Parking space distribution and fee calculation method and system, and electronic equipment

Abstract

The invention is suitable for the technical field of parking management, and provides a parking space distribution and cost calculation method and system, and electronic equipment, and the method comprises the steps: carrying out the encoding of a constructed user input text sequence through employing an encoder of a pre-training generative model, and generating a user feature vector; combining the constructed parking space feature matrix with the user feature vector, and generating a dynamic weight coefficient by using a decoder of the model; calculating a recommendation matching score of each parking space in the candidate parking space library and generating a recommendation list based on the dynamic weight coefficient in combination with a preset distance attenuation function and a historical similarity between the user and the parking space; in response to a selection instruction of a user for a target parking space, obtaining a basic rate and a rental time length of the target parking space, and calculating a rental fee; and in the lease period, predicting the lease renewal intention of the user by using the model, and if the prediction probability exceeds a preset threshold, generating a lease renewal order. The problems that an existing parking management system is rigid in distribution logic and single in charging mode are solved.

Description

Technical Field

[0001] This invention relates to the field of parking management technology, and in particular to a parking space allocation and fee calculation method, system and electronic equipment. Background Technology

[0002] With the acceleration of urbanization, parking difficulties have become a major pain point in urban traffic management. While existing intelligent parking systems have alleviated the pressure of finding parking spaces to some extent, with the diversification of user needs and the development of technology, there are still many technical bottlenecks and functional defects in practical applications, specifically manifested in the following aspects:

[0003] The rigid logic of parking space allocation and the lack of user demand awareness: Most mainstream parking systems on the market (such as ETCP and Jieshun Technology) rely on preset static rules for parking space allocation. For example, they may only be based on fixed area divisions or a simple first-come, first-served principle, or only optimize based on a single dimension (such as the vacancy status of parking spaces). These systems fail to dynamically adjust based on the dynamic identities of users (such as owners, long-term tenants, temporary visitors, and VIP users) and real-time needs (such as the urgency of temporary visitors and the preferred parking spaces of tenants). As a result, the average utilization rate of parking space resources is often less than 60%, and users often have to manually filter parking spaces from a large list, which is cumbersome and fails to achieve optimal resource allocation.

[0004] The pricing model is simplistic and lacks personalized pricing capabilities: Most existing systems adopt a rigid "time-based + fixed rate" model (e.g., a uniformly set rate of 5 yuan / hour), and the billing logic is not linked to user identity (tenant, visitor, VIP), rental duration (short-term, long-term), or parking space quality (e.g., distance from elevator, spaciousness). This rigid pricing system lacks flexibility, failing to protect the rights of long-term tenants and unable to regulate supply and demand imbalances during peak periods through price levers.

[0005] Redundant renewal process: In the scenario of renewing parking space rentals, users often need to repeat the entire process of "search-select-pay", which lacks the inheritance of historical rental records. The operation path is long, which reduces users' willingness to renew the rental.

[0006] In summary, existing parking systems generally suffer from technical problems such as rigid parking space allocation logic, a single cost calculation model, and redundant renewal processes. Summary of the Invention

[0007] Therefore, the purpose of this invention is to provide a parking space allocation and fee calculation method, system and electronic device to fundamentally solve the problems of rigid allocation logic and single billing mode in existing parking management systems.

[0008] According to an embodiment of the present invention, a method for parking space allocation and fee calculation includes:

[0009] Collect user identity information, historical parking behavior data and vehicle attribute information, construct a user input text sequence containing task identifiers, and use the encoder of a pre-trained generative model to perform feature encoding on the user input text sequence to generate a user feature vector containing user intent.

[0010] Real-time collection of parking space status data constructs a parking space feature matrix, and the parking space feature matrix is ​​combined with the user feature vector. The decoder of the pre-trained generative model adaptively generates dynamic weight coefficients for different parking space feature factors based on the business scenario identifier of the current time window.

[0011] Based on the dynamic weight coefficient, combined with the preset distance decay function and the historical similarity between the user and the parking space, the recommended matching score of each parking space in the candidate parking space database is calculated, and a recommendation list is generated based on the recommended matching score.

[0012] In response to the user's selection instruction for a target parking space in the recommended list, the basic rate and rental duration of the target parking space are obtained, and the recommendation matching score of the target parking space is introduced as a negative correlation factor for price calculation to calculate the final rental fee.

[0013] During the rental period, the pre-trained generative model is used to predict the user's intention to renew the rental. If the predicted probability exceeds a preset threshold, a renewal order is generated, and the user's final payment behavior and actual parking time are used as feedback signals to update the feature attention parameters of the pre-trained generative model.

[0014] In addition, the parking space allocation and fee calculation method according to the above embodiments of the present invention may also have the following additional technical features:

[0015] Furthermore, the steps of collecting user identity information, historical parking behavior data, and vehicle attribute information, constructing a user input text sequence containing a task identifier, and using the encoder of a pre-trained generative model to perform feature encoding on the user input text sequence to generate a user feature vector containing the user's intent include:

[0016] The user's identity tag, average parking time of historical orders and coordinates of frequently used parking spaces, as well as vehicle model and license plate location information are obtained, normalized and text-concatenated, and a user task identifier is added to the beginning of the concatenated sequence to construct a user input text sequence containing the task identifier.

[0017] The user input text sequence is input into the encoder of the pre-trained transformer model, and the multi-head self-attention mechanism is used to capture the dependencies between elements within the sequence, and the context-dependent hidden layer state vector is output.

[0018] Extract the output vector from the last hidden layer of the encoder corresponding to the user task identifier position, and use it as the initial user feature vector;

[0019] Obtain the feedback value of the previous round of recommendation results, calculate the feature adjustment coefficient based on the feedback value, and perform a dot product operation on each dimension of the initial user feature vector with the feature adjustment coefficient to obtain a weighted user feature vector, so as to dynamically adjust the weight ratio of the pre-trained generative model for specific user behavior features.

[0020] Furthermore, the step of combining the parking space feature matrix with the user feature vector and using the decoder of the pre-trained generative model to adaptively generate dynamic weight coefficients for different parking space feature factors based on the business scenario identifier of the current time window includes:

[0021] Real-time monitoring of traffic flow and parking space occupancy rate in the current area, and calculation of the area congestion index;

[0022] Based on the comparison between the regional congestion index and the preset threshold, a business scenario identifier representing the business strategy is generated.

[0023] The business scenario identifier is used as the driving instruction for the decoder of the pre-trained generative model, and the dimensional information of the parking space feature matrix and the generated user feature vector are used as context inputs. The decoder of the pre-trained generative model is used to perform autoregressive generation and output dynamic weight coefficients for different parking space feature factors.

[0024] Furthermore, the step of calculating the recommended matching score for each parking space in the candidate parking space database based on the dynamic weight coefficient, combined with a preset distance decay function and the historical similarity between the user and the parking space, includes:

[0025] Extract parking space vacancy time, physical distance between parking space and elevator entrance, historical usage frequency of parking space, and regional peak index from the parking space status matrix;

[0026] The scoring criteria for physical distance are calculated using a preset distance decay function;

[0027] Calculate the cosine similarity between the user feature vector and the historical usage frequency feature of the parking space to obtain the historical preference matching value;

[0028] The scoring items for parking space vacancy time, physical distance, and regional peak index are multiplied by the dynamic weight coefficients corresponding to different parking space characteristic factors to obtain their respective weighted attribute scores.

[0029] The final recommended matching score for the parking space is obtained by linearly summing all weighted attribute scores with the historical preference matching values.

[0030] Furthermore, the steps of responding to the user's selection instruction for a target parking space in the recommended list, obtaining the base rate and rental duration of the target parking space, and introducing the recommendation matching score of the target parking space as a negative correlation factor for price calculation, to calculate the final rental fee include:

[0031] It receives confirmation instructions from user terminals for selecting target parking spaces in the recommended list in real time, retrieves the base rate of the target parking space based on the current time period and parking space attributes, and obtains the rental duration set by the user.

[0032] The identity adjustment coefficient is determined based on the user's identity information, and the duration discount factor is calculated based on the rental duration using logarithmic decay logic.

[0033] Extract the calculated recommendation matching score of the target parking space and perform a weighted calculation with the preset priority discount coefficient to construct a recommendation incentive item that reflects the negative correlation between price and recommendation degree;

[0034] The basic rate, rental duration, identity adjustment coefficient, duration discount factor, and referral incentives are substituted into a preset multi-dimensional cost calculation model, and the final rental fee payable by the user is calculated by multiplying them together.

[0035] Furthermore, the step of predicting the user's renewal intention using the pre-trained generative model during the rental period, and generating a renewal order if the predicted probability exceeds a preset threshold, includes:

[0036] Within a preset time window before the lease ends, construct input prompt information that includes a lease renewal prediction instruction, user feature vector, and the future occupancy status of the current parking space;

[0037] The input prompt information is fed into a pre-trained generative model, and the decoder of the pre-trained generative model outputs the predicted renewal probability value and the suggested renewal duration.

[0038] Determine whether the renewal probability value is higher than a preset confidence threshold;

[0039] If the confidence level is higher than the preset confidence threshold, a one-click renewal confirmation request containing the suggested renewal duration will be sent to the user terminal.

[0040] Upon receiving confirmation from the user, the system updates the validity period and fee information of the rental order based on the suggested renewal duration, and simultaneously updates the access permissions of the barrier gate system.

[0041] Furthermore, the step of using the user's final payment behavior and actual parking duration as feedback signals to update the feature attention parameters of the pre-trained generative model includes:

[0042] After payment is completed, the deviation rate is calculated by comparing the user's actual parking departure time with the reservation end time; if the user actually fulfills the payment and the deviation rate is within the preset allowable range, the transaction data is marked as a positive incentive sample.

[0043] If a user defaults on the lease or refuses a parking space that is ranked high on the recommended list, this interaction data will be marked as a negative penalty sample.

[0044] A training dataset containing the positive incentive samples and negative penalty samples is constructed. The attention weight matrix in the encoder of the pre-trained generative model is fine-tuned using the backpropagation algorithm so that the pre-trained generative model automatically enhances its attention to the feature dimensions that generate positive feedback in subsequent predictions.

[0045] Furthermore, the step of calculating the physical distance scoring item using a preset distance attenuation function includes:

[0046] Determine the user's identity tag category to identify whether they are an emergency visitor or a long-term tenant;

[0047] If a user is identified as an emergency visitor, the decay rate parameter in the distance decay function is increased, so that the calculated rating item decreases rapidly with the increase of physical distance, thus prioritizing the presentation of parking spaces that are very close in the recommendation results;

[0048] If a user is identified as a long-term tenant, the decay rate parameter in the distance decay function is reduced, so that the calculated score item decreases slowly as the physical distance increases, thereby retaining parking spaces that are farther away but may conform to long-term usage habits in the recommendation results.

[0049] Another embodiment of the present invention aims to provide a parking space allocation and fee calculation system, the system comprising:

[0050] The user feature vector generation module is used to collect user identity information, historical parking behavior data and vehicle attribute information, construct a user input text sequence containing task identifiers, and use the encoder of a pre-trained generative model to perform feature encoding on the user input text sequence to generate a user feature vector containing user intent.

[0051] The dynamic weight coefficient generation module is used to collect parking space status data in real time to construct a parking space feature matrix, and combine the parking space feature matrix with the user feature vector. The decoder of the pre-trained generative model adaptively generates dynamic weight coefficients for different parking space feature factors based on the business scenario identifier of the current time window.

[0052] The recommendation list generation module is used to calculate the recommendation matching score of each parking space in the candidate parking space database based on the dynamic weight coefficient, combined with the preset distance decay function and the historical similarity between the user and the parking space, and generate a recommendation list based on the recommendation matching score.

[0053] The rental fee calculation module is used to respond to the user's selection instruction for the target parking space in the recommended list, obtain the basic rate and rental duration of the target parking space, and introduce the recommendation matching score of the target parking space as a negative correlation factor for price calculation to calculate the final rental fee.

[0054] The renewal update module is used to predict the user's renewal intention during the rental period using the pre-trained generative model. If the prediction probability exceeds a preset threshold, a renewal order is generated, and the user's final payment behavior and actual parking time are used as feedback signals to update the feature attention parameters of the pre-trained generative model.

[0055] Another embodiment of the present invention aims to provide an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the parking space allocation and fee calculation method as described above.

[0056] The parking space allocation and fee calculation method provided in this invention constructs a text sequence containing task identifiers by collecting multidimensional user data and performs deep feature encoding using a pre-trained generative model encoder. This achieves semantic-level intent understanding of users' explicit identities and implicit historical habits, breaking through the limitations of traditional systems that rely solely on license plates or simple rule matching, and significantly improving the accuracy of user feature characterization. By combining parking space status data and business scenario identifiers (such as peak / off-peak hours) in real time, the method uses a pre-trained generative model decoder to adaptively generate dynamic weight coefficients for different feature factors (such as idle time and physical distance). This enables the recommendation strategy to be flexibly adjusted according to macro traffic conditions (e.g., prioritizing high turnover during peak hours and prioritizing better experience during off-peak hours), effectively solving the problems of low efficiency and uneven resource allocation of static allocation rules during morning and evening peak hours. By introducing a recommendation matching score as a negative correlation factor for price calculation, a dynamic pricing mechanism is constructed where higher recommendation scores result in more favorable fees, thus using economic levers to guide users to actively choose the system. This system allocates globally optimal parking spaces (e.g., high matching scores), maximizing overall parking lot operational efficiency while ensuring user benefits. It addresses the issues of disconnect between recommendation algorithms and billing rules, and low user cooperation in existing technologies. By dynamically adjusting the decay rate parameter in the distance decay function based on user identity tags (e.g., emergency visitors or long-term tenants), it accurately responds to user expectations in specific scenarios (e.g., urgent parking or long-term rental preferences), providing highly personalized services while maintaining algorithm versatility and significantly improving user parking satisfaction. Furthermore, by predicting renewal intentions within the rental period and providing one-click renewal, it shifts from passively waiting for user input to proactively serving the system, significantly reducing user operational costs. Simultaneously, by constructing a closed loop using user's final payment behavior and actual parking duration as feedback signals, and fine-tuning the model's feature attention parameters using backpropagation, it achieves self-learning and self-evolution capabilities, ensuring the model can continuously adapt to changes in user behavior over long-term operation. This solves the problems of rigid allocation logic and a single billing model in existing parking management systems. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the parking space allocation and fee calculation method in the first embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of the parking space allocation and fee calculation system in the second embodiment of the present invention;

[0059] Figure 3 This is a schematic diagram of the structure of the electronic device in the third embodiment of the present invention;

[0060] The following detailed description of the embodiments will further illustrate the present invention in conjunction with the above-described accompanying drawings. Detailed Implementation

[0061] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0062] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0064] Example 1

[0065] Please see Figure 1 The diagram illustrates a parking space allocation and fee calculation method according to a first embodiment of the present invention. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown. The parking space allocation and fee calculation method provided by the embodiment of the present invention includes:

[0066] Step S10: Collect user identity information, historical parking behavior data and vehicle attribute information, construct a user input text sequence containing task identifier, and use the encoder of the pre-trained generative model to perform feature encoding on the user input text sequence to generate a user feature vector containing user intent.

[0067] In one embodiment of the present invention, step S10 specifically includes:

[0068] The user's identity tag, average parking time of historical orders and coordinates of frequently used parking spaces, as well as vehicle model and license plate location information are obtained, normalized and text-concatenated, and a user task identifier is added to the beginning of the concatenated sequence to construct a user input text sequence containing the task identifier.

[0069] The user input text sequence is fed into the encoder of the pre-trained transformer model. The multi-head self-attention mechanism is used to capture the dependencies between elements within the sequence and output the context-dependent hidden layer state vector.

[0070] Extract the output vector from the last hidden layer of the encoder corresponding to the user task identifier position, and use it as the initial user feature vector;

[0071] Obtain the feedback value of the previous round of recommendation results, calculate the feature adjustment coefficient based on the feedback value, and perform a dot product operation on each dimension of the initial user feature vector with the feature adjustment coefficient to obtain a weighted user feature vector, so as to dynamically adjust the weight ratio of the pre-trained generative model for specific user behavior features.

[0072] Specifically, this method is applied to a parking space allocation and fee calculation system. When a user registers for the first time or a vehicle enters the parking lot for the first time, the system obtains the user's identity information, historical parking behavior data, and vehicle attribute information through user self-filling, historical order accumulation, and automatic collection by vehicle recognition equipment. This raw data is typically structured or semi-structured. For example, identity information includes user ID (User_001) and identity tags (owner / long-term tenant / visitor / VIP). Historical parking behavior data includes average parking time over the past 30 days (e.g., 8.5 hours), frequently used parking space coordinates (e.g., B2 level, area A), and historical payment preferences. Vehicle attribute information includes vehicle type (SUV / sedan) and license plate location (local / out-of-town).

[0073] At this point, the currently collected data might include, for example, an identity tag of "owner," a historical average parking time of "eight hours," a frequently used area of ​​"basement level 2, zone A," and a vehicle type of "SUV." Next, to enable the pre-trained generative model to understand this diverse information, the system executes a standardized data transformation and text sequence construction process. Let the user-input text sequence be X, and its mathematical expression be defined as: X = {x1, x2, x3, ..., x...} n}, where: X represents the complete input sequence; where x1, x2, ..., x... nThe text identifiers in the sequence represent various elements, containing key information such as identity tags, historical order summaries, and vehicle models. Specifically, to instruct the model to perform feature extraction, the text sequence includes specially defined task identifiers. Specifically, the system transcribes the collected information into text, for example, converting "eight hours" of parking duration into the text fragment "long parking duration," and "basement level 2, zone A" into "frequently parked zone A." Then, the system sequentially concatenates these text fragments into a complete descriptive sentence according to preset grammatical rules. For example, the concatenated sentence might be: "User identity is owner, long parking duration, frequently parked zone A, vehicle type is SUV." To clearly inform the model that the current task is to extract user features, the system adds a special task identifier, such as "user feature extraction," at the beginning of this sentence.

[0074] Then, the system inputs this completed user input text sequence, containing the task identifier, into the encoder part of the pre-trained generative model. Specifically, this embodiment utilizes the "encoder-decoder" joint architecture of the pre-trained generative model to generate the user feature vector U. The encoder is responsible for understanding all information, while the decoder is responsible for focusing on the core intent. The formula for calculating the generated user feature vector U is as follows:

[0075] U=T5_Encode(X)⊙T5_Decode([CLS])

[0076] Where U represents the final generated high-dimensional feature vector containing the user intent (U∈R) d Where d is the vector dimension, representing the user's initial intent, is expressed as U = [U1, U2, ..., U...]. k ,…,U d ], where U k This represents the k-th value (the feature value of the k-th dimension) in this high-dimensional feature vector; T5_Encode(X) represents the hidden layer state matrix output by the encoder of the pre-trained generative model after fully encoding the input text sequence X, where the matrix captures the contextual dependencies between elements in the sequence; [CLS] represents the pre-set task identifier token, used to indicate that the pre-trained generative model focuses on the specific task of "identity core feature extraction"; T5_Decode([CLS]) represents the attention query vector generated by the decoder after receiving the [CLS] identifier; ⊙ represents the interactive attention operation.

[0077] Specifically, the encoder consists of multiple stacked attention layers. When a text sequence enters, the multi-head self-attention mechanism in each layer begins to operate. The core of this mechanism lies in calculating the correlation strength between each word in the sequence and all other words. For example, the model might analyze that the identity tag "owner" has a strong correlation with the behavioral habit of "frequently parked in Zone A," while the correlation with the vehicle attribute "sports utility vehicle" might be slightly weaker. In this way, the encoder progressively captures the complex semantic dependencies within the sequence, ultimately generating a set of hidden layer state vectors at the top layer that comprehensively reflect the contextual information. Furthermore, to obtain a condensed feature vector representing the entire user profile, the system uses the query vector generated by the decoder, corresponding to the special task identifier at the beginning of the sequence, to extract the global information from the encoder output. This vector, because it incorporates comprehensive information from the entire sequence during the calculation process, is considered an initial user feature vector that highly summarizes the user's overall intent.

[0078] Finally, the generated user feature vector is not static. To address the problem of rigid model parameters in existing technologies and enable it to dynamically reflect changes in user preferences, the system introduces a feedback-based adjustment mechanism. Based on the quality of the previous round of recommendations (feedback value R), the system dynamically adjusts the weights α of each dimension in the feature vector. i This enables the model to learn online.

[0079] The adjusted feature weights α′ i The calculation formula is as follows:

[0080] α′ i =α i ·(1+β·R i )

[0081] Where α i α′ represents the original weight value of the i-th dimension in the feature vector before adjustment. i R represents the adjusted updated weight value for the i-th dimension used in the next round of calculation; i R represents the feedback value of the recommendation result, which is automatically calculated by the system based on user behavior (e.g., if the user accepts the recommendation and completes the payment, R is positive; if the user rejects the recommendation, R is negative or a smaller value). The better the recommendation result, the higher the weight of the corresponding feature. β represents the feedback coefficient, used to control the sensitivity of weight adjustment. In this specific configuration, β is set to 0.1. When β = 0.1, if the recommendation result is excellent, the model will appropriately amplify the weight of the corresponding feature, allowing that feature to play a greater role in subsequent recommendations, thereby achieving an intelligent effect that becomes increasingly adept at understanding the user. Finally, the system utilizes the adjusted weight α... ′i Perform a weighted update on the feature vector:

[0082]

[0083] Where α ′ k This represents the adjustment factor for the k-th dimension. U ′ k Let represent the k-th value of the dynamic vector after the operation. Then, through element-wise multiplication—that is, multiplying each number in the vector by its corresponding adjustment coefficient—the adjusted feature vector can be obtained. Therefore, through the above mathematical model, the system completes the transformation from the original data to the dynamic feature vector U. ′ This transformation lays the data foundation for subsequent accurate recommendations. Specifically, the system obtains the user's satisfaction score from the previous round of parking recommendations. This score can be actively evaluated by the user or automatically calculated by the system based on the user's behavior, such as whether they accepted the recommendation and whether they parked successfully. The system generates a feature adjustment coefficient based on this satisfaction score; the higher the satisfaction, the larger the adjustment coefficient. Subsequently, the system multiplies each dimension of the initial user feature vector with this feature adjustment coefficient. Through this weighting process, if the user is very satisfied with the previous recommendations, the weight of the user feature vector used to generate the current recommendation will be strengthened overall; conversely, if the user is dissatisfied, the weight will be appropriately weakened. Through this series of processes, a user feature vector containing the user's intent can be generated that deeply understands the user's historical preferences and dynamically adapts to recent behavioral feedback.

[0084] Step S20: Collect parking space status data in real time to construct a parking space feature matrix, and combine the parking space feature matrix with the user feature vector. Use the decoder of the pre-trained generative model to adaptively generate dynamic weight coefficients for different parking space feature factors based on the business scenario identifier of the current time window.

[0085] In one embodiment of the present invention, the step of combining the parking space feature matrix with the user feature vector and using the decoder of the pre-trained generative model to adaptively generate dynamic weight coefficients for different parking space feature factors based on the business scenario identifier of the current time window specifically includes:

[0086] Real-time monitoring of traffic flow and parking space occupancy rate in the current area, and calculation of the area congestion index;

[0087] Based on the comparison between the regional congestion index and the preset threshold, a business scenario identifier representing the business strategy is generated.

[0088] The business scenario identifier is used as the driving instruction for the decoder of the pre-trained generative model, and the dimensional information of the parking space feature matrix and the generated user feature vector are used as context inputs. The decoder of the pre-trained generative model is used to perform autoregressive generation and output dynamic weight coefficients for different parking space feature factors.

[0089] Specifically, the system first collects multi-dimensional parking space status data for each physical parking space in the parking lot in real time, including idle time, location coordinates, and historical usage frequency. This multi-dimensional parking space status data is then structured and integrated to construct a parking space feature matrix. Specifically, the system maintains a dynamically updated database covering all available parking spaces in the parking lot. The system collects the current status of each parking space in real time and extracts feature values ​​from four core dimensions: idle time, physical distance, historical popularity, and regional attribute. Idle time is the time (in minutes) the parking space has been vacant since the last vehicle left; physical distance is the walking distance (in meters) from the nearest main entrance / exit or elevator lobby; historical popularity is the number of times the parking space has been selected and parked in the past 30 days (normalized value); and regional attribute is the congestion or priority level of the area where the parking space is located (e.g., 1 for VIP areas, 0 for regular areas). At this point, the system combines the above feature values ​​of N parking spaces to construct an N-row, 4-column parking space feature matrix S, where each row of the parking space feature matrix represents a parking space and each column represents a feature factor.

[0090] While constructing the parking space feature matrix, the system needs to perceive the current macro-business environment. By connecting to the entrance and exit gates, the system calculates the weighted sum of real-time inbound traffic flow and overall parking space occupancy rate to obtain a regional congestion index. Based on the comparison between the congestion index and a preset threshold, it generates corresponding peak-hour or off-peak-hour prompts as business scenario identifiers. The system constructs the decoder's input sequence, converting the business scenario identifiers into corresponding token sequences. The dimensional information of the parking space feature matrix and the weighted user feature vector generated in step S10 are used as context inputs to the decoder's interactive attention layer. The decoder runs a pre-trained generative model, performing autoregressive generation based on the business scenario identifiers and user feature vectors. It outputs numerical sequences for different parking space feature factors and maps these sequences to dynamic weight coefficients, enabling adaptive adjustment of weights according to business scenarios and user intent.

[0091] Specifically, this method requires the system to perceive the current macro-level business environment in order to make optimal decisions between efficiency and user experience. The system connects in real-time to parking space sensors and entrance / exit gate systems to obtain real-time parking space status data and traffic flow data. Parking space status data includes parking space vacancy time, physical location coordinates of the parking space, and historical usage frequency; traffic flow data includes the number of vehicles entering the parking lot per minute. Based on this raw data, the system first constructs an indicator reflecting the current environmental state, namely the regional congestion index.

[0092] At this point, in order to transform the digitized environmental indicators into instructions that the model can understand, the system executes the process of generating business scenario identifiers. Here, the regional congestion index is denoted as Index, and its calculation formula is defined as:

[0093] Index = a·Flow + b·Occupancy

[0094] Where: Flow represents the real-time inbound vehicle flow (vehicles / minute); Occupancy represents the current total parking space occupancy rate (percentage); a and b are preset balancing coefficients. The system sets a peak threshold T. high If the calculated Index > T high If the index is less than or equal to T, the system determines that it is currently in a congested state and generates a peak mode prompt word representing the "turnaround priority" strategy, such as "Mode:Peak"; if Index ≤ T high If the system determines that it is currently in an idle state, it generates a "mode: Off-peak" prompt word representing a "user experience priority" strategy. Specifically, the system uses the generated business scenario identifier (prompt word) as the core instruction driving the decoder. To achieve a fusion of personalization and scenario-based approaches, the decoder of the pre-trained generative model not only receives this prompt word but also the user feature vector U containing the user's personal intent generated in step S10. ′ The decoder, through its internal interactive attention mechanism, combines macroscopic scene instructions with microscopic user features. Then, the system utilizes the decoder of a pre-trained generative model to generate dynamic weight coefficients ω for different feature factors. j Specifically, this embodiment utilizes the autoregressive generation capability of the decoder to output a set of numerical sequences, which are then assigned to various evaluation dimensions. The dynamic weight coefficient ω is generated as follows. j The calculation formula is as follows:

[0095] ω j =T5_Generate(Prompt+S state ,U′)

[0096] Where ω jT5 represents the dynamic weighting coefficient for the j-th parking space's characteristic factors (such as idle time, physical distance, etc.); T5_Generate represents the decoder's generation function; Prompt represents the business scenario identifier (peak / off-peak prompt) generated based on the congestion index; S state The current parking space status description text is represented by U′; U′ represents the weighted user feature vector generated in step S10, which serves as the Key and Value input for the decoder's interactive attention layer. Specifically, if the input is a prompt word for peak mode, the weight of the idle time factor is higher than the weight of the location distance factor in the first set of dynamic weight coefficients output by the decoder; if the input is a prompt word for off-peak mode, the weight of the location distance factor is higher than the weight of the idle time factor in the second set of dynamic weight coefficients output by the decoder. In detail, during the calculation process, the decoder adjusts its attention distribution according to the Prompt content. When the input is "Mode:Peak" (peak mode), the parameter distribution within the decoder will cause the weight value ω corresponding to the "parking space idle time" factor in the generated numerical sequence to be higher. idle This is because during peak hours, the system strategy tends to guide users to park in long-vacant or easily accessible parking spaces to reduce the time spent searching for parking and improve overall turnover. Conversely, when the input is "Mode: Off-peak," the weight value ω corresponding to the "physical distance" factor in the numerical sequence generated by the decoder increases significantly. dist The number of users will increase significantly. At this point, the system strategy will tend to recommend parking spaces closest to the elevator or the user's destination to optimize the user's parking experience.

[0097] Furthermore, since the user feature vector u′ is introduced into the calculation, the generated weight coefficient ω j It is not only limited by the business scenario, but also subject to fine-tuning by individual user preferences. For example, even in peak mode, if the user feature vector u ′ The user is identified as a "VIP customer" with a "high-frequency payment" record. When generating weights, the decoder may appropriately retain some attention to "physical distance" to avoid recommending parking spaces that are too far away to high-value users, thereby achieving a dynamic balance between scenario strategy and user experience.

[0098] Finally, the system outputs a final set of dynamic weight coefficients {ω1,ω2,…,ω m These weights correspond to multiple evaluation dimensions, including parking space vacancy time, physical distance, historical popularity, and regional attributes. These weighting coefficients will be directly used in the comprehensive score calculation in subsequent step S30. Through this process, the system successfully transforms abstract business scenarios (peak / off-peak) and implicit user intentions into concrete, calculable mathematical weights, ensuring that every parking space allocation decision is the optimal solution based on the current environment and user needs.

[0099] Step S30: Based on the dynamic weight coefficient, combined with the preset distance decay function and the historical similarity between the user and the parking space, calculate the recommendation matching score of each parking space in the candidate parking space database, and generate a recommendation list based on the recommendation matching score;

[0100] In one embodiment of the present invention, the step of calculating the recommended matching score of each parking space in the candidate parking space database based on dynamic weight coefficients, combined with a preset distance decay function and the historical similarity between the user and the parking space includes:

[0101] Extract parking space vacancy time, physical distance between parking space and elevator entrance, historical usage frequency of parking space, and regional peak index from the parking space status matrix;

[0102] The scoring criteria for physical distance are calculated using a preset distance decay function;

[0103] Calculate the cosine similarity between the user feature vector and the historical usage frequency feature of the parking space to obtain the historical preference matching value;

[0104] The scoring items for parking space vacancy time, physical distance, and regional peak index are multiplied by the dynamic weight coefficients corresponding to different parking space characteristic factors to obtain their respective weighted attribute scores.

[0105] The final recommended matching score for the parking space is obtained by linearly summing all weighted attribute scores with historical preference matching values.

[0106] Furthermore, in one embodiment of the present invention, the step of calculating the physical distance scoring item using a preset distance attenuation function includes:

[0107] Determine the user's identity tag category to identify whether they are an emergency visitor or a long-term tenant;

[0108] If a user is identified as an emergency visitor, the decay rate parameter in the distance decay function is increased, so that the calculated rating item decreases rapidly with the increase of physical distance, thus prioritizing the presentation of parking spaces that are very close in the recommendation results;

[0109] If a user is identified as a long-term tenant, the decay rate parameter in the distance decay function is reduced, so that the calculated score item decreases slowly as the physical distance increases, thereby retaining parking spaces that are farther away but may conform to long-term usage habits in the recommendation results.

[0110] The specific steps of step S30 include: traversing each available parking space in the candidate parking space database, extracting multi-dimensional status attributes of the parking space, including idle time, physical distance from the target entrance, historical usage frequency, and peak index of the area; applying a preset distance decay function to perform nonlinear mapping calculation on the physical distance attribute to obtain a distance score item, wherein the decay rate parameter in the function is dynamically adjusted according to the user identity tag; calculating the cosine similarity between the weighted user feature vector generated in step S10 and the historical feature vector of the parking space to obtain the historical preference matching value; using the dynamic weight coefficients generated in step S20, weighting and summing the score items of each dimension, and superimposing the above historical preference matching value to obtain the final recommended matching score of the parking space, and generating a recommendation list by sorting the scores in descending order.

[0111] Specifically, this step transforms the abstract weighting strategy into concrete ranking results. The system first needs to construct the feature matrix of candidate parking spaces at the current time. For each available parking space S in the parking lot... i The system extracts the parking space's status data in real time. This data includes the parking space's vacancy time (e.g., the parking space has been vacant for 30 minutes), the physical distance between the parking space and the user's destination (e.g., elevator entrance or shopping mall entrance) (e.g., 50 meters), the frequency of the parking space's occupation over a period of time, and the current congestion level of the area where the parking space is located (area peak index).

[0112] To scientifically quantify the impact of physical distance, a key factor, on user experience, the system introduces a non-linear distance decay function. Unlike simple linear deduction, this function simulates the psychological changes in users' perception of distance. The distance scoring term f... dist The calculation formula is defined as follows:

[0113] f dist (d)=e -λ·d

[0114] Where: d represents the actual physical distance between the parking space and the target entrance; e is the base of the natural logarithm; λ is the decay rate parameter, used to control how quickly the score decreases with increasing distance. Specifically, to reflect the human-centered intelligent allocation logic, the decay rate parameter λ in the above formula is not fixed, but dynamically adjusted according to the user identity information extracted in step S10. The system determines the user's identity tag category: if the user is identified as an "emergency visitor" (e.g., in a hospital emergency room scenario or manually marked as an emergency by the user), the system will set λ to a large value (e.g., λ = 2.0). At this time, with a small increase in distance d, the score item f... distThe rate of decline due to distance decreases dramatically, causing the algorithm to heavily exclude distant parking spaces during calculations, ensuring that the best parking space closest to the user is recommended. Conversely, if the user is identified as a "long-term tenant" or "commuting employee," the system sets λ to a smaller value (e.g., λ = 0.5). In this case, the rate decay caused by distance becomes more gradual, and the algorithm tolerates greater distances, instead considering other factors such as the user's historical habits or the spaciousness of the parking space.

[0115] Then, in addition to objective physical attributes, the system also needs to consider the implicit compatibility between the user and the parking space. At this point, the system executes the historical similarity calculation process. Specifically, the system retrieves the weighted user feature vector U generated in step S10, which contains the user's deep intent. ′ And the historical feature vector S of the parking space, showing its past usage. hist The system uses a cosine similarity algorithm to calculate the degree of matching between the two, where the historical preference matching value Sim is calculated using the following formula:

[0116]

[0117] Where: the numerator is the dot product of the two vectors, and the denominator is the product of the magnitudes of the two vectors. The result Sim calculated by this formula is between -1 and 1. The closer the value is to 1, the more the attributes of the parking space (such as the degree of remoteness of the location, the surrounding environment, the types of vehicles that have been parked there in the past) match the user's historical behavior profile.

[0118] Finally, the system integrates the various indicators calculated above to calculate the final recommended matching score R for each parking space. i This calculation process not only considers individual scores but also deeply integrates the dynamic weight coefficients reflecting the current business strategy generated by the decoder in step S20. The recommendation matching score R... i The calculation formula is as follows:

[0119]

[0120] Where: m represents the total number of evaluation dimensions (such as free time, distance rating, etc.); ω j The dynamic weight coefficients for the j-th feature factor generated in step S20; V ij The normalized value of the i-th parking space on the j-th feature factor (where the distance factor is the f calculated above). dist ); γ is the preset historical preference balance coefficient (e.g., tenant γ = 0.3, visitor γ = 0.1); Sim is the historical preference matching value calculated above.

[0121] Specifically, using this formula, the system iterates through and calculates the R value for each parking space in the candidate parking space database. iValue. Subsequently, the system, based on R... i The system sorts all parking spaces in descending order of their values. It then extracts the top N (e.g., top 3) parking spaces from the sorted list, encapsulates this information to generate a final recommendation list, and pushes it to the user's device for display. Through these steps, the system achieves accurate parking space matching based on multi-factor dynamic weights and personalized user needs.

[0122] Step S40: In response to the user's instruction to select a target parking space in the recommended list, obtain the basic rate and rental duration of the target parking space, and introduce the recommendation matching score of the target parking space as a negative correlation factor for price calculation to calculate the final rental fee.

[0123] In one embodiment of the present invention, step S40 specifically includes:

[0124] It receives confirmation instructions from user terminals for selecting target parking spaces in the recommended list in real time, retrieves the base rate of the target parking space based on the current time period and parking space attributes, and obtains the rental duration set by the user.

[0125] The identity adjustment coefficient is determined based on the user's identity information, and the duration discount factor is calculated based on the rental duration using logarithmic decay logic.

[0126] Extract the calculated recommendation matching score of the target parking space and perform a weighted calculation with the preset priority discount coefficient to construct a recommendation incentive item that reflects the negative correlation between price and recommendation degree;

[0127] The basic rate, rental duration, identity adjustment coefficient, duration discount factor, and referral incentives are substituted into a preset multi-dimensional cost calculation model, and the final rental fee payable by the user is calculated by multiplying them together.

[0128] Specifically, this method is applied to the fee calculation and pricing module, aiming to change the traditional single-time charging model and guide user behavior through multi-dimensional price levers. When a user selects a target parking space from the recommended list on the app, the system immediately triggers the billing process. First, the system needs to determine the billing benchmark, namely, to obtain the base rate P0 (yuan / hour) and the rental duration T (hours). The base rate P0 can be dynamically set according to peak or off-peak periods, and the rental duration T is either input by the user or estimated by the system by default.

[0129] At this point, in order to reflect the differentiated rights and interests of users based on their identity, the system introduces an identity adjustment coefficient k. U The system assigns a value based on the user identity tag identified in step S10: if the user is a "registered tenant", the system will assign a value to k. U Set to a smaller discount value (e.g., k) UThe value ranges from 0.7 to 0.8 to protect the rights of long-term users; if the user is a "temporary visitor," the system will... U Set to a standard value (e.g., k) U =1.0); if it is a "VIP user", then set a lower value (e.g., k). U =0.6).

[0130] Furthermore, to encourage long-term leasing and reduce the management costs associated with frequent parking space turnover, the system introduces a duration discount factor d that varies non-linearly over time. T Specifically, this embodiment uses a logarithmic function to simulate the diminishing marginal discount effect, where the duration discount factor d... T The calculation formula is defined as follows:

[0131] d T = 1 - 0.05·log(T+1)

[0132] Where: T is the rental duration; log is the natural logarithm or common logarithm. According to this formula, as the rental duration T increases, the value of log(T+1) gradually increases, thus making d... T The value gradually decreases. This means that the longer the user rents, the lower the average rate per unit time, and this discount is a smooth transition, avoiding the abruptness of tiered pricing.

[0133] Then, the system deeply integrates the recommended matching score into the billing logic, constructing an incentive mechanism where higher cooperation leads to lower costs. The system obtains the recommended matching score R for the target parking space calculated in step S30. i A preset priority discount coefficient β (e.g., β = 0.05) is introduced. The system constructs the recommendation incentive term (1 - β·R). i Specifically, this item reflects a negative correlation factor: when the recommended matching score R... i The higher the value (e.g., close to 1.0), the better it is for the parking space, as it aligns with both the system's optimal scheduling and the user's historical habits. (1-β·R) i The smaller the value of (R), the greater the discount. For example, if a user selects the system-recommended first parking space (R... i (Higher), and the extra discount they will receive will be more than if they choose a parking space at the end of the recommended list.

[0134] Finally, the system integrates all the above factors and uses a multidimensional cost calculation model to calculate the final rental fee C. The formula for calculating the total cost is as follows:

[0135] C = P0·T·k U ·d T ·(1-β·R i )

[0136] Where: C is the final cost; P0 is the base rate; T is the lease duration; k U d is the identity adjustment coefficient; T The duration discount factor; (1-β·R i () is a recommended incentive item.

[0137] Specifically, the system performs the aforementioned multiplication operation to arrive at the final amount, and clearly displays the breakdown of the cost on the user's terminal interface (e.g., "Original price XX yuan, status discount XX yuan, long-term rental discount XX yuan, referral reward XX yuan"). Through this transparent and incentive-based pricing method, the system effectively utilizes economic means to guide users to choose the parking space that is most beneficial to the system's overall resource allocation (i.e., the highest recommendation score), achieving a dual improvement in parking space resource utilization and user satisfaction.

[0138] Step S50: During the rental period, the user's intention to renew the rental is predicted using a pre-trained generative model. If the predicted probability exceeds a preset threshold, a renewal order is generated, and the user's final payment behavior and actual parking time are used as feedback signals to update the feature attention parameters of the pre-trained generative model.

[0139] In one embodiment of the present invention, the step of predicting a user's renewal intention using a pre-trained generative model during the rental period, and generating a renewal order if the predicted probability exceeds a preset threshold, includes:

[0140] Within a preset time window before the lease ends, construct input prompt information that includes a lease renewal prediction instruction, user feature vector, and the future occupancy status of the current parking space;

[0141] Input prompts are fed into a pre-trained generative model, and the decoder of the pre-trained generative model outputs the predicted renewal probability and the suggested renewal duration.

[0142] Determine whether the renewal probability value is higher than the preset confidence threshold;

[0143] If the confidence level is higher than the preset confidence threshold, a one-click renewal confirmation request containing the suggested renewal duration will be sent to the user terminal.

[0144] Upon receiving confirmation from the user, the system updates the validity period and fee information of the rental order based on the suggested renewal duration, and simultaneously updates the access permissions of the barrier gate system.

[0145] In one embodiment of the present invention, the step of updating the feature attention parameters of the pre-trained generative model using the user's final payment behavior and actual parking duration as feedback signals includes:

[0146] After payment is completed, the deviation rate is calculated by comparing the user's actual parking departure time with the reservation end time;

[0147] If the user actually fulfills the payment and the deviation rate is within the preset allowable range, the transaction data will be marked as a positive incentive sample.

[0148] If a user defaults on the lease or refuses a parking space that is ranked high on the recommended list, this interaction data will be marked as a negative penalty sample.

[0149] A training dataset containing the positive incentive samples and negative penalty samples is constructed. The attention weight matrix in the encoder of the pre-trained generative model is fine-tuned using the backpropagation algorithm so that the pre-trained generative model automatically enhances its attention to the feature dimensions that generate positive feedback in subsequent predictions.

[0150] Specifically, step S50 includes the following steps: within a preset time window before the rental order expires, constructing a prompt text sequence containing a "renewal prediction" task instruction, current user characteristics, and the future occupancy status of the current parking space; inputting the prompt text sequence into a pre-trained generative model, and outputting the predicted renewal probability value and suggested renewal duration through the model's decoder; if the renewal probability value is higher than a preset confidence threshold, triggering an active service mechanism to send a renewal confirmation request containing the suggested duration to the user terminal, and generating a renewal order based on user feedback; after the service ends, collecting the user's actual payment actions and actual parking duration, comparing them with the system's predicted values, and constructing positive or negative feedback samples; using the feedback samples to fine-tune the pre-trained generative model, and updating the feature attention parameters in the model encoder through the backpropagation algorithm to improve the accuracy of subsequent predictions.

[0151] Specifically, the system monitors the status of all ongoing orders in real time. When it detects that the remaining time of a rental order is less than a preset threshold (e.g., 15 minutes remaining), the system automatically initiates the renewal intention prediction process. At this point, the system needs to integrate multi-dimensional contextual information to construct the model's input. The system extracts the user feature vector U generated in step S10, which represents the user's long-term preferences, and obtains the real-time status S of the current parking space. current (Including whether the parking space has been booked for a future time period, and the rate trend after the current time period).

[0152] At this point, to instruct the pre-trained generative model to perform the prediction task, the system constructs a standardized input prompt text. The structure of this text sequence is, for example, "Task: Renewal Prediction; User Features: [Vector Data]; Parking Space Status: [Status Data]". Subsequently, the system inputs this text sequence into the pre-trained generative model. This embodiment utilizes the generative capabilities of the pre-trained generative model to directly output two key predicted values: one is the probability R that the user will renew their lease. renew Another is the possible renewal period T for the user. sug The renewal probability R is... renew The computational logic is completed by neural network reasoning within the model, and its mathematical expression is:

[0153] R renew ,T sug =T5-Generate(Renewal_Predict+U+S current )

[0154] Where: R renew The probability value is between 0 and 1; T sug `Rrenewal_predict` is the suggested rental duration inferred by the model based on the user's historical average parking duration (e.g., "1 hour" or "30 minutes"); `Renewal_Predict` is the rental renewal prediction. Specifically, the system sets a confidence threshold (e.g., 0.8). The system determines the rental renewal probability Rrenewal_predict output by the model. renew Is it greater than this threshold? If R renew A value greater than 0.8 indicates that the model believes the user is highly likely to require a delay. In this case, the system no longer waits for manual user input but proactively pops up a "One-Click Renewal" window in the user's app, pre-filled with the model-suggested renewal duration T. sug Users simply need to click the "Confirm" button, and the system will execute step S40 to recalculate the outstanding fees and update the order end time and access permissions of the gate system. This design greatly simplifies the user's operation and avoids involuntary fines or exit obstacles caused by exceeding the time limit.

[0155] Furthermore, to endow the system with the ability to self-evolve, this embodiment of the invention constructs a closed-loop feedback mechanism based on real business data. After the user finally completes payment and drives the vehicle away, the system collects the user's final behavior data as the truth value. The system calculates the deviation rate between the actual parking time and the reserved time (including extension), and defines sample labels in conjunction with the payment status. Specifically, the definitions are as follows: if the user accepts the extension suggestion, successfully completes payment, and the deviation rate between the actual parking time and the system's suggested time is less than a preset range (e.g., 10%), then this prediction is marked as a positive feedback sample; if the user rejects the extension suggestion (chooses to leave immediately), or the user's actual parking time far exceeds the extension time, resulting in a breach of contract, then this prediction is marked as a negative feedback sample.

[0156] Finally, the system uses the constructed positive and negative sample sets to fine-tune the pre-trained generative model online. The system employs backpropagation to update the self-attention weight matrix in the model's encoder. Specifically, the update aims to adjust the model's attention parameters to input features. For example, if the model detects that "visitors with children" have a very high probability of renewing their leases in a "rainy day" scenario, but previous predictions ignored the "weather" feature, the backpropagation algorithm will increase the weight of the "weather" feature when processing this type of user through negative sample correction. Through this continuous feedback training, the pre-trained generative model can continuously adapt to changes in user behavior, ensuring that the feature attention parameters remain optimal, thus achieving an intelligent effect that becomes more accurate with use.

[0157] In summary, the parking space allocation and fee calculation method in the above embodiments of the present invention constructs a text sequence containing task identifiers by collecting multidimensional user data and uses the encoder of a pre-trained generative model for deep feature encoding. This achieves semantic-level intent understanding of users' explicit identities and implicit historical habits, thus breaking through the limitations of traditional systems that rely solely on license plates or simple rule matching, and significantly improving the accuracy of user feature characterization. By combining parking space status data and business scenario identifiers (such as peak / off-peak hours) in real time, the decoder of the pre-trained generative model adaptively generates dynamic weight coefficients for different feature factors (such as idle time and physical distance), realizing the elastic adjustment of recommendation strategies according to macro traffic conditions (e.g., prioritizing high turnover during peak hours and prioritizing better experience during off-peak hours). This effectively solves the problems of low efficiency and uneven resource allocation of static allocation rules during morning and evening peak hours. By introducing recommendation matching scores as a negative correlation factor for price calculation, a dynamic pricing mechanism is constructed in which higher recommendation scores result in more favorable fees, realizing the use of economic levers to guide users to actively choose. By selecting the globally optimal parking space (e.g., high matching score) based on user profiles, the system maximizes overall parking lot operational efficiency while ensuring user benefits. This addresses the issues of disconnect between recommendation algorithms and billing rules, and low user cooperation in existing technologies. By dynamically adjusting the decay rate parameter in the distance decay function based on user identity tags (e.g., emergency visitors or long-term tenants), the system accurately responds to user expectations in specific scenarios (e.g., urgent parking or long-term rental preferences), providing highly personalized services while maintaining algorithm versatility and significantly improving user parking satisfaction. Furthermore, by predicting renewal intentions within the rental period and providing one-click renewal, the system shifts from passively waiting for user input to proactively serving users, significantly reducing user operational costs. Simultaneously, by constructing a closed loop using user final payment behavior and actual parking duration as feedback signals, and fine-tuning the model's feature attention parameters using backpropagation, the system achieves self-learning and self-evolution capabilities, ensuring continuous adaptation to changes in user behavior over long-term operation. This solves the problems of rigid allocation logic and a single billing model in existing parking management systems.

[0158] Example 2

[0159] Please see Figure 2 This is a schematic diagram of a parking space allocation and fee calculation system provided in the second embodiment of the present invention. For ease of explanation, only the parts related to the embodiment of the present invention are shown. The system includes:

[0160] The user feature vector generation module 11 is used to collect user identity information, historical parking behavior data and vehicle attribute information, construct a user input text sequence containing task identifiers, and use the encoder of a pre-trained generative model to perform feature encoding on the user input text sequence to generate a user feature vector containing user intent.

[0161] The dynamic weight coefficient generation module 12 is used to collect parking space status data in real time to construct a parking space feature matrix, and combine the parking space feature matrix with the user feature vector. The decoder of the pre-trained generative model adaptively generates dynamic weight coefficients for different parking space feature factors based on the business scenario identifier of the current time window.

[0162] The recommendation list generation module 13 is used to calculate the recommendation matching score of each parking space in the candidate parking space database based on the dynamic weight coefficient, combined with the preset distance decay function and the historical similarity between the user and the parking space, and generate a recommendation list based on the recommendation matching score.

[0163] The rental fee calculation module 14 is used to respond to the user's selection instruction for the target parking space in the recommended list, obtain the basic rate and rental duration of the target parking space, and introduce the recommendation matching score of the target parking space as a negative correlation factor for price calculation to calculate the final rental fee.

[0164] The renewal update module 15 is used to predict the user's renewal intention during the rental period using the pre-trained generative model. If the prediction probability exceeds a preset threshold, a renewal order is generated, and the user's final payment behavior and actual parking time are used as feedback signals to update the feature attention parameters of the pre-trained generative model.

[0165] Furthermore, in one embodiment of the present invention, the user feature vector generation module 11 includes:

[0166] The user input text sequence construction unit is used to obtain the user's identity tag, the average parking time of historical orders and the coordinates of commonly used parking spaces, as well as the vehicle's model and license plate location information, and perform normalization and text concatenation. The user task identifier is added to the beginning of the concatenated sequence to construct a user input text sequence containing the task identifier.

[0167] The model output unit is used to input the user input text sequence into the encoder of the pre-trained transformer model, use a multi-head self-attention mechanism to capture the dependencies between elements within the sequence, and output a context-dependent hidden layer state vector.

[0168] The user feature vector extraction unit is used to extract the output vector corresponding to the user task identifier position in the last hidden layer of the encoder, as the initial user feature vector.

[0169] The user feature vector generation unit is used to obtain the feedback value of the previous round of recommendation results, calculate the feature adjustment coefficient based on the feedback value, and perform a dot product operation on each dimension of the initial user feature vector with the feature adjustment coefficient to obtain a weighted user feature vector, so as to dynamically adjust the weight ratio of the pre-trained generative model for specific user behavior features.

[0170] Furthermore, in one embodiment of the present invention, the dynamic weight coefficient generation module 12 includes:

[0171] The regional congestion index calculation unit is used to monitor the current traffic flow and parking space occupancy rate data in the region in real time and calculate the regional congestion index.

[0172] The business scenario identifier generation unit is used to generate a business scenario identifier representing a business strategy based on the comparison result between the regional congestion index and a preset threshold.

[0173] The dynamic weight coefficient output unit is used to take the business scenario identifier as the driving instruction of the decoder of the pre-trained generative model, and take the dimensional information of the parking space feature matrix and the generated user feature vector as context input. It uses the decoder of the pre-trained generative model to perform autoregressive generation and outputs dynamic weight coefficients for different parking space feature factors.

[0174] Furthermore, in one embodiment of the present invention, the recommendation list generation module 13 includes:

[0175] The parking space status matrix extraction unit is used to extract the parking space vacancy time, physical distance between the parking space and the elevator entrance, historical usage frequency of the parking space, and regional peak index from the parking space status matrix.

[0176] The scoring item calculation unit is used to calculate the scoring item of physical distance using a preset distance decay function;

[0177] The preference matching value calculation unit is used to calculate the cosine similarity between the user feature vector and the historical usage frequency feature of the parking space to obtain the historical preference matching value.

[0178] The weighted attribute score calculation unit is used to multiply the scoring items of parking space vacancy time, physical distance, and regional peak index by the dynamic weight coefficients corresponding to different parking space characteristic factors to obtain their respective weighted attribute scores.

[0179] The recommended matching score calculation unit is used to linearly sum all weighted attribute scores with the historical preference matching values ​​to obtain the final recommended matching score for the parking space.

[0180] Furthermore, in one embodiment of the present invention, the rental fee calculation module 14 includes:

[0181] The data acquisition unit is used to receive confirmation selection instructions for target parking spaces in the recommended list sent by the user terminal in real time, retrieve the basic rate of the target parking space according to the current time period and parking space attributes, and obtain the rental duration set by the user.

[0182] The discount coefficient determination unit is used to determine the identity adjustment coefficient based on the user's identity information and to calculate the duration discount factor based on the rental duration using logarithmic decay logic.

[0183] The recommendation incentive construction unit is used to extract the calculated recommendation matching score of the target parking space and perform a weighted operation with the preset priority discount coefficient to construct a recommendation incentive that reflects the negative correlation between price and recommendation degree.

[0184] The rental fee calculation unit is used to input the basic rate, rental duration, identity adjustment coefficient, duration discount factor and referral incentive into the preset multi-dimensional fee calculation model, and multiply them to calculate the final rental fee that the user needs to pay.

[0185] Furthermore, in one embodiment of the present invention, the lease renewal module 15 includes:

[0186] The input prompt information construction unit is used to construct input prompt information, including renewal prediction instructions, user feature vectors, and the future occupancy status of the current parking space, within a preset time window before the end of the lease.

[0187] The data output unit is used to input the input prompt information into the pre-trained generative model, and output the predicted renewal probability value and the suggested renewal duration through the decoder of the pre-trained generative model.

[0188] The renewal probability determination unit is used to determine whether the renewal probability value is higher than a preset confidence threshold.

[0189] The renewal confirmation request sending unit is used to send a one-click renewal confirmation request containing a suggested renewal duration to the user terminal if the confidence level is higher than a preset confidence threshold.

[0190] The rental order update unit is used to update the validity period and fee information of the rental order based on the suggested renewal period after receiving the user's confirmation instruction, and to simultaneously update the access permissions of the barrier gate system.

[0191] Furthermore, in one embodiment of the present invention, the lease renewal module 15 includes:

[0192] The deviation rate calculation unit is used to calculate the deviation rate by comparing the user's actual parking departure time with the reserved end time after payment is completed;

[0193] A positive incentive sample marking unit is used to mark the transaction data as a positive incentive sample if the user actually fulfills the payment and the deviation rate is within a preset allowable range.

[0194] The negative penalty sample labeling unit is used to mark the interaction data as a negative penalty sample if a user defaults on the lease or refuses a parking space ranked at the top of the recommendation list.

[0195] The weight matrix fine-tuning unit is used to construct a training dataset containing the positive incentive samples and negative penalty samples mentioned above. It uses the backpropagation algorithm to fine-tune the attention weight matrix in the encoder of the pre-trained generative model so that the pre-trained generative model can automatically enhance its attention to the feature dimensions that generate positive feedback in subsequent predictions.

[0196] Furthermore, in one embodiment of the present invention, the scoring item calculation unit includes:

[0197] The identity tag category determination subunit is used to determine the user's identity tag category and identify whether they belong to an emergency visitor or a long-term tenant.

[0198] The first decay rate parameter adjustment subunit is used to increase the decay rate parameter in the distance decay function if the user is identified as an emergency visitor, so that the calculated rating item decreases rapidly with the increase of physical distance, thereby giving priority to displaying parking spaces that are very close in the recommendation results.

[0199] The second decay rate parameter adjustment subunit is used to reduce the decay rate parameter in the distance decay function if the user is identified as a long-term tenant, so that the calculated score item decreases slowly as the physical distance increases, thereby retaining parking spaces that are farther away but may conform to long-term usage habits in the recommendation results.

[0200] The parking space allocation and fee calculation system provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0201] Example 3

[0202] In another aspect, the present invention also proposes an electronic device, please refer to [link to relevant documentation]. Figure 3 The electronic device shown is an electronic device according to the third embodiment of the present invention, including a memory 20, a processor 10, and a program 30 stored in the memory 20 and executable on the processor. When the processor 10 executes the program 30, it implements the IoT device repair process management method as described in the above embodiment.

[0203] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0204] The memory 20 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (such as SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of the electronic device, such as a SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the electronic device. Furthermore, the memory 20 can include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data installed on the electronic device, but also to temporarily store data that has been output or will be output.

[0205] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0206] This invention also provides a medium storing a program that, when executed by a processor, implements the parking space allocation and fee calculation method as described in the above embodiments.

[0207] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units or modules as needed, that is, the internal structure of the storage device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application.

[0208] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0209] More specific examples of media (a non-exhaustive list) include: electrical connections (electronic devices) with one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Additionally, the media can even be paper or other suitable media on which the program can be printed, as the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in memory.

[0210] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0211] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0212] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for parking space allocation and fee calculation, characterized in that, The method includes: Collect user identity information, historical parking behavior data and vehicle attribute information, construct a user input text sequence containing task identifiers, and use the encoder of a pre-trained generative model to perform feature encoding on the user input text sequence to generate a user feature vector containing user intent. Real-time collection of parking space status data constructs a parking space feature matrix, and the parking space feature matrix is ​​combined with the user feature vector. The decoder of the pre-trained generative model adaptively generates dynamic weight coefficients for different parking space feature factors based on the business scenario identifier of the current time window. Based on the dynamic weight coefficient, combined with the preset distance decay function and the historical similarity between the user and the parking space, the recommended matching score of each parking space in the candidate parking space database is calculated, and a recommendation list is generated based on the recommended matching score. In response to the user's selection instruction for a target parking space in the recommended list, the basic rate and rental duration of the target parking space are obtained, and the recommendation matching score of the target parking space is introduced as a negative correlation factor for price calculation to calculate the final rental fee. During the rental period, the pre-trained generative model is used to predict the user's intention to renew the rental. If the predicted probability exceeds a preset threshold, a renewal order is generated, and the user's final payment behavior and actual parking time are used as feedback signals to update the feature attention parameters of the pre-trained generative model.

2. The parking space allocation and fee calculation method according to claim 1, characterized in that, The steps of collecting user identity information, historical parking behavior data, and vehicle attribute information, constructing a user input text sequence containing a task identifier, and using a pre-trained generative model encoder to perform feature encoding on the user input text sequence to generate a user feature vector containing the user's intent include: The user's identity tag, average parking time of historical orders and coordinates of frequently used parking spaces, as well as vehicle model and license plate location information are obtained, normalized and text-concatenated, and a user task identifier is added to the beginning of the concatenated sequence to construct a user input text sequence containing the task identifier. The user input text sequence is input into the encoder of the pre-trained transformer model, and the multi-head self-attention mechanism is used to capture the dependencies between elements within the sequence, and the context-dependent hidden layer state vector is output. Extract the output vector from the last hidden layer of the encoder corresponding to the user task identifier position, and use it as the initial user feature vector; Obtain the feedback value of the previous round of recommendation results, calculate the feature adjustment coefficient based on the feedback value, and perform a dot product operation on each dimension of the initial user feature vector with the feature adjustment coefficient to obtain a weighted user feature vector, so as to dynamically adjust the weight ratio of the pre-trained generative model for specific user behavior features.

3. The parking space allocation and fee calculation method according to claim 1, characterized in that, The step of combining the parking space feature matrix with the user feature vector and using the decoder of the pre-trained generative model to adaptively generate dynamic weight coefficients for different parking space feature factors based on the business scenario identifier of the current time window includes: Real-time monitoring of traffic flow and parking space occupancy rate in the current area, and calculation of the area congestion index; Based on the comparison between the regional congestion index and the preset threshold, a business scenario identifier representing the business strategy is generated. The business scenario identifier is used as the driving instruction for the decoder of the pre-trained generative model, and the dimensional information of the parking space feature matrix and the generated user feature vector are used as context inputs. The decoder of the pre-trained generative model is used to perform autoregressive generation and output dynamic weight coefficients for different parking space feature factors.

4. The parking space allocation and fee calculation method according to claim 1, characterized in that, The step of calculating the recommended matching score for each parking space in the candidate parking space database based on the dynamic weight coefficient, combined with a preset distance decay function and the historical similarity between the user and the parking space, includes: Extract parking space vacancy time, physical distance between parking space and elevator entrance, historical usage frequency of parking space, and regional peak index from the parking space status matrix; The scoring criteria for physical distance are calculated using a preset distance decay function; Calculate the cosine similarity between the user feature vector and the historical usage frequency feature of the parking space to obtain the historical preference matching value; The scoring items for parking space vacancy time, physical distance, and regional peak index are multiplied by the dynamic weight coefficients corresponding to different parking space characteristic factors to obtain their respective weighted attribute scores. The final recommended matching score for the parking space is obtained by linearly summing all weighted attribute scores with the historical preference matching values.

5. The parking space allocation and fee calculation method according to claim 1, characterized in that, The steps for responding to a user's selection instruction for a target parking space in the recommended list, obtaining the base rate and rental duration of the target parking space, and incorporating the recommendation matching score of the target parking space as a negative correlation factor in price calculation, to calculate the final rental fee include: It receives confirmation instructions from user terminals for selecting target parking spaces in the recommended list in real time, retrieves the base rate of the target parking space based on the current time period and parking space attributes, and obtains the rental duration set by the user. The identity adjustment coefficient is determined based on the user's identity information, and the duration discount factor is calculated based on the rental duration using logarithmic decay logic. Extract the calculated recommendation matching score of the target parking space and perform a weighted calculation with the preset priority discount coefficient to construct a recommendation incentive item that reflects the negative correlation between price and recommendation degree; The basic rate, rental duration, identity adjustment coefficient, duration discount factor, and referral incentives are substituted into a preset multi-dimensional cost calculation model, and the final rental fee payable by the user is calculated by multiplying them together.

6. The parking space allocation and fee calculation method according to claim 1, characterized in that, The step of predicting a user's renewal intention using the pre-trained generative model during the rental period, and generating a renewal order if the predicted probability exceeds a preset threshold, includes: Within a preset time window before the lease ends, construct input prompt information that includes a lease renewal prediction instruction, user feature vector, and the future occupancy status of the current parking space; The input prompt information is fed into a pre-trained generative model, and the decoder of the pre-trained generative model outputs the predicted renewal probability value and the suggested renewal duration. Determine whether the renewal probability value is higher than a preset confidence threshold; If the confidence level is higher than the preset confidence threshold, a one-click renewal confirmation request containing the suggested renewal duration will be sent to the user terminal. Upon receiving confirmation from the user, the system updates the validity period and fee information of the rental order based on the suggested renewal duration, and simultaneously updates the access permissions of the barrier gate system.

7. The parking space allocation and fee calculation method according to claim 1, characterized in that, The step of updating the feature attention parameters of the pre-trained generative model by using the user's final payment behavior and actual parking duration as feedback signals includes: After payment is completed, the deviation rate is calculated by comparing the user's actual parking departure time with the reservation end time; If the user actually fulfills the payment and the deviation rate is within the preset allowable range, the transaction data will be marked as a positive incentive sample. If a user defaults on the lease or refuses a parking space that is ranked high on the recommended list, this interaction data will be marked as a negative penalty sample. A training dataset containing the positive incentive samples and negative penalty samples is constructed. The attention weight matrix in the encoder of the pre-trained generative model is fine-tuned using the backpropagation algorithm so that the pre-trained generative model automatically enhances its attention to the feature dimensions that generate positive feedback in subsequent predictions.

8. The parking space allocation and fee calculation method according to claim 4, characterized in that, The step of calculating the physical distance scoring item using a preset distance attenuation function includes: Determine the user's identity tag category to identify whether they are an emergency visitor or a long-term tenant; If a user is identified as an emergency visitor, the decay rate parameter in the distance decay function is increased, so that the calculated rating item decreases rapidly with the increase of physical distance, thus prioritizing the presentation of parking spaces that are very close in the recommendation results; If a user is identified as a long-term tenant, the decay rate parameter in the distance decay function is reduced, so that the calculated score item decreases slowly as the physical distance increases, thereby retaining parking spaces that are farther away but may conform to long-term usage habits in the recommendation results.

9. A parking space allocation and fee calculation system, characterized in that, The system includes: The user feature vector generation module is used to collect user identity information, historical parking behavior data and vehicle attribute information, construct a user input text sequence containing task identifiers, and use the encoder of a pre-trained generative model to perform feature encoding on the user input text sequence to generate a user feature vector containing user intent. The dynamic weight coefficient generation module is used to collect parking space status data in real time to construct a parking space feature matrix, and combine the parking space feature matrix with the user feature vector. The decoder of the pre-trained generative model adaptively generates dynamic weight coefficients for different parking space feature factors based on the business scenario identifier of the current time window. The recommendation list generation module is used to calculate the recommendation matching score of each parking space in the candidate parking space database based on the dynamic weight coefficient, combined with the preset distance decay function and the historical similarity between the user and the parking space, and generate a recommendation list based on the recommendation matching score. The rental fee calculation module is used to respond to the user's selection instruction for the target parking space in the recommended list, obtain the basic rate and rental duration of the target parking space, and introduce the recommendation matching score of the target parking space as a negative correlation factor for price calculation to calculate the final rental fee. The renewal update module is used to predict the user's renewal intention during the rental period using the pre-trained generative model. If the prediction probability exceeds a preset threshold, a renewal order is generated, and the user's final payment behavior and actual parking time are used as feedback signals to update the feature attention parameters of the pre-trained generative model.

10. An electronic device, characterized in that, It includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the parking space allocation and fee calculation method as described in any one of claims 1-8.

Images