Distributed component task scheduling strategy supporting cross-platform collaboration

Abstract

The invention discloses a distributed component task scheduling strategy supporting cross-platform collaboration. The distributed component task scheduling strategy comprises the following steps: step 1, collecting computing power, load conditions and network states of different platforms; step 2, carrying out cleaning and standardization treatment; step 3, optimizing task scheduling by using a deep Q network algorithm; 4, deploying the task scheduling model to a monitoring system; 5, adjusting rules according to the priorities and requirements of the tasks and platform load conditions, and dynamically adjusting a task scheduling strategy; step 6, performing anomaly detection on the task scheduling behavior by using a rule engine; 7, regularly analyzing task scheduling results, and optimizing the scheduling model by combining with newly added data; and step 8, optimizing a platform resource allocation strategy by adjusting scheduling algorithm parameters, and updating the model regularly. Through real-time resource perception, dynamic task allocation and an exception handling mechanism, efficient cooperation across multiple heterogeneous platforms is realized, and the resource utilization rate and the task execution efficiency are improved.

Description

Technical Field

[0001] This invention relates to the technical fields of distributed computing and task scheduling, specifically a distributed component task scheduling strategy and system that supports cross-platform collaboration, falling within the technical scope of computer system architecture and optimization. This invention is particularly applicable to multi-platform collaborative environments such as cloud computing, edge computing, and the Internet of Things. It aims to optimize resource utilization through intelligent task scheduling algorithms, improve task execution efficiency, and enhance system stability and adaptability, thereby providing technical support for distributed task management in multiple scenarios. Background Art

[0002] With the rapid development of cloud computing, edge computing, and the Internet of Things (IoT), distributed computing has become a core component of modern computing systems. Distributed component systems effectively improve resource utilization and task processing capabilities by distributing tasks across multiple heterogeneous platforms (such as cloud platforms, edge devices, and local servers). However, due to differences in computing power, network status, and resource load across these platforms, traditional static task scheduling strategies struggle to adapt to the dynamic changes in distributed environments, potentially leading to uneven task allocation, resource waste, or decreased system performance.

[0003] Existing task scheduling strategies are mostly based on rule engines or simple static algorithms, such as scheduling tasks based on fixed priorities or the current state of platform resources. While simple to implement, these methods lack adaptability to complex scenarios, particularly when platform resource loads fluctuate and task demands vary, making it difficult to adjust task allocation strategies in real time. Furthermore, in cross-platform collaboration scenarios, inter-platform communication latency, bandwidth limitations, and resource disparity further exacerbate the complexity of task scheduling, becoming a major bottleneck affecting distributed system performance.

[0004] In recent years, deep learning technology has been widely used in the computer field, demonstrating outstanding performance in pattern recognition, prediction, and decision optimization. In particular, the Deep Q-Network (DQN) algorithm, based on reinforcement learning, utilizes a "state-action-reward" decision-making framework to optimize complex task allocation strategies in dynamic environments. However, there is currently no mature method that combines DQN technology with distributed task scheduling to address the dynamic collaboration of cross-platform resources. Therefore, developing a distributed component task scheduling strategy based on deep learning that supports cross-platform collaboration has important practical significance and application value. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a distributed component task scheduling strategy that supports cross-platform collaboration. Through real-time resource perception, dynamic task allocation and exception handling mechanism, efficient collaboration across multiple heterogeneous platforms is achieved, thereby improving resource utilization and task execution efficiency, and solving the problems of inflexible scheduling and untimely response in the existing technology.

[0006] The object of the present invention is achieved through the following technical solutions: a distributed component task scheduling strategy supporting cross-platform collaboration, comprising:

[0007] Step 1: Deploy the platform resource monitoring module to collect the computing power, load and network status of different platforms in real time;

[0008] Step 2: Clean and standardize the collected data to form a standardized data set;

[0009] Step 3: Use the deep Q-network algorithm to optimize task scheduling, intelligently select the appropriate platform and predict the resource requirements for task execution;

[0010] Step 4: Deploy the optimized task scheduling model to the monitoring system;

[0011] Step 5: Adjust the rules based on the task priority, demand, and platform load, and dynamically adjust the task scheduling strategy;

[0012] Step 6: Use the rule engine to detect anomalies in task scheduling behavior and trigger an alarm mechanism if anomalies are found;

[0013] Step 7: Regularly analyze task scheduling results, optimize the scheduling model based on newly added data, and continuously improve scheduling accuracy and system adaptability through incremental learning.

[0014] Step 8. The system adjusts the scheduling algorithm parameters, optimizes the platform resource allocation strategy, and regularly updates the model to adapt to the dynamic environment and new task requirements to ensure continuous and efficient operation.

[0015] The beneficial effects of the present invention are as follows: the present invention significantly improves the efficiency and accuracy of cross-platform collaborative task scheduling by combining deep reinforcement learning with distributed computing technology, and shows strong adaptability and robustness under dynamic resource environments and diversified task requirements. By introducing the deep Q network (DQN) model and incremental learning mechanism, the system can efficiently optimize the task scheduling strategy, dynamically allocate tasks to the optimal platform, and predict resource requirements to avoid platform overload and resource waste. Combined with real-time monitoring and anomaly detection modules, the system can quickly identify abnormal behaviors such as task delays and insufficient resources, and trigger automated alarms and adjustment mechanisms to ensure the stable execution of tasks. Compared with traditional methods, the present invention has significant advantages in real-time, scheduling efficiency and abnormal response capabilities. At the same time, by continuously optimizing models and parameters, it ensures the long-term stable operation of the system in a complex environment, providing an efficient and intelligent solution for cross-platform distributed task management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a distributed component task scheduling strategy that supports cross-platform collaboration according to the present invention.

[0017] Figure 2 This is a flowchart for resource data cleaning and standardization.

[0018] Figure 3 Flowchart for optimizing task scheduling based on deep Q-network.

[0019] Figure 4 This is the flowchart of anomaly detection and dynamic adjustment. DETAILED DESCRIPTION

[0020] The technical solution of the present invention is further described below with reference to the accompanying drawings.

[0021] like Figure 1 As shown, a distributed component task scheduling strategy supporting cross-platform collaboration of the present invention includes:

[0022] Step 1: Deploy the platform resource monitoring module to collect the computing power, load and network status of different platforms (such as cloud platforms, edge computing nodes, etc.) in real time to provide real-time data support for task scheduling;

[0023] Based on task scheduling requirements, resource monitoring modules are deployed on heterogeneous platforms, including cloud platforms, edge computing nodes, and local servers, to collect real-time information on each platform's computing power, load status, and network conditions. By combining the platform's actual environmental characteristics with task allocation requirements, the system optimally configures the monitoring module's sampling frequency and monitoring scope, ensuring coverage of key resource information such as CPU, memory, and GPU utilization, as well as network latency and bandwidth. A lightweight monitoring agent runs on each node and transmits data to a central monitoring system via standard protocols (such as HTTP or GRPC), ensuring real-time and reliable data transmission. To adapt to dynamic task execution scenarios, the system incorporates a dynamic sampling mechanism. When node resource utilization exceeds a set threshold, the sampling frequency is automatically increased to capture resource fluctuations. For example, the sampling interval is 10 seconds when resource load is low, but when node load exceeds 80%, the sampling frequency is automatically increased to 1 second to capture sudden resource fluctuations. The collected data is then stored in a time series database for cleaning and standardization, providing high-quality, accurate, and real-time data support for subsequent task scheduling optimization.

[0024] In this example, an Amazon EC2 instance (m5.large type) with 2 vCPUs and 8GB of memory was selected in the cloud. NVIDIA Jetson AGX Xavier devices with an 8-core ARM CPU, 32GB of memory, and a 512-core GPU were used at the edge nodes. The Prometheus monitoring agent was installed on each of these nodes, and the time series database InfluxDB was used to store performance data. Data collection was performed on a 10-second cycle, and metrics collected included CPU utilization, memory usage, storage space usage, and network latency. Collected performance data was transmitted via a Kafka message queue, ensuring data reliability and real-time performance in a multi-node architecture. The central server ran the Prometheus service, receiving monitoring data via the gRPC protocol and storing it in a standardized JSON format for subsequent processing modules. Resource consumption of the monitoring agent was optimized for edge node devices. For example, irrelevant modules were removed from the Jetson device, retaining only the monitoring functions for CPU, memory, and network latency. This optimization reduced the monitoring agent's operational resource consumption by approximately 30%, while ensuring comprehensive data collection. The monitoring module also generates a real-time resource distribution chart and uses Matplotlib, a visualization tool developed in Python, to present the load situation, providing high-quality data support for subsequent scheduling algorithms.

[0025] Step 2: Clean and standardize the collected data to form a standardized data set; Figure 2 Shown, including:

[0026] Step 2.1: De-noise the collected data; then perform a cleaning operation to remove invalid data or outliers (such as missing values, duplicate values, or negative values); then correct missing data using preset rules or statistical methods. Mean filling, median replacement, or Lagrange interpolation can be used to process some missing data, and time series analysis methods can be used to correct deviations caused by network jitter; this lays the foundation for subsequent standardization operations.

[0027] Resource monitoring data is collected in real time through Prometheus, and indicators such as system load, CPU usage, memory usage, and network latency are obtained once a minute and stored in the central storage module in JSON format. First, the data is read through the Pandas library to detect and clean up outliers. Outlier detection uses a statistically based Z-score method, setting a threshold of ±3 times the standard deviation to identify load anomalies (such as CPU usage exceeding 100% or memory usage being negative) and eliminate these data. At the same time, the Savitzky-Golay filter is used to smooth the time series data to eliminate sudden noise signals (such as sudden delay spikes). The accuracy of the cleaned data is increased to 99.5%.

[0028] Missing values ​​due to packet loss or interrupted data collection are filled using Lagrange interpolation to ensure data continuity. Specifically, the NumPy library is used to locate missing data locations, and an interpolation model is constructed based on the regularity of time series data. For network delay fields, piecewise linear interpolation is used to minimize the impact of long interruptions on forecast performance. After filling, the integrity rate of key fields is guaranteed to exceed 99%.

[0029] Step 2.2: To eliminate the inconsistency of resource data distribution on different platforms, the cleaned data is standardized and the data consistency test is completed. Normalization uses a normalization method (such as Min-Max normalization) to adjust the data range to the [0, 1] interval to ensure the uniformity of data on different platforms and adapt to the input requirements of the deep Q network. The normalization formula is:

[0030]

[0031] Among them, X is the original data, X min and X max where are the minimum and maximum values ​​of the field, respectively, and X′ is the normalized data. The normalized data is saved in the PostgreSQL database for subsequent scheduling model input.

[0032] Step 2.3: Fuse historical task data with real-time monitoring data. In view of the differences between historical task data and real-time monitoring data, the system fuses the two types of data to build a unified feature representation model. By performing feature selection and dimensionality reduction on the data (principal component analysis (PCA) or Lasso regression can be used to perform feature selection and dimensionality reduction on the data), the most relevant feature information of the two types of data is extracted, and redundant features with little impact on task scheduling are eliminated. Then, data fusion is performed to reduce the computational complexity of the scheduling model and improve prediction efficiency.

[0033] Step 2.4: Store the processed and standardized data in a time series database (TSDB) and divide it into a training set and a validation set for training the load forecasting model and optimizing the scheduling model. The training set contains 80% of historical data and is used to optimize the parameter training of the deep Q network (DQN), while the validation set is used to evaluate the prediction accuracy and stability of the model. The normalized data needs to be converted into the tensor format (Tensor) of the deep Q network (DQN) model to adapt to the model input requirements. In the specific operation, the DataLoader module of PyTorch is used to divide the data into training and test sets, using a distribution ratio of 80% training data and 20% test data, and the size of each batch of data is set to 64. The generated Tensor data files are stored in Amazon S3 cloud storage and automatically loaded into the training module to ensure efficient data transfer.

[0034] Step 3: Use the deep Q network algorithm to optimize task scheduling, intelligently select the appropriate platform and predict the resource requirements for task execution. Figure 3 As shown, specifically including:

[0035] Step 3.1, Task status representation and model input construction; Use the processed standardized data set to extract key task features and platform resource status as input to the deep Q network model; Among them, task features include the priority P of the task t , Estimated execution time D t and resource requirements;

[0036] Define the state-action-reward structure for scheduling tasks. The state consists of the platform's real-time resource data (e.g., CPU utilization, memory usage, network latency, etc.) and the task's computational requirements (e.g., priority, execution time, etc.); the action is the decision to assign the task to the target platform; and the reward function is defined based on task completion time, resource utilization, and scheduling efficiency to guide the model's optimization of the scheduling strategy.

[0037] Step 3.2: Construct a deep Q network (DQN) model. The deep Q network model includes an online network and a target network, which have the same structure but different initial parameters. The network consists of three parts: an input layer, a hidden layer, and an output layer. The input layer receives a state vector S consisting of key task features and platform resource status. The hidden layer contains three fully connected networks with 128, 64, and 32 neurons, respectively. The activation function is ReLU (Rectified Linear Unit). The hidden layer is used to extract the deep relationship between the task and the platform status. The output layer outputs the task scheduling value Q corresponding to each platform. p (CPU requirement C t , memory requirement M t , Bandwidth requirement B t ); The network weight parameters are initialized using the Xavier method to ensure uniform distribution of weight values. The experience pool is initialized to empty.

[0038] Step 3.3, reward function design, the reward function measures the quality of the scheduling strategy, the goal is to maximize task efficiency and resource utilization, while minimizing the platform load imbalance. Reward value R w The calculation formula is as follows:

[0039]

[0040] Among them, T e is the task completion time (unit: seconds), the actual time from the start to the completion of the task; U r The amount of resources consumed by the current task (CPU + memory + bandwidth); U t is the total resource capacity of the platform (total CPU + total memory + total bandwidth); σ L is the standard deviation of the platform load; α1, α2, and α3 are weight coefficients used to adjust the contribution of completion time, resource utilization, and load balancing, and are usually set to 0.4, 0.4, and 0.2.

[0041] Step 3.4: Model Training and Optimization: Train the Deep Q-Network model. Using the experience replay mechanism, extract training samples from the training set obtained in Step 2 and optimize the model parameters. Specifically, the mean squared error (MSE) is used as the loss function, and the parameters are updated based on the target network and the online network. During training, an ε-greedy strategy is used to balance exploration and exploitation, gradually improving the model's scheduling decision-making capabilities in complex scenarios.

[0042] The loss function L(θ) used in the training process is as follows:

[0043]

[0044] in, Indicates expectation, Si is the current state vector; A i is the current action (selected platform); R i is the current reward value; S′ i is the next state vector; a′ is the next state S′ i The set of all possible actions in ; θ is the parameter of the current Q network; θ - is the parameter representing the target Q network; Q(S i ,A i ; θ) is the Q value predicted by the current Q network; Q(S′ i ,a′;θ - ) is the predicted Q value of the target Q network; γ is the discount factor, which is set to 0.9 in this embodiment; through the loss function, the optimal action is calculated in real time based on the input data.

[0045] Step 3.5, Scheduling decision generation and verification: After training, the deep Q network model outputs the task scheduling value Q based on the input state vector S p , select Q p The largest platform serves as the execution node for the task, effectively becoming the optimal target platform for task assignment. Feedback from scheduling decisions is also recorded to optimize the model's long-term performance. Verification using a validation set revealed a 96.5% scheduling success rate, a 23% increase in resource utilization, and significant improvements in load balancing.

[0046] To improve the model's adaptability and generalization capabilities, the system can also introduce an online learning mechanism during the real-time scheduling process. By continuously collecting feedback data on task execution results (such as task delays and resource utilization), the model parameters can be adjusted in real time to enable it to adapt to dynamic changes in the platform's resource status and continuously optimize scheduling efficiency.

[0047] Step 4: Deploy the optimized task scheduling model to the monitoring system. During deployment, the model is exported to the ONNX (Open Neural Network Exchange) format to support hardware acceleration. In specific implementation, the model is loaded onto edge computing devices (such as the NVIDIA Jetson Xavier NX) and cloud servers. The inference process is optimized using the TensorRT framework, ensuring that task allocation decision times are kept within 10 milliseconds. The monitoring system receives real-time task status and platform resource data via a message queue (such as Kafka), gradually inputs the model into inference, and outputs the optimal scheduling solution, including the target platform, estimated execution time, and resource allocation strategy. To ensure system stability and robustness, the deployed model is regularly evaluated through A / B testing, specifically analyzing scheduling performance in multi-task concurrent scenarios. Scheduling parameters are adjusted to accommodate complex task environments, meeting the real-time requirements of high-frequency scheduling scenarios. During deployment, the system incorporates an edge caching mechanism to locally store scheduling results for some common tasks, enabling rapid response to low-priority tasks. At the same time, the scheduling model supports dynamic updates. When the platform resource status mutates or the model version is updated, hot loading can be achieved without interrupting task execution, ensuring the stable operation of the system.

[0048] The task scheduling system employs a multi-level acceleration strategy to optimize inference performance. During deployment, the system uses quantization technology to reduce the storage space and computational complexity of model weights, replacing FP32 calculations with INT8 operations. Furthermore, a hierarchical allocation mechanism for inference tasks is designed between edge devices and cloud servers: high-real-time tasks are prioritized by edge nodes, while complex computational tasks are handled by cloud servers. This balances performance and resource consumption across different task requirements.

[0049] Step 5: Adjust the rules based on task priority, demand, and platform load, and dynamically adjust the task scheduling strategy to ensure efficient execution and avoid resource waste; specifically:

[0050] Step 5.1. When the task scheduling system receives a new task request, it first prioritizes the task according to its characteristics (such as computational complexity, execution time limit, resource requirements, etc.), matches it with the current load information of the platform, and selects a list of platforms that preliminarily meet the task execution conditions; the priority of the task is determined by a preset weight rule, for example, high-priority tasks are preferentially assigned to low-load platforms to reduce task delays. At the same time, the system dynamically analyzes the utilization of platform resources (CPU, memory, bandwidth) and adjusts the allocation order of low-priority tasks in the task queue to avoid resource waste or overload. During the task execution process, the system monitors the task completion progress and resource usage in real time. Once it finds that the platform resource pressure is too high or the task is running abnormally (such as timeout and incomplete), it will automatically adjust the scheduling plan for unassigned tasks to optimize the overall task execution efficiency.

[0051] Step 5.2: The system dynamically calculates the optimal task allocation path based on the platform's resource status, network latency, and task allocation history. It also uses a scheduling optimization strategy based on a deep Q network to prioritize nodes based on the real-time Q value of each platform while avoiding resource allocation conflicts.

[0052] Step 5.3. During the execution of the task, the system monitors the running status of the task and the resource load of the platform in real time; if a load peak or insufficient resources are detected during the execution of the task, the task will be reallocated through a dynamic adjustment mechanism. For example, the task is triggered to migrate to a node with more sufficient resources to ensure the continuity and stability of task execution. The system monitors the execution of the tasks assigned in real time, and dynamically adjusts the task scheduling strategy based on the feedback data of the task execution. For example, based on the deviation between the actual completion time of the task, resource consumption and scheduling expectations, the system determines whether there are any abnormalities in the task scheduling (such as execution delays, insufficient resource usage, etc.) through a rule engine or simple logic. Once an abnormality is detected, the system triggers the dynamic adjustment mechanism to re-evaluate the execution node of the task, or trigger the task to migrate to a platform with more sufficient resources to ensure the smooth completion of the task.

[0053] Step 6: Use the rule engine to detect anomalies in task scheduling behavior. If anomalies are found (such as task delays, insufficient resources, or node downtime), the alarm mechanism will be triggered and the anomaly handling will be performed. Figure 4As shown in the figure, the system dynamically identifies the type of abnormal event by analyzing the real-time feedback data of task execution (such as task completion time, resource consumption, etc.) and combining it with the preset normal behavior pattern. Once the abnormality is confirmed, the system records the detailed information of the abnormal event, including task ID, abnormality type, timestamp and the status of the relevant platform, and triggers the task migration or resource reallocation strategy to ensure that the system can resume normal operation in the shortest time. At the same time, the alarm mechanism triggers multiple notification methods, such as real-time notification of relevant persons in charge through SMS, email and system alarms, so as to intervene and handle it in time. In addition, the system will add the abnormal task to the rescheduling queue and dynamically adjust its resource allocation plan to ensure that other tasks are not affected and continue to run efficiently. For example, migrating the current task to other platforms with lower load can ensure the continuous execution of the task and reduce the impact of the abnormality on the system.

[0054] Step 7: Regularly analyze task scheduling results, optimize the scheduling model based on newly added data, and continuously improve scheduling accuracy and system adaptability through incremental learning.

[0055] Through the log collection module, the system conducts a comprehensive review of historical scheduling data, analyzes key data such as task completion rate, resource allocation efficiency, and platform load changes during the system-aggregated scheduling process, and generates monthly or quarterly analysis reports to evaluate the effectiveness of the current scheduling strategy. Combined with the newly added task execution data, the new data is processed by an automated annotation tool, and the training set and validation set are re-divided. The system uses incremental learning technology to fine-tune the deep Q network (DQN) model, dynamically adjusts the scheduling strategy parameters (such as reward function weights and learning rates), and improves the model's adaptability and scheduling accuracy, thereby adapting to resource changes and new task scenarios, and improving overall scheduling efficiency and reliability. During the model performance optimization process, scheduling delay, task completion rate, and resource utilization are used as evaluation indicators to ensure that the updated model performs better than the previous version in actual applications. In addition, the system uses version management tools to record detailed information for each model update, providing a reference for future model iterations.

[0056] Step 8. The system adjusts the scheduling algorithm parameters, optimizes the platform resource allocation strategy, and regularly updates the model to adapt to the dynamic environment and new task requirements to ensure continuous and efficient operation. During the optimization process, the system uses real-time monitoring data to analyze the adaptability of the current scheduling strategy. In response to large changes in task load, it dynamically adjusts model parameters (such as learning rate, loss function weight, etc.) to enhance the responsiveness of the scheduling model. At the same time, the system will optimize the resource allocation strategy based on resource allocation efficiency and platform performance trends, such as balancing task distribution on nodes with low resource utilization or prioritizing critical tasks during peak load periods to maximize resource utilization efficiency. The updated model and strategy will undergo rigorous performance testing before going online to ensure stability and robustness in complex scenarios, and ultimately achieve long-term and reliable operation of the task scheduling system.

[0057] In summary, the present invention significantly improves the efficiency and stability of distributed component task scheduling by combining the deep Q network (DQN) optimization model, the platform resource monitoring module and the intelligent scheduling strategy. Based on cross-platform resource load prediction, dynamic task allocation and anomaly detection mechanism, the present invention can efficiently adapt to complex distributed environments and show high robustness and accuracy in scenarios such as task priority scheduling, balanced resource allocation and real-time anomaly response. At the same time, the system ensures maximum resource utilization through software and hardware collaborative design and scheduling algorithm optimization, reduces task delays and resource waste, and significantly improves the system's operating efficiency and adaptability.

[0058] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A distributed component task scheduling strategy that supports cross-platform collaboration, characterized in that: include: Step 1: Deploy the platform resource monitoring module to collect the computing power, load and network status of different platforms in real time; Step 2: Clean and standardize the collected data to form a standardized data set; Step 3: Use the deep Q-network algorithm to optimize task scheduling, intelligently select the appropriate platform and predict the resource requirements for task execution; Step 4: Deploy the optimized task scheduling model to the monitoring system; Step 5: Adjust the rules based on the task priority, demand, and platform load, and dynamically adjust the task scheduling strategy; Step 6: Use the rule engine to detect anomalies in task scheduling behavior and trigger an alarm mechanism if anomalies are found; Step 7: Regularly analyze task scheduling results, optimize the scheduling model based on newly added data, and continuously improve scheduling accuracy and system adaptability through incremental learning. Step 8: By adjusting the scheduling algorithm parameters and optimizing the platform resource allocation strategy, the model is regularly updated to adapt to the dynamic environment and new task requirements to ensure continuous and efficient operation.

2. The distributed component task scheduling strategy according to claim 1, characterized in that: Step 2 includes: Step 2.1: De-noise the collected data; then perform a cleaning operation to remove invalid data or outliers; and then correct missing data using preset rules or statistical methods; Step 2.2: Normalize the data: Use normalization to adjust the data range to the [0,1] interval; Step 2.3: Fuse historical task data with real-time monitoring data; Step 2.4: Store the processed standardized data into the time series database to form a standardized data set.

3. The distributed component task scheduling strategy according to claim 1, characterized in that: Step 3 specifically includes: Step 3.1: Task state representation and model input construction: Using the processed standardized dataset, extract key task features and platform resource status as input to the deep Q-network model; Step 3.2: Build a deep Q network model; Step 3.3, reward function design, reward value R w The calculation formula is as follows: Among them, T e is the task completion time, which is the actual time from the start to the completion of the task; U r The amount of resources consumed by the current task; U t is the total resource capacity of the platform; σ L is the standard deviation of the platform load; α1, α2 and α3 are weight coefficients; Step 3.4: Model training and optimization: Train the deep Q-network model, extract training samples from the dataset obtained in step 2 using the experience replay mechanism, and optimize the model parameters. The loss function L(θ) used in the training process is as follows: in, Indicates expectation, S i is the current state vector; A i is the current action; R i is the current reward value; S′ i is the next state vector; a′ is the next state S′ i The set of all possible actions in ; θ is the parameter of the current Q network; θ - is the parameter representing the target Q network; Q(S i ,A i ; θ) is the Q value predicted by the current Q network; Q(S′ i ,a′;θ - ) is the predicted Q value of the target Q network; γ is the discount factor; through the loss function, the optimal action is calculated in real time based on the input data; Step 3.5, Scheduling decision generation and verification: After training, the deep Q network model outputs the task scheduling value Q based on the input state vector S p , select Q p The largest platform serves as the execution node of the task and is the optimal target platform for task allocation.

4. The distributed component task scheduling strategy according to claim 1, characterized in that: Step 5 specifically includes: Step 5.1: When the task scheduling system receives a new task request, it first classifies the priority of the task according to its characteristics, matches it with the current load information of the platform, and selects a list of platforms that preliminarily meet the task execution conditions; Step 5.2: The system dynamically calculates the optimal task allocation path based on the platform's resource status, network latency, and task allocation history. It also uses a scheduling optimization strategy based on a deep Q network to select priority allocation nodes based on the real-time Q value of each platform. Step 5.3: During the task execution process, the system monitors the task's running status and the platform's resource load in real time; if a load peak or insufficient resources are detected during task execution, the task is reallocated through a dynamic adjustment mechanism.

Images