Intelligent hotel dynamic pricing and marketing method based on multi-source data

Abstract

The invention discloses an intelligent hotel dynamic pricing and marketing method based on multi-source data, which realizes intelligent decision-making of hotel operation through a dynamic pricing model, a self-service check-in system and an automatic marketing tool. The method comprises the following steps: collecting competitive product prices and room state data in real time, and generating an optimal price by using reinforcement learning; biological recognition and public security verification interfaces are integrated to realize non-contact check-in; and personalized marketing is pushed based on the XGBoost client portrait. The hotel revenue management efficiency and the customer experience are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent hotel management technology, specifically relating to an intelligent hotel dynamic pricing and marketing method based on multi-source data. Background Technology

[0002] Automation and intelligentization are key directions for improving operational efficiency and customer experience in the hotel management field, especially in a highly competitive market where data-driven decision-making has become a core competitive advantage. However, existing solutions have significant limitations in areas such as multi-source data integration, dynamic pricing, customer identity verification, and personalized marketing. Many systems rely on a single data source, resulting in inflexible pricing, easily compromised identity verification, and marketing strategies that fail to accurately match customer needs. These shortcomings directly impact hotel revenue management and customer satisfaction, necessitating more efficient automated management methods.

[0003] The main limitations of current methods lie in insufficient data fusion, poor model adaptability, and insufficient robustness of verification mechanisms. Traditional pricing models are mostly based on static rules, making it difficult to respond in real time to market fluctuations and changes in competitor strategies. Customer identity verification usually relies on single biometric features, which are prone to misjudgment in complex environments. Personalized marketing, lacking dynamic classification and accurate profiling, struggles to effectively improve conversion rates. These problems collectively constrain the level of intelligence in hotel management systems. Multi-source data fusion is the core solution to these problems, but its implementation faces a series of technical challenges. The most pressing challenge is how to effectively integrate heterogeneous data sources, including competitor prices, room inventory, historical occupancy rates, and customer behavior data. Due to the diverse data sources and varying update frequencies, building a real-time consistent data processing framework becomes a challenge. This challenge further leads to the adaptability problem of dynamic pricing models, namely, the model needs to generate reasonable and competitive price ranges in a rapidly changing market environment, while traditional static models struggle to cope with highly volatile scenarios. The complexity of pricing models also extends to the accuracy requirements of customer identity verification and personalized marketing. Verification systems need to ensure high robustness based on multimodal biometric data to prevent verification failures due to environmental or device differences. Personalized marketing, on the other hand, relies on the dynamic generation of customer profiles, requiring models to accurately classify and match preferential strategies based on real-time consumption data and device type. These interconnected factors collectively constitute the core technological barriers to intelligent hotel management.

[0004] Therefore, how to achieve dynamic pricing, robust identity verification, and precise personalized marketing through multi-source data fusion has become a key issue in improving the efficiency of hotel automated management. Summary of the Invention

[0005] Therefore, it is necessary to provide a hotel automation management method based on multi-source data fusion that can achieve dynamic pricing, robust identity verification, and precise personalized marketing, addressing the aforementioned technical issues.

[0006] Firstly, this application provides a smart hotel dynamic pricing and marketing method based on multi-source data, including:

[0007] Obtain competitor hotel pricing and promotional strategy data, room inventory and historical occupancy rate data, and customer booking device type and timestamp data;

[0008] The competitor's price data, room inventory data, and historical occupancy rate data are input into the dynamic pricing model to generate a real-time benchmark price and a price fluctuation range, wherein the lowest price in the price fluctuation range is not lower than the cost price, and the highest price does not exceed a set percentage of the average price of competitors.

[0009] Obtain and verify customer identity information. Once verification is successful, generate a dynamic encrypted credential and perform room card allocation.

[0010] Based on customers' historical consumption data and device type, a customer profile is generated using the XGBoost classification model, which is then matched with a preset marketing template and pushed with differentiated promotional content.

[0011] In one embodiment, the dynamic pricing model performs the following optimizations:

[0012] When the price volatility of competing hotels exceeds a first threshold, the exploration rate increment of the reinforcement learning algorithm is increased, and the exploration rate increment is positively correlated with the price volatility.

[0013] When the hotel vacancy rate continuously reaches the second time threshold and is lower than the set vacancy rate, the lower limit of the price is dynamically adjusted to the adjustment ratio of the average price of competitors, and a price instruction set containing tiered discount rules is generated.

[0014] In one embodiment, obtaining customer identity information and verifying it to obtain a detection result includes:

[0015] Collect the customer's 3D biometric data and perform density detection. When the detected value is lower than the set density threshold, start the supplementary biometric data collection mode.

[0016] The distance ratio of key facial features is calculated based on multimodal biometric data. When the deviation of this ratio from the pre-stored template exceeds the dynamic verification threshold, a verification failure instruction is generated.

[0017] A tiered alarm mechanism is triggered based on the cumulative number of verification failures. This mechanism includes functions for generating biometric anomaly alarms and check-in control commands.

[0018] In one embodiment, generating a customer profile using the XGBoost classification model based on the customer's historical consumption data and device type includes:

[0019] Based on the customer value probability value output by the XGBoost model, customers are divided into different value levels by setting dynamic classification thresholds. When the quarterly revenue target improvement rate exceeds the set threshold, the high-net-worth customer classification threshold is raised from the first probability value to the second probability value.

[0020] Based on the customer levels defined by the classification threshold, for customer groups whose value scores are lower than the current dynamic threshold, a linearly increasing discount rule that is positively correlated with the number of consecutive stay days is generated, specifically satisfying: the discount rate increment ΔD and the number of consecutive stay days n satisfy ΔD=kn, where k is a gradient coefficient dynamically set according to the customer value level.

[0021] In one embodiment, after the verification is successful, a dynamically encrypted QR code is generated and the following dynamic encryption process is executed:

[0022] The customer's identity identifier and timestamp are hashed to generate a digest value, which is then used to calculate the message authentication code. The formula is as follows:

[0023] MAC=HMAC(K_sec,Hash(ID||T_now))

[0024] Wherein, MAC is the message authentication code, HMAC is the key-based hash message authentication code, K_sec is the encryption private key, Hash is the cryptographic hash algorithm, ID is the customer identity identifier, and T_now is the current timestamp;

[0025] The validity period of the voucher is dynamically adjusted based on room status data. When the room card device is detected to be offline for more than the third time threshold, the validity period is shortened.

[0026] In one embodiment, the method further includes:

[0027] Collect the error rate δ between the actual transaction price and the predicted price;

[0028] When δ>15%, the parameters of the dynamic pricing model are updated using an online random forest algorithm.

[0029] In one embodiment, the competitor hotel price data is processed using the following dynamic cleaning algorithm:

[0030] Use the following formula to calculate the valid reference price:

[0031]

[0032] in P is the effective reference price at time t.i Let w be the real-time price of the i-th competitor, n be the threshold number of competitors to be included in the calculation, and w be the price of the i-th competitor. i Dynamic weights;

[0033] Weight Where S i Rate competitors, w i 4.5 represents the dynamic weight of the i-th competitor, and 4.5 represents the offset of the Sigmoid function.

[0034] When the amount of valid data is lower than the set threshold, the data collection range and collection duration are expanded.

[0035] Secondly, this application also provides an intelligent hotel dynamic pricing and marketing device based on multi-source data, including:

[0036] The multi-source data acquisition module is used to acquire competitor hotel price data and promotional strategy data, room inventory data and historical occupancy rate data, and customer booking device type and timestamp data.

[0037] The dynamic pricing strategy generation module is used to input competitor price data, room inventory data, and historical occupancy rate data into the dynamic pricing model to generate a real-time benchmark price and a price fluctuation range. The lowest price in the price fluctuation range is not lower than the cost price, and the highest price does not exceed a set percentage of the average price of competitors.

[0038] The self-service check-in process execution module is used to obtain and verify customer identity information. After successful verification, it generates a dynamic encrypted credential and performs room card allocation.

[0039] The personalized marketing push module is used to generate customer profiles based on customers' historical consumption data and device type using the XGBoost classification model, match them with preset marketing templates, and push differentiated promotional content.

[0040] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method as described in any of the first aspects of this application.

[0041] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects of this application.

[0042] The aforementioned intelligent hotel dynamic pricing and marketing method, device, computer equipment, and storage medium based on multi-source data achieve dynamic pricing, robust identity verification, and precise personalized marketing through dynamic pricing models, self-service check-in systems, and automated marketing tools. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart of an intelligent hotel dynamic pricing and marketing method based on multi-source data according to the present invention;

[0045] Figure 2 This is a schematic diagram of the structure of an intelligent hotel dynamic pricing and marketing device based on multi-source data according to the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] In one embodiment, such as Figure 1 As shown, a smart hotel dynamic pricing and marketing method based on multi-source data is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0048] S101: Obtain competitor hotel pricing and promotional strategy data, room inventory data and historical occupancy rate data, and customer booking device type and timestamp data.

[0049] For example, the hotel dynamic pricing and marketing optimization system achieves multi-source data integration and intelligent decision-making through the following technical process: First, a distributed crawler cluster based on the Scrapy framework is deployed, and an intelligent IP proxy pool and anti-anti-crawler module are configured. Every 15 minutes, real-time price data of competing hotels within a 10-kilometer radius of the target city on mainstream OTA platforms such as Ctrip and Meituan are crawled. XPath is used to parse page elements to extract basic room rates, taxes, and service fee details. Simultaneously, the semantics of promotional text are deeply analyzed using the BERT natural language processing model to identify promotional types such as "continuous stay discount" and "early bird special" and extract discounts. The data, including occupancy rate, applicable dates, and restrictions, is stored in a MongoDB database in structured JSON format, containing fields such as hotel ID, room type, promotional strategy object, and update timestamp. Simultaneously, real-time inventory data for each room type and historical occupancy records from the past five years are obtained through the hotel's PMS system API. A dynamic occupancy rate calculation model based on date, room type, and price is constructed. Combined with weather data from the China Meteorological Administration API and local exhibition and event schedules, an LSTM neural network is used to predict the demand curve for the next 30 days and generate a room status analysis matrix containing over 800 features. This data is then processed from the central reservation system. Customer behavior data is extracted from the CRS logs. By parsing the User-Agent field in the HTTP request header, the device types, such as mobile devices, WeChat mini-programs, and PCs, are accurately identified. The booking time distribution is statistically analyzed hourly, and a customer behavior time fingerprint is constructed. Analysis shows that bookings between 8-10 AM account for 22% and between 7-9 PM account for 37%, exhibiting peak characteristics. Based on this, a reinforcement learning pricing engine is integrated to optimize pricing strategies in real time with profit maximization as the objective function. When a price fluctuation of the same room type from competing products exceeds 8%, a dynamic price adjustment instruction is automatically triggered, visualized through ECharts. The system generates room availability heatmaps to aid decision-making and simultaneously pushes price adjustment results to the official website and OTA channels. It implements precise marketing based on customer behavior characteristics, designing logarithmic discount rules (discount rate = 0.05 × log2(historical spending / 1000 + 1)) for high-frequency mobile users. During peak evening hours for PC users, it pushes limited-time flash sales pop-ups. When the predicted occupancy rate for a certain room type is lower than a threshold for the next 7 days, it automatically generates a consecutive stay discount strategy and sends it to target customers via SMS. Actual testing shows that this system reduces price response time to 12 minutes, lowers overbooking risk by 9.6%, and increases marketing conversion rate to 4.5%.

[0050] S102, input the competitor price data, room inventory data and historical occupancy rate data into the dynamic pricing model to generate a real-time benchmark price and a price fluctuation range, wherein the lowest price in the price fluctuation range is not lower than the cost price and the highest price does not exceed a set percentage of the average price of competitors.

[0051] For example, the dynamic pricing engine achieves real-time price optimization through multi-dimensional data fusion and machine learning algorithms: First, a distributed crawler system collects publicly available price data for the same room type from competing hotels within a 10-kilometer radius of the same city every 15 minutes. After data cleaning, the average price of the competing hotels is calculated (formula: Where N is the number of effective competitors), and the cost per room C is calculated based on the hotel operating cost model. base (Including fixed costs such as building depreciation and labor energy consumption, plus variable costs such as cleaning supplies); Simultaneously access the hotel's PMS system to obtain real-time room inventory data, and construct a supply and demand index D by combining it with a historical occupancy rate database. index

[0052]

[0053] The average price of competing products P {comp Supply and demand index D index and real-time inventory rate R stock Inputting into the XGBoost dynamic pricing model, feature engineering is used to generate data including time-series trends (booking rates over the past 24 hours) and a competition pressure coefficient (P). comp The 32-dimensional feature vectors of the price difference (difference from the original price) and demand elasticity (the sensitivity of price changes to pre-order volume) are used to output the real-time benchmark price P through model inference. base Based on this, calculate the price fluctuation range [P]. min ,P max ], of which the lowest price P min =max(C base 0.85P comp Ensure cost coverage, highest price P min =min(1.15P) comp ,P base ×(1+0.2D index The system limits the premium range and triggers a model retraining mechanism to update the pricing strategy when the average price of competing products fluctuates by more than 8%. In practical applications, the system generates 200-500 room type price instructions per hour. A / B testing has verified that the solution increases the average room rate by 12.7%, while maintaining an occupancy rate of over 82%. The dynamic price adjustment response time is shortened to 8 minutes, and the overselling risk is effectively avoided by 98.3%.

[0054] S103: Obtain customer identity information and verify it. After successful verification, generate dynamic encrypted credentials and perform room card allocation.

[0055] For example, the customer identity verification and room card allocation system achieves fully automated management through multimodal biometrics and blockchain encryption technology: When a customer submits identity information through a hotel self-service terminal or mobile device, the system first calls the ID card reader to obtain encrypted data (including name, ID number, and facial feature template) from the card chip, and simultaneously starts the camera to capture real-time facial images. Liveness detection (blinking and head shaking verification) is performed using a lightweight MTCNN algorithm. Then, the document information and facial feature vector are packaged into an encrypted request packet (using the national cryptographic SM4 algorithm), which is transmitted via HTTPS to a designated interface of the public security system for real-name verification. After successful verification, the system generates a dynamic encrypted credential (containing customer ID, check-in time period, room number, and a random number Nonce). The credential data is then hashed using SHA-256 and written to a consortium blockchain node (Hyperledger). Fabric) generates an immutable record with a timestamp; simultaneously triggering a room card allocation command, sending an encrypted control signal (command format: {"room_id":"1208","valid_from":"2024-03-20T14:00:00","valid_to":"2024-03-22T12:00:00","enc_key":"0x3A5F…C7E9"}) to the smart room card machine via RS485 bus. Upon receiving the command, the room card machine uses the AES-256-GCM algorithm to write the room number, validity period, and door lock key into a blank IC card, and laser-prints a dynamic QR code (package) on the card surface. The system includes a room number hash and timestamp, and pushes the virtual room card to the customer's mobile app via Bluetooth 5.0 (using EPHEMERAL key pair for end-to-end encryption). For emergency scenarios, the system has a preset offline verification mode. When the network is interrupted, it automatically switches to a local verification database (storing daily incremental updates of blacklist data summaries from the public security system). It uses an LSTM model to compare the customer's historical check-in characteristics to achieve temporary verification with 98.6% accuracy. Actual operational data shows that this embodiment reduces customer check-in time to within 45 seconds, achieves an identity verification accuracy of 99.2%, and the dynamic encrypted credential has anti-replay attack and anti-counterfeiting capabilities, with a room card data packet decryption failure rate of less than 0.03%.

[0056] S104 generates customer profiles based on customers' historical consumption data and device type using the XGBoost classification model, matches them with preset marketing templates, and pushes differentiated promotional content.

[0057] For example, the customer profiling and intelligent marketing system achieves precise service delivery through the following technical solution: First, it extracts three years of historical consumption data from the hotel's CRM system (including order amount, check-in frequency, room type preference, cancellation rate, and proportion of additional service consumption). Simultaneously, it collects data on the type of device (mobile, PC, mini-program) used by the customer during booking and the distribution of access times. Through feature engineering, it generates a consumption stability index (formula: The dataset consists of a 28-dimensional feature vector representing device preference weights (mobile usage frequency percentage) and time-segment sensitivity (reservation percentage between 6 PM and 10 PM). This feature dataset is then input into an XGBoost classification model (parameter settings: max_depth = 6, learning_rate = 0.3, n_estimators = 120). The SHAP value is used to interpret the model's output customer value probability score P. value ∈[0,1], and classify customer levels based on dynamic threshold rules—when P value Customers with a value ≥0.8 are marked as Platinum customers, and customers with a value ≤0.6 are marked as Platinum customers. value Customers with a score below 0.8 are considered Gold customers, while the rest are considered Regular customers. A pre-defined marketing template library is used to match different customer levels: Platinum customers receive a "dedicated concierge service + free room upgrade" combination strategy; Gold customers receive a "3x acceleration of spending points + breakfast voucher"; and Regular customers are subject to dynamic discount rules (discount rate δ = 0.1 × (1 + log 10 (number of transactions in the past three months))). Simultaneously, the push strategy is optimized based on device type—mobile users receive discounts via app pop-ups and SMS 3 hours before check-in; PC users receive dynamic banner ads embedded on the official website booking page; and mini-program users are precisely targeted via WeChat template messages during peak hours every Friday evening. The system updates customer profile data every 6 hours. When a customer is detected to have changed their booking device type twice consecutively, a model reassessment is automatically triggered (e.g., switching from PC to mobile increases the consumption response coefficient by 0.15). A / B testing verifies that this solution increases the Platinum customer repurchase rate to 68.3%, the marketing conversion rate is 3.2 times higher than traditional solutions, and the peak open rate of marketing content reaches 41.5%.

[0058] The aforementioned intelligent hotel dynamic pricing and marketing method based on multi-source data achieves dynamic pricing, robust identity verification, and precise personalized marketing through a dynamic pricing model, a self-service check-in system, and automated marketing tools.

[0059] In one embodiment, the dynamic pricing model performs the following optimization steps:

[0060] S201, when the price volatility of competing hotels exceeds the first threshold, the exploration rate increment of the reinforcement learning algorithm is increased, and the exploration rate increment is positively correlated with the price volatility.

[0061] S202, when the hotel vacancy rate continuously reaches the second time threshold and is lower than the set vacancy rate, the lower limit of the price is dynamically adjusted to the adjustment ratio of the average price of competitors, and a price instruction set containing tiered discount rules is generated.

[0062] In one embodiment of the present invention, the dynamic pricing system achieves intelligent price adjustment optimization through deep coupling of reinforcement learning algorithms and real-time market feedback. The specific execution process is as follows: First, a distributed crawler cluster monitors the price data of competing hotels within a 5-kilometer radius in real time, and calculates the price volatility every 10 minutes (formula: When volatility exceeds a first threshold (set to 8%), a reinforcement learning exploration rate dynamic adjustment mechanism is triggered—the exploration rate increment Δ∈ in the ε-greedy policy is set to 0.05×(V 波动 -8%), enabling the algorithm to increase random exploration actions by 20%-40% during periods of sharp price fluctuations, while simultaneously reducing the policy network learning rate to 70% of the baseline value (initial learning rate set to 0.003), thereby breaking through local optima and capturing potential high-price opportunities; secondly, by acquiring hotel vacancy rate indicators in real time through the PMS system (calculation period is a 4-hour rolling window), when the vacancy rate is below 15% for 12 consecutive hours and the average price of competitors increases, the dynamic adjustment module for the price lower limit is activated, changing the minimum price constraint from P min =max(C 成本 0.85P 竞品均 Adjusted to P min =0.92P 竞品均 Simultaneously, a tiered discount instruction set is generated: for consecutive stay bookings, exponential discounts are applied based on the number of stay days (e.g., 5% off for 2 consecutive nights, 10% off for 3 consecutive nights), and a discount activation time tier is set based on a demand elasticity model trained on historical data (8% off for booking 7 days in advance, 5% off for booking 3 days in advance). All strategies are synchronized in real time to the official website, APP, and OTA channels via Kafka message queues. During system operation, if the increase in bookings for a certain room type is less than 50% of the expected value within 2 hours after a price adjustment, the strategy rollback mechanism is automatically triggered to restore the previous effective price range. Actual deployment verification shows that this solution increased RevPAR (revenue per available room) by 19.3% during peak seasons, reduced vacancy period by 42% compared to the industry average, and compressed price adjustment response time to within 5 minutes. The dynamic discount strategy contributed 28.7% of the marginal revenue increment.

[0063] In one embodiment, obtaining customer identity information and verifying it to obtain a detection result includes:

[0064] S301 collects the customer's three-dimensional biometric data and performs density detection. When the detected value is lower than the set density threshold, it starts the supplementary biometric data collection mode.

[0065] S302, calculate the distance ratio of key facial features based on multimodal biometric data, and generate a verification failure instruction when the deviation of the ratio from the pre-stored template exceeds the dynamic verification threshold;

[0066] S303, trigger a graded alarm mechanism based on the cumulative number of verification failures. The mechanism includes biometric anomaly alarm and check-in control instruction generation functions.

[0067] In one embodiment of the present invention, the customer identity verification system achieves high-precision security verification through multimodal biometric fusion and dynamic threshold strategy: First, a multispectral 3D structured light camera is used to collect three-dimensional biometric features of the customer's face, iris, and palm veins (containing 256×256×1024 point cloud data), and the feature point cloud density is calculated in real time (formula: When the density value is lower than the set threshold of 0.88, the supplementary acquisition mode is automatically activated—the near-infrared compensation light source is started to enhance the iris texture details, and the angle of the multi-camera array is adjusted to obtain the zygomatic bone-mandibular bone stereoscopic topological data; the enhanced multimodal data is input into the ResNet-152 deep learning model to extract 128-dimensional feature vectors, and the distance ratios of key facial features are calculated (such as the interocular distance / nasal wing width ratio α = 1.62 ± 0.05, the lip height / chin length ratio β = 0.78 ± 0.03). The deviation index E is generated by comparing the cosine similarity with the pre-stored template. dev (Formula: E dev =1-cos(θ), where θ is the angle between the eigenvectors; when E dev When the dynamic verification threshold is exceeded (the initial threshold is set to 0.05, and is adaptively adjusted by ±0.02 according to the ambient light intensity), a verification failure command is triggered and a tiered response mechanism is initiated: the first failure only records an anomaly log; after three failures, a biometric anomaly alarm is sent to the hotel security system (confidence level ≥92%); and after five failures, a check-in control command is generated (locking the reservation system and pushing a manual verification work order). In actual deployment, the system still maintains a verification accuracy rate of 98.7% in strong backlight scenarios, the single verification time is reduced to 1.2 seconds, and 23 identity fraud attempts have been successfully intercepted (verified by data from the public security system). The false rejection rate is controlled below 0.11%, significantly improving the hotel's security management level.

[0068] In one embodiment, generating a customer profile using the XGBoost classification model based on the customer's historical consumption data and device type includes:

[0069] S401, based on the customer value probability value output by the XGBoost model, divides customers into different value levels by setting dynamic classification thresholds. When the quarterly revenue target improvement rate exceeds the set threshold, the high-net-worth customer classification threshold is raised from the first probability value to the second probability value.

[0070] S402, based on the customer levels divided by the classification threshold, for customer groups whose value scores are lower than the current dynamic threshold, generate a linearly increasing discount rule that is positively correlated with the number of consecutive stay days, specifically satisfying: the discount rate increment ΔD and the number of consecutive stay days n satisfy ΔD=kn, where k is a gradient coefficient dynamically set according to the customer value level.

[0071] In one embodiment of the present invention, the customer value segmentation and dynamic discount system achieves precise operation through an intelligent strategy that links machine learning with business objectives: First, it extracts customer consumption behavior data from the hotel CRM system over the past three years (including order amount, stay interval, cancellation rate, room type preference, and proportion of additional service consumption), and combines this with device usage characteristics (mobile / PAD / PC usage frequency, concentration of booking time period) to construct a consumption stability index. A 36-dimensional feature vector is generated for price sensitivity (the percentage of historical orders accepting price adjustments) and device stickiness weights (the percentage of mobile device usage). This feature dataset is then input into a pre-trained XGBoost classification model (parameter settings: max_depth = 5, learning_rate = 0.25, n_estimators = 150). The SHAP value is used to interpret the model's output customer value probability score P. value ∈[0,1], and establish a dynamic threshold adjustment mechanism—the quarterly revenue target improvement rate R target When it exceeds 10%, the high-net-worth client classification threshold changes from P. threshold1 =0.75 automatically adjusted to P threshold2 =0.82, and the gold customer threshold is simultaneously increased by 0.05 to optimize the customer segmentation structure; for customer groups whose value score is lower than the current dynamic threshold (such as P... value <0.65), the system generates consecutive stay discount rules: set the base discount rate D base =5%, the gradient coefficient kkk is set differently according to customer level (platinum customer k=0.015, gold customer k=0.01, ordinary customer k=0.005), then the total discount rate D for consecutive stays of nn days is n =D base+k×n (Example: When a Gold Customer stays for 5 consecutive days, D5 = 5% + 0.01×5 = 10%), and when real-time room status monitoring shows that the vacancy rate of the target room type exceeds 25% in the next 7 days, an additional 10% instant discount is automatically applied; the discount strategy is executed through a multi-channel reach engine: priority is given to pushing APP pop-ups and SMS dual-channel notifications to high-frequency mobile users (including a limited-time tag with a 3-hour countdown), and PC users are displayed a dynamic banner on the official website booking page (embedded with a JavaScript real-time discount calculation component), while combining historical behavior data to accurately target customers during their frequent visit times (such as 8-10 pm); the system updates customer profile data every 6 hours, and when it detects that a customer has rejected the discount proposal twice in a row, it triggers the strategy optimization module (automatically increasing the gradient coefficient kkk by 0.005 and extending the discount validity period by 24 hours). After A / B testing, this solution increased the repurchase rate of the target customer group to 41.3%, increased the proportion of orders for stays of more than 3 consecutive days by 19.7%, and the dynamic pricing strategy contributed 23.5% marginal revenue growth.

[0072] In one embodiment, after the verification is successful, a dynamically encrypted QR code is generated and the following dynamic encryption process is executed:

[0073] S501 performs a hash operation on the customer's identity identifier and timestamp, generates a digest value, and then calculates the message authentication code. The formula is as follows:

[0074] MAC=HMAC(K_sec,Hash(ID||T_now))

[0075] Wherein, MAC is the message authentication code, HMAC is the key-based hash message authentication code, K_sec is the encryption private key, Hash is the cryptographic hash algorithm, ID is the customer identity identifier, and T_now is the current timestamp;

[0076] S502, dynamically adjusts the validity period of the voucher based on room status data, and shortens the validity period when the room card device is detected to be offline for more than the third duration threshold.

[0077] In one embodiment of the present invention, the dynamic encrypted QR code generation system realizes intelligent credential management through a security mechanism that links cryptographic algorithms with real-time room status: First, after the customer's identity verification is passed, the encryption security module (HSM) executes a multi-factor dynamic encryption process—extracting the customer's identity identifier ID (a 32-bit hash value encrypted using the national cryptographic algorithm SM4) and a high-precision timestamp T_now (UTC time down to the millisecond level, in Unix Epoch format). The 13-bit numerical value is used to generate the original data string ID||T_now (example: SM4_9A3F…C2B1||1719878400000) through concatenation. The hash value Hash = H(ID||T_now) is calculated using the SHA-256 hash algorithm. Then, the 256-bit encryption private key K_sec (rotated every 24 hours by the key management system) stored in the hardware security module is used to perform HMAC-SHA512 operation to generate a 160-bit message authentication code MAC = HMAC(K_sec,Hash). Finally, a dynamic encrypted QR code data packet is synthesized (structure is Base64(Enc_AES-256(ID||T_now||MAC))), where the AES encryption key is dynamically derived from the hash value of the hotel room status data (formula: K_aes = HASH(room number||occupancy rate of the day||number of online devices)). At the same time, the validity period of the voucher is dynamically adjusted based on the real-time room status data, and the online status of the room card device is synchronized every 5 seconds through the hotel management system API. When the offline time of the target floor room card controller exceeds the third threshold (set to 15 minutes), an emergency encryption strategy is activated—the validity period of the QR code is shortened from the standard 24 hours to 5 minutes (algorithm parameter adjustment: the validity range of the timestamp T_now is compressed from ±30 minutes to ±2 minutes), and a dynamic expiration flag (field expire_flag=1) is embedded in the QR code data packet. During system operation, the encryption engine monitors the following risk indicators in real time: if the same MAC value appears more than 3 times in 10 minutes, the anti-replay attack mechanism is triggered (automatically invalidating the MAC and generating a new private key K_sec'). Actual deployment data shows that the dynamic QR code generated by this solution takes only 47ms for a single encryption (NIST test environment), and the computing power required to crack it exceeds 2^128 operations. Furthermore, through integration with the public security system, it achieves a 98.5% illegal credential interception rate, effectively preventing 92.3% of abnormal access attempts in the scenario where the room card device is offline, while ensuring a seamless access experience for legitimate customers.

[0078] In one embodiment, the method further includes:

[0079] S601, collects the error rate δ between the actual transaction price and the predicted price;

[0080] S602, when δ>15%, the parameters of the dynamic pricing model are updated using an online random forest algorithm.

[0081] In one embodiment of the present invention, the error monitoring and online learning mechanism of the dynamic pricing system achieves continuous model optimization through the following technical solution: First, the actual transaction price data stream is collected in real time through the hotel PMS system, and the predicted price error rate is calculated every 15 minutes (formula: The system constructs an error analysis vector containing time-series features (error trend over the past hour), market volatility factors (competitor average price change rate), and demand elasticity coefficients. When δ > 15% and the error direction is consistent for three consecutive calculation cycles, the online learning engine—based on the incremental learning framework's random forest algorithm (parameter settings: n_estimators = 200, max_depth = 8, warm_start = True)—is triggered to update the model parameters in real time. The specific process is as follows: 1) Extract the full data of the past 30 days (including 42-dimensional features such as price, room status, and customer behavior) from the historical database as the base model training set; 2) ... 1) Construct an incremental dataset from the feature data related to the current error event (including room type, booking channel, equipment type, and promotional tags of abnormal transactions), and dynamically adjust the weights of split nodes by re-ranking feature importance (adjustment coefficient α = 0.7 × δ); 2) Use a sliding window mechanism to retain the most recent 5000 high-error samples (δ > 10%) for reinforcement training, and introduce adversarial example generation technology (GAN network) to expand the diversity of feature space; During the model update process, the system simultaneously performs the following optimizations: exponential smoothing of the price sensitivity feature (historical price adjustment acceptance rate) (smoothing factor β = 0.85), and reconstruct the supply and demand balance index. The number of leaf nodes in the online random forest was dynamically expanded to 120% of the original model. The updated model was validated through A / B testing. The results showed that in the peak tourist season, the price prediction error rate decreased from 18.7% to 6.3%, the model iteration time was controlled within 90 seconds (AWS c5.4xlarge instance), and the revenue loss caused by dynamic price adjustment was reduced by 41.5%. The system generates a model health report every 24 hours. When a significant shift is detected in the top 5 indicators of feature importance (KL divergence > 0.15), the full data retraining process is automatically triggered to ensure that the model continues to adapt to market changes.

[0082] In one embodiment, the competitor hotel price data is processed using the following dynamic cleaning algorithm:

[0083] Use the following formula to calculate the valid reference price:

[0084]

[0085] S701, in which P is the effective reference price at time t. i Let w be the real-time price of the i-th competitor, n be the threshold number of competitors to be included in the calculation, and w be the price of the i-th competitor. i Dynamic weights;

[0086] S702, weight Where S i Rate competitors, w i 4.5 represents the dynamic weight of the i-th competitor, and 4.5 represents the offset of the Sigmoid function.

[0087] S703: When the amount of valid data is lower than the set threshold, the data collection range and collection duration are extended.

[0088] In one embodiment of the present invention, the dynamic cleaning and weighted calculation system for competitor price data achieves data quality optimization through the following technical solution: First, a distributed data stream processing engine (based on Apache Flink) is constructed to receive competitor hotel price data stream Pi(t) (where i represents the i-th competitor) from the crawler cluster in real time, and a multi-stage cleaning process is executed—1) Data verification stage: outliers exceeding the 3σ principle are removed (the calculation period is a 15-minute sliding window, formula: |P i (t)-μ(t)|>3σ(t) is marked as invalid data); 2) Weight calculation stage: based on the multi-dimensional evaluation indicators of competitors (including OTA platform scores) i ∈[0,5], historical price stability and geographical distance d i Dynamic weights are generated, and the weight calculation formula is as follows:

[0089]

[0090] Wherein, the parameter of the Sigmoid function is set to Sigmoid(x) = 1 / (1+e -(x-2.5) ), by adjusting the offset, the weight of competitors with a rating of 4.0 or higher is significantly increased; 3) Effective reference price calculation stage: when N is satisfied valid ≥N threshold (Set the minimum number of effective competitors N) threshold When =5), calculate the weighted effective reference price if the number of real-time effective competitors is N. valid If the value is less than 5, the data acquisition expansion mechanism is triggered—the geographical radius is expanded from the baseline 5 kilometers to 8 kilometers, the acquisition time window is extended from 15 minutes to 1 hour, and backup crawler nodes are activated (the number is increased by 50%) to improve the data capture volume; the cleaned data stream undergoes the following quality control cycle: the data confidence index is calculated every 30 minutes. When C(t) < 0.7, the heterogeneity of the data source is automatically enhanced (supplementary sources such as Meituan in-store data and telephone inquiry data are added). In actual deployment, the system can improve the stability of effective reference price calculation by 38% (variance reduced to 0.15σ) during peak periods (such as holidays). The dynamic weighting algorithm makes the influence weight of high-rated (≥4.5) competitors reach 0.28-0.35 (the baseline solution is a fixed weight of 0.2). When the collection range is expanded, the data validity recovery time is shortened to within 8 minutes. The cleaned data is pushed to the pricing model in real time through the Kafka message queue. Tests show that this solution reduces the price prediction error rate to 4.7% (the original solution was 7.2%) and effectively avoids 72% of abnormal price interference events.

[0091] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0092] Based on the same inventive concept, this application also provides an intelligent hotel dynamic pricing and marketing device based on multi-source data for implementing the aforementioned intelligent hotel dynamic pricing and marketing method based on multi-source data. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent hotel dynamic pricing and marketing device based on multi-source data provided below can be found in the limitations of the intelligent hotel dynamic pricing and marketing method based on multi-source data described above, and will not be repeated here.

[0093] In one exemplary embodiment, such as Figure 2 As shown, a smart hotel dynamic pricing and marketing device 400 based on multi-source data is provided, including:

[0094] The multi-source data acquisition module 401 is used to acquire competitor hotel price data and promotional strategy data, room inventory data and historical occupancy rate data, and customer reservation equipment type and timestamp data.

[0095] The dynamic pricing strategy generation module 402 is used to input competitor price data, room inventory data and historical occupancy rate data into the dynamic pricing model to generate a real-time benchmark price and a price fluctuation range. The lowest price in the price fluctuation range is not lower than the cost price, and the highest price does not exceed a set percentage of the average price of competitors.

[0096] The self-service check-in process execution module 403 is used to obtain customer identity information and verify it. After successful verification, a dynamic encrypted credential is generated and a room card allocation operation is performed.

[0097] The personalized marketing push module 404 is used to generate customer profiles based on customers' historical consumption data and device type using the XGBoost classification model, match them with preset marketing templates, and push differentiated promotional content.

[0098] In one embodiment, the multi-source data acquisition module 401 is used for:

[0099] The following dynamic cleaning algorithm is used to process competitor hotel price data:

[0100]

[0101] in P is the effective reference price at time t. i Let w be the real-time price of the i-th competitor, n be the threshold number of competitors to be included in the calculation, and w be the price of the i-th competitor. i For dynamic weights.

[0102] Weight Where S i Rate competitors, w i 4.5 represents the dynamic weight of the i-th competitor, and 4.5 represents the offset of the Sigmoid function.

[0103] When the amount of valid data is lower than the set threshold, the data collection range and collection duration are expanded.

[0104] In one embodiment, the dynamic pricing strategy generation module 402 is used to:

[0105] When the price volatility of competing hotels exceeds a first threshold, the exploration rate increment of the reinforcement learning algorithm is increased, and the exploration rate increment is positively correlated with the price volatility.

[0106] When the hotel vacancy rate continuously reaches the second time threshold and is lower than the set vacancy rate, the lower limit of the price is dynamically adjusted to the adjustment ratio of the average price of competitors, and a price instruction set containing tiered discount rules is generated.

[0107] In one embodiment, the self-service check-in process execution module 403 is used for:

[0108] Collect the customer's 3D biometric data and perform density detection. When the detected value is lower than the set density threshold, the supplementary biometric data collection mode is activated.

[0109] The distance ratio of key facial features is calculated based on multimodal biometric data. When the deviation of this ratio from the pre-stored template exceeds the dynamic verification threshold, a verification failure instruction is generated.

[0110] A tiered alarm mechanism is triggered based on the cumulative number of verification failures. This mechanism includes functions for generating biometric anomaly alarms and check-in control commands.

[0111] In one embodiment, the self-service check-in process execution module 403 is further configured to:

[0112] The customer's identity identifier and timestamp are hashed to generate a digest value, which is then used to calculate the message authentication code. The formula is as follows:

[0113] MAC=HMAC(K_sec,Hash(ID||T_now))

[0114] Wherein, MAC is the Message Authentication Code, HMAC is the Key-Based Hash Message Authentication Code, K_sec is the encrypted private key, Hash is the cryptographic hash algorithm, ID is the customer identity identifier, and T_now is the current timestamp.

[0115] The validity period of the voucher is dynamically adjusted based on room status data. When the room card device is detected to be offline for more than the third time threshold, the validity period is shortened.

[0116] In one embodiment, the personalized marketing push module 404 is used for:

[0117] Based on the customer value probability value output by the XGBoost model, customers are divided into different value levels by setting dynamic classification thresholds. When the quarterly revenue target improvement rate exceeds the set threshold, the high-net-worth customer classification threshold is raised from the first probability value to the second probability value.

[0118] Based on the customer levels defined by the classification threshold, for customer groups whose value scores are lower than the current dynamic threshold, a linearly increasing discount rule that is positively correlated with the number of consecutive stay days is generated, specifically satisfying: the discount rate increment ΔD and the number of consecutive stay days n satisfy ΔD=kn, where k is a gradient coefficient dynamically set according to the customer value level.

[0119] In one embodiment, the intelligent hotel dynamic pricing and marketing device 400 based on multi-source data further includes:

[0120] The data acquisition module collects the error rate δ between the actual transaction price and the predicted price.

[0121] The parameter update module updates the dynamic pricing model parameters using an online random forest algorithm when δ > 15%.

[0122] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the power supply safety management method as described above.

[0123] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0124] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0125] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A smart hotel dynamic pricing and marketing method based on multi-source data, characterized in that, include: Obtain competitor hotel pricing and promotional strategy data, room inventory and historical occupancy rate data, and customer booking device type and timestamp data; The competitor's price data, room inventory data, and historical occupancy rate data are input into the dynamic pricing model to generate a real-time benchmark price and a price fluctuation range, wherein the lowest price in the price fluctuation range is not lower than the cost price, and the highest price does not exceed a set percentage of the average price of competitors. Obtain and verify customer identity information. Once verification is successful, generate a dynamic encrypted credential and perform room card allocation. Based on customers' historical consumption data and device type, a customer profile is generated using the XGBoost classification model, which is then matched with a preset marketing template and pushed with differentiated promotional content.

2. The method according to claim 1, characterized in that, The dynamic pricing model performs the following optimizations: When the price volatility of competing hotels exceeds a first threshold, the exploration rate increment of the reinforcement learning algorithm is increased, and the exploration rate increment is positively correlated with the price volatility. When the hotel vacancy rate continuously reaches the second time threshold and is lower than the set vacancy rate, the lower limit of the price is dynamically adjusted to the adjustment ratio of the average price of competitors, and a price instruction set containing tiered discount rules is generated.

3. The method according to claim 1, characterized in that, The process of obtaining customer identity information and verifying it to obtain detection results includes: Collect the customer's 3D biometric data and perform density detection. When the detected value is lower than the set density threshold, start the supplementary biometric data collection mode. The distance ratio of key facial features is calculated based on multimodal biometric data. When the deviation of this ratio from the pre-stored template exceeds the dynamic verification threshold, a verification failure instruction is generated. A tiered alarm mechanism is triggered based on the cumulative number of verification failures. This mechanism includes functions for generating biometric anomaly alarms and check-in control commands.

4. The method according to claim 1, characterized in that, The process of generating customer profiles using the XGBoost classification model based on historical customer consumption data and device type includes: Based on the customer value probability value output by the XGBoost model, customers are divided into different value levels by setting dynamic classification thresholds. When the quarterly revenue target improvement rate exceeds the set threshold, the high-net-worth customer classification threshold is raised from the first probability value to the second probability value. Based on the customer levels defined by the classification threshold, for customer groups whose value scores are lower than the current dynamic threshold, a linearly increasing discount rule that is positively correlated with the number of consecutive stay days is generated, specifically satisfying: the discount rate increment ΔD and the number of consecutive stay days n satisfy ΔD=kn, where k is a gradient coefficient dynamically set according to the customer value level; The dynamic classification threshold division results are associated and mapped with the discount rules to generate a customer profile that includes value level labels and discount gradient parameters.

5. The method according to claim 1, characterized in that, After successful verification, a dynamically encrypted QR code is generated and the following dynamic encryption process is executed: The customer's identity identifier and timestamp are hashed to generate a digest value, which is then used to calculate the message authentication code. The formula is as follows: MAC=HMAC(K_sec,Hash(ID||T_now)) Wherein, MAC is the message authentication code, HMAC is the key-based hash message authentication code, K_sec is the encryption private key, Hash is the cryptographic hash algorithm, ID is the customer identity identifier, and T_now is the current timestamp; The validity period of the voucher is dynamically adjusted based on room status data. When the room card device is detected to be offline for more than the third time threshold, the validity period is shortened.

6. The method according to claim 1, characterized in that, The method further includes: Collect the error rate δ between the actual transaction price and the predicted price; When δ>15%, the parameters of the dynamic pricing model are updated using an online random forest algorithm.

7. The method according to claim 1, characterized in that, The competitor hotel price data was processed using the following dynamic cleaning algorithm: Use the following formula to calculate the valid reference price: in P is the effective reference price at time t. i Let w be the real-time price of the i-th competitor, n be the threshold number of competitors to be included in the calculation, and w be the price of the i-th competitor. i Dynamic weights; Weight Where S i Rate competitors, w i 4.5 represents the dynamic weight of the i-th competitor, and 4.5 represents the offset of the Sigmoid function. When the amount of valid data is lower than the set threshold, the data collection range and collection duration are expanded.

8. A smart hotel dynamic pricing and marketing device based on multi-source data, characterized in that, include: The multi-source data acquisition module is used to acquire competitor hotel price data and promotional strategy data, room inventory data and historical occupancy rate data, and customer booking device type and timestamp data. The dynamic pricing strategy generation module is used to input the competitor price data, room inventory data and historical occupancy rate data into the dynamic pricing model to generate a real-time benchmark price and a price fluctuation range, wherein the lowest price in the price fluctuation range is not lower than the cost price and the highest price does not exceed a set percentage of the average price of competitors. The self-service check-in process execution module is used to obtain and verify customer identity information. After successful verification, it generates a dynamic encrypted credential and performs room card allocation. The personalized marketing push module is used to generate customer profiles based on customers' historical consumption data and device type using the XGBoost classification model, match them with preset marketing templates, and push differentiated promotional content.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images