Intelligent parking process optimization scheduling method and system

Abstract

The invention discloses an intelligent parking process optimization scheduling method and system, and relates to the technical field of vehicle management, and the scheduling steps are as follows: S1, a user selects a target place parking lot for navigation, and obtains entrance and exit vehicle data in the current parking lot based on a parking lot entrance and exit sensor when approaching the target place parking lot; the vehicle data are converted into congestion indexes, and judgment and analysis are carried out; and S2, setting a dynamic bearing threshold value for each entrance and exit, and when the number of recommended vehicles is close to 80% of the bearing threshold value, reducing the recommendation priority of the entrance and exit, and recommending a secondary optimization item. According to the invention, real-time congestion data and dynamic recommendation are provided to reduce waiting and detour, vehicle-machine visual display improves operation convenience, parking experience is optimized, dynamic threshold adjustment balances entrance and exit loads, an LSTM model assists in prejudgment of congestion, management efficiency is enhanced, meanwhile, the parking space turnover rate is improved, benefits are increased, and pressure is relieved for regional traffic and reduction of surrounding invalid traffic flows.

Description

Technical Field

[0001] This invention relates to the field of vehicle management technology, specifically to a smart parking process optimization scheduling method and system. Background Technology

[0002] With the acceleration of urbanization and the surge in motor vehicle ownership, the problem of urban parking difficulties has become increasingly prominent, becoming a key bottleneck affecting traffic efficiency and residents' travel experience. In the current parking scenario, traditional parking lots often rely on fixed entrance and exit guidance signs or static parking space information, which are difficult to respond to the dynamic changes in vehicle flow in real time. This leads to vehicles being stranded when approaching the parking lot due to entrance and exit congestion and unreasonable recommended routes, which not only exacerbates traffic pressure on surrounding roads but also reduces parking efficiency. The parking process includes managing vehicle entry and exit and optimizing parking route scheduling. While some existing smart parking systems can collect parking space data through sensors, they have significant limitations in optimizing entrance and exit scheduling: First, most systems only provide overall parking space availability information, lacking accurate monitoring of real-time vehicle flow and congestion assessment at each entrance and exit, making it difficult for users to obtain targeted entrance and exit selection suggestions when approaching the parking lot; Second, entrance and exit recommendation strategies are mostly based on static rules, failing to consider the differences in vehicle growth rates at different times, and when an entrance or exit approaches saturation, the recommendation priority cannot be dynamically adjusted in a timely manner, easily leading to increased local congestion; Third, they lack the ability to predict entrance and exit congestion trends in the near future, making it difficult to guide vehicles to avoid potential congestion points in advance, resulting in delayed scheduling response. To address these issues, we provide a smart parking process optimization scheduling method and system. Summary of the Invention

[0003] To address the aforementioned technical problems, this paper provides a method and system for optimizing the scheduling of intelligent parking processes. This technical solution resolves the problems mentioned above.

[0004] To achieve the above objectives, the technical solution adopted by this invention is: a smart parking process optimization scheduling method, the scheduling steps of which are as follows: S1. When the user selects a target parking lot for navigation, and approaches the target parking lot, the system acquires vehicle data at the current parking lot entrance and exit based on the parking lot entrance and exit sensors, converts the vehicle data into congestion indicators, and performs judgment and analysis. S2. Set a dynamic carrying capacity threshold for each entrance / exit. When the number of recommended vehicles approaches 80% of the carrying capacity threshold, reduce the recommendation priority of that entrance / exit and recommend the next best option. S3. Based on the current vehicle growth rate, predict the congestion status of each entrance and exit in the next 3-5 minutes using the LSTM time series prediction model, formulate dynamic recommendation rules, and make dynamic adjustments to the recommendations. S4. The optimized scheduling process is based on real-time display on the vehicle's in-vehicle system and recommends routes for users to follow.

[0005] Preferably, the judgment and analysis step in step S1 is as follows: Coil loops are embedded in the ground of the entrance and exit lanes to detect vehicles entering and exiting through electromagnetic induction; Cameras are installed above the entrances and exits to identify license plates and vehicle models using computer vision, and to track queue length and traffic status in real time. Record the number of times the barrier gate is opened and closed, and obtain the number of vehicles passing through the entrance and exit within a unit of time; Calculate the current passage efficiency and congestion index of the entrance / exit per unit time; The congestion level is used to determine the current congestion situation at the entrance / exit.

[0006] Preferably, the traffic efficiency is calculated by dividing the total number of vehicles passing through the entrance / exit within a unit of time by the unit of time. The congestion index is calculated using a two-part weighted average. The first part is calculated using the following formula: Where Q is the current number of vehicles in the queue, N is the number of entrance and exit lanes, K is the maximum reasonable number of queues per lane, W1 is the weight of the impact of queuing vehicles, with a value range of 50-70, and A is the value calculated in the first part. The calculation formula for the second part is: Where B is the value calculated in the second part. For current traffic efficiency, For the sake of historical efficiency, The weight for the impact on traffic efficiency ranges from 30 to 50. Add the values ​​of A and B to get the congestion index. The higher the congestion index, the more vehicles are at the current entrance and exit and the more congested it is. Establish a congestion index range, including smooth flow, congested, and severe congestion. Substitute the obtained congestion index into the congestion index range for classification to obtain the current congestion situation at the entrance / exit. The range of the congestion index was determined through a combination of data modeling and manual calibration.

[0007] Preferably, the dynamic load threshold determination step in step S2 is as follows: The dynamic carrying capacity threshold is dynamically adjusted based on the hardware conditions and scenarios at entrances and exits. The basic carrying capacity threshold = number of lanes × maximum traffic volume per lane per unit time × α, where α is a buffer coefficient, a safety buffer coefficient calculated based on historical data, with a value range of 1.1 to 1.3. The carrying capacity threshold is dynamically adjusted in combination with factors such as the time of day, weather, and scenario. If the traffic efficiency decreases by β1 during the morning rush hour based on historical data analysis, the dynamic carrying capacity threshold is adjusted down by β1 accordingly. If the efficiency decreases by β2 during heavy rain, the dynamic threshold is also adjusted down by β2. If there is a large event around the target location and the traffic efficiency decreases by β3, the dynamic threshold is also adjusted down by β3. Here, β1 is the time period correction coefficient, β2 is the weather correction coefficient, and β3 is the special scenario correction coefficient.

[0008] Preferably, the next preferred step in step S2 is: The system monitors recommended vehicles in real time, including vehicles on navigation and arriving within 5 minutes, social vehicles within 500 meters heading towards the entrance / exit, and vehicles that have booked the entrance / exit within the next 5 minutes. The number of vehicles is added up to obtain the current number of recommended vehicles. The system updates the data every 30 seconds to reflect the traffic flow in real time. When the number of recommended vehicles reaches 80% of the dynamic capacity threshold for the entrance / exit on that day, the recommendation weight of that entrance / exit is reduced, and the next best entrance / exit is selected for recommendation.

[0009] Preferably, the specific steps in step S3 are as follows: Indicators are collected every 10 seconds, including the current number of vehicles queuing at each entrance / exit, the number of vehicles increasing, the passage efficiency of the entrance / exit, the number of vehicles en route to that entrance / exit, and vehicles that can be reached within 3 minutes by tracking through the navigation system. Based on the LSTM time series prediction model, input the vehicle growth rate and traffic efficiency data for the past 5 minutes, and output the predicted number of queuing vehicles and the predicted congestion index for each entrance and exit in the future time. The forecast results are corrected by combining real-time weather and time-of-day characteristics; The predicted congestion index is compared with the dynamic capacity threshold. If it is lower than the dynamic capacity threshold, the recommendation continues; if it is higher than the dynamic capacity threshold, the recommended entrance / exit is changed. Based on the predictions of the predictive model, different recommendation strategies are formulated for each entrance and exit, including: Prioritized recommendation level; expected to be smooth operation in the future; navigation will be recommended first at the current entrance / exit. Recommended access level suspended. Severe congestion is expected in the future, triggering a diversion warning. The current entrance / exit will be changed.

[0010] Preferably, the LSTM time series prediction model specifically predicts by preprocessing the collected data, building and training the model, designing an architecture containing an input layer, two LSTM hidden layers and an output layer, training with data from the past three months, adjusting parameters, using a validation set to validate the output results, and judging that the model meets the standard when the validation result MSE < 5%, and then using the model for prediction. During the prediction execution phase, a prediction is triggered once every minute. The preprocessed data is input and the prediction results and confidence scores for future time periods are output. The accuracy of the prediction results is judged based on the confidence scores. In the result correction stage, the results are corrected by taking into account weather and time period characteristics, and the corrected results are output. The input layer receives preprocessed time series data, and the input dimension is consistent with the set time step and the number of features. The number of LSTM units in the first layer of the LSTM hidden layer is set to 64, and the number in the second layer is set to 32, both using the ReLU activation function; The output layer is a fully connected layer, and the output dimension is the number of predicted targets.

[0011] Preferably, the confidence level is calculated by measuring the standard deviation σ of the model's prediction error distribution on the validation set, and the confidence level is defined by the relationship between the predicted value and the standard deviation. Let the model prediction value be The actual value is The error on the validation set is The mean error is The standard deviation is Predicted values ​​for new samples Its confidence level is defined as: in is a constant, taken as 1.96, where predictions with a confidence level greater than 80% are directly used as the basis for decision-making, while results with a confidence level less than 50% are re-evaluated.

[0012] Preferably, in step S4, the optimized scheduling process uses the vehicle's in-vehicle system as the core interaction carrier, and makes decisions through real-time data flow and intelligent algorithms.

[0013] A smart parking process optimization scheduling system, the scheduling system includes: The data acquisition module is configured to acquire vehicle data at the current parking lot entrances and exits, convert the vehicle data into congestion indicators, perform judgment and analysis, and display and dispatch the data in real time based on the vehicle's infotainment system. The recommendation module is configured to make intelligent recommendations for entrances and exits based on dynamic capacity thresholds; The dynamic adjustment module predicts the congestion status of each entrance and exit in the next 3-5 minutes based on the LSTM time series prediction model, formulates dynamic recommendation rules, and makes dynamic adjustments to the recommendations.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes real-time congestion data and dynamic recommendations to reduce waiting and detours, intuitive in-vehicle displays to improve ease of operation and optimize the parking experience, dynamic threshold adjustment to balance entrance and exit loads, and LSTM models to help predict congestion, thereby enhancing management efficiency. It also increases parking space turnover, boosts revenue, reduces surrounding ineffective traffic flow to alleviate pressure on regional traffic, and reduces vehicle idling to lower energy consumption and emissions, aligning with green principles. This multi-dimensional approach achieves intelligent and efficient parking, demonstrating significant practical value. Attached Figure Description

[0015] Figure 1 This is a flowchart of the scheduling steps of the present invention; Figure 2 This is a framework diagram of the scheduling system of the present invention. Detailed Implementation

[0016] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0017] Reference Figure 1 As shown, a smart parking process optimization scheduling method includes the following scheduling steps: S1. When the user selects a target parking lot for navigation, and approaches the target parking lot, the system acquires vehicle data at the current parking lot entrance and exit based on the parking lot entrance and exit sensors, converts the vehicle data into congestion indicators, and performs judgment and analysis. S2. Set a dynamic carrying capacity threshold for each entrance / exit. When the number of recommended vehicles approaches 80% of the carrying capacity threshold, reduce the recommendation priority of that entrance / exit and recommend the next best option. S3. Based on the current vehicle growth rate, predict the congestion status of each entrance and exit in the next 3-5 minutes using the LSTM time series prediction model, formulate dynamic recommendation rules, and make dynamic adjustments to the recommendations. S4. The optimized scheduling process is based on real-time display on the vehicle's in-vehicle system and recommends routes for users to follow.

[0018] This application acquires real-time vehicle data at entrances and exits and converts it into congestion indicators. Users can then know the congestion status of each entrance and exit as they approach the parking lot, avoiding blindly choosing congested entrances and reducing waiting time. Based on dynamic carrying capacity thresholds and LSTM time-series prediction models, the system can recommend better entrances and exits and routes for the current and next 3-5 minutes. Users do not need to make judgments or try and fail themselves, and the navigation routes are more in line with the actual situation. The optimized scheduling process is displayed in real time through the vehicle's infotainment system, and users can clearly see the recommended routes without additional queries or operations, improving the convenience and comfort of travel. Setting dynamic carrying capacity thresholds can prevent a certain entrance / exit from exceeding its carrying capacity due to too many recommended vehicles. When the number of recommended vehicles approaches 80% of the threshold, its recommendation priority is reduced, guiding vehicles to the secondary entrance / exit, thus achieving a balanced distribution of load among entrances / exits and reducing local congestion. Through accurate prediction and dynamic adjustment of recommendation rules, vehicles can enter the parking lot more quickly, reducing their stay at entrances / exits, improving the parking lot's vehicle turnover efficiency, and increasing the parking lot's capacity. Intelligent scheduling reduces reliance on manual guidance, lowers parking lot management costs, and also reduces vehicle collisions caused by congestion, thus mitigating management risks. Vehicles can quickly enter the parking lot, avoiding traffic congestion on surrounding roads caused by waiting to enter, improving the traffic environment around the parking lot, and enhancing overall traffic flow. Through reasonable vehicle scheduling, the entrance and exit resources of the parking lot are utilized more efficiently, indirectly promoting the optimal allocation of traffic resources in the region.

[0019] The judgment and analysis steps in step S1 are as follows: The coils embedded in the ground of the entrance and exit lanes are usually ring-shaped induction coils. When a vehicle drives over the coil, the vehicle's metal body changes the electromagnetic field around the coil, causing a change in the coil's inductance value. The system detects this change to determine that a vehicle has entered. Similarly, when the vehicle leaves the coil area, the electromagnetic field returns to its original state, and the inductance value returns to its initial state, thus identifying the vehicle's exit. This method can accurately capture the entry and exit movements of vehicles, providing basic data for traffic flow statistics. It is also unaffected by weather, lighting, or external environmental conditions, and has strong stability. Cameras installed above the entrances and exits capture real-time images of the lanes. Using computer vision technology, the system first locates and recognizes license plate areas in the images to collect license plate information, which can be used to associate with vehicle parking records. Simultaneously, image segmentation and target tracking algorithms are used to identify vehicle types, distinguishing between small and large vehicles, providing a basis for subsequent analysis of the impact of different vehicle types on traffic efficiency. Furthermore, the system continuously analyzes video frames to calculate the number of vehicles in queues (i.e., queue length), for example, by converting the pixel distance between the first and last positions of vehicles into actual distance, or by directly counting the number of vehicles in the image. It can also determine the traffic status of vehicles, such as whether they are queuing or passing through a barrier gate, thus providing real-time monitoring of traffic flow dynamics at the entrances and exits. The opening and closing of the barrier gate is directly related to vehicle passage. Each opening-closing cycle usually means that a vehicle has completed the entry or exit operation. By recording the number of times the barrier gate is opened and closed within a unit of time, the number of vehicles passing through the entrance and exit within that time period can be directly obtained. This data complements the coil detection and camera statistics, verifying traffic flow information from multiple dimensions and improving data accuracy. Traffic efficiency refers to the number of vehicles actually passing through an entrance / exit per unit time. The calculation formula is: total number of vehicles passing through ÷ statistical time. The higher the value, the stronger the traffic capacity of the entrance / exit. The congestion index is an indicator calculated by comprehensively considering factors such as queue length, traffic efficiency, and average vehicle waiting time. For example, it can be set that the longer the queue length, the lower the traffic efficiency, and the longer the average waiting time, the higher the congestion index. In the specific calculation, different weights can be assigned to each factor, and the final index can be obtained by weighted summation to quantify the degree of congestion. A pre-defined threshold range for the congestion index is established, defining 0-30 as smooth traffic, 30-60 as light congestion, 60-80 as moderate congestion, and 80-100 as severe congestion. Once the congestion index of the current entrance / exit is calculated, it is compared with the pre-defined threshold to clearly determine the current congestion level, providing an intuitive basis for subsequent recommendations and scheduling.

[0020] Traffic efficiency is calculated by dividing the total number of vehicles passing through the entrances and exits per unit time by the unit time. The congestion index is calculated using a two-part weighted average. The first part is calculated using the following formula: Where Q is the current number of vehicles in the queue, N is the number of entrance and exit lanes, K is the maximum reasonable number of queues per lane, W1 is the weight of the impact of queuing vehicles, with a value range of 50-70, and A is the value calculated in the first part. The calculation formula for the second part is: Where B is the value calculated in the second part. For current traffic efficiency, For the sake of historical efficiency, The weight for the impact on traffic efficiency ranges from 30 to 50. Add the values ​​of A and B to get the congestion index. The higher the congestion index, the more vehicles are at the current entrance and exit and the more congested it is. Establish a congestion index range, including smooth flow, congested, and severe congestion. Substitute the obtained congestion index into the congestion index range for classification to obtain the current congestion situation at the entrance / exit. The range of the congestion index was determined through a combination of data modeling and manual calibration.

[0021] By using historical congestion data, such as actual queue length, traffic efficiency, and user complaint rate corresponding to different indices, models can be built to classify thresholds for smooth flow, congestion, and severe congestion based on real patterns, avoiding subjective biases from human experience, such as avoiding setting ranges solely based on the feelings of managers. Data modeling may overlook special scenarios, such as extreme weather or abnormal congestion caused by temporary traffic control, while manual calibration can adjust the threshold based on actual operational experience, appropriately lower the congestion threshold for rainy or snowy days, provide early warnings, make the scope division more in line with on-site management needs, ensure that the index can effectively guide scheduling, and clearly indicate that vehicles should be immediately guided to other entrances and exits when there is severe congestion.

[0022] The steps for determining the dynamic load threshold in step S2 are as follows: The dynamic carrying capacity threshold is dynamically adjusted based on the hardware conditions and scenarios at entrances and exits. The basic carrying capacity threshold = number of lanes × maximum throughput of a single lane per unit time × α, where α is a buffer coefficient, a safety buffer coefficient calculated based on historical data, with a value range of 1.1 to 1.3. The carrying capacity threshold is dynamically corrected by combining the time of day, weather, and scenario factors. If the traffic efficiency decreases by β1 during the morning peak based on historical data analysis, the dynamic carrying capacity threshold is adjusted down by β1 accordingly. If the efficiency decreases by β2 during heavy rain, the dynamic threshold is also adjusted down by β2. If the traffic efficiency decreases by β3 due to large-scale events around the target location, the dynamic threshold is also adjusted down by β3. Here, β1 is the time period correction coefficient, β2 is the weather correction coefficient, and β3 is the special scenario correction coefficient. The time of day, weather, and surrounding environment directly affect vehicle traffic efficiency. By simultaneously lowering the dynamic capacity threshold, the threshold is matched with the actual traffic capacity. During the morning rush hour, due to the surge in traffic and drivers' impatient operation, historical data shows that traffic efficiency has decreased by 15%. If the basic threshold is still used at this time, it may lead to vehicle congestion due to insufficient actual traffic capacity. Therefore, the threshold is simultaneously lowered by 15%, which is equivalent to reducing the upper limit of the recommended number of vehicles and guiding vehicles to divert in advance. During heavy rain, drivers may slow down and the sensitivity of gate sensors may be affected, resulting in a 20% decrease in traffic efficiency. After the threshold is lowered accordingly, it can prevent recommended vehicles from exceeding the actual traffic flow capacity of the entrance and exit. When there are large-scale events around the target location, the influx of vehicles in a short period of time will cause a sudden increase in pressure on the entrances and exits, reducing traffic efficiency by 60%. A significant reduction in the threshold can prioritize the passage of existing vehicles and prevent complete blockage. This linkage mechanism, in which the threshold is adjusted according to the amount of efficiency reduction, ensures that the threshold always matches the carrying capacity limit of the real-time scenario. Unlike fixed thresholds, this approach, based on hardware infrastructure and scenario-based adjustments, allows the carrying capacity threshold to dynamically change according to actual conditions. For the same entrance / exit, the dynamic threshold during the morning rush hour is lower than that during off-peak hours, lower during heavy rain than during sunny days, and much lower when there are events nearby than on a normal day. This ensures that the rule of reducing priority when the number of recommended vehicles is close to 80% of the threshold can be accurately applied in any scenario, preventing congestion due to excessively high thresholds and resource waste due to excessively low thresholds.

[0023] The recommended second-most-priority selection step in step S2 is: The system monitors recommended vehicles in real time, including vehicles on navigation and arriving within 5 minutes, social vehicles within 500 meters heading towards the entrance / exit, and vehicles that have booked the entrance / exit within the next 5 minutes. The number of vehicles is added up to obtain the current number of recommended vehicles. The system updates the data every 30 seconds to reflect the traffic flow in real time. When the number of recommended vehicles reaches 80% of the dynamic capacity threshold for the entrance / exit on that day, the recommendation weight of that entrance / exit is reduced, and the next best entrance / exit is selected for recommendation.

[0024] The real-time monitoring scope of this application not only includes vehicles that are about to arrive according to the navigation, but also includes social vehicles within 500 meters of the entrance and exit and vehicles that have made reservations within the next 5 minutes. It covers all traffic flow that may flood into the entrance and exit in a short period of time. This full-dimensional statistics avoids omissions caused by relying solely on navigation data and ensures that the number of recommended vehicles can truly reflect the real-time pressure at the entrance. The high-frequency refresh mechanism that updates the data every 30 seconds can dynamically capture changes in traffic flow, so that the recommendation decision is always based on the latest situation and reduces misjudgments caused by data lag. When the number of recommended vehicles reaches 80% of the dynamic carrying capacity threshold, the recommendation priority is reduced, rather than waiting until the threshold limit is reached. This is equivalent to setting an early warning line. This design takes advantage of the fact that the dynamic carrying capacity threshold itself includes scenario adaptation (such as peak hours and weather effects). It can guide vehicles to the secondary option in advance before the entrance and exit are close to saturation, avoiding the risk of vehicles exceeding the carrying capacity limit after a concentrated influx. This reduces the risk of congestion from the source. The dynamic carrying capacity threshold of a certain entrance and exit is 100 vehicles / hour. When the number of recommended vehicles reaches 80, the traffic is diverted. This can not only retain a 20% buffer space to deal with sudden traffic flow, but also balance the pressure of each entrance and exit through the diversion of the secondary option.

[0025] The specific steps in step S3 are as follows: Indicators are collected every 10 seconds, including the current number of vehicles queuing at each entrance / exit, the number of vehicles increasing, the passage efficiency of the entrance / exit, the number of vehicles en route to that entrance / exit, and vehicles that can be reached within 3 minutes by tracking through the navigation system. Based on the LSTM time series prediction model, input the vehicle growth rate and traffic efficiency data for the past 5 minutes, and output the predicted number of queuing vehicles and the predicted congestion index for each entrance and exit in the future time. The forecast results are corrected by combining real-time weather and time-of-day characteristics; The predicted congestion index is compared with the dynamic capacity threshold. If it is lower than the dynamic capacity threshold, the recommendation continues; if it is higher than the dynamic capacity threshold, the recommended entrance / exit is changed. Based on the predictions of the predictive model, different recommendation strategies are formulated for each entrance and exit, including: Prioritized recommendation level; expected to be smooth operation in the future; navigation will be recommended first at the current entrance / exit. Recommended access level suspended. Severe congestion is expected in the future, triggering a diversion warning. The current entrance / exit will be changed.

[0026] This application collects data every 10 seconds, including the number of vehicles in queue, the increase in the number of vehicles, traffic efficiency, and the number of vehicles that can reach the entrance within 3 minutes. This high-frequency, multi-dimensional data collection is of great significance. High-frequency collection can accurately capture subtle changes in traffic flow and avoid missing key information due to long data intervals. For example, if a large number of vehicles suddenly head towards a certain entrance or exit in a short period of time, the data collection every 10 seconds can detect this trend in time. The collected indicators each have their own focus: the number of vehicles in queue reflects the current state of congestion; the increase in the number of vehicles reflects the growth rate of traffic flow and can determine whether congestion is worsening; traffic efficiency is the core manifestation of the entrance and exit's capacity to manage traffic; and the number of vehicles that can reach the entrance within 3 minutes predicts the increase in traffic flow in the near future. These data together form the basis of the LSTM model's prediction, ensuring that the input data can comprehensively reflect the dynamic changes at the entrance. LSTM (Long Short-Term Memory) networks are particularly well-suited for processing time-series data, effectively capturing long-term dependencies. Inputting vehicle growth rate and traffic efficiency data from the past 5 minutes is beneficial because these two indicators reflect the trends in entrance traffic flow and the changing patterns of the model's capacity to manage traffic. Vehicle growth rate reflects the increasing trend of traffic flow, while traffic efficiency reflects the fluctuations in the entrance and exit's ability to handle vehicles. Combining these two metrics allows the model to more accurately learn congestion evolution patterns under different conditions. The outputs predict the number of vehicles in the queue and the predicted congestion index, transforming abstract trends into concrete quantitative indicators, providing a clear basis for subsequent recommendation strategy formulation. Incorporating real-time weather and time-specific characteristics to correct predictions aims to eliminate the impact of external environments on prediction accuracy. Different weather conditions alter vehicle traffic behavior; for example, rain and snow can cause drivers to slow down, reducing traffic efficiency and making actual congestion more severe than predicted by the model based solely on historical data. Furthermore, traffic flow characteristics differ across time periods (such as morning rush hour, off-peak, and evening rush hour), and the model may not be able to fully capture these unique fluctuations. By incorporating real-time weather and time-specific characteristics for correction, predictions can better reflect real-world scenarios and improve their reliability. The logic of comparing the predicted congestion index with the dynamic capacity threshold to determine whether to continue recommending or change the entrance / exit ensures the scientific and forward-looking nature of the recommendations. When the predicted congestion index is less than the dynamic capacity threshold, it means that the entrance / exit still has sufficient capacity in the future, and continuing to recommend it can make full use of its resources. When the predicted congestion index is greater than the dynamic capacity threshold, it means that the entrance / exit may become congested in the future, and changing the recommended entrance / exit in advance can prevent vehicles from getting stuck in congestion. The priority recommendation level targets entrances and exits with smooth traffic flow in the future, making them the first choice to guide vehicles into the parking lot efficiently. The pause recommendation level targets entrances and exits with severe congestion in the future, triggering diversion warnings and changing the recommendation, which can reduce vehicle gathering at the source and avoid further congestion. The flexible application of these two recommendation strategies enables dynamic guidance of vehicles and ensures the overall traffic efficiency of the parking lot.

[0027] The LSTM time series prediction model makes predictions by preprocessing the collected data, building and training the model, designing an architecture with an input layer, two LSTM hidden layers and an output layer, training with data from the past three months, adjusting parameters, and using a validation set to validate the output results. If the validation result MSE < 5%, it is considered to have met the standard, and the model is used for prediction. During the prediction execution phase, a prediction is triggered once every minute. The preprocessed data is input and the prediction results and confidence scores for future time periods are output. The accuracy of the prediction results is judged based on the confidence scores. In the result correction stage, the results are corrected by taking into account weather and time period characteristics, and the corrected results are output. The input layer receives preprocessed time series data, and the input dimension is consistent with the set time step and the number of features. The number of LSTM units in the first layer of the LSTM hidden layer is set to 64, and the number in the second layer is set to 32, both using the ReLU activation function; The output layer is a fully connected layer, and the output dimension is the number of predicted targets.

[0028] This application triggers a prediction every 1 minute, which matches the characteristics of rapid changes in parking lot traffic flow. It can capture the latest trends in a timely manner and avoid the results becoming outdated due to excessively long prediction intervals. If the traffic flow at a certain entrance or exit suddenly surges within 5 minutes, the 1-minute prediction can quickly respond to this change and update the congestion forecast for the next 3-5 minutes. Providing confidence levels (such as the model's probability of predicting the outcome) along with the prediction results helps the system determine the reliability of the prediction. If the confidence level is low (e.g., <70%), it can be corrected by combining other auxiliary data (e.g., human observation) to avoid blindly relying on the model. If the confidence level is high (e.g., >90%), the prediction results can be used directly to improve decision-making efficiency.

[0029] The confidence level is calculated by taking the standard deviation σ from the model's prediction error distribution on the validation set, and the relationship between the predicted value and the standard deviation is used to define the confidence level. Let the model prediction value be The actual value is The error on the validation set is The mean error is The standard deviation is Predicted values ​​for new samples Its confidence level is defined as: in is a constant, taken as 1.96, where predictions with a confidence level greater than 80% are directly used as the basis for decision-making, while results with a confidence level less than 50% are re-evaluated.

[0030] This application correlates the predicted value of a new sample with the historical error characteristics by using the mean and standard deviation σ of the error on the validation set. This conforms to the common statistical law that the error follows a normal distribution. Especially when the model is sufficiently trained, the constant 1.96 corresponds to the critical value of the 95% confidence interval in the normal distribution. The confidence level is calculated in this way, which transforms the abstract reliability into a quantifiable value and avoids the arbitrariness of subjective judgment. The confidence level directly reflects the probability of the predicted value matching the actual value. When the deviation between the predicted value and the actual value is small, the confidence level calculated by substituting into the formula will be higher, indicating that the prediction result is more reliable. Conversely, if the deviation may be large, the confidence level will be lower, indicating that there is uncertainty in the prediction. This quantification method allows the system to clearly identify the reliability level of the prediction result.

[0031] In step S4, the optimized scheduling process uses the vehicle's in-vehicle infotainment system as the core interaction carrier, and makes decisions through real-time data flow and intelligent algorithms.

[0032] A smart parking process optimization scheduling system, the scheduling system includes: The data acquisition module is configured to acquire vehicle data at the current parking lot entrances and exits, convert the vehicle data into congestion indicators, perform judgment and analysis, and display and dispatch the data in real time based on the vehicle's infotainment system. The recommendation module is configured to make intelligent recommendations for entrances and exits based on dynamic capacity thresholds; The dynamic adjustment module predicts the congestion status of each entrance and exit in the next 3-5 minutes based on the LSTM time series prediction model, formulates dynamic recommendation rules, and makes dynamic adjustments to the recommendations.

[0033] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for optimizing and scheduling intelligent parking processes, characterized in that, The scheduling steps are as follows: S1. When the user selects a target parking lot for navigation, and approaches the target parking lot, the system acquires vehicle data at the current parking lot entrance and exit based on the parking lot entrance and exit sensors, converts the vehicle data into congestion indicators, and performs judgment and analysis. S2. Set a dynamic carrying capacity threshold for each entrance / exit. When the number of recommended vehicles approaches 80% of the carrying capacity threshold, reduce the recommendation priority of that entrance / exit and recommend the next best option. S3. Based on the current vehicle growth rate, predict the congestion status of each entrance and exit in the next 3-5 minutes using the LSTM time series prediction model, formulate dynamic recommendation rules, and make dynamic adjustments to the recommendations. S4. The optimized scheduling process is based on real-time display on the vehicle's in-vehicle system and recommends routes for users to proceed.

2. The intelligent parking process optimization scheduling method according to claim 1, characterized in that: The judgment and analysis steps in step S1 are as follows: Coil loops are embedded in the ground of the entrance and exit lanes to detect vehicles entering and exiting through electromagnetic induction; Cameras are installed above the entrances and exits to identify license plates and vehicle models using computer vision, and to track queue length and traffic status in real time. Record the number of times the barrier gate is opened and closed, and obtain the number of vehicles passing through the entrance and exit within a unit of time; Calculate the current passage efficiency and congestion index of the entrance / exit per unit time; The congestion level is used to determine the current congestion situation at the entrance / exit.

3. The intelligent parking process optimization scheduling method according to claim 2, characterized in that: Traffic efficiency is calculated by dividing the total number of vehicles passing through the entrances and exits per unit time by the unit time. The congestion index is calculated using a two-part weighted average. The first part is calculated using the following formula: Where Q is the current number of vehicles in the queue, N is the number of entrance and exit lanes, K is the maximum reasonable number of queues per lane, W1 is the weight of the impact of queuing vehicles, with a value range of 50-70, and A is the value calculated in the first part. The calculation formula for the second part is: Where B is the value calculated in the second part. For current traffic efficiency, For the sake of historical efficiency, The weight for the impact on traffic efficiency ranges from 30 to 50. Add the values ​​of A and B to get the congestion index. The higher the congestion index, the more vehicles are at the current entrance and exit and the more congested it is. Establish a congestion index range, including smooth flow, congested, and severe congestion. Substitute the obtained congestion index into the congestion index range for classification to obtain the current congestion situation at the entrance / exit. The range of the congestion index was determined through a combination of data modeling and manual calibration.

4. The intelligent parking process optimization scheduling method according to claim 1, characterized in that: The steps for determining the dynamic load threshold in step S2 are as follows: The dynamic carrying capacity threshold is dynamically adjusted based on the hardware conditions and scenarios at the entrances and exits. The basic carrying capacity threshold = number of lanes × maximum traffic volume of a single lane per unit time × α, where α is a buffer coefficient, a safe buffer coefficient calculated based on historical data, with a value range of 1.1 to 1.

3. The carrying capacity threshold is dynamically adjusted in combination with the time of day, weather and scenario factors. If the traffic efficiency during the morning peak decreases by β1 based on historical data analysis, the dynamic carrying capacity threshold is adjusted down by β1 accordingly. If efficiency decreases by β2 during heavy rain, the dynamic threshold is also lowered by β2. If there is a large event around the target location and traffic efficiency decreases by β3, the dynamic threshold is also lowered by β3. Here, β1 is the time period correction coefficient, β2 is the weather correction coefficient, and β3 is the special scenario correction coefficient.

5. The intelligent parking process optimization scheduling method according to claim 1, characterized in that: The recommended second-most-priority selection step in step S2 is: The system monitors recommended vehicles in real time, including vehicles on navigation and arriving within 5 minutes, social vehicles within 500 meters heading towards the entrance / exit, and vehicles that have booked the entrance / exit within the next 5 minutes. The number of vehicles is added up to obtain the current number of recommended vehicles. The system updates the data every 30 seconds to reflect the traffic flow in real time. When the number of recommended vehicles reaches 80% of the dynamic capacity threshold for the entrance / exit on that day, the recommendation weight of that entrance / exit is reduced, and the next best entrance / exit is selected for recommendation.

6. The intelligent parking process optimization scheduling method according to claim 1, characterized in that: The specific steps in step S3 are as follows: Indicators are collected every 10 seconds, including the current number of vehicles queuing at each entrance / exit, the number of vehicles increasing, the passage efficiency of the entrance / exit, the number of vehicles en route to that entrance / exit, and vehicles that can be reached within 3 minutes by tracking through the navigation system. Based on the LSTM time series prediction model, input the vehicle growth rate and traffic efficiency data for the past 5 minutes, and output the predicted number of queuing vehicles and the predicted congestion index for each entrance and exit in the future time. The forecast results are corrected by combining real-time weather and time-of-day characteristics; The predicted congestion index is compared with the dynamic capacity threshold. If it is lower than the dynamic capacity threshold, the recommendation continues; if it is higher than the dynamic capacity threshold, the recommended entrance / exit is changed. Based on the predictions of the predictive model, different recommendation strategies are formulated for each entrance and exit, including: Prioritized recommendation level; expected to be smooth operation in the future; navigation will be recommended first at the current entrance / exit. Recommended access level suspended. Severe congestion is expected in the future, triggering a diversion warning. The current entrance / exit will be changed.

7. The intelligent parking process optimization scheduling method according to claim 6, characterized in that: The LSTM time series prediction model makes predictions by preprocessing the collected data, building and training the model, designing an architecture with an input layer, two LSTM hidden layers and an output layer, training with data from the past three months, adjusting parameters, and using a validation set to validate the output results. If the validation result MSE < 5%, it is considered to have met the standard, and the model is then used for prediction. During the prediction execution phase, a prediction is triggered once every minute. The preprocessed data is input and the prediction results and confidence scores for future time periods are output. The accuracy of the prediction results is judged based on the confidence scores. In the result correction stage, the results are corrected by taking into account weather and time period characteristics, and the corrected results are output. The input layer receives preprocessed time series data, and the input dimension is consistent with the set time step and the number of features. The number of LSTM units in the first layer of the LSTM hidden layer is set to 64, and the number in the second layer is set to 32, both using the ReLU activation function; The output layer is a fully connected layer, and the output dimension is the number of predicted targets.

8. The intelligent parking process optimization scheduling method according to claim 7, characterized in that, The confidence level is calculated by taking the standard deviation σ from the model's prediction error distribution on the validation set, and the relationship between the predicted value and the standard deviation is used to define the confidence level. Let the model prediction value be The actual value is The error on the validation set is The mean error is The standard deviation is Predicted values ​​for new samples Its confidence level is defined as: in is a constant, taken as 1.96, where predictions with a confidence level greater than 80% are directly used as the basis for decision-making, while results with a confidence level less than 50% are re-evaluated.

9. The intelligent parking process optimization scheduling method according to claim 1, characterized in that: In step S4, the optimized scheduling process uses the vehicle's in-vehicle infotainment system as the core interaction carrier, and makes decisions through real-time data flow and intelligent algorithms.

10. A smart parking process optimization scheduling system, characterized in that, The scheduling system includes: The data acquisition module is configured to acquire vehicle data at the current parking lot entrances and exits, convert the vehicle data into congestion indicators, perform judgment and analysis, and display and dispatch the data in real time based on the vehicle's infotainment system. The recommendation module is configured to make intelligent recommendations for entrances and exits based on dynamic capacity thresholds; The dynamic adjustment module uses an LSTM time-series prediction model to predict the congestion status of each entrance and exit in the next 3-5 minutes, formulates dynamic recommendation rules, and makes dynamic adjustments to the recommendations.

Images