Apparatus for predicting demand based on historical accommodation data and controlling method thereof

Abstract

A server device disclosed in the present document includes: a communication unit for receiving room data for training a machine learning model related to room demand prediction; a memory for storing the room data; and a control unit which extracts a feature for adjusting a room sales price from the room data so as to generate preprocessed data, uses the preprocessed data as training data, combines a first learning model based on past reservation data and a second learning model based on the past reservation data and future reservation data so as to generate a room reservation number prediction model, inputs the room data to the room reservation number prediction model so as to derive a prediction reservation number as an output value, and derives a target cumulative reservation number based on the prediction reservation number and a reservation cumulative distribution during a lead time.

Description

Demand prediction device based on past stay date data and its control method

[0001] Cross-citation with related applications

[0002] This invention claims the benefit of priority to Korean Patent Application No. 10-2024-0024093, filed February 20, 2024, and Korean Patent Application No. 10-2024-0061449, filed May 9, 2024, the entire contents of which are incorporated herein by reference.

[0003] Technology field

[0004] The embodiments disclosed in this document relate to a server device for predicting room demand and a method for controlling the server device.

[0005] Accommodation products are characterized by a limited supply and a fixed sales deadline, and sellers aim to maximize profits by optimizing the selling price.

[0006] In particular, since the total supply of accommodation providers is fixed in the short term, accurate demand forecasting is necessary in a dynamic sales adjustment model that dynamically adjusts accommodation prices. Accordingly, research into algorithms for accurately predicting room demand is ongoing in the field of dynamic accommodation pricing technology.

[0007] According to one embodiment disclosed in this document, a server device and a control method of the server device are provided for predicting room demand using a machine learning algorithm based on room data.

[0008] The technical problems of the embodiments disclosed in this document are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the descriptions below.

[0009] According to one embodiment, a server device includes a communication unit that receives room data for training a machine learning model related to room demand prediction, a memory that stores the room data, and a control unit that extracts features for room demand prediction from the room data to generate preprocessed data, uses the preprocessed data as training data, generates a reservation number prediction model by combining a first learning model based on past reservation data and a second learning model based on the past reservation data and future reservation data, inputs the room data into the reservation number prediction model to derive a predicted reservation number as an output value, and derives a target cumulative reservation number based on the predicted reservation number and a cumulative distribution of reservations during a lead time.

[0010] The control unit can generate the reservation number prediction model by deriving weights based on the degree to which the output value of the first learning model and the output value of the second learning model are similar to the actual value of the reservation number.

[0011] The above control unit can derive the reservation cumulative distribution by increasing the weight of the past reservation data as it gets closer to the stay date during the lead time.

[0012] The above control unit can derive a cumulative distribution of business reservations and a cumulative distribution of business reservations by lead time and day of the week based on the cumulative distribution of reservations, and derive the maximum value among the cumulative distribution of business reservations and the cumulative distribution of business reservations derived by day of the week as a target cumulative distribution of reservations by day of the week.

[0013] The above control unit can derive the target cumulative reservation number by multiplying the target market share by the predicted number of reservations and the minimum value among the standard number of rooms sold by the target cumulative reservation distribution by day of the week.

[0014] The above control unit can derive a cumulative distribution of hard block reservations and a cumulative distribution of commercial reservations for the business by lead time and day of the week based on the room type included in the room data being a hard block type, and can derive a maximum value of the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations derived by day of the week as a target cumulative reservation distribution by day of the week.

[0015] The above control unit can derive the target cumulative number of reservations by multiplying the purchase amount per day of the room of the hard block type by the target cumulative reservation distribution.

[0016] The above first learning model may include an exogenous factor consideration model (SARIMAX) generated based on the past reservation data for a past reference time from a preset reference date.

[0017] The second learning model may include an additive advance booking model (ABM) generated based on the past reservation data from a preset reference date to a past reference time and the future reservation data from the reference date to a future reference time.

[0018] The above control unit can transmit the target cumulative reservation number to an external device through the communication unit.

[0019] A control method of a server device according to one embodiment includes receiving room data for training a machine learning model related to room demand prediction, extracting features for room demand prediction from the room data to generate preprocessed data, using the preprocessed data as training data, combining a first learning model based on past reservation data and a second learning model based on the past reservation data and future reservation data to generate a reservation count prediction model, inputting the room data into the room reservation count prediction model to derive a predicted reservation count as an output value, and deriving a target cumulative reservation count based on the predicted reservation count and a cumulative distribution of reservations during a lead time.

[0020] Creating the above reservation number prediction model may include creating the reservation number prediction model by deriving weights based on the degree to which the output value of the first learning model and the output value of the second learning model are similar to the actual value of the reservation number.

[0021] A control method of a server device according to one embodiment may further include deriving the cumulative reservation distribution by increasing a weight of the past reservation data as it gets closer to the stay date during the lead time.

[0022] A control method of a server device according to one embodiment may further include deriving a cumulative distribution of business reservations and a cumulative distribution of business reservations by lead time and day of the week based on the cumulative distribution of reservations, and deriving a maximum value among the cumulative distribution of business reservations and the cumulative distribution of business reservations derived by day of the week as a target cumulative reservation distribution.

[0023] Deriving the target cumulative reservation number may include deriving the target cumulative reservation number by multiplying the result of multiplying the target market share by the predicted number of reservations and the minimum value among the standard number of rooms sold by the target cumulative reservation distribution by day of the week.

[0024] Deriving the above target cumulative reservation distribution may include deriving the business's hard block reservation cumulative distribution and the above commercial area reservation cumulative distribution by lead time and day of the week based on the room type included in the above room data being a hard block type, and deriving the maximum value of the hard block reservation cumulative distribution and the above commercial area reservation cumulative distribution derived by day of the week as the target cumulative reservation distribution by day of the week.

[0025] The target cumulative reservation number can be derived by multiplying the daily purchase volume of the hard block type room by the target cumulative reservation distribution by day of the week.

[0026] The above first learning model may include an exogenous factor consideration model (SARIMAX) generated based on the past reservation data for a past reference time from a preset reference date.

[0027] The second learning model may include an additive advance booking model (ABM) generated based on the past reservation data from a preset reference date to a past reference time and the future reservation data from the reference date to a future reference time.

[0028] A control method of a server device according to one embodiment may further include transmitting the target cumulative reservation number to an external device through a communication unit.

[0029] According to a server device according to one embodiment, the number of reservations can be predicted by combining multiple reservation number prediction models, and a target cumulative number of reservations can be derived by considering the cumulative distribution of reservations during the lead time, thereby improving the accuracy of a dynamic sales adjustment system and maximizing sales.

[0030] FIG. 1 illustrates a control block diagram of a server device according to one embodiment.

[0031] FIG. 2 illustrates input and output values ​​of a prediction model utilized in a server device according to one embodiment.

[0032] FIG. 3 and FIG. 4 illustrate a schematic flowchart of a server device according to one embodiment predicting room demand and deriving a target cumulative number of reservations.

[0033] FIG. 5 illustrates a process of combining multiple learning models in a server device according to one embodiment.

[0034] Figure 6 illustrates the result of combining multiple learning models by a server device according to one embodiment.

[0035] FIG. 7 illustrates a flowchart of a method for controlling a server device according to one embodiment.

[0036] FIG. 8 continues to illustrate a flowchart regarding a method for controlling a server device according to one embodiment.

[0037] Hereinafter, various embodiments disclosed in this document will be described in detail with reference to the attached drawings. In this document, identical components in the drawings are designated by the same reference numerals, and redundant descriptions of identical components are omitted.

[0038] With respect to the various embodiments disclosed in this document, specific structural and functional descriptions are merely illustrative for the purpose of explaining the embodiments, and the various embodiments disclosed in this document may be implemented in various forms and should not be construed as being limited to the embodiments described in this document.

[0039] The expressions "first," "second," "first," or "second" used in various embodiments may describe various components, regardless of order and / or importance, and do not limit the components. For example, without departing from the scope of the embodiments disclosed herein, a first component may be renamed a second component, and similarly, a second component may also be renamed a first component.

[0040] The terms used in this document are intended solely to describe specific embodiments and may not be intended to limit the scope of other embodiments. Singular expressions may include plural expressions unless the context clearly indicates otherwise.

[0041] All terms used herein, including technical or scientific terms, have the same meaning as commonly understood by those of ordinary skill in the art of the embodiments disclosed herein. Terms defined in commonly used dictionaries may be interpreted as having the same or similar meaning in the context of the relevant technology, and unless explicitly defined herein, they shall not be interpreted in an idealized or overly formal sense. In some cases, even if a term is defined herein, it cannot be interpreted to exclude the embodiments disclosed herein.

[0042] The operating principle and embodiments of the present invention will be described with reference to the attached drawings below.

[0043] FIG. 1 illustrates a control block diagram of a server device according to one embodiment.

[0044] Referring to FIG. 1, a server device (1) according to one embodiment includes a control unit (100) including at least one processor (110) and a memory (120) and a communication unit (200), and can predict room demand by communicating with an external device (2) through the communication unit (200).

[0045] A server device (1) according to one embodiment may be implemented as various computing devices such as a workstation, a cloud, a data drive, a data station, etc. In addition, the server device (1) may be implemented as one or more server devices (1) that are physically or logically separated based on function, detailed configuration of function, data, etc., and data may be transmitted and received and the transmitted and received data may be processed through communication between each server device (1).

[0046] An external device (2) communicating with a server device (1) may include a user terminal and an external server device that transmits a room price adjusted by the server device (1).

[0047] Specifically, when the external device (2) is a user terminal, the control unit (100) of the server device (1) can transmit the prediction result of the room demand to the user terminal so that the user can check it. At this time, the user terminal may include, but is not limited to, a personal computer, a terminal, a portable telephone, a smart phone, a handheld device, a wearable device, etc.

[0048] In addition, when the external device (2) is an external server device, the external server device can be implemented as various computing devices, and the external server device can be implemented as one or more external server devices, and data can be transmitted and received and the transmitted and received data can be processed through communication between each external server device.

[0049] A server device (1) according to one embodiment may refer to any electronic device including a processor (110) and a memory (120), and the server device (1) may include a user terminal capable of performing server functions. Each component of the server device (1) will be described in detail below.

[0050] The communication unit (200) may include a wireless communication unit (210) and a wired communication unit (220) to communicate with an external device (2). The communication unit (200) may receive room data to be used as learning data from a separately provided external device (2), or may transmit and receive programs and various data for machine learning.

[0051] The wireless communication unit (210) may include at least one of a short-range communication module and a long-range communication module.

[0052] The short-range communication module can communicate with an external device (2) adjacent to the server device (1) using a short-range communication method. Here, the short-range communication module can utilize one of the following communication methods: Bluetooth, Bluetooth low energy, infrared data association (IrDA), Zigbee, Wi-Fi, Wi-Fi direct, Ultra Wideband (UWB), or near field communication (NFC).

[0053] The remote communication module may include a communication module that performs various types of remote communication and may include a mobile communication unit. The mobile communication unit may transmit and receive a wireless signal with at least one of a base station, an external terminal, and an external device (2) on a mobile communication network. In addition, the remote communication module may communicate with an external device (2) or an external device (2) such as another electronic device through a surrounding access point (AP). The access point (AP) may connect a local area network (LAN) to which a server device is connected to a wide area network (WAN) to which a communication server is connected. Accordingly, the server device (1) may be connected to the communication server through the wide area network (WAN) with the external device (2) so that the server device (1) may communicate with each other.

[0054] The wired communication unit (220) can connect to a wired communication network and communicate with an external device (2) through the wired communication network. For example, the wired communication unit (220) can connect to a wired communication network through Ethernet (IEEE 802.3 technology standard) or connect to a wired communication network through CAN communication, and transmit and receive data with the external devices (2) through the wired communication network.

[0055] A server device (1) according to one embodiment may include an input / output interface (not shown). An interface may be provided that connects an input device (not shown) such as a keyboard, mouse, or touch panel, an output device (not shown) such as a display, and a processor (110) to transmit and receive data.

[0056] The memory (120) can store various information necessary for operating the server device (1). Specifically, the memory (120) can store an operating system and a program necessary for operating the server device (1), or store data necessary for operating the server device (1).

[0057] Specifically, the memory (120) can store various programs related to room data preprocessing, deriving a target cumulative number of reservations, and machine learning algorithms. Furthermore, the memory (120) can store feature data utilized in machine learning.

[0058] The memory (120) may include volatile memory (120) such as Static Random Access Memory (S-RAM) and Dynamic Random Access Memory (D-RAM) for temporarily storing data. In addition, the memory (120) may include nonvolatile memory (120) such as Read Only Memory (ROM), Erasable Programmable Read Only Memory (EPROM), and Electrically Erasable Programmable Read Only Memory (EEPROM) for long-term storage of data.

[0059] The processor (110) outputs control signals to control the server device (1) as a whole. The processor (110) may include one or more central processing units (CPUs) and graphics processing units (GPUs). In this case, the processor (110) may be implemented as an array of a plurality of logic gates, or may be implemented as a combination of a general-purpose microprocessor (110) and a memory (120) storing a program that can be executed on the microprocessor (110).

[0060] The aforementioned memory (120) and processor (110) may be included in the control unit (100), and the control unit (100) may control the aforementioned components to determine the target cumulative reservation number.

[0061] Specifically, the control unit (100) can extract features for predicting room demand from room data to generate preprocessed data, use the preprocessed data as learning data, and generate a reservation count prediction model by combining a first learning model based on past reservation data and a second learning model based on past reservation data and future reservation data. That is, the control unit (100) can generate preprocessed data for efficient learning of the learning model, and generate a reservation count prediction model by combining the first learning model and the second learning model.

[0062] In this case, the first learning model may include an exogenous factor-aware model (SARIMAX) generated based on past reservation data from a preset reference date to a past reference time. Furthermore, the second learning model may include an additive advance booking model (ABM) generated based on past reservation data from a preset reference date to a past reference time and future reservation data from the reference date to a future reference time.

[0063] In addition, the first learning model may further include a model that improves prediction accuracy by adding exceptional variables such as public holidays as exogenous variables, including a model that considers exogenous factors (SARIMAX), and the second learning model may further include a model that can consider data on the number of pending reservations during the lead time of future stay dates.

[0064] In addition, the control unit (100) can generate a reservation number prediction model by deriving weights according to the degree to which the output values ​​of the first learning model and the output values ​​of the second learning model are similar to the actual number of reservations.

[0065] Here, the control unit (100) can improve the prediction accuracy of the overall reservation number prediction model by giving a greater weight to a model whose predicted result value is similar to the actual value among the first learning model and the second learning model.

[0066] In addition, the control unit (100) can derive a target cumulative number of reservations as a means for predicting room demand, and the target cumulative number of reservations can be derived by considering the target cumulative reservation distribution.

[0067] Specifically, the technology for predicting room demand must be able to consider the cumulative reservation distribution based on the characteristics of the accommodation facility and whether it is a weekday or weekend to improve prediction accuracy. To consider the cumulative reservation distribution, the control unit (100) can broadly categorize accommodations into two types: those with a steady flow of reservations over a long lead time (e.g., in tourist areas or on weekends), and those with a high concentration of reservations within the near past, starting from the day of check-in (e.g., in entertainment areas or on weekdays).

[0068] In the process of dynamically adjusting the room sales price, the price is adjusted so that the target number of reservations is received according to the characteristics and seasonality of the business during the lead time. Therefore, the server device (1) according to one embodiment can determine the cumulative distribution of reservations during the lead time for each business in order to set the target number of reservations for each lead time.

[0069] Specifically, the control unit (100) can derive a cumulative reservation distribution by increasing the weight of past reservation data as it approaches the check-in date during the lead time. For example, the control unit (100) can derive a final distribution by setting the weight for reservations made within the last four weeks to be greater than the weight for reservations made between the last 12 weeks and the last 4 weeks during a specific lead time.

[0070] Accordingly, in the process of deriving the cumulative reservation distribution, the control unit (100) can set the influence of the most recent reservation to be the greatest, and can give a large weight to the point in time closest to the current point in time, assuming that the point in time is most similar to the current reservation environment.

[0071] In addition, the control unit (100) can derive a cumulative distribution of business reservations and a cumulative distribution of business reservations by lead time and day of the week based on the cumulative distribution of reservations, and can derive the maximum value among the cumulative distribution of business reservations and the cumulative distribution of business reservations derived by day of the week as the target cumulative distribution of reservations by day of the week.

[0072] Thereafter, the control unit (100) can derive the target cumulative reservation number by multiplying the target market share by the predicted number of reservations and the minimum value among the standard number of rooms sold by the target cumulative reservation distribution by day of the week.

[0073] As another example, the control unit (100) may derive a cumulative distribution of business reservations and a cumulative distribution of business reservations by lead time based on the cumulative distribution of reservations, and may derive a target cumulative distribution of reservations by multiplying the maximum value of the cumulative distribution of business reservations and the cumulative distribution of business reservations by a weight for each day of the week.

[0074] Here, the weight by day of the week can mean a weight that gives a large value to days of the week with many reservations from Monday to Sunday.

[0075] Thereafter, the control unit (100) can derive the target cumulative reservation number by multiplying the target market share by the predicted number of reservations and the target cumulative reservation distribution by the minimum value among the standard number of rooms sold.

[0076] Here, the target cumulative number of reservations may mean a result value predicted by the server device (1) according to one embodiment regarding the demand for rooms, and may be used as a basis value for adjusting the sales price of rooms.

[0077] Additionally, the room types included in the room data may include soft block and hard block types, and if the room type is a hard block type, the target cumulative reservation distribution may be derived based on a separate algorithm.

[0078] Specifically, if the room type included in the room data is a hard block type, the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations of the business can be derived by lead time and day of the week, and the maximum value of the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations derived by day of the week can be used to derive the target cumulative reservation distribution by day of the week.

[0079] For another example, if the room type included in the room data is a hard block type, the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations for the business can be derived by lead time, and the target cumulative reservation distribution can be derived by multiplying the maximum value of the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations by the weight for each day of the week.

[0080] Accordingly, demand can be predicted more accurately by considering the characteristics of the pre-purchase quantity included in the hard block type among the room reservation types.

[0081] In addition, the control unit (100) can derive the target cumulative number of reservations by multiplying the target cumulative reservation distribution by the purchase amount per day of the hard block type room. Since the hard block type has the characteristic that the purchase amount per day of the room is fixed, the room demand can be predicted by considering the difference from the soft block.

[0082] Thereafter, the control unit (100) can transmit the target cumulative number of reservations to an external device (2) through the communication unit (200) to predict room demand and provide the target cumulative number of reservations to a user who wants to dynamically adjust the price of the room.

[0083] FIG. 2 illustrates input and output values ​​of a prediction model utilized in a server device (1) according to one embodiment.

[0084] The control unit (100) can train a prediction model (b) using potential room demand data, reservation information data, and seasonality data as training data (a). Here, training may include a fitting process that adjusts parameters of a statistical time series model and builds a model that fits the time series data.

[0085] When the control unit (100) includes an artificial intelligence-only processor (110) (e.g., NPU) for training the prediction model (b), the processor (110) can train the artificial neural network by utilizing the weight data stored in the memory (120) as training data for the artificial neural network.

[0086] Examples of learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0087] An artificial neural network can be composed of multiple neural network layers. Each of the multiple neural network layers has multiple weight values ​​and performs neural network operations through operations between the calculation results of the previous layer and the multiple weights. The multiple weights of the multiple neural network layers can be optimized based on the learning results of an artificial intelligence model. For example, during the learning process, the multiple weights can be updated to reduce or minimize the loss or cost values ​​obtained from the artificial intelligence model.

[0088] The artificial neural network may include a deep neural network (DNN), for example, a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN), a restricted boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), or a deep Q-network (900) (Deep Q-Networks), but is not limited to the examples described above.

[0089] The control unit (100) can learn the relationship between the potential demand data of the room, reservation information data, and seasonality data and the target cumulative number of reservations based on the selected artificial intelligence model.

[0090] That is, the control unit (100) can input potential demand data, reservation information data, and seasonality data of the room, and output room sales price adjustment data (c) based on a prediction model (b) including prediction techniques such as classification and regression.

[0091] FIG. 3 and FIG. 4 illustrate a schematic flowchart of a server device (1) according to one embodiment predicting room demand and deriving a target cumulative number of reservations.

[0092] The configurations 101 to 107 in FIGS. 3 and 4 are implemented in the form of software blocks, and can be stored in memory (120) and executed by the processor (110).

[0093] First, referring to FIG. 3, the control unit (100) can receive room data via the communication unit (200) and store the received room data in the raw data DB (121). At this time, the room data can include basic information, price information, reservation information, room location information, guest feedback information, service information, time-based reservation information, and availability information.

[0094] Specifically, basic information may include room number or name, room type (e.g., standard, suite, deluxe), room size and capacity (e.g., single bed, queen bed, twin bed), price information may include daily or hourly price information, seasonal price change information, discount information, and promotional information, and reservation information may include reservation status information, reservation date information, reservation time information, reservation customer information (e.g., name, contact information, email, etc.), and previous reservation history (preference) information.

[0095] Additionally, room location information may include the name of the accommodation, location information (city, region, etc.), nearby tourist attractions and facility information, and guest feedback information may include previous customer review information, evaluation information, and information on improvements made through feedback and reviews.

[0096] Additionally, service information may include information on in-room facilities (wireless internet, TV, refrigerator, bathroom, etc.), information on accommodation facilities (restaurant, swimming pool, fitness center, etc.), and information on room services and auxiliary facilities, and time-based reservation information and availability information may include information on availability of rooms at a specific time period, and information on the reserved time period and date.

[0097] The control unit (100) may perform feature engineering in the feature engineering execution unit (101) to process the guest room data received through the communication unit (200) into learning data. Feature engineering may refer to a task of improving the performance of a prediction model and enhancing prediction accuracy by appropriately modifying, selecting, or generating data characteristics or features in data analysis.

[0098] The control unit (100) can store features generated by feature engineering in a feature data DB (122) and perform feature selection in the feature data DB (122).

[0099] Specifically, the feature selection unit (102) of the control unit (100) can select suitable features to predict room demand. Feature selection is one of the data preprocessing processes used in machine learning and data analysis, and can refer to the task of selecting the most important features (characteristics) for model training and removing unnecessary features.

[0100] The control unit (100) can check the entire feature list extracted from the raw data DB (121) and evaluate feature importance to select suitable features for room demand prediction. That is, the control unit (100) can evaluate feature importance to determine which feature among various features related to room demand prediction plays the most important role in demand prediction.

[0101] As described above, the control unit (100) can utilize an exogenous factor consideration model (SARIMAX) as a first learning model to generate a reservation number prediction model, and can utilize an additive advance booking model (ABM) as a second learning model.

[0102] At this time, the feature selection unit (102) of the control unit (100) can select the most important variable that can improve model performance among the given exogenous variables. For example, the control unit (100) can analyze the correlation between the exogenous variable and time series data to determine which variables are important, evaluate the impact of a specific variable on the model's prediction, and measure the importance of the variable.

[0103] Additionally, the control unit (100) can consider the size or statistical significance of the regression coefficient, and the control unit (100) can perform feature selection by comparing the performance between a model including a specific variable and a model excluding the variable.

[0104] The control unit (100) can perform feature selection and store the selected features in a metadata DB (123), and the metadata DB (123) can mean a database that stores and manages metadata related to a project for predictive models and demand forecasting.

[0105] Thereafter, the prediction model fitting unit (103) of the control unit (100) can fit the prediction model to the given data. Specifically, the control unit (100) can fit the parameters of the prediction model to the data so that the model can operate well on the given data, and can adjust the parameters so that the model can generalize predictions for the given data, including model learning.

[0106] Specifically, the control unit (100) represents an optimization process for the coefficients and order of the prediction model, and the optimization process may mean a process of optimizing the performance of the prediction model by fitting it to specific time series data.

[0107] At this time, the order and coefficients of the prediction model optimized by the control unit (100) may include AR (AutoRegressive) and MA (Moving Average) orders, difference orders, Seasonal AR and Seasonal MA orders, Seasonal difference orders, and exogenous variables.

[0108] The control unit (100) can optimize model performance through predictive model fitting and find optimal parameter values ​​to enable the model to better generalize to data. Thereafter, the control unit (100) can store the parameters for which fitting has been performed in the metadata DB (123).

[0109] The predictive model tuning unit (104) of the control unit (100) can perform parameter tuning to further improve the performance of the learned model. Specifically, the control unit (100) can find an optimal parameter combination to improve model performance, and can select parameters that maximize the model's generalization performance using a validation algorithm, such as a cross-validation algorithm.

[0110] Thereafter, the control unit (100) can store the tuned prediction model based on the optimized parameters in the prediction model DB (124).

[0111] Next, referring to FIG. 4, the predicted reservation number derivation unit (105) of the control unit (100) can derive the predicted reservation number for each lead time by combining the first learning model and the second learning model with specific weights, which will be described in detail later in FIG. 5.

[0112] Afterwards, the target cumulative reservation distribution derivation unit (106) of the control unit (100) can derive the reservation cumulative distribution (Reservation Arrival Cumulative Distribution, RACD) during the lead time and then derive the target cumulative reservation distribution of the company.

[0113] Specifically, the server device (1) according to one embodiment can consider the cumulative distribution of reservations (RACD) during the lead time to derive the target cumulative number of reservations, and can improve the accuracy of prediction by reflecting the cumulative distribution of reservations by business and commercial district during the lead time in the system.

[0114] The control unit (100) can reflect weights such that the most recent reservation cumulative distribution (RACD) from the check-in date has a greater influence on the final distribution. For example, for confirmed reservation data from the past 12 weeks, the control unit (100) can assign a weight of 0.6 to reservations from the past 4 weeks, a weight of 0.3 to reservations from the past 5-8 weeks, and a weight of 0.1 to reservations from the past 9-12 weeks.

[0115] In addition, the control unit (100) can calculate the reservation cumulative distribution (RACD) for each day of the week from Monday to Sunday, and for non-weekend holidays, the reservation cumulative distribution (RACD) of the day of the week with the most similar reservation cumulative distribution (RACD) can be used for each company.

[0116] Accordingly, the control unit (100) can derive a target cumulative reservation distribution, and can derive the target cumulative reservation distribution differently depending on whether the room type is a soft block or a hard block.

[0117] Specifically, the control unit (100) can derive a cumulative distribution of business reservations and a cumulative distribution of commercial reservations by lead time based on the cumulative distribution of reservations when the room type is a soft block, and can derive a target cumulative reservation distribution using the maximum value of the cumulative distribution of business reservations and the cumulative distribution of commercial reservations, which can be expressed as in the following mathematical expression 1.

[0118] [Mathematical Formula 1]

[0119] Target cumulative reservation distribution for soft block rooms = MAX{cumulative distribution of business reservations, cumulative distribution of commercial reservations}

[0120] Next, if the room type is a hard block, the control unit (100) can derive the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations by lead time, and can derive the target cumulative reservation distribution by multiplying the maximum value of the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations by the weight for each day of the week, which can be expressed as in the following mathematical expression 2.

[0121] [Equation 2]

[0122] Target cumulative reservation distribution for hard block rooms = Weight by day of the week * MAX {company's cumulative distribution of hard block reservations, commercial area reservation cumulative distribution}

[0123] Thereafter, in the target cumulative reservation number derivation unit (107) of the control unit (100), in the case of a soft block room type, the target cumulative reservation number can be derived by multiplying the target market share by the result of multiplying the predicted number of reservations and the minimum value among the standard number of rooms sold by the target cumulative reservation distribution, and can be expressed as in the following mathematical expression 3.

[0124] [Equation 3]

[0125] Target cumulative reservations for soft block rooms = MIN{(predicted number of reservations * target market share), number of rooms sold by company} * target cumulative reservation distribution in Equation 1

[0126] For example, the target market share can be determined by multiplying the current market share by a certain ratio, and if the control unit (100) has a current market share of 10%, the target market share can be determined as 11% (10*1.1). In addition, the control unit (100) can determine the current market share as a weighted average of the market shares of the most recent 24 weeks. For example, the market share of the most recent 4 weeks can be weighted as 40%, the market share of the most recent 5 to 8 weeks as 30%, and the market share of the most recent 9 to 24 weeks as 30%, and the market share can be determined by calculating the average.

[0127] In addition, the target cumulative reservation number derivation unit (107) of the control unit (100) can derive the target cumulative reservation number by multiplying the target cumulative reservation distribution by the purchase amount per day of room stay of the hard block type, and can be expressed as in the following mathematical expression 4.

[0128] [Equation 4]

[0129] Target cumulative reservations for hard block rooms = Purchase volume per stay day * Target cumulative reservation distribution in Equation 2

[0130] Thereafter, the control unit (100) can transmit the target cumulative reservation number to an external device (2), such as a user terminal, through the communication unit (200).

[0131] According to a server device (1) according to one embodiment, the target cumulative reservation number can be derived based on the target cumulative reservation distribution, so that by dynamically adjusting the room sales price based on this, the maximum profit can be provided to the accommodation business owner.

[0132] Although each step described above in FIGS. 3 and 4 is depicted sequentially, each step may be performed in parallel by the CPU and GPU.

[0133] FIG. 5 illustrates a process of combining multiple learning models in a server device (1) according to one embodiment.

[0134] Referring to FIG. 5, the control unit (100) can derive the number of reservations to predict the target cumulative number of reservations, and can utilize multiple learning models to derive the number of reservations.

[0135] The control unit (100) can generate a prediction model that exhibits optimal performance by combining the first learning model (a) and the second learning model (b). Specifically, the first learning model (a) and the second learning model (b) may refer to models that exhibit different performance.

[0136] For example, the first learning model (a) may include an exogenous factor consideration model (SARIMAX) that utilizes confirmed reservation data for past stay dates, and the second learning model (b) may include an additive advance booking model (ABM) that utilizes confirmed reservation data for past stay dates and predicted reservation data for future stay dates.

[0137] The control unit (100) can use confirmed reservation data for past stay dates as learning data to train the first learning model (a). For example, the control unit (100) can use data from January 1, two years prior to the current date, to the day before to predict the number of reservations for 120 future stay dates.

[0138] Specifically, assuming that the current date is October 24, 2023, the control unit (100) can use reservation data for stay dates from January 1, 2021 to October 23, 2023 to predict the number of reservations for stay dates for 120 days in the future, including October 24, 2023.

[0139] In addition, the control unit (100) can utilize unconfirmed reservation data for future stay dates in addition to confirmed reservation data for past stay dates to train the second learning model (b).

[0140] Specifically, assuming the current date to be October 24, 2023, the control unit (100) can use the reservation data by lead time of the stay date from (2023 / 10 / 23 - 8 weeks) to (2023 / 10 / 23 + 120 days) to predict the total number of reservations by stay date for the next 120 days including October 24, 2023.

[0141] Thereafter, the control unit (100) can derive a weight (Wt*) that most ideally combines the first learning model (a) and the second learning model (b) through learning. That is, the control unit (100) can assign a larger weight (Wt*) to a model that predicts closer to the actual value, which is the actual number of reservations, among the first learning model (a) and the second learning model (b) in units of lead time (t).

[0142] The control unit (100) can utilize the actual number of reservations for the past 12 weeks of stay dates and the predicted values ​​of the first learning model (a) and the second learning model (b) from 120 days prior to the lead time for the stay dates to derive the weight (Wt*).

[0143] Accordingly, the control unit (100) can derive the predicted number of reservations (d) based on the weight (Wt*)(c-1) applied to the first learning model (a) and the weight (1-Wt*)(c-2) applied to the second learning model (b), and since the result value of each learning model may vary depending on the room data and the prediction accuracy may vary, a plurality of prediction models can be combined to derive an optimal prediction model.

[0144] Fig. 6 illustrates the result of combining multiple learning models by a server device (1) according to one embodiment.

[0145] Referring to FIG. 6, the control unit (100) can generate an optimal prediction model by combining multiple learning models, and shows the results according to the optimal prediction model.

[0146] Specifically, in Fig. 6, the horizontal axis may represent the past lead time (-60 days) based on the reference date, which is the accommodation date (day 0), and the vertical axis may represent the mean absolute percentage error, which may represent an indicator for evaluating the performance of the prediction model.

[0147] At this time, the Mean Absolute Percentage Error on the vertical axis has a smaller value, which means that the performance of the prediction model is excellent because the error from the actual value is relatively small.

[0148] The graph (a) of FIG. 6 may refer to the average absolute percentage error by lead time derived by the first learning model, the graph (b) may refer to the average absolute percentage error by lead time by the second learning model, and the graph (c) may refer to the average absolute percentage error by the final prediction model combining the first learning model and the second learning model.

[0149] In the room data utilized in Fig. 6, the average absolute percentage error of the graph (a) increases significantly around lead time -50, so the control unit (100) can assign a small weight (Wt*) to the first learning model.

[0150] On the other hand, since the average absolute percentage error of the (b) graph in the overall lead time is lower than that of the (a) graph, the control unit (100) can derive the predicted number of reservations by giving a large weight (1-Wt*) to the second learning model.

[0151] The control unit (100) can generate a final prediction model by assigning different weights to each of the first learning model and the second learning model, and the graph (c) by the generated final prediction model has a relatively small average absolute percentage error value and a small deviation in the average absolute percentage error, so it can be confirmed that the performance is superior to that of a case where prediction is made using only the first learning model or the second learning model.

[0152] FIG. 7 illustrates a flowchart regarding a control method of a server device (1) according to one embodiment, and FIG. 8 continues to illustrate a flowchart regarding a control method of a server device (1) according to one embodiment.

[0153] Referring to FIG. 7, the control unit (100) can receive room data via the communication unit (200) (700). Thereafter, the control unit (100) can extract features for room demand prediction from the room data and generate preprocessing data (710).

[0154] The control unit (100) can generate a first learning model and a second learning model that use preprocessed data as learning data (720), wherein the first learning model and the second learning model may mean a prediction model that uses time series data, or may mean a learning model that uses machine learning.

[0155] Thereafter, the control unit (100) can generate a reservation number prediction model by combining the first learning model and the second learning model (730), and input room data into the reservation number prediction model to derive the predicted reservation number as an output value (740).

[0156] Next, referring to FIG. 8, the control unit (100) can derive a cumulative reservation distribution by increasing the weight of past reservation data as it approaches the check-in date during the lead time (800). Here, the lead time can refer to a past date prior to the check-in date, and by considering the cumulative reservation distribution during the demand forecasting process, the period of concentrated reservations can be reflected, thereby improving forecast accuracy.

[0157] Thereafter, the control unit (100) can determine whether the room type included in the room data is a hard block type (810). If the room type included in the room data is a soft block type rather than a hard block type (No in 810), the control unit (100) can derive a cumulative distribution of business reservations and a cumulative distribution of commercial reservations by lead time.

[0158] Here, the cumulative distribution of business reservations refers to the distribution of when reservations are most frequently made for each business regardless of the business district, and the cumulative distribution of business district reservations refers to the distribution of when reservations are most frequently made in the business district to which each business belongs.

[0159] The control unit (100) can derive a target cumulative reservation distribution by multiplying the maximum value among the cumulative distribution of business reservations and the cumulative distribution of business reservations by a weight for each day of the week (830), and the control unit (100) can derive a target cumulative reservation number by multiplying the result of multiplying the target market share by the predicted number of reservations and the minimum value among the standard number of rooms sold by the target cumulative reservation distribution (840). However, as described above, the control unit (100) can derive a cumulative distribution of business reservations and a cumulative distribution of business reservations by lead time and day of the week based on the cumulative distribution of reservations, and can also derive the maximum value among the cumulative distribution of business reservations and the cumulative distribution of business reservations derived for each day of the week as a target cumulative reservation distribution for each day of the week.

[0160] On the other hand, if the room type included in the room data is a soft block type rather than a hard block type (example of 810), the control unit (100) can derive the cumulative distribution of the hard block reservations and the cumulative distribution of the commercial area reservations by lead time (850).

[0161] Thereafter, the control unit (100) can derive a target cumulative reservation distribution by multiplying the maximum value among the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations by a weight for each day of the week (860). Similarly, as described above, if the room type is a hard block type, the control unit (100) can derive the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations for each lead time and day of the week, and can derive the target cumulative reservation distribution for each day of the week by the maximum value among the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations derived for each day of the week.

[0162] The control unit (100) can derive the target cumulative number of reservations by multiplying the purchase amount per day of a hard block type room by the target cumulative reservation distribution (870).

[0163] In this way, the server device (1) according to one embodiment can determine the target cumulative number of reservations differently depending on whether the room type is a soft block or a hard block, thereby having the effect of minimizing the error between the actual value and the predicted value.

[0164] Meanwhile, the disclosed embodiments may be implemented in the form of a recording medium storing computer-executable instructions. The instructions may be stored in the form of program code, and when executed by a processor, may generate program modules to perform the operations of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.

[0165] Computer-readable storage media include all types of storage media that store instructions that can be deciphered by a computer. Examples include read-only memory (ROM), random access memory (RAM), magnetic tape, magnetic disks, flash memory, and optical data storage devices.

[0166] Additionally, a computer-readable recording medium may be provided in the form of a non-transitory storage medium. Here, the term "non-transitory storage medium" simply means a tangible device that does not contain signals (e.g., electromagnetic waves). This term does not distinguish between cases where data is permanently stored in the storage medium and cases where data is temporarily stored. For example, a "non-transitory storage medium" may include a buffer in which data is temporarily stored.

[0167] According to one embodiment, the method according to the various embodiments disclosed in the present document may be provided as included in a computer program product. The computer program product may be traded as a product between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable recording medium (e.g., compact disc read only memory (CD-ROM)), or may be distributed online (e.g., downloaded or uploaded) via an application store (e.g., Play Store™) or directly between two user devices (e.g., smartphones). In the case of online distribution, at least a portion of the computer program product (e.g., a downloadable app) may be at least temporarily stored or temporarily generated in a machine-readable recording medium, such as a memory of a manufacturer's server, an application store's server, or an intermediary server.

[0168] The above has illustrated and described specific embodiments. However, the invention is not limited to the above-described embodiments, and those skilled in the art will appreciate that various modifications and implementations can be made without departing from the spirit and scope of the invention as set forth in the claims below.

Claims

1. A communication unit that receives room data for training a machine learning model related to room demand prediction; A memory for storing the above room data; and Extract features for predicting room demand from the above room data to create preprocessed data, The above preprocessed data is used as learning data, and a reservation number prediction model is created by combining a first learning model based on past reservation data and a second learning model based on the past reservation data and future reservation data. A server device including a control unit that inputs the room data into the reservation number prediction model to derive a predicted reservation number as an output value, and derives a target cumulative reservation number based on the predicted reservation number and the cumulative distribution of reservations during the lead time.

2. In claim 1, The above control unit, A server device that generates a reservation number prediction model by deriving weights based on the degree to which the output values ​​of the first learning model and the output values ​​of the second learning model are similar to the actual number of reservations.

3. In claim 1, The above control unit, A server device that derives the cumulative distribution of reservations by increasing the weight of the past reservation data as it gets closer to the stay date during the lead time.

4. In claim 3, The above control unit, A server device that derives a cumulative distribution of business reservations and a cumulative distribution of business reservations by lead time and day of the week based on the cumulative distribution of reservations above, and derives the maximum value among the cumulative distribution of business reservations and the cumulative distribution of business reservations above derived by day of the week as a target cumulative distribution of reservations by day of the week.

5. In claim 4, The above control unit, A server device that derives the target cumulative reservation number by multiplying the target market share by the predicted number of reservations and the minimum value among the standard number of rooms sold by the target cumulative reservation distribution by day of the week.

6. In claim 3, The above control unit, A server device that derives a cumulative distribution of hard block reservations and a cumulative distribution of commercial reservations for a business by lead time and day of the week based on the fact that the room type included in the above room data is a hard block type, and derives the maximum value of the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations derived by day of the week as a target cumulative reservation distribution by day of the week.

7. In claim 6, The above control unit, A server device that derives the target cumulative reservation number by multiplying the purchase amount per day of the room of the hard block type by the target cumulative reservation distribution.

8. In claim 1, The above first learning model is, A server device including an exogenous factor consideration model (SARIMAX) generated based on the past reservation data for a past reference time from a preset reference date.

9. In claim 1, The second learning model above is, A server device including an additive advance booking model (ABM) generated based on past reservation data from a preset reference date to a past reference time and future reservation data from the reference date to a future reference time.

10. In claim 1, The above control unit, A server device that transmits the target cumulative reservation number to an external device through the above communication unit.

11. Receive room data for training a machine learning model related to room demand prediction; Extract features for predicting room demand from the above room data to generate preprocessed data; Using the above preprocessed data as learning data, a reservation number prediction model is created by combining a first learning model based on past reservation data and a second learning model based on the past reservation data and future reservation data; By inputting the above room data into the above reservation number prediction model, the predicted reservation number is derived as an output value; A control method of a server device, comprising: deriving a target cumulative number of reservations based on the above predicted number of reservations and the cumulative distribution of reservations during the lead time; 12. In claim 11, Creating the above reservation number prediction model is: A control method for a server device, comprising: generating a reservation number prediction model by deriving weights according to the degree to which the output values ​​of the first learning model and the output values ​​of the second learning model are similar to the actual number of reservations.

13. In claim 11, A control method of a server device further comprising: increasing a weight of the past reservation data as it gets closer to the check-in date during the lead time, thereby deriving the cumulative distribution of the reservations.

14. In claim 13, A control method of a server device further comprising: deriving a cumulative distribution of business reservations and a cumulative distribution of business reservations by lead time and day of the week based on the cumulative distribution of reservations, and deriving the maximum value of the cumulative distribution of business reservations and the cumulative distribution of business reservations by day of the week as a target cumulative reservation distribution by day of the week.

15. In claim 14, Deriving the above target cumulative reservation number is as follows: A control method for a server device, comprising: multiplying the target market share by the predicted number of reservations and the minimum value among the standard number of rooms sold by the target cumulative reservation distribution by the day of the week to derive the target cumulative reservation number.

16. In claim 13, Deriving the above target cumulative reservation distribution is: A control method for a server device, comprising: deriving a cumulative distribution of hard block reservations and a cumulative distribution of commercial reservations for a business by lead time and day of the week based on the fact that the room type included in the above room data is a hard block type; and deriving the maximum value of the cumulative distribution of hard block reservations and the cumulative distribution of commercial reservations derived by day of the week as a target cumulative reservation distribution by day of the week.

17. In claim 16, Deriving the above target cumulative reservation number is as follows: A control method for a server device that derives the target cumulative reservation number by multiplying the daily purchase amount of the above hard block type room by the target cumulative reservation distribution by day of the week.

18. In claim 11, The above first learning model is, A control method of a server device including an exogenous factor consideration model (SARIMAX) generated based on the past reservation data for a past reference time from a preset reference date.

19. In claim 11, The second learning model above is, A control method for a server device including an additive advance booking model (ABM) generated based on past reservation data from a preset reference date to a past reference time and future reservation data from the reference date to a future reference time.

20. In claim 11, A control method of a server device further comprising: transmitting the target cumulative reservation number to an external device through a communication unit;

Images