Vehicle periphery commodity recommendation method based on Internet of Things

Abstract

The invention relates to the field of Internet of Things technology and commercial data processing, and discloses an Internet of Things-based vehicle periphery commodity recommendation method, which comprises the following steps of: obtaining a future trajectory of a vehicle containing time arrival probability distribution and supply information of a merchant, and calculating an expected probability of time-space encounter of the future trajectory and the merchant to generate an expected encounter event, according to the method, a bidirectional information guiding mode based on future event prediction is established, so that an instant query structure which is only initiated by a single party after a demand is generated is changed; in order to deal with the supply and demand mismatching problem in the mobile environment, a dynamic robustness maintenance mechanism contained in the method maintains the continuity and reliability of the service when the vehicle driving path is uncertain.

Description

TECHNICAL FIELD

[0001] The application relates to a commodity recommendation method for a vehicle periphery based on an Internet of Things, and belongs to the fields of Internet of Things technology and commercial data processing. BACKGROUND

[0002] In current Internet of Things applications, providing location-based service recommendations for vehicles in motion is a common technical approach, which pushes nearby commercial information by obtaining the instant geographical position of the vehicle, thus meeting the instant needs of drivers. However, when this single-vehicle perspective and instant response approach is placed in a large-scale, high-dynamic real traffic environment, especially in scenarios such as highway service areas or popular tourist destinations that exhibit obvious passenger flow tidal effects, the design limitations result in a future information barrier between the supply side and the demand side. On the one hand, drivers are unaware of the real-time carrying capacity and potential queuing situation of the service point when approaching the service point, and the consumption decision is uncertain. On the other hand, service providers cannot predict the size of the incoming passenger flow and their demand preferences in the future, and can only make experience-based inventory and passive service, which directly leads to the mismatch of service resources and the loss of business opportunities.

[0003] To improve this situation, the industry has attempted to predict demand by increasing data interaction frequency or building large-scale centralized platforms. However, the former only speeds up the update of instant information and does not touch on the prediction of future states, while the latter introduces high system complexity, computation cost, and the risk of user journey privacy leakage. Therefore, these conventional improvement ideas do not provide a universally applicable solution.

[0004] Specifically, the existing technology mainly has the following deficiencies: 1. Lack of prediction mechanism for the future spatiotemporal state of vehicles, resulting in a lagging and responsive recommendation behavior that cannot guide supply and demand in advance; 2. Information interaction between the supply side and the demand side is one-way and static, and cannot reflect real-time changes in service capacity or form a closed-loop collaborative adjustment; 3. The individual future trajectories of a large number of vehicles cannot be converted into group prediction data that can be used for macro situational awareness, so that service point congestion caused by demand convergence cannot be foreseen and avoided. Therefore, how to establish a lightweight prediction mechanism that effectively associates the future trajectories of individual vehicles with the dynamic service capacity of businesses while protecting user privacy, achieve bidirectional and advance guidance of information between the supply and demand sides, and ensure that the mechanism can still operate reliably under various uncertain conditions of real driving, has become a technical problem to be solved by those skilled in the art. SUMMARY

[0005] The application provides a vehicle surrounding commodity recommendation method based on Internet of Things, which mainly aims to solve the problem that the future space-time state of a vehicle cannot be predicted in the prior art, resulting in one-way and static supply-demand information, difficulty in realizing pre-positioned collaborative guidance, and lack of robustness under real driving conditions.

[0006] To achieve the above-mentioned purpose, the application provides a vehicle surrounding commodity recommendation method based on Internet of Things, which comprises the following steps: Step a: obtaining a future trajectory line of a vehicle based on a navigation destination of the vehicle, the future trajectory line being a data structure containing a series of future space-time points and a time arrival probability distribution corresponding to the future space-time points; Step b: obtaining supply information published by a merchant, the supply information containing service content and an effective space-time window; Step c: calculating an expected probability of an encounter between the future trajectory line of the vehicle and the supply information of the merchant in space-time, to generate at least one expected encounter event; Step d: pushing guidance information to the vehicle or the merchant based on the probability of the expected encounter event; wherein the method further comprises a dynamic robustness maintenance mechanism, which performs the following operations: continuously comparing a real-time position of the vehicle with the future trajectory line, and determining that a trajectory deviation event occurs when a lateral deviation distance between the real-time position of the vehicle and the future trajectory line exceeds a reference threshold; in response to the trajectory deviation event, discarding all calculation results of the expected encounter events based on the future trajectory line; and activating an implicit trajectory deduction mode, which generates a probability vector field representing a short-term driving direction probability of the vehicle based on historical gravity points determined by clustering analysis of local historical trajectory data of the vehicle, and selects a historical gravity point with the highest weight in combination with a current time and date context, to maintain the commodity recommendation service by using the probability vector field.

[0007] Preferably, the dynamic robustness maintenance mechanism further performs the following operations: after activating the implicit trajectory deduction mode, continuously monitoring whether there is a re-anchoring event of the trajectory of the vehicle and a new or original navigation route, and in response to the re-anchoring event, automatically resuming the execution of steps a to d.

[0008] Preferably, after step d, the method further comprises: aggregating a plurality of future trajectory lines of vehicles pointing to the same merchant and overlapping in time window, to calculate an expected encounter flow density representing a future passenger flow density; comparing the expected encounter flow density with a service capacity of the merchant, to determine an encounter quality level; and adjusting the guidance information pushed to the vehicle in step d in content based on the encounter quality level.

[0009] Preferably, the time arrival probability distribution in step a is dynamically calculated and updated using a normal distribution model based on the current speed of the vehicle and real-time road condition information.

[0010] Preferably, the calculation of the expected probability in step c is to calculate the probability of each discrete time and space point on the future trajectory line falling within the merchant's effective time and space window, and perform a weighted sum of the probabilities of each point.

[0011] Preferably, the guidance information pushed to the vehicle in step d is an opportunity reminder including an expected encounter probability and a reservation or navigation detour suggestion operation.

[0012] Preferably, the guidance information pushed to the merchant in step d is a future passenger flow density heat map generated based on the expected encounter probability of multiple vehicles.

[0013] Preferably, the method also includes a supply-side service capacity dynamic modulation mechanism, which performs the following operations: when the expected encounter event is confirmed as a real transaction, an encounter confirmation signal associated with the real transaction is received; based on the encounter confirmation signal, the service capacity count value in the merchant's corresponding supply information is deducted in real time; and based on the deducted service capacity count value, subsequent expected probability calculation is performed.

[0014] Preferably, the guidance information pushed to the vehicle in step d is a binary timing opportunity including at least two alternative encounter events that differ in their core value propositions; the method also includes: receiving a behavioral feedback signal generated by the user's selection behavior for the binary timing opportunity; based on the accumulated behavioral feedback signal, constructing a personalized portrait for the user including time sensitivity and price sensitivity weight factors; and using the weight factors in the personalized portrait to sort or filter the subsequently generated expected encounter events.

[0015] Preferably, the calculation of the expected encounter traffic density follows the following rules: ,in, is the expected encounter traffic density, For the time window The total number of vehicles whose future trajectory lines point to the merchant, For the The expected encounter probability between a vehicle and a merchant.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. By acquiring the future trajectory line generated by the vehicle based on the navigation destination, and combining the effective space-time window supply information published by the merchant, the expected probability of the future space-time encounter between the vehicle and the merchant is calculated before the physical contact between the two, based on which the guiding information is pushed to the vehicle and the merchant, a bidirectional information guiding method based on future event prediction is established, and the information interaction structure of instant query by one party after demand is generated is changed.

[0017] 2. The dynamic robustness maintenance mechanism included in the method first discards the calculation result based on the old trajectory line when determining that the vehicle has deviated from the trajectory, and then activates the implicit trajectory deduction mode based on the local historical trajectory data of the vehicle. This design combines adaptability to the purposeless state, and maintains the continuity of the commodity recommendation service in two different scenarios where the core input of the main scheme is missing or invalid. Through monitoring of the trajectory redefinition event, the method switches between different deduction modes, forming a running logic that covers multiple driving states.

[0018] 3. The present application couples multiple technical mechanisms to form a multi-dimensional information verification and adjustment loop. On the one hand, by aggregating multiple future trajectory lines of vehicles pointing to the same merchant, the expected encounter flow density is calculated to predict possible micro-congestion in the future, and the content of the guiding information is adjusted accordingly. On the other hand, by receiving encounter confirmation signals after real transactions, the service capacity count value in the merchant supply information is updated in real time. The final recommendation decision is subject to the dual constraints of congestion evaluation based on group prediction and real-time service capacity based on transaction feedback, making the real feasibility of the recommendation result evident. BRIEF DESCRIPTION OF DRAWINGS

[0019] Fig. 1 is the interaction flow timing diagram of the vehicle surrounding commodity recommendation method based on the Internet of Things of the present application; Fig. 2 is the core logic function block diagram of the vehicle surrounding commodity recommendation method based on the Internet of Things of the present application; Fig. 3 is the function interaction use case diagram of the vehicle surrounding commodity recommendation system based on the Internet of Things of the present application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described in detail below. Obviously, the described embodiments are part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0021] The embodiment of the application provides a vehicle surrounding commodity recommendation method based on Internet of Things, which is constructed on a system comprising a vehicle edge computing node, a merchant information publishing terminal and a cloud or regional aggregation node, and data flow conversion of the method starts from probabilistic prediction of a future state of a vehicle, and via spatio-temporal calculation of supply and demand information, pre-posed information guidance to both supply and demand sides is completed; specifically, the method first acquires a future trajectory line generated based on a navigation destination on the vehicle side, and acquires supply information comprising dynamic service capability from the merchant side, then performs spatio-temporal dimension calculation on the cloud or regional aggregation node to generate an expected encounter event representing a possibility of future encounter, and pushes guidance information to the vehicle or the merchant according to the probability of the event, and the method further integrates a dynamic robustness maintenance mechanism to cope with the situation that the vehicle deviates from the intended navigation route during driving, so as to maintain the continuity of service in a real driving environment; in a specific application scenario, for example, a car is driving along a highway, and the driver has set a final destination through a vehicle-mounted navigation system, and the supply capability of commercial services such as catering and refueling in the service area along the highway and the demand of vehicles entering the service area often mismatch due to lack of prediction of future passenger flow; to cope with the imbalance between supply and demand caused by future information opacity, the method provided by the application is configured to perform the following steps.

[0022] Step a, acquiring a future trajectory line of a vehicle based on a navigation destination of the vehicle, the future trajectory line being a data structure comprising a series of future spatio-temporal points and a time arrival probability distribution corresponding to the future spatio-temporal points; specifically, the vehicle-mounted edge computing node uses the final destination and the planned path provided by the navigation system as basic input, generates a series of discrete spatio-temporal point sequences (t+1, loc+1), (t+2, loc+2) along the path within a future period of time, for example, 30 minutes, and for each spatio-temporal point, a normal distribution model is used to represent the probability of arrival time of the vehicle to quantify the uncertainty of actual arrival time of the vehicle, that is, a dynamically adjusted time arrival probability distribution is calculated for each point The expectation of the distribution is calculated based on the current speed, the remaining distance of the navigation path and real-time road condition information, and the variance The expectation of the distribution is calculated based on the current speed, the remaining distance of the navigation path and real-time road condition information, and the variance The fluctuation degree reflecting the current road condition, and the method predicts the future space-time state of the vehicle by this way; step b, obtaining the supply information published by the merchant, including service content and effective space-time window; the supply information of the merchant is designed as a data structure including core service content, real-time service capacity and service effective space-time window; the merchant publishes its service information through an application deployed on a smart phone or a special terminal, for example, a coffee shop in a service area publishes a latte, which is effective within 2 hours in the future, and the service capacity is 5 cups per minute, this information is packaged as a supply vector, which is broadcast to the regional server through 4G or 5G wireless communication network, and the supply vector contains digital service resource information which can be dynamically calculated and dispatched by the system.

[0023] The expected probability of the future trajectory of the vehicle and the supply information of the merchant is calculated to generate at least one expected encounter event; in order to realize this calculation, the system simplifies the integral process of the space-time encounter probability of the two to a discrete probability multiplication accumulation, and the specific procedure is to calculate the probability of each discrete space-time point on the future trajectory falling into the effective space-time window of the merchant, if the time arrival probability distribution of a future space-time point is , and the effective space-time window of the merchant supply information is and the geographical range , the probability of the point falling into the effective encounter is and its geographical position falls into , then the system will weight and sum up all the future points on the trajectory , so as to obtain a total expected encounter probability , when the probability value exceeds a preset trigger threshold, for example, 75%, the calculation finally generates an expected encounter event containing the vehicle, the merchant and the probability information, which is used for subsequent guiding decision; step d, based on the probability of the expected encounter event, the guiding information is pushed to the vehicle or the merchant; for the vehicle side, the guiding information is an opportunity reminder, which can include the estimated encounter probability and the suggested operation, for example, the A brand coffee in the front 15 kilometers service area has 85% probability without queuing, whether to book, this interaction is realized through the vehicle-mounted central control screen or mobile device; for the merchant, the system aggregates the expected encounter probability of multiple vehicles to generate a future passenger flow density heat map in the form of anonymous statistics, and pushes it to the merchant terminal, for example, it is expected that within 1 hour in the future, about 120 vehicles with coffee consumption preference will pass through, and it is suggested to adjust the stock, this two-way information interaction method is used to realize the information cooperation between supply and demand based on future prediction.

[0024] To cope with situations where the driver deviates from the established navigation route during real driving, resulting in the prediction results based on the original trajectory being no longer applicable, the method also includes a dynamic robustness maintenance mechanism; the mechanism performs the following operations: first, the vehicle's real-time position is continuously compared with the future trajectory. The comparison procedure is determined as follows: the on-board edge node compares the vehicle's real-time GPS position with the geometric path of the future trajectory at a frequency of 1Hz. When the vehicle's lateral offset distance continuously exceeds a benchmark threshold (e.g., 50 meters) set to filter out signal drift or normal lane changes for a specific duration (e.g., 3 seconds), the system determines that a trajectory deviation event has occurred; in response to the trajectory deviation event, to avoid guidance based on an invalid trajectory, the system performs a fuse operation, that is, immediately discards all calculation results of expected encounter events based on the future trajectory; then, to maintain service continuity, the system activates an implicit trajectory deduction mode, which is used to probabilistically deduce the vehicle's short-term driving intentions when the vehicle is in a state without a clear navigation destination. The specific procedure is as follows: when the vehicle is not in motion, the edge computing node performs density clustering analysis on the locally stored historical trajectory data to automatically identify the user's historical gravity points. This process is completed locally to protect user privacy. During driving, the mode combines the current time and date context, such as 8 a.m. on weekdays, selects the historical gravity point with the highest weight, such as assigning the company a gravity weight of 0.9, and generates a probability vector field representing the probability of the vehicle's short-term driving direction based on the vehicle's current position and the gravity point. The probability vector field is then used as a logically equivalent future trajectory input to maintain the product recommendation service; the dynamic robustness maintenance mechanism also continuously monitors whether there is a re-anchoring event in which the vehicle trajectory is re-anchored to a new or original navigation route. Once responding to this re-anchoring event, the system will automatically exit the implicit trajectory deduction mode and resume executing the process from step a to step d. Through this logical closed loop of deviation capture-fuse abandonment-mode switching-re-anchoring recovery, this method can adapt to route changes in real driving environments.

[0025] Meanwhile, in order to effectively prevent the privacy risk of reversing the user identity through the historical gravity point, the following multi-level privacy protection principles are followed in the design and operation of the implicit trajectory deduction mode, that is, the localization principle of data processing: such as the storage, clustering analysis and identification of historical gravity points of vehicle historical trajectory data, which are completely closed-loop completed on the edge computing node of the vehicle. The original trajectory data (Raw GPS Data) containing the accurate geographic position of the user is always left in the local and will not be uploaded to any cloud server, which physically isolates the risk of leakage of original sensitive data; anonymization and labeling of gravity point data: after identifying the historical gravity points (such as home address, company address) locally, the system does not directly use the real geographic coordinate information of the gravity points. Instead, it is anonymized, for example, the home address is abstracted as internal code gravity point A, and the company address is abstracted as gravity point B. In all subsequent calculations and weight distribution, these anonymous labels that do not contain any geographic information are used; probabilistic and abstract output results: when interacting with external systems, the vehicle terminal uploads not the specific gravity point coordinates or their anonymous labels, but the probability vector field generated according to the gravity weight of the gravity points, which represents the direction of the vehicle's short-term travel. The vector field is a pure mathematical data that only represents the probability distribution of the travel direction, for example, in the next 5 minutes, the probability of traveling north is 70%, and the probability of traveling east is 20%. The outside world (including the recommendation engine) cannot reverse any specific and sensitive geographic endpoint from this abstract vector field; user authorization and controllable design: this function is designed as an optional (Opt-in) mode, which will explicitly inform the user of its working principle and data processing method before first use and obtain the user's authorization. At the same time, the system provides a visual management interface for the user, who can view, delete or manually correct the historical gravity points identified by the system at any time, ensuring that the user has the final control over his personal data, which belongs to the extension implementation mode known to those skilled in the art.

[0026] In some embodiments, in order to improve the accuracy and practicality of the recommendation, the following steps are further included: in order to predict and avoid the service point congestion caused by demand convergence, before step d, the future trajectory lines of multiple vehicles pointing to the same merchant and overlapping in the time window are aggregated to calculate an expected encounter traffic density representing the future passenger flow density , which follows the rules of: , wherein, is the total number of vehicles whose future trajectory lines all point to the merchant within the time window is the ​the expected encounter probability, the system compares the calculated expected encounter traffic density with the service capacity in the merchant supply information to determine an encounter quality level, such as smooth, saturated, or congestion warning, and adjusts the content of the guidance information pushed to the vehicle based on the level; to maintain the real-time accuracy of the service capacity in the merchant supply information, the method further includes a supply-side service capacity dynamic modulation mechanism, when an expected encounter event is confirmed as a real transaction, the merchant system sends an encounter confirmation signal to the cloud, and the cloud deducts the service capacity count value in the supply information corresponding to the merchant in real time based on the signal, and performs subsequent expected probability calculation based on the deducted service capacity, which constitutes a closed-loop processing loop based on feedback adjustment of real transaction consumption; in addition, to explore the user's business intent, the guidance information pushed to the vehicle can be designed to include a binary timing opportunity of at least two alternative encounter events that differ in core value proposition, the system receives the behavior feedback signal generated by the user's selection behavior on the binary timing opportunity, and based on the accumulated signal, constructs a personalized portrait containing time sensitivity and price sensitivity weight factors for the user locally, and subsequently sorts or filters the events using the weight factors in the portrait when generating expected encounter events.

[0027] Embodiment 1: In a highway scene during a holiday return peak period, a vehicle A set with a navigation destination travels to a position about 50 kilometers away from a large service area X, which is the last major stop before entering the city ahead, according to historical traffic data, the service area often has problems of entrance congestion and long queuing time for internal commercial facilities during this period; at this time, the driver of vehicle A is faced with a parking decision dilemma, while the operator of coffee shop S in the service area cannot predict the passenger flow size in the next hour and can only make experience-based inventory and passive service; in this scenario, the method disclosed in the present application generates a future trajectory line containing a series of spatiotemporal points and their corresponding time arrival probability distribution based on the navigation destination of vehicle A at the edge computing node of vehicle A, at the same time, coffee shop S has published supply information containing effective service capacity for the next 2 hours, which is 5 customers per minute; after the cloud node of the system receives the future trajectory line of vehicle A and the supply information of coffee shop S, it calculates that the expected encounter probability of vehicle A and coffee shop S is 92%, and the generation of this calculation result triggers the system to evaluate the macro passenger flow state of service area X.

[0028] Specifically, the cloud node aggregates the expected encounter probability data of other vehicles that also point to service area X and arrive within a similar time window, and calculates the expected encounter traffic density according to the expected encounter traffic density calculation rule , it is concluded that the expected encounter traffic density of the coffee shop S in the period when vehicle A is expected to arrive is 8 customers per minute, which has exceeded its service capacity of 5 customers per minute; accordingly, the system determines the encounter quality level of this expected encounter event as a congestion warning; correspondingly, the guidance information pushed by the system to vehicle A is configured as a binary timing opportunity containing two differentiated options: option one, prompting the coffee shop S in service area X, expected to arrive, but there is a queue, and the estimated waiting time is about 20 minutes, option two, based on the synchronous operation of the alternative path, prompting the coffee shop T in service area Y along the current route to continue driving for 18 kilometers, with an expected encounter probability of 88%, and there is no queue, but the total journey will increase by about 10 minutes; the driver of vehicle A chooses option two according to the trade-off of time cost; finally, the navigation path of vehicle A is automatically updated to service area Y, and the driver obtains the service without waiting in service area Y, and the coffee shop S in service area X also pre-configures personnel and materials to cope with the upcoming passenger flow peak through the system and based on the future passenger flow density heat map of all potential passenger flows including vehicle A.

[0029] Example 2: To objectively verify the actual effectiveness of the technical solution of the present application in dealing with the dynamic supply-demand mismatch problem in a mobile environment, this example constructs a simulation test environment based on multi-agent microscopic traffic flow; the test environment simulates a two-way highway with a total length of 200 kilometers, with three service areas arranged thereon, of which the service area X located at 150 kilometers is set as the core commercial node, and the service capacity of the coffee shop S inside is set as 5 units per minute; the total test time is set to 8 hours to cover at least one complete passenger flow tidal period, i.e. including flat peak, peak and sub-peak periods, a total of 5000 simulated vehicles with randomly generated navigation destinations and service preferences, such as potential demand for coffee, are injected; the test sets up two groups for comparison: the control group, whose recommended services use the traditional method based on the real-time geographic location, i.e. when the vehicle enters the 5-kilometer range circle of the service area X, it pushes the service information to it; the test group, whose vehicles and business nodes completely follow the complete technical solution disclosed by the present application, including future trajectory prediction, expected encounter traffic density calculation and dynamic guidance; the two groups are run under the same traffic flow generation model, road network topology and business service capacity setting to exclude the interference of irrelevant variables; during the test, the system records and statistics the key performance indicators such as vehicle queue length, average waiting time, transaction success rate and service resource utilization rate of each service area at a period of 1 second.

[0030] At the 3rd hour of the test, when the first peak of return traffic is coming, the simulation platform observes a large number of vehicles with coffee demand in the control group arriving at the 5-kilometer recommended trigger circle of service area X at a similar time and receiving recommended information, resulting in a high convergence of vehicle decisions in a short time, and long queues are quickly formed at the entrance ramp of service area X and the door of coffee shop S. In the test group, the system calculates the expected encounter probability of each vehicle pointing to coffee shop S in advance, and aggregates to generate the expected encounter flow density. When the density exceeds the service capacity of coffee shop S, the system dynamically adjusts the guidance information for subsequent vehicles, such as providing differentiated opportunity options to the next service area or containing a waiting time warning. A part of vehicles choose to diverge according to the information, so that the peak of traffic flow to service area X is inhibited and smoothed.

[0031] In the middle of the peak period, the sampling data of the key performance indicators of the two groups are compared, and the results show that there are clear differences in the running efficiency of the two groups. In terms of service experience, the average waiting time of the control group is 22.1 minutes, while the average waiting time of the test group is 4.1 minutes. Correspondingly, the transaction abandonment rate caused by long queuing time is 41.5% in the control group and 5.8% in the test group. In terms of business efficiency, the proportion of successful transactions in the test group through guidance information is 89.7%, while that in the control group is 45.2%. The generation of these data differences is due to the technical solution adopted by the test group. By calculating the expected encounter flow density, the prediction ability of future congestion is brought forward to the recommended decision-making link of individual vehicles, so as to change the disordered individual decision-making behavior into group behavior coordinated and adjusted at the system level. In addition, the test data also records that the idle time of the service capacity of coffee shop S in the test group accounts for 11.3%, while that in the control group accounts for 2.3%. The reason for this phenomenon is that the system of the test group actively guides a part of the expected flow exceeding the carrying capacity of service area X to other service nodes at the cost of sacrificing the limit throughput capacity of a single node, in order to obtain the system adjustment behavior of stable overall operation of the road network service system.

[0032] Embodiment 3: This embodiment combines Figs. 1 to 3 to realize and describe a commodity recommendation method around a vehicle based on the Internet of Things, like Fig. 1As shown, the flow involves four core entities of vehicle, IoT platform, merchant and recommendation engine, which starts with the vehicle uploading information containing navigation destination to the IoT platform, and the recommendation engine generates future trajectory line containing a series of future spatiotemporal points and corresponding time arrival probability distribution based on this; at the same time, the merchant publishes supply information containing service content and effective spatiotemporal window through the IoT platform, which is transmitted to the recommendation engine; the recommendation engine performs matching calculation in spatiotemporal dimension based on the received vehicle future trajectory line and merchant supply information, to solve the expected probability of encounter between the two and generate expected encounter event; finally, according to the probability of the expected encounter event, the system pushes information to the vehicle and the merchant through the IoT platform, and the vehicle is pushed with opportunity reminder containing reservation or navigation detour suggestion, while the merchant is pushed with future passenger flow density heat map for passenger flow prediction.

[0033] As shown in Fig. 2 The figure starts with two parallel initial data input steps of obtaining vehicle future trajectory line and obtaining merchant supply information, the output of the former is a data structure containing time arrival probability distribution, and the output of the latter is data containing service content and effective spatiotemporal window; the two data streams converge into the core calculation spatiotemporal encounter expected probability function module, which aims to perform spatiotemporal dimension calculation on the information of supply and demand, and output one or more expected encounter events; then, the system enters the decision-making link of pushing guiding information, which makes decisions based on the probability of the generated expected encounter event, and distributes guiding information to two target objects, one is to push opportunity reminder to the vehicle, which contains specific reservation or navigation detour suggestion, and the other is to push passenger flow heat map to the merchant, which aims to present future passenger flow density to assist the merchant in making inventory decisions.

[0034] As shown in Fig. 3 The figure shows the core functions provided by the car surrounding goods recommendation system from the perspective of user interaction, and clearly defines two types of external interaction entities of vehicle / driver and merchant; for the vehicle / driver, it can interact with the system to receive personalized opportunity reminder and obtain dynamic path suggestion; for the merchant, it can publish supply information, receive future passenger flow warning and update real-time service capacity through the system, and the figure clearly defines the service boundary and core value provided by the system for users of supply and demand.

[0035] Example 4: In order to make the method of the present application consistent in deployment on different hardware platforms, the dynamic robustness maintenance mechanism and the implicit trajectory inference mode need to be calibrated and parameter configured; the configuration process includes setting a method based on objective data for determining the reference threshold of trajectory deviation event and the activation rules of the implicit trajectory inference mode; the calibration process first determines the reference threshold in the dynamic robustness maintenance mechanism, the setting of this threshold is to balance the sensitivity of identifying true deviation intention and the tolerance of normal driving behavior and signal noise, the calibration process collects GPS trajectory data of a specific vehicle model for more than 30 minutes of normal driving, merging and turning on various roads, by analyzing the data set, the maximum lateral error of the vehicle caused by signal drift in straight driving state , and the maximum lateral displacement when performing standard merging operation ; in order to avoid misjudging the above two cases as trajectory deviation events, the value of the reference threshold is calculated according to the rule , wherein is a safety factor greater than 1, such as 1.5, to provide a buffer zone, when the measured of a certain vehicle model is 5 meters, is 3.5 meters, then its is configured as 7.5 meters, the setting of this threshold has a quantifiable calculation basis.

[0036] Subsequently, the implicit trajectory inference mode is configured, which converts the user's historical driving habits into a calculable short-term intention prediction model; the configuration process first off-line mines historical attractors, this process uses a density clustering algorithm acting on a vehicle local historical trajectory data set containing at least 90 days, in this algorithm, the neighborhood radius and the minimum sample size parameters are set to 200 meters and 5, respectively, to aggregate parking points in the same geographic area and filter out parking point clusters with certain access frequency as candidate historical attractors; after the historical attractors are determined, the system further establishes a context-based weight activation rule library, the rule library is composed of a series of IF-THEN logic, wherein a rule is defined as: IF the current time is between 7am and 9am on weekdays, THEN set the weight factor of the historical attractor labeled as company to 0.9, set the weight factor of the home to 0.1, and the weight of all other attractors is 0; after the above calibration and configuration procedures, the dynamic robustness maintenance function and the implicit trajectory inference function of the method of the present application rely on the key parameters and logical rules, which have reproducible engineering setting basis when deployed on specific vehicle platforms.

[0037] Example 5: When the method of the present application is deployed, its personalized profiling mechanism needs to be initially configured and the iterative logic needs to be set to handle the initial recommendation for new users and the subsequent preference adaptation. This configuration process is used to generate the basic preference weights for the new user profile. In the offline stage, the system performs clustering analysis on the historical behavior feedback signals from the anonymous user group to identify several user preference profile prototypes, such as time priority type, price sensitivity type, and balanced type, and calculates the corresponding average time sensitivity and price sensitivity weight factors for each prototype. When a new user first enables the method, the system sets the initial weight factor of its personalized profile to the value of the balanced profile prototype, which serves as the baseline starting point for its personalized iteration.

[0038] In the process of user interaction with the system, the weight factor of the personalized profile follows a clear iterative update procedure. When the user selects the efficiency-oriented option in a binary timing opportunity, the time sensitivity weight is updated once, while the price sensitivity weight is updated once, and vice versa; the weight update follows the exponential moving average algorithm, and the update rule is where, is the updated weight, is the weight before updating, is the target value corresponding to this selection, and when the efficiency option is selected, the time sensitivity target value is 1 and the price sensitivity target value is 0, while is a fixed learning rate, whose value is used to control the impact of a single behavior on the overall profile; through the above initial value setting and iterative update procedure of the personalized profile, the recommendation function in the method of the present application has the ability to provide initial recommendations based on group statistics for new users and to adjust the personalized profile according to the behavior feedback of existing users.

[0039] Meanwhile, to verify the actual effectiveness of the personalized portrait mechanism, the application also conducts a special simulation test to quantify the recommendation accuracy before and after the optimization of the portrait. The test generates 1000 simulated driver users with different real preferences such as time priority, price sensitivity, and balance, and randomly divides them into two groups: one group is the control group using the fixed initial portrait, and the other group is the test group following the weight updating procedure described in Embodiment 5. During the simulation test period, the system pushes 20 binary time selection opportunities to all users, and records the number of accurate recommendations when the system's preferred recommendation is consistent with the user's real choice. The test results show that the average recommendation accuracy of the control group is stable at 51.2% throughout the cycle, while the accuracy of the test group improves from the initial similar level to 87.9% with the continuous optimization of the user portrait. This significant improvement of more than 35 percentage points verifies that the personalized portrait dynamic iteration mechanism adopted by the application has significant and quantifiable technical effects in improving user experience and service matching efficiency.

[0040] Embodiment 6: Before connecting a new service area containing multiple commercial facilities to the Internet of Things recommendation system of the application, in order to ensure that the system has uniform standards and reliability for service state evaluation and resource counting of all businesses in the service area, a set of pre-service business logic and fault tolerance mechanism configuration procedures need to be executed; the procedure first quantitatively configures the judgment logic of the encounter quality level, which maps the continuously changing expected encounter flow density into discrete service state levels; the core of the judgment logic is to calculate the ratio of the expected encounter flow density to the service capacity broadcast by the business , and based on the ratio, the service state is divided into three levels: when is less than 0.8, the system marks the encounter quality level of the corresponding expected encounter event as smooth; when is in the interval of 0.8 to 1.2, it is marked as saturated; when is greater than 1.2, it is marked as congestion warning.

[0041] ​Further, in order to deal with the situation that the service capacity count value is inaccurate due to the missing of encounter confirmation signal caused by the user's cancellation after booking or network communication delay, the procedure is configured with a set of fault-tolerant logic based on state and clock; when an expected encounter event is confirmed by the user through the booking operation, the system deducts one unit from the service capacity count value corresponding to the merchant while marking the unit as a time-limited reserved state, and the time length of the time-limited state is set according to the estimated arrival time calculated based on the future trajectory of the user's vehicle and a fixed grace period; if the system successfully receives the final encounter confirmation signal uploaded by the merchant and associated with the booking before the end of the time-limited period, the reserved state is cleared, otherwise, if the final confirmation signal is not received until the end of the period, the system determines that the booking is invalid and automatically performs a rollback operation, i.e. restores the service capacity count value of the merchant by one unit; after the above configuration procedure is completed, the service end logic of the newly accessed service area has a unified congestion level determination standard and a resource count mechanism with state rollback capability, providing a standardized data processing foundation for subsequent online operation.

[0042] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for recommending car-related products based on the Internet of Things, characterized in that: The method comprises the following steps: Step a: obtaining a future trajectory generated by the vehicle based on its navigation destination. The future trajectory is a data structure containing a series of future spatiotemporal points and the time arrival probability distribution corresponding to the future spatiotemporal points. Step b, obtaining supply information published by the merchant, including service content and effective time and space windows; Step c, calculating the expected probability of the vehicle's future trajectory encountering the merchant's supply information in time and space to generate at least one expected encounter event; Step d: Pushing guidance information to the vehicle or merchant based on the probability of the expected encounter event. The method further includes a dynamic robustness maintenance mechanism that performs the following operations: continuously comparing the vehicle's real-time position with the future trajectory line, and determining that a trajectory deviation event has occurred when the lateral offset distance between the vehicle's real-time position and the future trajectory line exceeds a reference threshold; In response to a trajectory deviation event, all calculation results of expected encounter events based on the future trajectory line are discarded; and an implicit trajectory deduction mode is activated. The implicit trajectory deduction mode is based on the historical gravity points determined by cluster analysis of the vehicle's local historical trajectory data, and the historical gravity points with the highest weight are selected in combination with the current time and date context. The mode generates a probability vector field representing the probability of the vehicle's short-term driving direction, so as to maintain the product recommendation service using the probability vector field.

2. The method for recommending car-related products based on the Internet of Things according to claim 1, characterized in that: The dynamic robustness maintenance mechanism further performs the following operations: after activating the implicit trajectory deduction mode, continuously monitors whether there is a resetting event in which the vehicle trajectory is re-anchored with a new or original navigation route, and automatically resumes execution of steps a to d in response to the resetting event.

3. The method for recommending car-related products based on the Internet of Things according to claim 1, characterized in that: After step d, the method further includes: aggregating future trajectory lines of multiple vehicles pointing to the same merchant and with overlapping time windows to calculate an expected encounter flow density representing the future passenger flow density; comparing the expected encounter flow density with the merchant's service capability to determine an encounter quality level; and based on the encounter quality level, adjusting the content of the guidance information pushed to the vehicle in step d.

4. The method for recommending car-related products based on the Internet of Things according to claim 1, characterized in that: The time arrival probability distribution in step a is dynamically calculated and updated using a normal distribution model based on the vehicle's current speed and real-time road condition information.

5. The method for recommending car-related products based on the Internet of Things according to claim 1, characterized in that: The calculation of the expected probability in step c is to calculate the probability of each discrete time and space point on the future trajectory line falling within the merchant's effective time and space window, and to perform a weighted sum of the probabilities of each point.

6. The method for recommending car-related products based on the Internet of Things according to claim 1, characterized in that: The guidance information pushed to the vehicle in step d is an opportunity reminder containing the expected encounter probability and the reservation or navigation detour suggestion operation.

7. The method for recommending car-related products based on the Internet of Things according to claim 1, characterized in that: The guidance information pushed to merchants in step d is a future passenger flow density heat map generated based on the expected encounter probability of multiple vehicles.

8. The method for recommending car-related products based on the Internet of Things according to claim 1, characterized in that: The method further includes a supply-side service capacity dynamic modulation mechanism, which performs the following operations: when an expected encounter event is confirmed as a real transaction, receiving an encounter confirmation signal associated with the real transaction; based on the encounter confirmation signal, deducting in real time the service capacity count value in the supply information corresponding to the merchant; And based on the deducted service capacity count value, the subsequent expected probability calculation is performed.

9. The method for recommending car-related products based on the Internet of Things according to claim 1, characterized in that: The guidance information pushed to the vehicle in step d is a binary timing opportunity that includes at least two alternative encounter events that differ in their core value propositions; the method also includes: receiving a behavioral feedback signal generated by the user's selection behavior for the binary timing opportunity; based on the accumulated behavioral feedback signal, constructing a personalized portrait for the user that includes time sensitivity and price sensitivity weight factors; and using the weight factors in the personalized portrait to sort or filter the subsequently generated expected encounter events.

10. The method for recommending car-related products based on the Internet of Things according to claim 3, characterized in that: The calculation of expected encounter traffic density follows the following rules: ,in, is the expected encounter traffic density, which represents the expected number of vehicles encountered per unit time. For the time window The total number of vehicles whose future trajectory lines point to the merchant, For the The expected encounter probability between a vehicle and a merchant.

Images