Vehicle transportation trip identification system based on path planning
By applying deep reinforcement learning and long-term memory network models in the vehicle transportation management system, the problems of unreasonable path planning and inaccurate time recognition in traditional systems are solved, and more efficient and accurate transportation management is achieved.
Patent Information
- Application Number
- CN202411937893.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-04-18
AI Technical Summary
The traditional vehicle transportation management system has problems such as unreasonable path planning and inaccurate identification of transportation trips, resulting in high logistics and transportation efficiency and cost.
The path planning model based on deep reinforcement learning and the time recognition model of long and short-term memory networks is adopted, and combined with data acquisition, preprocessing and model fusion technology, intelligent path planning and time recognition are achieved.
Through intelligent path planning and trip identification, the system can generate the optimal transportation path and accurately identify the trips, which improves transportation efficiency and accuracy and reduces transportation costs.
Smart Images
Figure CN119850074A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of vehicle transportation management, and in particular to a vehicle transportation trip identification system based on path planning. Background Art
[0002] With the rapid development of the logistics industry and the improvement of the level of intelligence, the efficiency and accuracy of vehicle transportation have become key competitive factors. Traditional vehicle transportation management systems often rely on manual scheduling and experience judgment, and have many limitations, such as unreasonable route planning and inaccurate identification of transportation trips. These problems directly affect the efficiency and cost of logistics transportation.
[0003] In terms of path planning, traditional methods are often based on simple rules or heuristic algorithms, which are difficult to fully consider the complex factors in the actual transportation process, such as road conditions, traffic congestion, weather changes, etc. As a result, the planned path is often not the optimal path, increasing transportation time and cost. At the same time, traditional path planning methods lack real-time and dynamic adjustment capabilities, and cannot be adjusted in time according to the actual situation during transportation, further reducing transportation efficiency.
[0004] In terms of trip identification, traditional methods mainly rely on manual recording and identification, which is not only labor-intensive but also prone to errors. Especially when transportation tasks are frequent and there are many vehicles, the accuracy and efficiency of trip identification are even more difficult to guarantee. In addition, traditional trip identification methods lack intelligent and automated means, and cannot achieve automatic identification and tracking of transportation trips, which brings great difficulties to logistics management and decision-making. Therefore, there is an urgent need for a system that can comprehensively consider multiple factors and realize intelligent path planning and trip identification to improve the efficiency and accuracy of logistics transportation. Summary of the invention
[0005] The purpose of the present invention is to provide a vehicle transport trip identification system based on path planning to solve the problems raised in the above background technology.
[0006] To achieve the above object, the present invention provides the following technical solution: a vehicle transport trip identification system based on path planning, the system comprising:
[0007] A data collection unit, used to collect multi-source data related to vehicle transportation, including vehicle GPS data, vehicle sensor data, transportation task data, and road condition data;
[0008] The data preprocessing unit cleans, converts the format, synchronizes the time and processes outliers of the collected multi-source data, and then extracts features to form a feature data set;
[0009] A path planning model building unit is used to build a path planning model based on a deep reinforcement learning (DRL) algorithm. The model is trained using a preprocessed feature data set to generate an optimal transportation path for the vehicle.
[0010] The trip identification model building unit is used to build a trip identification model based on the long short-term memory network LSTM algorithm. The model is trained using the historical transportation data and feature data set of the vehicle to identify the transportation trips of the vehicle;
[0011] Model fusion unit, which defines the model fusion algorithm, fuses the output results of the path planning model and the trip identification model, and optimizes the transportation path and trip identification results;
[0012] The result output unit is used to output the final transportation route planning and trip identification results, including route information, trip number, starting point and end point information.
[0013] Preferably, the vehicle GPS data includes vehicle location, speed and driving direction; the on-board sensor data includes fuel consumption, load and vehicle status; the transportation mission data includes cargo type, transportation volume and destination; and the road condition data includes road conditions, traffic congestion and weather conditions.
[0014] Preferably, the path planning model adopts a deep reinforcement learning algorithm, specifically a Deep Q-Network structure, whose network structure includes:
[0015] Input layer: receives preprocessed feature data sets, including vehicle status, road conditions, and transportation task information;
[0016] Convolution layer: multi-layer convolution operation is used to extract spatial features; the convolution kernel size is set to 3x3 and the step size is set to 1;
[0017] Fully connected layer: Flatten the feature map output by the convolutional layer into a one-dimensional vector and make decisions through multiple layers of fully connected layers;
[0018] Output layer: Output the Q value of each possible action, and select the action with the largest Q value as the next driving direction of the vehicle;
[0019] Reward function: Based on the vehicle’s driving distance, time consumption, and transportation task completion, a reward function is designed for model training.
[0020] Preferably, the training step of the path planning model includes:
[0021] F101: Initialize the parameters of the DQN model, including convolution kernel weights, fully connected layer weights, learning rate, and discount factor;
[0022] F102: Build an experience playback pool to store vehicle driving experience;
[0023] F103: In each training step, the vehicle selects an action based on the current state, executes the action and observes the new state and reward;
[0024] F104: Store the current state, selected action, reward, and new state into the experience replay pool;
[0025] F105: Randomly extract a batch of experiences from the experience replay pool for model training;
[0026] F106: Use the extracted experience to update model parameters through back propagation algorithm and gradient descent method;
[0027] F107: Repeat steps F103 to F106 until the model converges or reaches the preset number of training steps;
[0028] F108: Save the model parameters and obtain the trained path planning model.
[0029] Preferably, the trip identification model is constructed using a long short-term memory network LSTM algorithm, and its network structure includes:
[0030] Input layer: receives preprocessed vehicle historical transportation data and feature data sets, including vehicle GPS data, on-board sensor data, and transportation task data;
[0031] LSTM layer: uses multiple layers of LSTM units to capture long-term dependencies in time series data;
[0032] Fully connected layer: The output of the LSTM layer is passed through multiple layers of fully connected layers for decision making;
[0033] Output layer: Output trip recognition results, including trip number, starting point and end point information.
[0034] Preferably, the training step of the trip identification model includes:
[0035] F201: Divide the preprocessed vehicle historical transportation data and feature data set into a training set and a validation set;
[0036] F202: Initialize the parameters of the LSTM model, including LSTM unit weights, fully connected layer weights, and learning rate;
[0037] F203: Use the training set to iteratively train the LSTM model. In each iteration, the model output is calculated by forward propagation, and the loss between the predicted value and the true value is calculated according to the loss function; the loss function uses the cross entropy loss function to calculate the difference between the predicted times and the true times;
[0038] F204: Update model parameters based on loss value through back propagation algorithm and gradient descent method;
[0039] F205: Repeat steps F203 to F204 until the loss converges or the preset number of iterations is reached;
[0040] F206: Use the validation set to validate the trained LSTM model and evaluate the recognition accuracy of the model;
[0041] F207: When the model performance reaches the preset standard, the model parameters are saved to obtain the trained trip recognition model.
[0042] Preferably, the specific implementation steps of the model fusion algorithm defined in the model fusion unit include:
[0043] Step 1: Obtain the optimal transportation path output by the path planning model and the trip identification result output by the trip identification model;
[0044] Step 2: Match the optimal transportation path with the trip identification result to ensure that each trip corresponds to a valid transportation path;
[0045] Step 3: If there is a mismatch between the path and the trip, make adjustments based on the confidence of the path planning model and the accuracy of the trip identification model to ensure the consistency of the path and the trip;
[0046] Step 4: Optimize the fused transportation path and trip identification results, consider transportation cost, time and safety factors, and generate the final transportation plan.
[0047] Preferably, the transport route planning and trip identification results output by the result output unit include:
[0048] Path information: including the vehicle’s route, intersections passed, and estimated arrival time;
[0049] Trip number: Assign a unique trip number to each transport task for easy tracking and management;
[0050] Starting point and end point information: clearly define the starting and end points of each transport task;
[0051] Additional information: including the type and quantity of the transported goods and special considerations during transportation.
[0052] Preferably, it also includes a report generating unit, and the report generated by the report generating unit includes: vehicle number, transport trip, starting fence, end fence, passing fence, departure time, arrival time, total time, average speed and transport mileage.
[0053] Preferably, it also includes an abnormality detection unit for detecting abnormal behaviors during vehicle transportation, including deviation from a preset route, long-term stay and illegal entry into a restricted area.
[0054] Compared with the prior art, the present invention has the following beneficial effects:
[0055] The system uses a deep reinforcement learning (DRL) algorithm to build a path planning model, which can comprehensively consider multiple complex factors such as road conditions, traffic congestion, weather changes, etc. to generate the optimal transportation path for the vehicle. This dynamic, real-time path planning method greatly shortens transportation time and improves transportation efficiency. Through the construction of a trip identification model, the system can automatically and accurately identify the vehicle's transportation trips, avoiding the tediousness and errors of manual recording and identification, and further improving the smoothness and efficiency of the transportation process. The planning of the optimal path reduces unnecessary driving distance and time, thereby reducing fuel consumption and vehicle maintenance costs. The automation of trip identification reduces labor costs, while improving the accuracy of identification and avoiding additional transportation costs caused by trip identification errors.
[0056] The system integrates multiple modules such as data collection, preprocessing, model building and result output, and realizes the intelligent identification and planning of vehicle transportation routes and trips, greatly improving the automation of logistics management. Through the model fusion algorithm, the system optimizes and integrates the results of route planning and trip identification, providing more comprehensive and accurate information support for logistics decision-making. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] Figure 1 This is a working principle diagram of a vehicle transport trip identification system based on path planning according to the present invention;
[0058] Figure 2 This is a training flowchart for the path planning model based on the Deep Q-Network algorithm;
[0059] Figure 3 This is the training flowchart of the trip recognition model based on the LSTM algorithm. DETAILED DESCRIPTION
[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0061] See also Figure 1-3The present invention provides a technical solution: a vehicle transport trip identification system based on path planning, the system comprising:
[0062] Data collection unit: This unit is responsible for collecting vehicle transportation-related data from multiple sources, including vehicle GPS data to obtain real-time location information of the vehicle; vehicle sensor data to monitor the operating status of the vehicle; transportation task data, including the starting and ending points of the transportation, cargo information, etc.; and road condition data, covering road conditions, traffic congestion, construction information and weather forecasts, etc. These data are transmitted to the data preprocessing unit via wireless network or wired mode.
[0063] Data preprocessing unit: After receiving the raw data, the data preprocessing unit performs data cleaning operations, removes invalid or erroneous data, performs format conversion to ensure data consistency, and performs time synchronization processing on each data source. Subsequently, outlier detection and processing are performed on the cleaned data to ensure data accuracy. After completing preprocessing, key features are extracted from the data to form a feature data set, which provides a basis for subsequent model training.
[0064] Path planning model building unit: Use deep reinforcement learning (DRL) algorithms, such as deep Q network (DQN) or policy gradient method, to build a path planning model. Use the preprocessed feature data set as input, and train the model to learn how to choose the optimal path by simulating the vehicle's driving process under different road conditions. After the model training is completed, it can generate the optimal transportation path for the vehicle based on the real-time road conditions and transportation tasks.
[0065] Trip recognition model construction unit: The trip recognition model is constructed using the long short-term memory network (LSTM) algorithm. The model uses the historical transportation data and feature data sets of vehicles as training samples to learn the temporal characteristics of vehicle transportation trips, thereby realizing automatic recognition of transportation trips. Through the memory capacity of LSTM, the model can effectively handle the temporal dependency between transportation trips and improve recognition accuracy.
[0066] Model fusion unit: responsible for effectively integrating the output results of the path planning model and the trip identification model. Design a model fusion algorithm, such as weighted average, decision tree or neural network fusion, to comprehensively consider the results of path planning and trip identification, and jointly optimize the transportation path and trips. Through iterative calculation, continuously adjust the results of path planning and trip identification until the optimal solution is reached.
[0067] Result output unit: Output the integrated optimal transportation route planning and trip identification results in a visual or textual form. The output results include detailed route information, such as waypoints, estimated travel time; trip number, which is used to uniquely identify each transportation task; and start and end point information, which is convenient for logistics management personnel to schedule and monitor.
[0068] The present invention will be further described below in conjunction with Examples 1 to 4:
[0069] Embodiment 1:
[0070] This embodiment uses a deep reinforcement learning algorithm in the path planning model, and specifically implements the Deep Q-Network (DQN) structure to achieve intelligent planning of vehicle transportation paths. The network structure and training steps of the model are as follows:
[0071] The network structure of the model includes:
[0072] Input layer: The input layer receives the preprocessed feature data set, which integrates the vehicle status (such as speed, location, remaining fuel, etc.), road conditions (such as road conditions, traffic congestion, weather conditions, etc.) and transportation task information (such as starting and ending points, cargo type, urgency, etc.). The data is input in the form of a multidimensional array, where each dimension represents a different feature.
[0073] Convolutional layer: The convolutional layer uses multiple layers of convolution operations to extract spatial features, especially the spatial relationship between road conditions and vehicle status. The convolution kernel size is set to 3x3 and the stride is set to 1 to ensure that detailed local features can be captured. The ReLU activation function is used to increase nonlinearity, and the size of the feature map is reduced through pooling layers (such as maximum pooling) to improve computational efficiency.
[0074] Fully connected layer: The feature map output by the convolutional layer is flattened into a one-dimensional vector and then input into a multi-layer fully connected layer. The fully connected layer performs linear transformation through the weight matrix and bias term, and then combines the activation function (such as ReLU or Tanh) to make nonlinear decisions. The multi-layer fully connected structure helps to learn complex decision logic.
[0075] Output layer: The output layer contains neurons equal to the number of possible actions that the vehicle can take, and each neuron outputs a Q value. The Q value represents the expected cumulative reward for taking an action in the current state. The action with the largest Q value is selected as the next direction of the vehicle to achieve the goal of maximizing long-term rewards.
[0076] Reward function: The reward function is designed based on factors such as the vehicle's driving distance, time consumption, completion of the transportation task, and compliance with traffic rules. For example, a positive reward is given for reaching the destination, and a negative reward is given for driving too long or violating traffic rules. The reward function is the key to model learning, which guides the model to learn how to make the best decision.
[0077] The model training steps include:
[0078] F101: Initialize model parameters: Initialize the parameters of the DQN model, including convolution kernel weights, fully connected layer weights, learning rate (controls the step size of model updates), and discount factor (determines the importance of future rewards).
[0079] F102: Build an experience replay pool: Create an experience replay pool to store the vehicle's driving experience, including status, actions, rewards, and new status.
[0080] The size of the experience replay pool is limited. When the storage is full, new experience will replace the old experience.
[0081] F103: Select and execute actions: In each training step, the vehicle selects an action based on the current state (provided by the input layer). Action selection can adopt an ε-greedy strategy, that is, random actions are selected with a certain probability to explore new states, and the action with the largest Q value is selected in the rest of the cases to exploit known information.
[0082] F104: Store Experience: After performing an action, observe the new state and the reward obtained. Store the current state, the selected action, the reward, and the new state as one piece of experience in the experience replay pool.
[0083] F105: Extract experience for training: Randomly extract a batch of experience (called a batch) from the experience replay pool for model training. Random extraction helps break the temporal correlation between experiences and improve the generalization ability of the model.
[0084] F106: Update model parameters: Use the extracted experience to calculate the gradient of the loss function through the back propagation algorithm. Use gradient descent method (such as Adam optimizer) to update model parameters to minimize the loss function. The loss function is usually designed based on the error between the Q value and the target Q value.
[0085] F107: Repeat training: Repeat steps F103 to F106 until the model converges (i.e. the loss function no longer changes significantly) or the preset number of training steps is reached. During the training process, the learning rate can be gradually reduced to improve the stability of the model.
[0086] F108: Save model: After the model training is completed, save the model parameters (such as convolution kernel weights, fully connected layer weights, etc.) to obtain a trained path planning model that can generate the optimal transportation path for the vehicle based on the real-time input feature data set.
[0087] Embodiment 2:
[0088] This embodiment adopts the long short-term memory network (LSTM) algorithm in the trip identification model to realize intelligent identification of vehicle transport trips.
[0089] The network structure of the trip identification model includes:
[0090] Input layer: The input layer receives preprocessed historical vehicle transportation data and feature data sets, which integrate vehicle GPS data (such as location, speed, timestamp, etc.), vehicle sensor data (such as acceleration, fuel consumption, etc.) and transportation task data (such as starting and ending points, cargo information, transportation date, etc.). The data is input in the form of a time series, where each time step corresponds to a set of features.
[0091] LSTM layer: The LSTM layer uses multiple layers of LSTM units. These units have memory capabilities and can capture long-term dependencies in time series data. They are particularly suitable for processing data with time series characteristics such as vehicle transportation trips. Each LSTM layer contains multiple LSTM units. Each unit has an input gate, a forget gate, and an output gate. These gating mechanisms are used to control and regulate the flow of information. The multi-layer LSTM structure helps to learn more complex time series features.
[0092] Fully connected layer: The output of the LSTM layer is input into multiple layers of fully connected layers, which are linearly transformed through weight matrices and bias terms, and combined with activation functions (such as ReLU or Softmax) for nonlinear decision making. The role of the fully connected layer is to map the time series features extracted by the LSTM layer to the output space of pass recognition.
[0093] Output layer: The output layer contains neurons equal to the number of possible passes. Each neuron outputs a probability value, indicating the possibility that the current input belongs to a certain pass. The Softmax activation function is used to ensure that the sum of the output probabilities is 1. According to the probability distribution of the output layer, the pass with the largest probability is selected as the recognition result, and the pass number, starting point, and end point information are output.
[0094] The model training steps include:
[0095] F201. Data division: Divide the preprocessed vehicle historical transportation data and feature data set into a training set and a validation set. The training set is used for model training, and the validation set is used for model verification and performance evaluation.
[0096] F202. Initialize model parameters: Initialize the parameters of the LSTM model, including LSTM unit weights, fully connected layer weights, learning rate (controlling the step size of model updates), etc. Parameter initialization can be done by random initialization or pre-training initialization.
[0097] F203. Iterative training: The LSTM model is iteratively trained using the training set. In each iteration, the output of the model, i.e., the predicted number of passes, is calculated through forward propagation. The loss between the predicted value and the true value is calculated according to the loss function to measure the performance of the model in the current iteration. The loss function uses the cross entropy loss function, which calculates the difference between the predicted number of passes and the true number of passes. The cross entropy loss function is suitable for multi-classification problems and can measure the difference between the probability distribution predicted by the model and the true distribution.
[0098] F204. Parameter update: Based on the loss value, the gradient of the loss function to the model parameters is calculated through the back propagation algorithm. The gradient descent method (such as SGD, Adam and other optimizers) is used to update the model parameters to minimize the loss function. Parameter update is a key step in model learning. Through continuous iterative updates, the model gradually learns the mapping relationship from input to output.
[0099] F205. Repeat training: Repeat steps F203 to F204 until the loss converges (i.e. the loss value no longer changes significantly) or the preset number of iterations is reached. During the training process, the loss value and the performance on the validation set can be monitored to evaluate the training progress and generalization ability of the model.
[0100] F206. Model Validation: Use the validation set to validate the trained LSTM model and evaluate the recognition accuracy of the model. The performance on the validation set reflects the model's ability to generalize to unseen data.
[0101] F207. Save model: When the model performance reaches the preset standard (such as the recognition accuracy exceeds a certain threshold), save the model parameters (such as LSTM unit weights, fully connected layer weights, etc.). Get the trained trip recognition model, which can intelligently identify the vehicle's transportation trips based on the real-time input vehicle data and feature data sets.
[0102] Embodiment 3:
[0103] This embodiment designs a detailed model fusion algorithm in the model fusion unit to effectively integrate the results of the path planning model and the trip identification model, and optimizes them to generate the final transportation plan. The following is a detailed description of the specific implementation steps of the algorithm, as well as the specific content of the transportation path planning and trip identification results output by the result output unit:
[0104] Step 1. Get model output: Get the optimal transport path calculated for a specific vehicle or transport task from the path planning model. These paths are dynamically generated based on factors such as real-time road conditions, vehicle status, and transport task requirements. At the same time, get the trip identification results of the vehicle's historical transport data from the trip identification model, including key information such as trip number, starting point, and end point.
[0105] Step 2. Matching routes with trips: Match the optimal transport route output by the route planning model with the trip identification results output by the trip identification model. This is usually achieved by comparing the starting point and end point of the route with the starting point and end point in the trip identification results. Ensure that each trip corresponds to a valid transport route, and the starting point and end point of the route match the starting point and end point of the trip.
[0106] Step 3. Inconsistency adjustment: During the matching process, if there is a mismatch between the path and the trip (for example, the end point of the path is inconsistent with the end point of the trip), it is necessary to evaluate based on the confidence of the path planning model and the accuracy of the trip identification model. The model results with higher confidence and accuracy will be considered more reliable, and the results of the other model will be adjusted accordingly to ensure the consistency of the path and the trip.
[0107] For example, if the accuracy of the trip identification model is high, but the confidence of the path planning model for a certain path is low, it may be necessary to replan the path based on the trip identification results.
[0108] Step 4. Optimize the transportation plan: Comprehensively consider the integrated transportation path and trip identification results, including multiple factors such as transportation cost, time and safety. Use optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.) to iteratively optimize the transportation plan to find the transportation plan with the lowest cost, shortest time and highest safety under all constraints.
[0109] The optimization process may involve fine-tuning the route, rearranging trips, or redistributing cargo.
[0110] The output content of the result output unit includes:
[0111] ①Path information:
[0112] Driving route: Details the complete driving route of the vehicle from the starting point to the end point, including the roads, intersections and possible turns passed.
[0113] Intersections: Important intersections that the vehicle will pass through during driving are specially marked to facilitate the driver's advance preparation and navigation.
[0114] Estimated time of arrival: Based on the current road conditions, vehicle speed and transportation task requirements, the estimated time it will take for the vehicle to arrive at the destination.
[0115] ②Trip number:
[0116] Each transport task is assigned a unique trip number, which remains unchanged throughout the entire transport process, making it easier to track and manage the transport tasks.
[0117] ③Starting point and end point information:
[0118] Starting point: Clearly indicate the starting point of the transportation task, including specific address, coordinates or landmarks.
[0119] Destination information: It also clearly indicates the destination of the transport task to ensure that the driver knows the destination clearly.
[0120] ④Additional information:
[0121] Cargo type: Describe the type and nature of the transported cargo, such as dangerous goods, fragile goods, refrigerated goods, etc., so that the driver can take appropriate transportation measures.
[0122] Cargo quantity: clearly specify the quantity of cargo to be transported, so that the driver can arrange vehicle space and load appropriately.
[0123] Special precautions: List the matters that require special attention during transportation, such as keeping the goods stable, avoiding bumps, and regularly checking the status of the goods, etc., to ensure the safety and smoothness of the transportation process.
[0124] Embodiment 4:
[0125] The vehicle transport trip identification system of the present invention further includes a report generation unit and an abnormality detection unit. The report generated by the report generation unit includes:
[0126] Vehicle Number: A number that uniquely identifies the transport vehicle.
[0127] Transport trip: The unique identifier of each transport task.
[0128] Start fence and end fence: represent the starting and ending locations of a transport task, respectively, and are usually defined using geo-fencing technology.
[0129] Pass-through fences: Important locations or areas that a vehicle passes through during transportation are also marked using geo-fencing technology.
[0130] Departure time, arrival time: record the specific time when the vehicle departs from the starting point and arrives at the destination.
[0131] Total time: Calculate the total time it takes for a vehicle to depart and arrive.
[0132] Average speed: The average speed of the vehicle calculated based on the total driving distance and total time.
[0133] Transport mileage: record the total mileage travelled by the vehicle during transport.
[0134] The report generation unit extracts relevant data from the database on a regular basis (such as daily, weekly or monthly). Using the preset report templates and formatting tools, the data is organized into an easy-to-read report format. The report can be distributed to relevant managers or customers via email, cloud storage or printing.
[0135] The abnormal behavior detection of the anomaly detection unit includes:
[0136] Deviation from preset route: By comparing the actual driving route of the vehicle with the preset transportation route, it is detected whether the vehicle has deviated from the preset track. If it deviates, an abnormal alarm is triggered.
[0137] Long-term stay: Monitors the time a vehicle stays at a specific location. If it exceeds a preset threshold (such as 30 minutes or 1 hour), it is considered an abnormal stay and triggers an alarm.
[0138] Illegal entry into prohibited areas: Based on geo-fencing technology, prohibited areas are defined. If a vehicle enters these areas, an abnormal alarm is immediately triggered.
[0139] The anomaly detection unit receives and processes vehicle GPS data and transportation task data in real time. Using algorithms and rule engines, it analyzes and judges vehicle behavior in real time. When abnormal behavior is detected, the alarm mechanism is immediately triggered to inform relevant managers or drivers through SMS, email or system notification. At the same time, the anomaly detection unit can also record abnormal events to provide data support for subsequent analysis and improvement.
[0140] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0141] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A vehicle transport trip identification system based on path planning, characterized in that: The system comprises: A data collection unit, used to collect multi-source data related to vehicle transportation, including vehicle GPS data, vehicle sensor data, transportation task data, and road condition data; The data preprocessing unit cleans, converts the format, synchronizes the time and processes outliers of the collected multi-source data, and then extracts features to form a feature data set; A path planning model building unit is used to build a path planning model based on a deep reinforcement learning (DRL) algorithm. The model is trained using a preprocessed feature data set to generate an optimal transportation path for the vehicle. The trip identification model building unit is used to build a trip identification model based on the long short-term memory network LSTM algorithm. The model is trained using the historical transportation data and feature data set of the vehicle to identify the transportation trips of the vehicle; Model fusion unit, which defines the model fusion algorithm, fuses the output results of the path planning model and the trip identification model, and optimizes the transportation path and trip identification results; The result output unit is used to output the final transportation route planning and trip identification results, including route information, trip number, starting point and end point information.
2. A vehicle transport trip identification system based on path planning according to claim 1, characterized in that: The vehicle GPS data includes vehicle location, speed and driving direction; the on-board sensor data includes fuel consumption, load and vehicle status; the transportation mission data includes cargo type, transportation volume and destination; the road condition data includes road conditions, traffic congestion and weather conditions.
3. A vehicle transport trip identification system based on path planning according to claim 1, characterized in that: The path planning model adopts a deep reinforcement learning algorithm, specifically a Deep Q-Network structure, whose network structure includes: Input layer: receives preprocessed feature data sets, including vehicle status, road conditions, and transportation task information; Convolution layer: multi-layer convolution operation is used to extract spatial features; the convolution kernel size is set to 3x3 and the step size is set to 1; Fully connected layer: Flattens the feature map output by the convolutional layer into a one-dimensional vector and makes decisions through multiple layers of fully connected layers; Output layer: Output the Q value of each possible action, and select the action with the largest Q value as the next driving direction of the vehicle; Reward function: Design a reward function based on the vehicle’s driving distance, time consumption, and transportation task completion status.
4. A vehicle transport trip identification system based on path planning according to claim 3, characterized in that: The training steps of the path planning model include: F101: Initialize the parameters of the DQN model, including convolution kernel weights, fully connected layer weights, learning rate, and discount factor; F102: Build an experience playback pool to store vehicle driving experience; F103: In each training step, the vehicle selects an action based on the current state, executes the action and observes the new state and reward; F104: Store the current state, selected action, reward, and new state into the experience replay pool; F105: Randomly extract a batch of experiences from the experience replay pool for model training; F106: Use the extracted experience to update model parameters through back propagation algorithm and gradient descent method; F107: Repeat steps F103 to F106 until the model converges or reaches the preset number of training steps; F108: Save the model parameters and obtain the trained path planning model.
5. The vehicle transport trip identification system based on path planning according to claim 1 is characterized in that: The trip recognition model is constructed using the long short-term memory network LSTM algorithm, and its network structure includes: Input layer: receives preprocessed vehicle historical transportation data and feature data sets, including vehicle GPS data, on-board sensor data, and transportation task data; LSTM layer: uses multiple layers of LSTM units to capture long-term dependencies in time series data; Fully connected layer: The output of the LSTM layer is passed through multiple layers of fully connected layers for decision making; Output layer: Output trip recognition results, including trip number, starting point and end point information.
6. A vehicle transport trip identification system based on path planning according to claim 5, characterized in that: The training steps of the trip recognition model include: F201: Divide the preprocessed vehicle historical transportation data and feature data set into a training set and a validation set; F202: Initialize the parameters of the LSTM model, including LSTM unit weights, fully connected layer weights, and learning rate; F203: Use the training set to iteratively train the LSTM model. In each iteration, the model output is calculated by forward propagation, and the loss between the predicted value and the true value is calculated according to the loss function; the loss function uses the cross entropy loss function to calculate the difference between the predicted times and the true times; F204: Update model parameters based on loss value through back propagation algorithm and gradient descent method; F205: Repeat steps F203 to F204 until the loss converges or the preset number of iterations is reached; F206: Use the validation set to validate the trained LSTM model and evaluate the recognition accuracy of the model; F207: When the model performance reaches the preset standard, the model parameters are saved to obtain the trained trip recognition model.
7. The vehicle transport trip identification system based on path planning according to claim 1 is characterized in that: The specific implementation steps of the model fusion algorithm defined in the model fusion unit include: Step 1: Obtain the optimal transportation path output by the path planning model and the trip identification result output by the trip identification model; Step 2: Match the optimal transportation route with the trip identification result; Step 3: If there is a mismatch between the path and the trip, make adjustments based on the confidence of the path planning model and the accuracy of the trip identification model; Step 4: Optimize the fused transportation path and trip identification results, consider transportation cost, time and safety factors, and generate the final transportation plan.
8. The vehicle transport trip identification system based on path planning according to claim 1 is characterized in that: The transport route planning and trip identification results output by the result output unit include: Path information: including the vehicle’s route, intersections passed, and estimated arrival time; Trip number: Assign a unique trip number to each transport task for easy tracking and management; Starting point and end point information: clearly define the starting and end points of each transport task; Additional information: including the type and quantity of the transported goods and special considerations during transportation.
9. The vehicle transport trip identification system based on path planning according to claim 1, characterized in that: It also includes a report generating unit, and the report generated by the report generating unit includes: vehicle number, transport trip number, starting fence, end fence, passing fence, departure time, arrival time, total time, average speed and transport mileage.
10. The vehicle transport trip identification system based on path planning according to claim 1, characterized in that: It also includes an anomaly detection unit for detecting abnormal behavior during vehicle transportation, including deviation from the preset route, prolonged stay and illegal entry into prohibited areas.