Vehicle unsailing determination method and device and vehicle unsailing model training method and device

Abstract

The embodiment of the invention relates to the technical field of vehicles, and discloses a vehicle unsalable determination method and device and a vehicle unsalable model training method and device.The vehicle unsalable determination method comprises the steps that based on target features at the current moment, the estimated sales duration of a target type vehicle and the estimated stock at the future moment are determined; the pre-estimated sales duration is used for representing the pre-estimated duration from stock entry to final sales; the target features are used for representing features having influences on the unsalable condition of the target type vehicle; based on the estimated sales duration, determining the maximum allowable retention duration of the target type of vehicle in the warehouse; determining the maximum allowable stock of the vehicle of the target type at the current moment based on the estimated stock at the future moment; and based on the maximum allowable retention duration and the maximum allowable inventory, determining whether the target type vehicle is unsalable at the current moment. According to the technical scheme, whether the vehicle is unsalable or not can be accurately identified through the target features.

Description

Technical Field

[0001] This invention relates to the field of vehicle technology, specifically to a method and apparatus for determining unsold vehicles, and a method and apparatus for training a model of unsold vehicles. Background Technology

[0002] Against the backdrop of rapid iteration in the automotive industry and increasingly fierce market competition, enterprises face multiple severe challenges in inventory management. Currently, automobile manufacturers generally adopt a front-end production model to quickly respond to end-sales demand. However, due to uncertainties such as market demand fluctuations and changes in consumer preferences, actual sales are prone to falling short of expectations, leading to long-term inventory buildup and the risk of unsold vehicles. These unsold vehicles not only tie up a large amount of working capital and increase warehousing management and maintenance costs, but may also depreciate due to technological advancements and changes in market supply and demand, forcing companies to reduce prices to clear inventory, ultimately negatively impacting profitability and market competitiveness. Summary of the Invention

[0003] In view of the shortcomings of the prior art, the purpose of this application is to provide a method and apparatus for determining unsold vehicles, and a method and apparatus for training a model of unsold vehicles, which aims to solve the problem of unsold inventory caused by the contradiction between the uncertainty of pre-production and market demand in the prior art.

[0004] In a first aspect, embodiments of this application provide a method for determining unsold vehicles, comprising: determining the estimated sales duration and estimated inventory level of a target type of vehicle at a future time based on target characteristics at the current time; the estimated sales duration is used to characterize the expected time from entering the inventory to final sale; the target characteristics are used to characterize characteristics that have an impact on the unsold status of the target type of vehicle; the future time is the time after a preset time has elapsed from the current time; determining the maximum allowable retention time of the target type of vehicle in the warehouse based on the estimated sales duration; determining the maximum allowable inventory level of the target type of vehicle at the current time based on the estimated inventory level at a future time; and determining whether the target type of vehicle is unsold at the current time based on the maximum allowable retention time and the maximum allowable inventory level.

[0005] Based on the aforementioned technical means, by encompassing multi-dimensional target features that influence the slow-moving status of target vehicle types, the estimated sales duration and future estimated inventory levels of target vehicle types can be accurately quantified based on the inherent correlation between these features and sales duration and inventory changes. On this basis, the maximum allowable retention time is adjusted in real-time based on the estimated sales duration, and the maximum allowable inventory level is adjusted in real-time based on the future estimated inventory level. This ensures that the maximum allowable retention time and maximum allowable inventory level meet the actual market demand for target vehicle types. Furthermore, the dual verification of maximum allowable retention time and maximum allowable inventory level enables accurate identification of slow-moving vehicles, providing data-driven decision support for production planning optimization. This allows for precise control of the production scale of target vehicle types, effectively alleviating the contradiction between pre-production and market demand uncertainty, and reducing the risk of inventory backlog.

[0006] In one possible embodiment, determining the maximum allowable retention time of a target type of vehicle in the warehouse based on the estimated sales duration includes: determining a duration difference between a reference duration and the estimated sales duration when the estimated sales duration is greater than or equal to the lower limit of the sales duration and less than the upper limit of the sales duration; the reference duration is used to characterize the sum of the upper limit and the lower limit of the sales duration; determining the maximum allowable retention time based on the duration difference and a preset scaling factor; determining the maximum allowable retention time as the product of the lower limit of the sales duration and the preset scaling factor when the estimated sales duration is greater than or equal to the upper limit of the sales duration; and determining the maximum allowable retention time as the upper limit of the sales duration when the estimated sales duration is less than the lower limit of the sales duration.

[0007] Based on the aforementioned technical means, the maximum allowable retention time is determined by comparing the estimated sales duration, the lower limit of the sales duration, and the upper limit of the sales duration. When the estimated sales duration is long, a shorter maximum allowable retention time can be set accordingly. By tightening the threshold for determining the unsold inventory time in advance, the timeliness of early warning of the risk of unsold inventory for the target type of vehicle is significantly improved, providing enterprises with sufficient intervention window period. Conversely, when the estimated sales duration is short, a longer maximum allowable retention time can be set accordingly, giving vehicles more leeway in inventory retention. This avoids excessive control over vehicles with high circulation efficiency due to overly strict time thresholds, ensuring market supply stability while achieving a dynamic balance between unsold inventory risk control and inventory turnover efficiency.

[0008] In one embodiment, determining the maximum permissible inventory of a target type of vehicle at the current moment based on the estimated inventory at future moments includes: determining the inventory difference between the actual inventory of the target type of vehicle at the current moment and the estimated inventory at future moments; determining the theoretical maximum permissible inventory of the target type of vehicle at the current moment based on the lower limit of inventory, the baseline inventory, and the inventory difference; and determining the smaller value between the theoretical maximum permissible inventory and the upper limit of inventory as the maximum permissible inventory.

[0009] Based on the aforementioned technical means, and taking the estimated inventory level at future moments as the core basis, a dynamic quantitative model is constructed to adjust the maximum allowable inventory level in real time through benchmark inventory level calibration, inventory change trend correction (determined based on the inventory difference between the current actual inventory level and the future estimated inventory level), and rigid constraints on the upper and lower limits of inventory level. This allows the control threshold of the inventory dimension to accurately match the market supply and demand balance.

[0010] In some embodiments, the theoretical maximum allowable inventory of a target type of vehicle at the current moment is determined based on the lower limit of inventory, the baseline inventory, and the inventory difference, including: determining a target ratio between the inventory difference and the actual inventory; determining the product of the baseline inventory and the target ratio as the inventory correction increment; determining the sum of the baseline inventory and the inventory correction increment as the inventory correction value; and determining the larger of the inventory correction value and the lower limit of inventory as the theoretical maximum allowable inventory.

[0011] Based on the above technical means, the changing trend of inventory is captured by the target ratio between the inventory difference and the actual inventory, so that the determined inventory correction increment is adapted to the dynamics of supply and demand, and the inventory correction value accurately reflects the reasonable inventory threshold; and the lower limit of inventory is used as a safety net to avoid the risk of stockouts caused by the threshold being too low, and finally the scientific quantification of the theoretical maximum allowable inventory is achieved.

[0012] In an exemplary embodiment, determining the estimated sales duration of a target type of vehicle based on target features includes: inputting the target features into the first feature extraction layer of a sales duration prediction model to obtain a first key feature; wherein the first feature extraction layer is used to extract features from the target features that are more important to the sales duration prediction task than a first preset importance threshold; and inputting the first key feature into the first prediction layer of the sales duration prediction model to obtain the estimated sales duration.

[0013] Based on the above technical means, the first key feature with high importance for sales duration prediction is screened out through the first feature extraction layer, which can effectively eliminate redundant and weakly correlated features, reduce the computational complexity of the model and improve the quality of input data; then, through the first prediction layer's learning and quantification of the core features, the predicted sales duration can be accurately obtained, significantly improving the accuracy of judging sluggish sales.

[0014] In an exemplary embodiment, determining the estimated future inventory level of a target type of vehicle based on target features includes: inputting the target features into the second feature extraction layer of the inventory prediction model to obtain a second key feature; wherein the second feature extraction layer is used to extract features from the target features that are more important to the inventory prediction task than a second preset importance threshold; and inputting the second key feature into the second prediction layer of the inventory prediction model to obtain the estimated future inventory level.

[0015] Based on the above technical means, by selecting the second key features that are of high importance to inventory quantity prediction through the second feature extraction layer, redundant interference information can be eliminated, the relevance of input data can be improved, and the model calculation cost can be reduced; then, through the deep learning and temporal pattern capture of the core features by the second prediction layer, the estimated inventory quantity at future time can be accurately output, significantly improving the accuracy of the judgment of slow-moving inventory.

[0016] In an exemplary embodiment, determining whether a target type of vehicle is unsaleable at the current moment based on the maximum allowable retention time and the maximum allowable inventory includes: determining that the target type of vehicle is unsaleable at the current moment if the actual retention time of the target type of vehicle at the current moment is greater than the maximum allowable retention time and the actual inventory of the target type of vehicle at the current moment is greater than the maximum allowable inventory.

[0017] Based on the aforementioned technical means, the target type of vehicle that is not selling well is determined by two dimensions: the maximum allowable retention time and the maximum allowable inventory. This avoids misjudging scenarios where short-term inventory is high but vehicle circulation efficiency is normal (such as pre-peak season stockpiling) as not selling well, and also prevents excessive warnings caused by situations where vehicles remain for a slightly longer time but the inventory size is within a controllable range (such as reasonable turnover of niche models). Ultimately, this achieves a scientific definition of the not selling state and provides a reliable basis for enterprises to accurately initiate destocking intervention measures.

[0018] Secondly, embodiments of this application provide a training method for a vehicle unsold model. The training method includes: constructing a training dataset; the training dataset includes: sample target features of a target type vehicle at a first moment and sample labels corresponding to the sample target features; the sample labels are used to characterize whether the target type vehicle is unsold at the first moment; the sample labels are determined based on the vehicle unsold determination method described in the first aspect; training an initial vehicle unsold model based on the training dataset until the initial vehicle unsold model converges; the initial vehicle unsold model includes a sales duration prediction network, an inventory prediction network, and a vehicle unsold network; the sales duration prediction model is used to extract sales duration prediction tasks from the sample target features. The system extracts features whose importance to the inventory prediction task is greater than a first preset importance threshold and uses these extracted features to determine the estimated sample sales duration of the target type of vehicle. The inventory prediction network extracts features from the target features whose importance to the inventory prediction task is greater than a second preset importance threshold and uses these extracted features to determine the estimated sample inventory of the target type of vehicle at the second time step. The vehicle unsold inventory network determines the unsold inventory probability of the target type of vehicle at the first time step based on the estimated sample sales duration and estimated sample inventory, and optimizes the model parameters of the initial vehicle unsold inventory model using the difference between the unsold inventory probability and the sample label. The second time step is the time after a preset duration has elapsed since the first time step. The converged initial vehicle unsold inventory model is then determined as the vehicle unsold inventory model.

[0019] Based on the aforementioned technical means, by constructing a training dataset of sample target features and sample labels, and combining an initial vehicle unsold model with a sales duration prediction network, an inventory prediction network, and a vehicle unsold network, the model focuses on core predictive factors (features whose importance to the sales duration prediction task is greater than the first preset importance threshold and features whose importance to the inventory prediction task is greater than the second preset importance threshold) through a feature selection mechanism to improve prediction accuracy. Furthermore, by using real unsold sample labels as supervision, the model parameters are optimized in reverse through differences in unsold probability. This enables the trained vehicle unsold model to accurately quantify the probability of unsold inventory, ensuring the objectivity and reliability of unsold inventory identification results, and providing efficient and accurate model support for determining the status of unsold vehicles.

[0020] Thirdly, embodiments of this application provide a method for determining unsold vehicles. The method includes: obtaining target features at the current moment; the target features are used to characterize features that have an impact on the unsold status of target type vehicles; inputting the target features into the unsold vehicle model of the second aspect above to obtain the unsold probability of target type vehicles at the current moment; and determining that the target type vehicles are unsold at the current moment if the unsold probability is greater than or equal to a preset probability threshold.

[0021] Based on the aforementioned technical means, by inputting multi-dimensional target features that have a significant impact on vehicle sales stagnation, and by leveraging the feature learning and probability quantification capabilities of the vehicle sales stagnation model, the current sales stagnation probability of the target type of vehicle can be accurately output, thereby scientifically determining whether it is in a sales stagnation state. This avoids the bias of subjective experience judgment and improves the accuracy and objectivity of sales stagnation identification.

[0022] Fourthly, embodiments of this application provide a device for determining vehicle unsaleability. The device includes: a first determining module, a second determining module, a third determining module, and a fourth determining module. The first determining module is used to determine the estimated sales duration and estimated inventory level at future times for a target type of vehicle based on target characteristics at the current moment. The estimated sales duration represents the expected time from entering inventory to final sale. The target characteristics represent features that influence the unsaleability of the target type of vehicle. The future time is the time after a preset duration has elapsed from the current moment. The second determining module is used to determine the maximum allowable retention time of the target type of vehicle in the warehouse based on the estimated sales duration. The third determining module is used to determine the maximum allowable inventory level of the target type of vehicle at the current moment based on the estimated inventory level at future times. The fourth determining module is used to determine whether the target type of vehicle is unsaleable at the current moment based on the maximum allowable retention time and the maximum allowable inventory level.

[0023] Fifthly, embodiments of this application provide a training apparatus for a vehicle unsold model. The training apparatus includes: a construction module, a training module, and a fifth determination module. The construction module is used to construct a training dataset. The training dataset includes: sample target features of the target type vehicle at a first moment and sample labels corresponding to the sample target features. The sample labels are used to characterize whether the target type vehicle is unsold at the first moment. The sample labels are determined based on the vehicle unsold determination method described in the first aspect. The training module trains an initial vehicle unsold model based on the training dataset until the initial vehicle unsold model converges. The initial vehicle unsold model includes a sales duration prediction network, an inventory prediction network, and a vehicle unsold network. The sales duration prediction network is used to extract data from the sample target features... The system uses features whose importance to the sales duration estimation task is greater than a first preset importance threshold, and uses the extracted features to determine the estimated sample sales duration of the target type of vehicle; the inventory prediction network extracts features from the sample target features whose importance to the inventory estimation task is greater than a second preset importance threshold, and uses the extracted features to determine the estimated sample inventory of the target type of vehicle at the second time point; the vehicle unsold network determines the unsold probability of the target type of vehicle at the first time point based on the estimated sample sales duration and the estimated sample inventory, and uses the difference between the unsold probability and the sample label to optimize the model parameters of the initial vehicle unsold model; the second time point is the time when the first time point has elapsed for a preset duration; the fifth determination module is used to determine the converged initial vehicle unsold model as the vehicle unsold model.

[0024] In a sixth aspect, embodiments of this application provide a device for determining vehicle unsaleability. The device includes: an acquisition module, an input module, and a sixth determination module. The acquisition module is used to acquire target features at the current moment. The target features are used to characterize features that have an impact on the unsaleability of target type vehicles. The input module is used to input the target features into the vehicle unsaleability model of the second aspect described above to obtain the unsaleability probability of the target type vehicle at the current moment. The sixth determination module is used to determine that the target type vehicle is unsaleable at the current moment if the unsaleability probability is greater than or equal to a preset probability threshold.

[0025] In a seventh aspect, this application provides an electronic device comprising: a processor and a memory; the memory storing processor-executable instructions. When the processor is configured to execute the instructions, the electronic device causes the electronic device to implement any one of the methods described in the first to third aspects.

[0026] Eighthly, this application provides a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement any one of the methods described in the first to third aspects.

[0027] Ninthly, this application provides a computer program product including computer program instructions that, when executed by a processor, implement any one of the methods described in the first to third aspects.

[0028] It should be noted that the technical effects of any of the implementation methods in aspects four through nine can be found in the technical effects of the corresponding implementation methods in aspects one through four, and will not be repeated here.

[0029] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of this application or the background art, the accompanying drawings used in the embodiments of this application will be described below.

[0031] Figure 1 This is a schematic diagram of the structure of a device for determining unsold vehicles disclosed in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a training device for a vehicle slow-selling model disclosed in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of another vehicle unsold product determination device disclosed in the embodiments of this application; Figure 4 This is a flowchart illustrating a method for determining unsold vehicles disclosed in an embodiment of this application; Figure 5 This is a schematic diagram illustrating the composition of a target feature disclosed in an embodiment of this application; Figure 6 This is a flowchart illustrating a training method for a vehicle slow-selling model disclosed in an embodiment of this application. Figure 7 This is a flowchart illustrating another method for determining unsold vehicles disclosed in an embodiment of this application; Figure 8 This is a flowchart illustrating another method for determining unsold vehicles disclosed in an embodiment of this application; Figure 9 This is a flowchart illustrating another method for determining unsold vehicles disclosed in an embodiment of this application; Figure 10 This is a flowchart illustrating another method for determining unsold vehicles disclosed in an embodiment of this application; Figure 11 This is a flowchart illustrating another method for determining unsold vehicles disclosed in an embodiment of this application; Figure 12 This is a flowchart illustrating another method for determining unsold vehicles disclosed in an embodiment of this application; Figure 13 This is a flowchart illustrating another method for determining unsold vehicles disclosed in an embodiment of this application; Figure 14 This is a flowchart illustrating another method for determining unsold vehicles disclosed in an embodiment of this application; Figure 15 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application. Detailed Implementation

[0032] The terms “first,” “second,” etc., are used for descriptive purposes only and have no sequential or technical meaning, nor should they be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated.

[0033] In the description of the embodiments of this application, unless otherwise expressly specified and limited, the terms "installation" and "connection" should be interpreted broadly. For example, "connection" can be a detachable connection or a non-detachable connection; it can be a direct connection or an indirect connection through an intermediate medium. "Fixed connection" refers to a connection where the relative positional relationship remains unchanged after connection. "Rotary connection" refers to a connection where the two parts can rotate relative to each other after connection. "Sliding connection" refers to a connection where the two parts can slide relative to each other after connection.

[0034] The embodiments of this application are described below with reference to the accompanying drawings.

[0035] Please see Figure 1 , Figure 1 This is a schematic diagram of the structure of a device for determining unsold vehicles disclosed in an embodiment of this application.

[0036] In some embodiments, the vehicle unsold determination device 100 includes: an acquisition module 101, an input module 102, and a sixth determination module 103.

[0037] In some embodiments, the acquisition module 101 is used to acquire the target features at the current moment.

[0038] Among them, target features are used to characterize features that have an impact on the sluggish sales of target type vehicles.

[0039] Input module 102 is used to input the target features into the vehicle unsold model to obtain the unsold probability of the target type of vehicle at the current time.

[0040] The sixth determining module 103 is used to determine whether a target type of vehicle is unsold at the current moment when the probability of unsold inventory is greater than or equal to a preset probability threshold.

[0041] For example, the vehicle unsold determination device 100 can be any device or equipment capable of determining unsold status, such as a server cluster consisting of multiple servers, a single server, a computer, or a processor or processing chip in a server or computer. This application embodiment does not limit this.

[0042] In some embodiments, the vehicle unsold inventory determination device 100 receives a vehicle unsold inventory model uploaded by a vehicle unsold inventory model training device. For example, the vehicle unsold inventory determination device 100 may be connected to the vehicle unsold inventory model training device, or the vehicle unsold inventory model training device may be integrated into the vehicle unsold inventory determination device 100.

[0043] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a training device for a vehicle sales slump model disclosed in an embodiment of this application. The training device 200 for the vehicle sales slump model includes: a construction module 201, a training module 202, and a fifth determination module 203.

[0044] Module 201 is used to construct the training dataset. The training dataset includes: sample target features of target type vehicles at the first moment and sample labels corresponding to the sample target features. The sample labels are used to characterize whether the target type vehicles are unsaleable at the first moment. The sample labels are determined based on the method for determining unsaleable vehicles. Training module 202 trains the initial vehicle sales stagnation model based on the training dataset until the initial vehicle sales stagnation model converges.

[0045] The initial vehicle sales stagnation model includes a sales duration prediction network, an inventory prediction network, and a vehicle sales stagnation network. The sales duration prediction model extracts features from the sample target features that are more important than a first preset importance threshold for the sales duration prediction task, and uses these extracted features to determine the estimated sample sales duration for the target type of vehicle. The inventory prediction network extracts features from the sample target features that are more important than a second preset importance threshold for the inventory prediction task, and uses these extracted features to determine the estimated sample inventory of the target type of vehicle at the second time point. The vehicle sales stagnation network determines the probability of the target type of vehicle becoming unsold at the first time point based on the estimated sample sales duration and the estimated sample inventory, and optimizes the model parameters of the initial vehicle sales stagnation model using the difference between the unsold probability and the sample label. The second time point is the time after a preset duration has elapsed since the first time point. The fifth determination module 203 is used to determine the converged initial vehicle sales stagnation model as the vehicle sales stagnation model.

[0046] In some embodiments, the training device 200 for the vehicle unsold model receives sample tags uploaded by the vehicle unsold determination device 100.

[0047] Please see Figure 3 , Figure 3 This is a schematic diagram of another vehicle unsold inventory determination device disclosed in an embodiment of this application. The vehicle unsold inventory determination device 100 includes: a first determination module 301, a second determination module 302, a third determination module 303, and a fourth determination module 304.

[0048] The first determining module 301 is used to determine the estimated sales duration and estimated inventory level of the target type of vehicle based on the target characteristics at the current moment; the estimated sales duration is used to characterize the expected time from entering the inventory to final sale; the target characteristics are used to characterize the characteristics that have an impact on the sluggish sales of the target type of vehicle; the future moment is the time when the current moment has elapsed for a preset period of time.

[0049] The second determining module 302 is used to determine the maximum allowable retention time of the target type of vehicle in the warehouse based on the estimated sales duration.

[0050] The third determining module 303 is used to determine the maximum allowable inventory of the target type of vehicle at the current moment based on the estimated inventory at future times.

[0051] The fourth determination module 304 is used to determine whether the target type of vehicle is unsaleable at the current moment based on the maximum allowable retention time and the maximum allowable inventory.

[0052] Please see Figure 4 , Figure 4 This is a flowchart illustrating a method for determining unsold vehicles disclosed in an embodiment of this application.

[0053] In some embodiments, the method for determining unsold vehicles can be implemented by the following steps: S401, Obtain the target features at the current moment.

[0054] Among them, target features are used to characterize features that have an impact on the sluggish sales of target type vehicles.

[0055] As a feasible implementation method, such as Figure 5 As shown, the target features may include at least one of the following: vehicle attribute features, vehicle inventory and sales cycle (sales cycle is equivalent to sales duration) statistical features, regional attributes and inventory statistical features, holiday association features, time dimension features, and regional vehicle combination association statistical features.

[0056] Vehicle attribute characteristics are used to characterize the inherent technical parameters, configuration levels, and market positioning attributes of a target type of vehicle. For example, vehicle attribute characteristics may include: vehicle series identifier, model identifier, vehicle category, space level, price range, new energy attributes, pure electric range, intelligent driving configuration level, time on the market (unit: month / week / day), and whether it is a newly launched model, etc.

[0057] Statistical characteristics of vehicle inventory and sales cycle are used to characterize the fluctuation patterns of inventory quantity and related statistical attributes of circulation cycle of target type vehicles over historical periods. For example, the statistical characteristics of vehicle inventory and sales cycle may include: the maximum, minimum, mean, variance, coefficient of variation, moving average, growth rate, first difference, second difference, and linear regression slope of daily inventory quantity for the target type vehicles in the most recent 7 / 15 / 30 days; the year-on-year change rate and month-on-month change rate of inventory quantity; the maximum, minimum, mean, variance, coefficient of variation, moving average, growth rate, first difference, second difference, and linear regression slope of daily circulation cycle in the most recent 7 / 15 / 30 days; the longest historical circulation cycle; and the historical frequency of circulation cycles exceeding preset thresholds (e.g., 60 days, 90 days).

[0058] Regional attributes and inventory statistics are used to characterize the geographical attributes, economic development level, and inventory fluctuation characteristics of the target type of vehicles in a vehicle sales region. For example, regional attributes and inventory statistics may include: regional identifier, region of origin, geographical location, whether it is a coastal region, economic development level (e.g., first-tier, new first-tier), total regional inventory quantity; maximum, minimum, mean, variance, coefficient of variation, moving average, growth rate, first difference, second difference, and linear regression slope of the daily inventory quantity in the region over the past 7 / 15 / 30 days; and year-on-year and month-on-month changes in regional inventory quantity.

[0059] Holiday association features are used to characterize the relationship between the current moment and statutory holidays. For example, holiday association features may include: whether the current moment belongs to statutory holidays such as Spring Festival, Labor Day, Dragon Boat Festival, Mid-Autumn Festival, National Day, and New Year's Day; and whether the current moment is within the 1st / 2nd / 3rd week before or after statutory holidays such as Spring Festival, Dragon Boat Festival, and National Day.

[0060] The time dimension feature is used to characterize the time attributes of the current time point and the stage of the sales cycle. For example, the time dimension feature may include: the year, quarter, month, and week of the year to which the current time belongs; whether the current time is the end of the first half of the year, the end of the second half of the year, the end of the quarter, the end of the month, or the beginning of the month; whether the current date is within the 1st / 2nd / 3rd week before a preset promotional node (such as "Double 11"), etc.

[0061] The correlation statistics of regional vehicle combinations are used to characterize the correlation statistics of inventory quantity and sales cycle under the combination dimension of a specific region and a target type of vehicle. For example, the correlation statistics of regional vehicle combinations may include: the maximum, minimum, mean, variance, coefficient of variation, moving average, growth rate, first difference, second difference, and linear regression slope of the daily inventory quantity of the target type of vehicle in the specific region over the past 7 / 15 / 30 days; the year-on-year change rate and month-on-month change rate of the inventory quantity of the target type of vehicle in the specific region; the maximum, minimum, mean, variance, coefficient of variation, moving average, growth rate, first difference, second difference, and linear regression slope of the daily circulation cycle of the target type of vehicle in the specific region over the past 7 / 15 / 30 days; the longest historical circulation cycle of the target type of vehicle in the specific region; the historical frequency of the circulation cycle of the target type of vehicle in the specific region exceeding a preset threshold (such as 60 days, 90 days); and the growth rate, first difference, second difference, and linear regression slope of the sales of the target type of vehicle in the specific region over the past 7 days and the past 7 weeks.

[0062] It should be understood that, from the vehicle perspective, capturing inherent attributes such as vehicle configuration, technical level, and market positioning through vehicle attribute characteristics provides a basis for judging the vehicle's competitiveness and potential for slow sales; from the inventory circulation perspective, leveraging the statistical characteristics of vehicle inventory and sales cycles quantifies historical inventory fluctuation patterns and circulation efficiency, reflecting the vehicle's circulation status in the market; from the regional perspective, combining regional attributes and inventory statistics with regional geographical, economic, and consumption capacity characteristics and overall regional inventory levels reflects the supply and demand differences in the regional market; from the time perspective, incorporating time-dimensional characteristics and holiday-related features, integrating time-dimensional influencing factors such as sales cycle stages, promotional nodes, and holiday consumption patterns, adapts to the time-varying nature of market demand; and from the perspective of combination and coordination, focusing on the compatibility between specific regions and target vehicles through the correlation statistical characteristics of regional vehicle combinations, quantifies the exclusive circulation characteristics of target vehicles within a region, and compensates for the limitations of describing a single region or single vehicle dimension.

[0063] The systematic construction of the above-mentioned multi-dimensional features can achieve full-scenario coverage of the "region-vehicle" combined business status, including both static inherent attributes and dynamic changing trends; it reflects both single subject characteristics and takes into account the synergistic effect of combined dimensions, thereby providing comprehensive and accurate feature inputs for subsequent sales cycle prediction, inventory prediction and slow-moving judgment, ensuring that the prediction model and judgment logic can meet the complex needs of actual business scenarios.

[0064] S402. Input the target features into the vehicle sales stagnation model to obtain the sales stagnation probability of the target type of vehicle at the current time.

[0065] Among them, the vehicle unsold model is used to determine the unsold probability of a target type of vehicle based on the target characteristics of the target type of vehicle in a specific region.

[0066] The probability of unsold inventory is used to characterize the likelihood that a target type of vehicle will not be sold within a preset time period and will fall into an unsold inventory state in the current business scenario (target type of vehicle in a specific region) (the value range is usually [0,1] or [0,100%]).

[0067] It should be understood that by using quantitative methods to reflect the comprehensive impact of multiple factors such as the suitability of the target vehicle to the regional market and the inventory circulation trend on sales results, the closer the probability of unsold inventory is to the upper limit, the higher the risk of unsold inventory within the preset period; the closer the probability of unsold inventory is to the lower limit, the greater the probability of smooth circulation of vehicles and avoiding unsold inventory.

[0068] As a feasible implementation method, the vehicle unsold model can use a pre-trained prediction model. By learning the mapping relationship between the target features of vehicle combinations in historical regions and the actual unsold results, the vehicle unsold model has the ability to directly output the unsold probability based on the multi-dimensional target features of the input.

[0069] As another feasible implementation method, the vehicle overstock model can be trained based on the training dataset. For details, please refer to the description in steps S601-S602 below, which will not be repeated here.

[0070] S403. If the probability of unsold inventory is greater than or equal to a preset probability threshold, determine that the target type of vehicle is unsold at the current moment.

[0071] The preset probability threshold is used to represent the minimum probability of unsold vehicles.

[0072] In an exemplary embodiment, the preset probability threshold is set to 0.7 (i.e., 70%). If the calculated probability of a target type of vehicle being unsold in a specific area is 0.75 (75%), since this probability is greater than the preset probability threshold of 0.7, it is determined that the vehicle is currently in an unsold state, triggering an inventory optimization intervention process (such as regional transfer, adjustment of promotional strategies, etc.).

[0073] As another feasible approach, if the probability of unsold inventory is less than a preset probability threshold, it can be determined that the target type of vehicle is not unsold at the current moment.

[0074] In an exemplary embodiment, if the calculated probability of unsold vehicles of a target type in a specific area is 0.62 (62%), since it is less than the preset probability threshold of 0.7, it is determined that the vehicle is not currently unsold and the existing inventory management strategy is maintained.

[0075] By inputting multi-dimensional target features that significantly influence vehicle sales stagnation, and leveraging the feature learning and probability quantification capabilities of the vehicle sales stagnation model, the system can accurately output the current sales stagnation probability of a target type of vehicle, thereby scientifically determining whether it is in a sales stagnation state. This avoids the bias of subjective experience judgment and improves the accuracy and objectivity of sales stagnation identification.

[0076] The training process of the vehicle slow-selling model is described below.

[0077] In some embodiments, such as Figure 6 As shown, the training method for the vehicle slow-selling model can be implemented in the following steps: S601. Construct the training dataset.

[0078] The training dataset includes: the target features of the target type vehicle at the first moment and the sample labels corresponding to the target features.

[0079] The scope of the target vehicle type can be flexibly defined. It can be a single vehicle in a specific vehicle type (such as a single display vehicle of a certain brand of pure electric sedan) or a collection of multiple individuals in that type of vehicle (such as multiple pure electric sedans of a certain brand with the same configuration and batch), adapting to different data collection and model training scenarios.

[0080] In an exemplary embodiment, the sample label is used to characterize whether the target type of vehicle is unsaleable at the first moment.

[0081] For example, the implementation of sample labels can be determined by referring to the following steps S701-S704, which will not be elaborated here.

[0082] In an exemplary embodiment, the sample target features are used to represent historical data of the aforementioned target features.

[0083] S602. Train the initial vehicle sales slump model based on the training dataset until the initial vehicle sales slump model converges.

[0084] As a feasible implementation method, the initial vehicle unsold model includes a sales duration prediction network, an inventory prediction network, and a vehicle unsold network.

[0085] In an exemplary embodiment, the sales duration prediction network is used to extract features from the sample target features that are more important to the sales duration prediction task than a first preset importance threshold, and to use the extracted features to determine the estimated sample sales duration of the target type vehicle.

[0086] The first preset importance threshold is used to characterize the critical standard for the contribution of the target features of the segmented samples to the sales duration prediction task. Its core function is to quantify the boundary between key features strongly correlated with sales duration prediction and redundant features with low contribution. By using the first preset importance threshold, features with weak impact on the prediction results can be filtered out, ensuring that the features input to the sales duration prediction network have high relevance and effectiveness, thereby improving the accuracy of the predicted sales duration of the samples. For example, the first preset importance threshold is 70%.

[0087] The estimated sample sales time is used to represent the estimated time required for a target type of vehicle to go from being received to being shipped out.

[0088] As a feasible implementation method, the sales duration prediction network includes: a first feature extraction layer and a first prediction layer.

[0089] For example, the training process of the first feature extraction layer includes: constructing an extreme gradient boosting tree model (eXtremeGradient Boosting, XGBoost), using the target features of the samples as input and the actual sales duration of the vehicles as labels for model training; during the training process, quantifying the contribution of each target feature of the samples to the sales duration prediction task through the feature importance evaluation mechanism built into the XGBoost model (such as information gain and coverage based on split nodes); subsequently, combining the first preset importance threshold, selecting key features with importance higher than the first preset importance threshold to complete feature extraction and redundancy removal.

[0090] The training process of the first prediction layer includes: using key features selected by the XGBoost model as input data, and using the actual sales duration of the "region-vehicle" combination in historical inventory data as the supervision label (this sample label is determined by calculating the difference in natural days from the date of entry to the actual date of exit for a single vehicle in the target region, i.e., sales duration = actual exit date - actual entry date); constructing a Light Gradient Boosting Machine Regression (LightGBM regression model) based on the above key features and sample labels and conducting training. During the training process, the model parameters are optimized through the gradient boosting algorithm, so that the model can fully learn the nonlinear mapping relationship between key features and sales duration. At the same time, by leveraging its efficient fitting ability for high-dimensional features and its accurate capture ability for inventory cycle time series patterns, it can adapt to the turnover difference scenarios of different regions and different vehicle models; after the model training is completed, inputting new selected key features will output the estimated sample sales duration of the target type of vehicle.

[0091] It should be understood that the XGBoost model's strong feature filtering capability eliminates low-contribution redundant features, reducing the training complexity and overfitting risk of subsequent regression models; at the same time, the efficiency and fitting capability of the LightGBM regression model in continuous value prediction tasks are utilized to achieve accurate quantitative prediction of sales duration.

[0092] In an exemplary embodiment, the inventory prediction network is used to extract features from the sample target features that are more important to the inventory prediction task than a second preset importance threshold, and to use the extracted features to determine the estimated sample inventory of the target type vehicle at a second time.

[0093] The second preset importance threshold is used to characterize the critical standard for the contribution of the target features of the divided samples to the inventory prediction task. Its core function is to quantify the boundary between key features that are strongly correlated with inventory prediction and redundant features with low contribution. By using the second preset importance threshold, features that have a weak impact on the prediction results can be filtered out, irrelevant information interference can be reduced, and the features input to the prediction network can be highly correlated and effective, thereby improving the accuracy of the second time step in the predicted sample inventory.

[0094] The estimated sample inventory is used to represent the possible inventory of the target type of vehicle at the second time point.

[0095] As a feasible implementation method, the inventory prediction network includes: a second feature extraction layer and a second prediction layer.

[0096] In some embodiments, the training process of the second feature extraction layer includes: training the XGBoost model with sample target features as input and the actual inventory of target type vehicles at the second time step as supervision label; during the training process, using the feature importance evaluation mechanism built into the XGBoost model (such as information gain based on split nodes, feature coverage, weight ratio, etc.) to quantify the importance of each sample target feature to the inventory prediction task; combining the second preset importance threshold, selecting key features with importance indices higher than the second preset importance threshold, and eliminating low-contribution redundant features with importance indices lower than the second preset importance threshold, thus completing the training of the second feature extraction layer and achieving accurate extraction of key features related to inventory prediction.

[0097] The training process of the second prediction layer includes: using the key features filtered by the second feature extraction layer as input data, and using the short-term inventory of the "region-vehicle" combination in historical inventory data as the supervision label. This label can be specifically defined as the actual inventory of the "region-vehicle" combination corresponding to the current time point (e.g., day t) in the next 7 days (i.e., day t+7), calculated by accumulating the daily inbound, outbound, and inter-regional transfer volumes from day t to day t+6; training the LightGBM regression model based on the above input features and labels, using histogram discretization and leaf-based methods during the training process. The model employs optimization mechanisms such as leaf-wise growth and parallel training to fully leverage its ability to capture temporal fluctuations. It efficiently learns the nonlinear mapping relationship between key features and inventory levels over the next 7 days. Simultaneously, it iteratively optimizes model parameters such as learning rate, tree depth, and number of leaf nodes using gradient descent to reduce prediction errors. After training, inputting new, filtered key features yields the estimated sample inventory levels of target vehicle types at the corresponding future time (e.g., t+7 days), enabling dynamic prediction of short-term inventory levels for "region-vehicle" combinations and adapting to business scenarios with real-time inventory changes. The 7-day period is an example; the specific number of days can be set and configured independently.

[0098] It should be understood that by leveraging the strong feature selection capabilities of the XGBoost model, key features that are strongly correlated with inventory level prediction can be extracted in a targeted manner, reducing the training complexity and overfitting risk of the second prediction layer; at the same time, the advantages of the LightGBM regression model in terms of efficient fitting capability and low computational cost under large-scale data are utilized to balance the accuracy of inventory level prediction with the efficiency of engineering implementation.

[0099] In an exemplary embodiment, the vehicle unsold network is used to determine the unsold probability of a target type of vehicle at a first moment based on the estimated sample sales duration and the estimated sample inventory, and to optimize the model parameters of the initial vehicle unsold model by utilizing the difference between the unsold probability and the sample label.

[0100] For example, the training process of the vehicle unsold network includes: using region-vehicle combinations as the basic unit, constructing binary classification sample labels based on the core logic of unsold determination; using the core feature subset filtered by the differential feature selection mechanism as the model input, and using the constructed binary classification sample labels as the supervision target, constructing a LightGBM binary classification model and conducting training; and iteratively adjusting the parameters of the initial vehicle unsold model by calculating the difference between the unsold probability output by the model and the sample labels, continuously improving the accuracy and generalization ability of the model in predicting the unsold probability.

[0101] The construction of binary classification sample labels includes: if a region-vehicle combination sample meets the dual conditions of "actual sales duration is greater than the maximum allowable retention time and actual inventory is greater than the maximum allowable inventory", it is marked as a slow-moving sample and the sample label is set to 1; if the combination sample does not meet the above dual conditions (i.e., actual sales duration is less than or equal to the maximum allowable retention time or actual inventory is less than or equal to the maximum allowable inventory), it is marked as a non-slow-moving sample and the sample label is set to 0.

[0102] The construction and training of the LightGBM binary classification model includes: using a subset of core features selected by a differential feature selection mechanism as the model input data; secondly, using the constructed binary classification sample labels as the supervision target, initializing the basic parameters of the LightGBM binary classification model (such as tree depth, learning rate, number of leaf nodes, number of iterations, etc.); subsequently, using the gradient boosting algorithm to train the model. During the training process, leveraging LightGBM's histogram discretization, leaf-by-leaf growth, and parallel computing capabilities, the model efficiently learns the nonlinear correlation between core features and the "region-vehicle" combination of sluggish sales status, accurately capturing the driving effect of core features such as the estimated sample sales duration and estimated sample inventory on sluggish sales risk, until the model's fitting effect on the validation set tends to stabilize.

[0103] The calculation of the difference between the unsold probability output by the model and the sample label, and the iterative adjustment of the parameters of the initial vehicle unsold probability model, includes: using the cross-entropy loss function as the core evaluation index, calculating the difference between the unsold probability output by the model (value range [0,1] or [0,100%]) and the sample label (0 or 1), and quantifying the prediction error of the model; based on this loss value, backpropagating the error through the gradient descent algorithm, and iteratively adjusting the key parameters of the initial vehicle unsold probability model (such as adjusting the weight allocation of core features, optimizing the splitting strategy of the tree model, and correcting the learning rate); after each round of parameter adjustment, verifying the model performance (such as accuracy, recall, AUC value, etc.) on the validation set; if the performance index does not meet the preset standard, continuing iterative optimization; until the model performance meets the preset requirements (such as an AUC value greater than or equal to 0.85), stopping parameter adjustment, completing model training, and ensuring that the final model has accurate and stable unsold probability prediction capabilities.

[0104] Optionally, when performing feature screening at the first and second feature layers, the dynamic thresholds (maximum allowable retention time, maximum allowable inventory quantity) used for the final determination of slow-moving inventory have not yet been generated. Therefore, a fixed threshold is temporarily used to adapt to the scenario to complete the calculation and screening of feature importance (e.g., setting a fixed inventory days threshold of 90 days and a fixed inventory quantity threshold of 10 units). Based on the target features of the samples under the fixed threshold scenario, feature importance is quantified to ensure that the selected core feature subset is targeted and effective.

[0105] It should be understood that the second moment is the moment after the first moment has elapsed for a preset period of time.

[0106] S603. The converged initial vehicle sales stagnation model is determined as the vehicle sales stagnation model.

[0107] By constructing a training dataset of sample target features and sample labels, and combining an initial vehicle unsold model with a sales duration prediction network, an inventory prediction network, and a vehicle unsold network, the model focuses on core predictive factors (features whose importance to the sales duration prediction task is greater than the first preset importance threshold and features whose importance to the inventory prediction task is greater than the second preset importance threshold) through a feature selection mechanism to improve prediction accuracy. Furthermore, by using real unsold sample labels as supervision, the model parameters are optimized inversely through differences in unsold probability. This enables the trained vehicle unsold model to accurately quantify the probability of unsold inventory, ensuring the objectivity and reliability of unsold inventory identification results and providing efficient and accurate model support for determining the status of unsold vehicles.

[0108] Please see Figure 7 , Figure 7 This is a flowchart illustrating another method for determining unsold vehicles disclosed in an embodiment of this application.

[0109] In some embodiments, the method for determining unsold vehicles may also be implemented by the following steps: S701. Based on the target characteristics at the current moment, determine the estimated sales duration and estimated inventory level for the target type of vehicle at future moments.

[0110] Among them, the estimated sales duration is used to characterize the expected time from entering inventory to final sale.

[0111] Target features are used to characterize features that influence the sales performance of target type vehicles.

[0112] The future moment is the moment after a preset duration from the current moment.

[0113] Future time-period estimated inventory is used to characterize the inventory of a target type of vehicle at future time.

[0114] As a feasible approach, the complete target features at the current moment are directly input into an integrated vehicle sales stagnation model. This model integrates two core functional modules: a sales duration prediction network and an inventory prediction network. The two prediction networks can share the same main neural network, which is responsible for performing unified high-dimensional feature extraction, fusion, and abstract representation on the input target features. This efficiently mines the potential correlations and common information between features, avoiding computational redundancy and resource waste caused by repeated feature processing. Based on this, the prediction task is completed through two differentiated task-specific output layers.

[0115] The dedicated output layer of the sales duration prediction network is a regression task layer. Based on the high-dimensional feature vector output by the shared master network, it learns the non-linear mapping relationship between features and sales duration, outputting a continuous estimated sales duration. Similarly, the dedicated output layer of the inventory quantity prediction network is also a regression task layer. It optimizes parameter configuration for the business characteristics of inventory quantity prediction, learns the correlation logic between features and future inventory quantities, and outputs a continuous estimated inventory quantity for future times. During model training, a multi-task learning strategy can be adopted. The weighted sum of the sales duration prediction error and the inventory quantity prediction error is used as the total loss function. The parameters of the shared master network and the two dedicated output layers are optimized simultaneously, enabling the two prediction tasks to adapt collaboratively. While ensuring prediction accuracy, this significantly improves the model's computational efficiency and lightweight deployment, making it particularly suitable for scenarios with large data volumes and limited deployment resources.

[0116] As another feasible approach, an independent dual-model parallel architecture is adopted. The target features at the current moment are input into a specially trained sales duration prediction model and an inventory prediction model, respectively, to independently complete the two types of prediction tasks and obtain the corresponding estimated sales duration and estimated inventory at future moments.

[0117] S702. Based on the estimated sales duration, determine the maximum allowable retention time of the target type of vehicle in the warehouse.

[0118] Among them, the maximum allowable retention time is used to characterize the minimum sales duration for judging the unsold status of target type vehicles.

[0119] As a feasible approach, the maximum allowable retention time is negatively correlated with the estimated sales time. That is, the longer the estimated sales time, the lower the market circulation efficiency of the vehicle, the lower the consumer acceptance, or the greater the difficulty in selling it; correspondingly, the maximum allowable retention time should be set shorter. By compressing the tolerable retention time, early warnings of slow-moving inventory can be triggered, reducing costs such as capital occupation, warehousing losses, and vehicle depreciation caused by long-term inventory backlog. Conversely, the shorter the estimated sales time, the stronger the market demand and the faster the circulation of the vehicle; the maximum allowable retention time can be appropriately relaxed to avoid excessive intervention in normal inventory turnover due to overly strict thresholds, ensuring a reasonable sales time and customer choice.

[0120] In an exemplary embodiment, a first preset negative ratio coefficient is pre-calibrated for different vehicle models, regions, or market scenarios, and the maximum allowable retention time can be the product of the estimated sales duration and the first preset negative ratio coefficient.

[0121] It should be understood that the calibration of the first preset negative proportional coefficient is not fixed and can be flexibly adjusted according to multi-dimensional business scenarios. For example, it can be combined with factors such as the company's capital cost tolerance (the higher the capital cost, the smaller the coefficient value and the stricter the threshold), regional market consumption potential (first-tier city markets have faster absorption, so the coefficient can be higher than that of third- and fourth-tier cities), vehicle life cycle (the coefficient can be appropriately relaxed in the early stage of the launch of a new model, and the coefficient needs to be tightened when the model is about to be updated), and inventory backlog warning level (the coefficient is lowered when inventory pressure is high and raised when inventory is sufficient). Through historical data verification and cross-validation, the coefficient value is optimized to ensure that the setting of the maximum allowable retention time not only conforms to the core logic of matching circulation efficiency and risk tolerance, but also accurately adapts to the needs of unsold goods judgment in different scenarios, and achieves a balance between the timeliness and rationality of risk warning. At the same time, this quantitative calculation method has strong operability and can be directly embedded into the threshold generation module of the vehicle unsold goods model.

[0122] As another feasible implementation method, when the estimated sales duration is greater than or equal to the lower limit of sales duration and less than the upper limit of sales duration, the duration difference between the reference duration and the estimated sales duration is determined, and the maximum allowable retention duration is determined based on the duration difference and the preset scaling factor; when the estimated sales duration is greater than or equal to the upper limit of sales duration, the maximum allowable retention duration is determined to be the product of the lower limit of sales duration and the preset scaling factor; when the estimated sales duration is less than the lower limit of sales duration, the maximum allowable retention duration is determined to be the upper limit of sales duration. The specific implementation method is described in steps S1001-S1003 below, and will not be elaborated here.

[0123] S703. Based on the estimated inventory levels at future times, determine the maximum permissible inventory level for the target type of vehicle at the current time.

[0124] The maximum allowable inventory level is used to characterize the minimum inventory level required to determine if a target type of vehicle is slow to sell.

[0125] As a feasible implementation method, future inventory forecasts can reflect the inventory change trend of the target type of vehicles. If the future inventory forecast is higher than the current inventory, it indicates a continuous accumulation trend; if the future inventory forecast is lower than the current inventory, it indicates a gradual depletion trend. The inventory change trend is negatively correlated with the current maximum allowable inventory level. When the future inventory forecast is high and the inventory accumulation trend is obvious, it indicates that the subsequent market demand for vehicles is high, and the corresponding current maximum allowable inventory level needs to be set lower to control the inventory scale from the source and avoid further accumulation. When the future inventory forecast is low and the inventory depletion trend is good, it indicates that the market demand for vehicles is strong, and the current maximum allowable inventory level can be appropriately relaxed to ensure market supply and balance the risk of shortages and inventory costs.

[0126] In an exemplary embodiment, in order to achieve a quantitative correlation between the maximum allowable inventory and the estimated inventory at future time, a second preset negative proportional coefficient can be pre-calibrated between the two (this coefficient is a constant greater than 0 and less than 1, and its value is dynamically adapted to the inventory change trend reflected by the estimated inventory at future time, reflecting the core logic of negative correlation). Based on the product of the second negative proportional coefficient and the estimated inventory at future time, the maximum allowable inventory of the target type of vehicle at the current time is determined.

[0127] As another feasible implementation method, determine the inventory difference between the actual inventory of the target type of vehicle at the current moment and the estimated inventory at the future moment; based on the lower limit of inventory, the benchmark inventory, and the inventory difference, determine the theoretical maximum allowable inventory of the target type of vehicle at the current moment; determine the smaller value between the theoretical maximum allowable inventory and the upper limit of inventory as the maximum allowable inventory. The specific implementation method is described in steps S1201-S1203 below, which will not be elaborated here.

[0128] S704. Based on the maximum allowable retention time and the maximum allowable inventory, determine whether the target type of vehicle is unsaleable at the current moment.

[0129] As a feasible implementation method, the maximum allowable retention time determines the time threshold for judging the unsold status of the target type of vehicles (i.e., the longest acceptable retention time for vehicles from the time they enter the warehouse; if they are not sold after this time, an unsold status warning is triggered in the time dimension). The maximum allowable inventory level determines the inventory threshold for judging the unsold status of the target type of vehicles (i.e., the highest inventory limit that the warehouse can currently hold before reaching the unsold status; if this limit is exceeded, an unsold status warning is triggered in the inventory dimension). When the actual retention time of the target type of vehicles at the current time exceeds the maximum allowable retention time, and the current actual inventory level exceeds the maximum allowable inventory level, the results of the two dimensions are combined to determine that the target type of vehicles are unsold at the current time.

[0130] In an exemplary embodiment, if the actual retention time of the target type vehicle at the current moment is greater than the maximum allowable retention time, and the actual inventory of the target type vehicle at the current moment is greater than the maximum allowable inventory, it is determined that the target type vehicle is unsaleable at the current moment.

[0131] The actual retention time at the current moment is used to characterize the total time from when the target type vehicle enters the warehouse to the current moment.

[0132] The current inventory level is used to represent the total inventory level of the target type of vehicles at the current moment.

[0133] In another exemplary embodiment, if the actual retention time of the target type vehicle at the current moment is less than or equal to the maximum allowable retention time, or the actual inventory of the target type vehicle at the current moment is less than or equal to the maximum allowable inventory, it is determined that the target type vehicle is not unsold at the current moment.

[0134] For example, such as Figure 8 As shown, the target features of the input region-vehicle combination are used to obtain the maximum allowable retention time and the maximum allowable inventory. It is determined whether the actual retention time at the current moment is greater than the maximum allowable retention time. If not, the target type vehicle is directly output as a non-slow-moving vehicle. If so, it is determined whether the actual inventory at the current moment is greater than the maximum allowable inventory. If not, the target type vehicle is directly output as a non-slow-moving vehicle. If so, the target type vehicle is output as a slow-moving vehicle.

[0135] By encompassing multi-dimensional target features that influence the slow-moving status of target vehicle types, this system can accurately quantify the estimated sales duration and future estimated inventory levels of target vehicle types based on the inherent correlation between these features and sales duration and inventory changes. Furthermore, it adjusts the maximum allowable retention time in real-time based on the estimated sales duration and adjusts the maximum allowable inventory level in real-time based on the future estimated inventory levels. This ensures that the maximum allowable retention time and maximum allowable inventory level meet the actual market demand for target vehicle types. Through this dual verification of maximum allowable retention time and maximum allowable inventory level, it achieves accurate identification of slow-moving vehicles, providing data-driven decision support for production planning optimization. This enables precise control over the production scale of target vehicle types, effectively alleviating the contradiction between pre-production and market demand uncertainty, and reducing the risk of inventory backlog.

[0136] In some embodiments, the estimated sales duration can be determined using a sales duration prediction model, such as... Figure 9 As shown, this can be achieved through the following steps: S901. Input the target features into the first feature extraction layer of the sales duration prediction model to obtain the first key features.

[0137] The first feature extraction layer is used to extract features from the target features that are more important to the sales duration prediction task than a first preset importance threshold.

[0138] The first key feature refers to a subset of features that are strongly correlated with sales duration. Through specialized training and verification in the sales duration prediction task, the first key feature has accurately focused on the core factors affecting inventory sales duration. For example, the first key feature includes: the current regional sales growth / decline trend, the historical sales duration change pattern of the same model, the regional market supply and demand gap, the impact coefficient of seasonal factors on sales, and the rate of consumer preference iteration.

[0139] As a feasible implementation method, the first feature extraction layer first trains all target features using the XGBoost model, and uses the sum of information gain generated by each feature in the model split node as the importance quantification index; then sets a first preset importance threshold (such as the 70% quantile value of feature importance, which can be optimized by cross-validation based on the principle of minimizing the prediction error of the validation set), retains features with importance higher than the threshold, and removes redundant and weakly correlated features, and finally forms the first key feature.

[0140] Optionally, the first feature extraction layer may also introduce a feature interaction mechanism: after selecting high-importance features in a single dimension, feature interaction items (such as "current sales growth trend × seasonal influence coefficient" and "average historical sales duration × regional consumption capacity level") are constructed, and features that still meet the first preset importance threshold after interaction are retained through importance quantification again, further strengthening the representation ability of the first key feature on sales duration and improving the accuracy of subsequent predictions.

[0141] S902. Input the first key feature into the first prediction layer of the sales duration prediction model to obtain the estimated sales duration.

[0142] The first prediction layer is used to perform in-depth processing and pattern learning on the first key feature of the input to determine the estimated sales duration.

[0143] As a feasible implementation method, the time-series features (such as historical sales duration change trends and weekly sales fluctuation data) in the first key feature need to be processed by sliding window to extract time-series statistical features such as the average sales duration and sales fluctuation variance over the past 30 days. Then, the time-series statistical features are fused with the non-time-series features in the first key feature to form a complete input feature set, which is then input into the first prediction layer. The first prediction layer outputs the estimated sales duration based on the learned mapping relationship between features and sales duration.

[0144] In the exemplary embodiment, the time-series features (such as historical sales duration trends, weekly sales fluctuation data, and time-series fluctuations in regional market demand) in the first key features are processed using a sliding window (the sliding window size can be flexibly configured according to the business scenario, such as 7 days, 15 days, 30 days, etc.). Multi-dimensional time-series statistical features (including but not limited to the average sales duration of the past N days, sales fluctuation variance, month-on-month growth rate of sales duration, and cumulative growth rate of sales) are extracted through the sliding window to achieve in-depth mining and information fusion of time-series features. The fused time-series statistical features are integrated with the non-time-series core features in the first key features into a complete input feature set, which is input to the first prediction layer that has been trained. Based on the non-linear mapping relationship between the key features and sales duration learned during the training phase, and combined with the differences in scenarios such as the characteristics of different regional markets and the turnover patterns of vehicle models, the first prediction layer quickly outputs the estimated sales duration of the target type of vehicle.

[0145] In some embodiments, after determining the estimated sales duration, the maximum allowable retention time can be dynamically adjusted by the estimated sales duration, the upper limit of the sales duration, and the lower limit of the sales duration.

[0146] As a feasible implementation method, combined with Figure 7 ,like Figure 10 As shown, step S702 can be specifically implemented as follows (the following steps can be performed in parallel or sequentially, and this application embodiment does not limit this): S1001. If the estimated sales duration is greater than or equal to the lower limit of sales duration and less than the upper limit of sales duration, determine the duration difference between the reference duration and the estimated sales duration, and determine the maximum allowable retention duration based on the duration difference and the preset scaling factor.

[0147] The reference duration is used to represent the sum of the upper limit and the lower limit of the sales duration.

[0148] The sales duration limit is used to characterize the maximum acceptable sales cycle for a target type of vehicle under normal market conditions (i.e., the risk of unsold vehicles will increase significantly after exceeding this duration). For example, the sales duration limit can be 90.

[0149] The lower limit of sales duration is used to characterize the minimum sales cycle benchmark for a target type of vehicle under ideal market conditions (i.e., a duration lower than this indicates that vehicle circulation efficiency is far exceeding expectations). For example, the lower limit of sales duration can be 60.

[0150] The preset scaling factor is used to characterize the adjustment ratio of the duration difference to the maximum allowable retention time (the risk control intensity can be flexibly adapted through factor calibration). For example, the preset scaling factor can be 0.9.

[0151] As a feasible approach, if the estimated sales duration is greater than or equal to the lower limit of the sales duration but less than the upper limit, some regions may face the risk of unsold inventory due to the longer vehicle destocking cycle. Within the range of the lower limit of the sales market to the upper limit of the sales duration, the longer the estimated sales duration, the closer the vehicle circulation efficiency is to the risk threshold, and the higher the probability of unsold inventory. To accurately match risk levels and dynamically adjust control measures, a correlation logic can be constructed by introducing a duration difference (i.e., the difference between the reference duration and the estimated sales duration, where the reference duration is the sum of the upper and lower limits of the sales duration) and a preset scaling factor. This achieves a negative correlation adjustment where "the larger the estimated sales duration, the smaller the maximum allowable retention time." If the estimated sales duration is closer to the upper limit of the sales duration, the corresponding duration difference will be smaller, and the maximum allowable retention time calculated by the preset scaling factor will also be shorter. This allows for tightening the time judgment threshold in advance and strengthening the timeliness of early warning of unsold inventory risks. If the estimated sales duration is closer to the lower limit of the sales duration, the duration difference will be larger, and the maximum allowable retention time will be relatively longer. This can effectively avoid excessive control over vehicles within the normal circulation cycle and balance risk warning and inventory turnover efficiency.

[0152] In an exemplary embodiment, the product of the duration difference and the preset scaling factor is determined as the maximum allowed retention time.

[0153] S1002. If the estimated sales duration is greater than or equal to the upper limit of sales duration, determine the maximum allowable retention time as the product of the lower limit of sales duration and the preset scaling factor.

[0154] It should be understood that when the estimated sales duration is greater than or equal to the upper limit of the sales duration, it indicates that the destocking efficiency of the target type of vehicles is significantly low and the cycle is too long. If targeted intervention measures are not taken in time, it is very easy to cause long-term backlog of vehicles, resulting in a large number of inventory vehicles, which in turn increases the risks of capital occupation, storage losses, and vehicle depreciation. At this time, by forcibly setting the maximum allowable retention time as the product of the lower limit of the sales duration and the preset scaling factor, the time threshold for judging slow sales can be further tightened. Compared with the lower limit of the sales duration itself, the product of the lower limit of the sales duration and the preset scaling factor is shorter, which can give enterprises sufficient time for prevention and intervention in advance, and encourage them to start destocking strategies as early as possible (such as regional transfer, promotional discounts, and targeted marketing), effectively curbing the inventory backlog and reducing the operating losses caused by slow sales.

[0155] S1003. If the estimated sales duration is less than the lower limit of the sales duration, determine the maximum allowable retention duration as the upper limit of the sales duration.

[0156] It should be understood that when the estimated sales duration is less than the lower limit of the sales duration, it indicates that the market demand for the target type of vehicle is strong and the circulation efficiency is extremely high, with almost no risk of unsold inventory in the short term. In this case, setting the maximum allowable retention time as the upper limit of the sales duration can provide more flexible inventory retention space for vehicles. This avoids excessive intervention in the normal circulation of vehicles due to overly strict time thresholds, and can also adapt to short-term market fluctuations (such as periodic supply and demand imbalances, extended customer decision-making cycles, etc.), ensuring the flexibility of inventory turnover, while eliminating concerns about the accumulation of unsold inventory risks.

[0157] Specifically, assuming the upper limit of sales duration is 90 seconds, the lower limit of sales duration is 60 seconds, and the preset scaling factor is 0.9, the maximum allowable retention time can satisfy the following formula: ; in, Used to indicate the estimated sales duration; Used to indicate the maximum allowed retention time.

[0158] By comparing the estimated sales duration, the lower limit of the sales duration, and the upper limit of the sales duration, the maximum allowable retention time is determined. When the estimated sales duration is long, a shorter maximum allowable retention time can be set accordingly. By tightening the threshold for determining the unsold inventory time in advance, the timeliness of early warning of the risk of unsold inventory for the target type of vehicle is significantly improved, and sufficient intervention window is reserved for enterprises. When the estimated sales duration is short, a longer maximum allowable retention time can be set accordingly, giving vehicles more leeway in inventory retention. This avoids excessive control over vehicles with high circulation efficiency due to overly strict time thresholds, thus ensuring market supply stability and achieving a dynamic balance between unsold inventory risk control and inventory turnover efficiency.

[0159] In some embodiments, the estimated inventory levels at future time points can be determined based on an inventory level forecasting model, such as... Figure 11 As shown, this can be achieved through the following steps: S1101. Input the target features into the second feature extraction layer of the inventory prediction model to obtain the second key features.

[0160] The second feature extraction layer is used to extract features from the target features that are more important to the inventory estimation task than a second preset importance threshold.

[0161] The second key feature is used to characterize a subset of core features that are strongly correlated with changes in inventory levels at future times, focusing on key factors that affect inventory changes (such as historical inventory turnover data, regional inbound / outbound rhythm, cross-regional transfer plans, market demand forecasts, and capacity replenishment cycles).

[0162] As a feasible approach, all target features are first input into the XGBoost model, and the information gain or split contribution of features in the inventory prediction task is used as the quantitative indicator of importance. Then, a second preset importance threshold is set (such as the 70th percentile of feature importance, which can be optimized through cross-validation), features with importance higher than the threshold are retained, and redundant and weakly correlated features are eliminated, ultimately forming the second key feature.

[0163] Optionally, after XGBoost selects highly important single-dimensional features, feature interaction items such as "regional demand intensity × inbound frequency" and "transfer volume × inventory balance" are constructed, and effective interaction features are retained again through importance quantification. At the same time, variance inflation factor is used to detect and remove multicollinear features (such as features with variance inflation factor greater than 10) to ensure that the second key feature is both representative and independent, thereby improving the stability of subsequent predictions.

[0164] S1102. Input the second key feature into the second prediction layer of the inventory prediction model to obtain the estimated inventory level at future time.

[0165] The second prediction layer is used to perform time-series processing, pattern learning, and inventory quantity quantification and prediction on the second key feature of the input in order to determine the estimated inventory quantity at future times.

[0166] As the first feasible implementation method, the time-series features (such as historical daily average inventory, weekly outbound fluctuations, monthly transfer frequency, etc.) in the second key features are processed by sliding window aggregation (the sliding window size can be adapted according to the estimated period, such as 7 days or 15 days) to extract time-series statistical features such as mean, peak value, and variance of fluctuation. The time-series statistical features are then integrated with non-time-series core features (such as regional storage capacity, vehicle production capacity priority, policy impact coefficient, etc.) into a complete input set, which is then input into the trained second prediction layer. Based on the nonlinear mapping relationship between the learned features and the estimated inventory at future moments, the model outputs the estimated inventory of the target type of vehicle at future moments, adapting to the scenario of real-time inventory changes.

[0167] In some embodiments, after determining the estimated inventory level at future time, the inventory level change trend of the target type of vehicles can be determined by the estimated inventory level at future time, thereby determining the maximum allowable inventory level of the target type of vehicles at the current time.

[0168] As a feasible implementation method, such as Figure 12 As shown, step S703 above can be specifically implemented as follows: S1201. Determine the difference between the actual inventory of the target type of vehicle at the current moment and the estimated inventory at future moments.

[0169] It should be understood that the inventory difference can intuitively reflect the trend of vehicle inventory changes from the current moment to the future moment, and thus dynamically adjust the constraint of the judgment of slow-moving inventory: if the current actual inventory is greater than the estimated inventory at the future moment, the inventory difference is positive, indicating that the inventory is on a downward trend (the market absorption capacity is greater than the inventory accumulation rate). At this time, the constraint of the judgment of slow-moving inventory can be appropriately relaxed, and the maximum allowable inventory level can be increased to ensure the stability of market supply and avoid excessive control leading to shortages. If the current actual inventory is less than the estimated inventory at the future moment, the inventory difference is negative, indicating that the inventory is on an upward trend (the inventory accumulation rate is greater than the market absorption capacity). At this time, the constraint of the judgment of slow-moving inventory needs to be tightened, and the maximum allowable inventory level should be reduced to control the inventory scale from the source, provide early warning of the risk of backlog, and reduce the loss of slow-moving inventory.

[0170] S1202. Based on the lower limit of inventory, the benchmark inventory, and the inventory difference, determine the theoretical maximum allowable inventory of the target type of vehicle at the current moment.

[0171] The lower limit of inventory is used to characterize the minimum inventory threshold that the target type of vehicle needs to maintain to ensure normal market supply (i.e., the inventory cannot be lower than this value to avoid the risk of shortages and supply disruptions). For example, the lower limit of inventory can be 5.

[0172] The baseline inventory level is used to represent the acceptable, non-stagnant inventory limit for the target type of vehicles at the current moment when the inventory difference is zero (i.e., the current actual inventory level is equal to the estimated inventory level at a future time, and there is no significant upward or downward trend in inventory). For example, the baseline inventory level can be 10.

[0173] The theoretical maximum allowable inventory level is used to characterize the highest inventory threshold that can be tolerated at the current moment after dynamic adjustment based on inventory change trends (this threshold is the core calculation basis for the maximum allowable inventory level, and can be further calibrated in conjunction with scenario coefficients). Exceeding this threshold means that the risk of inventory backlog has increased significantly.

[0174] As a feasible implementation method, the calculation of the theoretical maximum allowable inventory is based on the benchmark inventory level, with the inventory difference as a dynamic adjustment factor and the lower limit of inventory as a safety net constraint. A quantitative formula is used to achieve precise linkage between inventory change trends and thresholds: when inventory is trending downward (inventory difference is positive), the threshold is appropriately relaxed by adjusting the coefficient upward; when inventory is trending upward (inventory difference is negative), the threshold is tightened by adjusting the coefficient downward; at the same time, the threshold is forced not to be lower than the lower limit of inventory to ensure supply stability.

[0175] For example, assuming the minimum inventory level is 5 and the baseline inventory level is 10, the theoretical maximum allowable inventory level can satisfy the following formula: ; in, Used to indicate actual inventory levels; Used to indicate the estimated inventory level at a future time.

[0176] S1203. The smaller of the theoretical maximum allowable inventory and the upper limit of inventory is determined as the maximum allowable inventory.

[0177] The inventory limit is used to characterize the absolute peak inventory that the target type of vehicles can handle at the current moment (i.e., exceeding this value will result in a serious risk of overstocking, regardless of inventory trends, and will necessitate triggering a slow-moving inventory warning). For example, the inventory limit can be 20.

[0178] As a feasible implementation method, the theoretical maximum allowable inventory level is dynamically adjusted based on inventory change trends to ensure the flexibility of the threshold; the inventory limit provides a rigid constraint to prevent extreme inventory backlog caused by excessively high thresholds due to theoretical calculations. Taking the smaller value of the two adapts to the dynamic management needs in normal scenarios while preventing the risk of excessive inventory in special circumstances, ensuring the rationality and safety of the maximum allowable inventory level.

[0179] For example, assuming the inventory limit is 20, the maximum allowable inventory can satisfy the following formula: ; in, Used to indicate the maximum allowable inventory level; Used to indicate actual inventory levels; Used to indicate the estimated inventory level at a future time.

[0180] Based on the estimated inventory levels at future moments, a dynamic quantitative model is constructed to adjust the maximum allowable inventory level in real time through benchmark inventory level calibration, inventory change trend correction (determined based on the inventory difference between the current actual inventory level and the future estimated inventory level), and rigid constraints on upper and lower limits of inventory levels. This allows the control threshold of the inventory dimension to accurately match the market supply and demand balance.

[0181] In some embodiments, such as Figure 13 As shown, step S1202 can be specifically implemented as follows: S1301. Determine the target ratio between the inventory quantity difference and the actual inventory quantity.

[0182] S1302. The product of the baseline inventory level and the target ratio is determined as the inventory level correction increment.

[0183] As a feasible implementation method, the inventory correction increment is dynamically adjusted according to the inventory change trend: when the current actual inventory is greater than the estimated inventory at a future time, the target ratio is positive and the correction increment is positive, which can be used as an upward supplement to the inventory threshold to adapt to the downward trend of inventory; when the current actual inventory is less than the estimated inventory at a future time, the target ratio is negative and the correction increment is negative, which can be used as a downward deduction to the inventory threshold to strengthen the control under the upward trend of inventory.

[0184] For example, the inventory adjustment increment can satisfy the following formula: ; in, Used to indicate actual inventory levels; Used to indicate the estimated inventory level at a future time.

[0185] S1303. The sum of the baseline inventory level and the inventory level correction increment is determined as the inventory level correction value.

[0186] As a feasible implementation method, the inventory level adjustment value is based on the benchmark inventory level, and the inventory level adjustment increment is adapted to the inventory change trend: when the inventory is declining (the inventory level adjustment increment is positive), the inventory level adjustment value is higher than the benchmark inventory level, and the inventory threshold is appropriately relaxed; when the inventory is rising (the inventory level adjustment increment is negative), the inventory level adjustment value is lower than the benchmark inventory level, and inventory control is tightened.

[0187] For example, the inventory level adjustment value can satisfy the following formula: ; in, Used to indicate actual inventory levels; Used to indicate the estimated inventory level at a future time.

[0188] S1304. The larger of the inventory level correction value and the inventory level lower limit is determined as the theoretical maximum allowable inventory level.

[0189] It should be understood that the inventory level adjustment value is the result of dynamic calibration based on the baseline inventory level and inventory change trend. It can adapt to the control needs under different inventory trends, but there may be cases where the adjustment value is too low due to a significant upward trend in inventory. When the inventory level adjustment value is higher than the lower limit of inventory level, the dynamically adjusted result is used as the theoretical maximum allowable inventory level to ensure that the threshold matches the inventory change trend. When the inventory level adjustment value is lower than the lower limit of inventory level, the lower limit of inventory level is forcibly used as the theoretical maximum allowable inventory level to avoid inventory falling below the minimum standard for ensuring normal market supply due to excessive tightening of the threshold. This effectively prevents the risk of stockouts and supply disruptions and achieves a balance between the control of unsold inventory risk and the stability of market supply.

[0190] In some embodiments, when the maximum allowable retention time and the maximum allowable inventory are determined, unsold vehicles are determined based on the actual retention time and the actual inventory of the target type of vehicle at the current moment.

[0191] As a feasible implementation method, step S704 can be specifically implemented as follows: if the actual retention time of the target type vehicle at the current moment is greater than the maximum allowable retention time, and the actual inventory of the target type vehicle at the current moment is greater than the maximum allowable inventory, then it is determined that the target type vehicle is unsaleable at the current moment.

[0192] It should be understood that, from a time perspective, if the actual retention time exceeds the maximum allowable retention time, it indicates that the vehicle inventory dwell period has exceeded the safe range, resulting in an imbalance in circulation efficiency. From an inventory perspective, if the actual inventory exceeds the maximum allowable inventory, it indicates that the inventory scale has exceeded the controllable threshold, posing a risk of overstocking. Only when both dimensions trigger the exceeding warning is the product deemed unsaleable. This avoids misjudging unsaleable simply because of short-term high inventory (but normal circulation efficiency), and also prevents excessive warnings due to slightly longer dwell time (but controllable inventory scale). Ultimately, this achieves a scientific definition of unsaleable status, providing a reliable basis for enterprises to accurately initiate destocking intervention measures.

[0193] In some embodiments, such as Figure 14 As shown, the method for determining unsold vehicles can be implemented through the following steps.

[0194] S1401, the sales duration prediction model determines the estimated sales duration based on target characteristics.

[0195] S1402. Dynamically adjust the maximum allowable retention time based on the estimated sales duration.

[0196] For example, the maximum allowed retention time can satisfy the following formula: .

[0197] S1403, the inventory forecasting model determines the estimated inventory level at future times based on target characteristics.

[0198] S1404. Obtain the actual inventory of the target type of vehicle at the current moment.

[0199] S1405. Based on the estimated inventory level and the actual inventory level at future moments, dynamically adjust the maximum allowable inventory level.

[0200] For example, the maximum allowable inventory level can satisfy the following formula: .

[0201] S1406. If the actual retention time of the target type of vehicle at the current moment is greater than the maximum allowable retention time, and the actual inventory of the target type of vehicle at the current moment is greater than the maximum allowable inventory, it is determined that the target type of vehicle is unsaleable at the current moment.

[0202] It should be understood that steps S1404, S1403, and S1404 can be processed in parallel; after steps S1402 and S1405 are executed, step S1406 is executed; after step S1403 is executed, step S1405 is executed.

[0203] Please see Figure 15 , Figure 15 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application. The electronic device 1500 may include a processor 1501 and a memory 1502. The processor 1501 and the memory 1502 are communicatively connected. The memory 1502 is used to store programs, and the processor 1501 is used to execute the programs, specifically performing the relevant steps in the above-described embodiment of the method for determining unsold vehicles.

[0204] Specifically, the program may include program code, which includes computer-executable instructions. Memory 1502 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device. Processor 1501 may be a central processing unit (CPU), a microcontroller unit (MCU), or an application-specific integrated circuit (ASIC).

[0205] This application also provides a computer-readable storage medium storing at least one executable instruction that, when executed on a vehicle unsold determination device, causes the vehicle unsold determination device to perform the vehicle unsold determination method in any of the above method embodiments.

[0206] This application provides a computer program product that can be executed by the processor 1501 of the electronic device 1500 to complete the method for determining unsold vehicles in the above embodiments.

[0207] It should be understood that the application of this application is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Those skilled in the art can understand that implementing all or part of the processes of the above embodiments and making equivalent changes according to the claims of this application still fall within the scope of this application.

Claims

1. A method for determining unsold vehicles, characterized in that, The methods for determining unsold vehicles include: Based on the target characteristics at the current moment, the estimated sales duration and estimated inventory level of the target type of vehicle are determined; the estimated sales duration is used to characterize the expected time from entering the inventory to final sale; the target characteristics are used to characterize the characteristics that have an impact on the unsold status of the target type of vehicle; the future time is the time after a preset time has elapsed from the current moment. Based on the estimated sales duration, determine the maximum allowable retention time of the target type of vehicle in the warehouse; Based on the estimated inventory levels at future times, determine the maximum permissible inventory level for the target type of vehicles at the current time. Based on the maximum allowable retention time and the maximum allowable inventory, determine whether the target type of vehicle is unsaleable at the current moment.

2. The method for determining unsold vehicles according to claim 1, characterized in that, The step of determining the maximum permissible retention time of the target type of vehicle in the warehouse based on the estimated sales duration includes: If the estimated sales duration is greater than or equal to the lower limit of sales duration and less than the upper limit of sales duration, a duration difference between a reference duration and the estimated sales duration is determined, and the maximum allowable retention duration is determined based on the duration difference and a preset scaling factor; the reference duration is used to represent the sum of the upper limit of sales duration and the lower limit of sales duration. If the estimated sales duration is greater than or equal to the upper limit of sales duration, the maximum allowable retention time is determined to be the product of the lower limit of sales duration and the preset scaling factor; If the estimated sales duration is less than the lower limit of the sales duration, the maximum allowable retention time is determined to be the upper limit of the sales duration.

3. The method for determining unsold vehicles according to claim 1, characterized in that, The step of determining the maximum allowable inventory of the target type of vehicles at the current time based on the estimated inventory at the future time includes: Determine the inventory difference between the actual inventory of the target type of vehicle at the current time and the estimated inventory at the future time; Based on the lower limit of inventory, the benchmark inventory, and the difference in inventory, determine the theoretical maximum allowable inventory of the target type of vehicle at the current moment; The smaller of the theoretical maximum allowable inventory and the upper limit of inventory is determined as the maximum allowable inventory.

4. The method for determining unsold vehicles according to claim 3, characterized in that, The determination of the theoretical maximum allowable inventory level for the target type of vehicle at the current moment, based on the lower limit of inventory level, the baseline inventory level, and the inventory level difference, includes: Determine the target ratio between the inventory quantity difference and the actual inventory quantity; The product of the baseline inventory level and the target ratio is determined as the inventory level correction increment; The sum of the baseline inventory level and the inventory level correction increment is determined as the inventory level correction value; The larger of the inventory level correction value and the inventory level lower limit is determined as the theoretical maximum allowable inventory level.

5. The method for determining unsold vehicles according to claim 1, characterized in that, Based on target characteristics, determine the estimated sales duration for the target type of vehicle, including: The target features are input into the first feature extraction layer of the sales duration prediction model to obtain the first key features; wherein, the first feature extraction layer is used to extract features from the target features that are more important to the sales duration prediction task than a first preset importance threshold; The first key feature is input into the first prediction layer of the sales duration prediction model to obtain the estimated sales duration.

6. The method for determining unsold vehicles according to claim 1, characterized in that, Based on target characteristics, determine the estimated future inventory levels of vehicles of the target type, including: The target features are input into the second feature extraction layer of the inventory prediction model to obtain the second key features; wherein, the second feature extraction layer is used to extract features from the target features that are more important to the inventory prediction task than a second preset importance threshold; The second key feature is input into the second prediction layer of the inventory prediction model to obtain the estimated inventory level at the future time.

7. The method for determining unsold vehicles according to claim 1, characterized in that, The step of determining whether the target type of vehicle is unsaleable at the current moment based on the maximum allowable retention time and the maximum allowable inventory includes: If the actual retention time of the target type of vehicle at the current moment is greater than the maximum allowable retention time, and the actual inventory of the target type of vehicle at the current moment is greater than the maximum allowable inventory, then the target type of vehicle is determined to be unsaleable at the current moment.

8. A training method for a vehicle slow-selling model, characterized in that, The training method for the vehicle slow-selling model includes: Construct a training dataset; the training dataset includes: sample target features of target type vehicles at a first time moment and sample labels corresponding to the sample target features; the sample labels are used to characterize whether the target type vehicles are unsaleable at the first time moment; the sample labels are determined based on the method for determining unsaleable vehicles as described in any one of claims 1-7; The initial vehicle sales stagnation model is trained based on the training dataset until it converges. The initial vehicle sales stagnation model includes a sales duration prediction network, an inventory prediction network, and a vehicle sales stagnation network. The sales duration prediction model extracts features from the sample target features that are more important to the sales duration prediction task than a first preset importance threshold, and uses these extracted features to determine the estimated sample sales duration of the target type of vehicle. The inventory prediction network extracts features from the sample target features that are more important to the inventory prediction task than a second preset importance threshold, and uses these extracted features to determine the estimated sample inventory of the target type of vehicle at a second time point. The vehicle sales stagnation network determines the probability of the target type of vehicle being unsold at the first time point based on the estimated sample sales duration and the estimated sample inventory, and optimizes the model parameters of the initial vehicle sales stagnation model using the difference between the unsold probability and the sample label. The second time point is the time after a preset duration has elapsed since the first time point. The converged initial vehicle sales stagnation model is determined as the vehicle sales stagnation model.

9. A method for determining unsold vehicles, characterized in that, The methods for determining unsold vehicles include: Obtain the target features at the current moment; the target features are used to characterize the features that affect the sluggish sales of the target type of vehicle; Input the target features into the vehicle sales stagnation model as described in claim 8 to obtain the sales stagnation probability of the target type of vehicle at the current moment; If the probability of unsold inventory is greater than or equal to a preset probability threshold, it is determined that the target type of vehicle is unsold at the current time.

10. A device for determining unsold vehicles, characterized in that, The device for determining unsold vehicles includes: a first determining module, a second determining module, a third determining module, and a fourth determining module; The first determining module is used to determine the estimated sales duration and estimated inventory level of the target type of vehicle based on the target characteristics at the current moment; the estimated sales duration is used to characterize the expected time from entering the inventory to final sale; the target characteristics are used to characterize the characteristics that have an impact on the unsold status of the target type of vehicle; the future time is the time after a preset time has elapsed from the current moment. The second determining module is used to determine the maximum allowable retention time of the target type of vehicle in the warehouse based on the estimated sales duration; The third determining module is used to determine the maximum allowable inventory of the target type of vehicle at the current time based on the estimated inventory at the future time. The fourth determining module is used to determine whether the target type of vehicle is unsaleable at the current moment based on the maximum allowable retention time and the maximum allowable inventory.

11. A training device for a vehicle slow-selling model, characterized in that, The training device for the vehicle slow-selling model includes: a construction module, a training module, and a fifth determination module; The construction module is used to construct a training dataset; the training dataset includes: sample target features of target type vehicles at a first moment and sample labels corresponding to the sample target features; the sample labels are used to characterize whether the target type vehicles are unsold at the first moment; the sample labels are determined based on the method for determining unsold vehicles as described in any one of claims 1-7; The training module trains the initial vehicle sales stagnation model based on the training dataset until the initial vehicle sales stagnation model converges. The initial vehicle sales stagnation model includes a sales duration prediction network, an inventory prediction network, and a vehicle sales stagnation network. The sales duration prediction network is used to extract features from the sample target features that are more important to the sales duration prediction task than a first preset importance threshold, and uses the extracted features to determine the estimated sample sales duration of the target type of vehicle. The inventory prediction network is used to extract features from the sample target features that are more important to the inventory prediction task than a second preset importance threshold, and uses the extracted features to determine the estimated sample inventory of the target type of vehicle at a second time point. The vehicle sales stagnation network is used to determine the sales stagnation probability of the target type of vehicle at the first time point based on the estimated sample sales duration and the estimated sample inventory, and to optimize the model parameters of the initial vehicle sales stagnation model using the difference between the sales stagnation probability and the sample label. The second time point is the time after a preset duration has elapsed since the first time point. The fifth determining module is used to determine the converged initial vehicle unsold model as the vehicle unsold model.

12. A device for determining unsold vehicles, characterized in that, The device for determining unsold vehicles includes: an acquisition module, an input module, and a sixth determination module; The acquisition module is used to acquire the target features at the current moment; the target features are used to characterize the features that have an impact on the sluggish sales of the target type of vehicle; The input module is used to input the target features into the vehicle sales slump model as described in claim 8 to obtain the sales slump probability of the target type of vehicle at the current time. The sixth determining module is used to determine that the target type of vehicle is unsold at the current time if the unsold probability is greater than or equal to a preset probability threshold.

Images