A parking lot recommendation method, device and distributed system based on historical data

Abstract

The application provides a parking lot recommendation method and device based on historical data and a distributed system, relates to the technical field of data processing, and comprises the following steps: an initial model set is established in advance, the initial model set comprises a plurality of initial models corresponding to respective parking lots, historical data of the respective parking lots is acquired as training data, the historical data of the respective parking lots is trained, a target model set is obtained, each target model is used for predicting the leaving time and leaving probability of each vehicle in a target time period of a corresponding parking lot, the vehicle distribution state of the parking lot in the target time period is determined, real-time data of the respective parking lots is acquired, a real-time parking database that dynamically changes is established based on the target model set, a parking score model is established, each target parking lot is scored, and a target vehicle is recommended to a parking lot, and the problem that the matching degree of an existing parking lot recommendation method is not enough is solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a parking lot recommendation method, apparatus, and distributed system based on historical data. Background Technology

[0002] In popular tourist attractions and shopping malls, due to high traffic volume and limited parking spaces, parking is often full, leading to long waiting times for users and negatively impacting their experience. Therefore, effectively managing surrounding parking spaces to maximize parking capacity and minimize waiting times is a crucial challenge for these locations.

[0003] Currently, most smart parking systems focus primarily on parking space management and navigation within the parking lot itself, neglecting to consider the real-time availability and information of other parking lots in the vicinity. This results in the system being unable to promptly recommend alternative parking options when the current parking lot is full, leading to recommendations that do not accurately match user needs and result in longer waiting times for users. Summary of the Invention

[0004] To overcome the aforementioned technical deficiencies, the present invention aims to provide a parking recommendation method, apparatus, and distributed system based on historical data, thereby addressing the problem of insufficient accuracy in existing parking recommendation methods.

[0005] This invention discloses a parking lot recommendation method based on historical data, comprising:

[0006] An initial model set is pre-established, which includes several initial models corresponding to each parking lot;

[0007] Historical data from each parking lot is used as training data to train each of the initial models to obtain a target model set. Each target model is used to predict the departure time and departure probability of each vehicle in the corresponding parking lot during the target time period, so as to determine the vehicle distribution status of the parking lot during the target time period.

[0008] Acquire real-time data from each parking lot and establish a dynamically changing real-time parking database based on the target model set;

[0009] A parking scoring model is established to score each target parking lot in order to recommend parking lots for target vehicles; wherein, the parking scoring model is used to score each parking lot based on real-time data obtained from the real-time parking database.

[0010] Obtain the location information of the target vehicle, and determine at least one target parking lot in the real-time parking database based on the location information of the target vehicle and the score.

[0011] Preferably, each of the target models performs the following operations:

[0012] Collect basic information and vehicle information of vehicles entering the corresponding parking lot, wherein the basic information includes weather data and time data;

[0013] The vehicle type is determined based on the vehicle information, wherein the vehicle type includes fixed parking users, temporary parking users, and long-term parking users;

[0014] Based on the vehicle type and the basic information, a trained target model is used to process the information to obtain the departure time and departure probability of the vehicle information in the target time period, wherein the target model is a long short-term memory model.

[0015] Preferably, after acquiring historical data from each parking lot, the process includes:

[0016] Historical data is cleaned and then standardized.

[0017] And / or, classify the information of each vehicle in the historical data;

[0018] And / or, calculate and analyze the parking duration data corresponding to each vehicle information in the historical data to predict the departure time and departure probability of each vehicle in the target time period of the corresponding parking lot.

[0019] Preferably, based on changes in the parking lot database, data is synchronized to the real-time parking database via Spark or Flink to update the real-time parking database.

[0020] Preferably, the step of establishing a scoring model to score each target parking lot includes:

[0021] Obtain the real-time number of available parking spaces for each of the target parking lots, and generate a first score for each of the target parking lots by weighting the real-time number of available parking spaces to the total number of parking spaces with a first weight.

[0022] The number of vehicles leaving each target parking lot and the departure probability of each leaving vehicle are determined from the real-time parking database according to the target model corresponding to each target parking lot during the target time period. The second score of each target parking lot is calculated based on the second weight multiplied by the preset first rule, wherein the preset first rule is the average departure probability of each leaving vehicle multiplied by the ratio of the number of empty spaces generated by leaving vehicles to the number of non-empty spaces in real time.

[0023] Vehicles within a preset range for each target parking lot are obtained from the real-time parking database. The number of vehicles entering each target parking lot during the target time period is determined. The location of each entering vehicle is obtained, and a third score for each target parking lot is calculated by multiplying a third weight by a preset second rule. The preset second rule is the ratio of the number of entering vehicles to the number of available spaces and the number of departing vehicles.

[0024] The distance of each target parking lot relative to the target vehicle and the destination is determined from the location information of each target parking lot in the real-time parking database, and a fourth score of each target parking lot is calculated based on the fourth weight.

[0025] Preferably, the step of establishing a scoring model to score each target parking lot includes:

[0026] The overall score for each target parking lot is determined based on its first, second, third, and fourth scores.

[0027] Preferably, the scoring of each target parking lot is repeated at predetermined time intervals to update the parking lot recommendations for the target vehicles.

[0028] Preferably, the evaluation information fed back by the target vehicle is received, and the parking scoring model and each of the target models are optimized in reverse based on the evaluation information.

[0029] The present invention also provides a parking recommendation device based on historical data, comprising:

[0030] The first processing module is used to pre-establish an initial model set, which includes several initial models corresponding to each parking lot; acquire historical data of each parking lot as training data, train each initial model to obtain a target model set, and each target model is used to predict the departure time and departure probability of each vehicle in the corresponding parking lot during a target time period, so as to determine the vehicle distribution status of the parking lot during the target time period.

[0031] The second processing module is used to acquire real-time data from each parking lot and establish a dynamically changing real-time parking database based on the target model set.

[0032] The execution module is used to establish a parking scoring model to score each target parking lot in order to recommend parking lots for the target vehicle; wherein, the parking scoring model is used to obtain real-time data of each target parking lot from the real-time parking database for scoring; obtain the location information of the target vehicle, and determine at least one target parking lot in the parking lot database based on the location information of the target vehicle and the score.

[0033] The present invention also provides a distributed system, including a master node and multiple slave nodes, wherein the master node and each slave node cooperate to apply the above-mentioned parking recommendation device based on historical data.

[0034] Compared with existing technologies, the above technical solution has the following advantages:

[0035] The parking recommendation method, device, and distributed system provided in this application based on historical data establish a model for each parking lot, predict the parking space distribution status of each parking lot under a target time period based on historical data, and score each parking lot from multiple dimensions such as the number of available parking spaces, location information, and vehicles leaving and entering the parking lot under the target time period to make parking lot recommendations. The combination of historical and real-time parking lot data solves the problem of insufficient matching degree of existing parking recommendation methods. Attached Figure Description

[0036] Figure 1 This is a flowchart of an embodiment of the parking recommendation method, apparatus and distributed system based on historical data described in this invention;

[0037] Figure 2 This is a schematic diagram of the model structure of a parking recommendation method, device and distributed system based on historical data according to an embodiment of the present invention;

[0038] Figure 3 This is a schematic diagram of a second embodiment of the parking lot recommendation method, device and distributed system based on historical data described in this invention.

[0039] Figure label:

[0040] 6- Parking recommendation device based on historical data; 61- First processing module; 62- Second processing module; 63- Execution module. Detailed Implementation

[0041] The advantages of the present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments.

[0042] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0043] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0044] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0045] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0046] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0047] In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the convenience of the description of the invention and have no specific meaning in themselves. Therefore, "module" and "part" can be used interchangeably.

[0048] Example 1: This invention discloses a parking lot recommendation method based on historical data. This embodiment models and analyzes historical data of each parking lot, combining it with real-time parking space availability and vehicle location information (specifically, through vehicle GPS navigation information networks) to perform a comprehensive parking score. For details, please refer to... Figure 1 and Figure 2 ,include:

[0049] S10: A preliminary model set is established in advance, which includes several preliminary models corresponding to each parking lot; historical data of each parking lot is obtained as training data, and each preliminary model is trained to obtain a target model set. Each target model is used to predict the departure time and departure probability of each vehicle in the corresponding parking lot during the target time period, so as to determine the vehicle distribution status of the parking lot during the target time period.

[0050] For illustration, the target model, obtained after training the initial model, is used to predict the departure time and probability of each vehicle in the corresponding parking lot during the target time period. It is based on a Long Short-Term Memory (LSTM) model. It should be noted that each parking lot corresponds to an initial model. Historical data for the parking lot is acquired, and the initial model set is trained to obtain the target model set, which is used to predict the departure time of each vehicle entering the parking lot. This can be viewed as a combination of multiple LSTM models. Multiple LSTM models can simultaneously predict the entering vehicles, or a single LSTM model can predict the entering vehicles one by one to determine the vehicle distribution in the parking lot during the target time period.

[0051] S20: Acquire real-time data from each parking lot and establish a dynamically changing real-time parking database based on the target model set. In this embodiment, historical data includes the historical data corresponding to each parking lot. Specifically, when collecting / generating historical data, a unified interface and format can be used for different parking lots to upload the parking historical data to the real-time parking database (e.g., a central data warehouse or a Sqoop-synchronized historical database) in real time. Then, based on the data changes in the parking lot database, it is synchronized to the real-time parking database via Spark or Flink to update the real-time parking database. Specifically, when a vehicle enters the parking lot, data is synchronized to the real-time parking database in real time based on the log increment of the parking lot database via Spark or Flink. Here, the parking lot database refers to the database where parking information is stored. Historical data may include, but is not limited to, all vehicle characteristic information, entry time, parking time, parking space number, and departure time of vehicles entering the parking lot during historical time periods. In addition to the above-mentioned pre-collected and generated data, pre-stored historical data can also be obtained from other devices / servers. Real-time data can also be continuously collected and supplemented to continuously improve and update the historical data.

[0052] As described above, the target model is used to predict the departure time and probability of any vehicle in a parking lot. Specifically, each target model performs the following operations: collecting basic information and vehicle information of vehicles entering the corresponding parking lot, wherein the basic information includes, but is not limited to, weather data and time data; identifying the vehicle type based on the vehicle information, wherein the vehicle type includes, but is not limited to, regular parking users, temporary parking users, and long-term parking users; specifically, for example, the parking time of regular parking users is mainly concentrated during peak hours on weekdays, and the number of users parking may increase in inclement weather conditions, therefore, basic information such as weather data and time data are taken into consideration, and the time data includes, but is not limited to, days of the week and whether it is a holiday. As an example, for any parking lot, an employee information database can be established within the real-time parking database. This database includes, but is not limited to, employee names, departments, positions, license plate numbers, and parking card numbers. This allows for the identification of fixed parking users, whose parking times are relatively fixed, mostly during weekday working hours, such as 8:00-17:00. In subsequent predictions, it can be determined that they will leave after 17:00. The license plate numbers in the employee database are matched with the cleaned parking record data to filter out historical parking data for employee vehicles. Parking durations exceeding 24 hours are defined as long-term parking, and the remaining data are considered temporary parking. Based on the vehicle type and the basic information, a trained target model is used to process the data to obtain the departure time and departure probability corresponding to the vehicle information within the target time period. The target model is a Long Short-Term Memory (LSTM) model. For explanation, the target time period is the time period during which parking lot recommendations are made for the target vehicle. Specifically, when a vehicle enters the parking lot, it is linked to the employee information database in the real-time parking database to determine the vehicle type. Based on the vehicle information (such as the license plate number), it is queried whether the vehicle has a parking record. If so, it is filtered out (e.g., data within the last year). If not, data of the same type of vehicle is filtered out (e.g., data of the same type of vehicle within the last 7 days). Then, the time data and weather data are fed into the long short-term memory model to obtain the vehicle's departure time and related probability.

[0053] Specifically, as an explanation, the aforementioned historical data needs to be preprocessed and analyzed to ensure that the target model's predictions based on historical data are more accurate. Therefore, after obtaining the historical data from each parking lot, the following steps are taken: cleaning and standardizing the historical data. Specifically, as an example, the historical data is cleaned by structuring, deduplication, and outlier removal, filtering out invalid data, standardizing the data from different parking lots according to a unified information structure and fields, and converting vehicle entry and exit times into half-hour intervals to facilitate subsequent statistical analysis and modeling. This can be achieved using existing format conversions or pre-defined conversion rules.

[0054] And / or, optionally, after the above normalization process, classify the vehicle information in the historical data; specifically, classify the vehicle information in the historical data of the parking lot, and the vehicle types include, but are not limited to, fixed parking users, temporary parking users, and long-term parking users. And / or, optionally, after the above classification / normalization process, calculate and analyze the parking duration data corresponding to each vehicle information in the historical data to predict the departure time and departure probability of each vehicle in the target time period of the corresponding parking lot. It should be noted that the vehicle parking duration data includes, but is not limited to, the calculation and analysis of the average, median, standard deviation, variance, and quartiles of the corresponding vehicle parking duration. In addition to the calculation and analysis of the above parameters, the parking records of different types of vehicles can also be summarized and statistically analyzed to calculate indicators such as the number of parkings, average parking duration, and total parking duration to obtain an overview of the parking behavior of vehicles in the parking lot.

[0055] It's necessary to supplement this information. The aforementioned mean, median, standard deviation, variance, and quartiles can all characterize parking distribution habits to some extent, helping us understand the central trend and dispersion of parking duration distribution. These can be used as primary statistical measures. In addition, other analytical methods can be employed. Specifically, for example, histograms can be used to view the distribution of parking duration. Histograms can be drawn for parking duration data of different vehicle types to observe their overall distribution characteristics. Box plots can also be used to view the distribution of parking duration. Drawing box plots for parking duration of different vehicle types clearly shows the quartiles, extreme values, and outliers of the duration distribution. Analyzing the variance of parking duration and comparing the variances of different vehicle types can illustrate the differences in the dispersion of parking duration. Line graphs can be used to observe the changing trends of parking duration. For parking duration data with time attributes, line graphs can be drawn to observe the changing trends and analyze the differences in vehicle parking duration over different time periods. All of the above analyses of historical data can help generate parameters of different dimensions within the model during subsequent training, thereby improving the accuracy of the model's predictions.

[0056] In addition, outlier analysis can be performed to identify and address anomalies in parking durations for different vehicle types, exploring their potential causes such as erroneous recordings or special vehicle usage needs. This allows for further adjustments to model parameters or dimensional data, ensuring comprehensive and rich historical data for improved training results. The aforementioned data analysis can be conducted using tools like SuperSet and Hive data warehouses, providing data references for later predictions of vehicle departure times.

[0057] S30: Establish a parking scoring model to score each target parking lot in order to recommend parking lots for the target vehicle; wherein, the parking scoring model is used to score each target parking lot by obtaining real-time data from the real-time parking database.

[0058] Specifically, the establishment of a scoring model to score each target parking lot mainly includes:

[0059] The system obtains the real-time number of available parking spaces for each target parking lot, and generates a first score for each target parking lot based on the proportion of real-time available spaces to the total number of parking spaces and a first weight. The system calculates the real-time number of available spaces for each parking lot; a higher real-time number of available spaces results in a higher first score, indicating that it is easier to find a parking space. Specifically, the first score is calculated using the formula: First weight × (Real-time number of available spaces / Total number of parking spaces). For example, if the first weight is 50%, there are 5 real-time available spaces, and there are 200 total parking spaces, then the first score is: 50% × (5 / 200) = 1.25%.

[0060] Based on the target model corresponding to each target parking lot, the number of vehicles leaving each target parking lot and the departure probability of each leaving vehicle during the target time period are determined from the real-time parking database. A second score for each target parking lot is calculated based on a second weight multiplied by a preset first rule, where the preset first rule is the average departure probability of each leaving vehicle multiplied by the ratio of the number of leaving vehicles to the number of real-time available parking spaces. The higher the second score, the more vehicles leaving the parking lot. Specifically, the calculation is based on the formula: Second weight × (Sum of departure vehicle probabilities / Number of leaving vehicles) × (Number of leaving vehicles / (Total number of parking spaces - Real-time available parking spaces)). For example: If the second weight is 40%, and the target model is used for prediction, and the vehicle is pre-set to have a departure probability exceeding 50%, and it is determined that 5 vehicles will leave, with their respective departure probabilities of 90%, 80%, 70%, 70%, and 60%, then the score is: 40% × ((90% + 80% + 70% + 70% + 60%) / 5) × (5 / (200 - 5)) = 0.76%.

[0061] Vehicles within a preset range for each target parking lot are retrieved from the real-time parking database. The number of vehicles entering each target parking lot during the target time period is determined. The location of each entering vehicle is obtained, and a third score for each target parking lot is calculated based on a third weight multiplied by a preset second rule. The preset second rule is the ratio of the number of entering vehicles to the number of available spaces and the number of departing vehicles. The number of vehicles about to enter during the target time period is obtained from the real-time parking database; the higher the number, the lower the third score. Considering entering vehicles when recommending parking lots can reduce the situation where there are available spaces initially, but they are already occupied by other vehicles upon arrival. Specifically, the score is calculated using the formula: negative weight × (number of entering vehicles / (number of available spaces + number of departing vehicles)). For example, if the third weight is 5% and 5 vehicles are about to enter, the score is -5% × (5 / (5 + 5)) = -2.5%.

[0062] The distances of each target parking lot relative to the target vehicle and the destination are determined from the location information of each target parking lot in the real-time parking database, and a fourth score is calculated for each target parking lot based on the fourth weight. The distance required to travel from the parking lot to an available space is obtained from the real-time parking database; the closer the distance, the sooner a newly released space can be obtained. Specifically, the score is calculated using the formula: Fourth weight / (maximum travel distance of the vehicle in the target time period × distance already traveled by the vehicle) - Fourth weight / (maximum walking distance to the destination × distance required to walk to the destination). For example, if the fourth weight is 5%, the maximum travel distance of the vehicle in the target time period is 15 kilometers, the farthest distance between the destination and the target parking lot is 1 kilometer, and the distance between the parking lot and an available space is 0.5 kilometers, then the score is (5% / 15 × (15-1)) - (5% / 1 × (1-0.5)) = 2.17%.

[0063] The overall score for each target parking lot is determined based on its first, second, third, and fourth scores. Specifically, for example, the overall score can be determined by adding or weighting the four scores together. The weights are adjusted based on vehicle data optimization. Optionally, to facilitate comparison of overall scores for parking lot recommendations, normalization can be performed, such as converting the scores to an overall score of 0-10.

[0064] Optionally, the scoring of each target parking lot can be repeated at predetermined time intervals to update the parking lot recommendation for the target vehicle. In this embodiment, to maintain the real-time nature of the recommendations and reduce the problem of inaccurate recommendations due to changes in parking lot availability over a period of time, the scoring of each target parking lot can be repeated at predetermined time intervals to update the parking lot recommendation for the target vehicle. For example, when a vehicle needs to go to a certain parking lot, Spark or Flink is used to call the parking scoring model and input relevant data to obtain the parking lot ranking, which is updated every 5 minutes.

[0065] S40: Obtain the location information of the target vehicle, and determine at least one target parking lot in the real-time parking database based on the location information of the target vehicle and the score.

[0066] In this embodiment, the aforementioned parking lots may include several, forming a parking lot network within a certain area. This network includes all (target) parking lots within the area. Similarly, parking lot recommendations can be made for all vehicles within the area, achieved by combining parking lot and vehicle GPS navigation. A scoring model is also built based on this to score each target parking lot. It should be understood that since parking lots farther away do not need to be included in the scoring when the target vehicle is located in different locations, a pre-defined area range can be set. That is, recommendations are made based on parking lots within the preset range, centered on the target vehicle, combined with the scores. This involves pre-selecting the recommended target parking lots.

[0067] It should be noted that the above scoring model is based not only on the various related data in the real-time parking database, but also on the location information of each vehicle. This allows the identification of vehicles located around each parking lot, and then the prediction of vehicles that may enter the corresponding parking lot during the target time period. Specifically, this can be achieved by directly obtaining the location information of the target vehicles. As mentioned above, recommendations can be made for different vehicles. By using different vehicles as target vehicles, the location information of all vehicles in the area can be obtained. This allows the identification of which vehicles are located within the preset range of which parking lots, thus establishing a dynamically changing vehicle distribution network.

[0068] Based on the above, after scoring each parking lot, the parking lots are ranked according to the obtained comprehensive scores, and then priorities are set to determine at least one target parking lot for parking lot recommendation. In this embodiment, a target model corresponding to each parking lot is established to predict the distribution of parking spaces in each parking lot during the target time period, including but not limited to real-time parking spaces, newly added parking spaces generated by vehicles leaving the parking lot, and parking spaces occupied by vehicles entering the parking lot. When recommending parking lots for target vehicles, the real-time number of parking spaces, location information, number of vehicles leaving the parking lot and number of vehicles entering the parking lot during the target time period are scored according to the predicted distribution of parking spaces in each parking lot. The scoring fully considers the real-time changes in the distribution of parking spaces, improves the accuracy of recommendations, and effectively reduces the user's parking waiting time.

[0069] As an optional further step in this embodiment, to further improve the accuracy of parking lot recommendation results, model optimization can be performed based on user feedback. Specifically, the evaluation information from the target vehicle is received, and the parking scoring model and each of the target models are optimized in reverse based on the evaluation information. For example, the model can be updated based on the parking lot ultimately selected by the target vehicle, or a scoring feedback questionnaire can be set up on the user's end, thereby adjusting the weights / evaluation dimensions of the scoring model and / or the parameters of each target model. This reverse optimization continuously improves the accuracy and matching degree of subsequent recommendation results, reducing user waiting time.

[0070] Example 2: This embodiment provides a parking lot recommendation device 6 based on historical data, executing the parking lot recommendation method based on historical data described in the above embodiment. (See attached document for details.) Figure 3 ,include:

[0071] The first processing module 61 is used to pre-establish an initial model set, which includes several initial models corresponding to each parking lot; acquire historical data of each parking lot as training data, train each initial model to obtain a target model set, and each target model is used to predict the departure time and departure probability of each vehicle in the corresponding parking lot during a target time period, so as to determine the vehicle distribution status of the parking lot during the target time period.

[0072] It should be noted that each parking lot corresponds to a target model, which is used to predict the accuracy of each vehicle entering the parking lot. Historical data is collected using a unified interface and format for different parking lots and then trained to obtain the target model. The historical data can also be preprocessed or analyzed to further optimize the accuracy of the target model.

[0073] The second processing module 62 is used to acquire real-time data from each parking lot and establish a dynamically changing real-time parking database based on the target model set.

[0074] Specifically, programs such as Spark or Flink can be used to synchronize real-time data from various parking lots and establish a dynamically changing real-time parking database. This real-time parking database can include, but is not limited to, real-time data of each parking lot (such as real-time parking distribution), changes in parking distribution in parking lots during the target time period (based on predictions from various target models), and remaining vacant spaces during the target time period, so that subsequent scoring can be based on this parking lot.

[0075] Execution module 63 is used to establish a parking scoring model to score each target parking lot in order to recommend parking lots for the target vehicle; wherein, the parking scoring model is used to obtain real-time data of each target parking lot from the real-time parking database for scoring; obtain the location information of the target vehicle, and determine at least one target parking lot in the real-time parking database based on the location information of the target vehicle and the score.

[0076] Specifically, the execution module scores each target parking lot based on four dimensions: the number of available parking spaces in real time, location information, the number of vehicles leaving and entering during the target time period. Different weights are set for different dimensions, and the comprehensive scores calculated for different dimensions are used to sort each target parking lot. Then, priorities are set to determine the target parking lots and make recommendations.

[0077] This implementation method, based on various modules, combines historical data modeling and analysis of each parking lot with real-time parking space availability and nearby vehicle information to generate a comprehensive parking score and real-time ranking (parking lot recommendation). It can also be updated in real-time (at preset time intervals) to maintain the real-time nature of the recommendations. Furthermore, the model can be optimized based on user feedback to improve accuracy, thereby enhancing the accuracy of the recommendation results, effectively reducing user parking waiting time, and increasing parking lot utilization and regional pedestrian traffic.

[0078] Example 3: This invention also provides a distributed system, including a master node and multiple slave nodes. The master node and each slave node cooperate to apply the parking recommendation device based on historical data described in Example 2 above to execute the parking recommendation method based on historical data described in Example 1. Using a distributed system can effectively alleviate data processing pressure, improve data processing efficiency, and enhance data storage security.

[0079] As an example, the master node can be used to configure the operations of the first processing module, the second processing module, and the execution module in Embodiment 2 above. Specifically, it is configured as a master server, used to predict the departure time and departure probability of each vehicle in a target time period of a corresponding parking lot using a target model, and to establish a dynamically changing parking lot database; to establish a scoring model to score each target parking lot in order to recommend parking lots for target vehicles. Specifically, it obtains the location information of the target vehicles, and determines at least one target parking lot in the real-time parking database based on the location information of the target vehicles and the scores. The slave nodes may include a first slave node corresponding to each parking lot and a second slave node corresponding to each vehicle. As an example, the first slave node obtains real-time data and historical data of the parking lot; the second slave node collects the location information of the vehicles and synchronizes it to the first slave node and the master node; the master node obtains the data from each slave node for scoring and recommendation; the processing of the target model can also be arranged on the slave nodes to share the data processing pressure.

[0080] This distributed system models and analyzes historical data from various parking lots, combining it with real-time data to recommend parking spaces for target vehicles. This approach accurately reflects changes in parking space availability and reduces user waiting time.

[0081] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A parking lot recommendation method based on historical data, characterized in that comprising: pre-establishing an initial model set, the initial model set comprising a plurality of initial models corresponding to respective parking lots; obtaining historical data of each parking lot as training data, training each of the initial models to obtain a target model set, the historical data comprising vehicle feature information, entry time, parking time, parking space number, and exit time of all vehicles entering the parking lot in a historical time period, the historical data being cleaned and normalized, each vehicle information in the historical data being classified to determine the vehicle type, and the average, median, standard deviation, variance, and quartile data of the parking time of each vehicle information in the historical data being calculated and analyzed, each target model being used to predict the exit time and exit probability of each vehicle in a target time period of the corresponding parking lot to determine the vehicle distribution state of the parking lot in the target time period, the target model performing the following operations: collecting basic information and vehicle information of the entering vehicle of the corresponding parking lot, identifying the vehicle type based on the vehicle information, and processing the vehicle type and the basic information based on a trained long short-term memory model to obtain the exit time and the exit probability corresponding to the vehicle information in the target time period, wherein the target model is a long short-term memory model; obtaining real-time data of each parking lot, updating the real-time parking database by synchronizing the data changes of the parking lot database to the real-time parking database through Spark or Flink, and establishing a dynamically changing real-time parking database based on the target model set; establishing a parking score model to score each target parking lot and recommend a parking lot to a target vehicle; wherein the parking score model is used to score each target parking lot based on the real-time data obtained from the real-time parking database, which includes: obtaining the real-time empty space number of each target parking lot, and generating a first score of each target parking lot based on the proportion of the real-time empty space number in the total number of parking spaces and the first weight; determining the number of vehicles leaving the target time period and the exit probability of each vehicle leaving the target time period from the real-time parking database based on the target model corresponding to each target parking lot, calculating a second score of each target parking lot based on a second weight multiplied by a preset first rule, wherein the preset first rule is the average of the exit probability of each vehicle leaving multiplied by the ratio of the number of vehicles leaving to the real-time non-empty space number; obtaining vehicles within a preset range of each target parking lot from the real-time parking database, determining the number of entering vehicles in the target time period of each target parking lot, and calculating a third score of each target parking lot based on a third weight multiplied by a preset second rule; wherein the preset second rule is the proportion of the number of entering vehicles to the real-time empty space number and the number of vehicles leaving. determine distances of each of the target parking lots from the target vehicle and the destination according to location information of each of the target parking lots from the real-time parking database, and calculate fourth scores of each of the target parking lots based on a fourth weight value; determine comprehensive scores of each of the target parking lots based on the first scores, the second scores, the third scores, the fourth scores of each of the target parking lots, and a preset third rule; obtain location information of a target vehicle, and determine at least one target parking lot in the real-time parking database according to the location information of the target vehicle and the scores.

2. The parking lot recommendation method according to claim 1, characterized by, Each of the target models performs the following operations: collect basic information and vehicle information of an incoming vehicle of a corresponding parking lot, wherein the basic information includes weather data and time data; identify a vehicle type based on the vehicle information, wherein the vehicle type includes a fixed parking user, a temporary parking user, and a long-term parking user; process the vehicle type and the basic information by using a trained target model to obtain the leaving time and the leaving probability of the vehicle information corresponding to the target time period, wherein the target model is a long short-term memory model.

3. The parking lot recommendation method according to claim 1, characterized by, After obtaining historical data of each parking lot, the following operations are performed: perform cleaning and normalization processing on the historical data; and / or, classify each vehicle information in the historical data; and / or, calculate and analyze vehicle parking duration data corresponding to each vehicle information in the historical data to predict the leaving time and the leaving probability of each vehicle in a target time period of a corresponding parking lot.

4. The parking lot recommendation method of claim 1, wherein: the scoring of each of the target parking lots is performed again at a predetermined time interval to update the parking lot recommendation for the target vehicle.

5. The parking lot recommendation method of claim 1, wherein: evaluation information fed back by the target vehicle is received, and the parking score model and each of the target models are optimized inversely according to the evaluation information.

6. A parking lot recommendation device based on historical data, comprising: The first processing module is configured to pre-establish an initial model set, the initial model set including a plurality of initial models corresponding to respective parking lots, obtain historical data of the respective parking lots as training data, train the respective initial models to obtain a target model set, the historical data including vehicle feature information, entry time, parking time, parking space number, and exit time of all vehicles entering the parking lots in a historical time period, the historical data being cleaned and normalized, vehicle types being determined by classifying the vehicle information in the historical data, and average value, median value, standard deviation, variance, and quartile data of parking time of each vehicle in the historical data being calculated and analyzed, each target model being configured to predict exit time and exit probability of each vehicle in a target time period of a corresponding parking lot to determine vehicle distribution state of the parking lot in the target time period, the target model being configured to collect basic information and vehicle information of an entering vehicle of the corresponding parking lot, identify the vehicle type based on the vehicle information, and process the vehicle type and the basic information based on a trained long short-term memory model to obtain the exit time and the exit probability corresponding to the vehicle information in the target time period, wherein the target model is a long short-term memory model; The second processing module is configured to obtain real-time data of the respective parking lots, update the real-time parking database based on data change of a parking database by synchronizing to the real-time parking database through Spark or Flink, and establish a dynamically changing real-time parking database based on the target model set; The execution module is configured to establish a parking score model, score the respective target parking lots, and recommend a target vehicle to a parking lot, including: obtaining real-time empty space number of the respective target parking lots, and generating a first score of each of the target parking lots based on a proportion of the real-time empty space number to total parking space number and a first weight value; determining a number of exiting vehicles and an exit probability of each exiting vehicle of each of the target parking lots in the target time period based on the target model corresponding to each of the target parking lots from the real-time parking database, and calculating a second score of each of the target parking lots based on a second weight value and a preset first rule, wherein the preset first rule is an average value of the exit probability of each exiting vehicle multiplied by a ratio of the number of exiting vehicles to real-time non-empty space number; obtaining vehicles within a preset range of each of the target parking lots from the real-time parking database, determining a number of entering vehicles of each of the target parking lots in the target time period, obtaining a position of each entering vehicle, and calculating a third score of each of the target parking lots based on a third weight value and a preset second rule, wherein the preset second rule is a proportion of the number of entering vehicles to real-time empty space number and the number of exiting vehicles; determining a distance of each of the target parking lots from a target vehicle and a destination based on position information of each of the target parking lots from the real-time parking database, and calculating a fourth score of each of the target parking lots based on a fourth weight value. The comprehensive score of each target parking lot is determined based on the first score, the second score, the third score, the fourth score of each target parking lot and a preset third rule, so as to recommend a parking lot for a target vehicle; wherein the parking score model is used to score real-time data of each target parking lot obtained from the real-time parking database; position information of a target vehicle is obtained, and at least one target parking lot is determined in the parking lot database based on the position information of the target vehicle and the score.

7. A distributed system, characterized by: The parking lot recommendation device based on historical data comprises a master node and a plurality of slave nodes, and the master node and each slave node cooperate to apply the parking lot recommendation device based on historical data of claim 6.

Images