A method and apparatus for predicting a clean preparation duration for a car delivery

Abstract

This invention discloses a method and apparatus for predicting the cleaning preparation time for car rental delivery, comprising: determining the current order information, historical order sequence information, real-time store status information, current order user profile information, and environmental parameter information of the vehicle to be cleaned; constructing a time-series decay and stain accumulation coupled model based on the historical order sequence information, and calculating the time-series coupled stain accumulation assessment value; calculating the basic cleaning time based on the current order information, user profile information, and the assessment value; constructing a queuing theory nonlinear load correction model based on the real-time store status information, and calculating the queuing correction factor; constructing an environmental stain solidification coupled correction model based on environmental parameters and dominant stain types, and calculating the environmental correction coefficient; synthesizing the total cleaning preparation time based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient; calculating the delivery readiness index based on the total cleaning preparation time and the scheduled pick-up time, and outputting the order decision based on an adaptive soft threshold.

Description

Technical Field

[0001] This application relates to the field of intelligent scheduling and operations optimization technology for car rental, and in particular to a method and apparatus for predicting the cleaning preparation time for car delivery. Background Technology

[0002] In the field of car rental and shared mobility technology, car rental platforms need to process thousands or even tens of thousands of orders every day, and users' requirements for the timeliness of car pick-up and drop-off are increasing. After completing the previous order, the vehicle must undergo a series of preparation processes such as internal and external cleaning, disinfection, and inspection before it can be delivered. The accurate estimation of the cleaning and preparation time directly determines whether the order can be confirmed in a timely manner and whether the user can pick up the car on time after arriving at the store.

[0003] In existing technologies, car rental platforms typically use fixed-duration estimation or simple linear fitting methods to estimate cleaning preparation time. This means setting a fixed cleaning time window in advance, or linearly calculating based on a few factors such as the driving time and mileage of the previous order. As a result, it is difficult to depict the congestion and delays that may occur when the workload of the workstation increases. It may also lead to a serious underestimation of cleaning preparation time during peak periods, resulting in order delays, while during off-peak periods, it may over-reserve cleaning resources, resulting in waste. Summary of the Invention

[0004] This specification provides a method and apparatus for predicting the cleaning and preparation time for car rental delivery, in order to solve at least one of the technical problems mentioned above.

[0005] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows: According to a first aspect of the embodiments of this specification, a method for predicting the cleaning preparation time for car rental delivery is provided, comprising: Determine the current order information, historical order sequence information, real-time store status information, current order user profile information, and environmental parameter information of the vehicles to be cleaned; Based on the historical order sequence information, a coupled model of time-series decay and stain accumulation between orders is constructed to calculate the time-series coupled stain accumulation evaluation value. Based on the current order information, the current order user profile information, and the time-coupled stain accumulation assessment value, calculate the basic cleaning duration; Based on the real-time status information of the stores, a nonlinear load correction model for store queuing based on the M / M / c queuing theory is constructed to calculate the queuing correction factor. Based on the dominant stain type corresponding to the environmental parameter information and the current order user profile information, an environment-stain curing coupling correction model is constructed to calculate the environmental correction coefficient; The total cleaning preparation time is calculated based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient. Calculate the delivery readiness index based on the total cleaning and preparation time and the user's scheduled vehicle pick-up time; Based on the delivery readiness index and the adaptive soft threshold, an order decision is output.

[0006] In some optional implementations, the step of synthesizing the total cleaning preparation time based on the base cleaning time, the queuing correction factor, and the environmental correction coefficient includes: The total cleaning preparation time is obtained by multiplying the basic cleaning time, the queuing correction factor, and the environmental correction coefficient, and adding the fixed vehicle inspection time and the fixed vehicle scheduling time. The fixed duration for vehicle inspection and the fixed duration for vehicle dispatch are predefined based on vehicle type or store layout.

[0007] In some optional implementations, the step of constructing a time-series decay and stain accumulation coupling model based on the historical order sequence information to calculate the time-series coupled stain accumulation assessment value includes: Extract the driving time, mileage, user stain index, and order end time of each historical order from the historical order sequence information; Calculate the stain contribution weight of each order based on the driving time and mileage of each historical order. ; Among them, symbols Indicates the first The travel time of each historical order, symbol This indicates the maximum driving time of this vehicle model in historical data, symbol... Indicates the first Mileage of a historical order, symbol This indicates the maximum mileage of this vehicle model in historical data, symbol... Indicates the driving time weighting coefficient, symbol This represents the mileage weighting coefficient; According to the current time With the end time of each historical order time interval Calculate the time-series decay factor , where the symbol Indicates the timing decay coefficient; The individual order stain contribution weights for each historical order. User Stain Index Vehicle model coefficient and the aforementioned time-series decay factor Sum the products and add them together with the baseline stain level. The sums are used to obtain the time-coupled stain accumulation assessment value. .

[0008] In some optional implementations, calculating the basic cleaning duration based on the current order information, the current order user profile information, and the time-coupled stain accumulation assessment value includes: Obtain baseline cleaning time Vehicle model coefficient The driving time of the previous order Mileage and its historical average and And the differentiated stain index of current order users. ; Based on the travel time of the previous order With the mileage Deviation from the historical average and The degree of elasticity, combined with the elastic coefficient and Calculate driving characteristic correction terms ; The benchmark cleaning time The vehicle model coefficient The driving characteristic correction item and the time-series coupled stain accumulation evaluation value And the differentiated stain index of the current order user Multiply by the product to obtain the basic cleaning time. : .

[0009] In some optional implementations, determining the current order user profile information of the vehicle to be cleaned includes: Extract the distribution vectors of user order frequency, average order duration, average mileage, and vehicle usage time within a preset time window. and vectors of common dot types , constitute user feature vector ; The user feature vector is processed using a clustering algorithm. Clustering is performed to categorize users into various profile types, including short-distance commuters, long-distance travelers, and high-frequency business travelers. For each image type, a differentiated set of stain pattern parameters is predefined, the set of parameters including the dominant stain weight vector. Stain depth baseline value and additional cleaning difficulty coefficient; Based on the user profile of the current order, dynamically determine their differentiated stain index. .

[0010] In some optional implementations, the step of constructing a nonlinear load correction model for store queuing based on the real-time status information of the store and calculating the queuing correction factor includes: Vehicle arrival rate estimated based on real-time store data. Service rate per workstation and number of effective service counters ; Based on the number of workstations currently in operation With the number of effective service counters Calculate workstation occupancy rate ; Queuing correction factor is calculated using piecewise functions. : Among them, symbols Represents the critical occupancy rate threshold, symbol Represents the linear queuing correction factor, symbol Represents the saturated nonlinear amplification factor, symbol This indicates the number of vehicles in the queue.

[0011] In some optional implementations, the step of constructing an environment-stain curing coupling correction model based on the dominant stain type corresponding to the environmental parameter information and the current order user profile information to calculate the environmental correction coefficient specifically includes: The dominant stain type is determined based on the current order user profile information. The dominant stain type includes organic stains, inorganic stains, grease stains, or a mixture of at least two types of stains. Get real-time ambient temperature relative humidity and seasonal factors ; When the dominant stain type is a single type, the corresponding correction function is selected; wherein, for organic stains, the environmental correction factor for inorganic stains is calculated according to the following formula. : Among them, symbols Represents the reference temperature, symbol Represents the reference relative humidity, symbol and Indicates the environmental sensitivity coefficient of organic stains; For inorganic stains, the environmental correction factor for inorganic stains is calculated using the following formula. : Among them, symbols Indicates the reference value for freezing point temperature, symbol and Indicates the environmental sensitivity coefficient of inorganic stains; For grease stains, calculate the environmental correction factor for grease stains using the following formula. : Among them, symbols Indicates the humidity reference scaling factor, symbol Indicates the environmental sensitivity coefficient of grease stains; When the dominant stain type is a mixed type, the correction coefficient for each individual stain type is calculated separately, and then the dominant stain weight vector is predefined in the user profile information. We perform a weighted summation to obtain the comprehensive environmental correction coefficient. ; Among them, symbols Indicates the stain type index, symbol Indicates the corresponding weight, symbol This represents the corresponding correction factor.

[0012] According to a second aspect of the embodiments of this specification, a device for predicting the cleaning preparation time for car rental delivery is provided, comprising: The information determination module is used to determine the current order information, historical order sequence information, real-time store status information, current order user profile information, and environmental parameter information of the vehicle to be cleaned; The evaluation value construction module is used to construct a coupling model of time-series decay and stain accumulation between orders based on the historical order sequence information to calculate the time-series coupled stain accumulation evaluation value; The basic cleaning time calculation module is used to calculate the basic cleaning time based on the current order information, the current order user profile information, and the time-coupled stain accumulation assessment value. The queuing correction factor calculation module is used to construct a nonlinear load correction model for store queuing based on the real-time status information of the store and calculate the queuing correction factor. The environmental correction coefficient calculation module is used to construct an environment-stain curing coupling correction model based on the dominant stain type corresponding to the environmental parameter information and the current order user profile information to calculate the environmental correction coefficient; The total cleaning preparation time synthesis module is used to synthesize the total cleaning preparation time based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient. The delivery readiness index calculation module is used to calculate the delivery readiness index based on the total cleaning preparation time and the user's scheduled vehicle pick-up time. The order decision output module is used to output order decisions based on the delivery readiness index and the adaptive soft threshold.

[0013] One embodiment of this specification can achieve at least the following beneficial effects: In the technical solution of this application, when predicting the vehicle cleaning preparation time, the car rental platform does not directly use a fixed duration or a simple linear fitting method. Instead, it first determines the current order information, historical order sequence information, real-time store status information, current order user profile information, and environmental parameter information of the vehicle to be cleaned. Then, based on the historical order sequence information, it constructs a coupling model of time-series decay and stain accumulation between orders to calculate the time-series coupled stain accumulation assessment value. Based on the current order information, user profile information, and this assessment value, it calculates the basic cleaning time. Simultaneously, based on the real-time store status information, it constructs a store queuing nonlinear load correction model based on M / M / c queuing theory to calculate the queuing correction factor. Furthermore, based on the environmental parameter information and the dominant stain type, it constructs an environment-stain curing coupling correction model to calculate the environmental correction coefficient. Finally, it synthesizes the total cleaning preparation time based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient. The system calculates the delivery readiness index based on the total cleaning preparation time and the user's scheduled pick-up time. Finally, it outputs order decisions based on the delivery readiness index and an adaptive soft threshold. This solution not only characterizes the accumulation and decay of stains between consecutive orders through a coupled model of time-series decay and stain accumulation, but also accurately reflects the congestion mutation characteristics when the store's workstation occupancy rate increases using the nonlinear piecewise function of the M / M / c queuing theory. At the same time, it achieves differentiated allocation of cleaning resources by combining user behavior profiles and environmental coupling correction. This is equivalent to establishing a full-link intelligent prediction mechanism for the car rental platform, from stain load assessment, queuing delay quantification, environmental factor depreciation to delivery readiness normalization decision-making. This enables the system to predict cleaning preparation time with minute-level accuracy and adaptively adjust the decision boundaries for order confirmation, delay monitoring, or rejection recommendation based on the store's real-time load, thereby improving order fulfillment success rate and user satisfaction. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a comparison curve between the nonlinear queuing correction model in this application and the existing linear model; Figure 2 This is a Sigmoid normalized curve of the delivery readiness index in the technical solution of this application; Figure 3 This is a convergence curve of the model gradient descent optimization weight parameters in the technical solution of this application; Figure 4This is a structural diagram of a device for predicting the cleaning and preparation time for car delivery, provided in the technical solution of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of one or more embodiments of this specification clearer, the technical solutions of one or more embodiments of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of one or more embodiments of this specification.

[0017] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another.

[0018] The following is a detailed description of the solution based on the accompanying drawings. This application provides a method for predicting the cleaning and preparation time for car rental delivery, which may include: Step 102: Determine the current order information, historical order sequence information, real-time store status information, current order user profile information, and environmental parameter information of the vehicle to be cleaned.

[0019] In the embodiments of this specification, the vehicle to be cleaned can refer to a vehicle that has just completed the previous order and needs cleaning and preparation before being delivered to the next user. For car rental platforms, each vehicle must undergo interior and exterior cleaning, disinfection, and inspection after each rental service before it can be delivered. Current order information may include data directly related to the current user's scheduled pick-up time, selected vehicle model, usage duration, pick-up and drop-off point identification, etc. This information provides the time constraints and vehicle specification constraints that the cleaning service must meet, and is the basis for determining whether delivery can be made on time. Historical order sequence information can refer to the collection of all orders completed by the vehicle within a past period. Each historical order record includes the order end time, driving duration, mileage, and the stain index of the user who placed the order. The length of the historical order sequence usually has a preset upper limit, because the impact of older historical orders on the current cleaning needs is negligible. This information is used to capture the accumulation and decay patterns of stains during multiple orders and is the input to the time-series decay model. Real-time store status information can include data reflecting the real-time load of the store's cleaning system, such as the current occupancy of cleaning workstations, the number of vehicles waiting in line for cleaning, the number of workers on duty, and the current time period. Current order user profile information refers to user type tags and expected stain behavior patterns obtained by analyzing the user's historical car rental behavior. This user profile information can reflect the differences in the type, quantity, and depth of stains that different users may generate when using vehicles. Environmental parameter information can include real-time ambient temperature, relative humidity, and seasonal factors of the store's location. These parameters can be used to quantify the coupled impact of environmental conditions on stain curing rate and cleaning efficiency.

[0020] Step 104: Based on the historical order sequence information, construct a time-series decay and stain accumulation coupling model between orders to calculate the time-series coupled stain accumulation evaluation value.

[0021] In the embodiments of this specification, the order-to-order temporal decay and stain accumulation coupling model can refer to a technical means for quantifying the level of stains accumulated by a vehicle during multiple consecutive orders. The design idea of ​​this model in the technical solution of this application is that the stains remaining after a vehicle completes an order do not exist indefinitely, but gradually decay over time. Simultaneously, the intensity of stains generated by different orders varies due to differences in driving characteristics and user behavior. The construction process of this model can include three key steps: calculation of single-order stain contribution weight, calculation of temporal decay factor, and synthesis of stain accumulation evaluation value. Specifically, for each historical order, the single-order stain contribution weight is first calculated based on its driving time and mileage. Driving time and mileage reflect the time and distance the vehicle is used, respectively, and both jointly determine the total amount of stains generated. Since the two have different dimensions and large differences in their numerical ranges, they need to be normalized by dividing by their maximum values ​​in historical data, and then multiplied by their respective preset weight coefficients before summing. The weight coefficients can be determined through training based on actual operating data, reflecting the relative importance of duration and mileage to stain contribution. Secondly, the time interval between the current moment and the end time of each historical order is calculated, and the time-series decay factor is calculated using an exponential function. The exponential decay function is a commonly used mathematical form to describe the natural dissipation process of stains, and its decay rate is controlled by a learnable coefficient. This design reflects observations in actual operation, namely that after a vehicle is returned, some stains will gradually dissipate due to natural settling, drying, or use by subsequent vehicles; the longer the interval, the smaller the impact of historical orders on the current cleaning needs.

[0022] Finally, the products of the stain contribution weight of each historical order, the user stain index, the vehicle model coefficient, and the time decay factor are summed and added to the baseline stain level to obtain the time-coupled stain accumulation assessment value. The baseline stain level represents the vehicle's basic cleaning needs starting from its standard maintenance state and can be a preset benchmark value. This assessment value integrates the usage history of multiple recent orders and considers the time decay effect, providing an accurate measure of stain load for subsequent cleaning time calculations. The higher the assessment value, the more stains the vehicle currently has accumulated, and the longer the cleaning time required.

[0023] Step 106: Calculate the basic cleaning duration based on the current order information, the current order user profile information, and the time-coupled stain accumulation assessment value.

[0024] In the embodiments of this specification, the basic cleaning time refers to the theoretical operating time required to clean the vehicle without considering store queuing congestion and environmental factors. In the technical solution of this application, this time is not calculated using a fixed value, but rather by coupling multiple factors. This multi-factor product design reflects the multiplicative coupling effect between factors; that is, when multiple adverse factors occur simultaneously, the basic cleaning time is drastically amplified, thus more realistically reflecting the actual cleaning needs under complex scenarios. Specifically, it starts with the baseline cleaning time (determined by the vehicle model). The baseline cleaning time is the basic time required to clean the vehicle model under standard conditions, derived from historical experience statistics of standard cleaning procedures. Different vehicle models have different values ​​due to differences in cleaning area and complexity. Then, it is multiplied by a vehicle model coefficient. The vehicle model coefficient is the first amplification of the baseline time, and different vehicle models have different values ​​due to differences in interior space size, number of seats, and cleaning standards. For example, large-space vehicles typically have a higher coefficient than standard vehicles due to their larger cleaning area and more cleaning steps.

[0025] Next, a driving characteristic correction term is applied, reflecting the deviation of the previous order's driving characteristics (duration and mileage) from the historical average. This correction term can be set to 1; when the driving time or mileage exceeds the historical average, the correction term is greater than 1, and vice versa. Elasticity coefficients control the marginal impact of the degree of out-of-average on cleaning time. These coefficients can be determined through training, and this design allows the model to impose a reasonable penalty extension on abnormally long trips.

[0026] Then multiply by the time-coupled stain accumulation assessment value calculated in step 104. This value reflects the stain load accumulated by the vehicle in the history of multiple orders. The larger the value, the longer the cleaning time required. For example, if the assessment value is significantly greater than 1, it means that the current stain load of the vehicle is higher than the standard state by a corresponding proportion, and the cleaning time also needs to be increased accordingly.

[0027] Finally, multiply by the differentiated stain index of the current order user. This index reflects the type and depth of stains that the user about to use the vehicle may produce, with significant differences between different user profiles. Through this multi-factor coupling method, differentiated base cleaning time can be output for different vehicles, different users, and different usage histories. Increasing any one factor will lead to a corresponding increase in base cleaning time, and when multiple factors increase simultaneously, their product effect will significantly increase the cleaning time.

[0028] Step 108: Based on the real-time status information of the stores, construct a nonlinear load correction model for store queuing based on M / M / c queuing theory and calculate the queuing correction factor.

[0029] In the embodiments of this specification, the nonlinear load correction model for store queuing based on M / M / c queuing theory can refer to a mathematical model used to characterize the nonlinear abrupt congestion characteristics of a store cleaning operation system under high load conditions. This model is constructed based on the classic multi-service station queuing system in queuing theory. Its core assumptions are that vehicle arrivals follow a Poisson distribution, cleaning service time follows an exponential distribution, and there are multiple service stations (cleaning workstations). In this scheme, the store cleaning operation system can be modeled as a multi-service station queuing system, where vehicles arrive randomly, and cleaning workstations serve as service stations for vehicles. The model first estimates the vehicle arrival rate, the service rate per workstation, and the number of effective service stations based on real-time store data. The vehicle arrival rate can be estimated in real-time by statistically analyzing the number of vehicles entering the cleaning queue over a past period. The service rate per workstation can be estimated by statistically analyzing the average number of vehicles completed per workstation over a past period to smooth individual efficiency fluctuations. The number of effective service stations needs to be adjusted in real-time considering the number of workers on duty.

[0030] Then, calculate the current workstation occupancy rate, which is the number of workstations currently in operation divided by the number of effective service counters. Workstation occupancy rate is a core indicator for measuring store load status, ranging from 0 to 1. When the occupancy rate is less than 1, the system is in a stable state. When the occupancy rate approaches 1, the queue length will theoretically grow indefinitely. The queue correction factor can be calculated using a piecewise function. When the workstation occupancy rate is below the critical occupancy rate threshold, the system is in a light load zone, and the waiting time is linearly related to the number of vehicles in the queue. The correction factor is 1 plus a linear coefficient multiplied by the number of vehicles in the queue. In this stage, for each additional vehicle in the queue, the waiting time increases by a fixed proportion. When the workstation occupancy rate reaches or exceeds the critical occupancy rate threshold, the system enters the saturation zone. At this time, an exponential saturation term can be added to the queue correction factor. The denominator of this saturation term is 1 minus the workstation occupancy rate. As the occupancy rate approaches 1, the denominator approaches 0, and the saturation term approaches infinity. This characteristic accurately depicts the theoretical phenomenon of infinitely increasing queue length when workstation utilization approaches saturation. The nonlinear piecewise function design in this solution allows the system to realistically reflect the actual operational experience where "even slight congestion can lead to severe delays," avoiding the flaw of linear models that significantly underestimate waiting times under high load. The critical occupancy threshold and various coefficients can be determined through training with historical data to adapt to the operational characteristics of different stores.

[0031] Step 110: Based on the dominant stain type corresponding to the environmental parameter information and the current order user profile information, construct an environment-stain curing coupling correction model to calculate the environmental correction coefficient.

[0032] In the embodiments of this specification, the environment-stain curing coupling correction model can refer to a correction mechanism used to quantify the influence of environmental temperature, humidity, and seasonal factors on the stain curing rate. Different types of stains have significantly different sensitivities to environmental conditions; therefore, it is necessary to select the corresponding correction function form according to the dominant stain type.

[0033] First, the dominant stain type is determined based on the current order user profile information. Different user profile types correspond to different dominant stains; for example, some user profiles are dominated by organic stains, some by inorganic stains, some by grease stains, and some by mixed stains. This determination determines which correction function to use subsequently. Next, real-time ambient temperature, relative humidity, and seasonal factors are obtained. Seasonal factors are used to capture the macro-environmental differences between seasons. For example, in cold seasons, additional corrections are needed due to de-icing agent salt stains, while in hot and humid seasons, high temperature and humidity may accelerate the curing of organic stains. For different types of stains, the corresponding correction function is selected. Organic stains ferment and cure faster in high temperature and humidity environments, and their correction coefficient increases with increasing temperature and humidity. Inorganic stains exhibit enhanced adhesion at low temperatures, and their correction coefficient increases with decreasing temperature. The curing and oxidation process of grease stains is affected by the interaction of temperature and humidity, and their correction coefficient includes cross terms. Each type of correction function can include a learnable sensitivity coefficient to control the degree of influence of various environmental factors on the correction coefficient.

[0034] When the dominant stain is of a mixed type, the correction coefficient for each individual stain type is calculated separately, and then the results are weighted and summed according to the predefined dominant stain weight vector in the user profile information to obtain the comprehensive environmental correction coefficient. For example, if a user profile contains both organic and inorganic stains, each with its corresponding weight, the comprehensive correction coefficient is the weighted sum of the organic and inorganic correction coefficients. This model allows cleaning schedules to be dynamically adjusted based on real-time environmental conditions, enabling advance scheduling of cleaning resources in high-risk solidification scenarios and optimized workstation scheduling in low-risk solidification scenarios.

[0035] Step 112: Based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient, synthesize the total cleaning preparation time.

[0036] In the embodiments of this specification, the total cleaning preparation time is synthesized by multiplying the base cleaning time, the queuing correction factor, and the environmental correction coefficient, and then adding the fixed vehicle inspection time and the fixed vehicle scheduling time. This "product-summation" synthesis method can take into account both the coupling amplification effect between factors and the independence of fixed process times.

[0037] Specifically, the basic cleaning time reflects the cleaning needs of the vehicle itself, the queue correction factor reflects the waiting delay caused by the current store load, and the environmental correction coefficient reflects the impact of environmental conditions on cleaning efficiency. Multiplying these three factors means that the cumulative effect of adverse factors is amplified non-linearly. For example, when the basic cleaning time is already long, the queue correction factor increases due to congestion, and the environmental correction factor increases due to inclement weather, the product of the three factors will be dramatically amplified, truly reflecting the surge in cleaning time caused by the superposition of multiple adverse factors. In addition, fixed vehicle inspection time and fixed vehicle dispatch time are added. Fixed vehicle inspection time includes standardized processes such as exterior environmental inspection, interior spot checks, and oil, water, and electrical checks; the value varies depending on the structural complexity of different vehicle models. Fixed vehicle dispatch time depends on the store area and the physical distance from the workstation to the delivery area. These two parts of the time are relatively independent of the basic cleaning process and are not significantly affected by stain load or queue status. Through this "product-sum" synthesis formula, this solution can output a complete cleaning preparation time window that comprehensively considers vehicle status, store load, environmental conditions, and standard operating procedures. This total time is the direct basis for subsequent delivery readiness calculations and decision outputs.

[0038] Step 114: Calculate the delivery readiness index based on the total cleaning preparation time and the user's scheduled vehicle pick-up time.

[0039] In the embodiments of this specification, the delivery readiness index can refer to a standardized indicator that transforms time margin into probabilistic readiness. The design goal of this index is to convert absolute time differences into probability values ​​between 0 and 1, facilitating unified decision comparison and threshold determination. Specifically, the time margin (which can be positive or negative) is obtained by subtracting the total cleaning preparation time and safety margin time from the user's scheduled pick-up time. The safety margin time is a buffer used to absorb minor fluctuations in the actual cleaning process, preventing promise failure due to prediction errors. The time margin is then divided by the safety margin time to obtain the basic readiness index. A positive basic readiness index indicates sufficient time margin (i.e., the expected completion time is at least one safety margin earlier than the scheduled pick-up time), while a negative index indicates that timely completion is not expected. Subsequently, the basic readiness index can be normalized using the Sigmoid function to obtain the delivery readiness index. The Sigmoid function is a commonly used probability mapping function, with an output range of (0,1). The curvature coefficient controls the steepness of the Sigmoid curve. The advantage of this normalization is that when the baseline readiness index is positive and larger, the delivery readiness index approaches 1; when the baseline readiness index is negative, the delivery readiness index approaches 0. When the baseline readiness index is close to 0, the Sigmoid function, under the control of the curvature coefficient, provides a smooth but steep transition. This allows small changes in preparation time in the critical region (where time margin is close to zero) to produce significant differences in decision signals, avoiding decision jitter near the linear threshold. Simultaneously, the standardized delivery readiness index eliminates comparability issues caused by differences in safety margin time settings across different stores and time periods, making this indicator a standardized metric that can be compared across stores.

[0040] Step 116: Output the order decision based on the delivery readiness index and the adaptive soft threshold.

[0041] In the embodiments of this specification, the adaptive soft threshold decision mechanism can refer to an intelligent decision-making method that dynamically adjusts the decision boundary based on the real-time load status of the store and outputs order confirmation, delay monitoring, or rejection recommendations. The design idea of ​​this mechanism in this solution is that the decision boundary is not fixed, but dynamically shifts with the store load, thereby achieving adaptive scheduling that actively accepts orders when resources are abundant and cautiously commits when resources are scarce. This solution can preset three basic decision thresholds: a confirmation threshold, a delay threshold, and a rejection threshold. These three thresholds divide the range of the delivery readiness index into three decision zones: a confirmation zone, a delay monitoring zone, and a rejection zone. The confirmation threshold and rejection threshold define the boundaries of high and low readiness, respectively, while the delay threshold lies between the two, used to identify intermediate states that require further observation. It should be noted that these three thresholds are not fixed but can be dynamically adjusted according to the current workstation occupancy rate of the store. Specifically, the adjustment rule can be: when the workstation occupancy rate is lower than a first preset value, the store is in a light-load state, all three thresholds are lowered, making orders easier to confirm, thereby making full use of idle cleaning resources. When the workstation occupancy rate exceeds the second preset value, the store is under heavy load. The system raises all three thresholds to more strictly screen orders and avoid orders that cannot be fulfilled. When the workstation occupancy rate is between the two, the thresholds remain unchanged. This "dynamic threshold offset" strategy in this solution allows the system's decision boundary to adaptively follow changes in store load, achieving intelligent scheduling that actively accepts orders when resources are abundant and makes promises cautiously when resources are scarce.

[0042] Finally, the delivery readiness index calculated in step 114 is compared with the three adjusted thresholds. Specifically, if the delivery readiness index is greater than or equal to the confirmation threshold, a confirmation order decision is output, and the system immediately confirms the order, promising the user that the vehicle will be ready before the scheduled time. If the delivery readiness index is between the delay threshold and the confirmation threshold, a delay monitoring decision is output, setting the order status to pending confirmation and initiating dynamic monitoring to automatically decide whether to confirm or reject based on subsequent changes in the readiness index. If the delivery readiness index is less than the rejection threshold, a rejection order and alternative vehicle recommendation decision is output, the system rejects the current order, and automatically searches for other deliverable vehicles to recommend to the user. Through this adaptive threshold mechanism, this solution can achieve a paradigm shift from a vague "probably deliverable" judgment to a precise decision driven by a quantified readiness index, thereby effectively solving the problem of the disconnect between "bookable" and "deliverable".

[0043] In the technical solution of this application, when predicting the vehicle cleaning preparation time, the car rental platform does not directly use a fixed duration or a simple linear fitting method. Instead, it first determines the current order information, historical order sequence information, real-time store status information, current order user profile information, and environmental parameter information of the vehicle to be cleaned. Then, based on the historical order sequence information, it constructs a coupling model of time-series decay and stain accumulation between orders to calculate the time-series coupled stain accumulation assessment value. Based on the current order information, user profile information, and this assessment value, it calculates the basic cleaning time. Simultaneously, based on the real-time store status information, it constructs a store queuing nonlinear load correction model based on M / M / c queuing theory to calculate the queuing correction factor. Furthermore, based on the environmental parameter information and the dominant stain type, it constructs an environment-stain curing coupling correction model to calculate the environmental correction coefficient. Finally, it synthesizes the total cleaning preparation time based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient. The system calculates the delivery readiness index based on the total cleaning preparation time and the user's scheduled pick-up time. Finally, it outputs order decisions based on the delivery readiness index and an adaptive soft threshold. This solution not only characterizes the accumulation and decay of stains between consecutive orders through a coupled model of time-series decay and stain accumulation, but also accurately reflects the congestion mutation characteristics when the store's workstation occupancy rate increases using the nonlinear piecewise function of the M / M / c queuing theory. At the same time, it achieves differentiated allocation of cleaning resources by combining user behavior profiles and environmental coupling correction. This is equivalent to establishing a full-link intelligent prediction mechanism for the car rental platform, from stain load assessment, queuing delay quantification, environmental factor depreciation to delivery readiness normalization decision-making. This enables the system to predict cleaning preparation time with minute-level accuracy and adaptively adjust the decision boundaries for order confirmation, delay monitoring, or rejection recommendation based on the store's real-time load, thereby improving order fulfillment success rate and user satisfaction.

[0044] Based on the technical solutions described above, this specification also provides some specific implementation schemes, which are described below.

[0045] In an optional embodiment, the step of synthesizing the total cleaning preparation time based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient may include: The total cleaning preparation time is obtained by multiplying the basic cleaning time, the queuing correction factor, and the environmental correction coefficient, and adding the fixed vehicle inspection time and the fixed vehicle scheduling time. The fixed duration for vehicle inspection and the fixed duration for vehicle dispatch are predefined based on vehicle type or store layout.

[0046] In this embodiment, the total cleaning preparation time can be calculated using a combination of multiplication and addition. Multiplying the base cleaning time, the queue correction factor, and the environmental correction factor reflects the combined impact of vehicle soil load, store queue congestion, and environmental conditions on the cleaning operation's time consumption. These three factors are coupled, amplifying or reducing each other. For example, when the base cleaning time is already long, the queue correction factor increases due to congestion, and the environmental correction factor increases due to inclement weather, the product of the three factors will be dramatically amplified, accurately reflecting the surge in cleaning time caused by the superposition of multiple adverse factors.

[0047] Building upon this, fixed timeframes for vehicle inspection and vehicle dispatch are added. These two timeframes are relatively independent of the basic cleaning process and are not significantly affected by stain load or queue status. Fixed vehicle inspection timeframes typically include standardized procedures such as exterior environmental inspection, interior spot checks, and oil, water, and electrical checks. The value varies depending on the vehicle model and its structural complexity; for example, sedans are typically 5 minutes, SUVs 6 minutes, and MPVs 8 minutes. Fixed vehicle dispatch timeframes depend on the store area and the physical distance from the workstation to the delivery area, usually ranging from 3 to 8 minutes. This product-sum formula preserves the non-linear coupling effects between factors without unnecessarily amplifying fixed process times, resulting in a complete and accurate total cleaning preparation time.

[0048] In an optional embodiment, the step of constructing a coupled model of time-series decay and stain accumulation between orders based on the historical order sequence information to calculate the time-series coupled stain accumulation assessment value may include: Extract the driving time, mileage, user stain index, and order end time of each historical order from the historical order sequence information; Calculate the stain contribution weight of each order based on the driving time and mileage of each historical order. ; Among them, symbols Indicates the first The travel time of each historical order, symbol This indicates the maximum driving time of this vehicle model in historical data, symbol... Indicates the first Mileage of a historical order, symbol This indicates the maximum mileage of this vehicle model in historical data, symbol... Indicates the driving time weighting coefficient, symbol This represents the mileage weighting coefficient; According to the current time With the end time of each historical order time interval Calculate the time-series decay factor , where the symbol Indicates the timing decay coefficient; The individual order stain contribution weights for each historical order. User Stain Index Vehicle model coefficient and the aforementioned time-series decay factor Sum the products and add them together with the baseline stain level. The sums are used to obtain the time-coupled stain accumulation assessment value. .

[0049] In this embodiment, the contribution weight of stains in a single order The design is based on an intuitive physical assumption: the longer the driving time and the greater the mileage, the more dirt will accumulate on the vehicle during use. Since driving time and mileage have different dimensions and significantly different numerical ranges, they are normalized by dividing by the maximum value in the historical data of this vehicle model, and then multiplied by a weighting coefficient. and Then add them together. In a typical implementation, and A weight of 0.5 can be applied to both duration and mileage, resulting in an equal weighting. This normalized weighting method ensures that features of different dimensions are fairly included in the contribution assessment.

[0050] The time-series decay factor uses an exponential decay function. ,in This represents the decay coefficient. The physical meaning of exponential decay is that the impact of a stain gradually weakens over time, and the decay rate is exponentially related to the time interval. For example, when... When the value is 0.045, the stain contribution from orders placed 30 minutes apart is only about 26%, while the contribution from orders placed 120 minutes apart is negligible. This design reflects observations from actual operation, namely that after a vehicle is returned, some stains (such as footprints and dust) will gradually dissipate due to natural settling or use by subsequent vehicles, while after a long period of parking, the stain impact of previous orders is almost negligible.

[0051] User Stain Index This reflects the contribution of historical order user behavior to stain generation; for example, tourists typically generate more mud and food scraps than commuters. Vehicle type coefficient. This reflects the amplified effect of different vehicle models on cleaning needs due to variations in interior space, number of seats, and cleaning standards. For example, the coefficient for SUVs is generally higher than that for sedans, and the coefficient for MPVs is even higher. Multiplying all the above factors and summing them, then adding a base stain level... (This represents the basic cleaning requirements of the vehicle starting from its standard maintenance condition, typically taken as 1.0), which yields the time-coupled stain accumulation assessment value. This value incorporates the usage history of multiple recent orders for the vehicle and takes into account the time decay effect, providing an accurate measure of stain load for subsequent cleaning duration calculations.

[0052] In an optional embodiment, the step of constructing a coupled model of time-series decay and stain accumulation between orders based on the historical order sequence information to calculate the time-series coupled stain accumulation assessment value may further include: Based on the workstation occupancy rate in the store's real-time status information Based on the current time period type, dynamically determine the maximum number of historical order sequences to retain. : When the workstation occupancy rate And when the current period is a low-peak period, the maximum number of items to be retained. Take the highest value; When the workstation occupancy rate Or, when the current time period is a peak period, the maximum number of items to be retained. Take the lowest value; When the workstation occupancy rate Between the first occupancy threshold With the second occupancy threshold When in between, the maximum number of retentions Take the median value, and the maximum number of retentions. With the aforementioned workstation occupancy rate It exhibits a monotonically decreasing linear relationship; Extract the most recent maximum retention quantity from the historical order sequence information. One historical order is used as the valid order sequence; For each historical order in the valid order sequence, based on the time interval between the current time and the order's end time. Calculate the dynamic truncation weight factor , where the symbol Represents the maximum time interval in the valid order sequence, symbol Indicates the steepness coefficient of the cutoff curve; The dynamic truncation weight factor Multiply by the time decay factor In this process, the corrected time-coupled stain accumulation assessment value is obtained. .

[0053] In this embodiment, considering the scarcity of cleaning resources during peak hours or periods of high store load, the system needs to prioritize the impact of recent orders on cleaning demand, while quickly truncating older historical orders to avoid prolonged waiting times due to interference from historical orders. To this end, this solution introduces a mechanism that dynamically adjusts the maximum number of historical orders to be retained based on workstation occupancy and time period type.

[0054] Specifically, when the workstation occupancy rate is low (e.g.) When cleaning resources are relatively abundant during off-peak hours (such as late at night or early morning), more historical orders can be retained (e.g., up to 10 orders) to finely depict stain accumulation trends and ensure prediction accuracy. However, when workstation occupancy exceeds a high threshold (e.g., ...), ... During peak periods (such as morning and evening rush hours or holidays), when cleaning resources are scarce, the system retains only the most recent few orders (e.g., the last 3 orders) to quickly respond to current cleaning needs and avoid "diluting" the current actual stain load by older historical orders. When the occupancy rate is between the middle value, the maximum number of orders retained has a monotonically decreasing linear relationship with the occupancy rate, achieving a smooth transition.

[0055] After determining the valid order sequence, this scheme further introduces a dynamic truncation weight factor. This factor uses Functional form ,in This represents the maximum time interval in the valid order sequence. A key characteristic of this function is that it is effective for time intervals close to... The oldest order (i.e., the oldest order). Approaching 0.5; for time intervals much smaller than The most recent order. Approaching 1. Combined with a steepness coefficient. Adjustments can achieve a smooth yet rapid transition zone. Multiply by the original timing decay factor This is equivalent to a secondary modulation of the exponential decay, that is, while retaining the exponential decay trend, additional decay is applied to long-interval orders based on the store load status. This allows the model to focus more on recent stain contributions in high-load scenarios, thereby optimizing the allocation efficiency of cleaning resources.

[0056] In an optional embodiment, calculating the basic cleaning duration based on the current order information, the current order user profile information, and the time-coupled stain accumulation assessment value may include: Obtain baseline cleaning time Vehicle model coefficient The driving time of the previous order Mileage and its historical average and And the differentiated stain index of current order users. ; Based on the travel time of the previous order With the mileage Deviation from the historical average and The degree of elasticity, combined with the elastic coefficient and Calculate driving characteristic correction terms ; The benchmark cleaning time The vehicle model coefficient The driving characteristic correction item and the time-series coupled stain accumulation evaluation value And the differentiated stain index of the current order user Multiply by the product to obtain the basic cleaning time. .

[0057] In this embodiment, the calculation of the basic cleaning time incorporates five independent factors, each characterizing the source of cleaning needs from different dimensions. This multi-factor product design reflects the multiplicative coupling effect between the factors; that is, when multiple adverse factors occur simultaneously, the basic cleaning time is dramatically amplified, thus more realistically reflecting the actual cleaning needs in complex scenarios.

[0058] Baseline cleaning time This is the basic cleaning time for that vehicle type under standard conditions; for example, 20 minutes for sedans, 25 minutes for SUVs, and 30 minutes for MPVs. (Vehicle type coefficient) This is the first scaling up of the baseline duration; for example, 1.25 for SUVs, 1.50 for MPVs, and 1.35 for luxury models. The driving characteristic correction term is designed based on a key insight: even two vehicles of the same model with different driving times and mileages after completing a previous order will produce significantly different amounts of dirt. Specifically, the correction term is based on 1; when the driving time or mileage exceeds the historical average, the correction term is greater than 1, and vice versa. Elasticity coefficient. and Controlling the marginal impact of superaverage degree on cleaning time, for example This means that for every 10% increase in driving time beyond the average, cleaning time increases by 3.5%. This design allows the model to impose a reasonable penalty extension on abnormally long trips.

[0059] Time-coupled stain accumulation assessment value This reflects the accumulated stain load of the vehicle over its multi-order history; the higher the value, the longer the cleaning time required. The current order user's differentiated stain index. This reflects the type and depth of stains that the user of the vehicle is likely to produce; for example, the stain index for tourists (typically 1.4) is higher than that for commuters (typically 1.0). The multiplication of the five factors creates a multiplicative coupling effect between them, accurately reflecting the amplifying effect of multiple factors on cleaning time.

[0060] In an optional embodiment, determining the current order user profile information of the vehicle to be cleaned may include: Extract the user's order frequency, average order duration, average mileage, vehicle usage time distribution vector, and commonly used service point type vector within a preset time window to form a user feature vector; The user feature vectors are clustered using a clustering algorithm to classify users into various profile types, including short-distance commuting, long-distance travel, and high-frequency business users. A differentiated set of stain pattern parameters is predefined for each portrait type, the set of parameters including a dominant stain weight vector, a stain depth baseline value and a cleaning difficulty additional coefficient; Based on the user profile of the current order, dynamically determine their differentiated stain index. .

[0061] In this embodiment, the construction of the user behavior-stain pattern association profile adopts an unsupervised clustering method. Its core is to automatically discover user groups with similar stain generation patterns from the user's historical car rental behavior, thereby realizing the differentiated allocation of cleaning resources.

[0062] First, five features are extracted from the user's historical car rental records: order frequency (reflecting usage activity), average order duration, average mileage, usage time distribution (the percentage of orders during morning peak hours, daytime, evening peak hours, nighttime, and early morning), and distribution of frequently used rental locations (the percentage of orders at airports, high-speed rail stations, commercial areas, residential areas, and scenic spots). These five dimensions together constitute a feature vector that comprehensively depicts the user's car rental behavior habits. For example, a commuter might have frequent short-distance orders during morning and evening peak hours or at residential rental locations; a tourist might have infrequent long-distance orders during daytime hours or at scenic spots or airport rental locations.

[0063] Then the K-Means clustering algorithm is used ( Users are categorized into four typical profile types. Short-distance commuter users typically use their vehicles frequently (e.g., ≥15 trips / 90 days), for short durations (e.g., ≤40 minutes), and for short distances (e.g., ≤25 kilometers). Their usage is concentrated during morning and evening rush hours and in residential areas, primarily generating interior dust, footprints, and breakfast food crumbs. Cleaning is relatively easy, and the baseline stain depth is set to 1.0. Long-distance tourist users typically use their vehicles infrequently (e.g., <8 trips / 90 days), for long durations (e.g., ≥120 minutes), and for long distances (e.g., ≥80 kilometers). Their usage is concentrated during the day, in scenic areas, and at airports, primarily generating mud, food scraps, and odors. Cleaning is more difficult, and the baseline stain depth can be set to 1.4, with a cleaning difficulty coefficient of 1.35. Business users with high frequency of use (e.g., ≥12 orders / 90 days), medium duration (e.g., 60-90 minutes), and medium-to-long mileage (e.g., 40-60 kilometers) mainly use their vehicles during the day, in commercial areas, and at high-speed rail stations. The main stains they generate are coffee stains, fiber shedding, and smoke odor. The cleaning difficulty is moderate to high, and they require a high level of cleanliness in the interior. The base value for stain depth can be set to 1.2. Other types can be used as a fallback and can use the default parameter, i.e., a base value for stain depth of 1.1.

[0064] In this solution, each profile type has a predefined set of differentiated stain pattern parameters. Each time an order is generated, the system queries the corresponding stain pattern parameter set in real time based on the current user's profile type and outputs a differentiated stain index. This is used for calculating the basic cleaning time in step 106. User profiles are recalculated monthly to adapt to the evolution of user behavior patterns and ensure that the profiles always reflect the user's latest car usage habits.

[0065] In an optional embodiment, determining the current order user profile information of the vehicle to be cleaned may further include: When the number of orders placed by the user within a preset time window is less than the minimum number of samples required for clustering, the cold start profile inference process is executed: Calculate the user's basic attribute feature vector, which includes the user's age range, type of city of residence, historical average car rental price range, and payment method preference; From the user group that has been clustered, select multiple similar users whose cosine similarity to the basic attribute feature vector is greater than the similarity threshold; The inferred stain pattern parameter set of the user is obtained by taking a weighted average of the stain pattern parameter sets of the multiple similar users, wherein the weighting coefficient is positively correlated with the cosine similarity. Based on the inferred stain pattern parameter set, the user's differentiated stain index is dynamically determined. ; Meanwhile, the user's profile type is marked as a type to be learned, and after the user completes a preset number of orders, clustering is re-executed to update its profile type.

[0066] In this embodiment, for newly registered users or users with insufficient historical order counts (e.g., fewer than 3 orders within 90 days), their profile type cannot be directly obtained through clustering due to a lack of sufficient car rental behavior data. Therefore, this solution designs a cold-start profile inference mechanism, utilizing the user's static basic attributes as the basis for inference, enabling the system to provide relatively reasonable personalized cleaning estimates when new users first enter the platform. The core of this mechanism is that although static basic attributes (such as age, place of residence, consumption level, and payment habits) do not have a direct causal relationship with specific car usage behavior, they are statistically correlated with users' car rental habits. For example, young business people from first-tier cities may prefer a high-frequency, short-duration, efficient car rental model, while older users from tourist cities may prefer a low-frequency, long-duration leisure car rental model. By finding the "neighbor" user group most similar to the current user's basic attributes, their stain patterns can be used to infer the current user's staining behavior.

[0067] The specific inference process can be as follows: First, construct the user's basic attribute feature vector, including the user's age range (e.g., 18-25, 26-35, etc.), the type of city of residence (first-tier city, second-tier city, tourist city, etc.), the historical average car rental price range (reflecting consumption level), and payment method preference (e.g., credit card, third-party payment, etc.). Then, from the user group that has completed clustering and has sufficient historical orders, select multiple similar users whose cosine similarity to the current user's basic attribute feature vector exceeds a preset threshold. The stain pattern parameter set of these similar users (including the dominant stain weight vector, stain depth baseline value, and cleaning difficulty additional coefficient) is extracted and weighted by a weighted average according to the similarity, with the weighting coefficient positively correlated with the cosine similarity. In this way, the stain pattern parameter set of the new user is inferred through "nearest neighbor voting," and then its differentiated stain index is output. At the same time, the user's profile type is marked as "to be learned," meaning that after the user completes a preset number of orders (e.g., 3 or 5 orders), the system will include its real order data in the training set and re-perform clustering to update its profile type. This design, which combines cold start with continuous learning, allows the system to provide relatively reasonable personalized cleaning estimates when new users first enter the platform. As data accumulates, it continuously self-corrects, gradually transitioning smoothly from an "inference mode" to a "clustering mode".

[0068] In an optional embodiment, the step of constructing a system based on the real-time status information of the store is... The nonlinear load correction model for store queuing in queuing theory can be used to calculate queuing correction factors, which may include: Vehicle arrival rate estimated based on real-time store data. Service rate per workstation and number of effective service counters ; Based on the number of workstations currently in operation With the number of effective service counters Calculate workstation occupancy rate ; Queuing correction factor is calculated using piecewise functions. : when hour, ; when hour, ; Among them, symbols Represents the critical occupancy rate threshold, symbol Represents the linear queuing correction factor, symbol Represents the saturated nonlinear amplification factor, symbol This indicates the number of vehicles in the queue.

[0069] In this embodiment, the store cleaning operation system is modeled as a A queuing system is in which vehicle arrivals follow a Poisson distribution, and cleaning service times follow an exponential distribution. There are a total of [number missing] queuing systems. This model framework, which includes a service counter (cleaning station), is suitable for describing multi-service counter systems with random arrivals and random service times.

[0070] Vehicle arrival rate The service rate per workstation is estimated in real time by counting the number of vehicles that entered the cleaning queue in the past hour. Estimated by analyzing the average number of vehicles completed per hour at a single workstation over the past two hours, to smooth out fluctuations in individual efficiency. Number of effective service counters. The number of workers on duty needs to be adjusted in real time; if any worker leaves their post for rest, the effective number of service stations should be reduced proportionally. Workstation occupancy rate. Defined as the ratio of the number of workstations actually in operation to the number of effective service stations, its value ranges from 0 to 1.

[0071] Queuing correction factor The calculation employs a piecewise function, designed to characterize the nonlinear, abrupt changes in congestion. When Less than the critical occupancy threshold When the value is 0.75 (e.g., 0.75), the system is in a light load zone, and the queuing time is related to the number of vehicles in the queue. The relationship is linear, and the correction factor is ,in This is a linear queuing correction factor (e.g., 0.08). In this phase, for every additional vehicle added to the queue, the waiting time increases by approximately 8%.

[0072] when Reaching or exceeding When the system enters the saturation region, an exponential saturation term is added to the queuing correction factor. The denominator of the saturation term is ,when As the value approaches 1, the denominator approaches 0, and the saturation term approaches infinity, thus realistically depicting the theoretical phenomenon of infinitely increasing queue length when the workstation utilization rate approaches saturation. For example, when... When the value increases from 0.75 to 0.9, the saturation term rises sharply from 0 to... ,like If this occurs, the value reaches 0.225, and the queuing correction factor increases from 1.16 to approximately 1.385. Compared to traditional linear correction models, this nonlinear piecewise function more accurately reflects the actual experience in store operations where "even slight congestion can lead to severe delays," thus effectively avoiding a significant underestimation of waiting time under high load. In actual systems, when... When the value reaches 0.95 or higher, the system will trigger the upper limit protection and send a "workstation saturation warning". It is recommended to activate backup cleaning resources or guide order diversion. Figure 1 The differences between the nonlinear queuing correction model of this invention and the existing linear model were further compared. The horizontal axis in the figure represents the workstation occupancy rate. The vertical axis represents the queuing correction factor. The dashed lines represent the calculation results of the existing linear model, and the solid lines represent the calculation results of the nonlinear model of this invention. From... Figure 1 It can be seen that when When the occupancy rate is below the critical threshold, the two curves basically overlap, and the queuing correction factor increases accordingly. The number of vehicles in queues increases linearly. When After exceeding the critical threshold, the curve of the model of this invention enters the nonlinear amplification region marked in the figure, where the queuing correction factor rises sharply, while the linear model maintains a gradual increase. For example, in the nonlinear amplification region, the queuing correction factor of the model of this invention is significantly higher than that of the linear model. This accurately depicts the abrupt change in the store from normal operation to congestion paralysis. The figure intuitively verifies the saturation term in the piecewise function of this scheme. The exponential amplification effect enables maintenance personnel to clearly identify risk inflection points under high load.

[0073] In an optional embodiment, the step of constructing a system based on the real-time status information of the store is... The nonlinear load correction model for store queuing in queuing theory can also include the following for calculating the queuing correction factor: Obtain the skill level vector of workers at each cleaning station in the store. And the cleaning difficulty coefficient vector for each vehicle in the current queue of vehicles to be cleaned. ; Construct a skill-difficulty matching matrix , where matrix elements Indicates the first The worker handles the first Factors that improve the basic cleaning efficiency of vehicles. , where the symbol Indicates the matching gain coefficient; The matching matrix is ​​solved using the Hungarian algorithm. The optimal workstation-vehicle pairing set is obtained by using the maximum overall efficiency allocation scheme. ; Based on the optimal workstation-vehicle pairing set Calculate the effective comprehensive service rate ; Using the aforementioned effective comprehensive service rate Replace the single workstation service rate Recalculate the workstation occupancy rate. And based on the workstation occupancy rate Recalculate the queuing correction factor .

[0074] In this embodiment, considering the traditional The queuing model assumes all service counters are homogeneous, meaning each station has the same service rate. However, in actual store operations, different cleaning workers have varying skill levels (e.g., a skilled worker might clean 4 cars per hour, while a novice can only clean 2), and different vehicles have varying cleaning difficulties (e.g., MPVs require more cleaning time than sedans). Matching highly skilled workers with vehicles requiring higher difficulty can improve overall cleaning efficiency. Conversely, inappropriate matching can reduce efficiency. Therefore, this solution introduces a skill-difficulty optimal matching mechanism to refine the queuing model.

[0075] First, obtain the skill level vector of workers at each workstation. The skill level can be an integer from 1 to 5 (5 represents the highest skill level), and the cleaning difficulty coefficient vector is the value of each vehicle in the queue to be cleaned. The cleaning difficulty coefficient can be estimated comprehensively based on information such as vehicle model, time-coupled stain accumulation assessment value, and vehicle model coefficient. For example, for SUVs... Higher-rise vehicles are more difficult to clean.

[0076] Then construct a skill-difficulty matching matrix. , where matrix elements The meaning of this formula is: when the worker's skill level... The higher the difficulty level, the greater the vehicle's difficulty. The lower the value, the higher the efficiency improvement factor. The larger the size, the higher the cleaning efficiency. This applies when workers have low skill levels and vehicles are more complex. It may be less than 1 (i.e., efficiency decreases). Matching gain coefficient Controlling the degree to which skill-difficulty differences affect efficiency, for example hour, (Skill 5 vs. Difficulty 2.5) will result in a 40% efficiency increase.

[0077] Next, the Hungarian algorithm can be used to solve for the maximum overall efficiency allocation scheme under this matching matrix, thus obtaining the optimal workstation-vehicle pairing set. The Hungarian algorithm is a classic combinatorial optimization algorithm that finds a perfect match that maximizes overall efficiency in polynomial time (i.e., assigning at most one vehicle to each workstation and one workstation to each vehicle). Using this algorithm ensures that, given a distribution of skills and difficulty, the allocation method that maximizes overall cleaning efficiency can be found.

[0078] Calculate the effective comprehensive service rate based on the optimal pairing results. That is, among all pairs The average value after multiplying by the corresponding efficiency improvement factor. This reflects the equivalent service rate of the store cleaning system under the optimal matching strategy. Finally, using... Replace the service rate of a single workstation in the original queuing model The workstation occupancy rate and queuing correction factor are recalculated. This correction ensures that the output of the queuing model reflects the actual queuing performance under optimal matching of intelligent scheduling, thereby avoiding queuing time estimation bias caused by simple assumptions of homogeneous workstations.

[0079] In an optional embodiment, the step of constructing an environment-stain curing coupling correction model to calculate the environment correction coefficient based on the dominant stain type corresponding to the environmental parameter information and the current order user profile information may include: The dominant stain type is determined based on the current order user profile information. The dominant stain type may include organic stains, inorganic stains, grease stains, or a mixture of at least two types of stains. Get real-time ambient temperature relative humidity and seasonal factors ; When the dominant stain type is a single type, select the corresponding correction function: For organic stains , where the symbol Represents the reference temperature, symbol Represents the reference relative humidity, symbol and Indicates the environmental sensitivity coefficient of organic stains; For inorganic stains , where the symbol Indicates the reference value for freezing point temperature, symbol and Indicates the environmental sensitivity coefficient of inorganic stains; For grease stains , where the symbol Indicates the humidity reference scaling factor, symbol Indicates the environmental sensitivity coefficient of grease stains; When the dominant stain type is a mixed type, the correction coefficient for each individual stain type is calculated separately, and then a weighted sum is performed according to the predefined dominant stain weight vector in the user profile information to obtain the comprehensive environmental correction coefficient. , where the symbol Indicates the stain type index, symbol Indicates the corresponding weight, symbol This represents the corresponding correction factor.

[0080] In this embodiment, different types of stains exhibit significantly different sensitivities to environmental conditions. Therefore, different correction functions need to be selected based on the dominant stain type. This differentiated design ensures that environmental correction accurately matches the physicochemical properties of the actual stain. For example, organic stains (such as food residue, beverage stains, and coffee stains) ferment, solidify, and breed bacteria more rapidly in high-temperature and high-humidity environments, thus increasing the difficulty of cleaning. Therefore, the correction function for organic stains can employ... In the form of, and The baseline conditions are (e.g., 25°C and 50% humidity). When the temperature or humidity exceeds the baseline value, the correction factor is greater than 1 and increases linearly with the degree of deviation. Sensitivity coefficient. and Determined through training, for example This indicates that for every 1°C increase in temperature, the organic curing rate increases by 1.2%.

[0081] Inorganic stains (such as mud, salt, and dust) become more difficult to remove at low temperatures due to ice crystal formation or increased adhesion. Therefore, a freezing point reference value is incorporated into the correction function. (For example, 0°C). When the temperature is below the freezing point, If the value is positive, the correction factor increases. When the temperature is above freezing, this term becomes negative, and the correction factor decreases. Inorganic stains are relatively less sensitive to humidity, but the humidity term is still retained to account for the difficulty of stains drying in high-humidity environments.

[0082] The curing and oxidation processes of grease stains are influenced by the interaction of temperature and humidity. High temperature or high humidity alone has a limited effect on promoting grease curing, but when high temperature and high humidity occur simultaneously, the oxidation rate of grease is significantly accelerated. Therefore, the correction function includes cross terms. When both temperature and humidity deviate from the baseline value, the cross term becomes positive and the product effect is amplified. (Denominator) It serves as a humidity reference scaling factor, used to adjust the sensitivity of interactive items.

[0083] Seasonal factors As a multiplicative correction term, it is used to capture the macro-environmental differences across seasons. For example, additional correction is needed in winter due to salt deposits caused by snow-melting agents, while in summer, the correction for organic stains may be amplified due to high temperature and humidity. When the dominant stain is a mixed type (e.g., long-distance travelers generate both organic and inorganic stains), the comprehensive environmental correction coefficients are weighted and summed according to the weight vector. For example, when the organic weight is 0.6 and the inorganic weight is 0.4, This differentiated, multi-dimensional environmental correction mechanism allows cleaning duration predictions to accurately reflect the actual impact of environmental conditions on the stain curing process.

[0084] In an optional embodiment, the step of calculating a delivery readiness index based on the total cleaning preparation time and the user's scheduled vehicle pickup time, and the step of outputting an order decision based on the delivery readiness index and an adaptive soft threshold, may include: Calculate the scheduled car pick-up time Total cleaning preparation time and safety margin time The difference is divided by the safety margin time. The basic readiness index was obtained. ; pass The function is applied to the basic readiness index. Perform normalization mapping to obtain the delivery readiness index. , where the symbol express Curvature coefficient; Get the current workstation occupancy rate of the store ; According to the workstation occupancy rate Dynamically adjust the confirmation threshold Delay threshold and rejection threshold When the workstation occupancy rate When the value is lower than the first preset value, the confirmation threshold, the delay threshold, and the rejection threshold are reduced; when the workstation occupancy rate is... When the value exceeds the second preset value, the confirmation threshold, the delay threshold, and the rejection threshold are increased. The delivery readiness index With the adjusted confirmation threshold The delay threshold The rejection threshold Comparison: If the delivery readiness index Output the order confirmation decision; like Output delay monitoring decisions; like It outputs a decision to reject the order and recommend an alternative vehicle.

[0085] In this embodiment, the delivery readiness index is calculated using... The normalization method maps time margins to probability values ​​between 0 and 1. Basic readiness index. This reflects the degree of leeway between the scheduled pick-up time and the estimated completion time. When When the value is positive and the larger it is, the more ample the time margin and the higher the delivery readiness index. Approaching 1; when When it is negative, Approaching 0; when When approaching 0, The function at the curvature coefficient It provides a smooth yet steep transition under control. The advantage of this normalization is that it can eliminate the influence of different stores and different time periods. The comparability issues caused by differences in settings make the delivery readiness index a standardized indicator that can be compared across stores.

[0086] In the decision-making output stage, this scheme presets three basic thresholds (e.g. , , ), and based on the real-time workstation occupancy rate of the store. Make dynamic adjustments. When When the threshold is below the first preset value (e.g., 0.6), the store is in a light-load state, and the system lowers all three thresholds (e.g., ...). (Lowered by 0.05 to 0.75), making orders easier to confirm and thus making full use of idle cleaning resources. When When the threshold is higher than the second preset value (e.g., 0.8), the store is under heavy load, and the system raises all three thresholds (e.g., ...). The threshold was increased by 0.05 to 0.85, allowing for more stringent order screening and preventing orders from being fulfilled. This "dynamic threshold offset" strategy enables the system's decision boundaries to adaptively follow changes in store load, achieving intelligent scheduling that actively accepts orders when resources are plentiful and makes promises cautiously when resources are scarce.

[0087] Meanwhile, in this embodiment, the three-branch decision-making (confirmation / delay monitoring / rejection) provides the system with refined order processing capabilities. Confirmation decisions directly commit to the user; delay monitoring decisions enter a dynamic tracking state (recalculating the readiness index every 5 minutes; if it rises back to the confirmation threshold, it automatically confirms; if it falls to the rejection threshold, it automatically rejects and recommends an alternative vehicle); and rejection decisions automatically search for other deliverable vehicles in the store, sorting them by vehicle type similarity, price, and user preference before pushing alternative recommendations to the user. This hierarchical decision-making mechanism maximizes user experience while avoiding fulfillment failures due to over-commitment.

[0088] To clearly illustrate the normalization process of the delivery readiness index and the adaptive soft threshold decision logic in this embodiment, the following will combine... Figure 2 Provide a visual explanation. Figure 2 In the middle, the horizontal axis represents the basic readiness index defined in this scheme. (Calculated from time margin and safety margin, with a value range of -3 to 3), the vertical axis represents the normalized delivery readiness index defined in this scheme. (Value range 0~1); The blue curve is the Sigmoid normalized curve described in this scheme, corresponding to the formula... , where parameters and Figure 2 The annotations are consistent. This curve embodies the design principle of "smooth but steep transition," meaning that when... hour, ;when hour, ;exist The curve changes most steeply in the region close to 0, corresponding to the critical region (decision-sensitive zone) marked in gray in the figure, demonstrating the effect of "avoiding linear threshold decision jitter." Meanwhile, Figure 2 The three horizontal dashed lines in the middle correspond to the adaptive soft threshold defined in this scheme, i.e., the green dashed lines are as follows: To confirm the order threshold, the orange dashed line... The red dashed line represents the delayed confirmation threshold. To reject orders, the four decision intervals divided by the three lines correspond perfectly to the three-branch decision logic, i.e. Confirm the order in time (corresponding to "Output Order Confirmation Decision"); Time delay monitoring (corresponding to "output delay monitoring decision"); Continuous monitoring (implied risk warning range); Reject the order if necessary (corresponding to "output order rejection decision").

[0089] In an optional embodiment, the step of outputting an order decision based on the delivery readiness index and an adaptive soft threshold may further include: When multiple orders awaiting decision exist simultaneously, a batch decision conflict resolution process is executed: According to the delivery readiness index of each order Sort the orders from highest to lowest to generate a candidate order sequence; Initialize the remaining cleaning resource vector ,in For the first The remaining available cleaning time for each workstation within the user's scheduled vehicle pickup time window; Iterate through each order in the candidate order sequence: Calculate the cleaning resource requirements for this order. ,in For order priority indicator factor, symbol This indicates the additional resource consumption coefficient for priority. From the remaining clean resource vector Search for whether a workstation exists. This allows for the remaining available cleaning time at the workstation. ; If it exists, then assign the workstation to the order and update the remaining cleaning resources of the workstation to [the required information]. Simultaneously, it outputs the order confirmation decision; If it does not exist, mark the order as a resource conflict order and enter the conflict resolution queue; For orders in the conflict resolution queue, perform a rollback negotiation: reduce their resource reservation amount according to order priority. Then, re-match the resources; if a match still cannot be found, output a delay monitoring decision or an order rejection decision, where the symbol... This represents the resource compression factor.

[0090] In this embodiment, when the system receives multiple order requests within a short period and the appointment pick-up time windows of these orders overlap, simply making decisions sequentially based on the readiness index may lead to resource allocation conflicts. That is, multiple orders attempt to compete for the cleaning capacity of the same workstation within the same time period. For example, if two orders both have an appointment pick-up time of 12:00 and both are expected to require approximately 60 minutes of cleaning preparation time, they will compete for the same workstation within the 10:00-11:00 time window. If decisions are made independently, both orders may be confirmed, resulting in insufficient resources. Therefore, this solution designs a batch decision-making conflict resolution mechanism.

[0091] First, candidate orders are sorted from highest to lowest according to their delivery readiness index, with the orders having the highest readiness index being processed first. This sorting principle ensures that, when resources are scarce, the orders most likely to be completed on time are confirmed first.

[0092] Then, the system initializes a vector of remaining clean resources. This vector records the time window for each cleaning station within the user's scheduled vehicle pickup time (e.g., a time window before the scheduled pickup time). The remaining available cleaning time for the interval. The initial value can be calculated based on the workstation's historical schedule, current work status, and estimated completion time.

[0093] For each pending decision order, calculate its required cleaning resource usage. This introduces a priority indicator factor. and priority additional resource consumption coefficient . It can be 0 or 1 (or a more granular level), when the order is of high priority (such as VIP users, urgent needs). Otherwise, it is 0. Control the resource reservation buffer ratio for high-priority orders, for example... This means that high-priority orders need to reserve an additional 20% of cleaning resources to ensure on-time delivery even in the event of slight delays. This design can provide resource allocation for high-value orders.

[0094] Next, the system will use the remaining clean resource vector Search for whether there exists a workstation whose remaining available time is greater than or equal to 1. If it exists, then assign that workstation to the order (usually selecting the first workstation that meets the criteria, or the one closest to the remaining resources). (To maintain resource balance) the corresponding resource amount is deducted from the workstations, and a confirmation decision is output. If no such decision exists, it means that the remaining resources of all workstations are insufficient to meet the cleaning requirements of this order, and the order is marked as a resource conflict order and enters the conflict resolution queue.

[0095] In the conflict resolution queue, the system performs rollback negotiation, specifically including: reducing the resource reservation amount according to the order priority, i.e. ,in This represents the resource compression factor (e.g., 0.1~0.3). Reducing the resource reservation means the system commits to the order with higher risk (potentially causing delays due to actual cleaning time exceeding the reservation), but it avoids outright order rejection. If no workstation can be matched after reducing the resource reservation, a delay monitoring or rejection decision will be output.

[0096] This batch conflict resolution mechanism ensures that clean resources are allocated reasonably in high-concurrency order scenarios, while providing resource allocation for high-priority orders and retaining orders as much as possible through negotiation when resources are insufficient, thereby maximizing system throughput and user satisfaction.

[0097] In an optional embodiment, the method for predicting the car rental delivery cleaning preparation time may further include a model iterative optimization step, specifically including: Extract daily cleaning records and construct a model containing input features and predicted outputs. Compared with the actual cleaning completion time The training samples; The regularized mean squared error is used as the loss function. ; Among them, symbols Indicates the number of samples, symbol Indicates the sample index, symbol Represents the regularization coefficient, symbol Indicates the model weight parameters; The learnable parameters in the model are updated using a gradient descent optimization algorithm, and the learnable parameters include the time-series decay coefficient. The queuing correction coefficient and The environmental sensitivity coefficient to The above Curvature coefficient The basic cleaning time and the elastic coefficient , ; The updated model was deployed online through a canary release mechanism, and A / B testing was conducted based on metrics such as fulfillment success rate, user complaint rate, and workstation utilization rate. The real-time status information of the store includes at least the number of vehicles in the queue. The workstation occupancy rate and the number of employees on duty The environmental parameter information includes at least the ambient temperature. The relative humidity and the aforementioned seasonal factors When the number of historical orders in the historical order sequence information exceeds a preset threshold, the most recently completed preset number of orders will be used.

[0098] In this embodiment, to ensure the model can continuously adapt to seasonal changes, customer flow fluctuations, and operational strategy adjustments, this solution introduces an iterative optimization closed loop of daily offline batch training and canary release. This closed loop allows the model to automatically learn and optimize from real daily operational data without the need for manual parameter recalibration. For example, every day at midnight (e.g., 2:00 AM), the system can extract all cleaning order records from the previous day from the platform database. Each sample can include the model input feature vector (18 dimensions, covering driving characteristics, vehicle attributes, store status, user profile, environmental parameters, etc.) and the predicted cleaning duration output by the model under the current parameters. And the actual cleaning completion time recorded by the geomagnetic sensor at the workstation. When constructing training samples, outlier cleaning (removing samples exceeding 3 standard deviations) and missing value imputation (filling in with the median of similar orders) can be performed to ensure the quality of the training data.

[0099] The loss function can be a band loss function. The regularized mean squared error. The first term... It is the mean squared error between the predicted and actual values, measuring the accuracy of the model's predictions. (Second term) It is the sum of squares of the weight parameters multiplied by the regularization coefficient. Regularization is used to prevent overfitting. When there are too many model parameters or insufficient training samples, the regularization term penalizes excessively large parameter values, forcing the model to remain concise. The set of learnable parameters can include time-decay coefficients. Queuing correction factor and Environmental sensitivity coefficient to , Curvature coefficient Basic cleaning time and elasticity coefficient , These parameters can be updated using gradient descent via the Adam optimizer. The Adam optimizer adaptively adjusts the learning rate, combining momentum and second-moment estimation to accelerate convergence and avoid getting trapped in local optima. After each training epoch, model performance can be evaluated on a validation set (20% of the data from the most recent 3 days). If the mean absolute percentage error (MAPE) on the validation set does not decrease for 5 consecutive epochs, the learning rate is decayed by 50%. Training stops when the MAPE on the validation set falls below 5% or the maximum number of epochs (50) is reached. After training, the new model can be deployed via a canary release mechanism, where the new model is made available to 10% of order traffic on the first day, while the old model continues to be used for 90% of traffic. The fulfillment success rate, user complaint rate, and workstation utilization rate of the two sets of orders are compared. If the new model outperforms the old model by more than 3 percentage points in all three metrics, a full switch is made; otherwise, the model is rolled back and the reasons are analyzed. This A / B testing-driven model update strategy ensures that each model update is empirically validated and will not cause a decline in online performance due to overfitting or data bias during training. Meanwhile, this solution allows for real-time monitoring of model prediction performance metrics: single-request computation latency (target <100ms), prediction MAPE (target <5%), and extreme scenario coverage (ensuring that the proportion of training samples is not less than 15% under extreme scenarios such as high load and severe weather). When the MAPE exceeds 8% for three consecutive days, a model degradation alarm is triggered, initiating an emergency retraining process. This daily iterative, continuous monitoring optimization loop enables the system to automatically adapt to seasonal changes, passenger flow fluctuations, and operational strategy adjustments. The model can enter a stable convergence state in the third month after deployment, and the prediction accuracy continues to improve over time.

[0100] To visually demonstrate the iterative optimization process of the model weight parameters in this embodiment, please refer to [link / reference]. Figure 3 . Figure 3 A convergence curve was plotted showing the relationship between the number of training epochs and the model loss value. From... Figure 3 It can be seen that the training loss (MSE) and validation loss (MSE) decrease rapidly in the first 10 rounds, indicating that the gradient descent algorithm can quickly capture the main patterns in the data. From rounds 10 to 30, the rate of loss decrease gradually slows down, and the model enters the fine-tuning stage. Figure 3 The target threshold (MAPE < 5%) marked in the code is used as the performance boundary. When the mean absolute percentage error of the validation set first falls below this threshold, the system records the current round. Meanwhile, Figure 3 It also identifies early stopping points (e.g., Epoch 35, MAPE=4.8%), where the validation loss has stabilized at a low level and no longer decreases significantly over multiple epochs, triggering the early stopping mechanism to stop training and prevent overfitting. Figure 3The training loss and validation loss decrease synchronously with a small difference, indicating that the model has good generalization ability. This figure fully presents the entire gradient descent optimization process from the initial state with high loss to the convergent stable state, verifying the effectiveness of using the Adam optimizer and the early stopping strategy.

[0101] It should be understood that in the methods described in one or more embodiments of this specification, the order of some steps may be adjusted according to actual needs, or some steps may be omitted.

[0102] Based on the foregoing technical solutions, the present invention also provides a device for predicting the cleaning preparation time for car rental delivery, such as... Figure 4 As shown, the device, from a macroscopic perspective, may include the following modules: The information determination module 402 is used to determine the current order information, historical order sequence information, real-time store status information, current order user profile information, and environmental parameter information of the vehicle to be cleaned; The evaluation value construction module 404 is used to construct a coupling model of time-series decay and stain accumulation between orders based on the historical order sequence information to calculate the time-series coupled stain accumulation evaluation value; The basic cleaning time calculation module 406 is used to calculate the basic cleaning time based on the current order information, the current order user profile information, and the time-coupled stain accumulation assessment value; Queuing correction factor calculation module 408 is used to calculate queuing correction factor based on the real-time status information of the store and to construct a nonlinear load correction model for store queuing based on M / M / c queuing theory. The environmental correction coefficient calculation module 410 is used to construct an environment-stain curing coupling correction model based on the dominant stain type corresponding to the environmental parameter information and the current order user profile information to calculate the environmental correction coefficient; The total cleaning preparation time synthesis module 412 is used to synthesize the total cleaning preparation time based on the basic cleaning time, the queuing correction factor and the environmental correction coefficient. The delivery readiness index calculation module 414 is used to calculate the delivery readiness index based on the total cleaning preparation time and the user's scheduled vehicle pick-up time. The order decision output module 416 is used to output an order decision based on the delivery readiness index and the adaptive soft threshold.

[0103] Those skilled in the art will understand that the modules in the apparatus of the foregoing embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules, that is, the module division can be flexibly performed to implement the method embodiments described above.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of predicting a clean preparation duration for a car rental delivery, characterized in that, include: Determine the current order information, historical order sequence information, real-time store status information, current order user profile information, and environmental parameter information of the vehicles to be cleaned; Based on the historical order sequence information, a coupled model of time-series decay and stain accumulation between orders is constructed to calculate the time-series coupled stain accumulation evaluation value. Based on the current order information, the current order user profile information, and the time-coupled stain accumulation assessment value, calculate the basic cleaning duration; Based on the real-time status information of the stores, a nonlinear load correction model for store queuing based on the M / M / c queuing theory is constructed to calculate the queuing correction factor. Based on the dominant stain type corresponding to the environmental parameter information and the current order user profile information, an environment-stain curing coupling correction model is constructed to calculate the environmental correction coefficient; The total cleaning preparation time is calculated based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient. Calculate the delivery readiness index based on the total cleaning and preparation time and the user's scheduled vehicle pick-up time; Based on the delivery readiness index and the adaptive soft threshold, an order decision is output.

2. The method of claim 1, wherein, The total cleaning preparation time, calculated based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient, includes: The total cleaning preparation time is obtained by multiplying the basic cleaning time, the queuing correction factor, and the environmental correction coefficient, and adding the fixed vehicle inspection time and the fixed vehicle scheduling time. The fixed duration for vehicle inspection and the fixed duration for vehicle dispatch are predefined based on vehicle type or store layout.

3. The method of claim 1, wherein the method further comprises: The step of constructing a coupled model of time-series decay and stain accumulation between orders based on the historical order sequence information to calculate the time-series coupled stain accumulation assessment value includes: Extract the driving time, mileage, user stain index, and order end time of each historical order from the historical order sequence information; According to the driving time length and the driving mileage of each historical order, the single-order stain contribution weight is calculated ; Among them, symbols Indicates the first The travel time of each historical order, symbol This indicates the maximum driving time of this vehicle model in historical data, symbol... Indicates the first Mileage of a historical order, symbol This indicates the maximum mileage of this vehicle model in historical data, symbol... Indicates the driving time weighting coefficient, symbol This represents the mileage weighting coefficient; According to the current time With the end time of each historical order time interval Calculate the time-series decay factor , where the symbol Indicates the timing decay coefficient; The individual order stain contribution weights for each historical order. User Stain Index Vehicle model coefficient and the aforementioned time-series decay factor Sum the products and add them together with the baseline stain level. The sums are used to obtain the time-coupled stain accumulation assessment value. .

4. The method of claim 1, wherein, The calculation of the basic cleaning duration based on the current order information, the current order user profile information, and the time-coupled stain accumulation assessment value includes: Obtain baseline cleaning time Vehicle model coefficient The driving time of the previous order Mileage and its historical average and And the differentiated stain index of current order users. ; Based on the travel time of the previous order With the mileage Deviation from the historical average and The degree of elasticity, combined with the elastic coefficient and Calculate driving characteristic correction terms ; The benchmark cleaning time The vehicle model coefficient The driving characteristic correction item and the time-series coupled stain accumulation evaluation value And the differentiated stain index of the current order user Multiply by the product to obtain the basic cleaning time. : 。 5. The method of claim 1, wherein, The user profile information for determining the current order of the vehicle to be cleaned includes: extract the order frequency, average order duration, average driving distance, and vehicle use time period distribution vector of the user within a preset time window and the common point type vector , to form a user feature vector ; The user feature vector is processed using a clustering algorithm. Clustering is performed to categorize users into various profile types, including short-distance commuters, long-distance travelers, and high-frequency business travelers. For each image type, a differentiated set of stain pattern parameters is predefined, the set of parameters including the dominant stain weight vector. Stain depth baseline value and additional cleaning difficulty coefficient; According to the current order user's portrait type, the difference of the stain index is dynamically determined .

6. The method of claim 1, wherein the method further comprises: The step of constructing a nonlinear load correction model for store queuing based on the real-time status information of the store and calculating the queuing correction factor includes: Vehicle arrival rate estimated based on real-time store data. Service rate per workstation and number of effective service counters ; According to the number of stations currently in operation With the effective number of service stations , the station occupancy rate is calculated ; Using piecewise functions to calculate queue correction factors : wherein the symbol represents a critical occupancy threshold, the symbol represents a linear queuing correction factor, the symbol represents a saturation non-linear amplification factor, and the symbol represents the number of queued vehicles.

7. The method of claim 1, wherein, The process of constructing an environment-stain curing coupling correction model based on the dominant stain type corresponding to the environmental parameter information and the current order user profile information to calculate the environmental correction coefficient specifically includes: The dominant stain type is determined based on the current order user profile information. The dominant stain type includes organic stains, inorganic stains, grease stains, or a mixture of at least two types of stains. Acquiring real-time ambient temperature , relative humidity and seasonal factor ; When the dominant stain type is a single type, the corresponding correction function is selected; wherein, for organic stains, the environmental correction factor for inorganic stains is calculated according to the following formula. : Among them, symbols Represents the reference temperature, symbol Represents the reference relative humidity, symbol and Indicates the environmental sensitivity coefficient of organic stains; For inorganic stains, the environmental correction factor for inorganic stains is calculated using the following formula. : Among them, symbols Indicates the reference value for freezing point temperature, symbol and Indicates the environmental sensitivity coefficient of inorganic stains; For grease stains, calculate the environmental correction factor for grease stains using the following formula. : Among them, symbols Indicates the humidity reference scaling factor, symbol Indicates the environmental sensitivity coefficient of grease stains; When the dominant stain type is a mixed type, the correction coefficient for each individual stain type is calculated separately, and then the dominant stain weight vector is predefined in the user profile information. We perform a weighted summation to obtain the comprehensive environmental correction coefficient. ; wherein the symbol represents a stain type index, the symbol represents a corresponding weight, and the symbol represents a corresponding correction factor.

8. The method of claim 1, wherein, The step of calculating the delivery readiness index based on the total cleaning preparation time and the user's scheduled vehicle pickup time, and the step of outputting an order decision based on the delivery readiness index and an adaptive soft threshold, includes: Calculating the reservation pickup time the difference between the total cleaning preparation time and the safety margin time , dividing the difference by the safety margin time to obtain a base preparation index ; normalizing mapping the base readiness index by a Sigmoid function to obtain a delivery readiness index wherein the symbol represents a Sigmoid curvature coefficient; Obtaining current station occupancy of a store ; According to the workstation occupancy rate Dynamically adjust the confirmation threshold Delay threshold and rejection threshold This includes: when the workstation occupancy rate When the value is lower than the first preset value, the confirmation threshold is reduced. The delay threshold and the rejection threshold When the workstation occupancy rate When the value is higher than the second preset value, the confirmation threshold is increased. The delay threshold and the rejection threshold ; The delivery readiness index With the adjusted confirmation threshold The delay threshold The rejection threshold Comparison: if the delivery readiness index , output a confirmed order decision; If , output the delay monitoring decision; If , the decision to output a rejected order and recommend an alternative vehicle.

9. The method of claim 1, wherein, It also includes model iterative optimization steps, specifically including: Daily extract cleaned order records, build training samples containing model input features, model prediction outputs vs. actual cleaning completion duration vs. actual cleaning completion duration Adopting the mean square error with regularization as the loss function ; Among them, symbols Indicates the number of samples, symbol Indicates the sample index, symbol Represents the regularization coefficient, symbol Indicates the model weight parameters; The learnable parameters in the model are updated using a gradient descent optimization algorithm, and the learnable parameters include the time-series decay coefficient. The queuing correction coefficient and The environmental sensitivity coefficient to The Sigmoid curvature coefficient The basic cleaning time and the elastic coefficient , ; The updated model was deployed online through a canary release mechanism, and A / B testing was conducted based on metrics such as fulfillment success rate, user complaint rate, and workstation utilization rate. The real-time status information of the store includes at least the number of vehicles in the queue. The workstation occupancy rate and the number of workers on duty The environmental parameter information includes at least the ambient temperature. The relative humidity and the aforementioned seasonal factors When the number of historical orders in the historical order sequence information exceeds a preset threshold, the most recently completed preset number of orders will be used.

10. An apparatus for predicting a cleaning preparation duration of a car delivery, characterized by, include: The information determination module is used to determine the current order information, historical order sequence information, real-time store status information, current order user profile information, and environmental parameter information of the vehicle to be cleaned; The evaluation value construction module is used to construct a coupling model of time-series decay and stain accumulation between orders based on the historical order sequence information to calculate the time-series coupled stain accumulation evaluation value; The basic cleaning time calculation module is used to calculate the basic cleaning time based on the current order information, the current order user profile information, and the time-coupled stain accumulation assessment value. The queuing correction factor calculation module is used to construct a nonlinear load correction model for store queuing based on the real-time status information of the store and calculate the queuing correction factor. The environmental correction coefficient calculation module is used to construct an environment-stain curing coupling correction model based on the dominant stain type corresponding to the environmental parameter information and the current order user profile information to calculate the environmental correction coefficient; The total cleaning preparation time synthesis module is used to synthesize the total cleaning preparation time based on the basic cleaning time, the queuing correction factor, and the environmental correction coefficient. The delivery readiness index calculation module is used to calculate the delivery readiness index based on the total cleaning preparation time and the user's scheduled vehicle pick-up time. The order decision output module is used to output order decisions based on the delivery readiness index and the adaptive soft threshold.

Images