time granularities are established.

[0030] Receive a standardized feature vector carrying customer demand characteristics and price sensitivity; use a clustering algorithm to group customers according to the standardized feature vector, and the clustering algorithm detects and processes outliers through the interquartile range method; obtain silhouette coefficients with different K values, determine the optimal number of clusters K, and obtain customer segmentation results; for each cluster group, obtain order information of different booking channels from a historical order database, and construct a feature vector reflecting the group's preference for different channels; aggregate the historical order data according to time granularity to obtain a standardized vector reflecting the group's demand characteristics at different time granularities; use a random forest regression model, input the feature vector and the standardized vector, and calculate the price sensitivity threshold of each demand preference group at different channels and time granularities.

[0031] Specifically, the K-means clustering algorithm is used to cluster customers based on the standardized feature vectors that describe the demand characteristics and price sensitivity of each customer segment. First, the interquartile range method is used to detect outliers, and the interquartile range IQR of each feature is calculated. Data points outside the range of Q1-1.5IQR and Q3+1.5IQR are regarded as outliers and processed by median replacement. Then, the optimal number of clusters K is determined by calculating the silhouette coefficient of different K values, and a random seed is set to ensure that the results are repeatable. The iterative calculation is performed until the cluster center converges or the maximum number of iterations is reached to obtain the customer segmentation result. For each demand preference group obtained by clustering, the historical order data of the group in different booking channels, such as official website, OTA platform, and telephone booking, is extracted, and the order proportion, average booking lead time, and average stay time of each channel are counted. These features are normalized by Min-Max, the values ​​are scaled to the interval [0,1], and then the normalized features are spliced ​​into a vector to construct a feature vector reflecting the preferences of the group in different channels. For each demand preference group, based on historical order data, order information is aggregated at different time granularities such as hours, days, weeks, and months, and the average order volume, average house price, and room type distribution at each time granularity are calculated. These features are also normalized by Min-Max and spliced ​​into vectors to construct standardized vectors that reflect the demand characteristics of the group at different time granularities. The price sensitivity threshold is calculated by combining the feature vectors of each demand preference group in different channels and different time granularities. The random forest regression model is used, and the input features include price change percentage, time factors such as weekends / weekdays, holidays, seasonal factors, etc., and the output is the percentage change in order volume. By setting a threshold for order volume change such as -10%, the critical point of price change that causes a significant change in order volume is determined, and this critical point is set as the price sensitivity threshold of the group at a specific channel and time granularity, and finally a complete data set is formed. When processing customer group data, the interquartile range method is first applied to the standardized feature vector to detect outliers. Taking price sensitivity as an example, Q1 is calculated to be 0.2, Q3 is 0.8, and IQR is 0.6. Values ​​less than -0.7 or greater than 1.7 are considered abnormal and replaced with the median 0.5. Then the K-means algorithm is used to cluster customers. By calculating the silhouette coefficient of K values ​​from 2 to 10, it is found that the silhouette coefficient is the highest when K=5, which is 0.68. The optimal number of clusters is determined to be 5. For each cluster group, order data from different channels are extracted. For example, the proportion of orders for a group on the official website, OTA platform, and telephone booking is 50%, 30%, and 20% respectively, the average booking lead time is 7 days, 5 days, and 3 days, and the average length of stay is 3 nights, 2 nights, and 4 nights. These features are normalized by Min-Max to [1,0.5,0], [1,0.5,0], [0.5,0,1], and spliced ​​into a feature vector of [1,0.5,0,1,0.5,0,0.5,0,1].Similar processing is performed on the time granularity data. For example, the hourly statistics show that the average number of orders at 1 a.m. is 10, the average room price is 200 yuan, and the proportion of standard rooms is 70%. These data are also normalized and spliced ​​into the feature vector. Finally, the random forest regression model is used to analyze price sensitivity. The input includes factors such as a 10% price increase, weekends, and summer vacation. The model outputs a 15% decrease in order volume. Set -10% as the order volume change threshold, and reversely calculate that the price increase that causes the order volume to decrease by 10% is 6.7%, which is set as the price sensitivity threshold for this group under specific conditions. Repeat this process to finally obtain a complete group feature and price sensitivity data set.

[0032] Step S104: For each demand preference group's housing demand feature vector dataset, a time series prediction model is used to dynamically adjust the time granularity of the training data to predict the future demand for housing types in different channels and for different groups. At the same time, the upper limit of the price adjustment range for different groups is determined based on the price sensitivity threshold dataset.

[0033] Obtain time series data, and adjust the time granularity of the training data according to the historical span of the time series data and the prediction target; if the prediction target is short-term, use hourly data; if the prediction target is medium-term, use daily data; if the prediction target is long-term, use weekly or monthly data. Use the adjusted time granularity data to predict the demand for housing types in different channels for the demand preference groups; determine whether the average absolute percentage error of the prediction results is lower than the preset threshold. Obtain price sensitivity data from the price sensitivity threshold data set; calculate the price sensitivity distribution under different channels and time granularities for the demand preference groups; take the preset quantile of the price sensitivity distribution as the upper limit of the price adjustment range. Use the linear programming method to optimize the price adjustment strategy; take maximizing the expected return as the objective function, and use the upper limit of the price adjustment range as a constraint to obtain the optimal price adjustment plan.

[0034] Specifically, for the room demand feature vector data set of each demand preference group, the ARIMA and Prophet time series forecasting models are used. The value range of the ARIMA model parameters p, d, and q is determined by calculating the autocorrelation function and the partial autocorrelation function. The grid search method is used to traverse different parameter combinations, and the parameter combination with the smallest AIC information criterion is selected as the optimal model parameter. At the same time, the cross-validation method is used to evaluate the performance of the two models, and the model with better performance is selected as the room demand forecasting model. According to the time span of historical data and the forecasting target, the time granularity of the training data is dynamically adjusted. For short-term forecasts of 1-7 days, hourly data is used, for medium-term forecasts of 8-30 days, daily data is used, and for long-term forecasts of 31-90 days, weekly or monthly data is used. The data is divided into training set and validation set by the sliding window method. Using the adjusted time granularity data, the room demand of each demand preference group in different channels is predicted, and the prediction results are compared with the actual data to calculate the mean absolute percentage error (MAPE) and the root mean square error (RMSE). When the MAPE is lower than 20% and the RMSE is lower than 0.5 times the standard deviation of the historical data, the model performance is judged to be good, otherwise the time granularity is readjusted. Based on the price sensitivity threshold data set, for each demand preference group, the price sensitivity distribution under different channels and time granularities is calculated, and the 95% quantile is taken as the upper limit of the price adjustment range of the group. This upper limit value is included in the demand forecasting model as a constraint to ensure that the price change considered in the forecast does not exceed this upper limit. The price adjustment strategy is optimized using the linear programming method to maximize the expected return, forming a complete demand forecasting and price adjustment decision support system. When dealing with the room demand forecasting of a business demand preference group, the historical data is first analyzed, and it is found that the order data of this group in the OTA channel shows obvious periodicity and trend. ARIMA and Prophet models were used for comparison. The optimal parameters of the ARIMA model were calculated to be (2,1,2) and the AIC value was 876.5. The Prophet model automatically captured the annual, weekly and holiday effects of the data. The two models were evaluated using 5-fold cross validation. The average MAPE of ARIMA was 15.3% and that of Prophet was 13.8%, so the Prophet model was selected for subsequent forecasts. For short-term forecasts within 7 days, hourly data was used; for medium-term forecasts of 8-30 days, daily data was used; and for long-term forecasts of 31-90 days, weekly data was used. For example, in the forecast for the next 30 days, 24-hour granularity was used for the first 7 days and daily granularity was used for the next 23 days. The forecast results show that the MAPE is 16.5% and the RMSE is 0.48 times the standard deviation of the historical data, which meets the preset threshold requirements. Based on the price sensitivity data of the group, the 95% quantile was calculated to be 8.2%, which is the upper limit of the price adjustment.This constraint is incorporated into the forecasting model to ensure that the price change considered in the forecast does not exceed 8.2%. Finally, the linear programming method is used to optimize the pricing strategy with the goal of maximizing the expected 30-day profit. The optimal price adjustment plan is a 3.5% increase on weekdays and a 6.8% increase on weekends, which is expected to bring a 12.4% profit increase.

[0035] Step S105, combining the hotel's real-time room status data, room type demand forecast results and price sensitivity thresholds, dynamically adjusting prices and shutting down sales of some room types.

[0036] Acquire a room demand prediction model and room status data, and calculate the future demand prediction values ​​of different room types based on the room demand prediction model and room status data; use a genetic algorithm to calculate the optimal pricing strategy, and the optimal pricing strategy is based on the demand prediction value and current room status data; receive real-time reservation data after price adjustment, and calculate the reservation volume and occupancy rate of each room type; if the reservation volume of a room type exceeds a preset threshold or the occupancy rate reaches a predetermined level, a progressive shutdown mechanism is triggered; the progressive shutdown mechanism includes restricting sales through some channels and completely shutting down sales of the room type; transmit the actual reservation data, occupancy rate data and customer feedback information after price adjustment and room closure to the room demand prediction model; and adjust the room demand prediction model parameters in real time according to the actual reservation data, occupancy rate data and customer feedback information.

[0037] Specifically, based on the room demand forecasting model and price sensitivity threshold, combined with the hotel's real-time room status data, the ARIMA and Prophet algorithms are used to build an integrated forecasting model, introducing external data such as weather and holidays to forecast the demand of different room types in the next 7 days, 30 days and 90 days. The forecast results are updated daily through the rolling forecast method to obtain the forecast value of room demand at each time granularity. Based on the predicted demand and current room status data, the genetic algorithm is used to calculate the optimal pricing strategy. Considering the nonlinear relationship between price and demand, different room types and different time periods are priced differently under the constraint of the price sensitivity threshold. The price sensitivity threshold is determined by analyzing the minimum percentage of significant demand changes caused by price changes in the past 90 days, and the adjusted price is updated in real time to each sales channel. By monitoring the real-time booking data after the price adjustment, the booking volume and occupancy rate of each room type are calculated, and the dynamic threshold is set: the booking volume threshold is 85% of the predicted demand, and the occupancy rate threshold is 90%. When the number of reservations for a room type exceeds the preset threshold or the occupancy rate reaches a predetermined level, the progressive shutdown mechanism is triggered, first restricting sales in some channels and then completely shutting down sales of the room type. The actual reservation data, occupancy rate data, and customer feedback information after price adjustment and room closure are fed back to the demand forecasting model, and the Online Gradient Descent online learning algorithm is used to update the model parameters in real time to improve forecasting accuracy. At the same time, the sliding window method is used to retain the data for the last 90 days, dynamically adjust the price sensitivity threshold, and form a closed-loop dynamic optimization process.

[0038] In the process of implementing room demand forecasting and dynamic pricing, the ARIMA and Prophet algorithms are first used to build an integrated forecasting model. For example, for a luxury suite, the ARIMA model parameters are (2,1,2), and the Prophet model captures the obvious weekend effect and summer peak. The introduction of weather data shows that when the temperature exceeds 30℃, the demand for this room type increases by 15%. Combining these factors, the model predicts that the average daily demand for this room type in the next 7 days will be 20, 18 in 30 days, and 15 in 90 days. Based on these prediction results, the genetic algorithm calculates the optimal pricing strategy: the price on weekdays is maintained at 1,200 yuan / night, increased to 1,500 yuan / night on weekends, and up to 1,800 yuan / night during the summer peak. Historical data analysis shows that price changes of more than 8% will lead to significant changes in demand, so the price sensitivity threshold is set to 8%. Real-time monitoring shows that the number of reservations for this room type reached 17 on a certain Friday, exceeding the 85% threshold of the predicted demand of 20 rooms, and the system automatically restricted sales in some channels. When the occupancy rate reaches 92%, the sales of this room type will be completely closed. The data after price adjustment and room closure show that the actual occupancy is 19, which is slightly lower than the predicted value. Using the OnlineGradient Descent algorithm, the model parameters are updated at a learning rate of 0.01, and the prediction accuracy is improved by 2%. At the same time, based on the data of the last 90 days, the price sensitivity threshold is adjusted to 7.8% for the next round of pricing decisions.

[0039] Based on the hotel's real-time room status data, different room types are classified according to their current booking status. For unbooked room types, their booking probability and room demand in the future are evaluated; for booked and reserved room types, whether there is overbooking or a sudden increase in demand is determined, and whether they need to be released in advance or locked at a higher price.

[0040] Receive real-time room status data carrying room status identifiers, the room status data is generated in real time by a hotel management system; classify the room types into three states: unbooked, booked and reserved according to the room status data, and calculate the proportion of rooms in each state; use a random forest algorithm to evaluate the booking probability of unbooked rooms, the input features of the random forest algorithm include historical booking data, seasonal factors, weekends / weekdays and current room prices; predict future room demand through an ARIMA model, the ARIMA model is trained based on historical booking data; determine whether there is overbooking of booked and reserved rooms, if the ratio of the current booking volume to the total number of rooms exceeds a preset threshold, it is determined that there is overbooking; calculate the probability of the reserved room type being converted into an actual booking, the conversion probability is obtained based on historical reservation data statistics; use a dynamic programming algorithm to calculate the optimal room allocation plan based on the overbooking judgment result, the demand surge judgment result, the booking probability, the demand volume forecast and the reservation conversion probability; determine whether it is necessary to release the reserved rooms in advance or to increase the price to lock some rooms, and generate a corresponding operation suggestion list.

[0041] Specifically, according to the hotel's real-time room status data, the room types are directly divided into three states: unbooked, booked, and reserved according to the label of the room status data. By calculating the proportion of rooms in each state, the overall booking status of the current hotel is obtained. For the unbooked room types, the random forest algorithm is used to evaluate their booking probability in the future. The input features include historical booking data, seasonal factors, weekends / weekdays, current room prices, etc., and the booking probability value between 0 and 1 is output. At the same time, the ARIMA model is used to predict the room demand in the next 7 days, 14 days and 30 days. For the booked and reserved room types, by comparing the ratio of the current booking volume to the total number of room types, it is determined whether there is an overbooking situation. The preset threshold is determined based on the average booking ratio when overbooking occurs in the past 90 days. The moving average method is used to calculate the recent demand growth rate. When the growth rate exceeds the threshold obtained by historical data analysis, it is determined to be a sudden increase in demand. For the reserved room types, the probability of converting the reservation into an actual reservation is calculated and incorporated into the subsequent decision-making process. Based on the judgment results of overbooking and sudden increase in demand, combined with the booking probability, demand forecast and reservation conversion probability of room types, the dynamic programming algorithm is used to calculate the optimal room allocation plan. The state is defined as the current available number of various room types, the action is the number of rooms released or locked, and the reward function considers the expected return and overbooking risk. The optimal strategy is calculated through reverse iteration to determine whether it is necessary to release the reserved rooms in advance or lock some rooms at a higher price, and generate a corresponding operation suggestion list. In the process of implementing room status management, the 100 guest rooms of a hotel are first divided into three states: unbooked, booked and reserved through real-time room status data, and the results are 50 unbooked, 40 booked and 10 reserved. For the 50 unbooked room types, the random forest algorithm is used to evaluate the booking probability. The input features include the average daily booking rate of 85% in the past 30 days, the current season is summer vacation, weekend, and the room price is 800 yuan / night. The output booking probability is 0.75. At the same time, the ARIMA (2,1,2) model is used to predict that the average daily demand for the next 7 days will be 45 rooms, 14 days will be 42 rooms, and 30 days will be 40 rooms. For the 40 booked rooms and 10 reserved rooms, the current booking ratio is calculated to be 50%, which is lower than the average booking ratio of 80% when overbooking occurred in the past 90 days, and it is determined that there is no overbooking. The demand growth rate calculated by the 5-day moving average method is 15%, which exceeds the 10% threshold obtained by historical data analysis, and it is determined that the demand has increased sharply. For the 10 reserved rooms, the probability of converting reservations to actual reservations is 0.6 calculated based on historical data. Based on the above data, the dynamic programming algorithm is used to calculate the optimal room allocation plan for the current state of 50 available rooms. The action space is defined as releasing 0-10 reserved rooms or locking 0-20 unbooked rooms. The reward function comprehensively considers expected income and overbooking risks. After 1,000 iterative calculations, the optimal strategy was found to be releasing 5 reserved rooms and locking 10 unbooked rooms, which is expected to increase revenue by 8% and control the overbooking risk within 3%.

[0042] Formulate room closure strategies based on the status of different room types and future demand forecast results. For room types with high future demand but currently unbooked, prioritize dynamic pricing strategies to increase prices to control demand; for room types with persistently low demand, temporarily close the room or reopen room sales channels. At the same time, update room status in real time based on customer booking behavior patterns to ensure the synchronization and accuracy of room status information and room demand forecast models.

[0043] Obtain the current reservation rate, future forecast demand and historical average occupancy rate, use the random forest algorithm to build a room closing strategy model, and obtain room closing recommendations. Based on the room closing recommendations, determine whether dynamic price adjustment or temporary room closing is required. If dynamic price adjustment is required, the demand elasticity formula is used to calculate the price elasticity value; based on the price elasticity value, the price adjustment range is determined by a linear programming method, which includes setting an objective function to maximize revenue and constraints on the number of rooms, and the highest and lowest prices. Analyze sales channels and identify inefficient channels. Use a distributed message queue to analyze reservation data in real time, optimize the demand forecasting model through an online gradient descent algorithm, and update room status information.

[0044] Specifically, according to the real-time status of different room types and the future demand forecast results, the random forest algorithm is used to build a room closing strategy model. The input features include the current booking rate, the predicted demand in the next 7 days, the historical average occupancy rate, etc., and the room closing suggestions are output to determine whether dynamic price adjustment or temporary room closing measures are needed. For room types with high future demand but currently unbooked, the demand elasticity formula E=(ΔQ / Q) / (ΔP / P) is used to calculate the price elasticity, and then the price is adjusted according to the elasticity value. E is the price elasticity coefficient of demand, which indicates the percentage of demand change when the price changes by one percent. ΔQ is the change in demand, which is the difference between the new demand and the original demand. Q is the original demand, which is the demand level before the price change. ΔP is the change in price, which is the difference between the new price and the original price. P is the original price, which is the price level before the demand change. The specific price adjustment range is determined by the linear programming method with the objective function (maximizing revenue) and constraints (number of rooms, highest and lowest prices) to achieve precise control of demand. At the same time, the price upper limit threshold is set to prevent excessive price adjustments from affecting the hotel's reputation. For rooms with persistently low demand, we set criteria for judging low demand, such as when the booking rate is less than 30% for 7 consecutive days, which triggers a temporary room closure mechanism. At the same time, we start the sales channel re-evaluation procedure, use the K-means clustering algorithm to analyze the existing sales channels, identify inefficient channels and suspend cooperation, and optimize the channel structure. We build a real-time room status update mechanism based on customer booking behavior patterns and sales channel analysis results, use distributed message queues to analyze booking data in real time, and continuously optimize the demand forecasting model through the online gradient descent algorithm to ensure the synchronization and accuracy of room status information and demand forecasting model, and realize dynamic room status management. In the process of implementing dynamic room status management, we first use the random forest algorithm to build a room closure strategy model. The input features include the current booking rate of 75%, the predicted demand of 90 rooms in the next 7 days, and the historical average occupancy rate of 85%. After training 1,000 decision trees, the output recommends taking dynamic price adjustment measures for luxury suites. Subsequently, we calculate the price elasticity for luxury suites. When the price is raised from 1,000 yuan to 1,100 yuan, the demand drops from 20 rooms to 18 rooms, and the price elasticity is -0.9. Based on this elasticity value, the optimal price adjustment range is determined to be 8% through linear programming, that is, the price is adjusted to 1,080 yuan. At the same time, if the reservation rate of standard rooms is less than 25% for 7 consecutive days, a temporary room closure mechanism is triggered. The K-means clustering algorithm is started to divide the 10 sales channels into 3 categories, identify 2 inefficient channels, and suspend their room allocation.Combining customer booking behavior and channel analysis results, the distributed message queue is used to process 1,000 booking data per second, and the demand forecasting model parameters are updated in real time through the online gradient descent algorithm. The learning rate is set to 0.01, and the model is updated every 100 data. The synchronization between room status information and the demand forecasting model is improved to 99.5%, and the accuracy is increased by 3 percentage points, realizing precise dynamic room status management.

[0045] Step S106, based on the price information after dynamic price adjustment and the price sensitivity threshold, combined with the key attributes of historical check-in frequency, evaluation score, and repurchase rate of different demand preference groups, a collaborative filtering algorithm is used to match the optimal room type combination from the hotel room type library, and personalized room type recommendations are pushed to target customers through different room status release channels and different time nodes.

[0046] Obtain a customer and room type matrix, which includes historical check-in frequency, evaluation score and repurchase rate; standardize the customer and room type matrix to obtain standardized customer and room type data. Use a user-based collaborative filtering algorithm to calculate the similarity between customers; calculate the recommendation score of each room type for target customers based on the historical preferences of similar customer groups and the current room type inventory status, combined with the current price of the room type, the customer's price sensitivity and historical evaluation; sort the recommendation scores and select the K room types with the highest scores as candidate recommendation results; filter the candidate recommendation results according to the price sensitivity threshold to obtain the final recommended room type. Receive customer historical behavior data; use the customer historical behavior data to predict the best push time; dynamically optimize the push strategies of different channels, and send room type recommendation information to target customers through the best channel at the best push time.

[0047] Specifically, according to the price information after dynamic price adjustment and the price sensitivity threshold, a customer and room type matrix is ​​constructed. The matrix elements include the customer's historical check-in frequency, evaluation score and repurchase rate for the room type. The Z-score standardization method is used to process the data and convert the attributes of different dimensions into a unified scale. The user-based collaborative filtering algorithm is used to calculate the similarity between customers. At the same time, cosine similarity and Pearson correlation coefficient are used as measurement indicators. For each target customer, 5% of the total number of customers and no less than 10 similar customers are found to form a similar customer group. Based on the historical preferences of the similar customer group and the current room inventory status, the weighted factor decomposition machine model is used to calculate the recommendation score of each room type for the target customer. The score calculation takes into account factors such as the current price of the room type, the customer's price sensitivity, and historical evaluation. According to the recommendation score sorting, the K room types with the highest scores are selected as the recommendation results. The price sensitivity threshold is used to filter the recommendation results to ensure that the recommended room type price does not exceed the acceptable range for customers. Combined with the characteristics of different room status release channels and the active time of customers, time series analysis is introduced to predict the best push time based on customer historical behavior data. Recursive neural networks are used to dynamically optimize the push strategies of different channels, and room recommendation information is pushed to target customers through the optimal channel at the appropriate time node. In the process of implementing personalized room recommendation, a matrix containing 1,000 customers and 50 room types is first constructed, and each element contains historical check-in frequency 0-20 times, evaluation score 1-5 points, and repurchase rate 0-100%. The Z-score standardization method is used to process the data, and these three attributes are converted into a standard normal distribution with a mean of 0 and a standard deviation of 1. Subsequently, the collaborative filtering algorithm is used to calculate customer similarity, while considering cosine similarity and Pearson correlation coefficient, and taking the weighted average of the two, with weights of 0.6 and 0.4 respectively as the final similarity. For each target customer, 50 customers with the highest similarity are selected as the similar customer group. Next, the weighted factor decomposition machine model is used to calculate the recommendation score. The input features include 10 dimensions such as the current price of the room type 500-5000 yuan, customer price sensitivity 0.1-2.0, and historical evaluation 1-5 points. The final model is obtained through 100 iterations of training. According to the recommendation scores output by the model, the top five room types with the highest scores are selected and filtered with a price sensitivity threshold, such as 1.5, to ensure that the price of the recommended room type does not exceed 1.5 times the customer's historical average consumption. Finally, the long short-term memory network (LSTM) is used to analyze the customer's behavior data for the past 90 days to predict the best push time. At the same time, the recursive neural network dynamically optimizes the selection strategy of the five push channels, such as APP, SMS, and email, and pushes personalized room recommendations to customers through the optimal channel at the predicted best time.

[0048] Furthermore, if the customer reserves a preferred room type and the booking price is within the price sensitivity threshold, the reservation of the room type is locked, and the room status association strategy of the room type is dynamically adjusted according to the customer's booking time and stay duration attributes. If the customer reserves a non-preferred room type, or the booking price exceeds the price sensitivity threshold, an alternative room type is dynamically recommended within the price sensitivity threshold based on the customer's demand characteristics and the hotel's room inventory.

[0049] Receive a reservation request message carrying a unique customer identification number, the reservation request message is sent by the customer; obtain a preferred room type list and price sensitivity threshold information from a customer portrait database according to the unique customer identification number; obtain the real-time price of the room type in the reservation request message, and determine whether the real-time price is lower than the price sensitivity threshold; if the real-time price is lower than the price sensitivity threshold, trigger a room type locking mechanism and mark the room type as booked; call a decision tree algorithm according to the booked status to dynamically adjust the availability and price of the associated room type; if the real-time price is higher than the price sensitivity threshold, start an alternative room type recommendation process and generate an alternative room type list; receive customer feedback on the alternative room type list, and update the customer preference information in the customer portrait database according to the feedback information.

[0050] Specifically, after receiving a customer's reservation request, the customer's preferred room type list and price sensitivity threshold information are extracted from the customer portrait database. The database structure contains fields such as customer ID, preferred room type list, price sensitivity threshold, etc. The SQL query language is used to obtain relevant information, and the real-time price of the current reserved room type is obtained at the same time. By comparing the relationship between the reservation price and the price sensitivity threshold, it is determined whether the reservation conditions are met. If the customer reserves a preferred room type and the reservation price is within the price sensitivity threshold, the room type locking mechanism is triggered, the room type is marked as booked, and the customer's reservation time and stay duration information are recorded. According to these attributes, the decision tree algorithm is called to dynamically adjust the availability and price of the associated room type, and update the association strategy of the room type. If the customer reserves a non-preferred room type or the reservation price exceeds the price sensitivity threshold, the alternative room type recommendation process is started. Combining collaborative filtering algorithms and content-based recommendation methods, room type characteristics such as area, facilities and customer preferences are comprehensively considered, and a weighted hybrid strategy is used to generate a list of alternative room types that meet the price sensitivity threshold. The list of alternative room types is pushed to customers through the communication channels preferred by customers. At the same time, the real-time monitoring mechanism is started, and distributed message queues are used to receive customer feedback in real time. Real-time data analysis and updates are performed through stream processing tools to track customer feedback and selection of recommended room types. These data are sent back to the customer portrait database for subsequent personalized recommendation optimization and dynamic adjustment of room status association strategies. In the implementation of the dynamic room type recommendation and booking process, the booking request with customer ID 10086 is first received, and the customer information is immediately extracted from the customer portrait database. The SQL query shows that the preferred room type is "Deluxe Sea View Room" and the price sensitivity threshold is 1,500 yuan / night. The real-time price of the currently booked "Deluxe Sea View Room" is 1,200 yuan / night, which is lower than the threshold, and the booking conditions are determined to be met. The room type locking mechanism is then triggered, and the room type is marked as booked, and the booking time is recorded as 2023-06-1520:30, and the check-in time is 3 nights. Based on this information, the decision tree algorithm raises the price of the 5 nearby "Deluxe Sea View Rooms" by 8% to 1,296 yuan / night, and at the same time raises the price of the 10 "Standard Sea View Rooms" by 5% to 945 yuan / night. If the customer chooses to book a "Standard Double Room" (1,000 yuan / night), the price sensitivity threshold is exceeded, and the recommendation of alternative room types is initiated. The collaborative filtering algorithm recommends the "Superior Double Room" based on the historical choices of similar customers. The content-based recommendation method takes into account the characteristics of the room type, such as an area of ​​80 square meters and a jacuzzi. The weights of the two algorithms are 0.6 and 0.4 respectively, and finally an alternative list including "Superior Double Room" (1,350 yuan / night) and "Deluxe Double Room" (1,480 yuan / night) is generated.Recommendations are pushed through the SMS channel preset by the customer, and real-time monitoring is enabled. 1,000 customer feedbacks are processed per second through the distributed message queue. Real-time analysis by the stream processing tool shows that the click-through rate of "Superior Double Room" is 30% and the booking conversion rate is 5%. These data are updated to the customer portrait in real time to optimize future recommendation strategies.

[0051] like Figure 2 The present invention also provides a multi-channel room reservation system for a hotel, comprising:

[0052] The data collection and preprocessing module is used to obtain historical booking data from different booking channels. The historical booking data includes room rate levels, price fluctuations, and room inventory data. The data is segmented according to the key attributes of customer booking behavior, length of stay, and average room rate of historical orders to obtain a room data set. Outliers and missing values ​​in the room data set are removed, and customer age, preferred room type, price sensitivity, and historical check-in frequency are extracted as key features. The key features are normalized to form a standardized feature vector that describes the demand characteristics and price sensitivity of each customer segment.

[0053] The customer clustering module is used to classify customers into different demand preference groups based on the standardized feature vectors that describe the demand characteristics and price sensitivity of each customer segment, and to establish a standardized feature vector dataset and price sensitivity threshold dataset for each demand preference group under different channels and time granularities;

[0054] The demand forecasting module is used to predict the future demand for housing types in different channels and for different groups based on the housing demand feature vector dataset of each demand preference group, using a time series forecasting model and dynamically adjusting the time granularity of the training data. At the same time, based on the price sensitivity threshold dataset, the upper limit of the price adjustment range for different groups is determined;

[0055] Dynamic pricing module, which combines the hotel's real-time room status data, room demand forecast results and price sensitivity thresholds to dynamically stop the sale of some room types;

[0056] The personalized recommendation module is used to match the optimal room type combination from the hotel room type library based on the price information and price sensitivity threshold after dynamic price adjustment, combined with the key attributes of historical check-in frequency, evaluation score, and repurchase rate of different demand preference groups, using a collaborative filtering algorithm, and push personalized room type recommendations to target customers through different room status release channels and at different time nodes.

[0057] It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, and that the present application can be implemented in other specific forms without departing from the spirit or essential features of the present application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the present application is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present application. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A multi-channel hotel reservation method, characterized in that: The method comprises: Obtain historical booking data from different booking channels. The historical booking data includes room rate levels, price fluctuations, and room inventory data. The data is segmented based on key attributes such as customer booking behavior, length of stay, and average room rate of historical orders to obtain a room data set. Remove outliers and missing values ​​from the room type data set, extract customer age, preferred room type, price sensitivity, and historical check-in frequency as key features, normalize the key features, and form a standardized feature vector that describes the demand characteristics and price sensitivity of each customer segment; Based on the standardized feature vectors describing the demand characteristics and price sensitivity of each customer segment, a clustering algorithm is used to divide customers into different demand preference groups. For each demand preference group, a standardized feature vector dataset and a price sensitivity threshold dataset of room type demand under different channels and time granularity are established. Based on the room demand feature vector dataset of each demand preference group, a time series prediction model is used to dynamically adjust the time granularity of the training data to predict the room demand of different channels and different groups in the future. At the same time, the upper limit of the price adjustment range for different groups is determined based on the price sensitivity threshold dataset. Combine the hotel's real-time room status data, room demand forecast results and price sensitivity thresholds to dynamically adjust prices and shut down sales of some room types; Based on the price information and price sensitivity threshold after dynamic price adjustment, combined with the key attributes of historical check-in frequency, evaluation score, and repurchase rate of different demand preference groups, a collaborative filtering algorithm is used to match the optimal room type combination from the hotel room type library, and push personalized room type recommendations to target customers through different room status release channels and at different time nodes.

2. The method according to claim 1, characterized in that The acquisition of historical booking data from different booking channels includes room rate levels, price fluctuations, and room inventory data. The data is segmented based on key attributes such as customer booking behavior, length of stay, and average room rate of historical orders to obtain a room data set, including: Acquire original data carrying a booking channel identifier, where the original data is sent by different booking channels; Preprocessing the raw data to obtain a feature matrix, wherein the feature matrix includes features of multiple time granularities; Performing clustering processing on the feature matrix to obtain clustering results of different room types; Construct a separate house type sub-dataset for each clustering result; Executing a time series decomposition instruction on the house type sub-dataset, wherein the time series decomposition instruction is used to separate trend, seasonality and random fluctuation components; Calculate a price volatility index based on the time series decomposition result, where the price volatility index includes multiple time granularities; The price fluctuation index is combined with the historical data to obtain a housing type data set.

3. The method according to claim 1, characterized in that The outliers and missing values ​​in the room type data set are removed, customer age, preferred room type, price sensitivity, and historical check-in frequency are extracted as key features, and the key features are normalized to form a standardized feature vector that describes the demand characteristics and price sensitivity of each customer segment, including: Preprocess the acquired room type data set, including deleting outliers and filling missing values, to obtain a complete data set; Extract customer age group, preferred room type, price sensitivity, and historical stay frequency as key features from the complete dataset; Price sensitivity is determined by calculating the standard deviation of the customer's past booking prices over a predefined period of time and dividing it by the average price; The extracted key features are normalized, and the minimum-maximum scaling method is used to map the feature values ​​into a predefined interval for continuous numerical features; The processed features are merged to obtain a standardized feature vector that describes the demand characteristics and price sensitivity of each customer segment.

4. The method according to claim 1, characterized in that: According to the standardized feature vectors describing the demand characteristics and price sensitivity of each customer segment group, a clustering algorithm is used to divide customers into different demand preference groups, and for each demand preference group, a standardized feature vector dataset and a price sensitivity threshold dataset of room type demand under different channels and different time granularities are established, including: receiving a normalized feature vector carrying customer demand characteristics and price sensitivity; Using a clustering algorithm to group customers according to the standardized feature vector, the clustering algorithm detects and processes outliers by using an interquartile range method; Obtain silhouette coefficients with different K values, determine the optimal number of clusters K, and obtain customer segmentation results; For each cluster group, obtain order information of different booking channels from the historical order database, and construct a feature vector reflecting the group's preferences in different channels; Aggregate historical order data by time granularity to obtain a standardized vector that reflects the demand characteristics of the group at different time granularities; A random forest regression model is adopted, the feature vector and the standardized vector are input, and the price sensitivity threshold of each demand preference group under different channels and time granularities is calculated.

5. The method according to claim 1, characterized in that The housing demand feature vector dataset for each demand preference group adopts a time series prediction model to dynamically adjust the time granularity of the training data to predict the housing demand for different channels and different groups in the future; at the same time, according to the price sensitivity threshold dataset, the upper limit of the price adjustment range for different groups is determined, including: Obtain time series data and adjust the time granularity of training data according to the historical span of the time series data and the prediction target; Use the adjusted time granularity data to predict the demand for room types in different channels by demand preference groups; Determine whether the mean absolute percentage error of the prediction results is lower than a preset threshold; Obtain price sensitivity data from the price sensitivity threshold dataset; Calculate the price sensitivity distribution under different channels and time granularities for the demand preference groups; Taking the preset quantile of the price sensitivity distribution as the upper limit of the price adjustment range; Use linear programming methods to optimize price adjustment strategies; Taking maximizing expected benefits as the objective function and the upper limit of price adjustment as a constraint, the optimal price adjustment plan is obtained.

6. The method according to claim 1, characterized in that The above-mentioned combination of the hotel's real-time room status data, room type demand forecast results and price sensitivity threshold, dynamic price adjustment and suspension of sales of some room types includes: Obtaining a room type demand prediction model and room status data, and calculating the future demand prediction values ​​of different room types according to the room type demand prediction model and room status data; Calculating an optimal pricing strategy using a genetic algorithm, wherein the optimal pricing strategy is based on the demand forecast value and current room status data; Receive real-time booking data after price adjustment and calculate booking volume and occupancy rate of each room type; If the booking volume of a room type exceeds a preset threshold or the occupancy rate reaches a predetermined level, a progressive shutdown mechanism will be triggered; The gradual closure mechanism includes restricting sales through some channels and completely shutting down sales of the room type; The actual booking data and occupancy rate data after price adjustment and room closure as well as customer feedback information are transmitted to the room demand forecasting model; Adjust room demand forecast model parameters in real time based on actual booking data, occupancy rate data and customer feedback; The method of dynamically shutting down the sale of some room types by combining the hotel's real-time room status data, room type demand forecast results and price sensitivity thresholds also includes: classifying different room types according to current booking status based on the hotel's real-time room status data, and evaluating the booking probability and room type demand in the future for unbooked room types; determining whether there is overbooking or a sudden increase in demand for booked and reserved room types, and evaluating whether they need to be released in advance or locked at a higher price; formulating room closure strategies based on the status of different room types and future demand forecast results, and giving priority to adopting a dynamic price adjustment strategy for room types that have high future demand but are currently unbooked, raising prices to control demand; temporarily closing the room or reopening the room sales channel for room types with continuously low demand, and updating the room status in real time in combination with customer booking behavior patterns to ensure the synchronization and accuracy of room status information and room type demand forecast models.

7. The method according to claim 6, characterized in that Based on the hotel's real-time room status data, different room types are classified according to the current booking status; for unbooked room types, their booking probability and room type demand in the future are evaluated; for booked and reserved room types, whether there is overbooking or a sudden increase in demand, and whether they need to be released in advance or locked at a higher price, including: Receiving real-time room status data carrying a room type status identifier, wherein the room status data is generated in real time by a hotel management system; Classify the room types into three states: unbooked, booked, and reserved according to the room status data, and calculate the proportion of rooms in each state; A random forest algorithm is used to evaluate the booking probability of an unbooked room type, wherein input features of the random forest algorithm include historical booking data, seasonal factors, weekends / weekdays, and current room rates; Predicting future room demand using an ARIMA model trained on historical booking data; Determine whether there is overbooking of the booked and reserved rooms. If the ratio of the current booking volume to the total number of rooms exceeds a preset threshold, it is determined that there is overbooking; Calculate the probability of the reserved room type being converted into an actual reservation, where the conversion probability is obtained based on historical reservation data statistics; The optimal room allocation plan is calculated using a dynamic programming algorithm based on the overbooking judgment results, demand surge judgment results, booking probability, demand forecast, and reservation conversion probability; Determine whether it is necessary to release reserved rooms in advance or to lock some rooms at a higher price, and generate a corresponding action suggestion list.

8. The method according to claim 6, Among them, the characteristics are that according to the status of different room types and the future demand forecast results, a room closing strategy is formulated; for room types with high future demand but currently unbooked, a dynamic price adjustment strategy is adopted first to increase the price to control demand; for room types with persistently low demand, the room is temporarily closed or the room sales channel is reopened; at the same time, the room status is updated in real time in combination with the customer booking behavior pattern to ensure the synchronization and accuracy of the room status information and the room demand forecast model, including: Obtain the current booking rate, future forecast demand, and historical average occupancy rate, use the random forest algorithm to build a room closure strategy model, and obtain room closure recommendations; Based on the room closure suggestion, determine whether dynamic price adjustment or temporary room closure is required; If dynamic price adjustment is required, the demand elasticity formula is used to calculate the price elasticity value; According to the price elasticity value, the price adjustment range is determined by a linear programming method, wherein the linear programming method includes setting an objective function of maximizing revenue and constraints of the number of rooms, and the maximum and minimum prices; Analyze sales channels and identify inefficient channels; Distributed message queues are used to analyze booking data in real time, and the demand forecasting model is optimized through an online gradient descent algorithm to update room status information.

9. The method according to claim 1, characterized in that: The method uses collaborative filtering algorithm to match the best room type combination from the hotel room type library based on the price information after dynamic price adjustment and the price sensitivity threshold, combined with the key attributes of historical check-in frequency, evaluation score, and repurchase rate of different demand preference groups, and pushes personalized room type recommendations to target customers through different room status release channels and different time nodes, including: Obtain a customer and room type matrix, wherein the customer and room type matrix includes historical check-in frequency, evaluation score, and repurchase rate; Standardize the customer and room type matrix to obtain standardized customer and room type data; A user-based collaborative filtering algorithm is used to calculate the similarity between customers; Calculate the recommendation score of each room type for target customers based on the historical preferences of similar customer groups and the current inventory status of room types, combined with the current price of the room type and the customer's price sensitivity and historical evaluation; Sorting the recommendation scores, and selecting K room types with the highest scores as candidate recommendation results; Filter the candidate recommendation results according to the price sensitivity threshold to obtain the final recommended room type; Receive customer historical behavior data; Using the customer historical behavior data to predict the best push time; Dynamically optimize the push strategies of different channels and send room recommendation information to target customers through the best channel at the best push time.

10. A multi-channel room reservation system for a hotel, characterized in that: The system comprises: The data collection and preprocessing module is used to obtain historical booking data from different booking channels. The historical booking data includes room rate levels, price fluctuations, and room inventory data. The data is segmented according to the key attributes of customer booking behavior, length of stay, and average room rate of historical orders to obtain a room data set. Outliers and missing values ​​in the room data set are removed, and customer age, preferred room type, price sensitivity, and historical check-in frequency are extracted as key features. The key features are normalized to form a standardized feature vector that describes the demand characteristics and price sensitivity of each customer segment. The customer clustering module is used to classify customers into different demand preference groups based on the standardized feature vectors that describe the demand characteristics and price sensitivity of each customer segment, and to establish a standardized feature vector dataset and price sensitivity threshold dataset for each demand preference group under different channels and time granularities; The demand forecasting module is used to predict the future demand for housing types in different channels and for different groups based on the housing demand feature vector dataset of each demand preference group, using a time series forecasting model and dynamically adjusting the time granularity of the training data. At the same time, based on the price sensitivity threshold dataset, the upper limit of the price adjustment range for different groups is determined; Dynamic pricing module, which is used to dynamically adjust prices and stop the sale of certain room types by combining the hotel's real-time room status data, room demand forecast results and price sensitivity thresholds; The personalized recommendation module is used to match the optimal room type combination from the hotel room type library based on the price information and price sensitivity threshold after dynamic price adjustment, combined with the key attributes of historical check-in frequency, evaluation score, and repurchase rate of different demand preference groups, using a collaborative filtering algorithm, and push personalized room type recommendations to target customers through different room status release channels and at different time nodes.

Images