Self-service refueling hierarchical guidance and vehicle dispatching system and method based on artificial intelligence

Abstract

The present application relates to the technical field of intelligent scheduling of gas stations, in particular to a self-service refueling hierarchical guidance and vehicle scheduling system and method based on artificial intelligence, a vehicle portrait construction unit acquires identity information, required oil products and vehicle physical characteristics of an incoming vehicle, an occupancy duration prediction unit is built-in with an artificial intelligence model trained by historical refueling data, and predicts the expected occupancy duration according to the vehicle physical characteristics and historical refueling behavior data, a parking space state detection unit detects the occupancy state and oil product type of each refueling position in real time, a driving-off convenience calibration unit stores the driving-off convenience of each refueling position pre-calibrated according to the adjacent position waiting and path conflict degree, a hierarchical guidance unit divides the expected occupancy duration into a low occupancy level and a high occupancy level by comparing with a set threshold, and a scheduling guidance unit allocates the low occupancy level vehicle to the idle refueling position with the lowest driving-off convenience, and allocates the high occupancy level vehicle to the idle refueling position with the highest driving-off convenience and generates a guidance signal.

Description

Technical Field

[0001] This invention relates to the field of intelligent dispatching technology for gas stations, and more specifically, to a self-service refueling tiered guidance and vehicle dispatching system and method based on artificial intelligence. Background Technology

[0002] In the daily operation of gas stations, guiding vehicles to quickly park at suitable refueling positions and improving overall traffic efficiency is a crucial management challenge in self-service refueling scenarios. Existing technologies include license plate recognition and parking space detection systems in some gas stations, capable of identifying vehicles and monitoring the occupancy status of each refueling position in real time. Some management systems can also guide vehicles to corresponding areas based on fuel type. However, existing guidance strategies typically employ a first-come, first-served or nearest-neighbor allocation approach, failing to differentiate between the different parking times of various vehicles or fully considering the inherent differences in ease of departure between different refueling positions. Positions near the exit or on the main traffic lane offer minimal obstacle removal, while positions deep within the aisle or adjacent to other positions require waiting. Waiting for neighboring vehicles to yield, navigating more intersections, and taking longer to leave can all lead to a mismatch between parking resources and vehicle demand. When vehicles with longer parking times occupy spaces with low ease of departure, or vehicles with shorter parking times occupy spaces with high ease of departure, it impacts the overall traffic flow efficiency of the gas station. Therefore, accurately predicting the occupancy time of vehicles entering the station and using ease of departure as the core consideration for parking space allocation, achieving differentiated matching between long-term vehicles and high-ease-of-delivery spaces, and short-term vehicles and low-ease-of-delivery spaces, is a key technical challenge for improving the traffic efficiency and parking space utilization of self-service gas stations. To address this issue, we provide an AI-based self-service refueling tiered guidance and vehicle dispatching system and method. Summary of the Invention

[0003] The purpose of this invention is to provide an artificial intelligence-based self-service refueling tiered guidance and vehicle dispatching system and method to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, an AI-based self-service refueling tiered guidance and vehicle dispatching system is provided, including: The vehicle profiling unit is used to obtain the identity information of the vehicles entering the station, the required fuel type, and the vehicle's physical characteristics, including the location of the fuel tank opening and the vehicle's dimensions. The occupancy time prediction unit has a built-in artificial intelligence model trained on historical refueling data, which is used to predict the estimated occupancy time of the current vehicle based on the vehicle's physical characteristics and historical refueling behavior data associated with identity information. The parking space status detection unit is used to detect the occupancy status of each refueling station and the type of fuel provided in real time; The departure convenience calibration unit is used to store the departure convenience of each pre-calibrated refueling station. The departure convenience is determined based on whether the vehicle needs to wait for the adjacent vehicle to leave when driving from the refueling station to the exit and the degree of path conflict. The lowest departure convenience indicates the greatest obstacle to departure. The graded guidance unit is used to compare the estimated occupancy time with a set time threshold and classify vehicles into low occupancy levels and high occupancy levels. The dispatching and guidance unit is used to assign vehicles with low occupancy levels to the vacant refueling position with the lowest ease of departure when there are multiple vacant refueling positions on the same side as the required fuel, and to assign vehicles with high occupancy levels to the vacant refueling position with the highest ease of departure, and to generate corresponding guidance signals.

[0005] The second objective of this invention is to provide a method for implementing an AI-based self-service refueling tiered guidance and vehicle dispatching system, comprising the following steps: Obtain the vehicle's identity information, required fuel type, and physical characteristics upon arrival at the station; The system inputs identity information, vehicle physical characteristics, and associated historical refueling behavior data into an artificial intelligence model trained on historical refueling data, and outputs the estimated time taken. Real-time monitoring of the occupancy status of each refueling station and the type of fuel provided; For multiple available refueling stations on the same side as the required fuel, the pre-marked ease of departure for each station is read. The ease of departure is determined based on whether the vehicle needs to wait for the adjacent vehicle to leave when driving from the refueling station to the exit and the degree of path conflict. Compare the estimated usage time with the set duration threshold to classify the system into low and high usage levels. Vehicles with low occupancy levels are assigned to the vacant refueling stations with the lowest ease of departure, while vehicles with high occupancy levels are assigned to the vacant refueling stations with the highest ease of departure. A guidance signal containing the target refueling station number and the guidance route is generated and sent.

[0006] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs vehicle profiles to obtain the physical characteristics and identity information of vehicles entering the station. It uses a regression neural network trained on historical refueling data to accurately predict the expected occupancy time of vehicles. It pre-determines the ease of leaving each refueling position based on the probability of waiting for adjacent positions and the degree of path conflict. Vehicles are divided into two levels according to their occupancy time: high and low. Among the available parking spaces on the same side of the same type of fuel, short-term refueling vehicles with low occupancy are assigned to low-convenience parking spaces with greater difficulty in leaving, while long-term refueling vehicles with high occupancy are assigned to high-convenience parking spaces with smooth departure and guidance signals are generated. The prediction model improves the prediction accuracy of long-term vehicles through weighted loss training, reduces the phenomenon of long-term vehicles blocking internal passages, and improves the overall parking space turnover rate and vehicle passage efficiency of the gas station. Attached Figure Description

[0007] Figure 1 This is an overall block diagram of the present invention; Figure 2 This is the overall flowchart of the present invention. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] This invention provides an AI-based self-service refueling tiered guidance and vehicle dispatching system. Please refer to [link / reference]. Figure 1 As shown, it includes: The vehicle profiling unit is used to obtain the identity information of the vehicles entering the station, the required fuel type, and the vehicle's physical characteristics, including the location of the fuel tank opening and the vehicle's dimensions. The occupancy time prediction unit has a built-in artificial intelligence model trained on historical refueling data, which is used to predict the estimated occupancy time of the current vehicle based on the vehicle's physical characteristics and historical refueling behavior data associated with identity information. The parking space status detection unit is used to detect the occupancy status of each refueling station and the type of fuel provided in real time; The departure convenience calibration unit is used to store the departure convenience of each pre-calibrated refueling station. The departure convenience is determined based on whether the vehicle needs to wait for the adjacent vehicle to leave when driving from the refueling station to the exit and the degree of path conflict. The lowest departure convenience indicates the greatest obstacle to departure. The graded guidance unit is used to compare the estimated occupancy time with a set time threshold and classify vehicles into low occupancy levels and high occupancy levels. The dispatching and guidance unit is used to assign vehicles with low occupancy levels to the vacant refueling position with the lowest ease of departure when there are multiple vacant refueling positions on the same side as the required fuel, and to assign vehicles with high occupancy levels to the vacant refueling position with the highest ease of departure, and to generate corresponding guidance signals.

[0010] Before training the AI ​​model used in the occupancy time prediction unit, historical refueling data is first collected. The historical refueling data comes from the license plate recognition equipment, fuel dispenser metering system, parking space detection device and payment settlement system linkage records of the gas station. The data coverage period is no less than twelve months. The historical refueling data includes the identity information, physical characteristics and actual occupancy time of each historical vehicle entering the station, as well as historical refueling behavior data.

[0011] The identity information is the unique identifier for the corresponding vehicle, specifically the license plate number, which is used to associate multiple historical refueling records of the same vehicle to achieve aggregation of behavioral data at the vehicle level.

[0012] Vehicle physical characteristics include two types of information: fuel tank opening location and vehicle dimensions. The fuel tank opening location records the vehicle's orientation, which is divided into four types: left and right, front and rear. Vehicle dimensions record the vehicle's length, width, and wheelbase, all in meters. The fuel tank opening location affects the difficulty and time required for parking and positioning the vehicle, while vehicle dimensions affect the space required for parking and the time required for driving away.

[0013] The actual occupancy time is the total time from when the corresponding vehicle stops at the refueling position to when it finishes refueling and leaves the refueling position, in seconds. The time covers all stages of the process, including vehicle parking and positioning, opening the fuel tank cap and refueling nozzle, refueling, settlement and payment, closing the fuel tank cap and preparing to leave. It is the core indicator for measuring the degree of vehicle occupancy of refueling space resources.

[0014] Historical refueling behavior data represents the statistical characteristics of all refueling records of vehicles with the same identity within the statistical period. Specifically, it includes the mean and standard deviation of the average number of refuelings per month, the average amount of refueling per refueling, the historical dwell time, and the commonly used payment methods, reflecting the vehicle's fixed refueling habits and operational time patterns.

[0015] After completing the raw data collection, data cleaning was performed to remove records with missing information fields, abnormal records with actual occupancy time exceeding three times the standard deviation of the overall distribution, and duplicate vehicle entry records, ensuring the validity and authenticity of all retained data. After data cleaning, a training dataset was constructed. The identity information, vehicle physical characteristics, and historical refueling behavior data from the historical refueling data were used as input features for the training samples, and the actual occupancy time was used as the label for the training samples. When constructing input features, different types of features were first converted into numerical values. Identity information was converted into a fixed-dimensional numerical vector through embedding encoding to realize the numerical expression of vehicle identity. The fuel tank opening position in the vehicle physical characteristics was converted into discrete values ​​using classification encoding, with different integer codes corresponding to different orientations. The length, width, and wheelbase of the vehicle dimensions were directly converted into continuous feature inputs in meters. The average monthly refueling frequency, average single refueling amount, and the mean and standard deviation of historical dwell time in the historical refueling behavior data were directly used as continuous numerical features. Common payment methods were converted into discrete values ​​using classification encoding.

[0016] All numerical features are concatenated in a fixed order to form a one-dimensional input feature vector. Each historical refueling record corresponds to a unique input feature vector, and the corresponding actual duration is used as the sample label. The label is a continuous numerical value in seconds.

[0017] All samples were randomly divided into training, validation, and test sets in a ratio of 7:1.5:1.5. The training set was used for iterative updates of model parameters, the validation set was used for monitoring generalization ability and preventing overfitting during training, and the test set was used for final accuracy evaluation after training was completed.

[0018] After completing the dataset construction, a regression neural network model was used as the initial model. The regression neural network is a neural network structure that takes continuous numerical values ​​as the output target. Unlike the classification neural network that outputs class probabilities, the regression neural network can output continuous duration prediction results, which is suitable for the regression task requirements of duration prediction.

[0019] In this invention, the regressive neural network learns the nonlinear mapping relationship between multi-dimensional input features and actual occupancy time, explores the influence of vehicle physical attributes and historical behavior on total occupancy time, and ultimately achieves accurate prediction of unknown vehicle occupancy time. This provides a reliable basis for subsequent tiered guidance and parking space scheduling. The initial regressive neural network has three hidden layers, each containing a fixed number of neurons. A linear rectified activation function is used between adjacent hidden layers for nonlinear transformation to enhance the model's feature representation ability. The output layer has one neuron and uses a linear activation function to output the final predicted occupancy time. The regressive neural network model is trained using a training dataset under supervised learning conditions. Supervised learning refers to a training method where the true labels of the training samples are known, and the model continuously adjusts its parameters by comparing the error between the predicted results and the true labels.

[0020] The training process employs a mini-batch gradient descent strategy. In each training round, the training set samples are divided into multiple fixed-size batches, which are then input into the model batch by batch for computation. For each input batch, the model calculates the feature transformation results layer by layer through a forward propagation process, ultimately outputting the predicted occupancy time for all samples within the batch. After forward propagation, the loss function value for the current batch is calculated. The loss function is a combination of a mean squared error (MSE) basic term and a long-term weighted penalty term. The MSE basic term is the average of the squares of the differences between the predicted and actual occupancy times for all samples within the batch, measuring the overall prediction error level of the model. The long-term weighted penalty term is an innovative optimization term designed to improve the prediction accuracy for high-occupancy vehicles. The calculation logic involves iterating through each sample within the batch and determining whether the actual occupancy time corresponding to the sample is greater than the predicted occupancy time. A set time threshold is set. If the time threshold is exceeded, the squared error value of the corresponding sample is multiplied by a preset penalty coefficient, which is greater than one. If the time threshold is not exceeded, the original squared error value remains unchanged. Finally, the average of all processed squared error values ​​is taken to obtain the total loss value. This weighted loss structure allows the model to pay more attention to the prediction error of long-duration vehicles during training, reducing the impact of prediction deviation of long-duration vehicles on the scheduling effect, adapting to the business requirements of hierarchical scheduling. Unlike the traditional training method of single mean square error loss, after obtaining the loss value, the weight parameters are adjusted through the backpropagation algorithm. The backpropagation algorithm is based on the chain rule of differentiation, and calculates the gradient of each layer's weight parameters and bias parameters with respect to the total loss value from the output layer forward. The gradient represents the direction and magnitude of the influence of parameter changes on the loss value.

[0021] After calculating the gradients of all parameters, the value of each parameter is updated according to the gradient descent rule. The parameter update direction is the opposite direction of the gradient, and the update step size is controlled by the learning rate. An adaptive learning rate optimizer is used during training. The learning rate is automatically adjusted according to the rate of loss descent. A larger learning rate is maintained when the loss decreases rapidly to accelerate convergence, and the learning rate is automatically reduced when the loss tends to stabilize to avoid training oscillations. Each batch of forward calculation and backward update completes one parameter iteration. Each complete traversal of the training set is called a training round. After each training round, the validation loss value is calculated using the validation set data. The validation loss value reflects the model's prediction performance on data not used in training and is used to judge the model's generalization ability. Training continues until the loss function converges. The convergence criterion includes two items. The conditions are as follows: First, the relative decrease in validation loss is less than one-thousandth for ten consecutive rounds; second, the number of training rounds reaches the preset maximum number of training rounds. Training can be stopped once either condition is met. An early stop mechanism is set during training: if the validation loss increases for three consecutive rounds, training is terminated early. The model parameters corresponding to the round with the lowest validation loss are saved to prevent overfitting. After training stops, the accuracy of the final model is tested using test set data. The mean absolute error and root mean square error are calculated as evaluation indicators. Once the indicators meet the preset accuracy requirements, the model is deployed to the occupancy time prediction unit to predict the expected occupancy time of vehicles entering the station in real time. This provides stable and reliable time prediction support for the implementation of subsequent tiered guidance strategies, ensuring the rationality of parking space scheduling and the traffic efficiency of gas stations.

[0022] After a vehicle enters the gas station entrance area, the vehicle profiling unit collects the vehicle's identity information and physical characteristics. The identity information specifically includes the license plate number, which is captured and identified by the license plate recognition equipment deployed at the entrance. This number serves as the unique identifier for the vehicle and is used to match the corresponding refueling record in the historical database.

[0023] Vehicle physical characteristics include two types of information: fuel tank opening location and vehicle dimensions. The fuel tank opening location is obtained by detecting the vehicle body through visual recognition equipment at the entrance, and is divided into four types: left and right, front and rear. The difference in the fuel tank opening location will affect the complexity of the alignment operation when the vehicle is parked at the refueling station, and thus affect the parking time. The vehicle dimensions are obtained by visual ranging technology combined with vehicle database matching, including three values: vehicle length, width, and wheelbase, all in meters. The vehicle dimensions directly determine the space required for parking and the difficulty of the departure operation, and are an important physical factor affecting the total occupancy time. This process completes the identification information and physical characteristics of the current vehicle. After data collection, all historical refueling records matching the corresponding vehicle are retrieved from the historical refueling behavior database based on the identity information. The refueling frequency, average refueling amount, and historical dwell time distribution characteristics corresponding to the identity information are extracted from the historical refueling behavior data. The refueling frequency is calculated by counting the total number of refuelings for the corresponding vehicle over the past twelve months, dividing the total number of refuelings by the total number of months in the statistical period to obtain the average number of refuelings per month. The refueling frequency reflects the stability of the vehicle's refueling habits; vehicles with regular refueling frequencies generally have more predictable operation times. The average refueling amount is calculated by counting the refueling amounts in all historical refueling records for the corresponding vehicle. The average refueling volume for a single record is calculated by taking the arithmetic mean of all single refueling volumes, in liters. This average volume directly determines the refueling time; a larger volume results in a longer refueling time and a longer total dwell time. The historical dwell time distribution characteristics include the mean and standard deviation of historical dwell time. The mean is calculated by taking the arithmetic mean of the actual dwell time across all historical records for the corresponding vehicle, in seconds, reflecting the vehicle's typical dwell time level. The standard deviation is calculated by first calculating the actual dwell time for each historical record and then... The difference between the means is used to calculate the arithmetic mean of the squared values ​​of all differences, and then the square root of the arithmetic mean is taken to obtain the standard deviation of the historical dwell time in seconds. The standard deviation of the historical dwell time reflects the degree of fluctuation in the dwell time of the corresponding vehicle. The larger the standard deviation value, the higher the uncertainty of the dwell time of the corresponding vehicle. If the current vehicle is an unfamiliar vehicle entering the station for the first time, and there is no record of corresponding identity information in the historical refueling behavior database, the statistical mean of the group of vehicles of the same model and fuel type is used as the substitute value of the three features to ensure that a complete feature set can be generated for all vehicles entering the station, and to avoid the prediction process from failing to execute normally due to missing features.

[0024] After extracting and calculating all features, the current vehicle's identity information, physical characteristics, and the extracted refueling frequency, average refueling amount, and historical dwell time distribution features are concatenated to form an input feature vector. Before concatenation, all features undergo numerical transformation and normalization processes identical to those used in the model training phase to ensure the input data distribution remains consistent with the training data distribution, thus improving the accuracy of prediction results. Identity information is converted into a fixed-dimensional numerical vector using the same embedding encoding method as in the training phase, achieving a numerical representation of identity. The fuel tank opening location is converted into a corresponding integer value using a classification encoding method, with different orientation types corresponding to different integer codes. The three values ​​for vehicle size, refueling frequency, average refueling amount, and two values ​​for historical dwell time distribution features are all continuous values, and the original calculation results are directly retained.

[0025] The normalization process uses the min-max normalization method. For each feature dimension, the minimum and maximum values ​​of the corresponding feature dimension in the training dataset are first obtained. The minimum value of the feature dimension is subtracted from the corresponding feature value of the current vehicle, and then divided by the difference between the maximum and minimum values ​​of the feature dimension. All feature values ​​are uniformly mapped to the numerical range of zero to one, eliminating the difference in dimensionality between different features and avoiding the dominance of features with large numerical ranges in the model's calculation process.

[0026] All normalized features are arranged in a fixed order determined during the training phase and concatenated to form a one-dimensional input feature vector. The dimension of the input feature vector is exactly the same as the dimension of the input features in the training samples, ensuring that the artificial intelligence model can receive and process the input data normally. The input feature vector is then input into the artificial intelligence model, which is a regression neural network trained on historical refueling data. The model has three hidden layers and one output layer. The input feature vector first enters the first hidden layer, which contains a predetermined number of neurons. Each neuron performs a dot product operation between the input feature vector and the corresponding weight vector, and adds the bias value of the corresponding neuron to obtain a linear transformation result. The linear transformation result is then input into a linear rectified activation function for nonlinear transformation. The calculation logic of the linear rectified activation function is to directly output the input value when the input value is greater than zero, and output the value zero when the input value is less than or equal to zero. Through nonlinear transformation, the model has the ability to learn complex nonlinear mapping relationships.

[0027] The output of the first hidden layer serves as the input data for the second hidden layer. The second hidden layer repeats the calculation process of weight vector dot product, bias numerical superposition, and activation function transformation to complete the feature extraction and abstraction at the second level. The output of the second hidden layer serves as the input data for the third hidden layer, and the same calculation process is performed. After the three layers of nonlinear transformation processing, the original multi-dimensional input features are transformed into higher-dimensional abstract feature representations, fully integrating multi-dimensional information such as vehicle identity attributes, physical attributes, and historical behavior attributes, and uncovering the deep correlation between various features and occupancy duration.

[0028] After processing by all hidden layers, the final abstract feature data is input to the output layer. The output layer has a single neuron node and uses a linear activation function for calculation. The calculation process involves performing a dot product operation between the abstract features output by the third hidden layer and the weight vector of the output layer, adding the bias value of the output layer, and obtaining the normalized prediction result. Then, an inverse normalization operation is performed, multiplying the normalized prediction result by the difference between the maximum and minimum actual occupancy time in the training dataset, and adding the minimum actual occupancy time in the training dataset, mapping the prediction result back to the actual time dimension in seconds. Finally, the output layer outputs the estimated occupancy time, which is directly transmitted to the hierarchical guidance unit as the core judgment basis for classifying vehicle occupancy levels, supporting subsequent parking space hierarchical scheduling decisions, and providing accurate duration data support for subsequent parking space hierarchical scheduling, thereby improving the overall traffic efficiency and parking space resource utilization rate of the gas station.

[0029] After completing the road network mapping and refueling point location marking of the gas station site, the ease of departure is calibrated and calculated for each refueling point within the gas station. Ease of departure is used to quantify the smoothness of a vehicle's departure from the corresponding refueling point to the gas station exit. A higher value indicates less obstacle to departure and a shorter overall departure time, while a lower value indicates greater obstacle to departure and a longer overall departure time. The calibration results of all refueling points are pre-stored in the ease of departure calibration unit, providing a fixed parking space attribute basis for subsequent parking space scheduling, avoiding repeated calculations during scheduling, and improving scheduling response speed.

[0030] The calibration process first collects the shortest driving path from the center point of the current refueling station to the exit. The center point of the refueling station refers to the geometric center of the refueling station parking area, and its corresponding planar coordinates are obtained through on-site surveying, in meters. The exit refers to the gate of the gas station used for vehicle exit, which also has fixed planar coordinates. The shortest driving path refers to the route with the shortest total driving distance from the center point of the refueling station to the exit within the designated drivable area road network of the gas station. When calculating the shortest driving path, the road network topology of the gas station is first constructed, and drivable road segments are abstracted as weighted edges, with the edge weight corresponding to the actual length of the road segment. The endpoints of the road segments and intersections are abstracted as topological nodes. The Dijkstra shortest path algorithm is used to calculate the shortest path to the exit node with the center point of the refueling station as the starting node, and the corresponding total length is obtained. The shortest driving path serves as the default driving route for vehicles to leave and is the basis for subsequent path conflict analysis.

[0031] After obtaining the shortest driving path, all nodes where vehicles intersect with other vehicles are marked on this path. Intersection nodes are locations where the shortest driving path of the current refueling station intersects with the shortest exit paths of other refueling stations. This also includes intersections and turning points within the gas station. When a vehicle reaches an intersection node, if other vehicles are exiting at the same time, a meeting and avoidance situation is likely, resulting in additional waiting time. To mark intersection nodes, the shortest exit path data for all other refueling stations within the gas station is first obtained. The shortest driving path of the current refueling station is then spatially compared with other paths to identify the locations of all path intersections. Each identified intersection point is marked as an intersection node. Simultaneously, the associated refueling station number corresponding to each intersection node is recorded to determine which vehicles from adjacent refueling stations will intersect with the current vehicle at the corresponding node. After marking the intersection nodes, the system calculates the number of nodes and the probability of waiting for adjacent vehicles to exit at each node. The path conflict level at the current refueling station is calculated. Each intersection node corresponds to a waiting probability. The waiting probability refers to the probability that at the moment a vehicle leaves, a vehicle at the corresponding associated refueling station also leaves at the same time, and the two vehicles meet at the intersection node and need to give way. The waiting probability is calculated based on the gas station's historical operating data, statistically analyzing the average occupancy rate of the corresponding associated refueling stations during the time period, and then combining this with the time difference between the two vehicles traveling from their respective parking spaces to the intersection node to calculate the time overlap ratio of the two vehicles arriving at the intersection node. The occupancy rate is multiplied by the time overlap ratio to obtain the waiting probability of the corresponding node. The path conflict level is calculated by assigning a corresponding node weight to each intersection node. The weight of a fully intersecting crossroads is higher than that of a T-shaped intersection with one-sided merging. The waiting probability of each intersection node is multiplied by the corresponding node weight, and then the weighted results of all intersection nodes are summed to obtain the original value of the path conflict level. The unit of the original value is related to the weight setting; the higher the value, the higher the risk of waiting at the intersection on the driving path.

[0032] After calculating the degree of path conflict, the distance between adjacent parking spaces on both sides of the current refueling position and the obstruction angle caused by the parking space when the adjacent parking space is parked on the vehicle leaving the current parking space are measured simultaneously. Adjacent parking spaces on both sides refer to refueling positions that are in the same row as the current refueling position and are adjacent to each other on the left and right. Refueling positions at both ends of the row have only one adjacent parking space on one side and no adjacent parking space on the other side. The distance between adjacent parking spaces refers to the horizontal distance between the center points of two adjacent refueling stations, measured in meters. The smaller the distance, the smaller the gap between the two vehicles after they are parked, and the more limited the lateral maneuvering space when the vehicle in this parking space leaves. The obstruction angle refers to the angle of obstruction formed by the body of the adjacent vehicle on the departure trajectory of the vehicle in this parking space after a standard-sized vehicle is parked in the adjacent parking space, measured in degrees. When measuring the obstruction angle, take the observation point of the driver's seat after the vehicle in this parking space is parked as the vertex, and draw rays to the outermost front and rear ends of the adjacent vehicles respectively. The angle formed between the two rays is the obstruction angle on the corresponding side. The larger the obstruction angle, the more serious the obstruction of the driving line and departure trajectory of the vehicle in this parking space by the adjacent vehicle, the more avoidance and adjustment operations are required when the vehicle leaves, and the longer the departure time.

[0033] For situations where there are no adjacent parking spaces on one side, the obstruction angle on that side is recorded as zero. After obtaining the obstruction angles on both sides, the larger of the two obstruction angles is taken as the comprehensive obstruction angle. This is because if there is severe obstruction on one side, it will become the main factor limiting the vehicle's departure speed. After obtaining the two indicators, path conflict degree and comprehensive obstruction angle, the two indicators are combined by weighted summation to obtain a comprehensive score. Before calculating the comprehensive score, the two indicators are normalized to eliminate dimensional differences and map the two values ​​to a uniform range of zero to one hundred. The normalization method for path conflict degree is as follows: the original values ​​of path conflict degree for all refueling positions at the gas station are counted, and the maximum value is taken as the normalization denominator. The original value of path conflict degree for the current refueling position is divided by the corresponding maximum value, and then multiplied by one hundred to obtain the normalized path conflict degree. The path conflict score is calculated, with a higher score indicating more severe obstruction at path intersections. The normalization method for the overall obstruction angle is as follows: using the maximum obstruction angle of 90 degrees in a standard parking space scenario as the normalization denominator, the overall obstruction angle of the current refueling station is divided by 90 degrees and then multiplied by 100 to obtain the normalized obstruction angle score. A higher obstruction angle score indicates more severe occlusion from adjacent vehicles. Subsequently, path conflict weight coefficients and obstruction angle weight coefficients are pre-set. Both weight coefficients are dimensionless constants, and their sum equals one. The values ​​of the weight coefficients can be adjusted according to the actual site layout of the gas station. If there are many road intersections and frequent intersections within the gas station, the value of the path conflict weight coefficient can be appropriately increased. If the parking spaces in the gas station are densely arranged and the spacing is small, the value of the obstruction angle weight coefficient can be appropriately increased. The overall score is calculated by multiplying the normalized path conflict score by the path conflict weighting coefficient, and adding the normalized obstruction angle score multiplied by the obstruction angle weighting coefficient to obtain the initial overall score for the current refueling location. The initial overall score ranges from zero to one hundred; a higher value indicates a greater total obstruction for the vehicle leaving the corresponding parking space. After obtaining the initial overall score, it is mapped to a departure convenience value. The mapping uses an inverse linear mapping method because a higher overall score indicates greater obstruction, and the corresponding departure convenience should be lower. The mapping calculation method is as follows: The ease of departure is equal to 100 minus the initial comprehensive score. After mapping, an initial comprehensive score of 0 means that the vehicle can leave without any obstacles, corresponding to an ease of departure value of 100, which is the highest level. An initial comprehensive score of 100 means that the vehicle is subject to the greatest obstacles, corresponding to an ease of departure value of 0, which is the lowest level. The ease of departure values ​​of all refueling positions are uniformly stored after calibration. When the scheduling and guidance unit performs parking space allocation, it can directly read the value of the corresponding refueling position without real-time calculation. This ease of departure calibration method, which integrates the risks of path intersections and the impact of adjacent occupancy, is more in line with the actual departure scenario of gas stations than the traditional method of classifying parking space levels solely based on distance from the exit. It can more accurately reflect the actual departure efficiency differences of different parking spaces, providing reliable parking space attribute support for subsequent hierarchical scheduling strategies. This ensures that long-occupied vehicles can leave quickly after being allocated to high-ease-of-delivery parking spaces, and that short-occupied vehicles will not have a significant impact on the overall traffic efficiency after being allocated to low-ease-of-delivery parking spaces, ultimately improving the parking space turnover efficiency and overall traffic capacity of gas stations.

[0034] When calibrating the path conflict level of refueling stations, the typical driving trajectory of the corresponding vehicle when leaving the exit is simulated for each refueling station. The typical driving trajectory represents the driving route of the vehicle from the corresponding refueling station to the gas station exit under driving habits. Compared with the shortest path, it is closer to actual driving operation and can reflect the path intersection risk in the actual departure process.

[0035] Before simulating typical driving trajectories, a road network survey of the gas station site was completed, marking the boundary positions of all drivable lanes, the radius of curvature of turning areas, and the passage range of the exit gate. Simultaneously, the most common type of passenger car in the gas station was selected as the reference model, and its minimum turning radius was used as a constraint for trajectory planning. The simulation process started from the center point of the refueling station, following the operational logic of a vehicle driving forward. First, the turning trajectory out of the parking space was planned, ensuring that the turning process did not exceed the lane boundaries and met the vehicle's turning radius requirements. Then, the straight section after merging into the main road was planned, and finally, the entry trajectory to the exit was planned. The process collects the planar coordinates of trajectory points at set intervals to form a typical driving trajectory sequence composed of continuous coordinate points. The interval between trajectory points is set to 0.5 meters to ensure that the spatial accuracy of the trajectory meets the needs of subsequent conflict identification. After the typical driving trajectory is generated, all conflict areas where the path intersects with that of vehicles in adjacent parking spaces are identified on the typical driving trajectory. Here, adjacent parking spaces refer to all refueling spaces where the driving path intersects with the current refueling station driving path. This includes adjacent parking spaces on the left and right sides of the same row, as well as parking spaces in the opposite lane and parking spaces around intersections. A corresponding typical driving trajectory is pre-generated for each adjacent parking space.

[0036] When identifying conflict areas, the typical driving trajectory of the current refueling station is spatially superimposed and compared with the typical driving trajectories of all adjacent parking spaces. The spatial positions of the two trajectories are compared segment by segment. When the lateral distance between the two trajectories is less than 1.2 times the width of the reference vehicle, it is determined that there is a driving conflict in the corresponding road segment. Continuous conflict road segments are marked as a conflict area. After identification, each conflict area is classified into three types: the first type is intersecting, the second type is merging, and the third type is parallel. Intersecting refers to a conflict area where the two trajectories intersect perpendicularly or at a large angle, corresponding to the scenario of two vehicles meeting at a cross. Merging refers to a conflict area where the two trajectories merge into the same lane from different directions, corresponding to the scenario of two vehicles merging. Parallel refers to a conflict area where the two trajectories extend parallel in the same direction, corresponding to the scenario of two vehicles driving side by side. Different types of conflicts correspond to different probabilities of waiting for oncoming traffic and different operational difficulties.

[0037] After identifying and classifying all conflict zones, the conflict weight for each conflict zone is calculated based on the occupancy probability of adjacent parking spaces and the spatiotemporal overlap of two vehicles simultaneously appearing in the corresponding conflict zone. The occupancy probability of adjacent parking spaces is derived from historical operational statistics of the gas station, calculated separately for peak and off-peak periods. Peak periods cover the morning and evening hours with high traffic volume, while off-peak periods cover other periods with lower traffic volume. For each period, the cumulative occupancy time of adjacent parking spaces is calculated, and the cumulative occupancy time is divided by the total duration of the corresponding period to obtain the occupancy probability for that period. The occupancy probability ranges from zero to one; a higher value indicates a greater likelihood that the adjacent parking space is occupied. The spatiotemporal overlap is used to quantify the degree of temporal overlap between two vehicles arriving at and passing through the conflict zone simultaneously. The calculation first calculates the time it takes for each vehicle to travel from its own parking space to the starting point of the conflict zone. The driving time is calculated by dividing the total length of the trajectory from the corresponding parking space to the starting point of the conflict area by the average driving speed of vehicles within the gas station. The average driving speed is taken as the value of meters per second corresponding to the normal driving speed of five kilometers per hour within the gas station. Then, the driving time for each vehicle to pass through the entire conflict area is calculated separately. The driving time is equal to the trajectory length of the conflict area divided by the average driving speed. Based on the driving time and passage time of the two vehicles, the time intervals for each vehicle to pass through the conflict area are determined. The starting point of the time interval is the time when the vehicle departs from the parking space and arrives at the starting point of the conflict area, and the ending point of the time interval is the time when the vehicle completely leaves the conflict area. The intersection time and union time of the two time intervals are calculated. The intersection time is divided by the union time to obtain the spatiotemporal overlap of the two vehicles in the corresponding conflict area. The spatiotemporal overlap ranges from zero to one. The higher the value, the higher the probability that the two vehicles are simultaneously in the conflict area, and the greater the possibility of oncoming traffic waiting.

[0038] To differentiate the impact of different conflict types, three type coefficients are pre-set for each type: 1.0 for cross-type conflicts, 0.7 for confluence-type conflicts, and 0.4 for parallel-type conflicts. These type coefficients are dimensionless constants used to reflect the varying severity of different conflict scenarios. The conflict weight of a single conflict region is calculated as follows: The conflict weight value for the corresponding conflict area is obtained by multiplying the time-period occupancy probability of adjacent parking spaces by the spatiotemporal overlap of the two vehicles in the corresponding conflict area, and then by the type coefficient of the corresponding conflict type. The conflict weights of all conflict areas corresponding to the current refueling station are accumulated to obtain the path conflict degree value of the corresponding refueling station. The path conflict degree value is a dimensionless comprehensive value. The higher the value, the higher the comprehensive risk of encountering other vehicles and waiting when the vehicle leaves the corresponding refueling station, and the greater the obstacles encountered in the departure process. This calculation method, which combines trajectory simulation, conflict type classification, and spatiotemporal overlap analysis, can more accurately quantify the departure conflict risk of different parking spaces compared with the traditional calculation method that only counts the number of intersection nodes. It fully considers the encounter probability in the time dimension and the type difference of conflict scenarios, improves the accuracy of path conflict degree calculation, provides a more reliable quantitative basis for the comprehensive calibration of departure convenience, and ultimately ensures the rationality and effectiveness of the parking space hierarchical scheduling strategy.

[0039] The system determines whether a vehicle needs to wait for an adjacent vehicle to leave a target refueling station. An adjacent station refers to a refueling station located on the same row of refueling islands and directly adjacent to the target station on both sides. Refueling stations at the ends of the refueling islands only have an adjacent station on one side, with no adjacent station on the other side. The status data required for this determination comes from the real-time detection results of the parking space status detection unit. This unit, through a visual recognition device deployed above the refueling station and linked to the real-time metering data of the refueling machine, can obtain the occupancy status of each refueling station and the refueling progress of the corresponding vehicle. The determination operation is performed separately for the left and right sides. For any adjacent station on one side, the occupancy status of the corresponding adjacent station is first detected. If the adjacent station is vacant, it means that there is no parked vehicle on the corresponding side, and it will not obstruct the exit trajectory of the vehicle in this station. Therefore, it is directly determined that the vehicle in this station does not need to wait for the adjacent vehicle on that side to leave.

[0040] If the adjacent parking space is occupied, the refueling progress data of the corresponding adjacent vehicle is retrieved, and the remaining refueling time of the adjacent vehicle is calculated. The remaining refueling time is equal to the remaining amount of fuel to be added to the vehicle divided by the current real-time refueling flow rate of the fuel nozzle, in seconds. At the same time, the remaining time for the vehicle in this parking space to reach a ready-to-leave state is calculated in combination with the estimated occupancy time of the vehicle in this parking space. The two time values ​​are compared. If the remaining refueling time of the adjacent vehicle is less than the remaining time for the vehicle in this parking space to leave, it means that when the vehicle in this parking space is ready to leave, the adjacent vehicle has already completed refueling and is ready to leave, and there will be no continuous obstruction. It is also determined that no waiting is required. If the remaining refueling time of the adjacent vehicle is greater than or equal to the remaining time for the vehicle in this parking space to leave, it means that when the vehicle in this parking space has completed refueling and is ready to leave, the adjacent vehicle is still parked and refueling, and its body will block the exit passage of this parking space. The vehicle cannot drive out directly, and it is determined that it is necessary to wait for the adjacent vehicle on this side to leave.

[0041] After determining the waiting time on both sides, the base score is adjusted. If no waiting is required on either side, it means there is sufficient operating space on both sides when the vehicle leaves, and there is no obstruction from adjacent vehicles. In this case, a base convenience score is awarded, which is preset to 100 points and serves as the baseline convenience score for scenarios without adjacent obstructions. If one side requires waiting while the other does not, the corresponding score is deducted based on the expected waiting time of the side requiring waiting. The expected waiting time refers to the average waiting time required when the vehicle in this parking space is about to leave due to obstruction from adjacent vehicles. In a single-side scenario, the expected waiting time is directly equal to the remaining refueling time of the corresponding adjacent vehicle. If the remaining refueling time cannot be obtained in real time, the historical average refueling time of vehicles of the same model and fuel type is used as a substitute value to ensure that the calculation process is not interrupted due to data loss. If waiting is required on both sides, the larger of the expected waiting times on both sides is taken as the overall expected waiting time. This is because the side with the longer waiting time becomes the core factor restricting the vehicle from leaving, and taking the larger value can more accurately reflect the actual waiting cost. Before the deduction calculation, a deduction value per unit time is preset, with the unit being minutes per second. This represents the convenience deduction amount corresponding to each second of waiting time. The expected waiting time is multiplied by the deduction value per unit time to obtain the total score to be deducted. The corresponding deduction score is subtracted from the basic convenience score to obtain the convenience score adjusted for adjacent influence. If the score after deduction is less than zero, it is uniformly set to zero to avoid the abnormal situation of negative scores.

[0042] After adjusting the score for adjacent location impact, the path conflict severity value is converted into a path conflict deduction value according to a preset ratio. The path conflict severity value is a pre-defined fixed attribute that reflects the comprehensive risk of intersection conflicts on the departure path of the corresponding refueling station. The higher the value, the higher the probability of encountering other vehicles waiting during the journey. The preset ratio is a pre-defined conversion coefficient used to convert the original value of the path conflict severity value into a deduction value that matches the convenience score system. The conversion coefficient is calibrated with reference to the actual impact of path conflict on departure time, so that the conflict deduction and adjacent location waiting deduction have a unified measurement scale. The calculation method of the conversion operation is to multiply the path conflict severity value by the preset conversion coefficient to obtain the corresponding path conflict deduction value.

[0043] Finally, the result of adding or deducting points from the basic convenience score is subtracted from the path conflict deduction to obtain the final departure convenience score. After the calculation, a range constraint is applied to the result, limiting the final value to the range of 0 to 100 points. Values ​​exceeding the boundary are taken as the corresponding boundary values. The final departure convenience score is directly transmitted to the scheduling and guidance unit as the core judgment basis for parking space allocation. The higher the value, the smoother the overall departure of the vehicle from the corresponding parking space, and the less the obstruction from adjacent spaces and path intersections. It is suitable for allocation to high-occupancy vehicles, ensuring that long-staying vehicles can leave quickly after refueling, reducing the obstruction to subsequent vehicles entering the station. The lower the value, the greater the departure obstruction, which is suitable for allocation to low-occupancy vehicles. Short-staying vehicles have short stopping times, and even if the departure speed is slightly slower, it will not have a significant impact on the overall traffic flow turnover efficiency. This improves the accuracy of parking space hierarchical scheduling and further optimizes the overall traffic efficiency and parking space turnover speed of the gas station.

[0044] When calculating the expected waiting time corresponding to the impact of adjacent waiting, the real-time detection data of the adjacent refueling positions on the left and right sides of the target refueling position is first obtained from the parking space status detection unit. The parking space status detection unit is composed of a high-definition visual acquisition device deployed above each refueling position and a real-time metering module of the refueling machine. It can synchronously output the occupancy status of each refueling position and the refueling progress data of the corresponding vehicle. The occupancy status is divided into two categories: idle and occupied. The idle status corresponds to no vehicle parked in the refueling position, and the occupied status corresponds to a vehicle parked in the refueling position. The vehicle refueling progress data can clearly distinguish between two states: the vehicle has not yet completed refueling and the vehicle has completed refueling. When the refueling nozzle is in the suspended state and the settlement process is completed, it is determined that the refueling has been completed. When the refueling nozzle is in the filling state or the settlement is not completed, it is determined that the refueling has not been completed.

[0045] Adjacent refueling positions to the target refueling position refer to those located on the same refueling island and adjacent to each other along the length of the island. Each refueling position can have a maximum of two adjacent positions. Refueling positions at both ends of the refueling island have only one adjacent position on one side, with no adjacent positions on the other side. The side without adjacent positions is automatically determined to be exempt from waiting and is not included in the subsequent waiting time calculation. For the left and right sides with adjacent refueling positions, independent waiting determinations are performed, with completely identical determination logic. For any adjacent refueling position on one side, the occupancy status of the corresponding refueling position output by the parking space status detection unit is read first. If the corresponding adjacent refueling position is in an vacant state, it means that no vehicle is parked on that side and will not obstruct the exit trajectory of the vehicle at the target refueling position. It is directly determined that when the vehicle at the target refueling position leaves, there is no need to wait for the adjacent vehicle on that side to leave, and the waiting time on that side is recorded as zero.

[0046] If the adjacent refueling spot is occupied, the refueling completion status of the vehicle in that spot is further read. If the vehicle has finished refueling, it means it can start and leave at any time and will not occupy the exit lane space for a long time. Therefore, it is determined that there is no need to wait for the adjacent vehicle to leave, and the waiting time on that side is recorded as zero. If the adjacent refueling spot is occupied and the vehicle in the adjacent spot has not yet finished refueling, it means the vehicle will continue to be parked in the adjacent refueling spot. The vehicle body will obstruct the exit view and turning space of the target refueling spot. After the vehicle in the target refueling spot finishes refueling, it cannot drive out directly and must wait for the adjacent vehicle to finish refueling and leave before it can obtain sufficient passage space. In this case, it is determined that it is necessary to wait for the adjacent vehicle to leave, and the waiting time value on the corresponding side is calculated. The calculation method for the waiting time on one side is as follows: First, obtain the remaining refueling time of the adjacent vehicle. The remaining refueling time is equal to the remaining amount of fuel to be added for the adjacent vehicle divided by the real-time refueling flow rate of the current refueling nozzle. The remaining amount of fuel to be added is calculated based on the difference between the historical average refueling amount of the corresponding vehicle and the current amount of fuel already added. If the historical refueling data of the corresponding vehicle cannot be obtained, the average refueling amount of vehicles of the same model and fuel type throughout the station is used, minus the current amount of fuel already added. Next, obtain the remaining operation time of the target refueling station vehicle, which is the time required for the target vehicle to complete refueling and be ready to leave from the current moment. The calculation logic is the same as that for the remaining refueling time of the adjacent vehicle. Subtract the remaining operation time of the target vehicle from the remaining refueling time of the adjacent vehicle to obtain the original calculated value of the waiting time on one side. If the original calculated value is less than or equal to zero, it means that the adjacent vehicle has already completed refueling when the target vehicle is ready to leave, and no waiting time will actually occur. The waiting time on the corresponding side is set to zero. If the original calculated value is greater than zero, the original calculated value is the waiting time value on the corresponding side, in seconds.

[0047] After completing the waiting determination and waiting time calculation for both sides, the final expected waiting time value is obtained by integrating the determination results from both sides. The integration process is divided into three scenarios. The first scenario is that there is no need to wait on either side. In this case, the expected waiting time value is directly set to zero, corresponding to the scenario where there is no obstruction from adjacent vehicles during the entire departure process. The second scenario is that one side requires waiting while the other side does not. In this case, the expected waiting time value is directly taken from the waiting time value of the side that requires waiting, corresponding to the scenario where there is only an adjacent obstruction on one side. The third scenario is that both sides require waiting. In this case, the larger value of the waiting time on both sides is taken as the expected waiting time value. The reason for choosing the larger value is that the normal departure of a vehicle requires sufficient passage space on both sides. The side with a longer waiting time becomes the core constraint restricting the departure of the vehicle. Only after the vehicle on the side with a longer waiting time has completed its departure can the target vehicle obtain a complete departure space. Therefore, taking the larger value can more accurately reflect the actual waiting time required, avoiding the underestimation of the waiting time due to taking the average or smaller value, and ensuring the accuracy of the departure convenience calculation results.

[0048] For refueling stations located at both ends of the refueling island, only one side has adjacent refueling stations. The expected waiting time is directly output based on the result of the judgment on that side. There is no scenario where both sides require waiting. If there is a partial data gap during the judgment process, such as the refueling progress data of an adjacent vehicle on one side not being available, the average remaining refueling time of the same type of vehicle at the same time is used as a substitute value to ensure that the judgment process is not interrupted due to partial data anomalies. The obtained expected waiting time is directly input into the subsequent calculation of the departure convenience score, serving as the core parameter for calculating the convenience score deduction corresponding to adjacent obstruction. The higher the expected value, the longer the duration of the adjacent obstruction, the higher the deducted convenience score, and the lower the final departure convenience score. This waiting judgment method based on real-time parking space status and refueling progress, compared with the static and fixed adjacent influence calibration method, can dynamically match the real-time traffic flow status of the gas station, accurately reflect the actual departure obstruction degree of different parking spaces at different times, and provide a more realistic decision basis for subsequent parking space hierarchical scheduling. This ensures that high-occupancy vehicles can be allocated to the parking space with the least departure obstruction at the current moment, further improving the overall parking space turnover efficiency and traffic capacity of the gas station.

[0049] After obtaining the estimated occupancy time output by the artificial intelligence model, the hierarchical guidance unit first acquires a pre-stored set time threshold. This set time threshold is the dividing line between vehicle occupancy levels, measured in seconds. The set time threshold value is pre-determined based on the median occupancy time of all vehicles in historical statistical data; it is a fixed value that is updated periodically. The median is a statistical indicator used to reflect the central tendency of data, and its calculation method is as follows: The actual occupancy time of all historical vehicles entering the station within the statistical period is arranged in ascending order. If the total number of statistical samples is odd, the value in the middle position after sorting is taken as the median. If the total number of statistical samples is even, the arithmetic mean of the two middle values ​​after sorting is taken as the median. The median is chosen as the tiered threshold because it is not affected by extreme large or small values ​​and reflects the intermediate level of traffic occupancy time better than the arithmetic mean. The number of vehicles corresponding to the two occupancy levels obtained by this method is relatively balanced, avoiding the situation where the proportion of vehicles in one level is too high, which would prevent the tiered scheduling strategy from being ineffective. The calibration period for the occupancy time threshold is set to one calendar month. At the beginning of each month, the actual occupancy time data of all valid vehicles entering the station within the previous full calendar month is extracted. First, a data cleaning operation is performed to remove invalid records with zero occupancy time and abnormal records exceeding three times the standard deviation. Then, the median occupancy time for the corresponding statistical period is obtained according to the median calculation method. As the set duration threshold for the next month, the updated value is stored in the storage module of the hierarchical guidance unit. The threshold update process will not affect the normal operation of the real-time scheduling process. After obtaining the set duration threshold, the estimated occupancy time of the current vehicle is compared with the set duration threshold to perform occupancy level classification. If the estimated occupancy time of the current vehicle is less than or equal to the set duration threshold, the current vehicle is marked as low occupancy level. Vehicles with low occupancy level have a shorter total refueling stop time and a faster parking space turnover rate. Even if they are parked in a refueling space with greater obstacles to leaving, they will not occupy the passageway resources for a long time, and the impact on the overall traffic flow of the gas station is low. If the estimated occupancy time of the current vehicle is greater than the set duration threshold, the current vehicle is marked as high occupancy level. Vehicles with high occupancy level have a longer total refueling stop time and a slower parking space turnover rate. If they are parked in a refueling space with greater obstacles to leaving, they will affect the departure of vehicles in the surrounding parking spaces for a long time, which can easily cause local traffic congestion and have a greater impact on the overall traffic flow of the gas station. After completing the level marking, the occupancy guidance unit transmits the vehicle's occupancy level result along with the vehicle's required fuel information to the dispatch guidance unit. This serves as the core basis for subsequent parking space allocation decisions. The dispatch guidance unit can match refueling spaces with corresponding convenience levels according to different occupancy levels, achieving the scheduling effect of allowing long-occupied vehicles to leave quickly and short-occupied vehicles to make full use of corner parking spaces. This provides a stable and reliable grading basis for subsequent grading dispatch strategies, ultimately improving the overall parking space resource utilization efficiency and traffic turnover speed of the gas station.

[0050] After classifying vehicle occupancy levels, the dispatching and guidance unit performs the allocation of target refueling positions. First, it obtains a list of all available refueling positions on the same side as the fuel type required by the current vehicle from the parking space status detection unit. This unit, through visual acquisition devices deployed at each refueling position and linked to the fuel dispenser's metering module, updates the occupancy status and available fuel types of each position in real time, with a data refresh interval of no more than one second. This ensures that the parking space information obtained during the dispatching process is consistent with the actual situation on-site. "Same side as the fuel type required by the current vehicle" means that the refueling nozzle's orientation matches the vehicle's fuel tank opening position, and the fuel type available at the refueling position is completely consistent with the vehicle's required fuel. Only available refueling positions that simultaneously meet both orientation and fuel type matching requirements are included in the candidate range, preventing issues such as the refueling nozzle not reaching the fuel tank opening or fuel incompatibility after the vehicle stops. After obtaining the list of available refueling positions, the unit first determines the number of refueling positions in the list. If there is only one available refueling spot in the list, the corresponding refueling spot is directly identified as the target refueling spot without the need for subsequent convenience filtering. If there are multiple available refueling spots in the list, the departure convenience value of each available refueling spot is read. The departure convenience value is a dynamic value calculated by combining the real-time status of adjacent spots and path conflict attributes. The higher the value, the less obstruction the vehicle will encounter when leaving the corresponding refueling spot, and the lower the value, the greater the obstruction the vehicle will encounter when leaving the corresponding refueling spot. For vehicles with low occupancy levels, the available refueling spot with the lowest departure convenience value is selected from the list of available refueling spots as the target refueling spot. Vehicles with low occupancy levels have a shorter expected parking time and can leave quickly after refueling. Parking in a parking spot with a lower departure convenience value will not have a long-term negative impact on the surrounding traffic flow. At the same time, high convenience parking space resources can be reserved for vehicles with longer parking times, thereby optimizing the allocation of parking space resources.

[0051] If multiple vacant refueling spots with the same exit convenience value appear during the screening process, for vehicles with low occupancy levels, the refueling spot with the shortest driving distance from the gas station entrance is selected as the target refueling spot. This reduces the vehicle's driving distance after entering the station and improves entry efficiency. For vehicles with high occupancy levels, the vacant refueling spot with the highest exit convenience value is selected from the list of vacant refueling spots as the target refueling spot. Vehicles with high occupancy levels are expected to stay for a longer period; parking in a spot with higher exit convenience allows them to quickly leave the gas station after refueling, reducing the probability of vehicles occupying the core traffic area for extended periods and lowering the risk of congestion in surrounding parking spaces. This effectively improves the overall traffic flow turnover efficiency of the gas station. If multiple vacant refueling spots with the same exit convenience value appear during the screening process, for vehicles with high occupancy levels, the refueling spot with the shortest driving distance from the gas station exit is selected as the target refueling spot. This further shortens the vehicle's exit path after refueling and speeds up the vehicle's departure. After determining the target refueling spot, the dispatching and guidance unit generates a corresponding guidance signal, which includes the target refueling spot information. The two core components are parking space numbering and guidance routes. The target refueling space number is a unique number uniformly assigned to each parking space within the gas station, with the numbering rules completely consistent with the on-site signage, facilitating drivers' quick identification of the corresponding parking space. The guidance route is generated based on the gas station's road network topology data, starting from the current vehicle's real-time location and ending at the target refueling space's stopping point. During route planning, it avoids driving conflict areas corresponding to currently occupied parking spaces, selecting the route with the shortest driving distance and lowest intersection risk as the guidance route. After the guidance signal is generated, it is sent to the vehicle display terminal or the driver's mobile device. The vehicle display terminal includes the guidance display screen at the gas station entrance and the guidance screen next to the refueling island, which can directly display guidance information to drivers in the area. The driver's mobile device includes the driver's mobile terminal, which can receive guidance information through the gas station's corresponding mini-program or SMS channel, guiding drivers to accurately park at the target refueling space according to the planned route. This can fully utilize the resource value of parking spaces with different levels of convenience, improving the overall parking space turnover efficiency and traffic capacity of the gas station without increasing hardware investment.

[0052] Please see Figure 2 As shown, the second objective of this invention is to provide a method for implementing an AI-based self-service refueling tiered guidance and vehicle dispatching system, comprising the following steps: Obtain the vehicle's identity information, required fuel type, and physical characteristics upon arrival at the station; The system inputs identity information, vehicle physical characteristics, and associated historical refueling behavior data into an artificial intelligence model trained on historical refueling data, and outputs the estimated time taken. Real-time monitoring of the occupancy status of each refueling station and the type of fuel provided; For multiple available refueling stations on the same side as the required fuel, the pre-marked ease of departure for each station is read. The ease of departure is determined based on whether the vehicle needs to wait for the adjacent vehicle to leave when driving from the refueling station to the exit and the degree of path conflict. Compare the estimated usage time with the set duration threshold to classify the system into low and high usage levels. Vehicles with low occupancy levels are assigned to the vacant refueling stations with the lowest ease of departure, while vehicles with high occupancy levels are assigned to the vacant refueling stations with the highest ease of departure. A guidance signal containing the target refueling station number and the guidance route is generated and sent.

[0053] 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 preferred examples and are not intended to limit 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 present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A self-service refueling tiered guidance and vehicle dispatching system based on artificial intelligence, characterized in that: include: The vehicle profiling unit is used to obtain the identity information of the vehicle entering the station, the required fuel type, and the vehicle's physical characteristics, including the location of the fuel tank opening and the vehicle's dimensions. The occupancy time prediction unit has a built-in artificial intelligence model trained on historical refueling data, which is used to predict the estimated occupancy time of the current vehicle based on the vehicle's physical characteristics and historical refueling behavior data associated with the identity information. The parking space status detection unit is used to detect the occupancy status of each refueling station and the type of fuel provided in real time; The departure convenience calibration unit is used to store the departure convenience of each pre-calibrated refueling station. The departure convenience is determined based on whether the vehicle needs to wait for the adjacent vehicle to leave when driving from the refueling station to the exit and the degree of path conflict. The lowest departure convenience indicates the greatest obstacle to departure. The graded guidance unit is used to compare the estimated occupancy time with a set time threshold and classify vehicles into low occupancy levels and high occupancy levels. The scheduling and guidance unit is used to allocate vehicles with low occupancy levels to the empty refueling position with the lowest ease of departure and vehicles with high occupancy levels to the empty refueling position with the highest ease of departure when there are multiple empty refueling positions on the same side as the required fuel, and to generate corresponding guidance signals.

2. The AI-based self-service refueling tiered guidance and vehicle dispatching system according to claim 1, characterized in that: The artificial intelligence model in the occupancy duration prediction unit is trained in the following way: Collect historical refueling data, which includes the identity information, physical characteristics, actual occupancy time, and historical refueling behavior data of each vehicle that has entered the station in history. Using identity information, vehicle physical characteristics, and historical refueling behavior data from historical refueling data as input features for training samples, and actual occupancy time as labels for training samples, a training dataset is constructed. A regression neural network model is used as the initial model, and supervised learning training of the regression neural network model is performed using the training dataset. During the training process, the weight parameters are adjusted through the backpropagation algorithm until the loss function converges, resulting in an artificial intelligence model.

3. The AI-based self-service refueling tiered guidance and vehicle dispatching system according to claim 2, characterized in that: The predicted estimated occupancy time of the current vehicle specifically includes: The current vehicle's identity information, physical characteristics, and the refueling frequency, average refueling amount, and historical dwell time distribution characteristics extracted from historical refueling behavior data are concatenated and combined to form an input feature vector. This input feature vector is then fed into the artificial intelligence model. After nonlinear transformation processing by the hidden layer of the artificial intelligence model, the estimated dwell time is output by the output layer.

4. The AI-based self-service refueling tiered guidance and vehicle dispatching system according to claim 1, characterized in that: For each refueling station within the gas station, the shortest driving path from the center point of that refueling station to the exit is collected. All nodes that intersect with other vehicles are marked on the shortest driving path. Based on the number of nodes and the probability of waiting for adjacent vehicles to leave at each node, the path conflict degree of that refueling station is calculated. At the same time, the distance between adjacent parking spaces on both sides of the refueling station and the obstruction angle caused by adjacent vehicles parking in the same parking space to the vehicle leaving the current parking space are measured. The path conflict degree and obstruction angle are combined and a comprehensive score is obtained by weighted summation. This comprehensive score is then mapped to a departure convenience value.

5. The AI-based self-service refueling tiered guidance and vehicle dispatching system according to claim 4, characterized in that: For each refueling station, the typical driving trajectory of the vehicle leaving the exit is simulated. All conflict areas where the path intersects with the vehicle in the adjacent parking space are identified on the typical driving trajectory. In each conflict area, the conflict weight of the conflict area is calculated based on the occupancy probability of the adjacent parking space and the spatiotemporal overlap of the two vehicles appearing in the conflict area at the same time. The conflict weights of all conflict areas are summed to obtain the path conflict degree value of the refueling station.

6. The AI-based self-service refueling tiered guidance and vehicle dispatching system according to claim 5, characterized in that: First, determine whether a vehicle needs to wait for a neighboring vehicle to leave when leaving the refueling station. If no waiting is required, add points to the basic convenience score. If waiting is required, deduct points based on the expected waiting time. Then, convert the path conflict score into a path conflict deduction score according to a preset ratio. Finally, subtract the path conflict deduction score from the result of the basic convenience score addition or deduction to obtain the final departure convenience score.

7. The AI-based self-service refueling tiered guidance and vehicle dispatching system according to claim 6, characterized in that: The occupancy status of the adjacent refueling spaces on the left and right sides of the refueling space is obtained in real time by the parking space status detection unit. If at least one side of the adjacent parking space is occupied and the vehicle in the adjacent parking space on that side has not yet finished refueling, it is determined that it is necessary to wait for the vehicle in the adjacent parking space on that side to leave. Otherwise, it is determined that there is no need to wait. When both sides need to wait, the larger value of the waiting time on both sides is taken as the expected waiting time value.

8. The AI-based self-service refueling tiered guidance and vehicle dispatching system according to claim 1, characterized in that: Obtain a set duration threshold, which is a fixed value predetermined based on the median duration of all vehicle occupancy in historical statistics. Vehicles with an expected occupancy time less than or equal to the set duration threshold are marked as low occupancy level, and vehicles with an expected occupancy time greater than the set duration threshold are marked as high occupancy level.

9. The AI-based self-service refueling tiered guidance and vehicle dispatching system according to claim 1, characterized in that: The system obtains a list of all available refueling stations on the same side as the fuel required by the current vehicle from the parking space status detection unit. It then reads the ease of leaving value for each available refueling station. For vehicles with low occupancy levels, the system selects the available refueling station with the lowest ease of leaving value from the list as the target refueling station. For vehicles with high occupancy levels, the system selects the available refueling station with the highest ease of leaving value from the list as the target refueling station. Finally, a guidance signal containing the target refueling station number and the guidance route is generated and sent to the vehicle display terminal or the driver's mobile device.

10. A method for implementing an AI-based self-service refueling tiered guidance and vehicle dispatching system according to any one of claims 1-9, characterized in that: Includes the following steps: Obtain the vehicle's identity information, required fuel type, and physical characteristics upon arrival at the station; The identity information, vehicle physical characteristics, and associated historical refueling behavior data are input into an artificial intelligence model trained on historical refueling data, and the estimated time taken is output. Real-time monitoring of the occupancy status of each refueling station and the type of fuel provided; For multiple vacant refueling stations on the same side as the required fuel, the pre-marked ease of departure for each station is read. The ease of departure is determined based on whether the vehicle needs to wait for the adjacent vehicle to leave when driving from the refueling station to the exit and the degree of path conflict. Compare the estimated usage time with the set duration threshold to classify the system into low and high usage levels. Vehicles with low occupancy levels are assigned to the vacant refueling stations with the lowest ease of departure, while vehicles with high occupancy levels are assigned to the vacant refueling stations with the highest ease of departure. A guidance signal containing the target refueling station number and the guidance route is generated and sent.

Images