Optimization System for Group Pickup and Drop-off of Autonomous Vehicles Based on Group Behavior Prediction

Abstract

The present invention discloses an optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction, including a passenger group data collection module, a group behavior prediction and pattern analysis module, a multi-modal large model optimization module, a group pick-up and drop-off point priority ranking and dynamic adjustment module, and a real-time group interaction and feedback module; By analyzing the behaviors, needs, locations, and future movement trajectories of multiple passengers in real time and combining group behavior prediction technology, the autonomous vehicle can dynamically adjust pick-up and drop-off points and driving routes to maximize pick-up and drop-off efficiency and passenger experience. The system of the present invention is particularly suitable for scenarios such as shared travel, multi-passenger carpooling, and dynamic multi-person pick-up and drop-off scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and more particularly, to an optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction. Background Art

[0002] With the development of autonomous driving technology, the level of vehicle intelligence and automation has been continuously improved, which provides a technical basis for group pick-up and drop-off of vehicles. Autonomous vehicles can perceive the environment and make decisions through sensors and algorithms, thus achieving more efficient and safe transportation in group pick-up and drop-off.

[0003] Traditional pick-up and drop-off systems often rely on fixed routes and schedules, lacking flexibility and real-time performance. This leads to waste of resources and increased passenger waiting time. In addition, these systems usually cannot effectively cope with emergencies such as traffic congestion or weather changes and cannot anticipate sudden traffic situations in advance.

[0004] Currently, the focus of research on traffic flow allocation tends to be on the mechanism of route selection or more effective solution algorithms under different information conditions, but less research has been done on its important basis, the "dynamic decision-making needs of group travel", which restricts the progress of research on dynamic traffic allocation theory and applications, resulting in the difficulty of existing route planning methods in reflecting group dynamic decision-making and unable to meet the needs of people in scientific travel, intelligent guidance, etc. Summary of the Invention

[0005] The purpose of the present invention is to provide an optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction to solve the above problems existing in the prior art.

[0006] Specifically, this application is as follows:

[0007] An optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction includes a passenger group data collection module, a group behavior prediction and pattern analysis module, a multi-modal large model optimization module, a group pick-up and drop-off point priority ranking and dynamic adjustment module, and a real-time group interaction and feedback module;

[0008] The passenger group data collection module is used to collect real-time data of each passenger in the group, including their current location information, predetermined pick-up and drop-off points, demand preferences, and historical behavior data;

[0009] The group behavior prediction and pattern analysis module analyzes the behavior patterns of each passenger in the group through a multi-modal large model, and combines the current environmental data and historical behaviors to predict the future action paths and time arrangements of each passenger in the group;

[0010] The multimodal large model optimization module is used to combine the group behavior prediction results of the group behavior prediction and pattern analysis module, integrate various input data through the multimodal large model, and dynamically generate optimal pick-up and drop-off points and route planning;

[0011] The group pick-up and drop-off point priority sorting and dynamic adjustment module dynamically adjusts the pick-up and drop-off order and the priority of pick-up and drop-off points by predicting the needs and behaviors of passengers in the group;

[0012] The real-time group interaction and feedback module maintains real-time interaction with each passenger through mobile devices, provides dynamic updates of the current pick-up and drop-off points, driving routes, and feedback on the group pick-up and drop-off order;

[0013] The passenger group data collection module, the group behavior prediction and pattern analysis module, the multimodal large model optimization module, the group pick-up and drop-off point priority sorting and dynamic adjustment module, and the real-time group interaction and feedback module are interconnected with each other.

[0014] Furthermore, the passenger group data collection module includes a location data collection sub-module, a demand input sub-module, and a behavior data integration sub-module;

[0015] The location data collection sub-module obtains the current location of passengers through their GPS, Wi-Fi signals, and Bluetooth;

[0016] The demand input sub-module is used for passengers to input pick-up and drop-off demands through mobile devices, including specific pick-up and drop-off points, time limits, and preferred routes;

[0017] The behavior data integration sub-module conducts personalized analysis through the historical behavior data of passengers, and the historical behavior data includes riding habits, frequently used pick-up and drop-off points, and preferred driving routes.

[0018] Furthermore, the specific implementation process of the location data collection sub-module is as follows:

[0019] L1: Data collection, obtain the outdoor location information of passengers through GPS to generate GPS data; construct a fingerprint database by collecting Wi-Fi signal characteristics for indoor location information of passengers to generate Wi-Fi data; collect location information through Bluetooth beacons by collecting beacon broadcast data packets received by the terminal devices of passengers to generate BLE data, and the beacon broadcast data packets include at least the MAC address and RSSI signal strength;

[0020] L2: Preprocessing, preprocess the collected GPS data, Wi-Fi data, and BLE data, including filtering out noise and data synchronization;

[0021] L3: Data fusion, which fuses GPS data, Wi-Fi data, and BLE data through the Kalman filter algorithm;

[0022] The specific implementation process of the above L3 is as follows:

[0023] L31: Set the initial state and covariance matrix of the Kalman filter, including the process noise covariance and the observation noise covariance;

[0024] L32: Predict the state and covariance at the current moment based on the state at the previous moment;

[0025] L33: Update the state estimate and covariance using the new observation data;

[0026] L34: Repeat steps L32 and L33 to achieve continuous tracking and optimization of the passenger's location;

[0027] The specific implementation process of the above demand input sub-module is as follows:

[0028] The passenger submits a pick-up and drop-off request through a mobile device. By using natural language processing technology to process the passenger's submitted pick-up and drop-off request, the pre-trained BERT language model is adopted to analyze the demand description in the passenger's pick-up and drop-off request, and the demand description is converted into pick-up and drop-off requirements; the geographical location and time in the pick-up and drop-off requirements are identified through the named entity recognition NER technology, and the geographical location and time are converted into parameters in the pick-up and drop-off plan;

[0029] The specific implementation process of the above behavior data integration sub-module is as follows:

[0030] Based on the passenger's historical behavior data, the Hidden Markov Model is used to model the passenger's behavior to generate a passenger future behavior prediction model;

[0031] The input of the training of the above passenger future behavior prediction model is the passenger's historical behavior data, and the output is the passenger's future behavior pattern. The passenger's future behavior pattern includes whether it is expected that the passenger will appear on time, the passenger's preferred route, and the common behavior of the passenger during a specific time period;

[0032] The formula of the above Hidden Markov Model is:

[0033] P(A|θ) = ∑ allstates P(A, Q|θ),

[0034] where A is the observed behavior sequence, θ is the model parameter, the model parameter is obtained through maximum likelihood estimation, and Q is the state of the behavior sequence.

[0035] Furthermore, the above group behavior prediction and pattern analysis module includes a behavior pattern analysis sub-module, a trajectory prediction sub-module, and a group behavior prediction sub-module;

[0036] The behavior pattern analysis sub-module identifies the behavior pattern of a passenger by analyzing the historical behavior data of the passenger. The behavior pattern at least includes the preference for getting on and off at certain specific times and locations.

[0037] The trajectory prediction sub-module predicts the future movement path of the passenger based on the current location and environmental information.

[0038] The group behavior prediction sub-module generates the optimal pick-up and drop-off sequence and path for the entire group through the behavior patterns of multiple passengers, and adjusts the plan according to the needs of different passengers.

[0039] Furthermore, the specific implementation process of the behavior pattern analysis sub-module is as follows:

[0040] Based on the historical behavior data of the passenger, the K-means clustering algorithm is used to analyze the behavior patterns of different passengers, and classification is performed based on the results of the pattern analysis.

[0041] The specific implementation process of the trajectory prediction sub-module is as follows:

[0042] The spatio-temporal trajectory prediction model based on the long short-term memory network (LSTM) is used to predict the movement trajectory of the passenger. The training input of the spatio-temporal trajectory prediction model is the historical trajectory of the passenger, and the output is the future movement trajectory of the passenger.

[0043] The specific implementation process of the group behavior prediction sub-module is as follows:

[0044] A model that combines the graph convolutional network (GCN) and the Transformer attention mechanism is used to globally optimize the behavior of multiple passengers. The graph convolutional network (GCN) predicts the overall behavior pattern by processing the relationship network between group members, and the Transformer is used to capture the complex dependencies of the relationship network to generate the optimal pick-up and drop-off sequence and path plan.

[0045] The specific implementation process of the optimal pick-up and drop-off sequence and path plan is as follows:

[0046] S1: Division of service areas for multiple passengers based on the K-Means++ algorithm;

[0047] S2: Determination of ride-sharing points for multiple passengers based on the DBSCAN algorithm;

[0048] S3: Generation of the optimal pick-up and drop-off sequence path based on maximum likelihood estimation;

[0049] The specific implementation process of S1 is as follows:

[0050] S11: Input the coordinate data set N containing the demand points of n passengers, N = {(xi , y i ), where \(i = 1, 2, 3, \cdots, n\). Input the preset number of categories \(k\), where the value of \(k\) is determined according to the required quantity, that is Q is the vehicle seat capacity;

[0051] S12: Randomly select a point \((x p , y p ) as the clustering center in the input data samples, and calculate the shortest distance between each sample and the current clustering center, denoted by \(D pi . The \(D pi uses a third-party map API interface to obtain the longitude and latitude coordinates of passenger demand points, and then obtains the driving distance between two passenger demand points;

[0052] S13: Calculate the probability that each sample point is selected as the next clustering center point

[0053] S14: Repeat steps S12 and S13 until \(k\) clustering centers are selected;

[0054] S15: Calculate the distances between each data sample and the \(k\) clustering centers, and select the clustering center with the closest distance to be grouped into one category;

[0055] S16: Calculate the distance mean in each cluster, update the cluster center through the distance means of the abscissa and ordinate, and repeat steps S15 and S16 until convergence;

[0056] S17: Extract the clustering results, and the clustering results where the number of data samples in each cluster is not greater than the single-vehicle passenger capacity are used as the single-time \(n\) passenger demand points;

[0057] The specific implementation process of the above S2 is as follows:

[0058] S21: Input the data set \(N=\{(x i , y i )|i = 1, 2, 3, \cdots, N\}, specify the Eps neighborhood radius as \(r\), and the minimum number of samples in the MinPts neighborhood as \(q\), where both MinPts and Eps are parameters in the DBSCAN clustering algorithm;

[0059] S22: Initialize the number of clusters \(k' = 0\), initialize the set of unvisited samples \(P = D\), and the carpool point set

[0060] S23: Randomly select a data sample \(o i =(x i , y i ). If \(|N Eps (o i )| \geq MinPts\), then \(oi is the core object, taking o i Add cluster C i to the set, where C i ={o i}, and add the neighborhood subset N Eps (o i ) to C i ;

[0061] S24: Update the data sample P = P - C i , repeat step S23 until each data sample point is visited, and obtain the final carpooling point set C k' ={C1, C2,..., C i};

[0062] S25: For the pick-up and drop-off points in the same cluster C i , randomly select one point as the final pick-up / drop-off point;

[0063] The specific implementation process of the above S3 is as follows:

[0064] Adopt the method of unary linear regression, fit all passengers' carpooling points by the least squares method to generate a uniform linear regression line of the optimal pick-up and drop-off order, and then use the Dijkstra algorithm to input the linear regression line into the actual road network route map for intersection calculation, finally generate the shortest path starting from cluster P0, and mark each node as P1, P2,..., P n .

[0065] Furthermore, the multimodal large model optimization module includes a multi-data input processing sub-module, a dynamic traffic analysis sub-module, and a personalized optimization sub-module;

[0066] The multi-data input processing sub-module parses the instructions and requirements input by passengers through natural language processing technology, and confirms the pick-up and drop-off point locations by combining image and text data;

[0067] The dynamic traffic analysis sub-module dynamically optimizes the pick-up and drop-off point selection and driving route by analyzing real-time traffic data, weather information, and road conditions;

[0068] The personalized optimization sub-module generates a personalized pick-up and drop-off plan for each passenger, and the personalized pick-up and drop-off plan needs to combine the specific needs of the passenger and the needs of other members in the group.

[0069] Furthermore, the specific implementation process of the multi-data input processing sub-module is as follows:

[0070] The input of passenger voice, images, and text is processed using a multimodal fusion CLIP model, and the output of the multimodal fusion CLIP model is the location information of the passenger, which is used to fuse real-time image data to confirm the pick-up and drop-off points;

[0071] The specific implementation process of the dynamic traffic analysis sub-module is as follows:

[0072] The real-time traffic information and road conditions are analyzed through a deep Q-network model and the proximal operator Proximal, and the pick-up and drop-off routes of the vehicle are dynamically adjusted;

[0073] The specific implementation process of the personalized optimization sub-module is as follows:

[0074] A genetic algorithm using multi-objective optimization is used to generate an optimization plan for each passenger's personalized needs, and the personalized needs include preferred pick-up and drop-off times and specific route preferences.

[0075] Furthermore, the group pick-up and drop-off point priority ranking and dynamic adjustment module includes a pick-up and drop-off point ranking sub-module and a dynamic adjustment sub-module;

[0076] The pick-up and drop-off point ranking sub-module ranks the pick-up and drop-off points according to each passenger's needs, and the needs at least include time limits and preferred locations;

[0077] The dynamic adjustment sub-module is used to automatically adjust the pick-up and drop-off order and route when the needs or locations of a certain passenger in the group change, and the location change at least includes being late or temporarily changing the pick-up and drop-off point.

[0078] Furthermore, the specific implementation process of ranking the pick-up and drop-off points in the pick-up and drop-off point ranking sub-module is as follows:

[0079] The priority ranking uses the multi-criteria decision-making TOPSIS algorithm. Based on the passengers' needs, the priority ranking of the pick-up and drop-off points is generated, the positive ideal solution and negative ideal solution of each pick-up and drop-off point are calculated, and a heat map is generated according to the positive ideal solution and negative ideal solution. According to the distribution of each pick-up and drop-off point in the heat map, those with dense distribution have high scores, and those with sparse distribution have low scores. The best pick-up and drop-off order is determined according to the scores, and the passengers' needs include time limits and location preferences;

[0080] The specific implementation process of the dynamic adjustment sub-module is as follows:

[0081] The deep Q-network (DQN) reinforcement learning algorithm is used to adjust the pick-up and drop-off order in real time. When the needs or locations of passengers change, the pick-up and drop-off order and path are recalculated according to the current environment, and the plan is adjusted according to the latest traffic and passenger data. The current environment includes passengers being late and temporarily changing the pick-up and drop-off point.

[0082] Further, the real-time group interaction and feedback module includes a real-time notification submodule and a feedback and adjustment submodule;

[0083] The real-time notification submodule is used to notify each passenger of the update of the pick-up point, estimated arrival time, and vehicle location through the application APP, and allow passengers to view the pick-up order and route in the group in real time;

[0084] The feedback and adjustment submodule is used for passengers to provide real-time feedback or adjust their pick-up and drop-off requirements through the APP, and dynamically adjust the group pick-up and drop-off plan based on the feedback or adjustment of the pick-up and drop-off requirements.

[0085] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0086] The embodiment of the present invention provides a passenger group data collection module for collecting real-time data of each passenger in the group, including their current location information, scheduled pick-up and drop-off points, demand preferences, and historical behavior data; the group behavior prediction and pattern analysis module analyzes the behavior pattern of each passenger in the group through a multimodal large model, and combines the current environmental data and historical behavior to predict the future action path and time schedule of each passenger in the group; the multimodal large model optimization module is used to combine the group behavior prediction results of the group behavior prediction and pattern analysis module, integrate various input data through the multimodal large model, and dynamically generate the optimal pick-up and drop-off points and path planning; group pick-up and drop-off The point priority sorting and dynamic adjustment module dynamically adjusts the pick-up order and pick-up point priority by predicting the needs and behaviors of passengers in the group; the real-time group interaction and feedback module maintains real-time interaction with each passenger through mobile devices, provides dynamic updates of current pick-up points and driving routes, and feedback on the group pick-up order; the present invention conducts real-time analysis of the behaviors, needs, locations and future movement trajectories of multiple passengers, combined with group behavior prediction technology, so that the autonomous driving vehicle can dynamically adjust the pick-up points and driving routes to maximize the pick-up efficiency and passenger experience. The system of the present invention is particularly suitable for shared travel, multi-passenger carpooling, and dynamically changing multi-person pick-up scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 is a system architecture diagram of an automatic driving vehicle group pick-up and drop-off optimization system based on group behavior prediction provided by an embodiment of the present invention;

[0088] Figure 2 This is a diagram of a passenger group data collection module of an automatic driving vehicle group pick-up and drop-off optimization system based on group behavior prediction provided by an embodiment of the present invention;

[0089] Figure 3 It is a group behavior prediction and pattern analysis module diagram of an automatic driving vehicle group pick-up optimization system based on group behavior prediction provided by an embodiment of the present invention;

[0090] Figure 4 It is a diagram of the multi-modal large model optimization module of the optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction provided by an embodiment of the present invention;

[0091] Figure 5 It is a diagram of the module for prioritizing and dynamically adjusting the group pick-up and drop-off points of the optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction provided by an embodiment of the present invention;

[0092] Figure 6 It is a diagram of the real-time group interaction and feedback module of the optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction provided by an embodiment of the present invention. Detailed implementation manners

[0093] The present invention will be described in detail below with reference to the accompanying drawings.

[0094] Embodiment 1

[0095] An embodiment of the present invention provides an optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction, as Figure 1 shown, which includes a passenger group data collection module, a group behavior prediction and pattern analysis module, a multi-modal large model optimization module, a module for prioritizing and dynamically adjusting group pick-up and drop-off points, and a real-time group interaction and feedback module;

[0096] The passenger group data collection module is used to collect real-time data of each passenger in the group, including their current location information, scheduled pick-up and drop-off points, demand preferences, and historical behavior data;

[0097] The group behavior prediction and pattern analysis module analyzes the behavior patterns of each passenger in the group through a multi-modal large model, and combines the current environmental data and historical behaviors to predict the future action paths and time arrangements of each passenger in the group;

[0098] The multi-modal large model optimization module is used to combine the group behavior prediction results of the group behavior prediction and pattern analysis module, integrate various input data through the multi-modal large model, and dynamically generate optimal pick-up and drop-off points and path planning;

[0099] The module for prioritizing and dynamically adjusting group pick-up and drop-off points dynamically adjusts the pick-up and drop-off order and the priority of pick-up and drop-off points through the prediction of the demands and behaviors of passengers in the group;

[0100] The real-time group interaction and feedback module maintains real-time interaction with each passenger through a mobile device, provides dynamic updates of the current pick-up and drop-off points, driving paths, and feedback on the group pick-up and drop-off order;

[0101] The passenger group data collection module, the group behavior prediction and pattern analysis module, the multi-modal large model optimization module, the group pick-up and drop-off point priority ranking and dynamic adjustment module, and the real-time group interaction and feedback module are interconnected with each other.

[0102] Further, the passenger group data collection module includes a location data collection sub-module, a demand input sub-module, and a behavior data integration sub-module. Please refer to Figure 2 ;

[0103] The location data collection sub-module obtains the current location of the passenger through the passenger's GPS, Wi-Fi signal, and Bluetooth; the demand input sub-module is used for the passenger to input pick-up and drop-off demands through a mobile device, including specific pick-up and drop-off points, time limits, and preferred routes.

[0104] The behavior data integration sub-module performs personalized analysis through the passenger's historical behavior data, and the historical behavior data includes riding habits, frequently used pick-up and drop-off points, and preferred driving routes.

[0105] Further, the specific implementation process of the location data collection sub-module is as follows:

[0106] L1: Data collection, obtain the outdoor location information of the passenger through GPS to generate GPS data; construct a fingerprint database by collecting Wi-Fi signal characteristics for the indoor location information of the passenger to generate Wi-Fi data; collect the location information by the Bluetooth beacon to obtain the beacon broadcast data packet received by the passenger's terminal device to generate BLE data, and the beacon broadcast data packet includes at least the MAC address and RSSI signal strength.

[0107] L2: Preprocessing, preprocess the collected GPS data, Wi-Fi data, and BLE data, including filtering out noise and data synchronization.

[0108] L3: Data fusion, fuse the GPS data, Wi-Fi data, and BLE data through the Kalman filter algorithm.

[0109] The specific implementation process of L3 is as follows:

[0110] L31: Set the initial state and covariance matrix of the Kalman filter, including the process noise covariance and the observation noise covariance.

[0111] L32: Predict the state and covariance of the current moment according to the state of the previous moment.

[0112] L33: Update the state estimate and covariance using the new observation data.

[0113] L34: Repeat steps L32 and L33 to achieve continuous tracking and optimization of the passenger's location;

[0114] The specific implementation process of the demand input sub-module is as follows:

[0115] The passenger submits a pick-up and drop-off request through a mobile device. By using natural language processing technology to process the passenger's pick-up and drop-off request, a pre-trained BERT language model is used to analyze the demand description in the passenger's pick-up and drop-off request, and the demand description is converted into a pick-up and drop-off demand; the geographical location and time in the pick-up and drop-off demand are identified through named entity recognition NER technology, and the geographical location and time are converted into parameters in the pick-up and drop-off plan;

[0116] The specific implementation process of the behavior data integration sub-module is as follows:

[0117] Based on the passenger's historical behavior data, a hidden Markov model is used to model the passenger's behavior, and a passenger future behavior prediction model is generated;

[0118] The input of the training of the passenger future behavior prediction model is the passenger's historical behavior data, and the output is the future behavior pattern of the passenger. The future behavior pattern of the passenger includes whether it is expected that the passenger will appear on time, the passenger's preferred route, and the common behavior of the passenger in a specific time period;

[0119] The formula of the hidden Markov model is:

[0120] P(A|θ) = ∑ allstates P(A,Q|θ),

[0121] where A is the observed behavior sequence, θ is the model parameter, the model parameter is obtained through maximum likelihood estimation, and Q is the state of the behavior sequence.

[0122] Furthermore, the group behavior prediction and pattern analysis module includes a behavior pattern analysis sub-module, a trajectory prediction sub-module, and a group behavior prediction sub-module. Please refer to Figure 3 ;

[0123] The behavior pattern analysis sub-module identifies the behavior pattern of the passenger through the analysis of the passenger's historical behavior data. The behavior pattern at least includes the preference for getting on and off at certain specific times and locations;

[0124] The trajectory prediction sub-module predicts the future movement path of the passenger based on the current location and environmental information;

[0125] The group behavior prediction sub-module generates the best pick-up and drop-off order and path for the entire group through the behavior patterns of multiple passengers, and adjusts the plan according to the needs of different passengers.

[0126] Specifically, the interaction process between the passenger group data collection module and the group behavior prediction and pattern analysis module is as follows:

[0127] Interaction description:

[0128] Data transmission: The passenger group data collection module first obtains the current location information, historical behavior data (such as the time and location of frequent rides), and demand inputs (such as preferred pick-up and drop-off points, route selection) of each passenger, and these data are transmitted to the group behavior prediction and pattern analysis module in real time;

[0129] Real-time update: According to the behavior changes of passengers, such as passengers temporarily changing pick-up and drop-off requirements or location information changing, the data collection module transmits the latest status of the data to the behavior prediction module to trigger the recalculation of behavior patterns and trajectory prediction;

[0130] Interaction algorithms and models:

[0131] Combination of Markov model HMM and long short-term memory network LSTM:

[0132] The hidden Markov model analyzes the long-term behavior patterns of passengers based on historical behavior data, including common pick-up and drop-off locations and frequencies.

[0133] The LSTM model performs spatio-temporal prediction in the short term. By analyzing the current location and historical trajectory of passengers, it predicts their future action routes;

[0134] Data fusion:

[0135] Through the complementary combination of LSTM, it is possible to predict short-term trajectories based on sequence data;

[0136] Dynamic data update: The data collection module updates the passenger location information in real time and inputs it into the LSTM network. The updated location information triggers the trajectory prediction module to re-evaluate the future path.

[0137] Furthermore, the specific implementation process of the behavior pattern analysis sub-module is as follows:

[0138] Based on the historical behavior data of passengers, the clustering algorithm K-means is used to analyze the behavior patterns of different passengers, and classification is performed based on the results of the pattern analysis;

[0139] The specific implementation process of the trajectory prediction sub-module is as follows:

[0140] Use the spatio-temporal trajectory prediction model based on the long short-term memory network LSTM to predict the movement trajectory of passengers. The training input of the spatio-temporal trajectory prediction model is the historical trajectory of passengers, and the output is the future movement trajectory of passengers;

[0141] The specific implementation process of the group behavior prediction sub-module is as follows:

[0142] Use a model that combines the graph convolutional network GCN and the Transformer attention mechanism to globally optimize the behaviors of multiple passengers. The graph convolutional network GCN predicts the overall behavior pattern by processing the relationship network among group members, and the Transformer is used to capture the complex dependencies of the relationship network to generate the optimal pick-up and drop-off sequence and path plan;

[0143] The specific implementation process of the optimal pick-up and drop-off sequence and path plan is as follows:

[0144] S1: Division of service areas for multiple passengers based on the K-Means++ algorithm;

[0145] S2: Determination of ride-sharing points for multiple passengers based on the DBSCAN algorithm;

[0146] S3: Generation of the optimal pick-up and drop-off sequence path based on maximum likelihood estimation;

[0147] The specific implementation process of S1 is as follows:

[0148] S11: Input the coordinate data set N containing the demand points of n passengers, N = {(x i , y i )|i = 1, 2, 3,..., n}, and input the preset number of categories k, where the value of k is determined according to the demand quantity, that is Q is the vehicle seat capacity;

[0149] S12: Randomly select a point (x p , y p ) in the input data sample as the clustering center, and calculate the shortest distance between each sample and the current clustering center, denoted by D pi . The D pi uses a third-party map API interface to obtain the longitude and latitude coordinates of the passenger demand points, and then obtains the driving distance between two passenger demand points;

[0150] S13: Calculate the probability that each sample point is selected as the next clustering center point

[0151] S14: Repeat steps S12 and S13 until k clustering centers are selected;

[0152] S15: Calculate the distances between each data sample and the k clustering centers, and select the nearest clustering center to classify it into a category;

[0153] S16: Calculate the average distance in each cluster type, update the cluster center through the average distance of the abscissa and ordinate, and repeat steps S15 and S16 until convergence;

[0154] S17: Extract the clustering results. The clustering results where the number of data samples in each cluster is not greater than the single-vehicle passenger capacity are used as the demand points of n passengers for a single trip;

[0155] The specific implementation process of the said S2 is as follows:

[0156] S21: Input the data set N = {(x i , y i ) | i = 1, 2, 3,..., N}, specify the Eps neighborhood radius as r, and the minimum number of samples in the MinPts neighborhood as q, where both MinPts and Eps are parameters in the DBSCAN clustering algorithm;

[0157] S22: Initialize the number of clusters k' = 0, initialize the set of unvisited samples P = D, and the carpool point set

[0158] S23: Randomly select a data sample o i = (x i , y i ). If |N Eps (o i )| ≥ MinPts, then o i is a core object. Add o i to the cluster C i set, C i = {o i}, and add the neighborhood subset N Eps (o i ) to C i ;

[0159] S24: Update the data sample P = P - C i , and repeat step S23 until each data sample point is visited, obtaining the final carpool point set C k' = {C1, C2,..., C i};

[0160] S25: For the demand points in the same cluster C i , randomly select one point as the final pick-up / drop-off point;

[0161] The specific implementation process of the said S3 is as follows:

[0162] By using the method of unary linear regression, all passenger carpooling points are fitted by the least squares method to generate a uniform linear regression line for the optimal pick-up and drop-off sequence. Then, through the Dijkstra algorithm, the linear regression line is input into the actual road network route map for intersection calculation, and finally the shortest path starting from cluster P0 is generated, and each node is marked as P1, P2,..., P n 。

[0163] Furthermore, the multimodal large model optimization module includes a multi-data input processing sub-module, a dynamic traffic analysis sub-module, and a personalized optimization sub-module. Please refer to Figure 4 ;

[0164] The multi-data input processing sub-module parses the instructions and requirements input by passengers through natural language processing technology, and confirms the pick-up and drop-off points in combination with image and text data;

[0165] The dynamic traffic analysis sub-module dynamically optimizes the pick-up and drop-off point selection and driving route by analyzing real-time traffic data, weather information, and road conditions;

[0166] The personalized optimization sub-module generates a personalized pick-up and drop-off plan for each passenger, and the personalized pick-up and drop-off plan needs to combine the specific needs of the passenger and the needs of other members in the group.

[0167] Specifically, the interaction process between the group behavior prediction and pattern analysis module and the multimodal large model optimization module is as follows:

[0168] Interaction description:

[0169] Input data: The group behavior prediction and pattern analysis module provides the future trajectory prediction of passengers, the analysis results of behavior patterns, and the preliminary path planning based on traffic conditions. These data are used by the multimodal large model optimization module for comprehensive optimization of pick-up and drop-off points and paths.

[0170] Environmental perception: The multimodal large model module further optimizes the results of group behavior prediction by combining real-time information such as traffic conditions, weather, and road closures, including adjusting the route to avoid congestion or delays;

[0171] Interaction algorithms and models:

[0172] Combination of graph convolutional network GCN and Transformer:

[0173] GCN is used to model the relationship network between passengers, including their geographical locations, the closeness of demand times, etc., and generate a preliminary group pick-up and drop-off sequence;

[0174] The Transformer attention mechanism helps to capture the complex dependencies between group members and optimize the final pick-up and drop-off points and routes;

[0175] Group behavior prediction generates a graph structure among groups through a graph convolutional network (GCN).

[0176] Traffic optimization and dynamic environment analysis: The multi-modal large model integrates the results of the group behavior prediction module and applies a deep reinforcement learning model. The deep reinforcement learning model includes, but is not limited to, DQN or PPO for path planning. By continuously exploring the interaction with the traffic and weather of the environment, the deep reinforcement learning model adjusts the pick-up and drop-off routes.

[0177] Further, the specific implementation process of the multi-data input processing sub-module is as follows:

[0178] Use a multi-modal fusion CLIP model to process the input of passenger voice, images, and text. The output of the multi-modal fusion CLIP model is the passenger's location information, which is used to confirm the pick-up and drop-off points by fusing real-time image data.

[0179] The specific implementation process of the dynamic traffic analysis sub-module is as follows:

[0180] Analyze real-time traffic information and road conditions through a deep Q-network model and the proximal operator Proximal, and dynamically adjust the pick-up and drop-off routes of the vehicle.

[0181] The specific implementation process of the personalized optimization sub-module is as follows:

[0182] Use a genetic algorithm for multi-objective optimization to generate an optimization plan for each passenger's personalized needs, where the personalized needs include preferred pick-up and drop-off times and specific route preferences.

[0183] Further, the group pick-up and drop-off point priority ranking and dynamic adjustment module includes a pick-up and drop-off point ranking sub-module and a dynamic adjustment sub-module. Please refer to Figure 5 ;

[0184] The pick-up and drop-off point ranking sub-module ranks the pick-up and drop-off points according to the needs of each passenger, and the needs at least include time limits and preferred locations.

[0185] The dynamic adjustment sub-module is used to automatically adjust the pick-up and drop-off order and routes when the needs or locations of a certain passenger in the group change. The change in location at least includes being late or temporarily changing the pick-up and drop-off points.

[0186] Specifically, the interaction process between the multi-modal large model optimization module and the group pick-up and drop-off point priority ranking and dynamic adjustment module is as follows:

[0187] Interaction description:

[0188] Optimize data transfer: The multi-modal large model module will transfer the preliminarily optimized pick-up and drop-off points and route plans to the pick-up and drop-off point priority sorting and dynamic adjustment module, which dynamically adjusts the priorities of the pick-up and drop-off points according to the priority requirements of passengers, real-time changing passenger demands or locations.

[0189] Real-time adjustment: According to the real-time data update, the priority sorting module feeds back the adjusted order of pick-up and drop-off points to the multi-modal large model, triggering the re-optimization of the route;

[0190] Interaction algorithms and models:

[0191] Combination of TOPSIS and deep reinforcement learning:

[0192] When determining the sorting of pick-up and drop-off points, the TOPSIS multi-criteria decision-making model sorts based on the needs of passengers, generates a series of priorities, and the needs of the passengers include time and location preferences for pick-up and drop-off points.

[0193] The reinforcement learning model DQN is used to adjust the order of pick-up and drop-off points in real time. When passenger demands or the environment change, the reinforcement learning model DQN re-adjusts the pick-up and drop-off points and routes according to the new feedback;

[0194] The formula of the TOPSIS multi-criteria decision-making model is:

[0195]

[0196] where S i is the priority score of the pick-up and drop-off point, w j is the weight of each factor, and a ij is the score of each pick-up and drop-off point under the j-th criterion;

[0197] Dynamic priority adjustment formula:

[0198] P(t + 1) = P(t) + αΔP(t), where P(t + 1) represents the updated priority of the pick-up and drop-off point, α represents the learning rate, and ΔP(t) represents the priority adjustment amount caused by real-time feedback.

[0199] Furthermore, the specific implementation process of prioritizing the pick-up and drop-off points in the pick-up and drop-off point sorting sub-module is as follows:

[0200] The priority sorting uses the multi-criteria decision-making TOPSIS algorithm. Based on the needs of passengers, it generates the priority sorting of pick-up and drop-off points, calculates the positive ideal solution and negative ideal solution of each pick-up and drop-off point, and generates a heat map according to the positive ideal solution and negative ideal solution. According to the distribution of each pick-up and drop-off point in the heat map, those with dense distribution have high scores, and those with sparse distribution have low scores. The best pick-up and drop-off order is determined according to the scores, and the needs of the passengers include time limits and location preferences;

[0201] The specific implementation process of the dynamic adjustment sub-module is as follows:

[0202] Use the deep Q-network (DQN) reinforcement learning algorithm to adjust the pick-up and drop-off order in real time. When the passenger demand or location changes, recalculate the pick-up and drop-off order and route according to the current environment, and adjust the plan based on the latest traffic and passenger data. The current environment includes passenger lateness and temporary changes to pick-up and drop-off points.

[0203] Furthermore, the real-time group interaction and feedback module includes a real-time notification sub-module and a feedback and adjustment sub-module. Please refer to Figure 6 ;

[0204] Specifically, the interaction process between the group pick-up and drop-off point priority sorting and dynamic adjustment module and the real-time group interaction and feedback module is as follows:

[0205] Interaction description:

[0206] Interaction data: The priority sorting and dynamic adjustment module pushes the adjusted pick-up and drop-off order and route to the real-time interaction and feedback module, which notifies each passenger of the new pick-up and drop-off plan and the dynamic itinerary information of the vehicle in real time;

[0207] Feedback mechanism: Passengers feedback their demand changes through the real-time interaction module, including temporary changes to pick-up and drop-off points. This feedback information is transmitted to the priority sorting and dynamic adjustment module through the interaction module, and further feedback to the multi-modal large model module to trigger the system's re-optimization process;

[0208] Interaction algorithms and models:

[0209] Real-time communication mechanism and intelligent feedback:

[0210] Ensure low-latency real-time communication through the WebSocket and MQTT protocols. The system can quickly push updated information to passengers when the pick-up and drop-off plan is adjusted.

[0211] The intelligent dialogue system Dialogflow receives passengers' feedback, parses passengers' instructions through natural language processing (NLP), converts them into instructions understood by the system, and further triggers data and algorithm linkages between modules within the system;

[0212] Real-time feedback formula:

[0213] T total = T notification + T response + T ajustment ,

[0214] where T total represents the total time for passenger feedback response, T notification represents the notification time, and T responseDenotes the passenger response time, T ajustment Denotes the time for the system to adjust the pick-up and drop-off sequence.

[0215] The real-time notification sub-module is used to notify each passenger of the update of the pick-up and drop-off point, the estimated arrival time, and the vehicle position through the application APP, and allows passengers to view the pick-up and drop-off sequence and route in the group in real time;

[0216] The feedback and adjustment sub-module is used for passengers to give real-time feedback or adjust pick-up and drop-off requirements through the application APP, and dynamically adjust the group pick-up and drop-off plan according to the feedback or adjusted pick-up and drop-off requirements.

[0217] The specific system comprehensive optimization process is as follows:

[0218] 1. Data collection: The location information, requirements, and behavior patterns of passengers are input into the system in real time through the data collection module;

[0219] 2. Behavior prediction: The group behavior prediction module predicts the trajectories of passengers based on the collected data, and initially generates the pick-up and drop-off sequence and route planning;

[0220] 3. Multimodal optimization: The multimodal large model optimization module combines real-time traffic, weather, and road information to dynamically adjust the initial route planning;

[0221] 4. Priority ranking and dynamic adjustment: Dynamically adjust the pick-up and drop-off point priorities and route order according to the changes in passengers' requirements and the results of behavior prediction;

[0222] 5. Real-time feedback and update: The system pushes the latest pick-up and drop-off plan to passengers through the real-time interaction module, and passengers' feedback further triggers the system to re-optimize.

[0223] Embodiment 2, Group Optimization in Multi-Passenger Carpooling

[0224] 1. Background description: The autonomous taxi receives carpooling requests from multiple passengers, and the system needs to dynamically optimize the pick-up and drop-off sequence according to the passengers' locations, requirements, and time limits;

[0225] 2. Group data collection:

[0226] Each passenger inputs pick-up and drop-off requirements through a smartphone and uploads the current location information;

[0227] The system confirms the specific locations of each passenger through GPS and Wi-Fi signals, and integrates their historical behavior data to predict the future movement trajectories of passengers;

[0228] 3. Group behavior prediction and optimization: The system analyzes the behavior patterns of each passenger in the group and generates the optimal pick-up and drop-off sequence and route planning according to the traffic conditions;

[0229] 4. Dynamic adjustment: If a passenger temporarily changes the pick-up point, the system recalculates the pick-up order in the group and pushes the updated plan to all passengers.

[0230] Example 3: Prioritization of Pickup and Dropoff Points in a Ride Sharing System

[0231] 1. Background description: Four passengers share a ride during peak hours. The system needs to generate the best pick-up order and driving route based on their pick-up needs and time constraints;

[0232] 2. Data input:

[0233] The system receives passengers' scheduled pick-up and drop-off points and time limits, and combines their historical behavior data for personalized optimization. Using GPS and real-time traffic data, the system calculates the optimal pick-up and drop-off sequence to ensure that time-sensitive passengers are picked up first;

[0234] 3. Group behavior prediction and feedback: The system pushes real-time updates of pick-up points and estimated arrival times to passengers, and allows passengers to adjust pick-up points according to their needs;

[0235] 4. Dynamic Optimization: In case of road closures or traffic congestion, the system automatically adjusts the pick-up order and notifies all passengers of the new route and estimated arrival time.

[0236] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages can be utilized to realize the content of the present invention described herein, and the description of the above specific languages is for disclosing the best mode of the present invention.

[0237] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0238] Similarly, it should be understood that, in order to streamline the present disclosure and assist in understanding one or more of the various inventive aspects, in the foregoing description of the exemplary embodiments of the present invention, the various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the following claims, the inventive aspects lie in less than all the features of the single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate embodiment of the present invention.

[0239] Those skilled in the art will appreciate that the modules in the devices in the embodiments can be adaptively changed and disposed in one or more devices different from the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and in addition, they can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all the features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all the processes or units of any method or device so disclosed. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) can be replaced by an alternative feature providing the same, equivalent, or similar purpose.

[0240] Furthermore, those skilled in the art will be able to understand that, although some of the embodiments herein include certain features included in other embodiments rather than other features, the combination of the features of different embodiments means that it is within the scope of the present invention and forms different embodiments. For example, in the following claims, any one of the claimed embodiments can be used in any combination.

[0241] The various component embodiments of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components in the device according to the embodiments of the present invention. The present invention can also be implemented as a device or device program for executing part or all of the methods described herein (for example, a computer program and a computer program product). Such a program implementing the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

Claims

1. An optimization system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction, characterized in that, It includes a passenger group data collection module, a group behavior prediction and pattern analysis module, a multi-modal large model optimization module, a group pick-up and drop-off point priority ranking and dynamic adjustment module, and a real-time group interaction and feedback module; The passenger group data collection module is used to collect real-time data of each passenger in the group, including their current location information, scheduled pick-up and drop-off points, demand preferences, and historical behavior data; The group behavior prediction and pattern analysis module analyzes the behavior patterns of each passenger in the group through a multi-modal large model, and combines the current environmental data and historical behavior to predict the future action paths and time arrangements of each passenger in the group; The multi-modal large model optimization module is used to combine the group behavior prediction results of the group behavior prediction and pattern analysis module, and dynamically generate the optimal pick-up and drop-off points and path planning by integrating various input data through the multi-modal large model; The group pick-up and drop-off point priority ranking and dynamic adjustment module dynamically adjusts the pick-up and drop-off order and pick-up and drop-off point priorities by predicting the needs and behaviors of the passengers in the group; The real-time group interaction and feedback module maintains real-time interaction with each passenger through a mobile device, provides dynamic updates of the current pick-up and drop-off points, driving paths, and feedback on the group pick-up and drop-off order; The passenger group data collection module, the group behavior prediction and pattern analysis module, the multi-modal large model optimization module, the group pick-up and drop-off point priority ranking and dynamic adjustment module, and the real-time group interaction and feedback module are interconnected; The group behavior prediction and pattern analysis module includes a behavior pattern analysis sub-module, a trajectory prediction sub-module, and a group behavior prediction sub-module; The behavior pattern analysis sub-module identifies its behavior pattern by analyzing the historical behavior data of the passenger, and the behavior pattern at least includes preferences for getting on and off at certain specific times and locations; The trajectory prediction sub-module predicts the future movement path of the passenger based on the current location and environmental information; The group behavior prediction sub-module generates the best pick-up and drop-off order and path for the entire group through the behavior patterns of multiple passengers, and adjusts the plan according to the needs of different passengers; The specific implementation process of the behavior pattern analysis sub-module is as follows: Based on the historical behavior data of the passengers, use the K-means clustering algorithm to perform pattern analysis on the behaviors of different passengers, and classify based on the results of the pattern analysis; The specific implementation process of the trajectory prediction sub-module is as follows: Use a spatio-temporal trajectory prediction model based on the long short-term memory network (LSTM) to predict the movement trajectory of the passenger. The training input of the spatio-temporal trajectory prediction model is the historical trajectory of the passenger, and the output is the future movement trajectory of the passenger; The specific implementation process of the group behavior prediction sub-module is as follows: Use a model that combines a graph convolutional network (GCN) and a Transformer attention mechanism to globally optimize the behaviors of multiple passengers. The graph convolutional network (GCN) predicts the overall behavior pattern by processing the relationship network between group members, and the Transformer is used to capture the complex dependencies of the relationship network to generate the optimal pick-up and drop-off order and path plan; The specific implementation process of the optimal pick-up and drop-off sequence and path plan is as follows: S1: Division of multiple passenger service areas based on the K-Means++ algorithm; S2: Determination of multiple passenger carpooling points based on the DBSCAN algorithm; S3: Generation of the optimal pick-up and drop-off sequence path based on maximum likelihood estimation; The specific implementation process of S1 is as follows: S11: Input the coordinate data set N containing the coordinates of n passenger demand points, N = {( , )|i = 1, 2, 3,..., n}, input the preset number of categories k, where the value of k is determined according to the demand quantity, that is, k = , and Q is the vehicle seat capacity; S12: Randomly select a point (x p , y p ) in the input data sample as the clustering center, and calculate the shortest distance between each sample and the current clustering center, denoted by . The D pi uses a third-party map API interface to obtain the longitude and latitude coordinates of the passenger demand points, and further obtains the driving distance between two passenger demand points; S13: Calculate the probability P that each sample point is selected as the next cluster center point, where P = ; S14: Repeat steps S12 and S13 until k clustering centers are selected; S15: Calculate the distances between each data sample and the k clustering centers, and assign the data sample to the cluster with the nearest distance; S16: Calculate the mean distance within each cluster, update the cluster center through the mean distances of the abscissa and ordinate, and repeat steps S15 and S16 until convergence; S17: Extract the clustering results, and use the clustering results where the number of data samples in each cluster is not greater than the single-vehicle passenger capacity as the demand points of n passengers for a single trip; The specific implementation process of S2 is as follows: S21: Input data set N = { (x i , y i ) | i = 1, 2, 3,..., N}, specify the Eps neighborhood radius as r, and the minimum number of samples in the MinPts neighborhood as q, where both MinPts and Eps are parameters in the DBSCAN clustering algorithm; S22: Initialize the number of clusters \(k' = 0\), initialize the set of unvisited samples \(P = D\), and the carpool point set \(C\) k = ; S23: Randomly select a data sample = (x i , y i ), if |N Eps (o i )| ≥ MinPts, then o i is a core object, add o i to the cluster C i set, C i = {o i}, and add the neighborhood subset N Eps (o i ) to C i ; S24: Update the data sample P = P - C i , repeat step S23 until each data sample point is visited to obtain the final carpooling point set C k = {C1, C2,..., C i}; S25: For the demand points within the same cluster C i select a point randomly as the final pick-up / drop-off point; The specific implementation process of S3 is as follows: Using the method of unary linear regression, all passenger carpooling points are fitted by the least squares method to generate a uniform linear regression line of the optimal pick-up and drop-off sequence. Then, through the Dijkstra algorithm, the linear regression line is input into the actual road network route map for intersection calculation, and finally the shortest path starting from cluster P0 is generated, and each node is marked as P1, P2,..., P n .

2. The optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction according to claim 1, wherein The passenger group data collection module includes a location data collection sub-module, a demand input sub-module, and a behavior data integration sub-module; The location data collection sub-module obtains the current location of the passenger through the passenger's GPS, Wi-Fi signal, and Bluetooth; The demand input sub-module is used for passengers to input pick-up and drop-off demands through mobile devices, including specific pick-up and drop-off points, time limits, and preferred routes; The behavior data integration sub-module conducts personalized analysis through the historical behavior data of passengers, and the historical behavior data includes riding habits, frequently used pick-up and drop-off points, and preferred driving routes.

3. The optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction according to claim 2, wherein, The specific implementation process of the location data collection sub-module is as follows: L1: Data collection. Obtain the outdoor location information of the passenger through GPS to generate GPS data; construct a fingerprint database by collecting Wi-Fi signal characteristics for the indoor location information of the passenger to generate Wi-Fi data; Collect the location information by the Bluetooth beacon to obtain the beacon broadcast data packet received by the passenger's terminal device to generate BLE data, and the beacon broadcast data packet includes at least the MAC address and RSSI signal strength; L2: Preprocessing. Preprocess the collected GPS data, Wi-Fi data, and BLE data, including noise filtering and data synchronization; L3: Data fusion. Perform fusion processing on the GPS data, Wi-Fi data, and BLE data through the Kalman filter algorithm; The specific implementation process of L3 is as follows: L31: Set the initial state and covariance matrix of the Kalman filter, including the process noise covariance and the observation noise covariance; L32: Predict the state and covariance at the current moment based on the state at the previous moment; L33: Update the state estimate and covariance using the new observation data; L34: Repeat steps L32 and L33 to achieve continuous tracking and optimization of the passenger's location; The specific implementation process of the demand input sub-module is as follows: The passenger submits a pick-up and drop-off request through a mobile device. By using natural language processing technology to process the passenger's pick-up and drop-off request, a pre-trained BERT language model is adopted to analyze the demand description in the passenger's pick-up and drop-off request, and the demand description is converted into pick-up and drop-off requirements; through named entity recognition NER technology, the geographical location and time in the pick-up and drop-off requirements are identified, and the geographical location and time are converted into parameters in the pick-up and drop-off plan. The specific implementation process of the behavior data integration sub-module is as follows: Based on the historical behavior data of the passenger, a Hidden Markov Model is used to model the passenger's behavior to generate a passenger future behavior prediction model. The input for training the passenger future behavior prediction model is the historical behavior data of the passenger, and the output is the future behavior pattern of the passenger. The future behavior pattern of the passenger includes whether it is expected that the passenger will appear on time, the passenger's preferred route, and the common behaviors of the passenger during specific time periods. The Hidden Markov Model formula is: P(A|θ) = , Where A is the observed behavior sequence, θ is the model parameter, the model parameter is obtained through maximum likelihood estimation, and Q is the state of the behavior sequence.

4. The optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction according to claim 3, wherein The multi-modal large model optimization module includes a multi-data input processing sub-module, a dynamic traffic analysis sub-module, and a personalized optimization sub-module. The multi-data input processing sub-module parses the instructions and requirements input by the passenger through natural language processing technology, and combines image and text data to confirm the pick-up and drop-off location. The dynamic traffic analysis sub-module dynamically optimizes the pick-up and drop-off point selection and driving route by analyzing real-time traffic data, weather information, and road conditions. The personalized optimization sub-module generates a personalized pick-up and drop-off plan for each passenger, and the personalized pick-up and drop-off plan needs to combine the specific needs of the passenger and the needs of other members in the group.

5. The optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction according to claim 4, wherein The specific implementation process of the multi-data input processing sub-module is as follows: The multi-modal fusion CLIP model is used to process the input of the passenger's voice, image, and text. The output of the multi-modal fusion CLIP model is the location information of the passenger, and the location information of the passenger is used to fuse real-time image data to confirm the pick-up and drop-off point. The specific implementation process of the dynamic traffic analysis sub-module is as follows: The deep Q-network model and the proximal operator Proximal are used to analyze real-time traffic information and road conditions to dynamically adjust the pick-up and drop-off route of the vehicle. The specific implementation process of the personalized optimization sub-module is as follows: The genetic algorithm with multi-objective optimization is used to generate an optimization plan for the personalized needs of each passenger. The personalized needs include preferred pick-up and drop-off times and specific route preferences.

6. The optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction according to claim 5, wherein, The group pick-up and drop-off point priority ranking and dynamic adjustment module includes a pick-up and drop-off point ranking sub-module and a dynamic adjustment sub-module. The pick-up and drop-off point ranking sub-module ranks the pick-up and drop-off points according to the needs of each passenger. The needs at least include time limits and preferred locations. The dynamic adjustment sub-module is used to automatically adjust the pick-up and drop-off order and route when the needs or locations of a certain passenger in the group change. The change in location at least includes being late or temporarily changing the pick-up and drop-off point.

7. An optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction according to claim 6, characterized in that The specific implementation process of ranking the pick-up and drop-off points in the pick-up and drop-off point ranking sub-module is: The priority ranking uses the multi-criteria decision-making TOPSIS algorithm. Based on the passengers' needs, it generates the priority ranking of pick-up and drop-off points, calculates the positive ideal solution and negative ideal solution of each pick-up and drop-off point, and generates a heat map according to the positive and negative ideal solutions. According to the distribution of each pick-up and drop-off point in the heat map, those with dense distribution have high scores, and those with sparse distribution have low scores. The best pick-up and drop-off order is determined according to the scores. The passengers' needs include time limits and location preferences; The specific implementation process of the dynamic adjustment sub-module is as follows: The deep Q-network (DQN) reinforcement learning algorithm is used to adjust the pick-up and drop-off order in real time. When the passengers' needs or locations change, the pick-up and drop-off order and route are recalculated according to the current environment, and the plan is adjusted according to the latest traffic and passenger data. The current environment includes passengers being late and temporarily changing pick-up and drop-off points.

8. The optimized system for group pick-up and drop-off of autonomous vehicles based on group behavior prediction according to claim 7, wherein The real-time group interaction and feedback module includes a real-time notification sub-module and a feedback and adjustment sub-module; The real-time notification sub-module is used to notify each passenger of the update of the pick-up and drop-off point, the estimated arrival time, and the vehicle location through the application APP, and allows passengers to view the pick-up and drop-off order and route in the group in real time; The feedback and adjustment sub-module is used for passengers to feedback or adjust pick-up and drop-off needs in real time through the application APP, and dynamically adjust the group pick-up and drop-off plan according to the feedback or adjusted pick-up and drop-off needs.

Images