Ski field automatic parking control method based on Internet of Things sensing and transport vehicle

Abstract

The embodiment of the invention discloses a ski field automatic parking control method based on Internet of Things sensing and a transport vehicle, and belongs to the technical field of Internet of Things control. According to the embodiment of the invention, the surrounding environment of the ski field transport vehicle can be comprehensively mastered by acquiring various kinds of Internet of Things sensing information, and the fusion of the ultrasonic sensing information, the geomagnetic sensing information, the visual sensing information and the laser radar sensing information can accurately reflect the vehicle state. An initial vehicle environment state vector is determined by using an intelligent driving control processing algorithm, and an accurate basis is provided for subsequent decision making. The determination of the parking path planning label helps to evaluate the risk of intersection with the control strategy path in advance, and determines whether to issue the automatic parking control strategy according to the label, thereby improving the parking safety and efficiency, avoiding the unnecessary collision risk, optimizing the resource utilization of the parking lot, and meeting the complex and variable parking demands of the ski field.

Description

Technical Field

[0001] The embodiments of the present invention belong to the field of Internet of Things control technology, and specifically relate to an automatic parking control method and a transport vehicle for a ski resort based on Internet of Things sensing. Background Art

[0002] In ski resort areas, automatic parking of transport vehicles faces many challenges. Traditional parking systems often cannot fully obtain information about the vehicle's surrounding environment, and it is difficult to accurately depict the complex ski resort environment by relying on a single type of sensor. For example, snow may cover part of the ground markings or affect the normal operation of the sensor, and a single sensor cannot accurately determine the vehicle's position and surrounding obstacles. Due to the lack of accurate initial environmental state judgment of the vehicle, the risk of intersection with the control strategy path cannot be effectively estimated during parking path planning, which can easily lead to low parking efficiency, increased collision risk, and ineffective use of parking lot resources. Therefore, a technical solution that can accurately analyze the vehicle status and reasonably plan the parking path is needed to solve the problem of automatic parking of ski resort transport vehicles. Summary of the invention

[0003] The embodiments of the present invention provide a ski resort automatic parking control method and a transport vehicle based on IoT sensing, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] An embodiment of the present invention provides a ski resort automatic parking control method based on IoT sensing, which is applied to an IoT sensing control system. The method comprises: acquiring first IoT sensing information of a transport vehicle to be controlled, wherein the transport vehicle to be controlled is a transport vehicle in a ski resort area; the first IoT sensing information comprises ultrasonic sensing information, geomagnetic sensing information, visual sensing information and laser radar sensing information; The first IoT sensor information of the transport vehicle to be controlled is processed using an intelligent driving control processing algorithm to determine an initial vehicle environment state vector of the transport vehicle to be controlled; the initial vehicle environment state vector of the transport vehicle to be controlled is processed using the intelligent driving control processing algorithm to determine a parking path planning label of the transport vehicle to be controlled, wherein the parking path planning label of the transport vehicle to be controlled represents a probability that the transport vehicle to be controlled has a path intersection with an automatic parking control strategy during an automatic parking process; and based on the parking path planning label of the transport vehicle to be controlled, it is determined whether to issue the automatic parking control strategy to the transport vehicle to be controlled.

[0005] An embodiment of the present invention provides a transport vehicle, which is communicatively connected to an Internet of Things sensor control system; the transport vehicle is used to receive an automatic parking control strategy issued by the Internet of Things sensor control system; the Internet of Things sensor control system includes at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the above method.

[0006] An embodiment of the present invention provides an Internet of Things sensor control system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above method.

[0007] An embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0008] In an embodiment of the present invention, by acquiring a variety of IoT sensor information, the surrounding environment of the ski resort transport vehicle can be fully understood, and the fusion of ultrasonic sensor information, geomagnetic sensor information, visual sensor information and lidar sensor information can accurately reflect the vehicle status. The initial vehicle environment state vector is determined using the intelligent driving control processing algorithm to provide an accurate basis for subsequent decision-making. The determination of the parking path planning label helps to assess the risk of intersection with the control strategy path in advance. The decision whether to issue the automatic parking control strategy based on the label can improve parking safety and efficiency, avoid unnecessary collision risks, optimize parking lot resource utilization, and adapt to the complex and changeable parking needs of the ski resort. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A flowchart of an automatic parking control method for a ski resort based on IoT sensing provided by an embodiment of the present invention.

[0010] Figure 2 A schematic diagram of the structure of an IoT sensor control system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0011] 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 part of the embodiments of the present invention, not all of them. Based on the embodiments in 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 embodiments of the present invention.

[0012] The terms "first", "second", etc. in the embodiments of the present invention are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the embodiments of the present invention can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more. In addition, in the embodiments of the present invention, "and / or" means at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0013] Figure 1 A ski resort automatic parking control method based on Internet of Things sensing is shown, which is applied to an Internet of Things sensing control system. The method includes the following steps 110-140.

[0014] Step 110: The IoT sensor control system obtains first IoT sensor information of the transport vehicle to be controlled, where the transport vehicle to be controlled is a transport vehicle in a ski resort area; the first IoT sensor information includes ultrasonic sensor information, geomagnetic sensor information, visual sensor information and lidar sensor information.

[0015] Step 120: The IoT sensor control system processes the first IoT sensor information of the transport vehicle to be controlled using an intelligent driving control processing algorithm to determine the initial vehicle environment state vector of the transport vehicle to be controlled.

[0016] Step 130: The IoT sensor control system uses the intelligent driving control processing algorithm to process the initial vehicle environment state vector of the transport vehicle to be controlled to determine the parking path planning label of the transport vehicle to be controlled. The parking path planning label of the transport vehicle to be controlled represents the probability that the transport vehicle to be controlled has a path intersection with the automatic parking control strategy during the automatic parking process.

[0017] Step 140: The IoT sensor control system determines whether to send the automatic parking control strategy to the transport vehicle to be controlled based on the parking path planning tag of the transport vehicle to be controlled.

[0018] In step 140, the parking path planning tag of the transport vehicle to be controlled includes a first parking path planning tag and a second parking path planning tag, wherein the first parking path planning tag represents the probability that the transport vehicle to be controlled matches the path intersection on the basis that the automatic parking control strategy is issued to the transport vehicle to be controlled, and the second parking path planning tag represents the probability that the transport vehicle to be controlled matches the path intersection on the basis that the automatic parking control strategy is not issued to the transport vehicle to be controlled. On this basis, the determining whether to issue the automatic parking control strategy to the transport vehicle to be controlled based on the parking path planning label of the transport vehicle to be controlled further includes: if it meets the first judgment mechanism and the second judgment mechanism, the automatic parking control strategy is issued to the transport vehicle to be controlled, and the first judgment mechanism is met when the weight of the first parking path planning label is greater than the weight of the second parking path planning label, and the second judgment mechanism is met when the weight difference between the first parking path planning label and the second parking path planning label is greater than the set weight; if it does not meet at least one of the first judgment mechanism and the second judgment mechanism, it is determined not to issue the automatic parking control strategy to the transport vehicle to be controlled.

[0019] In the embodiment of the present invention, the IoT sensor control system plays a key role in the automatic parking control of the ski resort transport vehicle. First, in step 110, the IoT sensor control system starts to obtain the first IoT sensor information of the transport vehicle to be controlled. The transport vehicle to be controlled in the embodiment of the present invention is a transport vehicle in the ski resort area, and the first IoT sensor information includes ultrasonic sensor information, geomagnetic sensor information, visual sensor information and laser radar sensor information.

[0020] For ultrasonic sensing information, in a ski resort environment, ultrasonic sensors can emit ultrasonic signals at a higher frequency, such as 20-50 times per second. These signals will be reflected back after encountering objects around the transport vehicle (such as other vehicles, snow piles, parking lot boundary facilities, etc.). By accurately measuring the time difference between the transmitted signal and the received reflected signal, the distance between the transport vehicle and the surrounding objects can be determined. For example, in a standard ski resort parking lot scenario, the effective detection range of the ultrasonic sensor can reach 0.2-5 meters, and its accuracy can reach the centimeter level, such as ±3 centimeters, which provides an important basis for the system to accurately judge the close-range environment around the transport vehicle.

[0021] Geomagnetic sensing information is obtained through geomagnetic sensors. Geomagnetic sensors are installed at specific locations on the transport vehicle or in corresponding areas of the parking lot. Due to the existence of the earth's magnetic field, when the transport vehicle is at different locations in the parking lot, the surrounding magnetic field will change to varying degrees due to the ferromagnetic materials of the vehicle itself (such as the engine, frame, etc.). For example, the geomagnetic sensor can detect changes in magnetic field strength between 0.3-0.6 Gauss, and accurately determine the position and posture of the transport vehicle based on the characteristics of geomagnetic changes. This method is not affected by environmental factors such as light and snow, and provides the system with stable position reference information.

[0022] Visual sensing information mainly relies on cameras installed around the transport vehicle or surveillance cameras in the parking lot. The camera acquires images at a certain frame rate (e.g., 30 frames per second). In a ski resort environment, the camera can capture the visual scene around the transport vehicle, including the layout of the parking lot (such as the division of parking spaces, the direction of the aisle), the position and movement status of other vehicles, and possible obstacles (such as temporarily placed ski equipment, snow piles, etc.). Through computer vision algorithms, the system can identify the shape, color, texture and other features of various objects from these images, and then determine their type and relative position. For example, the visual system can accurately identify other vehicles within 5-20 meters of the transport vehicle and determine their direction of travel.

[0023] LiDAR sensing information is provided by LiDAR equipment. LiDAR constructs a three-dimensional point cloud map of the surrounding environment by emitting laser beams and receiving reflected light. In complex environments such as ski resorts, LiDAR can quickly scan large areas around transport vehicles. For example, it can scan an area with a radius of 10-20 meters in 0.1-0.5 seconds, with a resolution of 1°-3°. By analyzing the time and intensity of reflected light, LiDAR can accurately depict the contours, heights and other information of surrounding objects, whether it is snow-covered objects or complex parking facility structures, it can be clearly presented, providing the system with comprehensive and accurate environmental data.

[0024] Then, in step 120, the IoT sensor control system uses the intelligent driving control processing algorithm to process the acquired first IoT sensor information to determine the initial vehicle environment state vector of the transport vehicle to be controlled. The intelligent driving control processing algorithm is an AI algorithm model that comprehensively considers the characteristics and interrelationships of various sensor information. For example, the algorithm constructs a multi-dimensional vector to describe the initial environmental state of the transport vehicle based on the distance data in the ultrasonic sensor information, the position and posture data in the geomagnetic sensor information, the object recognition and position data in the visual sensor information, and the environmental contour data in the lidar sensor information. This vector may include the distance vector between the transport vehicle and the surrounding objects, the coordinate vector of the transport vehicle in the parking lot, the type and distribution vector of the surrounding objects, etc. For example, in a specific scenario, the initial vehicle environment state vector can be represented as an array containing 10-20 elements, each element representing a different environmental state parameter. In this way, the system can fully and accurately grasp the current environmental state of the transport vehicle and provide detailed data support for subsequent parking path planning.

[0025] Then, in step 130, the IoT sensor control system continues to process the initial vehicle environment state vector using the intelligent driving control processing algorithm to determine the parking path planning label of the transport vehicle to be controlled. This parking path planning label represents the probability that the transport vehicle to be controlled has a path intersection with the automatic parking control strategy during the automatic parking process. During the processing, the intelligent driving control processing algorithm analyzes the various parameters in the initial vehicle environment state vector to predict the possible driving trajectory of the transport vehicle under different parking operations. For example, based on the current position of the transport vehicle, the parking situation of surrounding vehicles, and the layout of the parking lot, the algorithm simulates multiple possible parking paths. In this process, the algorithm takes into account the impact of various factors on the path, such as the width of the parking lot channel (for example, the standard channel width is 3-5 meters), the spacing of other vehicles (generally requiring a safe spacing of 0.5-1.5 meters), and the turning radius of the transport vehicle itself (the turning radius of different types of transport vehicles may be between 3-8 meters). Through a comprehensive analysis of these factors, the algorithm calculates the intersection of each possible path with the preset path in the automatic parking control strategy and converts it into a parking path planning label. The probability value in the parking path planning label can be represented by a value between 0 and 1. For example, 0.3 means that there is a 30% probability that there is a path intersection.

[0026] Finally, in step 140, the IoT sensor control system determines whether to issue an automatic parking control strategy to the transport vehicle to be controlled based on the parking path planning tag of the transport vehicle to be controlled. The parking path planning tag of the transport vehicle to be controlled in the embodiment of the present invention includes a first parking path planning tag and a second parking path planning tag. The first parking path planning tag represents the probability that the transport vehicle to be controlled matches the intersection of the path on the basis that the automatic parking control strategy is issued to the transport vehicle to be controlled; the second parking path planning tag represents the probability that the transport vehicle to be controlled matches the intersection of the path on the basis that the automatic parking control strategy is not issued to the transport vehicle to be controlled. In the specific discrimination process, if the first discrimination mechanism and the second discrimination mechanism are met, the automatic parking control strategy is issued to the transport vehicle to be controlled. Meeting the first discrimination mechanism means that the weight of the first parking path planning tag is greater than the weight of the second parking path planning tag. The weight in the embodiment of the present invention can be a value set according to the actual situation, for example, the weight of the first parking path planning tag is 0.6, and the weight of the second parking path planning tag is 0.3. The second discrimination mechanism is met when the weight difference between the first parking path planning label and the second parking path planning label is greater than the set weight. For example, the weight is set to 0.2. When the weight difference between the two is 0.3 (0.6-0.3=0.3), this condition is met. If at least one of the first discrimination mechanism and the second discrimination mechanism is not met, it is determined not to issue the automatic parking control strategy to the transport vehicle to be controlled. This decision-making mechanism can effectively avoid path conflicts during the parking process and improve the safety and efficiency of automatic parking. Through such a series of operations, the IoT sensor control system can accurately control and manage the automatic parking process of the transport vehicle to be controlled in a ski resort environment.

[0027] Throughout the entire process, the intelligent driving control processing algorithm plays a core role. It needs to be constantly updated and optimized to adapt to the complex and changing environment of the ski resort. For example, during the peak ski season, the parking lot has a large vehicle flow, and the algorithm needs to process sensor information more quickly and adjust the parking path planning labels and decision-making strategies in a timely manner. At the same time, with the continuous development of sensor technology, such as the emergence of higher-resolution visual sensors and longer-range laser radars, the intelligent driving control processing algorithm also needs to be improved accordingly to better utilize these new data and improve the accuracy and reliability of automatic parking control.

[0028] The IoT sensor control system also needs to consider the security of data transmission and storage. Since it involves the automatic parking control of many transport vehicles in the ski resort, a large amount of sensor information, vehicle status information and control strategy information need to be transmitted and stored in the system. In order to ensure the security of the data, the system can use encryption technology to encrypt the data. For example, the AES (Advanced Encryption Standard) algorithm is used to encrypt the data collected by the sensor. During the data transmission process, the SSL (Secure Sockets Layer) or TLS (Transport Layer Security) protocol is used for secure transmission to prevent the data from being stolen or tampered with. In terms of data storage, a secure database management system is used, and strict access rights are set. Only authorized personnel or equipment can access and modify relevant data.

[0029] In addition, the compatibility and scalability of the system are also important considerations. With the development of ski resorts and the increase in the number of transport vehicles, the IoT sensor control system needs to be compatible with different types of sensors and transport vehicle equipment. For example, the system should be able to easily incorporate newly introduced ultrasonic sensors with special functions or new electric transport vehicles into the control system. At the same time, the system also needs to have good scalability so that new functional modules can be added in the future, such as integration with other intelligent management systems of the ski resort (such as ticketing systems, ski equipment rental systems, etc.) to achieve more comprehensive intelligent management.

[0030] The layout and installation of sensors also need to be designed. For ultrasonic sensors, they need to be reasonably distributed around the vehicle body according to the shape and size of the transport vehicle to ensure all-round close-range detection. For example, multiple ultrasonic sensors are evenly installed on the front and rear bumpers and both sides of the vehicle body to form a complete close-range detection network. The installation location of the geomagnetic sensor needs to take into account the place with the least magnetic field interference to ensure the accuracy of the measurement. Visual sensors and lidars should be installed in a position where they can obtain the best field of view, generally on the top of the transport vehicle or at a higher position on the front of the vehicle to avoid being blocked by other objects.

[0031] In this way, through the above series of steps and technical means, effective control and management of the automatic parking process of the controlled transport vehicle in the ski resort environment is achieved, and the safety, efficiency and intelligence level of parking are improved.

[0032] In one possible embodiment, the intelligent driving control processing algorithm may be a Transformer algorithm. The following will describe in detail the process of the IoT sensor control system using the Transformer algorithm to process the first IoT sensor information to determine the initial vehicle environment state vector.

[0033] The Transformer algorithm is a deep learning model based on the attention mechanism. Its core advantage is that it can effectively process long sequence data and capture long-distance dependencies in the data. In the embodiment of the present invention, the first IoT sensor information of the transport vehicle can be regarded as a complex multimodal sequence data, and the Transformer algorithm can well mine the internal relationship therein.

[0034] 1. Characteristics and preprocessing of the first IoT sensor information 1. Ultrasonic sensing information Ultrasonic sensing information is presented as a series of distance values. For example, the distances to surrounding objects (other vehicles, snow piles, parking lot boundaries, etc.) are measured at regular intervals (e.g., every 0.1 seconds) in different directions around a ski resort transporter. These distance values ​​may range from 0.1 meters to 10 meters, depending on the sensor installation location and the surrounding environment.

[0035] Since ultrasonic sensors may have certain measurement errors (such as ±0.05 meters), the original distance data needs to be filtered. The mean filtering method can be used, for example, taking the average of five consecutive measurement values ​​as the current effective distance value to reduce the impact of noise. Then, the processed distance values ​​are arranged in the order of the sensor installation positions to form a distance value sequence as the input part of the Transformer algorithm.

[0036] 2. Geomagnetic sensor information The geomagnetic sensing information mainly reflects the changes in the transport vehicle relative to the earth's magnetic field. It may contain information such as magnetic field strength and magnetic field direction. For example, the value of magnetic field strength may be between 0.2 Gauss and 0.6 Gauss, and it changes with the position and posture of the transport vehicle.

[0037] First, the collected geomagnetic data is normalized and the magnetic field intensity value is mapped to the interval between 0 and 1 so that it can be processed on the same numerical scale as other sensor information. Then, according to the driving direction of the transport vehicle, the geomagnetic direction information is converted into a relative direction value relative to the transport vehicle, for example, the geomagnetic direction is converted into a clockwise angle value (0° to 360°) relative to the direction of the transport vehicle's head. The processed geomagnetic intensity and direction information are combined into a vector sequence as the input of the Transformer algorithm.

[0038] 3. Visual sensing information Visual sensing information is image data collected by cameras. In a ski resort environment, the image contains rich information, such as the layout of the parking lot (parking lines, channel markings, etc.), the appearance and location of other vehicles, the presence of skiers and equipment, etc. The resolution of the image may be 1920×1080 pixels, and each pixel has information in the three color channels of RGB.

[0039] First, the image is cropped and scaled to a size suitable for the Transformer algorithm, such as 224×224 pixels. Then, a convolutional neural network (CNN) is used to extract features from the image. CNN can extract features such as edges, textures, and object shapes in the image. The extracted feature vectors are arranged in a certain order to form an input sequence of visual sensor information.

[0040] 4. LiDAR sensing information LiDAR sensing information is 3D point cloud data. In a ski resort environment, it can accurately depict the 3D shape, height, and distance of objects around the transporter. Each point in the point cloud data contains 3D coordinates (x, y, z) and reflection intensity information. For example, in a single scan, tens of thousands of point cloud data points may be obtained, and their coordinate range may be within a spherical area with a radius of 10 meters centered on the transporter.

[0041] First, downsample the point cloud data to reduce the amount of data while retaining key shape information. The voxel grid filtering method can be used to divide the point cloud space into voxels of a certain size (such as 0.1m×0.1m×0.1m voxels), and only one representative point is retained in each voxel. Then, the processed point cloud data is converted into a format suitable for processing by the Transformer algorithm, for example, the coordinates and reflection intensity information of the point cloud are combined into a vector sequence.

[0042] 2. Input construction of Transformer algorithm 1) Multimodal data fusion The preprocessed ultrasonic sensor information sequence, geomagnetic sensor information vector sequence, visual sensor information feature vector sequence and lidar sensor information vector sequence are spliced ​​together in a certain order to form a unified input sequence. This input sequence contains various information about the environment around the transport vehicle and can fully describe the initial perception state of the vehicle.

[0043] 2) Positional encoding Since the attention mechanism in the Transformer algorithm is insensitive to the position information in the input sequence, it is necessary to add position encoding to the input sequence. Position encoding can be a combination of sine and cosine functions to add a position-related encoding value to each element in the input sequence. In this way, when processing the input sequence, the Transformer algorithm can distinguish sensor information elements at different positions, thereby better capturing the relationship between them.

[0044] 3. The processing of the Transformer algorithm 1. Multi-head attention mechanism 1) Calculate query, key and value In the encoder part of Transformer, each element in the input sequence is first transformed linearly to calculate the query, key, and value vectors. For each element in the input sequence, the query vector is used to match the key vectors of other elements to determine the degree of attention to the value vectors of other elements. For example, for a distance value element in ultrasonic sensor information, its query vector interacts with the key vectors of geomagnetic, visual, and lidar sensor information elements.

[0045] 2) Attention Calculation By calculating the dot product of the query vector and the key vector, and scaling and normalizing with the Softmax function, the attention weight of each element relative to other elements is obtained. Then, the attention weight is multiplied and summed with the corresponding value vector to obtain a new vector after attention weighting. Due to the use of a multi-head attention mechanism, multiple such attention calculation processes (for example, 8-head attention calculations) are performed in parallel, and each head focuses on a different representation subspace of the input sequence, thereby capturing richer relational information.

[0046] 2. Feedforward Neural Network 1) Fully connected layer processing The vector processed by the multi-head attention mechanism will pass through a feed-forward neural network (FFN). FFN consists of two fully connected layers with a nonlinear activation function (such as ReLU function) in the middle. The number of neurons in the fully connected layer can be set according to the actual situation. For example, the first fully connected layer has 512 neurons and the second fully connected layer has 256 neurons. FFN further performs nonlinear transformation on the vector after attention weighting to enhance the expression ability of the model.

[0047] (III) Layer Normalization and Residual Connection 1) Layer Normalization After each multi-head attention layer and feedforward neural network layer, layer normalization is performed. Layer normalization is to normalize the input of each neuron so that its mean is 0 and its variance is 1. This can speed up the training process of the model and improve the stability of the model.

[0048] 2) Residual Connection At the same time, residual connections are used to add the input and output of each sublayer (multi-head attention layer or feedforward neural network layer). Residual connections allow gradients to be back-propagated more effectively, avoiding the problem of gradient vanishing or gradient exploding in deep networks, and enabling the model to better learn the mapping relationship between input sequences and outputs.

[0049] 4. Determination of the Initial Vehicle Environment State Vector 1) Output layer design In the decoder part of the Transformer algorithm, an output layer is designed to generate the initial vehicle environment state vector. The number of neurons in the output layer is determined by the number of vehicle environment state parameters that need to be represented. For example, if you want to represent 10 parameters such as the position coordinates (x, y), the orientation angle, the type and distance of the surrounding objects of the transporter, then the output layer can be set to 10 neurons.

[0050] 2) Vector generation After being processed by the Transformer algorithm, the neurons in the output layer will generate corresponding values. These values ​​are combined to form the initial vehicle environment state vector. For example, the output position coordinate value may be the coordinate value relative to a fixed reference point in the parking lot, the heading angle value is the angle value relative to a certain direction (such as the north direction), the type of surrounding objects can be represented by coding (such as 0 for snow piles, 1 for other vehicles, etc.), and the distance value is the distance to the surrounding objects calculated based on the comprehensive calculation of different sensor information. This initial vehicle environment state vector comprehensively describes the initial state of the transport vehicle in the ski resort environment, and provides important basic data for subsequent parking path planning and other operations.

[0051] In the process of the IoT sensor control system using the Transformer algorithm to process the first IoT sensor information, the final generated initial vehicle environment state vector is a multi-dimensional vector that comprehensively reflects the initial state of the transport vehicle in the ski resort environment. The following will elaborate on this initial vehicle environment state vector through a specific numerical example.

[0052] First, considering the position information of the vehicle in three-dimensional space, the initial vehicle environment state vector may contain a numerical value representing the vehicle coordinates. For example, we use a fixed corner of the parking lot as the coordinate origin (0, 0, 0). In this coordinate system, the position coordinates of the transport vehicle may be (x = 10.5 meters, y = 15.2 meters, z = 0 meters). The x and y coordinates in the embodiment of the present invention represent the position of the vehicle on the horizontal plane relative to the origin, while the z coordinate is 0 meters because the vehicle is on flat ground. The accuracy of this coordinate value can reach the centimeter level, for example, ±0.01 meters, because the result of the fusion of multiple sensor information can provide relatively accurate positioning.

[0053] In addition to the position coordinates, the vehicle's heading angle is also an important parameter. In this example, the north direction is 0°, the clockwise direction is the positive direction, and the transport vehicle's heading angle is 45°. This angle value is also derived from factors such as the relative position of the lane line or other vehicles in the geomagnetic sensor information and the visual sensor information. The angle accuracy can be set to ±1°, which is sufficient to reflect the approximate direction of the vehicle for subsequent parking path planning.

[0054] Next, information about objects around the vehicle is also an important part of the initial vehicle environment state vector. Take the objects closest to the vehicle as an example. For example, the sensor detects a snow pile 3.2 meters in front of the vehicle, another transport vehicle 4.5 meters behind the left side of the vehicle, and a parking lot boundary facility (such as a guardrail) 2.8 meters to the right of the vehicle. The types of these objects can be represented by coding, such as setting the snow pile to 0, the transport vehicle to 1, and the parking lot boundary facility to 2. In this way, in the initial vehicle environment state vector, the representation of the snow pile in front may be (type = 0, distance = 3.2 meters, direction = 0°), the transport vehicle behind the left is represented as (type = 1, distance = 4.5 meters, direction = 225°), and the parking lot boundary facility on the right is represented as (type = 2, distance = 2.8 meters, direction = 90°). The accuracy of the distance can also reach the centimeter level, and the accuracy of the direction is ±1°.

[0055] The vehicle's own speed information is also included in this vector. For example, the transport vehicle is in a slow driving state at this time, and the speed is 0.5 m / s. This speed value is calculated by the vehicle's own speed sensor or based on continuous position information, with an accuracy of ±0.05 m / s. In some cases, the direction of the speed may also be included. If the vehicle is traveling in a straight line in a certain direction, the speed direction is the same as the vehicle's heading angle of 45°. If the vehicle is turning, the speed direction will change according to the turning situation.

[0056] The vehicle's body size information is also important for parking planning. For example, the length of this transport vehicle is 6 meters, the width is 2.5 meters, and the height is 3 meters. These values ​​can be pre-set and stored in the system, or verified through visual sensor information or other methods. The accuracy of the body size may be based on the actual manufacturing standard of the vehicle, with an accuracy of ±0.05 meters for length and width and ±0.1 meters for height.

[0057] Consider the vehicle's wheel status, such as the steering angle of the wheel. For example, the current steering angle of the wheel is 5°, which represents the offset angle of the wheel relative to the vehicle when it is traveling in a straight line, with an accuracy of ±0.5°. The steering angle of the wheel is crucial to determining the vehicle's driving trajectory and parking operations.

[0058] In addition, some special conditions of the vehicle's surroundings will also be reflected in the vector. For example, the ground conditions around the vehicle are detected through visual sensor information, and it is found that there is snow on the ground within 2-5 meters in front of the vehicle and the snow depth is 0.1 meters. The snow depth can be measured by lidar or other special sensors with an accuracy of ±0.01 meters. This ground condition information will affect the vehicle's driving and parking operations, for example, it needs to be considered when calculating braking distance and steering friction.

[0059] At the same time, the initial vehicle environment state vector may also contain information related to the parking lot layout. For example, the type of parking area the vehicle is in is coded as 0 if it is a perpendicular parking area and 1 if it is a parallel parking area. The distance and direction of the vehicle to the nearest parking space are also included. For example, the nearest perpendicular parking space is 6 meters to the left front of the vehicle and the direction is 30°.

[0060] In terms of the relative speed between the vehicle and surrounding vehicles or objects, if there are moving objects around, this information will also be reflected. For example, the transport truck behind the vehicle on the left is moving away from the vehicle at a speed of 0.3 m / s. This relative speed information is also calculated through continuous position monitoring with an accuracy of ±0.05 m / s.

[0061] The vehicle's acceleration information is also part of the vector. For example, if the vehicle's current acceleration is 0.1 m / s², this value represents the rate of change of the vehicle's speed with an accuracy of ±0.01 m / s². The direction of the acceleration can also be included, for example, if the vehicle is accelerating forward, the acceleration direction is the same as the vehicle's heading angle, which is 45°.

[0062] In addition, the state of the vehicle's suspension system may also be reflected in the vector. For example, the compression degree of the vehicle's suspension system is detected by the sensor, such as the compression of the left front wheel suspension of the vehicle by 0.03 meters and the compression of the right rear wheel suspension by 0.02 meters. This value reflects the load distribution of the vehicle and the impact of the flatness of the road on the vehicle, with an accuracy of ±0.005 meters.

[0063] The vehicle's lighting state can also be part of the initial vehicle environment state vector. For example, the vehicle's headlights are on, with 1 representing on and 0 representing off; the brake lights are off. Lighting state information is useful in interacting with other vehicles and driving safely in low-light ski resort environments.

[0064] Regarding the battery or fuel status of the vehicle, if it is an electric transport vehicle, the remaining power is 50%, with an accuracy of ±5%; if it is a fuel transport vehicle, the remaining fuel in the tank is 30 liters, with an accuracy of ±1 liter. Although this information does not seem to be directly related to parking path planning, it is useful in some cases, such as considering whether there is enough energy to move to the designated parking location or whether the charging equipment needs to be started during parking.

[0065] The vehicle's communication status may also be included. For example, the communication signal strength between the vehicle and the IoT sensor control system is -60dBm, which means that the communication connection is normal but the signal strength is at a medium level. The communication status information ensures that the vehicle and the control system can exchange data stably so that parking instructions can be received in time or vehicle status information can be uploaded.

[0066] The vehicle's internal environment information, such as the temperature inside the vehicle is 20°C with an accuracy of ±1°C; whether there are passengers in the vehicle, for example, if there is 1 passenger in the vehicle, 1 indicates there is a passenger and 0 indicates there is no passenger. Although this internal environment information may not have much impact on the parking operation itself, it may be considered in some comprehensive management scenarios (such as selecting a more suitable parking location based on the number of people in the vehicle to facilitate passengers getting on and off the vehicle, etc.).

[0067] The vehicle's sensor status also needs to be considered in the initial vehicle environment state vector. For example, the working status of an ultrasonic sensor, such as normal operation is represented by 1, and failure is represented by 0. If an ultrasonic sensor fails, this will affect the vehicle's accurate perception of the surrounding environment, especially in close-range obstacle detection. Therefore, the system needs to know the status of each sensor in order to make corresponding adjustments or prompt maintenance when processing sensor information.

[0068] External identification information of the vehicle may also be included, for example, unique identification information such as the vehicle's license plate number or identification code, which is very important for accurately identifying vehicles and recording parking information in parking management systems.

[0069] The initial vehicle environment state vector contains such multi-dimensional information with specific values, which comprehensively and meticulously describes the initial state of the transport vehicle in the ski resort environment. Each element in these numerical feature vectors does not exist in isolation, but is interrelated and mutually influential. For example, the vehicle's position coordinates and the distance and direction information of surrounding objects jointly determine the feasible path of the vehicle during parking; the vehicle's speed, acceleration, wheel steering angle and other information reflect the dynamic driving state of the vehicle, and the surrounding environment information needs to be combined when planning the parking path to ensure safe and efficient parking operations; the vehicle's internal environment, power or fuel volume, communication status and other information, although seemingly auxiliary in some cases, also play an indispensable role in overall vehicle management and optimization of parking experience. Moreover, the values ​​in this vector are not static. With the movement of the vehicle, changes in the surrounding environment and continuous monitoring of the sensor, the IoT sensor control system will continuously update this vector to provide the latest and most accurate basic data for subsequent parking path planning, formulation of automatic parking control strategies and other operations.

[0070] In actual system operation, due to the complexity of sensor information and the dynamic changes of the environment, each value of the initial vehicle environment state vector may be affected by multiple factors. For example, in a snowy environment, visual sensor information may cause certain errors in the judgment of the distance and type of surrounding objects due to the interference of snowflakes, which requires supplementation and correction by other sensor information (such as lidar sensor information). Similarly, geomagnetic sensor information may be interfered by metal facilities or other magnetic field sources around the ski resort, thereby affecting the accuracy of the vehicle's position and orientation angle. Therefore, the IoT sensor control system needs to continuously optimize algorithms and fusion strategies to ensure that the initial vehicle environment state vector can reflect the actual state of the vehicle as accurately as possible.

[0071] At the same time, this initial vehicle environment state vector may have different focuses in different ski resort environments and different types of transport vehicle scenarios. In some large ski resorts, where the vehicle flow is large and the parking lot layout is complex, the relative speed of the vehicle and the surrounding vehicles, parking lot layout information, etc. may be more important; while in some small ski resorts or specific transport vehicle scenarios (such as small vehicles specifically used to transport ski equipment), the vehicle body size, cargo loading (if any), etc. may become more critical information. Therefore, the IoT sensor control system needs to flexibly adjust the composition of the initial vehicle environment state vector and the weight of each value according to the specific application scenario to achieve the most optimized automatic parking control.

[0072] In terms of data transmission and storage, this initial vehicle environment state vector containing many numerical features needs to be effectively managed. Since the amount of data is large and needs to be updated in real time, the use of efficient data compression algorithms can reduce the bandwidth requirements for data transmission and the space occupied by storage. For example, a lossless compression algorithm can be used to compress the values ​​in the vector to improve the efficiency of data transmission and storage without losing data accuracy. At the same time, in order to ensure the security of the data, encryption technology, such as the AES (Advanced Encryption Standard) encryption algorithm, can be used during the transmission process to encrypt the vector data to prevent the data from being stolen or tampered with during the transmission process. In terms of storage, the data is stored in a reliable database system and strict access rights are set. Only authorized devices and personnel can access and modify the data.

[0073] In summary, the initial vehicle environment state vector comprehensively reflects the initial state of the transport vehicle in the ski resort environment through a series of specific numerical feature vectors. These numerical values ​​cover various information such as the vehicle's position, posture, surrounding environment, and its own state, and play a fundamental and key role in the automatic parking process of the entire IoT sensor control system.

[0074] On the basis of the above steps 110 to 140, the debugging method of the intelligent driving control processing algorithm includes: obtaining the second IoT sensor information of multiple historical transport vehicles and the prior basis of each historical transport vehicle, the prior basis of the historical transport vehicle characterizing whether there is a relevant path intersection between the historical transport vehicle and the automatic parking control strategy during the automatic parking process; using the Transformer algorithm to mine the state characteristics of the second IoT sensor information of the multiple historical transport vehicles to obtain the initial vehicle environment state vector of each historical transport vehicle; using the Transformer algorithm to determine the initial vehicle environment state vector of each historical transport vehicle according to the initial vehicle environment state vector of each historical transport vehicle. The parking path planning label of the historical transport vehicle represents the probability that the historical transport vehicle matches the path intersection; the prior confidence of each historical transport vehicle is determined based on the initial vehicle environment state vector of each historical transport vehicle, and the prior confidence of the historical transport vehicle is used to adjust the initial vehicle environment state vector of the historical transport vehicle to a target mode; based on the prior basis, parking path planning label and prior confidence of each historical transport vehicle, the Transformer algorithm is debugged to obtain an intelligent driving control processing algorithm, and the intelligent driving control processing algorithm is used to determine the probability that the transport vehicle to be controlled matches the path intersection.

[0075] In the entire technical solution, the debugging of the intelligent driving control processing algorithm is crucial to improving the accuracy and effectiveness of automatic parking control.

[0076] First, obtain the second IoT sensor information of multiple historical transport vehicles and the prior basis of each historical transport vehicle. The second IoT sensor information in the embodiment of the present invention covers multiple types of sensor data, which is similar in nature and function to the first IoT sensor information mentioned above. For example, ultrasonic sensor information can provide distance information of objects within a certain range around the vehicle. In the historical data, it is possible that at a certain moment in time, the ultrasonic sensor in front of a historical transport vehicle detects that the distance to the obstacle in front is 1.2 meters, while the sensor on the side detects that the distance to another object is 2.5 meters. Geomagnetic sensor information reflects the state of the vehicle in the geomagnetic field. If expressed in terms of the numerical value of the geomagnetic intensity, the geomagnetic intensity measurement value at a certain position may be 0.4 Gauss, which helps to determine the approximate position of the vehicle in the parking lot. Visual sensor information contains rich image data content, and the layout elements of the parking lot can be identified from the image, such as parking space lines, passages, and the relative positions of other vehicles. For example, it can be known from visual sensor information that there are three other vehicles around a historical transport vehicle, and the lateral distance from the nearest vehicle is 1.8 meters. The lidar sensing information constructs a three-dimensional point cloud map of the vehicle's surrounding environment. For example, in the lidar scanning data of a historical transport vehicle, the outline of an object with a height of 0.5 meters at a distance of 5 meters from the vehicle in a certain direction is accurately depicted.

[0077] At the same time, the prior evidence of each historical transport vehicle is an important information, which clearly characterizes whether there are relevant path intersections with the automatic parking control strategy during the automatic parking process of the historical transport vehicle. This information is obtained based on the actual parking operation records. Some prior evidence of historical transport vehicles indicates that there are indeed path intersections with the control strategy during the parking process, while others indicate that there are no such situations.

[0078] Next, the Transformer algorithm is used to mine the state features of the second IoT sensor information of multiple historical transport vehicles, and then the initial vehicle environment state vector of each historical transport vehicle is obtained. Taking a specific historical transport vehicle as an example, its initial vehicle environment state vector contains a lot of numerical information with practical significance. The position coordinates of the vehicle may be (9.5 meters, 13.2 meters, 0 meters), where the z coordinate is 0 meters, indicating that the vehicle is on the horizontal ground. The vehicle's orientation angle is, for example, 40°, which is relative to a certain reference direction (such as the north direction). For the object information around the vehicle, there may be a snow pile 3.0 meters in front of the right side of the vehicle (indicated by type code 0), and another transport vehicle 4.0 meters behind the left side of the vehicle (type code 1). The vehicle's own speed value is 0.4 meters / second, and this speed information reflects the vehicle's driving state at the time. The steering angle of the wheel is, for example, 8°, which is of great significance for analyzing the vehicle's driving trajectory and parking operations. The vehicle has a body length of 5.8 meters, a width of 2.3 meters, and a height of 2.8 meters. These body size data are indispensable in parking path planning. For example, when judging whether a vehicle can smoothly enter a parking space, it is necessary to consider the relationship between the body width and the parking space width.

[0079] Then, based on this initial vehicle environment state vector, the Transformer algorithm determines the parking path planning label of each historical transport vehicle, which represents the probability of the historical transport vehicle matching the path intersection. For example, the parking path planning label of a historical transport vehicle may indicate that there is a 0.35 probability of a path intersection. This probability is calculated by comprehensively considering multiple factors in the vehicle's initial environment state vector. For example, the vehicle is located close to the intersection of the aisles of the parking lot, the surrounding vehicles are parked densely, and the vehicle's own speed and steering angle work together, causing the algorithm to determine that there is a certain path intersection risk.

[0080] Then, the prior confidence of each historical transport vehicle is determined based on the initial vehicle environment state vector of each historical transport vehicle. The prior confidence is used to adjust the initial vehicle environment state vector of the historical transport vehicle to the target mode. For example, the prior confidence of a historical transport vehicle is 0.7, which reflects the reliability of the initial environment state vector of the vehicle or the importance weight in the overall data. When the prior confidence is high, when debugging the Transformer algorithm, more attention will be paid to each value in the initial environment state vector of the vehicle. For example, for key values ​​such as vehicle position coordinates and the distance of surrounding objects, their influence weight in the decision-making process will be adjusted according to the prior confidence during the algorithm debugging process. If the prior confidence is 0.7, then when calculating the probability related to the path intersection, key information such as vehicle position coordinates may be given a higher weight ratio, which may be increased from the original 30% weight to 40% weight, so as to more accurately adjust the initial vehicle environment state vector to the target mode to better meet the algorithm debugging needs.

[0081] Finally, the Transformer algorithm is debugged based on the prior basis, parking path planning label and prior confidence of each historical transport vehicle to obtain the intelligent driving control processing algorithm. In this process, if the prior basis of a historical transport vehicle indicates that there is a path intersection, and the parking path planning label shows a high path intersection probability (such as 0.6), and the prior confidence is also high (such as 0.8), then when debugging the Transformer algorithm, the relevant parameters in the algorithm will be adjusted. For example, the weight distribution of factors such as vehicle speed and surrounding object distribution when calculating the path intersection probability may be adjusted. For example, the original weight of vehicle speed in calculating the path intersection probability is 0.1. Due to the existence of the above situation, its weight may be adjusted to 0.15, so that the algorithm can more accurately determine the probability of matching the path intersection when processing the transport vehicle to be controlled. This debugging process is an iterative optimization process. By comprehensively analyzing and processing the data of multiple historical transport vehicles, the parameters of the Transformer algorithm are continuously adjusted, and finally an intelligent driving control processing algorithm suitable for determining the probability of matching the path intersection of the transport vehicle to be controlled is obtained.

[0082] By using the rich information of historical transport vehicles to debug the Transformer algorithm to obtain the intelligent driving control processing algorithm, there are many significant advantages. In terms of accuracy, taking the values ​​in the vehicle environment state vector as an example, precise information such as vehicle position and distance to surrounding objects are effectively used in the algorithm debugging process, so that the intelligent driving control processing algorithm can more accurately determine the probability of the controlled transport vehicle matching the path intersection. In terms of parking safety, the accurate determination of the probability of parking path planning labels helps to identify potential path intersection risks in advance, thereby avoiding dangerous situations such as collisions during automatic parking. The introduction of prior confidence provides a reasonable basis for the adjustment of the initial vehicle environment state vector, so that the algorithm can be optimized in a targeted manner according to the reliability of different historical transport vehicle data. Overall, this debugging method optimizes the algorithm's control of the automatic parking of transport vehicles, improves the reliability and efficiency of the entire parking system, reduces the occurrence of parking failures or safety accidents caused by inaccurate algorithms, and can better adapt to the needs of automatic parking in complex environments such as ski resorts.

[0083] In a preferred embodiment, the multiple historical transport vehicles include multiple first transport vehicles to which the automatic parking control strategy is issued and multiple second transport vehicles to which the automatic parking control strategy is not issued; the using the Transformer algorithm to determine the parking path planning label of each historical transport vehicle according to the initial vehicle environment state vector of each historical transport vehicle includes: using the Transformer algorithm to determine the parking path planning label of each first transport vehicle based on a first debugging annotation and the initial vehicle environment state vector of the multiple first transport vehicles, the first debugging annotation representing the result of the automatic parking control strategy being issued; using the Transformer algorithm to determine the parking path planning label of each second transport vehicle based on a second debugging annotation and the initial vehicle environment state vector of the multiple second transport vehicles, the second debugging annotation representing the result of not having the automatic parking control strategy issued.

[0084] In this technical solution, multiple historical transport vehicles are divided into multiple first transport vehicles to which automatic parking control strategies are issued and multiple second transport vehicles to which automatic parking control strategies are not issued. This classification helps to analyze parking-related situations in different situations in more detail.

[0085] For the first transport vehicle, when determining its parking path planning label, the Transformer algorithm is used based on the first debugging annotation and the initial vehicle environment state vectors of multiple first transport vehicles. The first debugging annotation represents the result of the automatic parking control strategy being issued. For example, in the initial vehicle environment state vector of a first transport vehicle, the coordinates of the vehicle are (10.2 meters, 14.5 meters, 0 meters), the orientation angle is 45°, there is an object around it that is 2.8 meters away from the vehicle and the type is parking facility coded as 2, and the vehicle speed is 0.3 meters / second. Based on such an initial vehicle environment state vector, combined with the first debugging annotation, for example, the first debugging annotation contains the interaction information of the transport vehicle with other vehicles within a specific time after the automatic parking control strategy is issued, such as the minimum safe distance from another vehicle during parking is 0.5 meters, etc., the Transformer algorithm determines the parking path planning label of the first transport vehicle by comprehensively analyzing these factors. For example, this label may indicate that the first transport vehicle has a probability of 0.4 of having a path intersection. This probability is derived by considering multiple factors such as the vehicle's initial position, speed, surrounding objects, and the interaction results after the automatic parking control strategy is issued.

[0086] For the second transport vehicle, the Transformer algorithm is used to determine its parking path planning label based on the second debugging annotation and the initial vehicle environment state vectors of multiple second transport vehicles, wherein the second debugging annotation represents the result of not being issued an automatic parking control strategy. Taking a second transport vehicle as an example, the vehicle coordinates in its initial vehicle environment state vector are (8.8 meters, 11.3 meters, 0 meters), the orientation angle is 35°, there is an object around it that is 3.2 meters away from the vehicle and the type is snow pile coded 0, and the vehicle speed is 0.2 meters / second. The second debugging annotation may contain relevant information such as the reason why the automatic parking control strategy was not issued, such as the surrounding environment is too complex or the vehicle's own state is not suitable for automatic parking. The Transformer algorithm determines the parking path planning label of the second transport vehicle based on this information and the initial vehicle environment state vector. For example, this label may indicate that the second transport vehicle has a probability of 0.3 to have a path intersection. The probability in the embodiment of the present invention is obtained by comprehensively considering the initial state of the vehicle and the relevant factors that the automatic parking control strategy has not been issued.

[0087] In the whole process, by performing such processing on the first transport vehicle and the second transport vehicle respectively, the parking path planning of the vehicle under different strategies can be analyzed more comprehensively and meticulously. For the first transport vehicle, since the automatic parking control strategy has been issued, the relevant information in the first debugging annotation can reflect various situations related to the automatic parking control strategy in the actual parking operation. These situations, combined with the initial vehicle environment state vector, can more accurately determine the parking path planning label. For the second transport vehicle, although the automatic parking control strategy has not been issued, the information in the second debugging annotation is equally important. It can provide relevant factors about the vehicle not performing the automatic parking operation, and determine the parking path planning label together with the initial vehicle environment state vector. This helps to understand the parking possibility of the vehicle in the parking lot environment from another perspective.

[0088] By determining the parking path planning labels for the first transport vehicle and the second transport vehicle based on different debugging annotations and the initial vehicle environment state vector, the accuracy of parking path planning can be improved. From the numerical example, the values ​​in the state vector such as the position of the vehicle, the type of surrounding objects and the distance are combined with the information in the debugging annotation to more accurately reflect the parking path of the vehicle in different situations. This helps to optimize the automatic parking control strategy, better cope with complex parking environments, improve the overall performance and reliability of the parking system, and reduce risks and uncertainties in the parking process.

[0089] In an optional embodiment, determining the prior confidence of each historical transport vehicle based on the initial vehicle environment state vector of each historical transport vehicle includes: performing feature mapping on the initial vehicle environment state vector of each historical transport vehicle to obtain the driving control decision mapping vector of each historical transport vehicle; determining the prior confidence of each historical transport vehicle based on the driving control decision mapping vector of each historical transport vehicle.

[0090] Furthermore, the driving control decision mapping vector of the historical transport vehicle includes local control decision mapping vectors of the historical transport vehicle in multiple attention indicators; the prior confidence of each historical transport vehicle is determined based on the driving control decision mapping vectors of each historical transport vehicle, including: for any historical transport vehicle, according to the local control decision mapping vector of any historical transport vehicle in a first attention indicator and the local control decision mapping vector of each historical transport vehicle in the first attention indicator, determining a first mapping vector comparison result, the first attention indicator being any one of the multiple attention indicators; according to the local control decision mapping vector of any historical transport vehicle in the first attention indicator The second mapping vector comparison result is determined by comparing the local control decision mapping vectors of the two attention indicators and the local control decision mapping vectors of each historical transport vehicle in the second attention indicator, where the second attention indicator is any attention indicator among the multiple attention indicators except the first attention indicator; based on the comparison result of the first mapping vector and the comparison result of the second mapping vector, the local control decision mapping correlation between the first attention indicator and the second attention indicator of any historical transport vehicle is determined; based on the local control decision mapping correlation between each two attention indicators of each historical transport vehicle, the prior confidence of each historical transport vehicle is determined.

[0091] Furthermore, the first mapping vector comparison result is determined based on the local control decision mapping vector of any historical transport vehicle in the first attention index and the local control decision mapping vectors of each historical transport vehicle in the first attention index, including: determining the local control decision mapping mean vector of the first attention index based on the basic confidence of each historical transport vehicle and the local control decision mapping vector of each historical transport vehicle in the first attention index; determining the local control decision mapping reinforcement vector of any historical transport vehicle in the first attention index based on the basic confidence of any historical transport vehicle and the local control decision mapping vector of any historical transport vehicle in the first attention index; and taking the vector operation value between the local control decision mapping reinforcement vector of any historical transport vehicle in the first attention index and the local control decision mapping mean vector of the first attention index as the first mapping vector comparison result.

[0092] In this embodiment, the entire technical solution focuses on the key link of determining the prior confidence based on the initial vehicle environment state vector of each historical transport vehicle, and its specific operation process involves multiple levels of complex calculations and logical judgments.

[0093] First, feature mapping is performed on the initial vehicle environment state vector of each historical transport vehicle to obtain the driving control decision mapping vector of each historical transport vehicle. The initial vehicle environment state vector of the historical transport vehicle contains rich vehicle-related information, which is an important basis for determining the vehicle driving control decision. For example, in the initial vehicle environment state vector of a certain historical transport vehicle, the coordinate position of the vehicle is (11.5 meters, 16.2 meters, 0 meters). The coordinates in the embodiment of the present invention represent the position of the vehicle in the specific coordinate system of the parking lot, where the z coordinate is 0 meters, indicating that the vehicles are on the same horizontal plane. The vehicle's heading angle is 50°, which reflects the direction of the vehicle relative to a fixed direction (such as the north direction). The object information around the vehicle is also an important part of the vector, such as there is an object 3.8 meters behind the right side of the vehicle, its type is other vehicles (for example, coded as 1), the current driving speed of the vehicle is 0.4 meters / second, and the steering angle of the wheel is 10°, etc. Through feature mapping operations, the information in these initial vehicle environment state vectors is converted into a driving control decision mapping vector. This mapping vector contains mapping information related to how the vehicle makes driving control decisions. It is no longer just the original vehicle state information, but information that has been converted through a certain specific mapping relationship and is more directly related to the driving control decision.

[0094] The driving control decision mapping vector of the historical transport vehicle includes the local control decision mapping vectors of multiple attention indicators. This means that in this vector, for different attention indicators, there are corresponding local control decision mapping parts. For any historical transport vehicle, the process of determining the prior confidence is a multi-step and interrelated process.

[0095] When determining the comparison result of the first mapping vector, it is necessary to determine the local control decision mapping mean vector of the first attention index based on the basic confidence of each historical transport vehicle and the local control decision mapping vector of the first attention index. For example, there is a group of historical transport vehicles, of which the basic confidences of three historical transport vehicles are 0.6, 0.7 and 0.8 respectively. For the local control decision mapping vectors of these three historical transport vehicles under the first attention index, for example, they are vector A=[a1, a2, a3], vector B=[b1, b2, b3] and vector C=[c1, c2, c3] respectively (the vector elements in the embodiment of the present invention represent the values ​​related to the local control decision mapping, and the specific values ​​depend on the vehicle state and the mapping logic of the algorithm). For example, according to the idea of ​​weighted average (but the specific weighting coefficient is determined by the algorithm), the local control decision mapping mean vector of the first attention index is calculated. At the same time, the local control decision mapping reinforcement vector of the transport vehicle in the first attention index is determined based on the basic confidence of any historical transport vehicle and its local control decision mapping vector in the first attention index. For example, for a historical transport vehicle with a basic confidence of 0.7, its local control decision mapping vector under the first attention index is vector D=[d1, d2, d3]. Through a specific algorithm operation (determined by the internal logic of the algorithm), the basic confidence is calculated with the elements in vector D to obtain a local control decision mapping reinforcement vector. Then, the vector operation value between the local control decision mapping reinforcement vector of the first attention index and the local control decision mapping mean vector of the first attention index of the historical transport vehicle is used as the first mapping vector comparison result. The calculation method of this vector operation value is based on the operation rules pre-defined by the algorithm, which reflects the degree of difference between the historical transport vehicle and the overall mean under the first attention index.

[0096] Then, according to the local control decision mapping vector of any historical transport vehicle under the second attention index and the local control decision mapping vector of each historical transport vehicle under the second attention index, the second mapping vector comparison result is determined. The second attention index in the embodiment of the present invention is any attention index other than the first attention index among the multiple attention indexes. For example, under the second attention index, there is another group of relevant data of historical transport vehicles. For a specific historical transport vehicle, its local control decision mapping vector under the second attention index is vector E=[e1, e2, e3], and the local control decision mapping vectors of other historical transport vehicles under the second attention index are respectively vector F=[f1, f2, f3], vector G=[g1, g2, g3], etc. According to a method similar to that for determining the first mapping vector comparison result, firstly, the local control decision mapping mean vector of the second attention index is determined based on the basic confidence of each historical transport vehicle and the local control decision mapping vector of the second attention index, and then the local control decision mapping reinforcement vector of the transport vehicle in the second attention index is determined based on the basic confidence of the specific historical transport vehicle and its local control decision mapping vector of the second attention index, and finally the vector operation value between the two vectors is calculated to obtain the second mapping vector comparison result. This result also reflects the difference between the historical transport vehicle and the overall mean under the second attention index.

[0097] Then, based on the comparison result of the first mapping vector and the comparison result of the second mapping vector, the correlation degree of the local control decision mapping between the first attention index and the second attention index of any historical transport vehicle is determined. This correlation degree is determined based on the comparison results of the two mapping vectors obtained previously, and is calculated through a specific algorithm logic. For example, the comparison result of the first mapping vector is vector H=[h1, h2, h3], and the comparison result of the second mapping vector is vector I=[i1, i2, i3]. The algorithm obtains the correlation degree of the local control decision mapping between the first attention index and the second attention index of the historical transport vehicle by performing a specific combination operation (defined by the algorithm) on the elements of the two vectors. This correlation degree reflects the closeness of the relationship between the local control decision mapping of the historical transport vehicle under these two attention indicators. It is an indicator that comprehensively considers the comparison results of the vehicle and the overall situation under two different attention indicators.

[0098] Finally, the prior confidence of each historical transport vehicle is determined based on the correlation between the local control decision mapping of each historical transport vehicle between each two attention indicators. For example, the local control decision mapping correlation between a certain historical transport vehicle and multiple different attention indicator pairs is correlation 1, correlation 2, correlation 3, etc. (each correlation is a value obtained by the previous calculation method). The algorithm obtains the prior confidence of the historical transport vehicle by comprehensively calculating these correlations (according to the internal logic of the algorithm, it may involve weighted summation, normalization and other operations). For example, the prior confidence of a certain historical transport vehicle is calculated to be 0.75. This prior confidence is of great significance in the entire technical solution. It is used to adjust the initial vehicle environment state vector of the historical transport vehicle to the target mode, thereby playing a key role in the subsequent algorithm debugging and other processes. For example, when adjusting the initial vehicle environment state vector, the prior confidence may affect the weight adjustment of different elements in the vector, or modify certain elements, so that the initial vehicle environment state vector is more in line with the actual control requirements.

[0099] The prior confidence is determined through such a complex and systematic calculation process. From the perspective of the eigenvector value, the calculation is based on the initial vehicle environment state vector value such as the vehicle's position and speed. This method greatly improves the accuracy of the prior confidence determination. The improved accuracy of the prior confidence can more accurately adjust the initial vehicle environment state vector to the target mode. In this way, it helps to improve the accuracy and reliability of the entire algorithm in automatic parking control, so that the automatic parking system can take into account different vehicle states and environmental conditions in more detail. In actual automatic parking scenarios, more accurate prior confidence can optimize the formulation of parking-related decisions, reduce parking risks caused by inaccurate decisions, improve parking efficiency, and better adapt to complex and changeable parking environments. Whether it is a dense vehicle situation or a situation with special obstacles, it can perform automatic parking operations more effectively.

[0100] In an exemplary technical scheme, the Transformer algorithm is debugged according to the prior basis, parking path planning label and prior confidence of each historical transport vehicle to obtain an intelligent driving control processing algorithm, including: for any historical transport vehicle, according to the prior basis of any historical transport vehicle and the parking path planning label of any historical transport vehicle, the first path planning node error corresponding to any historical transport vehicle is determined; according to the first path planning node error corresponding to any historical transport vehicle and the prior confidence of any historical transport vehicle, the weighted error of the first path planning node of any historical transport vehicle is determined; according to the weighted errors of the first path planning nodes of each historical transport vehicle, a first training error is determined; according to the first training error, the Transformer algorithm is debugged to obtain the intelligent driving control processing algorithm.

[0101] In this exemplary technical solution, the entire process revolves around debugging the Transformer algorithm based on the prior evidence of each historical transport vehicle, the parking path planning label and the prior confidence to obtain the intelligent driving control processing algorithm.

[0102] First, for any historical transport vehicle, the first path planning node error is determined based on its prior basis and parking path planning label. The prior basis of the historical transport vehicle characterizes whether there is a path intersection related to the automatic parking control strategy during the automatic parking process, and the parking path planning label characterizes the probability of the vehicle matching the path intersection. For example, the prior basis of a historical transport vehicle indicates that there is a path intersection, and its parking path planning label shows that there is a probability of 0.6 that there is a path intersection. If the actual situation is that there is indeed a path intersection, then the first path planning node error may be a value related to the difference between the two. This value is determined based on the logic inside the algorithm and is related to the comparison between the specific representation of the prior basis and the parking path planning label and the actual results.

[0103] Next, the weighted error of the first path planning node of any historical transport vehicle is determined based on the first path planning node error corresponding to the historical transport vehicle and its prior confidence. For example, the prior confidence of a historical transport vehicle is 0.7, and its first path planning node error is a certain specific value (obtained from the previous calculation). The weighted error of the first path planning node is obtained by calculating the first path planning node error and the prior confidence in the manner specified by the algorithm (this calculation method is based on the logic of the algorithm). The prior confidence plays the role of weighting the first path planning node error here, so that the errors of historical transport vehicles with different confidences have different weights in subsequent calculations.

[0104] Then, the first training error is determined based on the weighted error of the first path planning node of each historical transport vehicle. For example, there are multiple historical transport vehicles, and each historical transport vehicle has its own weighted error of the first path planning node. These weighted errors are combined in a specific way to form the first training error. For example, the weighted errors of the first path planning nodes of all historical transport vehicles may be summed or the first training error may be obtained in a weighted average manner (determined by the internal logic of the algorithm). This first training error reflects the overall error of all historical transport vehicles in terms of path planning nodes, and comprehensively considers the influence of the prior basis, parking path planning label and prior confidence of each historical transport vehicle.

[0105] Finally, the Transformer algorithm is debugged according to the first training error to obtain the intelligent driving control processing algorithm. The first training error is used as a feedback information to adjust the parameters in the Transformer algorithm. Since the Transformer algorithm plays a key role in determining the initial vehicle environment state vector of the historical transport vehicle, the parking path planning label, etc., debugging it through the first training error can optimize the performance of the algorithm so that the algorithm can more accurately determine the probability of matching path intersections when processing the controlled transport vehicle. For example, if the first training error is large, it means that the previous Transformer algorithm has a large deviation when processing historical transport vehicle data, then it is necessary to adjust the parameters in the algorithm, such as adjusting the weights of factors such as vehicle position and surrounding objects when calculating the parking path planning label, or adjusting the coefficients in certain calculation steps when determining the prior confidence, etc., so as to obtain the optimized intelligent driving control processing algorithm.

[0106] Through this method of determining the training error and debugging the Transformer algorithm based on the prior basis of historical transport vehicles, parking path planning labels and prior confidence, from the perspective of specific numerical examples, such as the probability value of the parking path planning label of the historical transport vehicle and the prior confidence value, they are all involved in the optimization process of the algorithm. This helps to improve the accuracy of the intelligent driving control processing algorithm, so that the algorithm can more accurately judge the automatic parking situation of the controlled transport vehicle, reduce the error in parking path planning, thereby improving the reliability and efficiency of the entire automatic parking system, and better adapting to different parking scenarios and vehicle conditions.

[0107] In another exemplary technical scheme, the multiple historical transport vehicles include multiple first transport vehicles to which automatic parking control strategies have been issued and multiple second transport vehicles to which the automatic parking control strategies have not been issued; the Transformer algorithm is debugged based on the prior basis, parking path planning labels and prior confidence of each historical transport vehicle to obtain an intelligent driving control processing algorithm, including: determining a first training error based on the prior basis, parking path planning labels and prior confidence of each historical transport vehicle; determining a second training error based on an initial state vector comparison result between an initial vehicle environment state vector of the multiple first transport vehicles and an initial vehicle environment state vector of the multiple second transport vehicles; and debugging the Transformer algorithm based on the first training error and the second training error to obtain the intelligent driving control processing algorithm.

[0108] In this exemplary technical solution, the entire technical solution focuses on debugging the Transformer algorithm based on relevant information of multiple historical transport vehicles (including multiple first transport vehicles to which automatic parking control strategies have been issued and multiple second transport vehicles to which automatic parking control strategies have not been issued) to obtain an intelligent driving control processing algorithm.

[0109] First, the first training error is determined based on the prior basis, parking path planning label and prior confidence of each historical transport vehicle. For each historical transport vehicle, the prior basis clarifies whether there is a path intersection related to the automatic parking control strategy during the automatic parking process. The parking path planning label represents the probability of the vehicle matching the path intersection. The prior confidence is used to measure the reliability of the vehicle's initial vehicle environment state vector or its importance weight in the overall data. For example, the prior basis of a first transport vehicle indicates that there is a path intersection, and its parking path planning label shows that there is a probability of 0.6 that there is a path intersection, and the prior confidence is 0.7. The prior basis of another first transport vehicle indicates that there is no path intersection, and the parking path planning label shows that there is a probability of 0.2 that there is a path intersection, and the prior confidence is 0.6. A similar situation exists for the second transport vehicle.

[0110] When determining the first training error, first determine an intermediate error based on the prior basis of each historical transport vehicle and the parking path planning label (similar to the first path planning node error mentioned earlier, but here all historical transport vehicles are considered comprehensively). This intermediate error is a value determined according to the internal logic of the algorithm based on the relationship between the two. Then, this intermediate error and the prior confidence are calculated in the manner specified by the algorithm to obtain the weighted error corresponding to each historical transport vehicle. These weighted errors are combined in a specific manner (such as summation or weighted average, etc., determined by the internal logic of the algorithm) to form the first training error.

[0111] Next, the second training error is determined based on the initial state vector comparison results between the initial vehicle environment state vectors of multiple first transport vehicles and the initial vehicle environment state vectors of multiple second transport vehicles. The initial vehicle environment state vector of the first transport vehicle contains various information such as the vehicle's position coordinates, orientation angle, surrounding object information, vehicle speed, and body size. For example, the vehicle position coordinates in the initial vehicle environment state vector of a first transport vehicle are (10.5 meters, 13.2 meters, 0 meters), the orientation angle is 45°, there is an object (type 1) 2.8 meters ahead, the vehicle speed is 0.3 meters / second, the body length is 5.8 meters, and the width is 2.3 meters. The initial vehicle environment state vector of another first transport vehicle has different values ​​on these parameters. The initial vehicle environment state vector of the second transport vehicle also has similar components and has its own values.

[0112] Determining the initial state vector comparison result is a complex process. For example, for the vehicle position coordinates, it is possible to calculate the average value or variance of the coordinate difference between multiple first transport vehicles and multiple second transport vehicles on a certain coordinate axis (the specific calculation method is determined by the algorithm). For vehicle speed, it is possible to calculate the distribution of the difference between the two speeds. By comprehensively calculating these different comparison results according to the internal logic of the algorithm, the second training error is obtained. This second training error reflects the impact of the difference in the initial vehicle environment state between the first transport vehicle and the second transport vehicle on the algorithm debugging.

[0113] Finally, the Transformer algorithm is debugged according to the first training error and the second training error to obtain the intelligent driving control processing algorithm. The first training error and the second training error, as feedback information from two different sources, act together on the debugging of the Transformer algorithm. If the first training error is large, it means that there is a large deviation in the algorithm based on the prior basis, parking path planning label and prior confidence, and it may be necessary to adjust the weights or internal logical relationships in the algorithm when calculating these parameters. If the second training error is large, for example, it is caused by the difference in the initial vehicle environment state vector between the first transport vehicle and the second transport vehicle, then it may be necessary to adjust the relevant parameters of the algorithm when processing the initial vehicle environment state of different types of transport vehicles (the first transport vehicle and the second transport vehicle), such as the importance of different parameters of different types of vehicles in the feature mapping process. Through such a debugging process, the Transformer algorithm can be optimized into an intelligent driving control processing algorithm, so that when processing the controlled transport vehicle, the probability of matching the intersection of the path can be more accurately determined.

[0114] By comprehensively considering the first training error and the second training error to debug the Transformer algorithm, from the numerical example, the vehicle's parking path planning label probability, prior confidence, and the value in the initial vehicle environment state vector are all involved in the algorithm debugging. This helps to improve the accuracy of the intelligent driving control processing algorithm, so that the algorithm can more accurately handle the automatic parking of different types of transport vehicles (the first transport vehicle and the second transport vehicle), reduce errors in parking path planning, improve the reliability and efficiency of the entire automatic parking system, and enhance the system's adaptability to different parking scenarios and vehicle states.

[0115] In another exemplary technical scheme, the Transformer algorithm is debugged based on the prior basis, parking path planning labels and prior confidence of each historical transport vehicle to obtain an intelligent driving control processing algorithm, including: determining a first training error based on the prior basis, parking path planning labels and prior confidence of each historical transport vehicle; determining hyper-parameter data based on the algorithm weights of the Transformer algorithm, the hyper-parameter data characterizing the long-distance dependency performance of the Transformer algorithm; determining a third training error based on the hyper-parameter data and the number of historical transport vehicles; debugging the Transformer algorithm based on the first training error and the third training error to obtain the intelligent driving control processing algorithm.

[0116] In this yet another exemplary technical solution, the entire process revolves around debugging the Transformer algorithm based on the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence level to obtain the intelligent driving control processing algorithm. This process is of great significance in the automatic parking scenario at the ski resort.

[0117] The ski resort environment is quite special, with factors such as a lot of snow, complex terrain, and frequent personnel and equipment movement, which puts high demands on the automatic parking of transport vehicles. In such a scenario, the relevant data of historical transport vehicles is particularly critical for optimizing the automatic parking algorithm.

[0118] First, the first training error is determined based on the prior basis, parking path planning label and prior confidence of each historical transport vehicle. For each historical transport vehicle that performs the automatic parking task at the ski resort, the prior basis reflects whether there is a path intersection with the automatic parking control strategy during the automatic parking process. In the ski resort parking lot, since the parking space layout may be relatively compact, and there may be snow piles, temporary stacking of ski equipment and irregular parking of other vehicles around, the existence of path intersections is directly related to the safety and efficiency of parking. The parking planning label indicates the probability that the vehicle matches the path intersection. For example, the prior basis of a historical transport vehicle indicates that there is a path intersection, and its parking path planning label has a probability of 0.55 that there is a path intersection, and the prior confidence is 0.65.

[0119] When determining the first training error, it is necessary to operate according to a specific algorithm logic. First, an intermediate value is determined based on the relationship between the prior basis and the parking path planning label. This intermediate value is related to the specific values ​​and relationship between the prior basis and the parking path planning label, and is obtained according to the calculation rules within the algorithm. In the ski resort scenario, the calculation of this intermediate value needs to take into account the impact of the special environment of the ski resort. For example, due to snow cover, there may be a certain deviation in the sensor's measurement of the distance of surrounding objects. This deviation will be reflected in the numerical relationship between the prior basis and the parking path planning label. Then, this intermediate value and the prior confidence are calculated according to the prescribed algorithm to obtain an error-related value corresponding to each historical transport vehicle. Finally, the first training error is determined by comprehensively processing these error-related values ​​of all historical transport vehicles (for example, summing, weighted averaging, etc., determined by the internal logic of the algorithm).

[0120] Next, the hyperparameter data is determined based on the algorithm weights of the Transformer algorithm. The hyperparameter data characterizes the long-distance dependency performance of the Transformer algorithm. In the automatic parking scenario at a ski resort, the algorithm weights in the Transformer algorithm are a key part of the algorithm's internal structure. These weights determine how the algorithm processes input data and how to transfer information between different neurons or modules. For example, in a ski resort parking lot, a transport vehicle may need to drive from a farther entrance to the parking location, during which it needs to process environmental information over a longer distance. In the multi-head attention mechanism part of the Transformer algorithm, different heads may have different weights, which affect the degree of attention paid to different representation subspaces of the input. By performing specific analysis and processing on these algorithm weights, hyperparameter data can be obtained. This hyperparameter data reflects the performance of the algorithm in processing long-distance dependencies. Long-distance dependencies are very important when processing multi-sensor data such as in the automatic parking scenario at a ski resort, because the vehicle's state and surrounding environment information may have complex associations at different time and space scales. For example, a snow pile in the distance or other vehicles in motion may affect the parking path of the current transport vehicle, and the algorithm needs to accurately process this long-distance association information.

[0121] Then, the third training error is determined based on the hyperparameter data and the number of historical transport vehicles. The hyperparameter data, as an indicator reflecting the performance of the algorithm's long-distance dependency, is combined with the number of historical transport vehicles to determine the third training error. In the automatic parking scenario of a ski resort, if the hyperparameter data shows that the algorithm has a certain deviation in the processing of long-distance dependencies, and the number of historical transport vehicles is large, this may mean that this deviation will be amplified in the process of processing a large amount of data. For example, there may be a large number of transport vehicles in a ski resort that perform parking tasks at different time periods and accumulate a lot of historical data. When there is a deviation in the algorithm's processing of long-distance dependencies, as the number of historical transport vehicles increases, this deviation will produce a larger value when determining the third training error. The specific determination method is to follow the logic set inside the algorithm, which may involve some mathematical operations or logical judgments on the hyperparameter data and the number of historical transport vehicles to obtain a third training error that can reflect the impact of this relationship on algorithm debugging.

[0122] Finally, the Transformer algorithm is debugged based on the first training error and the third training error to obtain the intelligent driving control processing algorithm. The first training error and the third training error are both important bases for debugging the Transformer algorithm. The first training error reflects the deviation of the algorithm in processing vehicle automatic parking related information from the perspective of the prior basis of historical transport vehicles, parking path planning labels and prior confidence, while the third training error reflects another deviation of the algorithm from the perspective of the Transformer algorithm's own long-distance dependency performance and the number of historical transport vehicles.

[0123] During the debugging process of automatic parking at a ski resort, if the first training error is large, it means that there are major problems in processing the relevant parking information of historical transport vehicles. In the special environment of the ski resort, it may be necessary to adjust the calculation part of the algorithm related to the prior basis, parking path planning label and prior confidence. For example, since snow accumulation may affect the vehicle's driving trajectory and the detection of surrounding objects, when calculating the parking path planning label, it is necessary to adjust the weights of factors such as the distance of objects around the vehicle (such as the distance of obstacles partially covered by snow), the vehicle speed (considering the special situation of driving on snow), or adjust the calculation method of the prior confidence. If the third training error is large, it may be necessary to adjust the weights in the Transformer algorithm, especially the weights related to the processing of long-distance dependencies. For example, adjust the weight distribution of different heads in the multi-head attention mechanism, or adjust the weights of different position information when processing sequence data, so as to optimize the algorithm's long-distance dependency processing ability. By comprehensively considering these two training errors, the Transformer algorithm is fully debugged, and finally an intelligent driving control processing algorithm that can more accurately handle the automatic parking of the transport vehicle to be controlled at the ski resort is obtained.

[0124] By combining the first training error and the third training error to debug the Transformer algorithm, we can see from the numerical examples that the parking path planning label probability of the historical transport vehicle, the prior confidence, and the hyperparameter data related to the weight of the Transformer algorithm are all involved. In the automatic parking scenario at the ski resort, this helps to improve the accuracy of the intelligent driving control processing algorithm, so that the algorithm can better handle the complex relationships in automatic parking at the ski resort, such as the impact of snow accumulation, irregular parking, long-distance driving, etc., reduce parking errors, improve the reliability and efficiency of the parking system, and enhance the adaptability to different parking situations at the ski resort.

[0125] In other exemplary technical schemes, the Transformer algorithm is debugged based on the prior basis, parking path planning labels and prior confidence of each historical transport vehicle to obtain the intelligent driving control processing algorithm, including: determining the first training error based on the prior basis, parking path planning labels and prior confidence of each historical transport vehicle; determining the fourth training error based on the prior confidence of each historical transport vehicle and the number of historical transport vehicles; debugging the Transformer algorithm based on the first training error and the fourth training error to obtain the intelligent driving control processing algorithm.

[0126] In detail, the above technical solution revolves around debugging the Transformer algorithm based on the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence to obtain the intelligent driving control processing algorithm. The special environment of the ski resort brings many challenges to automatic parking. The layout of the parking lot may become complicated due to the accumulation of snow piles, the temporary parking of ski equipment, etc., and the snow covering the ground will affect the driving of the vehicle and the perception ability of the sensor. In such a scenario, debugging the Transformer algorithm based on the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence is the key to optimizing the automatic parking system.

[0127] First, the first training error is determined based on the prior basis, parking path planning label and prior confidence of each historical transport vehicle. For each historical transport vehicle that performs an automatic parking task at a ski resort, the prior basis is a representation of whether there is a path intersection with the automatic parking control strategy during the automatic parking process. In a ski resort, the determination of path intersections is more complicated because in addition to conventional vehicles and parking facilities, there may also be factors such as skiers passing through and snow piles blocking the path. The parking path planning label represents the probability that the vehicle matches the path intersection, and the prior confidence reflects the reliability of the vehicle's initial vehicle environment state vector. For example, the prior basis of a historical transport vehicle indicates that there is a path intersection, and its parking path planning label has a probability of 0.6 that there is a path intersection, and the prior confidence is 0.7.

[0128] When determining the first training error, it is necessary to follow a specific algorithm logic. First, an intermediate result is obtained based on the relationship between the prior basis and the parking path planning label. This intermediate result is related to the specific values ​​and mutual relationship of the prior basis and the parking path planning label, and is calculated according to the internal rules of the algorithm. In the ski resort environment, due to the influence of snow on the sensor, for example, the ultrasonic sensor may produce deviations in measuring the distance in the snow environment, which will be reflected in the numerical relationship between the prior basis and the parking path planning label, thereby affecting the calculation of the intermediate result. Then this intermediate result and the prior confidence are calculated according to the prescribed algorithm to obtain the error-related value corresponding to each historical transport vehicle. For example, if the intermediate result is a certain value, it is calculated according to the algorithm with the prior confidence of 0.7 to obtain a specific error-related value. Finally, these error-related values ​​of all historical transport vehicles are comprehensively processed, and the processing method is determined by the internal logic of the algorithm, such as summation or weighted average, so as to determine the first training error.

[0129] Next, the fourth training error is determined based on the prior confidence of each historical transport vehicle and the number of historical transport vehicles. The prior confidence reflects information such as the reliability of the initial vehicle environment state vector of each historical transport vehicle. In the automatic parking scenario at the ski resort, the initial vehicle environment state vector of the historical transport vehicle contains more special factors. For example, the location coordinates of the vehicle may be difficult to accurately determine because the landmarks are covered with snow, and the speed of the vehicle may be unstable due to snowy road conditions. For example, there are multiple historical transport vehicles, and the prior confidence of some of the historical transport vehicles is 0.6, 0.7, 0.8, etc. The number of historical transport vehicles is also an important factor in this process. If the prior confidence is generally high and the number of historical transport vehicles is large, or the prior confidence is generally low and the number of historical transport vehicles is large, it will have different effects on the fourth training error. The specific method of determining the fourth training error is based on the logic set within the algorithm, which may involve performing some mathematical operations on the prior confidence (such as summing, averaging, etc.) and performing logical judgment based on the number of historical transport vehicles, thereby obtaining a fourth training error that can reflect the impact of this relationship on algorithm debugging.

[0130] Finally, the Transformer algorithm is debugged based on the first training error and the fourth training error to obtain the intelligent driving control processing algorithm. The first training error and the fourth training error are both important bases for debugging the Transformer algorithm. The first training error reflects the deviation of the algorithm in processing the information related to the automatic parking of the vehicle from the perspective of the prior basis of the historical transport vehicle, the parking path planning label and the prior confidence. For example, in a ski resort environment, if the first training error is large, it may mean that there are major problems in processing the logical relationship between the prior basis of the historical transport vehicle, the parking path planning label and the prior confidence. Due to the special conditions of the ski resort, such as the interference of snow on the sensor and the unpredictability of skiers, these relationships may become more complicated and difficult to handle accurately.

[0131] The fourth training error reflects another deviation of the algorithm from the perspective of the prior confidence of historical transport vehicles and the number of historical transport vehicles. During the debugging process, if the first training error is large, it may be necessary to adjust the calculation part of the algorithm related to the prior basis, parking path planning label and prior confidence. For example, when calculating the parking path planning label, the weight of factors such as the distance of objects around the vehicle and the speed of the vehicle, or adjust the calculation method of the prior confidence. In the ski resort scenario, the measurement of the distance of objects around the vehicle may be inaccurate due to snow accumulation, and the control of vehicle speed also needs to consider factors such as the friction of the snow. If the fourth training error is large, it may be necessary to adjust the part of the algorithm related to the prior confidence and the number of historical transport vehicles, such as adjusting the algorithm logic when considering the ratio of the sum of prior confidence and the number. By comprehensively considering these two training errors, the Transformer algorithm is fully debugged, and finally an intelligent driving control processing algorithm that can more accurately handle the automatic parking of the transport vehicle to be controlled at the ski resort is obtained.

[0132] By combining the first training error and the fourth training error to debug the Transformer algorithm, we can see from the numerical example that the parking path planning label probability and prior confidence of the historical transport vehicle are all involved. In the automatic parking scenario at the ski resort, this helps to improve the accuracy of the intelligent driving control processing algorithm, so that the algorithm can better handle the complex situations in automatic parking at the ski resort, such as the environmental complexity caused by snow accumulation affecting sensors, skiers and snow piles, etc., thereby reducing parking errors, improving the reliability and efficiency of the parking system, and enhancing the adaptability to different parking scenarios and vehicle conditions at the ski resort.

[0133] In summary, the embodiments of the present invention can fully grasp the surrounding environment of the ski resort transport vehicle by acquiring a variety of IoT sensor information. The fusion of ultrasonic sensor information, geomagnetic sensor information, visual sensor information and lidar sensor information can accurately reflect the vehicle status. The initial vehicle environment state vector is determined using the intelligent driving control processing algorithm to provide an accurate basis for subsequent decision-making. The determination of the parking path planning label helps to assess the risk of intersection with the control strategy path in advance. The decision whether to issue the automatic parking control strategy based on the label can improve parking safety and efficiency, avoid unnecessary collision risks, optimize parking lot resource utilization, and adapt to the complex and changeable parking needs of ski resorts.

[0134] An embodiment of the present invention provides a transport vehicle, which is communicatively connected to an Internet of Things sensor control system; the transport vehicle is used to receive an automatic parking control strategy issued by the Internet of Things sensor control system; the Internet of Things sensor control system includes at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the above method.

[0135] Further, Figure 2 FIG. 2 is a schematic diagram of a structure of an IoT sensor control system 200 provided in an embodiment of the present invention. Figure 2 The IoT sensor control system 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method in the embodiment of the present invention.

[0136] Alternatively, if Figure 2 As shown, the IoT sensor control system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present invention.

[0137] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0138] Alternatively, if Figure 2 As shown, the IoT sensor control system 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, may send information or data to other devices, or receive information or data sent by other devices.

[0139] Optionally, the IoT sensor control system 200 can implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present invention. For the sake of brevity, they are not repeated here.

[0140] It should be understood that the processor of the embodiment of the present invention may be an integrated circuit chip with signal processing capability.

[0141] It is understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0142] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0143] The embodiments of the embodiments of the present invention are described above in conjunction with the accompanying drawings, but the embodiments of the present invention are not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the embodiments of the present invention, ordinary technicians in this field can also make many forms without departing from the purpose of the embodiments of the present invention and the scope of protection of the embodiments of the present invention, all of which are within the protection of the embodiments of the present invention.

Claims

1. A ski resort automatic parking control method based on IoT sensing, characterized in that: The method is applied to an IoT sensor control system, and the method comprises: Acquire first IoT sensor information of the transport vehicle to be controlled, where the transport vehicle to be controlled is a transport vehicle in a ski resort area; the first IoT sensor information includes ultrasonic sensor information, geomagnetic sensor information, visual sensor information, and laser radar sensor information; Using an intelligent driving control processing algorithm to process the first IoT sensor information of the transport vehicle to be controlled, so as to determine an initial vehicle environment state vector of the transport vehicle to be controlled; The initial vehicle environment state vector of the transport vehicle to be controlled is processed by using the intelligent driving control processing algorithm to determine a parking path planning label of the transport vehicle to be controlled, wherein the parking path planning label of the transport vehicle to be controlled represents the probability that the transport vehicle to be controlled has a path intersection with the automatic parking control strategy during the automatic parking process; According to the parking path planning tag of the transport vehicle to be controlled, it is determined whether to issue the automatic parking control strategy to the transport vehicle to be controlled.

2. The method according to claim 1, characterized in that The parking path planning tag of the transport vehicle to be controlled includes a first parking path planning tag and a second parking path planning tag, wherein the first parking path planning tag represents the probability that the transport vehicle to be controlled matches the path intersection on the basis that the automatic parking control strategy is issued to the transport vehicle to be controlled, and the second parking path planning tag represents the probability that the transport vehicle to be controlled matches the path intersection on the basis that the automatic parking control strategy is not issued to the transport vehicle to be controlled; The step of determining whether to issue the automatic parking control strategy to the transport vehicle to be controlled based on the parking path planning tag of the transport vehicle to be controlled includes: If the first discrimination mechanism and the second discrimination mechanism are met, the automatic parking control strategy is issued to the transport vehicle to be controlled. The first discrimination mechanism is met when the weight of the first parking path planning label is greater than the weight of the second parking path planning label. The second discrimination mechanism is met when the weight difference between the first parking path planning label and the second parking path planning label is greater than the set weight. If at least one of the first discrimination mechanism and the second discrimination mechanism is not met, it is determined not to issue the automatic parking control strategy to the transport vehicle to be controlled.

3. The method according to claim 1 or 2, characterized in that The method further comprises: Acquire second IoT sensor information of multiple historical transport vehicles and a priori basis of each historical transport vehicle, wherein the a priori basis of the historical transport vehicle indicates whether the historical transport vehicle has a path intersection related to the automatic parking control strategy during the automatic parking process; Using the Transformer algorithm to mine state features of the second IoT sensor information of the plurality of historical transport vehicles, to obtain an initial vehicle environment state vector of each of the historical transport vehicles; Determine the parking path planning label of each historical transport vehicle according to the initial vehicle environment state vector of each historical transport vehicle by using the Transformer algorithm, wherein the parking path planning label of the historical transport vehicle represents the probability that the historical transport vehicle matches the path intersection; Determining the prior confidence of each historical transport vehicle according to the initial vehicle environment state vector of each historical transport vehicle, wherein the prior confidence of the historical transport vehicle is used to adjust the initial vehicle environment state vector of the historical transport vehicle to a target mode; According to the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence, the Transformer algorithm is debugged to obtain an intelligent driving control processing algorithm, which is used to determine the probability that the transport vehicle to be controlled matches the path intersection.

4. The method according to claim 3, characterized in that The plurality of historical transport vehicles include a plurality of first transport vehicles to which the automatic parking control strategy is issued and a plurality of second transport vehicles to which the automatic parking control strategy is not issued; The method of using the Transformer algorithm to determine the parking path planning label of each historical transport vehicle according to the initial vehicle environment state vector of each historical transport vehicle includes: Determining a parking path planning label of each of the first transport vehicles using the Transformer algorithm based on a first debugging annotation and the initial vehicle environment state vectors of the plurality of first transport vehicles, wherein the first debugging annotation represents a result of the issued automatic parking control strategy; The parking path planning label of each second transport vehicle is determined by using the Transformer algorithm based on a second debugging annotation and the initial vehicle environment state vectors of the plurality of second transport vehicles, wherein the second debugging annotation represents a result of an automatic parking control strategy not being issued.

5. The method according to claim 3, characterized in that Determining the prior confidence of each historical transport vehicle based on the initial vehicle environment state vector of each historical transport vehicle includes: Performing feature mapping on the initial vehicle environment state vectors of each of the historical transport vehicles to obtain the driving control decision mapping vectors of each of the historical transport vehicles; Determining the prior confidence of each of the historical transport vehicles according to the driving control decision mapping vectors of each of the historical transport vehicles; The driving control decision mapping vector of the historical transport vehicle includes local control decision mapping vectors of the historical transport vehicle in multiple attention indicators; Determining the prior confidence of each historical transport vehicle according to the driving control decision mapping vector of each historical transport vehicle includes: For any historical transport vehicle, determining a first mapping vector comparison result according to a local control decision mapping vector of any historical transport vehicle at a first attention index and a local control decision mapping vector of each historical transport vehicle at the first attention index, wherein the first attention index is any one of the multiple attention indexes; Determining a second mapping vector comparison result according to a local control decision mapping vector of any one of the historical transport vehicles at a second attention index and a local control decision mapping vector of each of the historical transport vehicles at the second attention index, wherein the second attention index is any one of the multiple attention indexes except the first attention index; Determining a local control decision mapping correlation between the first attention index and the second attention index for any historical transport vehicle according to the comparison result of the first mapping vector and the comparison result of the second mapping vector; Determining the prior confidence of each historical transport vehicle according to the local control decision mapping correlation between each two attention indicators of each historical transport vehicle; Determining a first mapping vector comparison result based on the local control decision mapping vector of any one of the historical transport vehicles in the first attention index and the local control decision mapping vectors of each of the historical transport vehicles in the first attention index includes: Determining a local control decision mapping mean vector of the first attention index according to the basic confidence of each of the historical transport vehicles and the local control decision mapping vector of each of the historical transport vehicles in the first attention index; Determine a local control decision mapping reinforcement vector of any one of the historical transport vehicles in the first attention index according to a basic confidence of any one of the historical transport vehicles and a local control decision mapping vector of any one of the historical transport vehicles in the first attention index; The vector operation value between the local control decision mapping enhancement vector of the first attention index and the local control decision mapping mean vector of the first attention index of any historical transport vehicle is used as the first mapping vector comparison result.

6. The method according to claim 3, characterized in that The Transformer algorithm is debugged based on the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence to obtain the intelligent driving control processing algorithm, including: For any historical transport vehicle, determine the first path planning node error corresponding to any historical transport vehicle according to the prior basis of any historical transport vehicle and the parking path planning label of any historical transport vehicle, and determine the first path planning node weighted error of any historical transport vehicle according to the first path planning node error corresponding to any historical transport vehicle and the prior confidence of any historical transport vehicle; Determining a first training error according to the weighted errors of the first path planning nodes of each of the historical transport vehicles; The Transformer algorithm is debugged according to the first training error to obtain the intelligent driving control processing algorithm.

7. The method according to claim 3, characterized in that The plurality of historical transport vehicles include a plurality of first transport vehicles to which the automatic parking control strategy is issued and a plurality of second transport vehicles to which the automatic parking control strategy is not issued; The Transformer algorithm is debugged based on the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence to obtain the intelligent driving control processing algorithm, including: Determining a first training error according to the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence; Determining a second training error according to a comparison result of the initial state vectors between the initial vehicle environment state vectors of the plurality of first transport vehicles and the initial vehicle environment state vectors of the plurality of second transport vehicles; Based on the first training error and the second training error, the Transformer algorithm is debugged to obtain the intelligent driving control processing algorithm.

8. The method according to claim 3, characterized in that The Transformer algorithm is debugged based on the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence to obtain the intelligent driving control processing algorithm, including: Determining a first training error according to the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence; Determining hyperparameter data according to the algorithm weight of the Transformer algorithm, wherein the hyperparameter data characterizes the long-distance dependency performance of the Transformer algorithm; Determining a third training error according to the hyperparameter data and the number of historical transport vehicles; Based on the first training error and the third training error, the Transformer algorithm is debugged to obtain the intelligent driving control processing algorithm.

9. The method according to claim 3, characterized in that The Transformer algorithm is debugged based on the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence to obtain the intelligent driving control processing algorithm, including: Determining a first training error according to the prior basis of each historical transport vehicle, the parking path planning label and the prior confidence; Determining a fourth training error according to the prior confidence of each of the historical transport vehicles and the number of the historical transport vehicles; Based on the first training error and the fourth training error, the Transformer algorithm is debugged to obtain the intelligent driving control processing algorithm.

10. A transport vehicle, characterized in that: The transport vehicle is communicatively connected to the Internet of Things sensor control system; the transport vehicle is used to receive the automatic parking control strategy issued by the Internet of Things sensor control system; the Internet of Things sensor control system includes at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the method described in any one of claims 1-9.

Images