A method and system for multi-task collaborative processing of an industrial vehicle-mounted computer

Abstract

This invention proposes a multi-task collaborative processing method and system for industrial vehicle-mounted computers, relating to the field of data processing. The method includes: extracting the original time-series sequence and preprocessing it to obtain a smoothed driving behavior sequence; extracting the original features and performing component analysis to obtain a driver habit feature vector; classifying several driving behavior patterns and constructing priority allocation rules; extracting real-time driving features, matching the real-time driving features with the driving behavior patterns, and retrieving the priority allocation rules to obtain a first task execution order; adjusting the first task execution order to obtain a second task execution order; allocating resources according to the second task execution order and the priority allocation rules, and optimizing the resources to obtain an optimized resource allocation scheme; and adjusting the second task execution order to obtain a third task execution order. This invention improves system resource utilization while ensuring the real-time performance of critical tasks, effectively balancing the real-time performance and resource efficiency of the vehicle-mounted system.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to a multi-task collaborative processing method and system for an industrial vehicle-mounted computer. Background Technology

[0002] As a core component in the fields of intelligent transportation and autonomous driving, industrial vehicle computers play a crucial role in improving vehicle operating efficiency and safety. The multi-task collaborative processing capability of industrial vehicle computers directly determines the real-time response and resource optimization effect of the vehicle system in complex driving environments.

[0003] Driver behavior data reflects individual driving habits, and integrating historical driving data allows for personalized optimization to create driving plans more aligned with driver preferences. However, current industrial vehicle computers often face challenges in multi-task collaborative processing, including uneven resource allocation, insufficient dynamic adjustment of task priorities, and low data processing efficiency. These issues make it difficult for vehicle systems to simultaneously meet the requirements of timeliness and processing accuracy under high load scenarios, particularly when integrating historical driving data for personalized optimization. These problems collectively constitute the core challenge of multi-task collaborative processing in industrial vehicle computers: how to dynamically optimize multi-task collaborative execution based on driver habit characteristics within limited resource capacity, while balancing the conflict between functional degradation and high-speed processing.

[0004] Therefore, how to achieve multi-task collaborative processing based on the driver's driving habits under limited resource capacity is a key problem that urgently needs to be solved in the field of industrial vehicle computers. Summary of the Invention

[0005] In view of this, the present invention proposes a multi-task collaborative processing method and system for industrial vehicle computers, in order to solve the problem that industrial vehicle computers cannot simultaneously meet the requirements of task timeliness and processing accuracy when performing multi-task collaborative processing.

[0006] The technical solution of this invention is implemented as follows:

[0007] According to a first aspect, embodiments of the present invention provide a multi-task collaborative processing method for an industrial vehicle-mounted computer, the method comprising:

[0008] Acquire historical driving data, extract raw time series from the historical driving data, preprocess the raw time series data to obtain smooth driving behavior sequences;

[0009] The original features are extracted from the smooth driving behavior sequence, and the driver habit feature vector carrying the driver's driving habits is determined from the original features.

[0010] The driver's habit feature vector is classified into several driving behavior patterns. A corresponding priority allocation rule is constructed for each driving behavior pattern. The priority allocation rule includes the tasks involved, the order of each task, and the priority weight.

[0011] Acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain matching driving behavior patterns, retrieve the corresponding priority allocation rules according to the matching driving behavior patterns, and obtain the first task execution order according to the priority allocation rules.

[0012] Determine whether there is a circular dependency among tasks based on the priority allocation rules, adjust the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task, and allocate resources according to the execution order of the second task and the priority allocation rules to obtain the initial resource allocation scheme;

[0013] The resource occupancy threshold of the task is obtained based on the initial resource allocation scheme, and the initial resource allocation scheme is adjusted based on the resource occupancy threshold to obtain the optimized resource allocation scheme;

[0014] The execution order of the second task is adjusted according to the optimized resource allocation scheme to obtain the execution order of the third task, and the tasks are executed according to the execution order of the third task.

[0015] In conjunction with the first aspect, in the first embodiment of the first aspect, the step of classifying the driver's habitual feature vector to obtain several driving behavior patterns, and constructing a corresponding priority allocation rule for each driving behavior pattern, specifically includes:

[0016] The driver habit feature vectors are clustered to obtain several clusters. If the variance within a cluster is lower than a preset threshold, the clusters with variances lower than the preset threshold are merged to obtain an initial behavior pattern. Each cluster represents a pattern in the initial behavior pattern, and each pattern contains several driver habit feature vectors.

[0017] The behavioral weights of each driver habit feature vector in the initial behavior pattern are calculated by the entropy method, and the initial behavior patterns with behavioral weights are divided into patterns by the hierarchical clustering method to obtain the driving behavior patterns.

[0018] The distance between each driving behavior pattern is determined based on the behavior weight. If the distance is determined to be lower than the first preset value, the driving behavior patterns that are mutually less than the first preset value are merged.

[0019] For each driving behavior pattern, a corresponding priority allocation rule is constructed, and the driving behavior pattern is stored in a pre-set behavior pattern library, and the priority allocation rule is stored in a pre-set rule library.

[0020] In conjunction with the first embodiment of the first aspect, in the second embodiment of the first aspect, the step of acquiring real-time driving data, extracting real-time driving features from the real-time driving data, matching the real-time driving features with driving behavior patterns in a behavior pattern library to obtain matching driving behavior patterns, retrieving the corresponding priority allocation rule according to the matching driving behavior pattern, and obtaining the first task execution order according to the priority allocation rule specifically includes:

[0021] Real-time driving data is obtained through time-series sampling methods to obtain a first data stream carrying timestamp information;

[0022] The first data stream is processed by a moving average filter to obtain the second data stream;

[0023] Principal component analysis is performed on the second data stream to obtain a third data stream containing real-time feature vectors;

[0024] Determine the Euclidean distance of the real-time feature vectors in the third data stream. If the Euclidean distance exceeds a preset environmental complexity threshold, perform vector normalization on the real-time feature vectors that exceed the preset environmental complexity threshold to generate real-time feature vectors.

[0025] Determine the Euclidean distance between real-time driving features and each driving behavior pattern in the behavior pattern library, and determine the driving behavior pattern that matches the real-time driving features based on the Euclidean distance.

[0026] Based on the matching driving behavior pattern, the corresponding priority allocation rule is retrieved from the rule base, a task priority table is generated according to the priority allocation rule, and the execution order of the first task is obtained according to the task priority table.

[0027] In conjunction with the first aspect, in the third embodiment of the first aspect, the step of determining whether there is a circular dependency among tasks according to the priority allocation rule, adjusting the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task, and allocating resources according to the execution order of the second task and the priority allocation rule to obtain the initial resource allocation scheme specifically includes:

[0028] Determine the dependencies between tasks based on priority allocation rules, and determine the estimated duration of each task based on the dependencies and the completion time of each task.

[0029] Based on the expected duration and dependency order of the tasks, a topological sorting sequence of tasks is constructed. Based on the topological sorting sequence, it is determined whether there is a circular dependency between the tasks. If a circular dependency is determined, the execution order of the first task is adjusted according to the circular dependency and priority weight to obtain the execution order of the second task.

[0030] Obtain resource capacity, allocate resources using a forward recursive strategy based on the execution order of the second task and the resource capacity, and if it is determined that the resource capacity is insufficient, reallocate resources according to the expected duration and priority weight of the tasks to obtain an initial resource allocation scheme.

[0031] In conjunction with the first aspect, in the fourth embodiment of the first aspect, the step of obtaining the resource occupancy threshold of the task according to the initial resource allocation scheme, and adjusting the initial resource allocation scheme according to the resource occupancy threshold to obtain an optimized resource allocation scheme specifically includes:

[0032] Based on the initial resource allocation scheme, obtain the resource usage ratio of all tasks whose response time does not exceed the preset time, and obtain the resource usage threshold.

[0033] If the resource occupancy threshold is determined to be lower than the preset standard, a genetic algorithm is used to reallocate the resource capacity to obtain an optimized resource allocation scheme.

[0034] In conjunction with the first aspect, in the fifth embodiment of the first aspect, the step of adjusting the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task specifically includes:

[0035] Based on the optimized resource allocation scheme and the use of conflict detection algorithms, it is determined whether there is resource competition between tasks. If resource competition exists, i.e. conflict, the execution path of the non-conflicting task is determined.

[0036] By using conflict-free task execution paths and the estimated duration of tasks, a linear programming algorithm is used to adjust the allocation of resource capacity to obtain the allocation efficiency with the highest resource utilization. Based on the allocation efficiency and conflict-free task execution paths, the execution order of the second task is adjusted to obtain the execution order of the third task.

[0037] In conjunction with the first aspect, in the sixth embodiment of the first aspect, the method further includes the following steps:

[0038] Based on the third task execution order, the latency of tasks with priority weights lower than the preset priority is obtained. When the latency is determined to exceed the preset degradation threshold, the resource capacity of tasks with priority weights lower than the preset priority is reduced to obtain the fourth task execution order, and the tasks are executed according to the fourth task execution order.

[0039] In conjunction with the first aspect, in the seventh embodiment of the first aspect, the step of extracting original features from a smooth driving behavior sequence and determining a driver habit feature vector carrying the driver's driving habits from the original features specifically includes:

[0040] Data from the smoothed driving behavior sequence were aligned to a unified time axis, and raw features were extracted from the aligned smoothed driving behavior sequence; the raw features included the rate of change of acceleration, steering angular velocity, and braking frequency.

[0041] The acceleration rate of change and steering angular velocity are normalized, and the braking frequency is discretized and normalized to obtain standardized features;

[0042] Determine the variance among standardized features. If the variance among standardized features exceeds a preset variance, calculate the feature weights corresponding to each standardized feature using the entropy method. If the variance among standardized features does not exceed a preset variance, equally divide the feature weights corresponding to each standardized feature.

[0043] The standardized features with feature weights are subjected to dimensionality reduction processing to obtain dimensionality-reduced features. The information retention rate of the dimensionality-reduced features is determined, and the dimensionality-reduced features with information retention rates exceeding the preset retention rate are used as driver habit feature vectors.

[0044] In conjunction with the first aspect, in the eighth embodiment of the first aspect, the step of acquiring historical driving data, extracting the original time-series sequence from the historical driving data, and preprocessing the original time-series data to obtain a smooth driving behavior sequence specifically includes:

[0045] Acquire historical driving data and extract the raw time series from the historical driving data;

[0046] Data cleaning is performed on the original time series to remove duplicate and outlier values, and missing points in the original time series are marked to obtain the first time series.

[0047] The first time series is subjected to a moving average filter to remove noisy data, resulting in the second time series.

[0048] Linear interpolation is performed on the second time series to fill in the missing points in the second time series, resulting in the third time series.

[0049] Time series data analysis is performed on the third time series to extract driving behavior features. Fast Fourier Transform is then applied to the driving behavior features to determine their frequency distribution. Based on the frequency distribution, smoothed driving behavior features are obtained. The driving behavior features include the distribution of acceleration change rate, steering angle change rate, and braking frequency.

[0050] According to a second aspect, embodiments of the present invention provide a multi-task collaborative processing system for an industrial vehicle-mounted computer, the system comprising:

[0051] The sequence acquisition module is used to acquire historical driving data, extract raw time series sequences from the historical driving data, and preprocess the raw time series data to obtain smooth driving behavior sequences.

[0052] The habit acquisition module is used to extract raw features from the smooth driving behavior sequence and determine the driver habit feature vector carrying the driver's driving habits from the raw features.

[0053] The behavior classification module is used to classify the driver's habit feature vectors into several driving behavior patterns, and to construct a corresponding priority allocation rule for each driving behavior pattern. The priority allocation rule includes the tasks included, the order of each task, and the priority weight.

[0054] The first sequence module is used to acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain the matching driving behavior patterns, and retrieve the corresponding priority allocation rules according to the matching driving behavior patterns to obtain the first task execution order according to the priority allocation rules.

[0055] The second sequence module is used to determine whether there is a circular dependency between tasks according to the priority allocation rules, adjust the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task, and allocate resources according to the execution order of the second task and the priority allocation rules to obtain the initial resource allocation scheme.

[0056] The resource optimization module is used to obtain the resource occupancy threshold of the task based on the initial resource allocation scheme, and adjust the initial resource allocation scheme according to the resource occupancy threshold to obtain an optimized resource allocation scheme.

[0057] The sequence determination module is used to adjust the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, and then execute the task according to the execution order of the third task.

[0058] The multi-task collaborative processing method and system for industrial vehicle-mounted computers of the present invention have the following advantages over the prior art:

[0059] By extracting raw time-series data from historical driving data and preprocessing it to obtain smooth driving behavior sequences, raw features are extracted from these sequences. Principal component analysis is then used to accurately extract driver habit feature vectors containing driver driving habits from these raw features. These driver habit feature vectors are then subjected to two separate behavior classifications to obtain several driving behavior patterns. These patterns can provide personalized driving suggestions to drivers, helping to optimize the classification and management of driving behavior. Next, priority allocation rules are constructed for each driving behavior pattern. Real-time driving features are then extracted from real-time driving data and compared with those in the behavior pattern library. The system matches driving behavior patterns to obtain matching patterns and retrieves corresponding priority allocation rules based on these patterns to determine the first task execution order. This first task execution order is then optimized by considering task timeliness requirements and optimal task execution time to arrive at the final third task execution order. This third task execution order ensures reasonable allocation of resources, effectively avoids resource waste, and guarantees the completion of high-priority tasks. It enables high-speed collaborative processing of multiple tasks and can adaptively adjust the vehicle system's task scheduling strategy based on driver behavior characteristics and the current driving environment. This ensures the real-time performance of critical tasks while improving system resource utilization, effectively balancing the real-time performance and resource efficiency of the vehicle system. Attached Figure Description

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0061] Figure 1 This is one of the flowcharts illustrating the multi-task collaborative processing method for an industrial vehicle-mounted computer according to the present invention;

[0062] Figure 2 This is a second flowchart illustrating the multi-task collaborative processing method for the industrial vehicle-mounted computer of the present invention.

[0063] Figure 3 This is the third flowchart illustrating the multi-task collaborative processing method for the industrial vehicle-mounted computer of the present invention.

[0064] Figure 4 This is the fourth flowchart illustrating the multi-task collaborative processing method for the industrial vehicle-mounted computer of the present invention.

[0065] Figure 5 This is the fifth flowchart illustrating the multi-task collaborative processing method for the industrial vehicle-mounted computer of the present invention.

[0066] Figure 6 This is the sixth flowchart illustrating the multi-task collaborative processing method for the industrial vehicle-mounted computer of the present invention;

[0067] Figure 7 This is a schematic diagram of the multi-task collaborative processing system of the industrial vehicle computer of the present invention. Detailed Implementation

[0068] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0069] As a core component in the fields of intelligent transportation and autonomous driving, industrial vehicle computers play a crucial role in improving vehicle operating efficiency and safety. The multi-task collaborative processing capability of industrial vehicle computers directly determines the real-time response and resource optimization effect of the vehicle system in complex driving environments.

[0070] Driver behavior data reflects individual driving habits, and integrating historical driving data allows for personalized optimization to create driving plans more aligned with driver preferences. However, current industrial vehicle computers often face challenges in multi-task collaborative processing, including uneven resource allocation, insufficient dynamic adjustment of task priorities, and low data processing efficiency. These issues make it difficult for vehicle systems to simultaneously meet the requirements of timeliness and processing accuracy under high load scenarios, particularly when integrating historical driving data for personalized optimization. These problems collectively constitute the core challenge of multi-task collaborative processing in industrial vehicle computers: how to dynamically optimize multi-task collaborative execution based on driver habit characteristics within limited resource capacity, while balancing the conflict between functional degradation and high-speed processing.

[0071] Specifically, current industrial vehicle computers generally lack efficient dynamic modeling mechanisms when extracting driver habit feature vectors from historical driving data, making it difficult to reflect subtle changes in driving behavior in real time. Furthermore, a significant conflict exists between task timeliness and resource capacity allocation. For example, high-priority tasks may be delayed due to resource contention, while low-priority tasks may consume excessive resources, leading to a decline in the overall data processing performance of the vehicle system.

[0072] Therefore, how to achieve multi-task collaborative processing based on the driver's driving habits under limited resource capacity is a key problem that urgently needs to be solved in the field of industrial vehicle computers.

[0073] The multi-task collaborative processing method for industrial vehicle computers provided in this manual aims to accurately extract driving behavior patterns from historical driving data, dynamically calculate the optimal task execution time in combination with task timeliness requirements, and achieve high-speed collaborative processing of multiple tasks through resource capacity reallocation and function degradation.

[0074] Please see Figure 1 , Figure 1 A flowchart illustrating a multi-task collaborative processing method for an industrial vehicle-mounted computer according to an embodiment of the present invention is shown. The method may include the following steps:

[0075] S101. Obtain historical driving data, extract the original time series from the historical driving data, preprocess the original time series data, and obtain a smooth driving behavior sequence.

[0076] Historical driving data can be stored in electronic devices beforehand, or it can be acquired by electronic devices from external sources. For example, electronic devices can acquire historical driving data from various types of onboard sensors, such as those recording information including acceleration, steering angle, and braking frequency when the vehicle is driving in urban driving scenarios.

[0077] Vehicle-mounted sensors typically include accelerometers, gyroscopes, and brake pedal position sensors. Accelerometers record the vehicle's acceleration values ​​along the X, Y, and Z axes; gyroscopes capture dynamic changes in the steering angle; and brake sensors determine braking action based on pedal pressure or position. In acquiring acceleration, steering angle, and braking data from these sensors, they can be set at a preset acquisition frequency (e.g., 10Hz, recording acceleration, steering angle, and braking status every 0.1 seconds). This high-frequency acquisition ensures the temporal continuity of the data, providing rich foundational information for subsequent analysis.

[0078] There are no restrictions on the specific form of historical driving data acquisition, as long as the electronic device can acquire the historical driving data.

[0079] In this embodiment, the original time-series sequence includes information such as historical acceleration, historical steering angle, and historical braking frequency. When extracting the original time-series sequence containing information such as historical acceleration, historical steering angle, and historical braking frequency from historical driving data, the original time-series sequence often contains noisy data. This is due to abnormal spikes caused by vehicle sensor jitter, data loss due to data acquisition interruptions, etc. To address the fact that the original time-series sequence also contains noisy data, this embodiment preprocesses the original time-series sequence to remove the noisy data.

[0080] Preferred preprocessing methods include, but are not limited to: removing duplicate and outlier values, moving average filtering, interpolation, and fast Fourier transform. For example, if a historical acceleration data segment is 0, 2, 2, 10, 2, 10, it is clear that the historical acceleration data deviates from the normal range and can be considered outlier and removed. If a historical acceleration data segment contains duplicate values, such as consecutive 2s, these can be merged into a single valid point, and the duplicate values ​​can be removed. If there are missing data points at a certain moment, such as null values ​​appearing in a data sequence containing historical braking frequencies, these are marked as missing data points.

[0081] S102. Extract the original features from the smooth driving behavior sequence, and determine the driver habit feature vector carrying the driver's driving habits from the original features. In this embodiment, the original features include the rate of change of acceleration, steering angular velocity, and braking frequency.

[0082] In this embodiment, the feature discriminant degree between each of the original features is determined, and a corresponding feature weight is assigned to each original feature based on the feature discriminant degree. Then, principal component analysis (PCA) is used to perform dimensionality reduction on the original features with assigned feature weights, retaining principal components with an information retention rate higher than a preset retention rate. Therefore, the driver habit feature vector generated in this way retains the driver's driving habits and can be used to analyze driving style. For example, a high rate of change of acceleration and a high steering angular velocity in the vector may reflect aggressive driving, while a low braking frequency indicates smooth driving. This multi-dimensional analysis of feature vectors provides a reliable basis for subsequent driving behavior evaluation.

[0083] S103. Classify the driver's habitual feature vector to obtain several driving behavior patterns. Construct a corresponding priority allocation rule for each driving behavior pattern. It can be understood that the constructed driving behavior patterns and the corresponding priority allocation rules have a mapping relationship. Among them, driving behavior patterns can include aggressive, smooth, conservative driving modes, etc.

[0084] Preferably, driving behavior patterns can be stored in a pre-set behavior pattern library and priority allocation rules can be stored in a pre-set rule library to facilitate subsequent matching of driving behavior patterns and retrieval of priority allocation rules.

[0085] In this embodiment, the behavior classification includes two steps. The first behavior classification divides the driver's habit feature vector into initial behavior patterns based on similarity. The second behavior classification is based on the classification results obtained from the first behavior classification and further divides the driver's behavior patterns, thereby refining the granularity of the behavior classification.

[0086] Priority allocation rules include the tasks contained in the driving behavior mode, the order of each task, and the priority weight. For example, the priority allocation rule for the smooth driving mode is "cruise control > navigation prompts > entertainment system"; the priority allocation rule for the aggressive driving mode is "brake assist > warning prompts > navigation prompts", so as to respond to potential risks in a timely manner, that is, each task has a corresponding priority weight.

[0087] S104. Acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns to obtain matching driving behavior patterns, retrieve the corresponding priority allocation rules based on the matching driving behavior patterns, and obtain the first task execution order according to the priority allocation rules. Since the first task execution order is based on the priority allocation rules, it can meet the timeliness requirements of the tasks.

[0088] In this embodiment, step S104 specifically includes:

[0089] S1041. Real-time driving data is acquired through a time-series sampling method to obtain a first data stream carrying timestamp information, and the first data stream is a structured data stream. Specifically, real-time driving data can be acquired through a time-series sampling method, which collects this data at a fixed frequency, such as 10Hz, and each data point is accompanied by a timestamp.

[0090] S1042. Perform moving average filtering on the first data stream to remove noise and smooth the data, resulting in a second data stream. This second data stream retains the real-time driving behavior trend while removing abrupt noise. For example, if the difference between real-time acceleration, real-time steering angle, and real-time braking frequency and the preceding and following data exceeds a set threshold, moving average filtering can replace it with the average of the preceding and following data. In sharp turning scenarios, the original steering angle data may fluctuate due to vibration. Moving average filtering can filter out high-frequency noise, retaining only low-frequency signals that reflect the vehicle's true movement. After this processing, the second data stream better reflects the true driving intention, improving the accuracy of subsequent analysis.

[0091] S1043. Perform principal component analysis on the second data stream to obtain a third data stream containing real-time feature vectors. For the second data stream, extract real-time feature vectors of key features such as real-time acceleration, real-time steering angle, and real-time braking frequency using the principal component analysis algorithm, generate high-dimensional feature vectors containing at least three dimensions, and thus obtain the third data stream.

[0092] Preferably, a time-domain analysis method can be used to perform the aforementioned principal component analysis. That is, the first three principal components can be extracted by calculating the covariance matrix of the data to form a high-dimensional feature vector. For example, in urban driving scenarios, the principal components may reflect a strong correlation between acceleration and braking frequency. The three-dimensional vector of the third data stream is such as [0.65, 0.32, 0.15]. This reduces data redundancy and highlights key behavioral features.

[0093] S1044. If the Euclidean distance of the real-time feature vectors in the third data stream exceeds a preset environmental complexity threshold, vector normalization is performed on the real-time feature vectors exceeding the preset environmental complexity threshold to generate standardized real-time feature vectors, thus obtaining the real-time driving features. It is understandable that in complex driving environments, such as congested roads or sharp turns, the feature vectors of the third data set may fluctuate significantly due to environmental changes.

[0094] In this embodiment, a preset environmental complexity threshold, such as 0.5, can be set in advance. By calculating the Euclidean distance between vectors, if it exceeds the preset environmental complexity threshold, vector normalization is performed, mapping the vector values ​​to the [0,1] interval. For example, a vector [0.8,0.4,0.2] is normalized to [0.62,0.31,0.15]. The normalized real-time driving features are more suitable for subsequent behavior pattern analysis and can adapt to different driving scenarios.

[0095] Among them, time series sampling ensures the temporal integrity of the data, moving average filtering improves the data quality, principal component analysis focuses on core features, and vector normalization enhances environmental adaptability. The processing in steps S1041 to S1044 jointly supports the accurate construction and acquisition of real-time driving features, providing a reliable foundation for subsequent applications.

[0096] S1045. Determine the Euclidean distance between real-time driving features and various driving behavior patterns in the behavior pattern library, and determine the driving behavior pattern that matches the real-time driving features based on the Euclidean distance, i.e., the current driving behavior pattern.

[0097] After the above steps, real-time driving features for Euclidean distance calculation are obtained. These feature vectors retain the core information of the data while reducing dimensionality complexity. When calculating the Euclidean distance, the real-time feature vectors need to be compared with various pre-stored driving behavior patterns in the behavior pattern library.

[0098] If the calculated Euclidean distance to a certain driving behavior pattern is less than a preset distance, then a pattern matching rule is applied to take that driving behavior pattern as the current driving behavior pattern. This matching rule is intuitive and efficient, and can quickly identify the behavior type. The behavior pattern library may contain multiple driving behavior patterns, such as the vector corresponding to the smooth driving mode being (0.2, 0.03, 0.5), while the vector corresponding to the aggressive driving mode is (0.8, 0.1, 1.2).

[0099] Specifically, assuming the vector of real-time driving features is (0.3, 0.05, 0.8), and its Euclidean distance from the smooth driving mode is small and less than a preset distance, then the current mode is determined to be smooth driving. It should be noted that the threshold setting needs to be adjusted according to the actual scenario; for example, it may be stricter in urban roads, while it can be appropriately relaxed in highway scenarios. The rule base plays a crucial role in generating the task priority table based on the current driving behavior pattern.

[0100] S1046. Based on the matching driving behavior pattern, retrieve the corresponding priority allocation rule from the rule base, and generate a task priority table based on the task, task order, and task weight contained in the priority allocation rule. In this way, the first task execution order can be obtained based on the task priority table.

[0101] After obtaining the priority allocation rules, the task execution order is generated by sorting the tasks in descending order based on the sorting algorithm and priority weights (e.g., cruise control weight 90, navigation prompts weight 70, entertainment system weight 50). This dynamic adjustment method ensures that critical tasks are executed first, improving driving efficiency and significantly enhancing the system's adaptability, ensuring a high degree of match between task allocation and the driving environment. In complex driving environments, the above process, through a complete chain from raw signals to task sorting, effectively supports real-time decision-making. Matching driving behavior patterns and prioritizing tasks further optimizes the logic and timeliness of task execution.

[0102] S105. Determine if there are circular dependencies among tasks based on the priority allocation rules. Adjust the execution order of the first task according to the circular dependencies and priority weights to obtain the execution order of the second task. Allocate resources according to the execution order of the second task and the priority allocation rules to obtain the initial resource allocation scheme. The execution order of the second task quantifies the task duration and dependency order to ensure that the scheduling logic is rigorous, thereby obtaining the optimal task execution time.

[0103] S106. Obtain the resource occupancy threshold for the task based on the initial resource allocation scheme, and adjust the initial resource allocation scheme according to the resource occupancy threshold to obtain the optimized resource allocation scheme. The resource occupancy threshold reflects the degree of resource stress in the system and helps to identify potential resource bottlenecks.

[0104] In this embodiment, the resource occupancy thresholds for high-time-sensitivity tasks are specifically obtained. This is achieved by receiving the resource capacity allocation for high-time-sensitivity tasks from the initial allocation scheme, then iterating through the resource capacity occupancy ratios of each allocated task, and summing the results to obtain the aforementioned resource occupancy thresholds. If the resource occupancy threshold is lower than a preset standard, a genetic algorithm is used to initialize the population and iteratively optimize it to generate an optimized allocation scheme.

[0105] In this embodiment, high time-sensitive tasks typically refer to tasks with strict requirements for response time, such as real-time data processing or critical business operations, i.e., tasks whose response time does not exceed a preset time.

[0106] S107. Adjust the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, which is the final task execution schedule. Finally, execute the tasks according to the execution order of the third task to achieve layer-by-layer optimization of dependency relationships and resource allocation.

[0107] In this embodiment, step S107 specifically includes:

[0108] S1071. Based on the optimized resource allocation scheme and using the conflict detection algorithm, determine whether there is resource competition between tasks. If resource competition exists, i.e., a conflict, determine the execution path of the non-conflicting task and adjust the execution order of the second task accordingly.

[0109] S1072. Using the conflict-free task execution path and the expected duration of the task, the allocation of resource capacity is adjusted by linear programming algorithm to obtain the allocation efficiency with the highest resource utilization. Then, the execution order of the second task is adjusted according to the allocation efficiency and the conflict-free task execution path to obtain the execution order of the third task.

[0110] For example, a project includes Task A (requirements analysis task, estimated duration 2 days), Task B (design task, estimated duration 3 days, dependent on Task A), Task C (coding task, estimated duration 5 days, dependent on Task B), and Task D (testing task, estimated duration 2 days, dependent on Task C). Task A is determined to have the highest priority because it has no prerequisite dependencies; Task B is next because it depends on Task A; and so on. Therefore, the execution order of the second task is: Task A → Task B → Task C → Task D. This second task execution order ensures that tasks are arranged according to their dependencies, avoiding scheduling failures due to missing dependencies.

[0111] Based on the optimized resource allocation scheme, a conflict detection algorithm is used to determine resource contention. Assume the project has two servers: tasks A and B each require one server, and task C requires two servers. In the second task execution sequence, task C requires two servers, but if task D is scheduled in parallel with C, a conflict may occur due to insufficient servers. The conflict detection algorithm will identify this problem and adjust the scheduling order, for example, postponing task D until task C is completed, ensuring conflict-free resource allocation. Using conflict-free task execution paths and estimated durations, a linear programming algorithm is used to optimize resource allocation. Again using this project as an example, assuming the total project duration needs to be controlled within 12 days, linear programming analyzes the resource consumption and time constraints of each task. For example, task C takes longer, so two servers can be allocated first to ensure progress, while tasks A and B can share a single server to reduce resource consumption. This optimization ensures maximum resource utilization while meeting the project schedule requirements.

[0112] The third task execution order generated based on allocation efficiency and conflict-free task execution paths can be:

[0113] Task A is scheduled for days 1-2, Task B for days 3-5, Task C for days 6-10, and Task D for days 11-12. The final execution order of the third task needs to be verified to ensure that it meets time constraints, such as a total duration of 12 days and no violation of task dependencies. If it does, a final task execution schedule is generated; if it does not, the scheduling order or resource allocation can be adjusted backtracked.

[0114] Priority calculation ensures clear task logic, conflict detection avoids resource bottlenecks, linear programming improves resource efficiency, and the final schedule guarantees controllable project timelines. Each step supports the others, forming a complete task scheduling system suitable for scenarios with high timeliness requirements, such as software development or production line task management.

[0115] A specific implementation of the conflict detection algorithm can be based on resource occupancy checks within a time window. For example, it checks whether the time windows of tasks C and D overlap; if they do, it calculates whether the resource requirements exceed two servers. Conflict detection algorithms are intuitive and easy to implement, making them particularly suitable for resource-constrained scenarios. In linear programming optimization, a virtual resource allocation scenario can be introduced, assuming an additional server is added, to analyze whether this can shorten the time consumed by task C, thereby optimizing the overall project duration. Linear programming optimization enriches the flexibility of resource allocation while maintaining consistency with the core solution.

[0116] The multi-task collaborative processing method for industrial vehicle computers of this invention extracts the original time-series sequence from historical driving data, preprocesses the original time-series data to obtain a smooth driving behavior sequence, extracts the original features from the smooth driving behavior sequence, and then uses principal component analysis to accurately extract driver habit feature vectors containing driver driving habits from the original features. Next, by performing two behavior classifications on the driver habit feature vectors, several driving behavior patterns are obtained. These driving behavior patterns can provide personalized driving suggestions to the driver, which helps to optimize the classification and management of driving behavior. Then, a corresponding priority allocation rule is constructed for each driving behavior pattern, and real-time driving features are extracted from real-time driving data. The system matches driving behavior patterns in the feature and behavior pattern library to obtain matching driving behavior patterns. Based on the matching driving behavior patterns, the corresponding priority allocation rules are retrieved to obtain the first task execution order. Then, combined with the timeliness requirements of the tasks and the optimal task execution time, the system is optimized to obtain the final third task execution order. The third task execution order can reasonably allocate resource capacity, effectively avoid resource waste, and ensure the completion of high-priority tasks. It can realize high-speed collaborative processing of multiple tasks and can adaptively adjust the task scheduling strategy of the vehicle system according to the driver's behavior characteristics and the current driving environment. While ensuring the real-time performance of critical tasks, it can improve the system resource utilization and effectively balance the real-time performance and resource efficiency of the vehicle system.

[0117] Please see Figure 2 The method may also include the following steps:

[0118] S2011. Obtain historical driving data and extract the raw time sequence from the historical driving data. The raw time sequence can be directly extracted from the vehicle's CAN bus, etc.

[0119] S2012. Perform data cleaning on the original time series, removing duplicate and outlier values ​​and marking missing points to obtain the first time series. Data cleaning removes duplicate and outlier values ​​from the original time series and marks missing points, thus obtaining the first time series. This first time series has undergone preliminary standardization, removing redundant and erroneous data points to ensure the accuracy of subsequent data analysis.

[0120] S2013. Perform a moving average filter on the first time series to remove noisy data and obtain the second time series. For the first time series, a moving average filter is used for denoising. Specifically, if the difference between a point in the first time series and its preceding and following points exceeds a preset difference, the average of the preceding and following points is used to replace that point, thus obtaining the second time series. This denoising method smooths out abrupt changes in the first time series, reflecting a more realistic historical driving behavior trend and helping to reduce misjudgments in subsequent data analysis.

[0121] S2014. Linear interpolation is performed on the second time series to fill in the missing points, resulting in the third time series. For the second time series, linear interpolation is used to fill in the missing points. Specifically, if there are marked missing points in the second time series, linear interpolation is calculated based on the values ​​of the preceding and following points to obtain the third time series. This method of filling in missing points preserves the continuity of the data, avoids abrupt jumps, and improves the reliability of the analysis.

[0122] S2015. Perform time-series data analysis on the third time series to extract driving behavior features. Perform Fast Fourier Transform on the driving behavior features to determine the frequency distribution of the driving behavior features. Obtain smoothed driving behavior features based on the frequency distribution. In this embodiment, the driving behavior features include the distribution of acceleration change rate, steering angle change rate, and braking frequency.

[0123] Frequency analysis of the rate of change of acceleration may show a predominance of low-frequency components, indicating smooth driving behavior. Conversely, a high proportion of high-frequency components in the rate of change of acceleration may suggest frequent sudden maneuvers. Sudden acceleration or braking leads to a surge in the rate of change, while smooth driving results in a gradual change. Therefore, in this embodiment, when performing time-series data analysis on the third time-series sequence, the main features extracted are the rate of change of acceleration (e.g., from 1 to 2), the rate of change of steering angle (e.g., from 10 to 15), and the distribution characteristics of braking frequency (e.g., braking frequency of 2 to 3 times per minute). These driving behavior features reflect the smoothness of driving. By performing time-series data analysis on the third time-series sequence, the periodicity of driving behavior can be determined, which helps in evaluating driver style or optimizing autonomous driving algorithms.

[0124] Fast Fourier Transform (FFT) can convert these driving behavior features into frequency distributions. By reflecting the smoothness of driving behavior, these features can be used to determine whether a driver tends to operate smoothly. This can be used for insurance pricing or driver training, which not only improves data utilization but also provides reliable support for intelligent transportation systems.

[0125] Steps S2011 to S2015, from cleaning to feature extraction, adopt a progressive data processing approach to ensure data quality while uncovering deeper information contained in the data.

[0126] S202. Extract the original features from the smooth driving behavior sequence, and determine the driver's habit feature vector from the original features. Refer to step S102 for details.

[0127] S203. Classify the driver's habit feature vector to obtain several driving behavior patterns, and construct a corresponding priority allocation rule for each driving behavior pattern. Refer to step S103 for details.

[0128] S204. Acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain matching driving behavior patterns, retrieve the corresponding priority allocation rules according to the matching driving behavior patterns, and obtain the first task execution order according to the priority allocation rules. Refer to step S104 for details.

[0129] S205. Determine whether there is a circular dependency among tasks according to the priority allocation rules. Adjust the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task. Allocate resources according to the execution order of the second task and the priority allocation rules to obtain the initial resource allocation scheme. Refer to step S105 for details.

[0130] S206. Obtain the resource occupancy threshold of the task based on the initial resource allocation scheme, and adjust the initial resource allocation scheme according to the resource occupancy threshold to obtain an optimized resource allocation scheme. Refer to step S106 for details.

[0131] S207. Adjust the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, and execute the tasks according to the execution order of the third task. For details, refer to step S107.

[0132] Please see Figure 3 The method may also include the following steps:

[0133] S301. Obtain historical driving data, extract the raw time series from the historical driving data, and preprocess the raw time series data to obtain a smooth driving behavior sequence. Refer to step S101 for details.

[0134] S3021. Align the data in the smooth driving behavior sequence to a unified time axis, and extract the original features from the data-aligned smooth driving behavior sequence.

[0135] In this embodiment, a timestamp alignment algorithm is used to handle the time discrepancies between different vehicle sensor data. Since different vehicle sensors may have slight time synchronization errors, such as accelerometer data lagging behind brake data by 0.02 seconds, timestamp alignment will use linear interpolation or nearest neighbor methods to align all data to a unified time axis.

[0136] For example, the acceleration at a certain moment is 2.3 m / s². 2 The steering angle is 12°, but the braking data is missing. This can be estimated as 0 by interpolating the front and rear braking states. The data alignment method of timestamp alignment ensures the accuracy of subsequent feature calculations.

[0137] Alternatively, the rate of change of acceleration, steering angular velocity, and braking frequency can be calculated using the sliding window method to obtain the required raw features. These raw features are then aggregated to form the raw feature set. For example, the rate of change of acceleration is calculated by dividing the acceleration difference within the window by the time interval. Assuming the acceleration within the window starts at 2.0 m / s²... 2 Change to 2.5 m / s 2 Therefore, the rate of change of acceleration is 0.5 m / s². 3 The calculation method for steering angular velocity is similar; assuming the steering angle within the window ranges from 10° to 15°, the steering angular velocity is 5° / s. Braking frequency is calculated by counting the number of braking actions within the window. This feature extraction method can effectively capture the dynamic changes in driving behavior.

[0138] S3022. Normalize the acceleration change rate and steering angular velocity in the original features, and discretize and normalize the braking frequency to obtain standardized features. In this embodiment, for the original feature set, the z-score normalization method is used to normalize the acceleration change rate and steering angular velocity, and the frequency interval division method is used to discretize and normalize the braking frequency to obtain standardized features. The above standardized features are then combined to obtain a standardized feature set.

[0139] Preferably, the acceleration rate of change and steering angular velocity are normalized using the z-score normalization method, which eliminates dimensional differences and facilitates subsequent analysis. It is assumed that the mean value of the acceleration rate of change is 0.4 m / s². 3 The standard deviation is 0.1 m / s 3 After standardization, the value is 1; dividing the braking frequency into frequency ranges divides the frequency into three ranges: low, medium, and high. Discretization simplifies the processing of frequency characteristics. For example, 0-1Hz is set as low frequency, 1-3Hz as medium frequency, and above 3Hz as high frequency. If a certain frequency is actually calculated to be 2Hz, it can be classified into the medium frequency range.

[0140] S3023. Determine the variance among standardized features. If the variance among standardized features exceeds the preset variance, calculate the feature weight corresponding to each standardized feature using the entropy method. If the variance among standardized features does not exceed the preset variance, equally divide the feature weight corresponding to each standardized feature.

[0141] In this embodiment, a preset variance, such as 0.8, can be set in advance. If the variance of the standardized feature set is greater than the preset variance, the weights are calculated using the entropy method. The entropy method is based on the information entropy of the features; the lower the entropy, the higher the feature's discriminative power, and the greater the weight. If the variance does not exceed the preset variance, the weights are equally distributed at 0.33. This dynamic weight allocation highlights the contribution of key features.

[0142] S3024. Perform dimensionality reduction processing on the standardized features with feature weights to obtain dimensionality-reduced features, determine the information retention rate of the dimensionality-reduced features, and take the dimensionality-reduced features with information retention rates exceeding the preset retention rate as driver habit feature vectors.

[0143] In this embodiment, a preset retention rate, such as 90%, can be set in advance. Principal component analysis (PCA) is used to reduce the dimensionality of the standardized feature set with assigned feature weights, retaining principal components with an information retention rate higher than the preset retention rate. Assuming the standardized feature set contains three dimensions—acceleration rate of change, steering angular velocity, and braking frequency—PCA may project the data onto two principal components, retaining 95% of the information. For example, the dimensionality-reduced feature vector might be represented as [0.7, 0.4], reflecting the driver's smoothness and steering habits. This dimensionality reduction reduces computational complexity while preserving core driving features. Specifically, if the dimensionality of the reduced features is higher than the preset dimension, the higher-dimensional features are then regularized to obtain feature vectors of uniform dimension. This avoids overfitting of high-dimensional data and effectively reduces computational complexity.

[0144] S303. Classify the driver's habit feature vector to obtain several driving behavior patterns, and construct a corresponding priority allocation rule for each driving behavior pattern. Refer to step S103 for details.

[0145] S304. Acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain matching driving behavior patterns, retrieve the corresponding priority allocation rules according to the matching driving behavior patterns, and obtain the first task execution order according to the priority allocation rules. Refer to step S104 for details.

[0146] S305. Determine whether there is a circular dependency among tasks according to the priority allocation rules. Adjust the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task. Allocate resources according to the execution order of the second task and the priority allocation rules to obtain the initial resource allocation scheme. Refer to step S105 for details.

[0147] S306. Obtain the resource occupancy threshold of the task according to the initial resource allocation scheme, and adjust the initial resource allocation scheme according to the resource occupancy threshold to obtain the optimized resource allocation scheme. For details, refer to step S106.

[0148] S307. Adjust the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, and execute the tasks according to the execution order of the third task. For details, refer to step S107.

[0149] Please see Figure 4 The method may also include the following steps:

[0150] S401. Obtain historical driving data, extract the raw time series from the historical driving data, preprocess the raw time series data to obtain a smooth driving behavior sequence. Refer to step S101 for details.

[0151] S402. Extract the original features from the smooth driving behavior sequence, and determine the driver's habit feature vector from the original features. Refer to step S102 for details.

[0152] S4031. Cluster the driver's habitual feature vectors to obtain several clusters. If the variance within a cluster is determined to be below a preset threshold, then merge the clusters with variance below the preset threshold to obtain the initial behavior pattern. This is understandable.

[0153] Each cluster represents one of the initial behavioral patterns, and each pattern contains several driver habit feature vectors. Clustering algorithms in unsupervised learning are generally used as the basis for other algorithm analyses. Clustering data allows for the analysis of some characteristics of the data as a whole. Specifically, the K-means algorithm for unsupervised learning clustering divides driver habit feature vectors into multiple clusters based on the similarity within the data, maximizing the similarity within clusters and minimizing the similarity between clusters. Preferably, the initial number of clusters can be set to 3, representing aggressive, stable, and conservative driving patterns, respectively. Aggressive driving patterns may be characterized by large speed fluctuations, high acceleration peaks, and high braking frequency; stable driving patterns have gradual speed changes and low braking frequency. After clustering, if the variance within a cluster is less than 0.01, it indicates that the samples within the cluster are highly similar and can be merged into neighboring clusters. For example, two conservative driving clusters can be merged into one cluster because their speed fluctuations are both below 5 km / h; this merging improves the discriminative power of the behavioral patterns.

[0154] S4032. Calculate the behavioral weights of each driver habit feature vector in the initial behavior pattern using the entropy method, and analyze the contribution of the features to the clustering results. Then, use hierarchical clustering to divide the initial behavior pattern with behavioral weights into patterns to obtain the driving behavior pattern.

[0155] S4033. Determine the distance between each driving behavior mode based on the behavior weight. If the distance is lower than the first preset value, merge the driving behavior modes that are lower than the first preset value to obtain the final set of driving behavior modes.

[0156] In this embodiment, a first preset value, such as 0.05, can be pre-set. When performing a secondary division of the initial behavior patterns using hierarchical clustering, the distance between driving behavior patterns can be calculated based on feature weights. For example, if the speed feature weights of two driving behavior patterns are both 0.3, and the difference in acceleration weights is less than 0.1, and the distance is lower than a second preset value, then they are merged into a single pattern. Assuming there are initially 5 driving behavior patterns, after the secondary division, they are merged into 3 driving behavior patterns, generating a set of driving behavior patterns including aggressive, stable, and mixed patterns. This setting refines the granularity of behavior classification. In particular, if the behavior weight of a driver's habit feature vector is lower than the pre-set second preset value, the corresponding feature is deleted, and the weight of each driver's habit feature vector in the initial behavior pattern is obtained.

[0157] In this embodiment, a second preset value, such as 0.25, can be pre-set. If the braking frequency weight is lower than the second preset value, the feature is deleted, generating a feature weight set containing only speed, acceleration, and steering angle. This filtering method ensures that only the features most discriminative of the behavior pattern are retained. Taking the analysis of urban taxi driving behavior as an example, a driver who brakes frequently and exhibits large acceleration variations during peak hours is classified as having an aggressive driving mode; while maintaining a stable speed during low-traffic periods at night, they are classified as having a stable driving mode. By classifying driving behavior patterns, personalized driving suggestions can be provided to drivers, which helps optimize the classification and management of driving behavior.

[0158] S4034. Construct corresponding priority allocation rules for each driving behavior mode, and store the driving behavior mode in a pre-set behavior mode library and the priority allocation rules in a pre-set rule library.

[0159] S404. Acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain matching driving behavior patterns, retrieve the corresponding priority allocation rules according to the matching driving behavior patterns, and obtain the first task execution order according to the priority allocation rules. Refer to step S104 for details.

[0160] S405. Determine whether there is a circular dependency among tasks according to the priority allocation rules. Adjust the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task. Allocate resources according to the execution order of the second task and the priority allocation rules to obtain the initial resource allocation scheme. Refer to step S105 for details.

[0161] S406. Obtain the resource occupancy threshold of the task according to the initial resource allocation scheme, and adjust the initial resource allocation scheme according to the resource occupancy threshold to obtain the optimized resource allocation scheme. For details, refer to step S106.

[0162] S407. Adjust the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, and execute the tasks according to the execution order of the third task. For details, refer to step S107.

[0163] Please see Figure 5 The method may also include the following steps:

[0164] S501. Obtain historical driving data, extract the raw time series from the historical driving data, preprocess the raw time series data to obtain a smooth driving behavior sequence. Refer to step S101 for details.

[0165] S502. Extract the original features from the smooth driving behavior sequence, and determine the driver's habit feature vector from the original features. Refer to step S102 for details.

[0166] S503. Classify the driver's habit feature vector to obtain several driving behavior patterns, and construct a corresponding priority allocation rule for each driving behavior pattern. Refer to step S103 for details.

[0167] S504. Acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain matching driving behavior patterns, retrieve the corresponding priority allocation rules according to the matching driving behavior patterns, and obtain the first task execution order according to the priority allocation rules. Refer to step S104 for details.

[0168] S5051. Determine the dependencies between tasks according to the priority allocation rules, and determine the estimated duration of each task based on the dependencies and the completion time of the tasks.

[0169] In this embodiment, the task order and priority weight are obtained from the task priority table. The task order and priority weight can reflect the dependencies between tasks. Then, the estimated duration of each task is calculated through the state transition equation, which is defined as:

[0170]

[0171] in, Indicates the first The estimated duration of each task; Indicates the relationship with the first The estimated duration of the dependent tasks that have dependencies on the first task. It should be noted that for the first task... Each task may have multiple dependent tasks with dependencies. Indicates the first The completion time of each task.

[0172] For example, in a task scheduling scenario of an autonomous driving system, suppose there are three tasks: A, B, and C. Task A is path planning, task B is obstacle detection, and task C is speed adjustment. The task priority table shows that task A has a priority weight of 0.8, task B has a priority weight of 0.7, and task C has a priority weight of 0.5. The task order indicates that task B depends on the result of task A, while task C depends on the output of task B. Based on this, the state transition equation can be used to calculate the estimated duration of these three tasks. Specifically, assuming task A takes 3 seconds to complete, task B takes 2 seconds, and task C takes 1 second, then the estimated time for task B is the 3 seconds of task A plus its own 2 seconds, which is 5 seconds; the estimated time for task C is the 5 seconds of task B plus its own 1 second, which is 6 seconds. This method clearly quantifies the task duration and dependency order, ensuring rigorous scheduling logic.

[0173] S5052. Based on the expected duration and dependency order of the tasks, construct a topological sorting sequence of the tasks. Determine whether there is a circular dependency between the tasks based on the topological sorting sequence. If a circular dependency is determined to exist, adjust the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task without cycles.

[0174] When constructing a topological sorting sequence, if there are no circular dependencies, the execution order of the first task is directly Task A → Task B → Task C. Suppose a new task D is added, and Task D depends on Task C, while Task C in turn depends on Task D, forming a circular dependency. In this case, priority weights need to be adjusted. However, if the priority weight of Task C (0.5) is lower than the priority weight of Task D (0.6), Task D can be executed first, breaking the original circular dependency. The execution order of the first task is then adjusted to Task A → Task B → Task D → Task C, resulting in a new execution order of the second task. This adjustment ensures a cycle-free task execution sequence, solves the task loop problem, optimizes the priority execution of critical tasks, and improves system response efficiency.

[0175] S5053. Obtain resource capacity. Based on the execution order of the second task and the resource capacity, allocate resources using a forward recursive strategy. If it is determined that the resource capacity is insufficient, reallocate resources according to the expected duration and priority weight of the task to obtain an initial resource allocation scheme.

[0176] After obtaining the execution order and resource capacity of the second task, the forward recursion strategy can be used for resource allocation. Assume the system has 10 computing units. Task A requires 4 computing units, Task B requires 3, Task C requires 2, and Task D requires 3. The forward recursion starts by allocating 4 computing units to Task A, leaving 6 units remaining; then allocates 3 more to Task B, leaving 3 more; then allocates 2 more to Task C, leaving 1 more; Task D requires 3 computing units but lacks sufficient resources. At this point, resource allocation is redistributed based on the expected duration and priority of the tasks. For example, the computing units for Task C might be reduced to 1, freeing up 1 unit for Task D, ensuring all tasks execute smoothly.

[0177] In this embodiment, after the initial resource allocation scheme is generated, its feasibility needs further verification. For example, if task A is a critical task, insufficient allocation may lead to path planning delays, affecting subsequent tasks. By ensuring sufficient resources for task A through priority weighting, and dynamically adjusting other tasks, both real-time requirements are met and overall stability is improved. This resource allocation scheme effectively avoids resource waste and ensures the completion of high-priority tasks. It can also adapt to the insertion of sudden tasks, enhancing system flexibility. At the same time, by quantifying duration, optimizing sorting, and dynamically allocating, it significantly improves the rationality of task scheduling and execution efficiency. For example, during peak hours, quickly adjusting resources for obstacle detection tasks can effectively reduce collision risks. This multi-faceted resource allocation scheme, from core dependency analysis to resource optimization, forms a consistent scheduling logic, ensuring the efficient operation of the autonomous driving system.

[0178] S506. Obtain the resource occupancy threshold of the task according to the initial resource allocation scheme, and adjust the initial resource allocation scheme according to the resource occupancy threshold to obtain the optimized resource allocation scheme. For details, refer to step S106.

[0179] S507. Adjust the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, and execute the tasks according to the execution order of the third task. For details, refer to step S107.

[0180] Please see Figure 6 The method may also include the following steps:

[0181] S601. Obtain historical driving data, extract the raw time series from the historical driving data, preprocess the raw time series data to obtain a smooth driving behavior sequence. Refer to step S101 for details.

[0182] S602. Extract the original features from the smooth driving behavior sequence, and determine the driver's habit feature vector from the original features. Refer to step S102 for details.

[0183] S603. Classify the driver's habit feature vector to obtain several driving behavior patterns, and construct a corresponding priority allocation rule for each driving behavior pattern. Refer to step S103 for details.

[0184] S604. Acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain matching driving behavior patterns, retrieve the corresponding priority allocation rules according to the matching driving behavior patterns, and obtain the first task execution order according to the priority allocation rules. Refer to step S104 for details.

[0185] S605. Determine whether there is a circular dependency among tasks according to the priority allocation rules. Adjust the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task. Allocate resources according to the execution order of the second task and the priority allocation rules to obtain the initial resource allocation scheme. Refer to step S105 for details.

[0186] S606. Obtain the resource occupancy threshold of the task according to the initial resource allocation scheme, and adjust the initial resource allocation scheme according to the resource occupancy threshold to obtain the optimized resource allocation scheme.

[0187] S6061. Obtain the resource occupancy ratio of all tasks with response times not exceeding a preset time according to the initial resource allocation scheme, and obtain the resource occupancy threshold. In this embodiment, high-time-sensitivity tasks refer to tasks with strict response time requirements, such as real-time data processing or critical business operations, i.e., tasks with response times not exceeding a preset time. Specifically, receive the resource capacity allocation amount of high-time-sensitivity tasks in the initial allocation scheme, and obtain the above-mentioned resource occupancy threshold by traversing the resource capacity occupancy ratio of each allocated task. If the resource occupancy threshold is lower than the preset standard, use a genetic algorithm to initialize the population and iteratively optimize to generate an optimized allocation scheme.

[0188] In a task scheduling system, task A is for real-time log analysis, initially allocated 4 CPU cores, while task B is for database synchronization, initially allocated 2 CPU cores. Recording the resource allocation for each task provides foundational data for subsequent analysis. This method ensures transparency in resource allocation, facilitating future optimization. When iterating through the resource allocation and utilization ratios, the resource allocation for each task can be compared to the total system resources. Assuming the total system resources are 16 CPU cores, task A utilizes 4 / 16 = 25%, and task B utilizes 2 / 16 = 12.5%. Assuming both tasks A and B are time-sensitive, summing the resource utilization ratios of all time-sensitive tasks yields a resource utilization threshold, such as 25% + 12.5% ​​= 37.5%.

[0189] S6062. If the resource occupancy threshold is determined to be lower than the preset standard, a genetic algorithm is used to redistribute the resource capacity to obtain an optimized resource allocation scheme.

[0190] Understandably, if the new resource occupancy threshold is close to the preset standard, it indicates that the system's resource utilization is more efficient, avoiding resource waste or task delays. In this embodiment, a preset standard can be set in advance, such as 50%. If the obtained resource occupancy threshold is lower than the preset standard, a genetic algorithm is used to optimize the allocation. The genetic algorithm initializes a set of resource allocation schemes as a population by simulating natural selection, and iterates and optimizes each generation through crossover and mutation operations. For example, the initial population includes scheme 1: task A is allocated 5 CPU cores, and task B is allocated 1 CPU core; scheme 2: task A is allocated 3 CPU cores, and task B is allocated 3 CPU cores. After iteration, a new scheme is generated: task A is allocated 4 CPU cores, and task B is allocated 2 CPU cores, increasing the overall occupancy threshold to 43.75%. Iterative optimization is a multi-round optimization process, which balances resource allocation efficiency and task timeliness through multiple rounds of optimization.

[0191] Simultaneously, the optimized allocation scheme can map resource requirements to a resource pool, taking into account the dynamic nature of the resource pool. The resource pool may contain different types of computing nodes, such as high-performance nodes and low-power nodes. Assuming the optimized scheme requires task A to allocate 4 CPU cores, high-performance nodes are prioritized to meet timeliness requirements, while task B is allocated 2 CPU cores, which low-power nodes can handle. This mapping process ensures maximum resource utilization while meeting task performance requirements. After reallocating resource capacity, the resource allocation for high-timeliness tasks needs to be iterated again to calculate the new resource occupancy ratio. Assuming task A still has 4 CPU cores, and task B is adjusted to 3 CPU cores, the new occupancy threshold is 4 / 16 + 3 / 16 = 43.75%, ensuring that resource allocation better matches task requirements. By optimizing resource allocation around high-timeliness tasks, a closed-loop process is formed. This closed-loop design ensures continuous improvement of the resource allocation scheme, adapting to dynamic task needs.

[0192] S607. Adjust the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, and execute the tasks according to the execution order of the third task. For details, refer to step S107.

[0193] The method may further include, after step S107:

[0194] S108. Based on the third task execution order, obtain the latency of tasks with priority weights lower than the preset priority. If the latency exceeds the preset degradation threshold, reduce the resource capacity of tasks with priority weights lower than the preset priority to obtain the fourth task execution order, and execute the tasks according to the fourth task execution order. Specifically, obtain the latency of low-priority tasks with priority weights lower than the preset priority from the third task execution order. If the latency exceeds the preset degradation threshold, a trigger signal is generated. The rule engine is then activated based on the trigger signal. The rule engine is responsible for obtaining the pre-allocated resource capacity of the aforementioned low-priority tasks from the third task execution order, and then dynamically reducing the resource capacity using a computational cost reduction algorithm to determine the simplified resource capacity. A lightweight task execution topology can be generated based on the simplified resource capacity. Then, topology optimization tools are used to reorder the dependencies in the topology sorting sequence to obtain an optimized topology sorting sequence. Finally, the task execution order is extracted from the optimized topology sorting sequence to obtain the final optimized task processing flow, i.e., the fourth task execution order. The computational load reduction algorithm reduces the processing volume of secondary data, such as keeping only backups of critical files, thus reducing the computational load, i.e. the required resource capacity, to 50GB and 2000 queries.

[0195] Lightweight topology reduces unnecessary dependency waiting and avoids resource idleness. This adaptive dynamic optimization is especially suitable for scenarios with scarce resources or large fluctuations in workload, and has strong practical value.

[0196] For example, suppose a low-priority task is data backup, with a normal latency of 10 minutes. The real-time monitoring tool collects latency data every minute and finds that the execution latency reaches 15 minutes. The preset degradation threshold is 12 minutes, which has obviously exceeded the threshold, thus generating the trigger signal mentioned above.

[0197] Preferably, the trigger signal can also be obtained by comparing real-time latency with historical latency. For example, by comparing it with the historical average latency (such as the average of 9 minutes over the past week), it can be determined whether the adaptive degradation condition is met. This comparison method is not only based on single data points but also incorporates historical trends, which can more accurately reflect the abnormal state of the task.

[0198] The system provided by the embodiments of the present invention will be described below. The system described below can be referred to in correspondence with the method described above.

[0199] Please see Figure 7 , Figure 7 This diagram illustrates the structure of a multi-task collaborative processing system for an industrial vehicle-mounted computer according to an embodiment of the present invention. The system may include:

[0200] The sequence acquisition module 10 is used to acquire historical driving data, extract the original time series from the historical driving data, preprocess the original time series data, and obtain a smooth driving behavior sequence.

[0201] The habit acquisition module 20 is used to extract the original features from the smooth driving behavior sequence and determine the driver habit feature vector from the original features.

[0202] The behavior classification module 30 is used to classify the driver's habit feature vector to obtain several driving behavior patterns. For each driving behavior pattern, a corresponding priority allocation rule is constructed. The priority allocation rule includes the included tasks, the order of each task, and the priority weight.

[0203] The first sequence module 40 is used to acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain matching driving behavior patterns, retrieve the corresponding priority allocation rules according to the matching driving behavior patterns, and obtain the first task execution order according to the priority allocation rules.

[0204] The second sequence module 50 is used to determine whether there are circular dependencies among tasks according to the priority allocation rules, adjust the execution order of the first task according to the circular dependencies and priority weights, and obtain the execution order of the second task. Resources are then allocated according to the execution order of the second task and the priority allocation rules to obtain an initial resource allocation scheme.

[0205] The resource optimization module 60 is used to obtain the resource occupancy threshold of the task according to the initial resource allocation scheme, and adjust the initial resource allocation scheme according to the resource occupancy threshold to obtain an optimized resource allocation scheme.

[0206] The sequence determination module 70 is used to adjust the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, and execute the task according to the execution order of the third task.

[0207] The multi-task collaborative processing system for industrial vehicle computers of this invention extracts raw time-series data from historical driving data, preprocesses the raw time-series data to obtain smooth driving behavior sequences, extracts raw features from the smooth driving behavior sequences, and then uses principal component analysis to accurately extract driver habit feature vectors containing driver driving habits from the raw features. Next, it performs two behavior classifications on the driver habit feature vectors to obtain several driving behavior patterns. These driving behavior patterns can provide drivers with personalized driving suggestions, which helps to optimize the classification and management of driving behavior. Then, it constructs corresponding priority allocation rules for each driving behavior pattern and extracts real-time driving features from real-time driving data. The system matches driving behavior patterns in the feature and behavior pattern library to obtain matching driving behavior patterns. Based on the matching driving behavior patterns, the corresponding priority allocation rules are retrieved to obtain the first task execution order. Then, combined with the timeliness requirements of the tasks and the optimal task execution time, the system is optimized to obtain the final third task execution order. The third task execution order can reasonably allocate resource capacity, effectively avoid resource waste, and ensure the completion of high-priority tasks. It can realize high-speed collaborative processing of multiple tasks and can adaptively adjust the task scheduling strategy of the vehicle system according to the driver's behavior characteristics and the current driving environment. While ensuring the real-time performance of critical tasks, it can improve the system resource utilization and effectively balance the real-time performance and resource efficiency of the vehicle system.

[0208] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-task collaborative processing method for an industrial vehicle-mounted computer, characterized in that: The method includes: Acquire historical driving data, extract raw time series from the historical driving data, preprocess the raw time series data to obtain smooth driving behavior sequences; The original features are extracted from the smooth driving behavior sequence, and the driver habit feature vector carrying the driver's driving habits is determined from the original features. The driver's habit feature vector is classified into several driving behavior patterns. A corresponding priority allocation rule is constructed for each driving behavior pattern. The priority allocation rule includes the tasks involved, the order of each task, and the priority weight. Acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain matching driving behavior patterns, retrieve the corresponding priority allocation rules according to the matching driving behavior patterns, and obtain the first task execution order according to the priority allocation rules. Determine whether there is a circular dependency among tasks based on the priority allocation rules, adjust the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task, and allocate resources according to the execution order of the second task and the priority allocation rules to obtain the initial resource allocation scheme; The resource occupancy threshold of the task is obtained based on the initial resource allocation scheme, and the initial resource allocation scheme is adjusted based on the resource occupancy threshold to obtain the optimized resource allocation scheme; The execution order of the second task is adjusted according to the optimized resource allocation scheme to obtain the execution order of the third task, and the tasks are executed according to the execution order of the third task.

2. The multi-task collaborative processing method for an industrial vehicle-mounted computer as described in claim 1, characterized in that: The process of classifying driver habit feature vectors into several driving behavior patterns and constructing corresponding priority allocation rules for each driving behavior pattern specifically includes: The driver habit feature vectors are clustered to obtain several clusters. If the variance within a cluster is lower than a preset threshold, the clusters with variances lower than the preset threshold are merged to obtain an initial behavior pattern. Each cluster represents a pattern in the initial behavior pattern, and each pattern contains several driver habit feature vectors. The behavioral weights of each driver habit feature vector in the initial behavior pattern are determined, and the initial behavior patterns with behavioral weights are divided into patterns by hierarchical clustering method to obtain the driving behavior patterns. The distance between each driving behavior pattern is determined based on the behavior weight. If the distance is determined to be lower than the first preset value, the driving behavior patterns that are mutually less than the first preset value are merged. For each driving behavior pattern, a corresponding priority allocation rule is constructed, and the driving behavior pattern is stored in a pre-set behavior pattern library, and the priority allocation rule is stored in a pre-set rule library.

3. The multi-task collaborative processing method for an industrial vehicle-mounted computer as described in claim 2, characterized in that: The process of acquiring real-time driving data, extracting real-time driving features from the real-time driving data, matching the real-time driving features with driving behavior patterns in a behavior pattern library to obtain matching driving behavior patterns, retrieving the corresponding priority allocation rules based on the matching driving behavior patterns, and obtaining the first task execution order based on the priority allocation rules specifically includes: Real-time driving data is obtained through time-series sampling methods to obtain a first data stream carrying timestamp information; The first data stream is processed by a moving average filter to obtain the second data stream; Principal component analysis is performed on the second data stream to obtain a third data stream containing real-time feature vectors; Determine the Euclidean distance of the real-time feature vectors in the third data stream. If the Euclidean distance exceeds a preset environmental complexity threshold, perform vector normalization on the real-time feature vectors that exceed the preset environmental complexity threshold to generate real-time feature vectors. Determine the Euclidean distance between real-time driving features and each driving behavior pattern in the behavior pattern library, and determine the driving behavior pattern that matches the real-time driving features based on the Euclidean distance. Based on the matching driving behavior pattern, the corresponding priority allocation rule is retrieved from the rule base, a task priority table is generated according to the priority allocation rule, and the execution order of the first task is obtained according to the task priority table.

4. The multi-task collaborative processing method for an industrial vehicle-mounted computer as described in claim 1, characterized in that: The process of determining whether there is a circular dependency among tasks based on priority allocation rules, adjusting the execution order of the first task according to the circular dependency and priority weights to obtain the execution order of the second task, and allocating resources according to the execution order of the second task and priority allocation rules to obtain an initial resource allocation scheme, specifically includes: Determine the dependencies between tasks based on priority allocation rules, and determine the estimated duration of each task based on the dependencies and the completion time of each task. Based on the expected duration and dependency order of the tasks, a topological sorting sequence of tasks is constructed. Based on the topological sorting sequence, it is determined whether there is a circular dependency between the tasks. If a circular dependency is determined, the execution order of the first task is adjusted according to the circular dependency and priority weight to obtain the execution order of the second task. Obtain resource capacity, allocate resources using a forward recursive strategy based on the execution order of the second task and the resource capacity, and if it is determined that the resource capacity is insufficient, reallocate resources according to the expected duration and priority weight of the tasks to obtain an initial resource allocation scheme.

5. The multi-task collaborative processing method for an industrial vehicle-mounted computer as described in claim 1, characterized in that: The step of obtaining the resource occupancy threshold of the task based on the initial resource allocation scheme, and adjusting the initial resource allocation scheme based on the resource occupancy threshold to obtain an optimized resource allocation scheme specifically includes: Based on the initial resource allocation scheme, obtain the resource usage ratio of all tasks whose response time does not exceed the preset time, and obtain the resource usage threshold. If the resource occupancy threshold is determined to be lower than the preset standard, a genetic algorithm is used to reallocate the resource capacity to obtain an optimized resource allocation scheme.

6. The multi-task collaborative processing method for an industrial vehicle-mounted computer as described in claim 1, characterized in that: The step of adjusting the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, and then executing the tasks according to the execution order of the third task, specifically includes: Based on the optimized resource allocation scheme and the use of conflict detection algorithms, it is determined whether there is resource competition between tasks. If resource competition exists, i.e. conflict, the execution path of the non-conflicting task is determined. By using conflict-free task execution paths and the estimated duration of tasks, a linear programming algorithm is used to adjust the allocation of resource capacity to obtain the allocation efficiency with the highest resource utilization. Based on the allocation efficiency and conflict-free task execution paths, the execution order of the second task is adjusted to obtain the execution order of the third task.

7. The multi-task collaborative processing method for an industrial vehicle-mounted computer as described in claim 1, characterized in that: The method also includes the following steps: Based on the third task execution order, the latency of tasks with priority weights lower than the preset priority is obtained. When the latency is determined to exceed the preset degradation threshold, the resource capacity of tasks with priority weights lower than the preset priority is reduced to obtain the fourth task execution order, and the tasks are executed according to the fourth task execution order.

8. The multi-task collaborative processing method for an industrial vehicle-mounted computer as described in claim 1, characterized in that: The process of extracting original features from a smooth driving behavior sequence and determining a driver habit feature vector carrying the driver's driving habits from the original features specifically includes: Data from the smoothed driving behavior sequence were aligned to a unified time axis, and raw features were extracted from the aligned smoothed driving behavior sequence; the raw features included the rate of change of acceleration, steering angular velocity, and braking frequency. The acceleration rate of change and steering angular velocity are normalized, and the braking frequency is discretized and normalized to obtain standardized features; Determine the variance among standardized features. If the variance among standardized features exceeds a preset variance, calculate the feature weights corresponding to each standardized feature using the entropy method. If the variance among standardized features does not exceed a preset variance, equally divide the feature weights corresponding to each standardized feature. The standardized features with feature weights are subjected to dimensionality reduction processing to obtain dimensionality-reduced features. The information retention rate of the dimensionality-reduced features is determined, and the dimensionality-reduced features with information retention rates exceeding the preset retention rate are used as driver habit feature vectors.

9. The multi-task collaborative processing method for an industrial vehicle-mounted computer as described in claim 1, characterized in that: The process of acquiring historical driving data, extracting raw time-series sequences from the historical driving data, and preprocessing the raw time-series data to obtain smooth driving behavior sequences specifically includes: Acquire historical driving data and extract the raw time series from the historical driving data; Data cleaning is performed on the original time series to remove duplicate and outlier values, and missing points in the original time series are marked to obtain the first time series. The first time series is subjected to a moving average filter to remove noisy data, resulting in the second time series. Linear interpolation is performed on the second time series to fill in the missing points in the second time series, resulting in the third time series. Time series data analysis is performed on the third time series to extract driving behavior features. Fast Fourier Transform is then applied to the driving behavior features to determine their frequency distribution. Based on the frequency distribution, smoothed driving behavior features are obtained. The driving behavior features include the distribution of acceleration change rate, steering angle change rate, and braking frequency.

10. A multi-task collaborative processing system for an industrial vehicle-mounted computer, characterized in that: The system includes: The sequence acquisition module is used to acquire historical driving data, extract raw time series sequences from the historical driving data, and preprocess the raw time series data to obtain smooth driving behavior sequences. The habit acquisition module is used to extract raw features from the smooth driving behavior sequence and determine the driver habit feature vector carrying the driver's driving habits from the raw features. The behavior classification module is used to classify the driver's habit feature vectors into several driving behavior patterns, and to construct a corresponding priority allocation rule for each driving behavior pattern. The priority allocation rule includes the tasks included, the order of each task, and the priority weight. The first sequence module is used to acquire real-time driving data, extract real-time driving features from the real-time driving data, match the real-time driving features with driving behavior patterns in the behavior pattern library to obtain the matching driving behavior patterns, and retrieve the corresponding priority allocation rules according to the matching driving behavior patterns to obtain the first task execution order according to the priority allocation rules. The second sequence module is used to determine whether there is a circular dependency between tasks according to the priority allocation rules, adjust the execution order of the first task according to the circular dependency and priority weight to obtain the execution order of the second task, and allocate resources according to the execution order of the second task and the priority allocation rules to obtain the initial resource allocation scheme. The resource optimization module is used to obtain the resource occupancy threshold of the task based on the initial resource allocation scheme, and adjust the initial resource allocation scheme according to the resource occupancy threshold to obtain an optimized resource allocation scheme. The sequence determination module is used to adjust the execution order of the second task according to the optimized resource allocation scheme to obtain the execution order of the third task, and then execute the task according to the execution order of the third task.

Images