AI-based Hotel Warehouse Scheduling and Predictive Optimization Method
By constructing a multi-regional spatiotemporal correlation dataset and conducting causal impact analysis, a heat map of quality failure risk was generated, optimizing hotel warehouse management, solving the problem of food consumption and dynamic quality changes, and achieving the optimization of inventory costs and quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- QINGDAO HOTEL MANAGEMENT VOCATIONAL & TECH COLLEGE
- Filing Date
- 2026-04-15
- Publication Date
- 2026-06-02
AI Technical Summary
Existing hotel warehouse management systems cannot effectively distinguish the causal relationship of food consumption, leading to prediction failures in the event of emergencies, and ignoring the dynamic changes in food quality, resulting in waste and a decline in food quality.
By constructing a multi-regional spatiotemporal correlation dataset, identifying causal influences and transmission delay times between regions, generating a heatmap of quality failure risk, and combining it with an inventory cost optimization model, dynamically generating an adaptive replenishment strategy to achieve synchronous prediction of consumption and quality.
This effectively reduces food waste and losses, optimizes warehouse scheduling, achieves optimal overall operating costs, and improves food quality and customer satisfaction.
Smart Images
Figure CN122134259A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of management and scheduling technology, and in particular to a method for hotel warehouse scheduling and predictive optimization based on artificial intelligence. Background Technology
[0002] In the daily operation of hotels, food and beverage services occupy a core position, and the storage and scheduling of ingredients and beverages are directly related to the quality of dishes, cost control, and customer satisfaction. This is especially true for high-end hotels, where the inventory management of valuable ingredients, such as imported seafood, premium beef, and expensive beverages, faces more severe challenges. On the one hand, these ingredients are high-value, have short shelf lives, and are sensitive to quality fluctuations. On the other hand, the demand for ingredients varies significantly across different areas of the hotel and is interconnected. Inventory imbalances in any area can lead to a decline in quality, an increase in costs, or service interruptions. Currently, hotel warehouse management mainly relies on traditional inventory control methods, such as the economic order quantity model and safety stock formula, combined with simple historical consumption statistics for replenishment decisions. Some hotels with a higher level of informatization have introduced warehouse management systems, which have achieved digital recording of inventory data and basic early warning functions.
[0003] However, these methods have the following technical drawbacks when dealing with complex hotel warehousing and scheduling problems:
[0004] Some hotels have introduced machine learning models to predict food consumption. However, these models can only capture the statistical correlation between features and consumption, and cannot distinguish between true causal relationships and spurious associations caused by confounding factors, leading to prediction failures in the event of unforeseen circumstances.
[0005] The quality of valuable ingredients deteriorates non-linearly over storage time, and the deterioration curves vary for different ingredients. Existing inventory models usually only focus on the quantity dimension, treating quality as static and triggered by simple thresholds, ignoring the dynamic characteristics of quality deterioration of valuable ingredients. This results in ingredients not being used in time during their optimal quality control period, causing waste and affecting the quality of dishes.
[0006] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention
[0007] The purpose of this invention is to: construct a multi-regional spatiotemporal correlation dataset and identify the transmission delay time and causal influence coefficient moments of business impact between regions to achieve the purpose of prediction deviation analysis; construct a quality degradation curve and output a quality failure risk heat map to dynamically incorporate food freshness into inventory optimization decisions, solve the problem of food spoilage losses caused by long-term neglect of quality factors, achieve end-to-end collaborative prediction of consumption uncertainty and dynamic quality changes; and dynamically generate adaptive replenishment strategies through a safety stock decision model aimed at minimizing inventory holding costs, stockout penalty costs, and quality spoilage losses, thereby achieving global optimization of warehouse scheduling and minimization of operating costs.
[0008] To achieve the above objectives, the present invention adopts the following technical solution: a hotel warehouse scheduling and prediction optimization method based on artificial intelligence, comprising the following steps:
[0009] Step 1: Collect data on the hotel area's target food consumption, business event intensity, supply chain indicators, and IoT quality sensing data, and align the collected data with spatial area identifiers to output a standardized spatiotemporal correlation dataset.
[0010] Step 2: Based on the spatiotemporal correlation dataset, calculate the temporal correlation of consumption fluctuations in different regions, identify the time offset between fluctuations as the transmission delay time of business impact between regions, perform confounding factor separation processing, and calculate the causal influence coefficient matrix after deconfounding correction.
[0011] Step 3: Based on the causal influence coefficient matrix, perturb the intensity of business events to generate a counterfactual consumption sample library under different hypothetical scenarios. Based on real-time quality sensing data, establish the degradation curve of food quality and output a heat map of quality failure risk when the quality index exceeds the warning threshold in the future period.
[0012] Step 4: Input the counterfactual consumption sample library and the quality failure risk heat map into the predictive analysis network to perform probability consumption and quality analysis, and obtain the probability distribution of future consumption of valuable ingredients and the quality evolution results.
[0013] Step 5: Construct a multi-objective cost function with the goal of minimizing inventory holding costs, stockout penalty costs, and quality scrap losses. Solve for the dynamic safety stock threshold based on the consumption probability distribution and quality evolution results, generate an adaptive replenishment strategy, and output it to the procurement management end.
[0014] Furthermore, data on the intensity of target business events, environmental disturbance factors, supply chain indicators, and IoT quality sensing data in the hotel area are collected. The specific collection process is as follows:
[0015] Based on the outbound records of target ingredients in each area of the warehouse and the sales back-calculation data, standardized consumption time-series data is formed;
[0016] The daily food consumption and time-point business indicators are weighted into an event intensity sequence, and data is continuously collected at the granularity of a period to form time series data of business activity.
[0017] By comprehensively analyzing the on-time delivery rate of ingredients, batch quality level, and supplier reliability index, we can obtain supply chain stability index data.
[0018] By collecting data on remaining shelf life, total bacterial count, and sensory scores using temperature and humidity sensors, dedicated quality testing instruments, and manual data entry terminals deployed in cold storage and refrigerated trucks, a comprehensive quality index data of 0-1 is set up to integrate multiple quality parameters.
[0019] Spatial region identification and alignment are performed on the collected consumption time series data, time series data, supply chain stability index data, and quality index data to output a standardized spatiotemporal correlation dataset.
[0020] Furthermore, based on the spatiotemporal correlation dataset, the temporal correlation of consumption fluctuations in different regions is calculated, and the time offset between fluctuations is identified as the transmission delay time of inter-regional business impact. The specific process is as follows:
[0021] By extracting time-series data on the consumption of valuable ingredients in any two regions at the same time from a standardized spatiotemporal correlation dataset, and calculating the cross-correlation coefficient between the two sequences based on the time offset using the transmission delay window of the business unit, a cross-correlation function curve is generated with the delay time as the horizontal axis and the correlation coefficient as the vertical axis.
[0022] On the cross-correlation curve, filter out the delay time point that makes the absolute value of the correlation coefficient reach the peak, determine the direction of influence transmission based on the sign of the delay time point at the peak, and record the absolute value of the delay time point at the peak as the business influence transmission delay time between the two regions.
[0023] Based on the obtained propagation delay time, the data is classified and labeled according to different regions to form a delay time dataset.
[0024] Furthermore, confounding factors are separated, and the causal influence coefficient matrix after deconfounding correction is calculated. The specific process is as follows:
[0025] The hotel promotional activities, holidays, and out-of-stock events were used as the criteria for screening confounding factors. The spatiotemporal correlation datasets of each hotel region were analyzed and filtered to obtain the spatiotemporal correlation data separated for each region.
[0026] The separated screening data are statistically analyzed to determine the duration of promotional activities and the duration of stockouts in each region within the time window of the cycle. This data serves as an observable confounding factor vector. The consumption residual sequence after removing the confounding effects is used as a proxy indicator for the real business intervention variable, representing the consumption changes driven by the region's own operational fluctuations after removing external interference.
[0027] Using confounding factors as control variables and actual intervention variables as treatment variables, a causal impact analysis of sales volume in a region was conducted to obtain the causal impact coefficients between regions. The effect coefficients of all region pairs were arranged according to the source region and the affected region to form a matrix of causal impact coefficients between regions.
[0028] Furthermore, based on the causal influence coefficient matrix, the intensity of business events is analyzed to generate counterfactual samples of the scenarios. Based on the consumption change trajectories under different scenarios, a counterfactual consumption sample library is generated. The specific process is as follows:
[0029] Based on the causal influence coefficient matrix, the key business event types that affect the consumption of the target area are identified. According to the distribution characteristics of historical data, the disturbance amplitude range and disturbance step size are set to form a disturbance parameter combination. For each selected disturbance target area, multiple hypothetical scenarios are set for disturbance transformation.
[0030] The intensity of the disturbed business events is used to calculate the trajectory of consumption changes through the net effect coefficient and propagation delay time in the coefficient matrix, generating a complete consumption time series for each disturbance scenario, thus forming a preliminary counterfactual sample.
[0031] All counterfactual samples and their corresponding confidence weights and perturbation parameter labels are stored in a structured manner to construct a counterfactual consumption sample library.
[0032] Furthermore, based on real-time quality indicators, a quality status is established, a quality degradation curve is constructed, and a heatmap of quality failure risk where the quality index exceeds the warning threshold in the future is output. The specific process is as follows:
[0033] Acquire real-time data from IoT sensors in the storage area, and plot a scatter plot of quality decay for the current batch of ingredients with storage time on the horizontal axis and quality index on the vertical axis.
[0034] Analyze all historical batch data, statistically analyze the differences in food decay rate under different temperature and humidity ranges, use the decay rate under standard conditions as a benchmark, calculate the accelerated decay coefficient under high temperature and high humidity conditions and the decelerated decay coefficient under low temperature and low humidity conditions, and establish a mapping relationship table between environmental parameters and decay coefficients.
[0035] Based on the mapping relationship table between environmental parameters and decay coefficients, starting with the latest measured quality index of the current batch of ingredients, and combined with the predicted environmental parameters for future periods, the real-time decay rate at each future time point is calculated to form a dynamic quality degradation curve. On the degradation curve, the quality index is checked point by point to see if it is lower than the preset warning threshold and scrap threshold. The specific time point when it first falls below the threshold is recorded, which is the predicted failure time.
[0036] Using future time as the horizontal axis and batches of food from each storage area as the vertical axis, a two-dimensional risk matrix is constructed to generate an intuitive heat map of quality failure risk, with key information of each batch of food marked on the map.
[0037] Furthermore, the counterfactual consumption sample library and the heat map of quality failure risk are input into the predictive analysis network to perform probabilistic consumption and quality analysis, obtaining the probability distribution of future consumption of each valuable ingredient in each region and the results of quality evolution. The specific process is as follows:
[0038] The counterfactual consumption sample library and the quality failure risk heatmap are time-aligned and preprocessed to construct a multi-dimensional feature vector, which is then used as input features to the predictive analysis network.
[0039] Construct a predictive analysis network to perform two prediction tasks: consumption probability distribution prediction and quality evolution result prediction. Based on the causal relationship in the counterfactual sample library, output multiple quantiles of consumption at each future time point. Based on the current quality index and the failure probability provided by the heatmap, output the mean of the future quality index.
[0040] By summarizing the predicted quantiles for each region and time period, a probability density curve is generated. The predicted values of the quality index are then connected to form a curve, and the expected time points when the quality index falls below the warning threshold are marked. This yields the probability distribution of future consumption and the results of quality evolution for each valuable ingredient.
[0041] Furthermore, a multi-objective cost function is constructed with the objectives of minimizing inventory holding costs, stockout penalty costs, and quality scrap losses. Based on the consumption probability distribution and quality evolution results, a dynamic safety stock threshold is calculated, generating an adaptive replenishment strategy which is then output to the procurement management system. The specific process is as follows:
[0042] Based on the probability distribution of consumption and the quality evolution trajectory, three cost functions are defined respectively: inventory holding cost, stockout penalty cost, and quality scrap loss;
[0043] Using the next 30 days as a rolling window, the consumption forecasts at different quantiles are substituted into the cost function, and the inventory level critical point that minimizes the sum of the three costs is searched.
[0044] The system compares the current inventory level with the dynamic safety stock threshold in real time. When the inventory level falls below the threshold, it automatically triggers a replenishment suggestion, calculates the replenishment quantity, determines the order date, outputs quality risk warnings and replenishment instructions, and automatically pushes them to the procurement management personnel's terminal daily.
[0045] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0046] This AI-based hotel warehouse scheduling and predictive optimization method calculates the temporal correlation of consumption fluctuations in different regions using a spatiotemporally correlated dataset. It identifies the time offset between fluctuations as the transmission delay time of business impacts between regions, performs confounding factor separation, and calculates a deconfounded causal influence coefficient matrix to achieve the goal of real business causal correlation between regions. It generates counterfactual samples of scenarios, creating a counterfactual consumption sample library based on consumption change trajectories under different scenarios. It establishes quality status based on real-time quality indicators, constructs quality degradation curves, and outputs a heatmap of quality failure risk when the quality index exceeds the warning threshold in future periods, achieving early visualization and warning of quality risks. The counterfactual consumption sample library and the quality failure risk heatmap are input into a predictive analysis network for probabilistic consumption and quality analysis, obtaining the future consumption probability distribution and quality evolution results of valuable ingredients. Dynamic quality decay is incorporated into inventory optimization, effectively reducing food scrap losses and achieving simultaneous prediction of consumption uncertainty and dynamic quality evolution. A safety stock replenishment strategy aimed at minimizing inventory holding costs, stockout penalty costs, and quality scrap losses ultimately achieves optimal warehouse scheduling and costs. Attached Figure Description
[0047] Figure 1 A schematic diagram of the overall structure of the method flow of the present invention is shown. Detailed Implementation
[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0049] Example:
[0050] like Figure 1 As shown, the AI-based hotel warehousing scheduling and predictive optimization method includes the following steps:
[0051] Step 1: Collect data on the hotel area's target food consumption, business event intensity, supply chain indicators, and IoT quality sensing data, and align the collected data with spatial area identifiers to output a standardized spatiotemporal correlation dataset.
[0052] Step 2: Based on the spatiotemporal correlation dataset, calculate the temporal correlation of consumption fluctuations in different regions, identify the time offset between fluctuations as the transmission delay time of business impact between regions, perform confounding factor separation processing, and calculate the causal influence coefficient matrix after deconfounding correction.
[0053] Step 3: Based on the causal influence coefficient matrix, perturb the intensity of business events to generate a counterfactual consumption sample library under different hypothetical scenarios. Based on real-time quality sensing data, establish the degradation curve of food quality and output a heat map of quality failure risk when the quality index exceeds the warning threshold in the future period.
[0054] Step 4: Input the counterfactual consumption sample library and the quality failure risk heat map into the predictive analysis network to perform probability consumption and quality analysis, and obtain the probability distribution of future consumption of valuable ingredients and the quality evolution results.
[0055] Step 5: Construct a multi-objective cost function with the goal of minimizing inventory holding costs, stockout penalty costs, and quality scrap losses. Solve for the dynamic safety stock threshold based on the consumption probability distribution and quality evolution results, generate an adaptive replenishment strategy, and output it to the procurement management end.
[0056] This solution obtains daily outbound records from the hotel management system for the restaurant and room service areas. Combined with sales back-calculation data, standardized consumption time-series data for each ingredient in each area is generated. Daily ingredient consumption is weighted with corresponding business indicators at specific times to form an event intensity sequence. Data is collected at a periodic granularity to create a business event intensity sequence. The on-time rate is collected through the supplier management system and procurement platform, representing the deviation between the supplier's actual delivery time and the agreed-upon time, and the on-time delivery ratio is calculated. Batch quality level is the pass rate of each batch of delivered ingredients. The supplier reliability index is calculated based on historical cooperation records, combining on-time rate, pass rate, and response speed to form a reliability score of 0-100, updated daily to create a supply chain stability indicator sequence. Temperature and humidity sensors are deployed in cold storage, refrigerators, and refrigerated trucks to record temperature and relative humidity. Remaining shelf life and total bacterial count are collected using dedicated quality testing instruments and manual input terminals. All collected data is aligned by region identifier and timestamp to generate a spatiotemporal correlation dataset containing region ID, timestamp, consumption quantity (C), event intensity (E), supply chain indicator (S), and quality index (Q). ;
[0057] Based on the spatiotemporal correlation dataset, the propagation delay time of business impacts between regions is identified, and the causal influence coefficient matrix after removing confounding factors is calculated. Time-series data on the consumption of valuable ingredients for any two regions i and j are extracted from the standardized spatiotemporal correlation dataset. and Calculate the cross-correlation coefficient between two sequences within the possible propagation window of the business event. :
[0058]
[0059] In the above formula, Let be the mean of the consumption time series in region i. Let J be the mean of the consumption time series in region j, with a time delay. Using the x-axis as the plot and the correlation coefficient R as the y-axis, a cross-correlation function curve is generated. The time delay points on the curve that cause the absolute value of the correlation coefficient to reach its maximum value are then selected. ,according to The positive or negative sign of the signal affects the direction of propagation; a positive sign indicates that i leads j, and a negative sign indicates that j leads i. The delay time for the transmission of business impact between two regions is recorded. This process is repeated for all regions to form a delay time matrix. ;
[0060] The hotel promotional activities, holidays, and stockout periods were used as the criteria for screening confounding factors. The datasets of each region were analyzed, and time windows that simultaneously included promotions or stockouts were marked as confounding periods. Data fluctuations during confounding periods were removed, and the consumption sequence reflecting the natural correlation between regions was retained. The intensity of promotional activities and the duration of stockouts in each region within each time window were statistically analyzed and used as observable confounding factor vectors. The consumption residual sequence after removing confounding effects was used as an indicator of the real business intervention variable.
[0061] Using confounding factors as control variables and actual intervention variables as treatment variables, a dual machine learning framework was employed to estimate the causal effects for each pair of region combinations. The specific steps are as follows:
[0062] Construct the first machine learning model to predict the intervention variable Z and obtain the predicted value. ;
[0063] Construct a second machine learning model to predict the outcome variable C and obtain the predicted value. ;
[0064] Calculate the residuals of the intervention variables and residuals of outcome variables ;
[0065] Regression of residuals The regression coefficients were obtained. This is the net effect coefficient of the change in event i on the consumption of region j. This process is repeated for all regions, and the effect coefficients are arranged according to the source region and the affected region to form an inter-regional causal influence coefficient matrix. ;
[0066] Based on the causal influence coefficient matrix, the key business event types with the greatest impact on consumption in the target area are identified. According to the distribution characteristics of historical data, the disturbance amplitude range and disturbance step size are set to form a disturbance parameter combination. Through the net effect coefficient and transmission delay time in the coefficient matrix, the consumption change trajectory is iteratively calculated step by step to generate a complete consumption time series corresponding to each disturbance scenario, forming a preliminary counterfactual sample. Real-time data from IoT sensors in the storage area are obtained. With storage time t as the horizontal axis and quality index Q as the vertical axis, a scatter plot of the quality decay of the current batch of food is drawn. Starting from the latest measured quality index of the current batch of food, combined with the predicted environmental parameters of the future period, the real-time decay rate of each future time point t is calculated. On the degradation curve, the quality index is checked point by point to see if it is lower than the preset warning threshold and scrap threshold. The time point when it first falls below the threshold is recorded as the predicted failure time. With future time as the horizontal axis and batches of food in each storage area as the vertical axis, a two-dimensional risk matrix is constructed.
[0067] When constructing the predictive analysis network, the feature tensor is divided into training and validation sets in an 8:2 ratio. A multi-task learning network is constructed, adopting an architecture design that shares a bottom-level feature extraction layer and a dual-branch output layer.
[0068] Composed of three stacked long short-term memory networks, each with 128 hidden units, it is responsible for capturing the time dependence of consumption sequences and the long-term memory effect of event influences. The input time window is set to 30 days, and the output is the prediction results for the next 7, 15, and 30 days.
[0069] Output the predicted future consumption values at the 10th, 50th, and 90th percentiles, forming a complete probability distribution interval, and the loss function. Quantile loss:
[0070]
[0071] , The actual consumption at time t. The consumption at quantile τ at time t predicted by the model;
[0072] A temporal convolutional network is used to output a sequence of daily quality index predictions for the next 30 days. The output is mapped to the [0,1] interval through an activation function, and the loss function is the mean squared error.
[0073] During network training, an uncertainty-weighted strategy is adopted, dynamically adjusting the weights of the respective loss functions based on the performance of the two tasks on the validation set. The latest real-time data is input into the trained multi-task network, and the probability distribution of consumption of each valuable ingredient in each region over the next 7, 15, and 30 days is calculated through forward propagation. The results are output in the form of 10%, 50%, and 90% quantiles, as well as the predicted daily quality index value within the corresponding time window. Finally, the probability distribution map of consumption of each ingredient in each region and the quality evolution trajectory results are output.
[0074] When constructing a system that dynamically calculates the safety stock threshold and generates an adaptive replenishment strategy with the goal of minimizing inventory holding costs, stockout penalty costs, and quality scrap losses, the following three costs are defined:
[0075] Inventory holding cost is the daily storage fee per unit of food ingredient. Forecast inventory Storage days The predicted inventory level is calculated based on the replenishment plan and consumption forecast, and the stockout penalty cost P is the profit loss caused by a unit of food shortage. The duration of the shortage and the loss due to spoilage (W) represent the cost of food procurement. Quality index below scrap threshold Expected scrap volume Simultaneously input the maximum inventory capacity for each region. Supplier minimum order quantity (MOQ), lead time (L), fixed cost per purchase Actual business constraints form a complete basis for cost analysis;
[0076] Using the next 30 days as a rolling window, different quantiles of the consumption probability distribution are substituted into the cost to calculate the expected total cost under different inventory levels I. According to the minimum critical point of TC(I) inventory level The current inventory of each ingredient in each region. Compare in real time with the calculated dynamic safety stock threshold I∗I∗:
[0077] when When the time comes, a replenishment suggestion will be automatically triggered;
[0078] Replenishment quantity ,in The order date is determined based on the supplier's working hours and transportation cycle to ensure that the delivery time is no later than the point when the inventory is depleted.
[0079] Finally, the replenishment strategy is pushed to the purchasing system or management terminal in a standardized instruction format and is automatically updated daily. The strategy output includes complete information such as replenishment area, replenishment food category, suggested replenishment quantity, suggested order date, expected arrival date, and expected cost savings analysis.
[0080] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.
[0081] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.
[0082] In the two embodiments provided in this application, it should be understood that the disclosed apparatus and system can be implemented in other ways; for example, the apparatus embodiments described above are merely illustrative, and the division of modules is merely a logical functional division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed; furthermore, the coupling or direct coupling or communication connection between the shown or discussed mutuals can be through some interfaces, and the indirect coupling or communication connection between the apparatus or modules can be electrical, mechanical or other forms.
[0083] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A hotel warehousing scheduling and predictive optimization method based on artificial intelligence, characterized in that, Includes the following steps: Step 1: Collect data on the hotel area's target food consumption, business event intensity, supply chain indicators, and IoT quality sensing data, and align the collected data with spatial area identifiers to output a standardized spatiotemporal correlation dataset. Step 2: Based on the spatiotemporal correlation dataset, calculate the temporal correlation of consumption fluctuations in different regions, identify the time offset between fluctuations as the transmission delay time of business impact between regions, perform confounding factor separation processing, and calculate the causal influence coefficient matrix after deconfounding correction. Step 3: Based on the causal influence coefficient matrix, perturb the intensity of business events to generate a counterfactual consumption sample library under different hypothetical scenarios. Based on real-time quality sensing data, establish the degradation curve of food quality and output a heat map of quality failure risk when the quality index exceeds the warning threshold in the future period. Step 4: Input the counterfactual consumption sample library and the quality failure risk heat map into the predictive analysis network to perform probability consumption and quality analysis, and obtain the probability distribution of future consumption of valuable ingredients and the quality evolution results. Step 5: Construct a multi-objective cost function with the goal of minimizing inventory holding costs, stockout penalty costs, and quality scrap losses. Solve for the dynamic safety stock threshold based on the consumption probability distribution and quality evolution results, generate an adaptive replenishment strategy, and output it to the procurement management end.
2. The hotel warehousing scheduling and predictive optimization method based on artificial intelligence according to claim 1, characterized in that, The following is a detailed data collection process: Intensity of target business events, environmental disturbance factors, supply chain indicators, and IoT quality sensing data for the hotel area. Based on the outbound records of target ingredients in each area of the warehouse and the sales back-calculation data, standardized consumption time-series data is formed; The daily food consumption and time-point business indicators are weighted into an event intensity sequence, and data is continuously collected at the granularity of a period to form time series data of business activity. By comprehensively analyzing the on-time delivery rate of ingredients, batch quality level, and supplier reliability index, we can obtain supply chain stability index data. By collecting data on remaining shelf life, total bacterial count, and sensory scores using temperature and humidity sensors, dedicated quality testing instruments, and manual data entry terminals deployed in cold storage and refrigerated trucks, a comprehensive quality index data of 0-1 is set up to integrate multiple quality parameters. Spatial region identification and alignment are performed on the collected consumption time series data, time series data, supply chain stability index data, and quality index data to output a standardized spatiotemporal correlation dataset.
3. The hotel warehouse scheduling and predictive optimization method based on artificial intelligence according to claim 1, characterized in that, Based on a spatiotemporal correlation dataset, the temporal correlation of consumption fluctuations in different regions is calculated, and the time offset between fluctuations is identified as the transmission delay time of business impact between regions. The specific process is as follows: By extracting time-series data on the consumption of valuable ingredients in any two regions at the same time from a standardized spatiotemporal correlation dataset, and calculating the cross-correlation coefficient between the two sequences based on the time offset using the transmission delay window of the business unit, a cross-correlation function curve is generated with the delay time as the horizontal axis and the correlation coefficient as the vertical axis. On the cross-correlation curve, filter out the delay time point that makes the absolute value of the correlation coefficient reach the peak, determine the direction of influence transmission based on the sign of the delay time point at the peak, and record the absolute value of the delay time point at the peak as the business influence transmission delay time between the two regions. Based on the obtained propagation delay time, the data is classified and labeled according to different regions to form a delay time dataset.
4. The hotel warehousing scheduling and predictive optimization method based on artificial intelligence according to claim 1, characterized in that, The confounding factors are separated, and the causal influence coefficient matrix after deconfounding correction is calculated. The specific process is as follows: The hotel promotional activities, holidays, and out-of-stock events were used as the criteria for screening confounding factors. The spatiotemporal correlation datasets of each hotel region were analyzed and filtered to obtain the spatiotemporal correlation data separated for each region. The separated screening data are statistically analyzed to determine the duration of promotional activities and the duration of stockouts in each region within the time window of the cycle. This data serves as an observable confounding factor vector. The consumption residual sequence after removing the confounding effects is used as a proxy indicator for the real business intervention variable, representing the consumption changes driven by the region's own operational fluctuations after removing external interference. Using confounding factors as control variables and actual intervention variables as treatment variables, a causal impact analysis of sales volume in a region was conducted to obtain the causal impact coefficients between regions. The effect coefficients of all region pairs were arranged according to the source region and the affected region to form a matrix of causal impact coefficients between regions.
5. The hotel warehouse scheduling and predictive optimization method based on artificial intelligence according to claim 1, characterized in that, Based on the causal impact coefficient matrix, the intensity of business events is analyzed to generate counterfactual samples of scenarios. According to the consumption change trajectory under different scenarios, a counterfactual consumption sample library is generated. The specific process is as follows: Based on the causal influence coefficient matrix, the key business event types that affect the consumption of the target area are identified. According to the distribution characteristics of historical data, the disturbance amplitude range and disturbance step size are set to form a disturbance parameter combination. For each selected disturbance target area, multiple hypothetical scenarios are set for disturbance transformation. The intensity of the disturbed business events is used to calculate the trajectory of consumption changes through the net effect coefficient and propagation delay time in the coefficient matrix, generating a complete consumption time series for each disturbance scenario, thus forming a preliminary counterfactual sample. All counterfactual samples and their corresponding confidence weights and perturbation parameter labels are stored in a structured manner to construct a counterfactual consumption sample library.
6. The hotel warehouse scheduling and predictive optimization method based on artificial intelligence according to claim 1, characterized in that, The quality status is established based on real-time quality indicators, a quality degradation curve is constructed, and a heatmap of quality failure risk when the quality index exceeds the warning threshold in the future period is output. The specific process is as follows: Acquire real-time data from IoT sensors in the storage area, and plot a scatter plot of quality decay for the current batch of ingredients with storage time on the horizontal axis and quality index on the vertical axis. Analyze all historical batch data, statistically analyze the differences in food decay rate under different temperature and humidity ranges, use the decay rate under standard conditions as a benchmark, calculate the accelerated decay coefficient under high temperature and high humidity conditions and the decelerated decay coefficient under low temperature and low humidity conditions, and establish a mapping relationship table between environmental parameters and decay coefficients. Based on the mapping relationship table between environmental parameters and decay coefficients, starting with the latest measured quality index of the current batch of ingredients, and combined with the predicted environmental parameters for future periods, the real-time decay rate at each future time point is calculated to form a dynamic quality degradation curve. On the degradation curve, the quality index is checked point by point to see if it is lower than the preset warning threshold and scrap threshold. The specific time point when it first falls below the threshold is recorded, which is the predicted failure time. Using future time as the horizontal axis and batches of food from each storage area as the vertical axis, a two-dimensional risk matrix is constructed to generate an intuitive heat map of quality failure risk, with key information of each batch of food marked on the map.
7. The hotel warehousing scheduling and predictive optimization method based on artificial intelligence according to claim 1, characterized in that, The counterfactual consumption sample library and the heat map of quality failure risk are input into the predictive analysis network to perform probabilistic consumption and quality analysis, obtaining the probability distribution of future consumption of each valuable ingredient in each region and the results of quality evolution. The specific process is as follows: The counterfactual consumption sample library and the quality failure risk heatmap are time-aligned and preprocessed to construct a multi-dimensional feature vector, which is then used as input features to the predictive analysis network. Construct a predictive analysis network to perform two prediction tasks: consumption probability distribution prediction and quality evolution result prediction. Based on the causal relationship in the counterfactual sample library, output multiple quantiles of consumption at each future time point. Based on the current quality index and the failure probability provided by the heatmap, output the mean of the future quality index. By summarizing the predicted quantiles for each region and time period, a probability density curve is generated. The predicted values of the quality index are then connected to form a curve, and the expected time points when the quality index falls below the warning threshold are marked. This yields the probability distribution of future consumption and the results of quality evolution for each valuable food ingredient.
8. The hotel warehousing scheduling and predictive optimization method based on artificial intelligence according to claim 1, characterized in that, A multi-objective cost function is constructed with the objectives of minimizing inventory holding costs, stockout penalty costs, and quality scrap losses. Based on the consumption probability distribution and quality evolution results, a dynamic safety stock threshold is calculated, and an adaptive replenishment strategy is generated and output to the procurement management system. The specific process is as follows: Based on the probability distribution of consumption and the quality evolution trajectory, three cost functions are defined respectively: inventory holding cost, stockout penalty cost, and quality scrap loss; Using the next 30 days as a rolling window, the consumption forecasts at different quantiles are substituted into the cost function, and the inventory level critical point that minimizes the sum of the three costs is searched. The system compares the current inventory level with the dynamic safety stock threshold in real time. When the inventory level falls below the threshold, it automatically triggers a replenishment suggestion, calculates the replenishment quantity, determines the order date, outputs quality risk warnings and replenishment instructions, and automatically pushes them to the procurement management personnel's terminal daily.