Smart park vehicle scheduling system based on AI identification

Abstract

The invention discloses a smart park vehicle scheduling system based on AI identification, and relates to the technical field of smart park vehicle scheduling, and the system comprises a vehicle feature extraction module, a space-time prediction module, a path planning module, an energy consumption calculation module, a scheduling verification module and an instruction execution module. The vehicle feature extraction module is used for collecting vehicle visual features through a multispectral camera, obtaining electronic tag data in combination with an RFID reader-writer, and generating unique fingerprint features of a vehicle by adopting a cross-modal contrast learning algorithm; the space-time prediction module is used for receiving vehicle fingerprint features and historical trajectory data, extracting spatial topological features through a three-dimensional convolutional neural network, and predicting vehicle demand distribution in each region in combination with a time attention mechanism; the path planning module is used for generating an initial path set according to the demand distribution data, and calling the energy consumption parameters output by the energy consumption calculation module to perform path cost evaluation; the energy consumption calculation module is used for obtaining the state of the battery through a vehicle-mounted sensor.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent park vehicle scheduling, and particularly relates to an intelligent park vehicle scheduling system based on AI recognition. BACKGROUND

[0002] Intelligent park vehicle scheduling technology is to use advanced information technology, communication technology and intelligent algorithms to intelligently manage and schedule the operation of vehicles in the park. The technology aims to improve transportation efficiency, reduce energy consumption, optimize resource allocation, and ensure traffic safety and smoothness. Therefore, how to use advanced technical means to improve the intelligent level and safety of intelligent park vehicle scheduling has become one of the problems to be solved at present.

[0003] In the field of intelligent park vehicle scheduling, existing methods often rely on historical data or static models to predict traffic demand in the future period of time, which cannot adapt to dynamic changes in traffic conditions in real time, and traditional path planning usually only considers the shortest distance or the least time, ignoring the influence of energy consumption cost and charging facilities, resulting in electric vehicles stopping halfway due to insufficient power. At the same time, the fixed energy consumption setting in the existing technology cannot reflect the changes of speed, load, slope and other factors in the actual driving process, which is easy to cause inaccurate estimation and increase the risk of power failure. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides an intelligent park vehicle scheduling system based on AI recognition to solve the problem that existing methods often rely on historical data or static models to predict traffic demand in the future period of time, which cannot adapt to dynamic changes in traffic conditions in real time, and traditional path planning usually only considers the shortest distance or the least time, ignoring the influence of energy consumption cost and charging facilities, resulting in electric vehicles stopping halfway due to insufficient power.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides an intelligent park vehicle scheduling system based on AI recognition, which comprises:

[0008] a vehicle feature extraction module, a space-time prediction module, a path planning module, an energy consumption calculation module, a scheduling verification module and an instruction execution module;

[0009] The vehicle feature extraction module is used to collect vehicle visual features through a multispectral camera, obtain electronic tag data in combination with an RFID reader, and generate a vehicle unique fingerprint feature by using a cross-modal contrast learning algorithm.

[0010] The space-time prediction module is configured to receive vehicle fingerprint features and historical trajectory data, extract spatial topology features through a three-dimensional convolutional neural network, and predict vehicle demand distribution in each region by combining a time attention mechanism;

[0011] The path planning module is configured to generate an initial path set according to demand distribution data, and perform path cost evaluation by calling energy consumption parameters output by the energy consumption calculation module;

[0012] The energy consumption calculation module is configured to obtain battery state, load and slope data through a vehicle-mounted sensor, and calculate a road section energy consumption coefficient based on a nonlinear dynamics model;

[0013] The scheduling verification module is configured to import a scheduling scheme output by the path planning module into a digital twin environment to detect conflicts, and output a feasible scheme that passes Monte Carlo testing;

[0014] The instruction execution module is configured to convert the verified scheduling scheme into control instructions to drive the vehicle to perform a transportation task and feed back real-time state data to the vehicle feature extraction module.

[0015] As a preferred scheme of the AI recognition-based intelligent park vehicle scheduling system, the vehicle feature extraction module performs the following steps:

[0016] Visible light images, infrared thermal images and polarized light images of the vehicle are collected by a multispectral camera, and an improved MobileNetV3 network is used to extract features from the three-channel images to output a visual feature vector f v ;

[0017] An encrypted signal of a vehicle-mounted electronic tag is read by an RFID reader and writer, and a 128-bit AES decryption algorithm is used to analyze the tag ID to generate a radio frequency feature vector f r ;

[0018] A cross-modal contrast learning model is constructed, and f v and f r are input into a double-flow neural network to optimize feature space alignment through a loss function , and the expression is:

[0019]

[0020] where s(.) is a cosine similarity function, τ is a temperature coefficient, N is the number of batch samples, and k is a summation index variable;

[0021] When a change in the appearance of the vehicle is detected, an online learning mechanism is triggered, and an elastic weight solidification algorithm is used to update network parameters to generate dynamic fingerprint features F.

[0022] As a preferred scheme of the intelligent park vehicle scheduling system based on AI recognition, the space-time prediction module performs the following steps:

[0023] The vehicle fingerprint feature F and the historical trajectory data are received, and a four-dimensional input tensor X is constructed T×H×W×C , wherein T represents the time step, HxW represents the park grid division, and C represents the feature channel number;

[0024] A three-dimensional convolution kernel K is adopted 5×5×3×64 The spatial feature of the input tensor X is extracted, and the spatial feature map S is output T×46×46×64 .

[0025] The spatial feature map S is input into the time attention mechanism layer, and the query matrix Q, the key matrix K and the value matrix V are calculated respectively, and the expression is:

[0026] Q=S·W q ,K=S·W k ,V=S·W v .

[0027] Wherein, W q , W k , W v are trainable weight matrices;

[0028] The space-time correlation weight between each time step is calculated by the multi-head attention mechanism, and the demand distribution matrix D is output H ×W×2 .

[0029] As a preferred scheme of the intelligent park vehicle scheduling system based on AI recognition, the path planning module performs the following steps:

[0030] According to the demand distribution matrix D output by the space-time prediction module, the Dijkstra algorithm is adopted to generate an initial path set P={p1,...,p n}, wherein each path p k contains a node sequence;

[0031] The road section energy consumption coefficient output by the energy consumption calculation module is called to calculate the total cost of the path p k , and the expression is:

[0032]

[0033] Wherein, d(.) represents the Euclidean distance, λ1 is the energy consumption weight coefficient, λ2 is the charging point penalty term, I 充电站 (v i ) is an indicator function;

[0034] Select minC(pk ) and the total energy consumption E total <SOC的路径作为候选方案,其中,SOC表示当前电池剩余电量。

[0035] As a preferred solution of the AI-based smart park vehicle dispatching system of the present invention, the energy consumption calculation module performs the following steps:

[0036] Obtain real-time data from vehicle sensors via the CAN bus, including remaining battery capacity textSOC, load mass m, and slope angle θ;

[0037] The energy consumption per unit distance traveled is calculated based on the vehicle dynamics model. The expression is:

[0038] e=0.01v 2 +0.2msinθ+0.005m;

[0039] Where v is the current vehicle speed;

[0040] For any two adjacent nodes v on the path i →v j , integrally calculate the total energy consumption of the road section, the expression is:

[0041]

[0042] Among them, E i→j represents the total energy consumed when traveling from node i to node j, e(t) is the instantaneous energy consumption rate at time t, and t i and t j They represent the time when the vehicle arrives at node i and node j respectively.

[0043] As a preferred solution of the AI-based smart park vehicle dispatching system of the present invention, the dispatching verification module performs the following steps:

[0044] Build a digital twin simulation environment, import the park's 3D model and physical parameters, including the road friction coefficient μ and the signal light cycle T light ;

[0045] 2000 sets of simulated abnormal events were injected for Monte Carlo testing, including:

[0046] Type A is sudden braking of the vehicle;

[0047] Type B is communication delay

[0048] The success rate of the statistical scheduling plan is expressed as:

[0049]

[0050] Among them, if and only if R≥99.9%, the scheduling plan is determined to be a feasible plan and output to the instruction execution module.

[0051] As a preferred solution of the AI-based smart park vehicle dispatching system of the present invention, the instruction execution module performs the following steps:

[0052] Encoding the feasible path solution output by the scheduling verification module into a control instruction set, including a speed instruction and a charging instruction;

[0053] The charging instruction triggers a charging request when the remaining battery capacity SOC is less than 0.3;

[0054] Send control instructions to the vehicle terminal through the 5G C-V2X communication protocol;

[0055] Receive the positioning data (x, y)∈WGS84 coordinate system fed back by the vehicle terminal in real time and update it to the spatiotemporal prediction module.

[0056] As a preferred solution of the AI-based smart park vehicle dispatching system of the present invention, the system further includes a dynamic weight adjustment unit for performing the following steps:

[0057] Monitor the operating status of each module, including feature extraction delay t f , prediction error ε p , path deviation δ;

[0058] Calculate the adaptive weight parameter, the expression is:

[0059]

[0060] Among them, t f,max , t f,min is the historical maximum / minimum value of feature extraction delay, ε p,max , ε p,min is the historical extreme value of the forecast error;

[0061] The updated weights λ′1 and λ′2 are fed back to the path planning module to replace the original fixed weights and used to achieve dynamic optimization of the path cost function.

[0062] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the smart park vehicle dispatching system based on AI recognition as described in the first aspect of the present invention.

[0063] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the smart park vehicle dispatching system based on AI recognition as described in the first aspect of the present invention.

[0064] The beneficial effects of the present invention are as follows: visible light, infrared thermal imaging and polarized light images are collected by a multispectral camera, and the improved MobileNetV3 network is used to extract features from the three-channel images and output visual feature vectors, thereby realizing high-precision, multi-source fusion vehicle identity recognition capabilities. The method breaks through the limitations of traditional single-sensor recognition methods in complex environments, ensuring that vehicle identities can still be accurately identified under conditions such as occlusion, night, and bad weather. The initial path set is generated by using the Dijkstra algorithm based on the demand distribution matrix output by the spatiotemporal prediction module; the road section energy consumption coefficient provided by the energy consumption calculation module is called, combined with the Euclidean distance, energy consumption weight coefficient and charging point penalty item, to construct a path cost function, evaluate the total cost of each path, and realize an intelligent path selection mechanism based on real-time demand and energy consumption cost, solving the problem that traditional path planning ignores energy consumption factors and lacks dynamic adjustment capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0066] Figure 1 This is a flowchart of the smart park vehicle dispatching system based on AI recognition in Example 1.

[0067] Figure 2 This is a flowchart of vehicle feature extraction in Example 1.

[0068] Figure 3 Schematic diagram of the coupling between path planning and energy consumption in Example 1. DETAILED DESCRIPTION

[0069] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0070] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the "one embodiment" or "an embodiment" referred to herein means a specific feature, structure, or characteristic under discussion. Thus, "an embodiment" appearing in various places in the specification are not necessarily all referring to the same embodiment, nor are they mutually exclusive of other embodiments.

[0072] Embodiment, with reference to Figure 1 , Figure 2 and Figure 3 , is an embodiment of the present application, which provides an AI recognition-based smart park vehicle scheduling system, comprising:

[0073] a vehicle feature extraction module, a space-time prediction module, a path planning module, an energy consumption calculation module, a scheduling verification module and an instruction execution module;

[0074] The vehicle feature extraction module is configured to collect vehicle visual features through a multi-spectral camera, obtain electronic tag data in combination with an RFID reader, and generate a vehicle unique fingerprint feature by using a cross-modal contrast learning algorithm.

[0075] Further, the multi-spectral camera is configured to collect a visible light image, an infrared thermal image and a polarized light image of the vehicle, and an improved MobileNetV3 network is used to extract features of the three-channel images and output a visual feature vector f v .

[0076] The RFID reader is configured to read an encrypted signal of a vehicle-mounted electronic tag, and a 128-bit AES decryption algorithm is used to analyze the tag ID and generate a radio frequency feature vector f r .

[0077] A cross-modal contrast learning model is constructed, and f v and f r are input into a double-flow neural network, and a loss function is used to optimize the alignment of the feature space, and the expression is as follows:

[0078]

[0079] where s(.) is a cosine similarity function, τ is a temperature coefficient, N is the number of batch samples, and k is a summation index variable.

[0080] When a change in the appearance of the vehicle is detected, an online learning mechanism is triggered, an elastic weight solidification algorithm is used to update the network parameters, and a dynamic fingerprint feature F is generated.

[0081] The multi-spectral camera is deployed at the entrances and exits of the park and key road sections, and each frame of image contains three dimensions of RGB visible light channel, infrared thermal imaging channel and polarized light channel, which are respectively used to capture color information, heat source distribution and surface material texture.

[0082] The improved MobileNetV3 network introduces the SE-block attention module to enhance the focus on the vehicle subject in complex backgrounds;

[0083] The RFID reader uses the UHF frequency band, with a communication range of up to 15 meters. It supports concurrent reading of multiple tags, ensuring stable identification in dense traffic. The AES decryption algorithm is completed using a hardware acceleration module, with a decryption delay of less than 1ms.

[0084] The cross-modal contrastive learning model uses a hard negative sample mining strategy during the training phase to enhance the ability to distinguish between different vehicles;

[0085] The elastic weight consolidation algorithm is based on the ElasticWeightConsolidation method to prevent the forgetting of old task knowledge during online learning.

[0086] It should be noted that this module generates unique vehicle fingerprint features by fusing multispectral images and radio frequency tag data, combined with a cross-modal contrastive learning algorithm, effectively improving the accuracy and robustness of vehicle identification in complex environments. It adopts an improved MobileNetV3 network with the SE attention mechanism to enhance the anti-interference ability against background interference. The introduction of an online learning mechanism enables the system to have dynamic update capabilities and adapt to changes in vehicle appearance, providing a stable and reliable identity recognition foundation for subsequent path prediction and scheduling, and significantly improving the recognition accuracy and response speed of the entire scheduling system.

[0087] The spatiotemporal prediction module receives vehicle fingerprint features and historical trajectory data, extracts spatial topological features through a three-dimensional convolutional neural network, and combines the temporal attention mechanism to predict the distribution of vehicle demand in each region.

[0088] Furthermore, the vehicle fingerprint feature F and historical trajectory data are received to construct a four-dimensional input tensor X∈R T ×H×W×C , where T represents the time step, H×W represents the park grid division, and C represents the number of feature channels;

[0089] Use three-dimensional convolution kernel K∈R 5×5×3×64 Extract spatial features from the input tensor X and output spatial feature map S∈R T×46×46×64 ;

[0090] The spatial feature map S is input into the temporal attention mechanism layer to calculate the query matrix Q, key matrix K and value matrix V respectively. The expressions are:

[0091] Q=S·W q ,K=S·W k ,V=S·W v ;

[0092] Among them, W q , W k , W v is the trainable weight matrix;

[0093] The spatiotemporal correlation weights between each time step are calculated through the multi-head attention mechanism, and the demand distribution moment D∈R is output. H ×W×2 ;

[0094] Historical trajectory data is uploaded by the vehicle terminal and contains timestamps, GPS coordinates, speed and direction information, which is mapped to the campus grid system after preprocessing;

[0095] The 3D convolution operation slides simultaneously in the time, row, and column dimensions to effectively extract the spatial correlation of road intersections and curve structures;

[0096] The temporal attention mechanism adopts a multi-head design, setting up 8 parallel attention heads to focus on traffic patterns in different time periods;

[0097] The multi-head attention results are linearly weighted fused to generate a demand distribution matrix with spatiotemporal joint perception capabilities, which is used to guide the path planning module to prioritize high-demand areas.

[0098] It should be noted that this module is based on a combination of a three-dimensional convolutional neural network and a temporal attention mechanism to achieve high-precision spatiotemporal prediction of traffic demand within the park. Through spatial topology modeling and temporal evolution analysis of historical trajectory data, it can perceive regional traffic hotspots in the future in advance and provide forward-looking guidance for path planning. The design of the multi-head attention mechanism enhances the model's ability to capture traffic patterns in different time periods, making the scheduling strategy more flexible and efficient, and improving the system's adaptability and resource utilization in a dynamic environment.

[0099] The path planning module is used to generate an initial path set based on the demand distribution data and use the energy consumption parameters output by the energy consumption calculation module to evaluate the path cost;

[0100] Furthermore, according to the demand distribution matrix D output by the spatiotemporal prediction module, the Dijkstra algorithm is used to generate the initial path set P = {p1, ..., p n}, where each path p k Contains a sequence of nodes;

[0101] Call the energy consumption calculation module to output the energy consumption coefficient of the road section and calculate the path p k The total cost is expressed as:

[0102]

[0103] Where d(.) represents the Euclidean distance, λ1 is the energy consumption weight coefficient, λ2 is the charging point penalty term, and I 充电站 (v i ) is the indicator function;

[0104] Choose the one that satisfies minC(p k ) and the total energy consumption E total <SOC的路径作为候选方案,其中,SOC表示当前电池剩余电量;

[0105] The Dijkstra algorithm is implemented using priority queue optimization, which improves path search efficiency by more than 30%;

[0106] The road section energy consumption coefficient comes from the real-time feedback of the energy consumption calculation module and is updated every 5 seconds;

[0107] In the path cost function, the Euclidean distance term ensures the shortest path length, the energy consumption term balances energy consumption, and the charging point penalty term avoids frequent detours to charging stations;

[0108] In the path selection logic, if the energy consumption of the current path exceeds the remaining battery capacity, the path will be automatically eliminated and a new suboptimal path will be found;

[0109] The path set P is dynamically generated before each scheduling to ensure that it adapts to changes in the park's traffic conditions.

[0110] It should be noted that this module constructs a multi-objective cost function by comprehensively considering path length, energy consumption cost and the impact of charging points, thus achieving the transition from "shortest path" to "optimal path". The Dijkstra algorithm combined with the real-time energy consumption feedback mechanism ensures the efficiency and accuracy of path search. The battery power constraint judgment mechanism is embedded in the path selection logic, which effectively avoids the problem of mid-route stagnation caused by insufficient energy. The design of dynamically generated path sets improves the system's response ability to changes in the traffic status of the park, thereby achieving more intelligent and efficient path scheduling.

[0111] Energy consumption calculation module, which is used to obtain battery status, load and slope data through on-board sensors and calculate the energy consumption coefficient of the road section based on a nonlinear dynamic model;

[0112] Furthermore, the real-time data of the vehicle sensors is obtained through the CAN bus, including the remaining battery capacity textSOC, load mass m, and slope angle θ;

[0113] The energy consumption per unit distance traveled is calculated based on the vehicle dynamics model. The expression is:

[0114] e=0.01v 2 +0.2msinθ+0.005m;

[0115] Where v is the current vehicle speed;

[0116] For any two adjacent nodes v on the path i →v j , integrally calculate the total energy consumption of the road section, the expression is:

[0117]

[0118] Among them, E i→j represents the total energy consumed when traveling from node i to node j, e(t) is the instantaneous energy consumption rate at time t, and t i and t j Represent the time when the vehicle arrives at node i and node j respectively;

[0119] The CAN bus interface complies with the ISO 11898 standard, with a sampling frequency of 10Hz to ensure real-time data.

[0120] The remaining battery power is obtained through the BMS system with an accuracy error of less than ±2%;

[0121] The slope angle is estimated by the IMU inertial measurement unit combined with high-precision maps;

[0122] All coefficients in the dynamic model are obtained through real vehicle experimental calibration, reflecting the energy consumption pattern of a typical electric logistics vehicle;

[0123] The integral calculation is implemented using the trapezoidal method approximation and is completed in an embedded manner in the vehicle controller;

[0124] Road section energy consumption E i→j The final result is normalized to watt-hours per kilometer for unified comparison by the path planning module;

[0125] It should be noted that this module is based on real-time data collected by on-board sensors and combines nonlinear dynamic models to perform energy consumption modeling, achieving a refined estimate of actual energy consumption under different road conditions. The total energy consumption of the road section is calculated in an integral manner, enabling the path planning module to obtain more accurate energy consumption parameters. The normalized energy consumption indicators facilitate horizontal comparison between paths and help optimize energy utilization efficiency. The introduction of this module effectively prevents the risk of power outages caused by inaccurate energy consumption estimates, and provides strong support for the long-term operation of electric vehicles in park scenarios.

[0126] The scheduling verification module is used to import the scheduling plan output by the path planning module into the digital twin environment for conflict detection and output a feasible plan that passes the Monte Carlo test;

[0127] Furthermore, a digital twin simulation environment is constructed, and the park 3D model and physical parameters are imported, including the road friction coefficient μ and the signal light cycle T light ;

[0128] 2000 sets of simulated abnormal events were injected for Monte Carlo testing, including:

[0129] Type A is when the vehicle brakes suddenly;

[0130] Type B is communication delay

[0131] The success rate of the statistical scheduling plan is expressed as:

[0132]

[0133] Among them, if and only if R ≥ 99.9%, the scheduling plan is determined to be a feasible plan and output to the instruction execution module;

[0134] The digital twin simulation platform is developed based on the Unity3D engine, integrates the physics engine PhysX, and supports millimeter-level precision collision detection;

[0135] The park's 3D model was reconstructed through LiDAR scanning, accurately restoring road width, turning radius, and traffic light location geometry.

[0136] In the Monte Carlo test, Type A events simulate vehicle braking behavior under sudden road conditions, and Type B events simulate control command delays caused by 5G network fluctuations.

[0137] The test process automatically records the number of path conflicts, communication packet loss rate, and response delay indicators, and generates a visual report for manual review;

[0138] It should be noted that this module uses a digital twin simulation platform to perform conflict detection and Monte Carlo testing on the scheduling plan, which can detect potential problems and risk points before actual deployment. The injection mechanism of simulated abnormal events improves the system's ability to respond to emergencies. The success rate threshold setting (≥99.9%) ensures the safety and reliability of the scheduling plan. The visual report output mechanism provides an intuitive basis for manual review, enhancing the system's operability and fault tolerance in complex campus environments.

[0139] The instruction execution module is used to convert the verified scheduling plan into control instructions, drive the vehicle to perform the transportation task and feed back real-time status data to the vehicle feature extraction module;

[0140] Furthermore, the feasible path solution output by the scheduling verification module is encoded into a control instruction set, including speed instructions and charging instructions;

[0141] The charging instruction triggers a charging request when the remaining battery capacity SOC is less than 0.3;

[0142] Send control instructions to the vehicle terminal through the 5G C-V2X communication protocol;

[0143] Receive the positioning data (x, y)∈WGS84 coordinate system fed back by the vehicle terminal in real time and update it to the spatiotemporal prediction module;

[0144] The system further includes a dynamic weight adjustment unit, configured to perform the following steps:

[0145] Monitor the operating status of each module, including feature extraction delay t f , prediction error ε p , path deviation δ;

[0146] Calculate the adaptive weight parameter, the expression is:

[0147]

[0148] Among them, t f,max , t f,min is the historical maximum / minimum value of feature extraction delay, ε p,max , ε p,min is the historical extreme value of the forecast error;

[0149] The updated weights λ′1 and λ′2 are fed back to the path planning module to replace the original fixed weights and used to achieve dynamic optimization of the path cost function.

[0150] The control command format is a JSON structure, which includes the path node sequence, target speed, charging suggestion and obstacle avoidance action fields;

[0151] The 5G C-V2X communication protocol adopts the NR-V2X direct communication mode, with end-to-end latency less than 20ms and packet loss rate less than 0.1%;

[0152] The vehicle terminal has a built-in Beidou / GPS dual-mode positioning chip, with positioning accuracy reaching sub-meter level;

[0153] Feedback data is reported every 1 second, including latitude and longitude, speed, heading angle, SOC and operation status code information;

[0154] The state data is filtered and used to update the vehicle fingerprint feature library, forming a closed-loop learning mechanism;

[0155] It should be noted that this module realizes closed-loop execution from scheduling decisions to actual control instructions, ensuring efficient response and precise control of the scheduling system. The 5G C-V2X communication protocol ensures low latency and high reliability of command transmission, supports multi-vehicle collaborative operations in the park, and the real-time feedback mechanism of positioning data forms a complete closed-loop learning chain, continuously optimizing vehicle feature recognition and path prediction capabilities. The dynamic weight adjustment unit further enhances the system's adaptive adjustment capabilities in the face of performance fluctuations, making the entire scheduling system more flexible and stable.

[0156] This embodiment also provides a computer device, which is suitable for the case of a smart park vehicle dispatching system based on AI recognition, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the smart park vehicle dispatching system based on AI recognition proposed in the above embodiment.

[0157] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0158] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the smart park vehicle dispatching system based on AI recognition as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.

[0159] In summary, the present invention collects visible light, infrared thermal imaging and polarized light images through a multispectral camera, and uses an improved MobileNetV3 network to extract features from the three-channel images and output visual feature vectors, thereby realizing high-precision, multi-source fusion vehicle identity recognition capabilities. The method breaks through the limitations of traditional single-sensor recognition methods in complex environments, ensuring that vehicle identities can still be accurately identified under conditions such as occlusion, night, and bad weather. The initial path set is generated using the Dijkstra algorithm based on the demand distribution matrix output by the spatiotemporal prediction module; the road section energy consumption coefficient provided by the energy consumption calculation module is called, and the Euclidean distance, energy consumption weight coefficient and charging point penalty item are combined to construct a path cost function to evaluate the total cost of each path, thereby realizing an intelligent path selection mechanism based on real-time demand and energy consumption cost, and solving the problem that traditional path planning ignores energy consumption factors and lacks dynamic adjustment capabilities.

[0160] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A smart park vehicle dispatching system based on AI recognition, characterized by: include: Vehicle feature extraction module, spatiotemporal prediction module, path planning module, energy consumption calculation module, scheduling verification module and instruction execution module; The vehicle feature extraction module is used to collect vehicle visual features through a multispectral camera, obtain electronic tag data in combination with an RFID reader, and generate a unique fingerprint feature of the vehicle using a cross-modal contrastive learning algorithm; The spatiotemporal prediction module is used to receive vehicle fingerprint features and historical trajectory data, extract spatial topological features through a three-dimensional convolutional neural network, and combine the temporal attention mechanism to predict the distribution of vehicle demand in each area; The path planning module is used to generate an initial path set based on the demand distribution data and to perform path cost evaluation by calling the energy consumption parameters output by the energy consumption calculation module; The energy consumption calculation module is used to obtain battery status, load and slope data through vehicle-mounted sensors and calculate the energy consumption coefficient of the road section based on a nonlinear dynamic model; The scheduling verification module is used to import the scheduling plan output by the path planning module into the digital twin environment for conflict detection and output a feasible plan that passes the Monte Carlo test; The instruction execution module is used to convert the verified scheduling plan into control instructions, drive the vehicle to perform the transportation task and feed back real-time status data to the vehicle feature extraction module.

2. The AI-based smart park vehicle dispatching system according to claim 1, characterized in that: The vehicle feature extraction module performs the following steps: The multispectral camera collects the visible light image, infrared thermal image and polarized light image of the vehicle, and uses the improved MobileNetV3 network to extract features from the three-channel image and output the visual feature vector f v ; The encrypted signal of the vehicle electronic tag is read by the RFID reader, and the tag ID is parsed using the 128-bit AES decryption algorithm to generate the radio frequency feature vector f r ; Construct a cross-modal contrastive learning model and transform f v and f r Input two-stream neural network, through the loss function Optimize feature space alignment, the expression is: Where s(.) is the cosine similarity function, τ is the temperature coefficient, N is the number of batch samples, and k is the sum index variable; When a change in the vehicle's appearance is detected, the online learning mechanism is triggered, and the elastic weight solidification algorithm is used to update the network parameters to generate a dynamic fingerprint feature F.

3. The AI-based smart park vehicle dispatching system according to claim 2, characterized in that: The spatiotemporal prediction module performs the following steps: Receive vehicle fingerprint features F and historical trajectory data, and construct a four-dimensional input tensor X∈R T×H×W×C , where T represents the time step, H×W represents the park grid division, and C represents the number of feature channels; Use three-dimensional convolution kernel K∈R 5×5×3×64 Extract spatial features from the input tensor X and output spatial feature map S∈R T ×46×46×64 ; The spatial feature map S is input into the temporal attention mechanism layer to calculate the query matrix Q, key matrix K and value matrix V respectively. The expressions are: Q=S·W q ,K=S·W k ,V=S·W v ; Among them, W q , W k , W v is the trainable weight matrix; The spatiotemporal correlation weights between each time step are calculated through the multi-head attention mechanism, and the demand distribution moment D∈R is output. H×W×2 .

4. The AI-based smart park vehicle dispatching system according to claim 3, characterized in that: The path planning module performs the following steps: According to the demand distribution matrix D output by the spatiotemporal prediction module, the Dijkstra algorithm is used to generate the initial path set P = {p1,...,p n }, where each path p k Contains a sequence of nodes; Call the energy consumption calculation module to output the energy consumption coefficient of the road section and calculate the path p k The total cost is expressed as: Where d(.) represents the Euclidean distance, λ1 is the energy consumption weight coefficient, λ2 is the charging point penalty term, and I 充电站 (v i ) is the indicator function; Select the path that satisfies minC(p k ) and has a total energy consumption E total < the SOC as the candidate solution, where SOC represents the remaining battery charge of the current battery.

5. The AI-based smart park vehicle dispatching system according to claim 4, characterized in that: The energy consumption calculation module performs the following steps: Obtain real-time data from vehicle sensors via the CAN bus, including remaining battery capacity textSOC, load mass m, and slope angle θ; The energy consumption per unit distance traveled is calculated based on the vehicle dynamics model. The expression is: e=0.01v 2 +0.2msinθ+0.005m; Where v is the current vehicle speed; For any two adjacent nodes v on the path i →v j , integrally calculate the total energy consumption of the road section, the expression is: Among them, E i→j represents the total energy consumed when traveling from node i to node j, e(t) is the instantaneous energy consumption rate at time t, and t i and t j They represent the time when the vehicle arrives at node i and node j respectively.

6. The AI-based smart park vehicle dispatching system according to claim 5, characterized in that: The scheduling verification module performs the following steps: Build a digital twin simulation environment, import the park's 3D model and physical parameters, including the road friction coefficient μ and the signal light cycle T ligh t ; 2000 sets of simulated abnormal events were injected for Monte Carlo testing, including: Type A is sudden braking of the vehicle; Type B is communication delay The success rate of the statistical scheduling plan is expressed as: Among them, if and only if R≥99.9%, the scheduling plan is determined to be a feasible plan and output to the instruction execution module.

7. The AI-based smart park vehicle dispatching system according to claim 6, characterized in that: The instruction execution module performs the following steps: Encoding the feasible path solution output by the scheduling verification module into a control instruction set, including a speed instruction and a charging instruction; The charging instruction triggers a charging request when the remaining battery capacity SOC is less than 0.3; Send control instructions to the vehicle terminal through the 5G C-V2X communication protocol; Receive the positioning data (x, y)∈WGS84 coordinate system fed back by the vehicle terminal in real time and update it to the spatiotemporal prediction module.

8. The AI-based smart park vehicle dispatching system according to claim 7, characterized in that: The system further includes a dynamic weight adjustment unit, configured to perform the following steps: Monitor the operating status of each module, including feature extraction delay t f , prediction error ε p , path deviation δ; Calculate the adaptive weight parameter, the expression is: Among them, t f,max , t f,min is the historical maximum / minimum value of feature extraction delay, ε p,max , ε p,min is the historical extreme value of the forecast error; The updated weights λ′1 and λ′2 are fed back to the path planning module to replace the original fixed weights and used to achieve dynamic optimization of the path cost function.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the smart park vehicle dispatching system based on AI recognition are implemented in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the smart park vehicle dispatching system based on AI recognition according to any one of claims 1 to 8 are implemented.

Images