Hotel room speed prediction system based on hidden Markov model and index curve

Abstract

The invention provides a hotel room speed prediction system based on a hidden Markov model and an index curve. The system belongs to the technical field of artificial intelligence software. According to the system, the psychological trend of hotel user groups in order submitting time is accurately insighted by deeply analyzing historical room speed data and a hidden Markov chain. Through scientific calculation, the system can update the expected value of the house-going speed everyday, thereby effectively solving the defects of an existing house-going speed prediction algorithm in the aspect of providing personalized services. The innovative prediction algorithm not only can provide customized services for hotels of different user groups, but also is helpful for the hotels to plan operation more reasonably, and the operation efficiency is remarkably improved.

Description

Technical field:

[0001] The present invention relates to the field of artificial intelligence software, and in particular to a hotel room occupancy rate prediction system based on a hidden Markov model and an exponential curve. Background technology:

[0002] For a future check-in date, the number of orders for that check-in date every day in the near future is a key indicator of hotel operations. Generally speaking, as a check-in date approaches, the number of orders received by the hotel for that date will gradually increase, and the speed of room occupancy will also increase accordingly. This reflects the customer's increased willingness to book and the clarity of decision-making when the check-in date is approaching.

[0003] At present, many hotels use room booking speed prediction methods that mainly rely on the time and booking number curve fitting technology based on Poisson distribution, but this method ignores individual factors and is not responsive to market dynamics. This method usually presets a smooth exponential curve based on experience that increases the number of bookings over time. It does not take into account the characteristics of different hotels and room types, as well as changes in customer mentality during special periods such as holidays; existing prediction methods often fail to fully respond to rapid changes in the market, such as competitor strategy adjustments, the impact of tourism off-seasons, etc., which may lead to a large deviation between the actual room booking speed and the predicted results.

[0004] Therefore, in order to improve the accuracy and practicality of forecasts, it is necessary to explore more flexible and sophisticated forecasting technologies and methods to better adapt to the dynamic changes in hotel business and the diversity of customer needs.

[0005] Existing hotel intelligent management systems generally use smooth curve fitting technology based on Poisson distribution when predicting the room occupancy rate on a certain date in the future. This technology builds a mathematical model based on historical data and mathematical statistical methods, assuming that the room occupancy rate gradually increases over time. However, this method has some significant limitations. It relies on fixed historical average data and builds a relatively static model, which is difficult to accurately capture the dynamic changes in user booking behavior in reality. At the same time, it does not fully take into account the differences in booking habits between different user groups. For example, business travelers and leisure tourists may have completely different booking patterns. In addition, in a rapidly changing market environment, such as the emergence of new competitors, policy adjustments or emergencies, this traditional method is often difficult to adjust the forecasting strategy in a timely manner, and its ability to respond to market changes is weak. Summary of the invention:

[0006] In order to overcome these limitations, the present invention adopts a more advanced hidden Markov model (HMM). HMM can not only deal with the randomness in time series data, but also deeply explore the hidden state behind the data, that is, the user's booking mentality and tendency. Compared with the traditional smooth curve fitting method, the model of the present invention is more dynamic and adaptive. It can be updated in real time, and the prediction strategy can be adjusted according to the latest market data and user behavior to ensure the timeliness and accuracy of the prediction. At the same time, through HMM, the system can deeply learn the hidden patterns of user booking behavior, such as user's booking time preference, price sensitivity, etc., so as to more accurately predict the speed of room booking. The following technical solutions are specifically adopted:

[0007] The hotel sales database of this system is updated in real time through Java Springboot technology and automatically refreshed every 5 minutes. During each update, the system will accurately calculate the room occupancy rate for each check-in date and each room type based on the latest hotel sales data, thereby ensuring the timeliness and accuracy of the data.

[0008] The room occupancy rate prediction model is implemented in Python programming, combining the LSTM model in deep learning and the hidden Markov model in statistics. Among them, the future occupancy prediction uses the LSTM model in the PyTorch library to capture the long-term dependencies in time series data: while the room occupancy rate prediction is implemented through the exponential curve fitting model in the statsmodel library and the hidden Markov model in the hmmlearn library. To ensure the continued effectiveness of the prediction model, the system retrains the hidden Markov model every day to adapt to market changes and user needs.

[0009] Before training the hidden Markov model, the system first uses the exponential curve model in the R language to fit a smooth curve based on the historical hotel room occupancy data. This curve reflects the law of the change of the room occupancy rate with the number of days from the check-in date (t) without external interference. Among them, λ(t) represents the room occupancy rate when the check-in date is t days; a and b are parameters obtained by fitting historical data. The fitting formula is:

[0010] λ(t)=e^(a+bt)

[0011] The hidden Markov model of this system sets the difference between the observed room occupancy rate and the theoretical curve as the random variable Y. The model contains three hidden states, corresponding to three different user reservation mindsets: booking long in advance, booking short in advance, and booking close to the travel date. Through training data, the model determines a set of parameters (μ k , σ k), respectively representing the expected number of orders submitted by the user group in this state and the standard deviation. The model formula is:

[0012]

[0013] When making predictions, the model infers the most likely user booking mentality (i.e., implicit state) based on the current room booking speed data, and predicts the future room booking speed based on the transition probability of the hidden Markov chain. This prediction method not only takes into account the statistical laws of historical data, but also incorporates the dynamic changes of user behavior patterns, thereby achieving a more accurate and personalized room booking speed prediction service. Description of the drawings:

[0014] Figure 1 This is a flow chart of the house entry speed prediction system based on hidden Markov model and exponential curve.

[0015] Figure 2 Schematic diagram of hidden Markov model. Specific implementation method:

[0016] The present invention proposes a hotel room occupancy rate prediction and pricing optimization system based on big data analysis, which can dynamically predict and adjust according to data from different years. The specific implementation steps are as follows:

[0017] Step 1: Hotel sales database update and room occupancy rate calculation S1 Data update: The system will first update the order information in the hotel sales database in real time according to the selected year, especially the information related to the calculation and prediction of room occupancy rate, which includes the booked check-in date (Dj), the date of order submission (Bj), room type, number of rooms (Rj), etc. S2 initializes the room occupancy rate array: For each check-in date (Di) in the selected year, the system initializes an array (Xi) of length 30, which represents the room occupancy rate on that date. The kth element Xi[k] of the array represents the total number of rooms booked for that check-in date k days in advance. S3 Data Processing and Room Filling Speed ​​Calculation: Whenever there is a new order, the system calculates the number of days in advance (k = Dj-Bj). If k is greater than 30, it is considered a 30-day advance booking. Then, the number of rooms in the order (Rj) is accumulated to the corresponding Xi[k] to update the room filling speed data. S4 Multi-Room Type Processing: For hotels with multiple room types, the system will independently calculate and predict the room occupancy rate for each room type.

[0018] Step 2: Predicting room availability and optimizing pricing strategies S5 Obtain pre-training data: The system allows users to select historical data of a specific year as pre-training data. The data is classified into off-season and peak-season room occupancy rates through the k-means clustering algorithm. S6 Establish a benchmark model: Use the off-season and peak season data of the selected year to fit a smooth exponential curve respectively, which serves as the benchmark housing occupancy rate model for different market demand stages. S7 Real-time prediction and adjustment: During the operation of the revenue forecasting and pricing system, the model will adjust the room availability prediction every day through the hidden Markov model. The model takes into account the different booking mentality of users and infers the mentality of the current dominant user group based on the difference between the room availability speed and the benchmark curve over the past period of time, and then predicts the most likely expected room availability speed. S8 Dynamic Optimization of Pricing Strategies: Based on the updated room availability prediction, the system will combine competitors’ pricing strategies to provide users with pricing suggestions. At the same time, the LSTM deep learning model is used to predict the impact of different pricing strategies on occupancy rates based on historical housing prices and occupancy rate data for the selected year, thereby dynamically optimizing pricing strategies.

Claims

1. The hotel room occupancy rate prediction system based on hidden Markov model and exponential curve is characterized by the following steps: Step 1: Use the Java Springboot framework to write a hotel sales database update program to obtain and update the detailed list of hotel sales data in real time to ensure the timeliness and accuracy of the data. Step 2: By fitting the real hotel sales data, an expected room occupancy rate curve based on Poisson distribution is constructed. The difference between the current user's hotel occupancy rate and the expected curve is further calculated to provide basic data for subsequent predictions. Step 3: Based on the hotel sales data details obtained in step 1, use Python programming language to write a room booking speed prediction model. This model can predict the difference between the user's hotel room booking speed and the expected curve, providing input for the subsequent hidden Markov model. Step 4: Use the difference predicted in step 3 as a random variable to construct a hidden Markov model with three hidden states. These three hidden states correspond to three different user groups with different booking psychology tendencies: booking long in advance, booking short in advance, and booking close to the travel date. Through training data, a set of parameters is determined for each hidden state. When predicting, the model will infer the most likely user group mentality based on the current room booking speed, and output the difference random variable predicted in step 3 based on the hidden Markov chain. Step 5: Add the difference predicted in step 4 to the expected value of the room check-in speed of the expected curve to obtain the final predicted value of the user's room check-in speed. This predicted value can provide an important basis for the hotel's operational decision-making and help improve the hotel's service quality and operational efficiency. The present invention not only retains the traditional smooth curve fitting method as the basis, but also cleverly integrates advanced machine learning technology. This combination enables the prediction model to take into account historical trends while adapting to the rapid changes in the modern market and the complexity of user behavior. In addition, the present invention also focuses on comprehensive data fusion and analysis. In addition to the hotel's internal sales data, it also actively integrates external market data, user feedback, social media public opinion and other multi-dimensional information to provide richer references for prediction. Through in-depth mining and analysis of these data, the system can more accurately identify subtle changes in market trends and user needs. A hotel room occupancy rate prediction system based on hidden Markov model and exponential curve is implemented, which provides a novel and effective room occupancy rate prediction method for the hotel industry.

Images