Resource intelligent bus and task parallel hydrogen energy data scheduling optimization method and system

Abstract

The invention provides a resource intelligent bus and task parallel hydrogen energy data scheduling optimization method and system, relates to the technical field of energy, and aims to solve the problems of resource heterogeneity, lack of real-time performance and dynamic adjustment, insufficient energy efficiency optimization, poor adaptability to abnormal working conditions and the like in the prior art. According to the method, a resource intelligent bus architecture is constructed, and the resource intelligent bus architecture comprises a resource virtualization encapsulation layer, a resource state sensing layer and a global prediction and scheduling layer; in the resource state sensing layer, performing multi-granularity task decomposition on hydrogen energy business requirements, and establishing an inter-task dependency relationship model by adopting a directed acyclic graph; in the global prediction and scheduling layer, a multi-objective optimization model is constructed based on a dependency relationship between resource states and tasks, and resources and tasks are dynamically matched according to an optimal solution of the model; and dynamically scheduling the resources according to a matching result of the resources and the tasks. The problems in the prior art are solved, and the hydrogen energy data scheduling optimization performance is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of energy, and particularly relates to a resource intelligent bus and task parallel hydrogen energy data scheduling optimization method and system. BACKGROUND

[0002] The statements in this section merely provide background information related to the application and do not necessarily constitute prior art.

[0003] With the rapid development of the hydrogen energy industry, hydrogen energy system data scheduling faces multi-dimensional technical challenges. The hydrogen energy system contains diversified nodes such as electrolytic hydrogen production equipment, storage and transportation equipment, and fuel cells, generating multi-source heterogeneous data such as time series data and equipment logs. The traditional centralized scheduling architecture has a single point of failure risk and cannot adapt to the dynamic fluctuations of edge computing resources. Although the existing containerization technology based on Kubernetes can realize resource abstraction, it lacks real-time sensing capability for physical resources such as GPU computing power and network bandwidth.

[0004] Hydrogen energy application scenarios have strict real-time requirements for tasks, such as millisecond-level leak detection and second-level supply and demand matching. However, the static resource allocation and coarse-grained task division mechanism in existing scheduling cannot coordinate the parallel needs of computing-intensive tasks (such as hydrogen flow field CFD simulation) and communication-sensitive tasks (such as hydrogenation station safety monitoring), resulting in problems such as task deadline violation rate. And the existing scheduling method does not fully consider the energy efficiency index specific to the hydrogen energy system, resulting in energy waste and insufficient utilization of system resources.

[0005] The mobility of hydrogen energy equipment (such as on-board hydrogen storage tanks) causes unstable network connections, and the existing star-shaped communication architecture based on TCP / IP protocol cannot adapt to dynamic topology changes, causing an increase in data transmission packet loss rate and scheduling jitter amplification.

[0006] The existing scheduling method is inefficient in task coordination. Although heuristic algorithms (such as genetic algorithms or particle swarm optimization algorithms) can optimize local task delays, they ignore the data dependency between tasks (such as the need for pressure sensor data preprocessing before hydrogen production efficiency analysis), which can easily cause task blocking and redundant communication. At the same time, the existing scheduling method lacks adaptability to low-probability high-risk scenarios such as extreme weather and new types of pollutants, for example, there are thousands of potential variants in the aging pattern of the proton exchange membrane of hydrogen fuel cells. Traditional supervised learning is limited by the scarcity of labeled samples and cannot cover all failure modes.

[0007] Currently, although the resource intelligent bus (RIB) technology has shown potential for heterogeneous resource collaboration in the industrial Internet of Things field, the existing RIB implementation scheme does not deeply integrate task parallel optimization mechanisms, especially when dealing with directed acyclic graph (DAG) structured tasks, there are defects such as low accuracy in identifying critical paths and inaccurate modeling of communication overhead. Summary of the Invention

[0008] To overcome the shortcomings of the existing technologies, this invention provides a hydrogen energy data scheduling optimization method and system that integrates intelligent resource bus and task parallelism. It constructs a new scheduling framework that integrates dynamic resource perception and parallel task optimization, realizes fine-grained collaboration of heterogeneous resources in the hydrogen energy system through intelligent bus architecture, and generates the optimal scheduling strategy based on the dual constraints of task dependency and resource status, thereby improving the performance of hydrogen energy data scheduling optimization.

[0009] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a hydrogen energy data scheduling optimization method based on a resource intelligent bus and task parallelism, comprising: Construct a resource intelligent bus architecture, which includes a resource virtualization encapsulation layer, a resource status awareness layer, and a global prediction and scheduling layer; At the resource status perception layer, the hydrogen energy business requirements are decomposed into multi-granularity tasks, and a directed acyclic graph is used to establish a dependency model between tasks. At the global prediction and scheduling layer, a multi-objective optimization model is constructed based on resource status and inter-task dependencies, and resources and tasks are dynamically matched according to the optimal solution of the model. Dynamic resource scheduling is performed based on the matching results between resources and tasks.

[0010] As one implementation method, a resource intelligent bus architecture is constructed, and the specific process is as follows: A resource virtualization encapsulation layer is constructed by standardizing the definition of various resources through a unified resource descriptor. A resource status awareness layer is constructed by using lightweight probes to collect CPU utilization, available memory, real-time bandwidth and link jitter parameters, and setting data transmission strategies. A global prediction and scheduling layer is constructed based on a spatiotemporal joint prediction model to predict resource load changes.

[0011] As one implementation method, a unified resource descriptor includes resource type, unique identifier, computing power, geographical location, and dynamic energy consumption characteristics.

[0012] As one implementation method, the hydrogen energy business requirements are decomposed into multi-granularity tasks. The specific process is as follows: The task semantics of hydrogen energy business requirements are analyzed, and the tasks of hydrogen energy business are divided into computationally intensive tasks, communication-sensitive tasks and data-dependent tasks. A dynamic weight evaluation function is constructed to dynamically evaluate the task.

[0013] As one implementation method, a directed acyclic graph is used to establish a dependency model between tasks. The specific process is as follows: Construct a directed acyclic graph where each node is an atomic task, label the multidimensional attributes of the atomic task nodes, and model the edge weights by calculating the communication time between nodes; Based on the directed acyclic graph, the critical path is identified and optimized.

[0014] As one implementation method, based on a directed acyclic graph, the critical path is identified and optimized. The specific process is as follows: Traverse all execution paths in the directed acyclic graph and calculate the total execution time for each path; Select the path with the longest execution time as the critical path; Prioritize deploying critical path tasks to CPU / high-performance computing acceleration nodes and reserve dedicated bandwidth channels for data transmission; For critical path tasks, a preemptive scheduling mechanism is used to prioritize resource acquisition; Establish checkpoints and task replication mechanisms for critical path tasks to check for path interruptions. By monitoring the task execution status, the scheduling strategy is dynamically updated when the path time relationship is reconstructed.

[0015] As one implementation method, a multi-objective optimization model is constructed based on resource status and inter-task dependencies, and resources and tasks are dynamically matched according to the optimal solution of the model. The specific process is as follows: Multi-objective optimization modeling takes minimizing global execution time and minimizing total system energy consumption as dual objectives, and constructs a multi-objective optimization function. Based on resource status and inter-task dependencies, computationally intensive tasks are allocated to CPU / high-performance computing acceleration nodes, end-to-end optimized paths are built for data-dependent tasks, dedicated communication channels are reserved, and remote direct memory access technology is used for data transmission. For non-critical path tasks, a preemptive delayed scheduling strategy is adopted; Energy consumption optimization is performed from both computing and communication perspectives. Construct constraints for the multi-objective optimization function, including resource capacity constraints, task dependency constraints, deadline constraints, and time series constraints. Under the constraints, the multi-objective optimization function is solved to obtain the optimal solution of the model; Resources and tasks are dynamically matched based on the optimal solution.

[0016] As one implementation method, dynamic resource scheduling is performed based on the matching results of resources and tasks. The specific process is as follows: A priority-based preemptive scheduling mechanism is adopted for resources and tasks, and a multi-dimensional scoring function is constructed to quantify the urgency of tasks, including deadline urgency, security level classification, and resource demand intensity. Establish a preemption protocol so that when a high-priority task arrives, low-priority tasks are paused and the intermediate state is preserved, the occupied computing and channel resources are released, and the resources are reallocated to urgent tasks. For communication-sensitive tasks, a dual-engine collaborative mechanism of bandwidth prediction and path selection is adopted, and bandwidth prediction is performed through a neural network model. A multi-objective decision-making model is constructed to select dynamic paths, and a cross-protocol fast switching technology is used to complete session migration and data resume transmission within a specified time when the path is switched. The task is dynamically and adaptively adjusted based on fault response and resilience indicators; Based on the results of dynamic adaptive adjustment, a closed-loop control of monitoring, decision-making, and action is implemented.

[0017] As one implementation method, the multi-objective decision model formula is: ; in, For transmission time, D / B, where D is the data volume and B is the available bandwidth of the link; Energy consumption per bit; For connection stability index; , , These are all weighting coefficients, indicating the emphasis of the scheduling strategy. For low latency, For high energy efficiency, For high stability, and .

[0018] A second aspect of the present invention provides a hydrogen energy data scheduling optimization system with intelligent resource bus and parallel task operation, comprising: The resource intelligent bus architecture construction module is used to build the resource intelligent bus architecture, which includes a resource virtualization encapsulation layer, a resource status awareness layer, and a global prediction and scheduling layer. The task decomposition and modeling module is used to decompose hydrogen energy business requirements into multi-granularity tasks at the resource status awareness layer, and to establish a dependency model between tasks using a directed acyclic graph. The dynamic matching decision module is used to construct a multi-objective optimization model based on resource status and inter-task dependencies at the global prediction and scheduling layer, and to dynamically match resources and tasks according to the optimal solution of the model. The scheduling execution model is used to dynamically schedule resources based on the matching results of resources and tasks.

[0019] The above one or more technical solutions have the following beneficial effects: In this embodiment, by constructing a Resource Intelligent Bus (RIB) architecture, the dynamic status and load characteristics of computing nodes, storage devices and communication links are perceived in real time. This breaks through the limitations of traditional static resource allocation, realizes the elastic combination and global scheduling of heterogeneous resources at the edge, effectively avoids the risk of single point of failure, and improves resource utilization.

[0020] In this embodiment, a multi-granularity task decomposition and DAG modeling mechanism is designed. Combining the temporal constraints and priority rules of the hydrogen energy scenario, the optimal scheduling scheme is generated through critical path identification and communication overhead modeling, which effectively reduces system scheduling jitter and ensures end-to-end latency of millisecond-level safety monitoring tasks. A resource-task dynamic matching algorithm is proposed, which integrates a bandwidth prediction model and a link jitter evaluation strategy to achieve dynamic optimization of data transmission paths in the network fluctuation environment of mobile hydrogen energy devices, thereby reducing the task interruption rate.

[0021] In this embodiment, the energy efficiency evaluation index of the hydrogen energy system is embedded into a multi-objective optimization model. Through the dual-dimensional collaborative optimization of computational energy consumption and communication energy consumption, the overall energy efficiency is improved compared with traditional methods, thus meeting the low-carbon requirements of green hydrogen energy systems.

[0022] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0024] Figure 1 This is a flowchart of the hydrogen energy data scheduling optimization method with resource intelligent bus and task parallelism according to Embodiment 1 of the present invention; Figure 2 This is a flowchart illustrating the construction process of the Resource Intelligent Bus (RIB) architecture according to Embodiment 1 of the present invention. Figure 3 This is a schematic diagram of the multi-granularity task decomposition and DAG modeling process in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the resource-task dynamic matching process in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the dynamic scheduling execution process in Embodiment 1 of the present invention. Detailed Implementation

[0025] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0026] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0027] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0028] Example 1 This embodiment discloses a hydrogen energy data scheduling optimization method that combines resource intelligent bus and task parallelism.

[0029] To more clearly illustrate this embodiment, the hydrogen energy data scheduling optimization process of resource intelligent bus and task parallelism can be specifically described as follows: A resource intelligent bus and task-parallel hydrogen energy data scheduling optimization method includes: S1. Construct a resource intelligent bus architecture, which includes a resource virtualization encapsulation layer, a resource status awareness layer, and a global prediction and scheduling layer. S2. At the resource status perception layer, the hydrogen energy business demand is decomposed into multi-granularity tasks, and a directed acyclic graph is used to establish a dependency relationship model between tasks. S3. In the global prediction and scheduling layer, a multi-objective optimization model is constructed based on resource status and inter-task dependencies, and resources and tasks are dynamically matched according to the optimal solution of the model. S4. Perform dynamic resource scheduling based on the matching results of resources and tasks.

[0030] like Figure 1 , Figure 2 As shown, in step S1, a resource intelligent bus architecture is constructed, which includes a resource virtualization encapsulation layer, a resource status awareness layer, and a global prediction and scheduling layer.

[0031] Building a Resource Intelligent Bus (RIB) architecture is the core foundation for achieving efficient resource management and scheduling in hydrogen energy systems. Its core objective is to unify and dynamically coordinate distributed, heterogeneous physical resources, thereby providing a global resource view and real-time status support for upper-layer task scheduling. The construction of this architecture first addresses the heterogeneity of physical resources. Hydrogen energy systems contain diverse hardware devices such as edge computing nodes, GPU accelerators, distributed storage units, and 5G / NB-IoT communication modules. These devices differ significantly in computing power, storage capacity, communication protocols, and energy consumption characteristics.

[0032] In this embodiment, the resource intelligent bus architecture is constructed as follows: (1) Standardize the definition of various resources by using a unified resource descriptor and construct a resource virtualization encapsulation layer.

[0033] The Resource Intelligent Bus (RIB) architecture uses virtualization technology to logically abstract various resources, encapsulating them into a unified, elastically schedulable logical resource pool. Each resource entity is standardized and defined using a Uniform Resource Descriptor (NTG), which includes resource type, unique identifier, computing power, geographical location, and dynamic energy consumption characteristics.

[0034] Specifically, this descriptor not only includes basic attributes such as resource type and unique identifier, but also deeply integrates multi-dimensional dynamic feature parameters. The computing power dimension needs to clearly define quantitative indicators such as the number of CPU cores, memory capacity, and peak floating-point operation speed. Geographical location information is used to optimize cross-node data transmission paths and reduce communication latency caused by physical distance. Energy consumption characteristic modeling needs to distinguish between basic power consumption and dynamic frequency-related power consumption. For devices such as GPU accelerators, their power consumption is positively correlated with the cube of the operating frequency, while the energy efficiency of communication modules is linearly correlated with the square of the transmission distance and the data packet size.

[0035] The formula for the GPU dynamic power consumption model is as follows: (1) in, This refers to the GPU's base power consumption, measured in watts (W). The current operating frequency, in MHz. This is a process-related proportional coefficient.

[0036] This sophisticated modeling enables the system to accurately predict energy consumption during resource use, laying a data foundation for the formulation of green scheduling strategies.

[0037] By mapping real-time operating parameters to power consumption values ​​using formula (1), resource type, unique identifier, computing power, and geographical location can be directly collected or queried.

[0038] (2) Use lightweight probes to collect CPU utilization, available memory, real-time bandwidth and link jitter parameters, and set data transmission strategies to build a resource status awareness layer.

[0039] After completing the resource virtualization encapsulation, a Resource Intelligent Bus (RIB) architecture is constructed to establish a real-time dynamic resource status awareness mechanism. Each resource node deploys a lightweight probe to continuously collect four types of key status parameters with millisecond-level precision: CPU utilization, available memory, real-time bandwidth, and link jitter parameters.

[0040] CPU utilization is dynamically calculated by monitoring task queue length and processor occupancy to determine the proportion of current load to the maximum theoretical load, thus reflecting the real-time stress of computing resources. Available memory statistics require periodic scanning of the physical memory page table to accurately identify unoccupied memory blocks, providing a capacity assessment basis for distributing memory-intensive tasks. Real-time bandwidth measurement is based on a sliding window algorithm, statistically analyzing the average transmission rate of network interfaces to dynamically characterize the instantaneous communication capability of the link. Link jitter is assessed by calculating the standard deviation of historical latency data; when increased network fluctuations are detected, the system automatically adjusts the sampling window to quickly capture sudden changes.

[0041] The collection of this real-time data is not a simple accumulation, but rather a deeply optimized design with a defined data transmission strategy: the probes adopt an event-driven model, triggering data transmission only when the state change exceeds a preset threshold, significantly reducing system overhead. Simultaneously, a data compression algorithm is introduced to differentially encode highly repetitive monitoring information, further reducing network bandwidth consumption.

[0042] The formula for calculating the standard deviation of link jitter is as follows: (2) in, The delay for the t-th sampling is expressed in milliseconds. Let be the average delay of N samples.

[0043] (3) Based on the spatiotemporal joint prediction model, predict resource load changes and construct a global prediction and scheduling layer.

[0044] The integration of global resource status is the central component of the RIB architecture, its core being the construction of a dynamic resource matrix and the achievement of forward-looking predictions. The dynamic resource matrix is ​​organized in a two-dimensional table structure, with each row corresponding to a resource node and each column recording instantaneous state values ​​such as CPU utilization, available memory, real-time bandwidth, and link jitter. This matrix not only serves as a snapshot of the current system state but also extends its value to the time dimension through time series prediction algorithms. For the LightGBM prediction model expression: (3) in, This is the ensemble prediction value after the m-th iteration. This is the output of the m-th decision tree. This is the learning rate.

[0045] When using the LightGBM model to model historical monitoring data, the CPU load trend and bandwidth fluctuation pattern of each node can be predicted in the future, providing a forward-looking decision-making basis for dynamic scheduling. In this process, the input data of the prediction model needs to undergo rigorous screening and preprocessing. The historical state sequence of the most recent sampling period is selected as the feature vector, and a sliding window mechanism is used to ensure temporal continuity. Abnormal data points are weighted and smoothed to avoid distortion of the prediction results.

[0046] In addition, for the unique mobility scenarios of hydrogen energy systems (such as on-board hydrogen storage tank monitoring units), the RIB architecture also introduces geographic location correlation analysis, which combines electronic map data to predict changes in network coverage of mobile nodes, triggering resource migration or communication path switching in advance to ensure the continuity of critical tasks.

[0047] Through the synergistic effect of the three-layer architecture described above, RIB achieves full lifecycle management of heterogeneous resources. From the virtualization and abstraction of physical devices to millisecond-level state awareness, and then to global situational inference based on predictive models, it constructs the core infrastructure supporting intelligent scheduling. The innovation of this architecture is reflected in three aspects: First, the Unified Resource Descriptor (NTG) breaks through the limitation of traditional resource models that only focus on static attributes, incorporating dynamic energy consumption, geographical location, and other dimensions into a unified framework; Second, the adaptive probe mechanism achieves the optimal balance between monitoring accuracy and system overhead by dynamically adjusting the sampling frequency and data transmission strategy; Third, the spatiotemporal joint prediction model combines the temporal evolution law of resource status with the spatial distribution characteristics, significantly improving the foresight and robustness of scheduling decisions.

[0048] like Figure 1 , Figure 3 As shown, in step S2, in the resource status perception layer, the hydrogen energy business demand is decomposed into multi-granularity tasks, and a directed acyclic graph is used to establish a dependency relationship model between tasks.

[0049] Multi-granularity task decomposition and DAG modeling are core components of intelligent scheduling in hydrogen energy systems. Essentially, they break down complex business requirements into parallelizable and schedulable atomic task units, and reveal the inherent relationships and execution constraints between tasks through formal modeling. This process closely integrates with the business characteristics of the hydrogen energy industry, such as the differentiated computing, communication, and data requirements of scenarios like fluid simulation for hydrogen electrolysis, safety monitoring of hydrogen refueling stations, and status analysis of hydrogen storage tanks, resulting in domain-adaptive task partitioning and scheduling strategies.

[0050] S2-1. Decompose hydrogen energy business requirements into multi-granularity tasks.

[0051] In this embodiment, the specific process is as follows: (1) Perform task semantic analysis on hydrogen energy business requirements and divide hydrogen energy business tasks into computationally intensive tasks, communication-sensitive tasks and data-dependent tasks.

[0052] Task semantic parsing accurately classifies tasks based on the physical characteristics and computational features of the business scenario. For the full lifecycle management needs of hydrogen energy systems, task decomposition focuses on distinguishing three core business types: computationally intensive tasks, communication-sensitive tasks, and data-dependent tasks.

[0053] Among them, computationally intensive tasks are represented by electrolytic cell fluid dynamics simulation. Their computational complexity grows approximately linearly to superlinearly with the grid size and time steps. A single task may involve solving unsteady flow fields with hundreds of millions of grid cells, and should be preferentially allocated to nodes equipped with GPU accelerators or high-performance computing clusters.

[0054] Communication-sensitive tasks focus on scenarios with extremely high real-time requirements, such as safety monitoring of hydrogen refueling stations. In the millisecond-level processing of pressure sensor data and abnormal alarms, end-to-end communication latency must be strictly controlled within the specified value. Such tasks are extremely sensitive to network jitter and require the selection of low-latency, high-stability transmission paths such as direct fiber optic connections or 5G private networks.

[0055] Data-dependent tasks are common in scenarios such as hydrogen production efficiency optimization analysis. Their execution is strictly dependent on the data output of the preceding processes. For example, energy efficiency modeling and parameter optimization can only be started after the electrolyzer temperature monitoring data has been collected, cleaned and feature extracted.

[0056] (2) Construct a dynamic weight evaluation function to dynamically evaluate the task.

[0057] In this embodiment, the dynamic weight evaluation function formula is: (4) in, , , All are weighting coefficients, and + + =1; This represents the system's maximum computing power. The computational load required for a fluid simulation task is proportional to the mesh resolution and the number of time steps. The estimated / real-time communication time for this task, including "data volume" when crossing nodes. "Bandwidth" plus jitter latency; The communication latency threshold that the task can tolerate; The amount of data or the depth of dependency for a task; data-dependent tasks need to wait for the output of preceding data. Input the total amount of data for the task.

[0058] S2-2. Use a directed acyclic graph to establish a dependency model between tasks.

[0059] Task dependency modeling relies on Directed Acyclic Graphs (DAGs), using graph theory tools to transform the spatiotemporal constraints between tasks into a computable topological structure. The specific process is as follows: (1) Construct a directed acyclic graph, where each node is an atomic task. Label the multidimensional attributes of the atomic task nodes and model the edge weights by calculating the communication time between nodes.

[0060] 1) Each node represents an atomic task and is labeled with multidimensional attributes, including computational and data volume dimensions.

[0061] Specifically, the computational complexity dimension needs to be quantified based on hardware characteristics. The computational complexity of fluid simulation tasks is modeled as a cubic function of grid resolution (positively correlated with memory usage) multiplied by the number of time steps. The data volume dimension determines the size and format of the input and output datasets and is labeled with the encoding format to optimize transmission efficiency.

[0062] 2) Edge weight modeling is performed by calculating the communication time between nodes.

[0063] Directed edges between nodes represent the order of task execution and the direction of data flow. Their weights need to comprehensively consider communication overhead and network environment. When two tasks are assigned to nodes in different geographical locations, the communication time includes not only the theoretical transmission time of data volume divided by link bandwidth, but also random delays caused by network jitter. The formula for communication time between cross-regional nodes including random disturbances is: (5) in, The standard deviation of network jitter; For nodes To the node The estimated communication delay is composed of the theoretical transmission time and random jitter. For the node The amount of data sent to node j For nodes and The available effective bandwidth of the link between them. The random delay term introduced for network jitter.

[0064] (2) Based on the directed acyclic graph, identify and optimize the critical path.

[0065] Critical path identification is an advanced stage of directed acyclic graph (DAG) modeling, using global path analysis to pinpoint the core task chain that constrains overall execution efficiency. The specific process is as follows: 1) Traverse all execution paths in the directed acyclic graph and calculate the total execution time for each path.

[0066] Specifically, traverse all possible execution paths in the DAG, and calculate the total execution time of each path (the sum of the computation time of each node and the edge weight). The formula for the total execution time of a path is: (6) in, For execution path Total execution time This represents a feasible execution path from source to sink in a DAG; path Upper Task nodes For path Upper A communication edge indicates the direction of data flow between tasks; To calculate time, This refers to the communication time.

[0067] 2) Select the path with the longest execution time as the critical path.

[0068] The path that takes the longest time is the critical path.

[0069] 3) Optimize the critical path.

[0070] In a typical electrolytic hydrogen production optimization task chain, its Directed Acyclic Graph (DAG) includes four types of tasks: data acquisition, preprocessing, simulation calculation, and result analysis. Through traversal, it was found that the path "high-precision mesh generation → multiphysics coupled simulation → energy efficiency parameter optimization" has an excessively long total time consumption, far exceeding the time of other paths. Therefore, this path is identified as the critical path. A triple optimization strategy is implemented for this type of path: Prioritize deploying critical path tasks to CPU / HPC acceleration nodes and reserve dedicated bandwidth channels for data transmission.

[0071] A preemptive scheduling mechanism is used to prioritize resource acquisition for critical path tasks.

[0072] Establish checkpoints and task replication mechanisms for critical path tasks to check for path interruptions.

[0073] Specifically, at the resource allocation level, tasks on the path are prioritized for deployment to high-performance computing nodes, and dedicated bandwidth channels are reserved for data transmission; at the scheduling strategy level, a preemptive scheduling mechanism is adopted to ensure that critical tasks obtain resources first; at the fault tolerance mechanism level, checkpoint and task replication mechanisms are established to prevent path interruptions caused by node failures.

[0074] 6) By monitoring the task execution status, the scheduling strategy is dynamically updated when the path time relationship is reconstructed.

[0075] The dynamic nature of the critical path requires the system to continuously monitor the task execution status. When node performance fluctuations or network environment changes lead to a reconstruction of path time relationships, the critical path determination should be updated in real time and the scheduling strategy adjusted. This dynamic adjustment capability is the core guarantee for hydrogen energy systems to cope with complex industrial environments.

[0076] like Figure 1 , Figure 4 As shown, in step S3, in the global prediction and scheduling layer, a multi-objective optimization model is constructed based on resource status and inter-task dependencies, and resources and tasks are dynamically matched according to the optimal solution of the model.

[0077] Resource-task dynamic matching is the decision-making center for intelligent scheduling of hydrogen energy systems. Its core lies in constructing a multi-objective optimization model to find the Pareto optimal solution in the trade-off between time and energy consumption, while simultaneously satisfying multi-dimensional constraints. This process fully considers the different scenario-based needs of hydrogen energy operations, balancing the stringent real-time requirements of safety monitoring scenarios with the emphasis on green energy conservation in historical data analysis scenarios, achieving flexible adaptation of scheduling strategies through a dynamic weighting mechanism. The specific process is as follows: (1) Multi-objective optimization modeling takes minimizing global execution time and minimizing total system energy consumption as dual objectives and constructs a multi-objective optimization function.

[0078] The objective optimization model has two dual objectives: minimizing global execution time and minimizing total system energy consumption. There is a natural trade-off between these two objectives. The time optimization strategy prioritizes resource allocation for critical path tasks, forcibly allocating computationally intensive tasks such as fluid dynamics simulations to GPU clusters or high-performance computing nodes. This leverages their parallel computing capabilities to shorten the core path execution time. The multi-objective optimization function is as follows: (7) in, For the global execution time target, For the total energy consumption target of the system, This represents the feasible solution space.

[0079] (2) Based on resource status and inter-task dependencies, computationally intensive tasks are allocated to CPU / high-performance computing acceleration nodes, end-to-end optimized paths are constructed for data-dependent tasks, dedicated communication channels are reserved, and remote direct memory access technology is used for data transmission; a preemptive delay scheduling strategy is adopted for non-critical path tasks; and energy consumption is optimized from both computational and communication dimensions.

[0080] In this embodiment, an end-to-end optimized path is constructed for data-dependent tasks, and a dedicated communication channel is reserved between the electrolytic cell temperature monitoring task and the simulation task.

[0081] Remote Direct Memory Access (RDMA) technology is used to bypass the operating system kernel and achieve microsecond-level data transfer. The RDMA transfer time model function is as follows: (8) in, For protocol efficiency factor, For connection establishment time, The total time required to complete one data block transfer using RDMA. The amount of data to be transmitted This represents the peak bandwidth of the physical link.

[0082] Non-critical path tasks (such as historical data archiving) adopt a preemptive delayed scheduling strategy, which executes asynchronously only during the idle period of the node, and reduces resource contention through a time window segmentation mechanism.

[0083] Energy consumption optimization is approached from both computing and communication dimensions. On the computing side, batch data processing tasks are prioritized for allocation to energy-efficient edge computing devices rather than cloud servers with higher peak performance but higher power consumption. On the communication side, a transmission energy consumption model is constructed, selecting short-range 5G direct connections instead of multi-hop Wi-Fi relays for monitoring data from vehicle-mounted hydrogen storage tanks, thereby reducing energy consumption per unit of data. The dynamic weight adjustment mechanism is the core hub balancing these two aspects. The system automatically configures the objective function weights based on service type identification. This differentiated strategy enables the system to improve overall energy efficiency compared to traditional scheduling while ensuring the real-time performance of critical services.

[0084] (3) Construct the constraints of the multi-objective optimization function, including resource capacity constraints, task dependency constraints, deadline constraints and time series constraints; under the constraints, solve the multi-objective optimization function to obtain the optimal solution of the model; and dynamically match resources and tasks according to the optimal solution.

[0085] Constraint handling is crucial throughout the entire lifecycle of dynamic matching, requiring the construction of a multi-dimensional constraint space to ensure the feasibility of the scheduling scheme. This includes resource capacity constraints, task dependency constraints, deadline constraints, and timing constraints.

[0086] 1) Resource capacity constraints are implemented through hard limits by a real-time resource matrix. For example, the number of concurrent tasks of a single edge node must not exceed the number of its CPU cores, memory usage must be controlled below a safe threshold, and communication bandwidth allocation follows the token bucket algorithm to prevent overload.

[0087] Specifically, resource capacity constraints: For each physical / virtual node r and each managed resource type k, the total resources consumed by concurrently running tasks within the system at any given time must not exceed the node's upper limit, as specified in the formula: , (9) in, Let k be the unit requirement of resource k for task j. The indicator variable represents whether task j is scheduled to run on node r at time t. The capacity limit of node r in the resource k dimension, and t is the discrete scheduling period.

[0088] 2) Task dependency constraints strengthen the timing logic through DAG topology order and use an event-driven mechanism to trigger task execution: When task A completes the calculation and generates output data, the system immediately starts data compression and encrypted transmission. Only after the node where task B is located confirms that it has received the complete data packet and the verification is passed will it be added to the ready queue.

[0089] Specifically, the task dependency constraint is that if there is a directed edge (i,j) in the DAG, then task j can only start after its direct predecessor i has completed computation and data transmission, as shown in the formula: , (10) in, Let this be the earliest start time of task j. Let be the completion time of task i. The communication latency of data generated for task i being transmitted to the node where task j resides.

[0090] 3) For real-time tasks (such as pressure vessel anomaly alarms), a two-stage verification mechanism is adopted for deadline constraints. During the scheduling phase, it is determined whether the execution time meets the hard deadline, and overdue solutions are directly eliminated. During the operation phase, progress tracking is implemented. If the actual execution progress lags behind the expected planned prediction value, dynamic rescheduling is triggered (such as enabling redundant computing nodes to execute in parallel).

[0091] 4) DAG timing constraint function, the formula is: (11) in, The earliest start time for task j; For the pioneer mission Completion time; Let J be the set of direct predecessor tasks of task j in the DAG. Task j can only start after all tasks in the set have been completed. The communication latency is the time required for task i's output data to be transmitted from its node to the node executing task j.

[0092] like Figure 1 , Figure 5 As shown, in step S4, resources are dynamically scheduled based on the matching results of resources and tasks.

[0093] Dynamic scheduling execution is the final implementation stage of resource scheduling in hydrogen energy systems. Its core lies in translating the calculation results of the optimization model into specific resource allocation actions through a real-time decision-making mechanism, and continuously adapting to environmental changes during operation. This stage requires building a flexible and scalable scheduling framework that ensures both the immediate response capability of high-priority tasks and the stability of the system under complex operating conditions.

[0094] The specific process is as follows: (1) A priority preemption scheduling mechanism is adopted for resources and tasks, and a multi-dimensional scoring function is constructed to quantify the urgency of tasks, including the urgency of deadlines, the classification of security levels, and the intensity of resource demand.

[0095] Priority-based preemptive scheduling mechanisms use multi-dimensional scoring functions to quantitatively assess the urgency of tasks.

[0096] 1) Construct a multidimensional scoring function.

[0097] The formula for constructing the multidimensional scoring function is as follows: (12) in, The score is based on the overall urgency level; the higher the score, the more urgent the task is in terms of securing resources. This is a normalization function for the urgency of deadlines; For security level mapping functions; This is a normalization function for resource demand intensity.

[0098] 2) Multidimensional scoring functions quantify the urgency of tasks.

[0099] The scoring function integrates three core dimensions: deadline urgency, security level classification, and resource demand intensity.

[0100] The urgency of deadlines is assessed using a non-linear decay model, mapping remaining task time to an exponentially increasing priority coefficient. Tasks with closer deadlines are scored accordingly, while tasks with remaining hours see their scores surge to high values. Safety level classification utilizes a hydrogen energy equipment risk map, automatically labeling safety-critical tasks such as pressure vessel monitoring and hydrogen leak detection as Level 5, triggering a red alert. Resource demand intensity is quantified by calculating the sum of normalized CPU, memory, and bandwidth requirements for each task, thus assessing the intensity of resource competition.

[0101] (2) Construct a preemption protocol. When a high-priority task arrives, pause the low-priority task and retain the intermediate state, release the occupied computing and channel resources, and reallocate the resources to the urgent task. When a high-priority task arrives, the system scans resource occupancy in real time. If resources are insufficient, a preemption protocol is initiated: first, low-priority tasks are paused and their intermediate states are saved to persistent storage; then, the occupied computing units and communication bandwidth are released; and finally, resources are reallocated to urgent tasks. This mechanism reduces response latency for safety-critical tasks and improves timeliness compared to traditional polling scheduling.

[0102] (3) For communication-sensitive tasks, a dual-engine collaborative mechanism of bandwidth prediction and path selection is adopted, and bandwidth prediction is performed through a neural network model.

[0103] Optimization for communication-sensitive tasks focuses on adapting to the dynamic network environment of mobile hydrogen energy devices. For mobile terminals such as onboard hydrogen storage tank monitoring units, the system constructs a dual-engine collaborative mechanism for bandwidth prediction and path selection: bandwidth prediction uses an LSTM neural network model, inputting historical bandwidth sequences, geographical location information, base station load status, and other multi-dimensional features, and outputting a bandwidth change curve for the next 30 seconds. The prediction accuracy is high, and the LSTM prediction accuracy quantification function formula is as follows: (13) in, For LSTM to predict bandwidth values, This is the actual bandwidth measurement.

[0104] (4) Construct a multi-objective decision model to select a dynamic path, and use cross-protocol fast switching technology to complete session migration and data continuation within a specified time when the path is switched.

[0105] Dynamic path selection establishes a multi-objective decision-making model. Among multiple available links such as 5G, NB-IoT, and satellite communication, it comprehensively evaluates indicators such as transmission time, energy consumption cost, and connection stability. When the 5G signal strength is lower than the specified value, it will automatically switch to the NB-IoT link. Although the bandwidth is reduced, the real-time performance of key parameters can still be guaranteed through data compression and fragmented transmission technology, while reducing the risk of connection interruption.

[0106] The standard function for dynamic link selection, i.e., the formula for the multi-objective decision model, is as follows: (14) in, For transmission time, D / B, where D is the data volume and B is the available bandwidth of the link; Energy consumption per bit; For connection stability index; , , These are all weighting coefficients, indicating the emphasis of the scheduling strategy. For low latency, For high energy efficiency, For high stability, and .

[0107] Formula (14) combines the three indicators of "latency, energy efficiency, and link stability" into a single comparable scalar. .

[0108] In addition, the system introduces cross-protocol fast switching technology. When a path switching event is detected, the session migration and data resume transmission are completed within a specified number of milliseconds to avoid data packet loss caused by network switching.

[0109] (5) Based on the fault response and resilience indicators, the task is dynamically and adaptively adjusted; based on the results of the dynamic adaptive adjustment, a closed-loop control of monitoring-decision-action is carried out.

[0110] The dynamic scheduling execution layer achieves resilient scheduling of the hydrogen energy system in complex environments through closed-loop control of real-time monitoring, decision-making, and action. When fluctuations in fuel cell power supply cause performance degradation at edge nodes, the system detects the decrease in computing power within seconds and triggers a task migration mechanism to dynamically distribute some of the load to neighboring nodes. When a sudden grid failure causes cascading data center downtime, the system relies on pre-built disaster recovery plans to transfer core tasks to geographically dispersed backup nodes within a specified time, minimizing the impact of service interruptions. This dynamic adaptive capability enables the system to maintain a high on-time task completion rate even under extreme conditions with large fluctuations in resource availability, significantly improving the operational reliability of hydrogen energy infrastructure.

[0111] Example 2 The purpose of this embodiment is to provide a hydrogen energy data scheduling and optimization system that integrates a resource intelligent bus and parallel tasks, including: The resource intelligent bus architecture construction module is used to build the resource intelligent bus architecture, which includes a resource virtualization encapsulation layer, a resource status awareness layer, and a global prediction and scheduling layer. The task decomposition and modeling module is used to decompose hydrogen energy business requirements into multi-granularity tasks at the resource status awareness layer, and to establish a dependency model between tasks using a directed acyclic graph. The dynamic matching decision module is used to construct a multi-objective optimization model based on resource status and inter-task dependencies at the global prediction and scheduling layer, and to dynamically match resources and tasks according to the optimal solution of the model. The scheduling execution model is used to dynamically schedule resources based on the matching results of resources and tasks.

[0112] The method steps in Example 1 are implemented based on a hydrogen energy data scheduling and optimization system that provides intelligent resource bus and parallel task execution.

[0113] Example 3 The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method.

[0114] Example 4 The purpose of this embodiment is to provide a computer-readable storage medium.

[0115] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the above method.

[0116] The steps and methods involved in the apparatus of the above embodiments correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0117] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0118] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A hydrogen energy data scheduling optimization method based on resource intelligent bus and parallel task operation, characterized in that, include: Construct a resource intelligent bus architecture, which includes a resource virtualization encapsulation layer, a resource status awareness layer, and a global prediction and scheduling layer; At the resource status perception layer, the hydrogen energy business requirements are decomposed into multi-granularity tasks, and a directed acyclic graph is used to establish a dependency model between tasks. At the global prediction and scheduling layer, a multi-objective optimization model is constructed based on resource status and inter-task dependencies, and resources and tasks are dynamically matched according to the optimal solution of the model. Dynamic resource scheduling is performed based on the matching results between resources and tasks.

2. The hydrogen energy data scheduling optimization method with resource intelligent bus and task parallelism as described in claim 1, characterized in that, The specific process of building a resource intelligent bus architecture is as follows: A resource virtualization encapsulation layer is constructed by standardizing the definition of various resources through a unified resource descriptor. A resource status awareness layer is constructed by using lightweight probes to collect CPU utilization, available memory, real-time bandwidth and link jitter parameters, and setting data transmission strategies. A global prediction and scheduling layer is constructed based on a spatiotemporal joint prediction model to predict resource load changes.

3. The hydrogen energy data scheduling optimization method with resource intelligent bus and task parallelism as described in claim 2, characterized in that, A Uniform Resource Descriptor includes resource type, unique identifier, computing power, geographic location, and dynamic energy consumption characteristics.

4. The hydrogen energy data scheduling optimization method with resource intelligent bus and task parallelism as described in claim 1, characterized in that, The hydrogen energy business needs are decomposed into multi-granular tasks, and the specific process is as follows: The task semantics of hydrogen energy business requirements are analyzed, and the tasks of hydrogen energy business are divided into computationally intensive tasks, communication-sensitive tasks and data-dependent tasks. A dynamic weight evaluation function is constructed to dynamically evaluate the task.

5. The hydrogen energy data scheduling optimization method with resource intelligent bus and task parallelism as described in claim 1, characterized in that, A task dependency model is established using a directed acyclic graph. The specific process is as follows: Construct a directed acyclic graph where each node is an atomic task, label the multidimensional attributes of the atomic task nodes, and model the edge weights by calculating the communication time between nodes; Based on the directed acyclic graph, the critical path is identified and optimized.

6. The hydrogen energy data scheduling optimization method with resource intelligent bus and task parallelism as described in claim 5, characterized in that, Based on a directed acyclic graph, the critical path is identified and optimized. The specific process is as follows: Traverse all execution paths in the directed acyclic graph and calculate the total execution time for each path; Select the path with the longest execution time as the critical path; Prioritize deploying critical path tasks to CPU / high-performance computing acceleration nodes and reserve dedicated bandwidth channels for data transmission; For critical path tasks, a preemptive scheduling mechanism is used to prioritize resource acquisition; Establish checkpoints and task replication mechanisms for critical path tasks to check for path interruptions. By monitoring the task execution status, the scheduling strategy is dynamically updated when the path time relationship is reconstructed.

7. The hydrogen energy data scheduling optimization method with resource intelligent bus and task parallelism as described in claim 1, characterized in that, Based on resource status and inter-task dependencies, a multi-objective optimization model is constructed, and resources and tasks are dynamically matched according to the optimal solution of the model. The specific process is as follows: Multi-objective optimization modeling takes minimizing global execution time and minimizing total system energy consumption as dual objectives, and constructs a multi-objective optimization function. Based on resource status and inter-task dependencies, computationally intensive tasks are allocated to CPU / high-performance computing acceleration nodes, end-to-end optimized paths are built for data-dependent tasks, dedicated communication channels are reserved, and remote direct memory access technology is used for data transmission. For non-critical path tasks, a preemptive delayed scheduling strategy is adopted; Energy consumption optimization is performed from both computing and communication perspectives. Construct constraints for the multi-objective optimization function, including resource capacity constraints, task dependency constraints, deadline constraints, and time series constraints. Under the constraints, the multi-objective optimization function is solved to obtain the optimal solution of the model; Resources and tasks are dynamically matched based on the optimal solution.

8. The hydrogen energy data scheduling optimization method with resource intelligent bus and task parallelism as described in claim 1, characterized in that, Based on the matching results between resources and tasks, resources are dynamically scheduled. The specific process is as follows: A priority-based preemptive scheduling mechanism is adopted for resources and tasks, and a multi-dimensional scoring function is constructed to quantify the urgency of tasks, including deadline urgency, security level classification, and resource demand intensity. Establish a preemption protocol so that when a high-priority task arrives, low-priority tasks are paused and the intermediate state is preserved, the occupied computing and channel resources are released, and the resources are reallocated to urgent tasks. For communication-sensitive tasks, a dual-engine collaborative mechanism of bandwidth prediction and path selection is adopted, and bandwidth prediction is performed through a neural network model. A multi-objective decision-making model is constructed to select dynamic paths, and a cross-protocol fast switching technology is used to complete session migration and data resume transmission within a specified time when the path is switched. The task is dynamically and adaptively adjusted based on fault response and resilience indicators; Based on the results of dynamic adaptive adjustment, a closed-loop control of monitoring, decision-making, and action is implemented.

9. The hydrogen energy data scheduling optimization method with resource intelligent bus and task parallelism as described in claim 1, characterized in that, The formula for the multi-objective decision model is: ； in, For transmission time, D / B, where D is the data volume and B is the available bandwidth of the link; Energy consumption per bit; For connection stability index; , , These are all weighting coefficients, indicating the emphasis of the scheduling strategy. For low latency, For high energy efficiency, For high stability, and .

10. A hydrogen energy data scheduling and optimization system with intelligent resource bus and parallel task operation, characterized in that: include: The resource intelligent bus architecture construction module is used to build the resource intelligent bus architecture, which includes a resource virtualization encapsulation layer, a resource status awareness layer, and a global prediction and scheduling layer. The task decomposition and modeling module is used to decompose hydrogen energy business requirements into multi-granularity tasks at the resource status awareness layer, and to establish a dependency model between tasks using a directed acyclic graph. The dynamic matching decision module is used to construct a multi-objective optimization model based on resource status and inter-task dependencies at the global prediction and scheduling layer, and to dynamically match resources and tasks according to the optimal solution of the model. The scheduling execution model is used to dynamically schedule resources based on the matching results of resources and tasks.

Images