Task scheduling method, electronic equipment, vehicle and storage medium

Abstract

The embodiment of the invention provides a task scheduling method, electronic equipment, a vehicle and a storage medium, and the method comprises the steps: obtaining the multi-mode environment characteristics of terminal equipment, the load characteristics of the terminal equipment, the state information of an edge node cluster, and the information of a task queue, the multi-modal environment characteristics comprise at least one of terrain complexity, communication stability and positioning precision; inputting the multi-modal environment characteristics, the terminal equipment load characteristics, the edge node cluster state information and the task queue information into a task scheduling model, and determining a scheduling strategy of each task in a task queue, the scheduling strategy of each task indicates that each task is executed on the terminal equipment or each task is unloaded to an edge node in an edge node cluster. According to the embodiment of the invention, the scheduling strategy of the task is determined based on the fused various types of environment characteristics, and the real-time performance, the resource utilization rate and the scene adaptability of task processing are effectively improved.

Description

Technical Field

[0001] This application relates to the field of edge computing technology, and in particular to a task scheduling method, electronic device, vehicle, and storage medium. Background Technology

[0002] With the development of the low-altitude economy, the application of low-altitude equipment such as drones and low-altitude communication terminals is becoming increasingly widespread. Low-altitude equipment is affected by factors such as power consumption and communication stability, and typically collaborates with edge devices to handle low-altitude tasks. Task scheduling between low-altitude equipment and edge devices is usually based on priority; for example, high-priority tasks are scheduled for low-altitude equipment, and low-priority tasks are scheduled for edge devices. This relatively singular scheduling criterion may lead to low resource utilization and an inability to adapt to the operating environment of low-altitude equipment. Summary of the Invention

[0003] In view of the above, it is necessary to provide a task scheduling method, electronic device, vehicle and storage medium to solve the problem that task scheduling between the low-altitude equipment and the edge terminal is usually based on priority, the basis of task scheduling is relatively simple, which may lead to low resource utilization and inability to adapt to the operating environment of low-altitude equipment.

[0004] In a first aspect, embodiments of this application provide a task scheduling method, the method comprising: The system acquires multimodal environment characteristics of the terminal device, terminal device load characteristics, edge node cluster status information, and task queue information. The multimodal environment characteristics include at least one of terrain complexity, communication stability, and positioning accuracy. The multimodal environment characteristics, terminal device load characteristics, edge node cluster status information, and task queue information are input into the task scheduling model to determine the scheduling strategy for each task in the task queue. The scheduling strategy for each task instructs each task to be executed on the terminal device or to be offloaded to an edge node in the edge node cluster.

[0005] In one possible implementation, the multimodal environment data of the terminal device includes at least one of point cloud data, attitude data, communication data, state data, and positioning data. Obtaining the multimodal environment features of the terminal device includes: The statistical features of the point cloud data are extracted, and the statistical features are normalized to obtain the terrain complexity. A communication quality score is calculated based on the communication data, and the communication quality score is normalized to obtain the communication stability. The deviation data of the positioning data is determined as the positioning accuracy.

[0006] In one possible implementation, the multimodal environment features also include obstacle density, IMU three-attitude angles, wind speed, weather conditions, task type, and task time window.

[0007] In one possible implementation, the multimodal environment features are multi-dimensional feature vectors, where each dimension of the feature vector has a corresponding weight. The method further includes: In the time dimension, a self-attention mechanism is used to set the weights corresponding to the features in each dimension, including: Calculate the query matrix, key matrix, and value matrix corresponding to the features of each dimension based on the weight matrices WQ, WK, and WV; Based on the query matrix, the key matrix, and the value matrix, the weights corresponding to the features of each dimension are calculated.

[0008] In one possible implementation, the multimodal environment features are multi-dimensional feature vectors, where each dimension of the feature vector has a corresponding weight. The method further includes: In the spatial dimension, the weights corresponding to the features in each dimension are set using a cross-attention mechanism, including: The query matrix corresponding to the feature of each dimension is calculated based on the weight matrix WQ. Calculate the key matrix and value matrix corresponding to the task type based on the weight matrices WK and WV; Based on the query matrix, the key matrix, and the value matrix, the weights corresponding to the features of each dimension are calculated.

[0009] In one possible implementation, the edge node cluster status includes remaining computing power and a first communication latency. The step of inputting the multimodal environment characteristics, the terminal device load characteristics, the edge node cluster status information, and the task queue information into a task scheduling model to determine the scheduling strategy for each task in the task queue includes: The task scheduling model determines the second communication delay for offloading each task to the edge node based on the weighted multimodal environment features and the first communication delay of the edge node cluster. The task scheduling model also determines the computation latency based on the remaining computing power of the edge node cluster and the task queue information; The first weight of the second communication delay and the second weight of the calculated delay are determined based on the communication stability. The total latency for offloading each task to the edge node is determined based on the second communication latency, the first weight, the calculation latency, and the second weight. The edge node with the minimum total latency is selected as the edge node for processing each task.

[0010] In one possible implementation, the method further includes: Create a low-priority task queue based on low-priority tasks; Calculate the rate of change of multimodal environmental characteristics over multiple consecutive preset periods; Calculate the priority value of each low-priority task in the low-priority task queue based on the environmental change rate; Based on the priority value of each low-priority task, the multiple low-priority tasks in the low-priority task queue are reordered.

[0011] Secondly, embodiments of this application provide an electronic device, which includes a memory and a processor; wherein the memory is used to store program instructions; and the processor is used to read and execute the program instructions stored in the memory, such that when the program instructions are executed by the processor, the electronic device performs the task scheduling method described above.

[0012] Thirdly, embodiments of this application provide a vehicle, the vehicle including a memory and a processor: wherein the memory is used to store program instructions; the processor is used to read and execute the program instructions stored in the memory, and when the program instructions are executed by the processor, the vehicle performs the above-described task scheduling method.

[0013] Fourthly, embodiments of this application provide a task scheduling device, the device comprising: an acquisition module, configured to acquire multimodal environment characteristics of a terminal device, load characteristics of the terminal device, status information of an edge node cluster, and task queue information, wherein the multimodal environment characteristics include at least one of terrain complexity, communication stability, and positioning accuracy; and a determination module, configured to input the multimodal environment characteristics, the load characteristics of the terminal device, the status information of the edge node cluster, and the task queue information into a task scheduling model, and determine a scheduling strategy for each task in the task queue, wherein the scheduling strategy for each task instructs each task to be executed on the terminal device or to be offloaded to an edge node in the edge node cluster.

[0014] Fifthly, embodiments of this application provide a computer storage medium that, when the program instructions are run on an electronic device or a vehicle, causes the processor of the electronic device or the vehicle to execute the above-described task scheduling method.

[0015] The task scheduling method, apparatus, electronic device, vehicle, and storage medium provided in this application determine the task scheduling strategy based on the fusion of multiple types of environmental features, thereby achieving accurate task scheduling and effectively improving the real-time performance, resource utilization, and scene adaptability of task processing. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This is a schematic diagram illustrating an application scenario of the task scheduling method provided in an embodiment of this application.

[0018] Figure 2 This is a flowchart of a task scheduling method provided in an embodiment of this application.

[0019] Figure 3 This is a schematic diagram of the structure of a task scheduling method apparatus provided in an embodiment of this application.

[0020] Figure 4 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application.

[0021] Figure 5 This is a schematic diagram of the hardware structure of a vehicle provided in one embodiment of this application. Detailed Implementation

[0022] The terms "first" and "second" used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this application's specification is for the purpose of describing particular embodiments only and is not intended to limit the application. It should be understood that, unless otherwise stated, " / " in this application means "or". For example, A / B can mean A or B. "And / or" in this application is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. "At least one" refers to one or more. "More than one" refers to two or more. For example, at least one of a, b, or c can represent: a, b, c, a and b, a and c, b and c, and a, b, and c. Where there is no conflict, the following embodiments and features described herein can be combined with each other.

[0024] With the development of the low-altitude economy, the application of low-altitude equipment such as drones and low-altitude communication terminals is becoming increasingly widespread. Low-altitude equipment is affected by factors such as power consumption and communication stability, and typically collaborates with edge devices to handle low-altitude tasks. Task scheduling between low-altitude equipment and edge devices is usually based on priority; for example, high-priority tasks are scheduled for low-altitude equipment, and low-priority tasks are scheduled for edge devices. This relatively singular scheduling criterion may lead to low resource utilization and an inability to adapt to the operating environment of low-altitude equipment.

[0025] To address the issue that task scheduling between low-altitude equipment and edge devices is typically based on priority, resulting in a single scheduling criterion and potentially low resource utilization, and an inability to adapt to the operating environment of low-altitude equipment, this application provides a task scheduling method. This method can determine the task scheduling strategy based on a combination of various environmental characteristics, achieving precise task scheduling and effectively improving the real-time performance, resource utilization, and scene adaptability of task processing.

[0026] See Figure 1 The diagram illustrates an application scenario of a task scheduling method provided in an embodiment of this application. The task scheduling provided in this embodiment can be applied to a task scheduling system 100, which includes, but is not limited to, an electronic device 1, a terminal device 2, and an edge node cluster 3. The edge node cluster 3 includes multiple edge nodes 31, and the electronic device 1, terminal device 2, and multiple edge nodes 31 are connected for communication. The electronic device 1 can be a server or a vehicle; for example, the server can be a cloud server, a server cluster, or a single server, used to create tasks and perform task scheduling. The terminal device 2 is a low-altitude device, such as a drone or a low-altitude communication terminal, used to execute high-priority tasks. The edge nodes 31 can be vehicle-mounted devices, mobile devices, servers, etc., used to execute low-priority tasks.

[0027] See Figure 2 The diagram shown is a flowchart of a task scheduling method provided in an embodiment of this application. The method is applied in an electronic device and includes: S101, acquire the multimodal environment characteristics of the terminal device, the load characteristics of the terminal device, the status information of the edge node cluster, and the task queue information.

[0028] In one embodiment of this application, multimodal environmental data of the terminal device is collected, and multimodal environmental features of the multimodal environmental data are extracted. For example, multimodal environmental features include at least one of terrain complexity, communication stability, positioning accuracy, obstacle density, IMU three-attitude angles, wind speed, weather conditions, task type, and task time window. The multimodal environmental features and the terminal device load features constitute a multi-dimensional feature vector, for example, a 12-dimensional feature vector.

[0029] In one embodiment of this application, the multimodal environment data includes at least one of point cloud data, attitude data, communication data, and positioning data. Specifically, the point cloud data is the lidar point cloud of the terminal device, with a sampling rate of 10Hz and a resolution of 0.1 meters. The attitude data is IMU (Inertial Measurement Unit) attitude data, including pitch angle, roll angle, yaw angle, etc., with a sampling rate of 100Hz and an error of ±0.1°. The communication data is 5G communication parameters, including signal-to-noise ratio (SNR), packet loss rate, etc., with a sampling rate of 50Hz. The positioning data is BeiDou-5G fused positioning data, providing centimeter-level positioning and microsecond-level timing.

[0030] In one embodiment of this application, multimodal environmental data is preprocessed. This preprocessing includes at least one of time synchronization, spatial unification, and noise reduction. Based on BeiDou-5G microsecond-level timing, multiple types of environmental data are timestamped, with an error of less than 0.1 seconds, resolving the "spatiotemporal misalignment" problem. Point cloud data and attitude data are mapped to the WGS84 coordinate system, and point cloud offsets caused by terminal device tilt are corrected based on IMU attitude data. For example, when the UAV pitch angle is 5°, longitudinal errors in the point cloud data are compensated in reverse to ensure data alignment with the actual terrain or location. A Gaussian filtering algorithm (e.g., with a window radius of 0.5 meters) is used to filter out noisy data in the point cloud data, and a Kalman filtering algorithm (with a state equation based on the second derivative of acceleration / angular velocity) is used to smooth the drift error of the attitude data.

[0031] In one embodiment of this application, statistical features of point cloud data are extracted and normalized to obtain terrain complexity. Specifically, the variance and extreme value difference of the height difference of lidar point clouds within the same spatial region are calculated, and the variance and extreme value difference are normalized and mapped to the range [0,1]. Higher variance and extreme value difference indicate more complex terrain.

[0032] In one embodiment of this application, a communication quality score is calculated based on communication data, and the communication quality score is normalized to obtain the communication stability. For example, the communication quality score = (normalized SNR value × 0.6 + (1 - normalized packet loss rate value) × 0.4).

[0033] In one embodiment of this application, the deviation data of the positioning data is determined as the positioning accuracy. For example, the BeiDou-5G positioning deviation (usually less than 10cm) is directly used as a feature, and when the deviation is >8cm, it is marked as "low accuracy state".

[0034] In one embodiment of this application, status data of the terminal device is collected, including CPU utilization, remaining battery power, etc., with a sampling rate of 10Hz. Load characteristics of the terminal device are extracted based on the status data. Specifically, the status data is normalized to obtain the device load characteristics. For example, CPU utilization and remaining battery power are normalized to [0,1], and a weighted sum of CPU utilization and remaining battery power is performed to obtain the device load characteristics, where the sum of the weights of CPU utilization and remaining battery power is 1, for example, 0.5 for both.

[0035] In one embodiment of this application, weights for multi-dimensional feature vectors are set using an attention mechanism based on time and spatial dimensions. Specifically, in the time dimension, a self-attention mechanism is used to set the weights corresponding to each feature dimension. The query matrix, key matrix, and value matrix corresponding to each feature dimension are calculated based on preset weight matrices WQ, WK, and WV. Specifically, the multi-dimensional feature vector sequence spanning T consecutive time periods is denoted as X = [x1, x2, …, x…]. t ], where each element x in the sequence t ∈R d Let represent the d-dimensional eigenvector of the t-th period. The query matrix Q = X * WQ, where WQ ∈ R. (d×d_k) The bond matrix K = X * WK, where WK ∈ R (d×d_k) The value matrix V = X * WV, where WV ∈ R (d×d_v) .

[0036] Based on the query matrix, key matrix, and value matrix, the weights corresponding to the features in each dimension are calculated. Specifically, the attention score matrix A = softmax(QK^T / √d_k) is calculated, where the softmax function normalizes row by row to obtain the attention weight distribution of each time period to other periods. The attention score matrix is ​​multiplied by the value matrix to obtain the weighted feature representation Z = AV ∈ R^(T×d_v). The average of Z along the time dimension or the output of the last time step is taken to obtain the fused feature vector z ∈ R^d_v. The fused feature vector z is mapped to the weight vector w ∈ R^d of each dimension feature through a fully connected layer, where each element w_i corresponds to the weight of the i-th dimension feature. w is normalized so that the sum of the weights of each dimension feature is 1.

[0037] For example, in the time dimension, a two-head self-attention mechanism is used to strengthen the weight of strongly correlated data within three adjacent cycles (e.g., 0.3s). For instance, if the communication quality score decreases by more than 20% for three consecutive cycles, the weight of the communication stability feature increases from 0.1 to 0.4; if the IMU attitude data changes by less than 0.5° within 500ms, the attitude feature weight decreases to 0.05.

[0038] In the spatial dimension, a cross-attention mechanism is used to set the weights corresponding to the features in each dimension. Specifically, the query matrix corresponding to the features in each dimension is calculated based on the weight matrix WQ. Specifically, the current multi-dimensional feature vector x∈R^d and the task type embedding vector c∈R^(m×d_c) are obtained, where m is the number of task types. The multi-dimensional feature vector x is expanded into a query matrix Q'=x*WQ, where WQ∈R^(d×d_k) and Q'∈R^(1×d_k).

[0039] The key matrix and value matrix corresponding to the task type are calculated based on the weight matrices WK and WV. Specifically, the key matrix K' = c * WK, where c ∈ R^(m × d_c) is the embedding matrix for all task types, WK ∈ R^(d_c × d_k), and K' ∈ R^(m × d_k). The value matrix V' = c * W_V, where WV ∈ R^(d_c × d_v), and V' ∈ R^(m × d_v).

[0040] Based on the query matrix, key matrix, and value matrix, the weights corresponding to the features in each dimension are calculated. Specifically, the attention score a = softmax(Q'K'^T / √d_k) ∈ R^(1×m) is calculated to obtain the attention distribution of the adjacent pairs of multi-dimensional features for each task type. The context vector h = aV' ∈ R^(1×d_v) is calculated, where the context vector h represents the feature representation weighted according to the current task type. The context vector h is concatenated or residually connected with the environment feature vector x to obtain the enhanced feature representation x' = [x; h] ∈ R^(d+d_v). The enhanced feature representation x' is mapped to the weight vector w' ∈ R^d of each dimension feature through a fully connected layer and a softmax function, where each element w'i corresponds to the weight of the i-th dimension feature, and the sum of the weights of each dimension feature is 1.

[0041] For example, for obstacle avoidance tasks (priority P1), the weight of lidar obstacle density is set to 0.3, terrain complexity to 0.2, communication stability to 0.1, IMU attitude angle to 0.2, and the total weight of the remaining features to 0.2. For data transmission tasks (priority P2), the weight of communication quality score is set to 0.4, device load characteristics to 0.25, BeiDou-5G positioning deviation to 0.15, and the total weight of the remaining features to 0.2. For path planning tasks (priority P3), the weight of terrain complexity is set to 0.3, positioning accuracy to 0.2, obstacle density to 0.1, and the total weight of the remaining features to 0.4.

[0042] Through data acquisition, feature extraction, and weight allocation, the final output is an environmental feature vector with 12-dimensional quantitative features (such as terrain complexity, obstacle density, IMU three attitude angles, communication quality score, positioning deviation, device load, etc.), and the feature vectors of each dimension have been dynamically weighted through spatiotemporal attention.

[0043] In one embodiment of this application, the edge node cluster status information includes remaining computing power and a first communication delay. The remaining computing power is calculated as: total computing power - allocated computing power. The first communication delay is RTT (Round-Trip Time).

[0044] S102, input the multimodal environment characteristics, terminal device load characteristics, edge node cluster status information, and task queue information into the task scheduling model to determine the scheduling strategy for each task in the task queue. The scheduling strategy for each task indicates whether to execute the task on the terminal device or offload the task to an edge node in the edge node cluster.

[0045] In one embodiment of this application, weighted environmental feature vectors, edge node cluster status information, and terminal device task queue information are input into the task scheduling model. The task queue includes multiple tasks to be executed, each task having a priority identifier, which includes P1, P2, and P3. Identifier P1 indicates high priority, while P2 and P3 indicate low priority. For example, P1-level tasks include obstacle avoidance and emergency communication, P2-level tasks include data backhaul, and P3-level tasks include path planning.

[0046] In one embodiment of this application, before receiving data input, the task scheduling model filters out unqualified edge nodes in the edge node cluster based on at least one constraint to narrow down the candidate range of edge nodes. The at least one constraint includes, but is not limited to, task constraints, computing power constraints, and resource reservation constraints. Specifically, the task constraint may be the maximum communication latency required by the task for candidate edge nodes. For example, if a P1-level task requires a maximum communication latency of 20ms for candidate edge nodes, then edge nodes with a communication latency greater than 20ms are filtered out. If none of the candidate edge nodes meet the requirements, the task is forced to execute on the terminal device. The computing power constraint may be the computing power requirement of the task, meaning the remaining computing power of the candidate edge nodes must be greater than the task's computing power requirement. The resource reservation constraint may be that candidate edge nodes need to reserve resources for high-priority tasks; even if the computing power of a candidate edge node is sufficient, it still needs to be filtered out.

[0047] In one embodiment of this application, the task scheduling model is based on the improved Dijkstra algorithm. It calculates the second communication delay of unloading each task in the task queue to the edge node based on the weighted multimodal environment feature vector and the first communication delay of the edge node cluster. Based on the remaining computing power of the edge node cluster and the task queue information, it determines the computation delay. Based on the communication stability, it determines the first weight of the second communication delay and the second weight of the computation delay. Based on the second communication delay, the first weight, the computation delay and the second weight, it determines the total delay of unloading each task to the edge node. The edge node with the smallest total delay is determined as the edge node that processes each task.

[0048] In one embodiment of this application, the weights of the second communication delay and the calculation delay are dynamically adjusted based on communication stability. For example, communication stability is a communication quality score. When the communication quality score is ≥0.8, the first weight of the second communication delay is 30%, and the second weight of the calculation delay is 70%. When the communication quality score is between 0.5 and 0.8, both the first and second weights are 50%. When the communication quality score is <0.5, the first weight is 80%, and the second weight is 20%. The first weight of the forced communication delay for P1-level tasks is 80%, and the delay is guaranteed to be <20ms through 5G slicing. In this case, the second weight is 20%.

[0049] In one embodiment of this application, the second communication delay and the calculation delay are weighted and summed based on the first weight and the second weight to obtain the total delay, and the edge node with the smallest total delay is determined as the edge node for processing each task, thereby obtaining the scheduling strategy for each task in the task queue.

[0050] In one embodiment of this application, a high-priority (P1) task retention list and the edge node allocation results of low-priority tasks (P2 / P3) are output, a global task unloading instruction is sent to the terminal device, and the low-priority tasks of the terminal device are unloaded to the edge nodes for processing based on the scheduling strategy.

[0051] In one embodiment of this application, a P1-level task reserve list and real-time sensor data from the terminal device are used as inputs. The real-time sensor data includes ultrasonic radar obstacle distance, battery voltage, etc. A high-priority task queue is created based on high-priority tasks. The terminal device is controlled to process high-priority tasks according to the queue order, and preemption is triggered when the terminal device processes tasks in the high-priority task queue, for example, allocating more than 90% of CPU resources to high-priority tasks. For example, when the IMU detects a sudden change in attitude (e.g., pitch angle change > 2° per second), an attitude correction task or a return-to-home task is created, set as high priority, and preemption is forcibly triggered, thereby prioritizing the processing of the attitude correction task or the return-to-home task.

[0052] In one embodiment of this application, an interrupt priority mask can also be set to block tasks with priority values ​​lower than the priority mask. For example, fatal interrupts such as power failure and ultrasonic radar emergency alarm (obstacle distance < 5m) can be retained, while non-emergency interrupts such as log printing and low-priority sensor acquisition can be blocked to ensure a response time of < 5ms.

[0053] In one embodiment of this application, the method further includes: creating a low-priority task queue based on low-priority tasks, forming a double-buffered queue with the high-priority task queue, calculating the environmental change rate of multimodal environmental characteristics for two consecutive first preset periods, calculating the priority value of each low-priority task in the low-priority task queue based on the environmental change rate, and re-sorting the multiple low-priority tasks in the low-priority task queue based on the priority value of each low-priority task.

[0054] Specifically, the environmental change rate is obtained by using the Euclidean distance between the environmental feature vector of the current first preset period and the environmental feature vector of the previous preset period (0.1 seconds interval), and normalizing the calculated Euclidean distance to the range [0, 1]. The priority value = base priority (e.g., P3=3, P2=2) + environmental change rate × 30% + task relevance adjustment (e.g., for path planning tasks, if the environmental change rate > 0.5, add an additional 0.5). Every second preset period (e.g., 0.5 seconds), based on the priority value of each low-priority task, multiple low-priority tasks in the low-priority task queue are reordered to prevent low-priority tasks from being "starved" and left unprocessed for extended periods.

[0055] In one embodiment of this application, the terminal device processes high-priority tasks, and the edge node processes low-priority tasks, resulting in the execution results of P1-level tasks (e.g., obstacle avoidance command: turn left 30°, speed reduced to 2m / s), the execution progress of P2 / P3-level tasks, and terminal status feedback (including remaining battery percentage, CPU load rate, and IMU current attitude).

[0056] In one embodiment of this application, the method further includes: using the remaining computing power of the edge node cluster, the load rate of the terminal device, and the environmental complexity score as the state space, using the current task offloading ratio and the latency weight adjustment coefficient as the action space, using the latency optimization value, energy consumption reduction value, and resource utilization improvement value as the reward function, and using the reinforcement learning proximal policy optimization (PPO) algorithm to determine the optimized task offloading ratio and latency weight adjustment coefficient.

[0057] Specifically, Principal Component Analysis (PCA) is used to reduce the dimensionality of the environmental feature vectors, obtaining the top two principal components with a cumulative contribution rate ≥90%, which are then weighted, summed, and mapped to [0,1]. The current task offloading ratio is 30%~60%, and the latency (including communication latency and estimated latency) weight adjustment coefficient is ±20%. In the reward function, the weight of latency optimization is 50%, the weight of energy consumption reduction is 30%, and the weight of resource utilization improvement is 20%. The parameters of the policy network are updated every ten task cycles (e.g., each task cycle is 1 second), and the optimized task offloading ratio and latency weight adjustment coefficient are output. For example, when the load rate of the terminal device is greater than 80%, the optimized task offloading ratio is 60%.

[0058] This application's embodiments utilize spatiotemporal attention to enable task scheduling decisions to accurately pinpoint critical scenario requirements (e.g., prioritizing obstacle signals during obstacle avoidance). The preemptive scheduling and interruption protection of dual-buffered queues ensure exclusive access to high-priority task resources, reducing blind decision-making and resource contention. This effectively reduces task processing latency and response time for high-priority tasks, meeting the emergency needs of low-altitude equipment (e.g., drone obstacle avoidance).

[0059] In this embodiment, tasks are dynamically allocated at the edge based on computing power and communication latency. Low-priority tasks on the terminal device side are adjusted according to the rate of environmental change to avoid idle edge computing power or overload of terminal devices. The terminal device only performs high-energy-consuming tasks (such as path planning tasks) when necessary, reducing ineffective energy consumption, effectively reducing the CPU load rate of the terminal device, improving battery life, and improving the utilization rate of edge node resources.

[0060] The multimodal data in this embodiment covers dimensions such as terrain, communication, equipment, and positioning. Spatiotemporal attention is dynamically weighted according to the scenario (e.g., terrain is emphasized in mountainous areas and communication is emphasized in urban areas), avoiding the limitation of a single feature adapting to all scenarios, effectively improving the accuracy of scenario adaptation, and can be stably applied to diverse low-altitude scenarios such as plains, mountains, and urban canyons.

[0061] See Figure 3 The diagram shown is a structural schematic of a task scheduling device provided in one embodiment of this application. In one embodiment of this application, the task scheduling device 200 may include multiple functional modules composed of computer program segments. The computer program segments in the task scheduling device 200 may be stored in the memory of an electronic device and executed by at least one processor to perform the task scheduling method function.

[0062] In one embodiment of this application, the task scheduling device 200 can be divided into multiple functional modules according to the functions it performs. The functional modules of the task scheduling device 200 may include: an acquisition module 201 and a determination module 202. In this embodiment, a module refers to a series of computer program segments that can be executed by at least one processor and perform a fixed function, and which are stored in memory.

[0063] The acquisition module 201 is used to acquire the multimodal environment characteristics of the terminal device, the load characteristics of the terminal device, the status information of the edge node cluster, and the task queue information. The multimodal environment characteristics include at least one of terrain complexity, communication stability, and positioning accuracy.

[0064] The determination module 202 is used to input the multimodal environment characteristics, the terminal device load characteristics, the edge node cluster status information and the task queue information into the task scheduling model, and determine the scheduling strategy for each task in the task queue. The scheduling strategy for each task instructs each task to be executed on the terminal device or to be offloaded to an edge node in the edge node cluster.

[0065] See Figure 4 The diagram shown is a hardware structure schematic of an electronic device provided in an embodiment of this application. The task scheduling method provided in this embodiment can be applied to electronic device 1, which includes, but is not limited to, a processor 110 and a memory 120 connected via a communication bus 130. Figure 4 This is merely an example of an electronic device and does not constitute a limitation thereof. In other embodiments, the electronic device may include more components than those shown in the figure.

[0066] See Figure 5 The diagram shown is a schematic of the hardware structure of a vehicle provided in an embodiment of this application. The task scheduling method provided in this embodiment can also be applied to vehicle 2, which includes, but is not limited to, a processor 210 and a memory 220 connected via a communication bus 230. Figure 5 This is merely an example of a vehicle and does not constitute a limitation thereof. In other embodiments, the vehicle may include more components than those shown in the figure.

[0067] Memory may include one or more Random Access Memory (RAM) and one or more Non-Volatile Memory (NVM). RAM can be directly read and written by the processor and can be used to store executable programs (such as machine instructions) of the operating system or other running programs, as well as user and application data. RAM may include Static Random-Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), and Double Data Rate Synchronous Dynamic Random Access Memory (DDR SDRAM), etc.

[0068] Non-volatile memory can also store executable programs and user and application data, and can be pre-loaded into random access memory for direct read and write operations by the processor. Non-volatile memory can include disk storage devices and flash memory.

[0069] The memory is used to store one or more computer programs. These one or more computer programs are configured to be executed by a processor. The one or more computer programs include multiple instructions that, when executed by the processor, enable a task scheduling method to be performed on an electronic device or vehicle.

[0070] In other embodiments, the electronic device or vehicle also includes an external memory interface for connecting to an external memory to expand the storage capacity of the electronic device or vehicle.

[0071] A processor may include one or more processing units, such as an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0072] The processor provides computing and control capabilities; for example, the processor is used to execute computer programs stored in memory to implement the task scheduling method described above.

[0073] A communication bus is used at least to provide a channel for communication between memory and processor in electronic devices or vehicles.

[0074] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device or vehicle. In other embodiments of this application, the electronic device or vehicle may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0075] This application also provides a computer storage medium storing computer instructions. When the computer instructions are executed on an electronic device or vehicle, the electronic device or vehicle performs the aforementioned related method steps to implement the task scheduling method described in the above embodiments.

[0076] This application also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned steps to implement the task scheduling method described in the above embodiments.

[0077] In addition, embodiments of this application also provide an apparatus, which may specifically be a chip, component or module. The apparatus may include a connected processor and a memory; wherein the memory is used to store computer execution instructions, and when the apparatus is running, the processor may execute the computer execution instructions stored in the memory to cause the chip to execute the task scheduling methods in the above method embodiments.

[0078] In this application, the electronic devices, vehicles, computer storage media, computer program products or chips provided in the embodiments are all used to execute the corresponding methods provided above. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0079] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0080] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0081] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0082] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0083] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts or all or part of the technical solutions that contribute to the prior art, can be embodied in the form of a software product. The software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application.

Claims

1. A task scheduling method, characterized in that, The method includes: The system acquires multimodal environment characteristics of the terminal device, terminal device load characteristics, edge node cluster status information, and task queue information. The multimodal environment characteristics include at least one of terrain complexity, communication stability, and positioning accuracy. The multimodal environment characteristics, terminal device load characteristics, edge node cluster status information, and task queue information are input into the task scheduling model to determine the scheduling strategy for each task in the task queue. The scheduling strategy for each task instructs each task to be executed on the terminal device or to be offloaded to an edge node in the edge node cluster.

2. The task scheduling method according to claim 1, characterized in that, The multimodal environment data of the terminal device includes at least one of point cloud data, attitude data, communication data, state data, and positioning data. Obtaining the multimodal environment features of the terminal device includes: The statistical features of the point cloud data are extracted, and the statistical features are normalized to obtain the terrain complexity. A communication quality score is calculated based on the communication data, and the communication quality score is normalized to obtain the communication stability. The deviation data of the positioning data is determined as the positioning accuracy.

3. The task scheduling method according to claim 1, characterized in that, The multimodal environment features also include obstacle density, IMU three-attitude angles, wind speed, weather conditions, task type, and task time window.

4. The task scheduling method according to claim 1, characterized in that, The multimodal environment features are multi-dimensional feature vectors, where each dimension of the feature vector has a corresponding weight. The method further includes: In the time dimension, a self-attention mechanism is used to set the weights corresponding to the features in each dimension, including: Calculate the query matrix, key matrix, and value matrix corresponding to the features of each dimension based on the weight matrices WQ, WK, and WV; Based on the query matrix, the key matrix, and the value matrix, the weights corresponding to the features of each dimension are calculated.

5. The task scheduling method according to claim 1, characterized in that, The multimodal environment features are multi-dimensional feature vectors, where each dimension of the feature vector has a corresponding weight. The method further includes: In the spatial dimension, the weights corresponding to the features in each dimension are set using a cross-attention mechanism, including: The query matrix corresponding to the feature of each dimension is calculated based on the weight matrix WQ. Calculate the key matrix and value matrix corresponding to the task type based on the weight matrices WK and WV; Based on the query matrix, the key matrix, and the value matrix, the weights corresponding to the features of each dimension are calculated.

6. The task scheduling method according to claim 4 or 5, characterized in that, The edge node cluster status includes remaining computing power and first communication latency. The step of inputting the multimodal environment characteristics, terminal device load characteristics, edge node cluster status information, and task queue information into the task scheduling model to determine the scheduling strategy for each task in the task queue includes: The task scheduling model determines the second communication delay for offloading each task to the edge node based on the weighted multimodal environment features and the first communication delay of the edge node cluster. The task scheduling model also determines the computation latency based on the remaining computing power of the edge node cluster and the task queue information; The first weight of the second communication delay and the second weight of the calculated delay are determined based on the communication stability. The total latency for offloading each task to the edge node is determined based on the second communication latency, the first weight, the calculation latency, and the second weight. The edge node with the minimum total latency is selected as the edge node for processing each task.

7. The task scheduling method according to claim 1, characterized in that, The method further includes: Create a low-priority task queue based on low-priority tasks; Calculate the rate of change of multimodal environmental characteristics over multiple consecutive preset periods; Calculate the priority value of each low-priority task in the low-priority task queue based on the environmental change rate; Based on the priority value of each low-priority task, the multiple low-priority tasks in the low-priority task queue are reordered.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor: The memory is used to store program instructions; The processor is configured to read and execute the program instructions stored in the memory, and when the program instructions are executed by the processor, cause the electronic device to perform the task scheduling method as described in any one of claims 1 to 7.

9. A vehicle, characterized in that, The vehicle includes a memory and a processor: The memory is used to store program instructions; The processor is configured to read and execute the program instructions stored in the memory, and when the program instructions are executed by the processor, cause the vehicle to perform the task scheduling method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium stores program instructions that, when executed on an electronic device or vehicle, cause the processor of the electronic device or vehicle to perform the task scheduling method as described in any one of claims 1 to 7.

Images