An ultra-intelligent fusion-based power industry cross-domain computing power dynamic scheduling system

Abstract

The application relates to the technical field of cable laying management, and discloses a power industry cross-domain computing power dynamic scheduling system based on super intelligence fusion. The system acquires computing power demand data of each business domain of the power industry in real time, identifies computing resource demand characteristics in different business scenarios, and generates a cross-domain computing power demand feature map. Meanwhile, the system continuously tracks real-time load states and available resource amounts of each computing node, and constructs a distributed computing power resource state matrix. Then, the system analyzes the matching relationship between computing resource demand and available resources, generates a multi-objective optimization computing power scheduling strategy scheme, allocates power business computing tasks to optimal computing nodes according to the scheme, triggers a cross-domain execution process, and finally monitors task execution states and resource usage in real time, generates an evaluation report and feeds back the evaluation report to a strategy generation module, forms a closed-loop optimization, and realizes efficient cooperation and dynamic scheduling of power industry cross-domain computing power resources.

Description

Technical Field

[0001] This invention relates to the field of cable laying management technology, specifically to a cross-domain computing power dynamic scheduling system for the power industry based on super-intelligent fusion. Background Technology

[0002] Cable laying is a critical construction phase in infrastructure development, including power systems and communication networks. Traditional cable laying management relies heavily on the experience and manual operation of construction personnel, lacking systematic and quantifiable monitoring and control methods. During laying, the lateral pressure on the cable is a significant factor affecting its installation quality and service life. Insufficient pressure may cause the cable to sag and droop, affecting aesthetics and safety; excessive pressure may damage the cable insulation layer, deform the metal shielding layer, or even damage the internal conductors, leading to potential operational failures. However, current technology struggles to monitor lateral pressure in real-time and accurately. The use of mechanical pressure gauges or manual judgment is common, but suffers from low accuracy, delayed feedback, and inability to continuously record data, making it difficult to effectively warn of abnormal pressure.

[0003] Measuring the cable laying length is equally crucial. Accurate length data is the foundation for material management, schedule control, and quality acceptance. Traditional methods often rely on tape measures or rough estimates based on the number of cable reel rotations, resulting in significant errors and an inability to perform real-time correlation analysis with other parameters during the laying process (such as pressure and angle). The complexity of the laying path, including turns, inclines, and conduits, means that changes in the cable's angle as it passes through pulley systems directly affect its stress state. Currently, pulley system angle adjustments are largely based on experience, lacking an optimization mechanism based on real-time pressure feedback, making it difficult to achieve dynamic optimization of the laying process.

[0004] Monitoring construction progress often relies on periodic manual records and reports, resulting in fragmented information and difficulty in timely detection of progress deviations. Laying quality assessments are typically conducted after construction is completed, lacking supporting process data and thus constituting a retrospective judgment that cannot correct problems during construction. Throughout the management process, key parameters such as pressure, length, angle, and progress are isolated, lacking a unified data analysis model and closed-loop control mechanism, leading to low management efficiency and weak controllability of construction quality. Therefore, there is an urgent need for a cable laying management method capable of real-time monitoring of key parameters, intelligent status analysis, prediction of potential risks, and support for dynamic adjustments. Summary of the Invention

[0005] The purpose of this invention is to provide a cross-domain computing power dynamic scheduling system for the power industry based on super-intelligent fusion, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a cross-domain computing power dynamic scheduling system for the power industry based on super-intelligent fusion, the system comprising:

[0007] The computing power demand perception module collects computing power demand data from various business domains in the power industry in real time, identifies the characteristics of computing resource demand under different business scenarios, and generates a cross-domain computing power demand characteristic map.

[0008] The resource status monitoring module continuously tracks the real-time load status and available resources of each computing node, obtains the resource utilization, response latency and energy consumption indicators of the computing nodes, and constructs a distributed computing power resource status matrix.

[0009] The scheduling strategy generation module analyzes the matching relationship between computing resource requirements and available resources in different business scenarios based on the cross-domain computing power demand feature map and the distributed computing power resource status matrix, and generates a multi-objective optimized computing power scheduling strategy scheme.

[0010] The task assignment and execution module is based on a multi-objective optimization computing power scheduling strategy. It allocates power business computing tasks to the optimal computing nodes according to priority and resource demand characteristics, triggering the cross-domain execution process of computing tasks.

[0011] The dynamic feedback adjustment module monitors the execution status and resource usage of computing tasks in real time, collects task execution progress and resource consumption data, generates a computing power scheduling effect evaluation report, and feeds it back to the scheduling strategy generation module.

[0012] Preferably, the computing power demand sensing module includes:

[0013] The business feature extraction submodule parses the computing task requests from various business domains in the power industry, extracts the characteristics of task type, data scale, processing timeliness and computing complexity, and forms a description of business computing power demand characteristics.

[0014] The demand correlation analysis submodule identifies data dependencies and execution order constraints between computing tasks in different business domains, and establishes a cross-business domain computing power demand correlation network.

[0015] The demand graph construction submodule integrates the description of business computing power demand characteristics with the cross-business domain computing power demand association network to generate a cross-domain computing power demand characteristic graph with a hierarchical structure.

[0016] Preferably, the resource status monitoring module includes:

[0017] The node performance acquisition submodule periodically polls the CPU utilization, memory usage, storage I / O performance, and network bandwidth usage of each computing node, and records the real-time performance metrics of the computing nodes.

[0018] The state matrix update submodule compares and analyzes the real-time performance indicators of computing nodes with historical operating data, identifies abnormal fluctuations in resource usage, and updates the node state information in the distributed computing power resource state matrix.

[0019] The resource prediction submodule predicts the available resources and performance changes of each node in the future period based on the historical load change patterns and current resource usage trends of computing nodes.

[0020] Preferably, the scheduling strategy generation module includes:

[0021] The demand matching analysis submodule performs multi-dimensional matching between the business computing needs in the cross-domain computing power demand feature map and the node capabilities in the distributed computing power resource status matrix, and calculates the adaptability score of each node for a specific task.

[0022] The strategy optimization submodule comprehensively considers multiple objectives such as task priority, resource utilization balance, energy efficiency and response latency to generate a set of candidate computing power scheduling strategies that meet the needs of power business.

[0023] The strategy evaluation submodule simulates the execution effect of different candidate computing power scheduling strategies in a distributed environment and selects the scheme with the highest comprehensive score as the final multi-objective optimized computing power scheduling strategy.

[0024] Preferably, the task assignment and execution module includes:

[0025] The task priority sorting submodule prioritizes the queue of tasks waiting to be scheduled based on the urgency and importance of the computing tasks in the power business.

[0026] The resource allocation submodule, following a multi-objective optimized computing power scheduling strategy, prioritizes the allocation of high-priority tasks to computing nodes that meet resource requirements, ensuring the timeliness of critical task execution.

[0027] The task triggering submodule sends task execution instructions to the target computing node, initiating the distributed processing flow of the cross-domain computing task.

[0028] Preferably, the dynamic feedback adjustment module includes:

[0029] The execution monitoring submodule tracks the execution progress and resource consumption of tasks on each computing node in real time, and collects data on task processing rate and resource utilization efficiency.

[0030] The performance evaluation submodule compares the actual execution of tasks with the expected scheduling goals, analyzes the execution deviation of the computing power scheduling strategy, and generates a computing power scheduling performance evaluation report.

[0031] The strategy adjustment submodule triggers the optimization process of the scheduling strategy generation module based on the deviation analysis results in the computing power scheduling effect evaluation report, thereby realizing the dynamic adjustment of the computing power scheduling strategy.

[0032] Preferably, the business feature extraction submodule includes:

[0033] The task parsing unit breaks down the workflow of power business calculation tasks and identifies the calculation characteristics and data processing requirements of each stage of the task.

[0034] The feature quantization unit transforms the computational complexity, data throughput, and timeliness requirements of a task into quantifiable computing power demand indicators.

[0035] The feature integration unit normalizes computing power demand indicators from different dimensions to form a standardized description of business computing power demand characteristics.

[0036] Preferably, the node performance acquisition submodule includes:

[0037] The metrics acquisition unit obtains real-time performance data of computing nodes, including processor load, memory usage, and network latency, through remote API calls.

[0038] The data cleaning unit filters out outliers and noise from the collected data to ensure the accuracy and reliability of performance indicators.

[0039] The indicator storage unit stores the cleaned performance data in a distributed database according to time series, supporting historical data backtracking and analysis.

[0040] Preferably, the strategy optimization submodule includes:

[0041] The target weight allocation unit dynamically adjusts the weight ratios of optimization targets such as resource utilization, energy efficiency, and response latency based on the characteristics of the power business scenario.

[0042] The strategy generation unit, based on a multi-objective optimization algorithm, explores the space of feasible computing power scheduling strategies that satisfy the constraints of each objective.

[0043] The strategy screening unit removes obviously inferior solutions from the feasible strategy space and retains a set of candidate computing power scheduling strategies with optimization potential.

[0044] Preferably, the resource allocation submodule includes:

[0045] The node selection unit determines the set of target nodes most suitable for executing the current task based on the characteristics of task resource requirements and the computing node suitability score.

[0046] The load balancing unit takes into account the current load of the target node and avoids excessive concentration of resources on a few high-performance nodes;

[0047] The fault-tolerant allocation unit configures backup execution nodes for critical tasks, automatically switching to the backup node to continue execution when the primary node fails.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] This invention enables real-time and accurate monitoring and early warning of pressure anomalies on the cable laying side. By integrating sensing technology into an adjustable pulley system, dynamic pressure data is continuously collected and compared with preset safety thresholds, promptly identifying any pressure anomalies. This changes the previous reliance on manual judgment, making pressure control more precise and timely, effectively reducing the risk of cable damage due to improper pressure, and providing crucial data support for ensuring laying quality.

[0050] By integrating and correlating multi-dimensional parameters such as pressure, angle, and length, a unified laying status matrix was constructed. This method breaks away from the isolation of parameters in traditional management, clearly expressing the intrinsic relationships between key parameters through a matrix format. This data fusion lays a solid foundation for in-depth analysis of laying status and diagnosis of root causes of problems, enabling management decisions to shift from being based on single information to being based on the correlation analysis of multi-source information.

[0051] Based on state matrix analysis, angle optimization suggestions are generated and the pulley block angle is dynamically adjusted, forming a closed-loop control mechanism of "monitoring-analysis-optimization". The system can automatically identify the angle operation range that causes abnormal pressure and provide specific adjustment suggestions. By performing angle optimization, the stress state of the cable can be improved from the source, realizing dynamic adaptive adjustment during the laying process, and improving the level of intelligence and quality control capabilities of construction.

[0052] By performing trend analysis and predicting fluctuations in pressure parameters, the system achieves a leap from static monitoring to dynamic forecasting. It not only focuses on current pressure values ​​but also analyzes their trends over time, predicting the likelihood of future fluctuations. This proactive analysis helps identify potential risks in advance, providing a window of opportunity for preventative measures and enhancing the initiative in risk management.

[0053] By constructing quality assessment indicators and comparing them with historical data, collaborative monitoring of construction progress and quality was achieved. This method combines process quality data (stress conditions) with progress data (laying length) to form a comprehensive quality assessment indicator, which is then compared with historical benchmarks. This allows for timely detection of signs of schedule deviation or quality decline and the issuance of early warnings. This integrated management helps ensure that the project progresses on time and to the required standard.

[0054] The resulting cable laying management report integrates key information from the entire process, providing comprehensive and objective decision-making support for project management. The report covers multi-level information, from real-time anomalies and process trends to final assessments, enabling managers to quickly grasp the current construction status, optimize resource allocation, and improve overall project management efficiency. This method promotes the digital and intelligent transformation of cable laying management. Attached Figure Description

[0055] Figure 1 This is a timing diagram of a cross-domain computing power dynamic scheduling system for the power industry based on super-intelligent fusion, as described in this invention.

[0056] Figure 2 A flowchart of the computing power demand perception module;

[0057] Figure 3 This is a flowchart of the resource status monitoring module.

[0058] Figure 4 This is a diagram illustrating the scheduling strategy analysis. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Please see Figure 1This invention provides a cross-domain computing power dynamic scheduling system for the power industry based on super-intelligent fusion. The system includes a computing power demand perception module that collects computing power demand data from various business domains in the power industry in real time, identifies the characteristics of computing resource demand under different business scenarios, and generates a cross-domain computing power demand characteristic map; a resource status monitoring module that continuously tracks the real-time load status and available resources of each computing node, obtains the resource utilization rate, response latency, and energy consumption indicators of the computing nodes, and constructs a distributed computing power resource status matrix; a scheduling strategy generation module that analyzes the matching relationship between computing resource demand and available resources in different business scenarios based on the cross-domain computing power demand characteristic map and the distributed computing power resource status matrix, and generates a multi-objective optimized computing power scheduling strategy; a task assignment and execution module that, based on the multi-objective optimized computing power scheduling strategy, allocates power business computing tasks to the optimal computing nodes according to priority and resource demand characteristics, triggering the cross-domain execution process of computing tasks; and a dynamic feedback adjustment module that monitors the execution status and resource usage of computing tasks in real time, collects task execution progress and resource consumption data, generates a computing power scheduling effect evaluation report, and feeds it back to the scheduling strategy generation module, forming a closed-loop control mechanism. The system adopts a distributed architecture, with each module exchanging data through message queues and API interfaces to ensure low latency and high reliability. The super-intelligent fusion technology is reflected in the intelligent identification of computing power requirements and the adaptive optimization of resource scheduling by artificial intelligence algorithms, thereby realizing efficient dynamic scheduling of cross-domain computing power in the power industry.

[0061] Example 1: See Figure 2 The computing power demand perception module, through the collaborative operation of multiple sub-modules, achieves in-depth analysis and structured expression of computing power demands across various business domains in the power industry. The business feature extraction sub-module, acting as the entry point, receives and processes raw computing task requests from business systems such as power dispatching, fault diagnosis, and load forecasting. Within this sub-module, the task parsing unit uses workflow decomposition to deconstruct tasks. For example, for a real-time power grid security analysis task, it identifies multiple stages such as topology analysis, power flow calculation, and stability assessment, and marks the computational characteristics and data processing requirements of each stage. The feature quantification unit then transforms the qualitative descriptions into quantifiable numerical indicators. The quantification of computational complexity may refer to historical execution time or theoretical algorithm complexity; data throughput is statistically analyzed based on the size and frequency of data packets processed by the task; and timeliness requirements are converted into maximum allowable latency according to the business service level agreement. These quantification processes rely on a predefined indicator dictionary and a normalization rule base. Finally, the feature integration unit standardizes the indicators from different dimensions, eliminates differences in units, and combines them into a structured record of business computing power demand features. This record adopts a unified JSON or XML format and includes task ID, resource requirement vector, and constraints.

[0062] The operation of the demand correlation analysis submodule begins with continuous monitoring of the standardized task description stream produced by the business feature extraction submodule. This submodule maintains a real-time updated task metadata cache. Whenever a new computing task is parsed by the feature extraction submodule and its computing power requirement feature description is generated, the task's metadata is immediately pushed to this cache. An asynchronous processing pipeline then retrieves task metadata records sequentially from the cache. At the core of this pipeline is a graph computing engine-based correlation identifier. This identifier scans the input data source identifiers declared in the new task record and matches these identifiers with the data target identifiers of existing task outputs in the cache. For example, when the identifier processes a new task called "T002 - Fault Location", it finds that the input data source for this task is identified as "DS001". The cache records show that the data source "DS001" was produced by an earlier task called "T001 - Fault Recording Analysis". The identification logic will automatically create a directed edge in the internal graph model from task node "T001" to task node "T002" and mark the attribute of this edge as "data dependency".

[0063] Beyond these explicit data flow dependencies, this submodule is also configured with an extensible business rule base to capture more implicit execution order constraints. The rule base predefines constraint templates from power industry regulations, such as "a task of type 'protection action analysis' can only start after a task of type 'switch position change confirmation' is completed." The rule engine continuously matches the type attributes of newly arriving tasks with patterns in the rule base. When a match is successful, even if there is no direct data transfer relationship between the two tasks, the system will establish corresponding temporal dependency edges in the graph model. This rule-driven association discovery compensates for the shortcomings of simple data flow analysis and can characterize the inherent logic at the business process level. All identified dependencies are updated in real time to a computing power demand association network stored in an in-memory graph database. This network is a dynamically growing directed graph with rich node attributes, including not only basic task information but also resource demand vectors obtained from the business feature extraction phase. Its edges carry attributes such as dependency type (data dependency, temporal dependency), strength, and constraints. The graph database's indexing mechanism supports fast queries on complex patterns, such as efficiently answering questions like "find the set of all task nodes that directly or indirectly depend on a specific data source task."

[0064] A key function of this interconnected network is to reveal potential collaborative opportunities across business domains. The network construction algorithm periodically executes a community discovery algorithm to identify clusters of nodes with tight internal connections and sparse external connections. These clusters often correspond to specific business scenarios. For example, the algorithm might identify a cluster that simultaneously contains a "short-term load forecasting" task from the scheduling domain and a "large user electricity consumption behavior analysis" task from the marketing domain. Although they belong to different systems, the interconnected network shows that they are highly dependent on the same set of raw electricity collection data. This discovery is recorded as a high-order property of the graph. When these tasks are simultaneously in a pending scheduling state, the demand graph construction submodule can use this information as an optimization suggestion. The subsequent scheduling strategy generation module may then generate a strategy to allocate these two tasks to the same computing node or the same rack, reducing network transmission overhead through data localization, thereby achieving cross-domain collaborative optimization of computing power. The entire interconnected analysis process is incremental. As new tasks arrive and old tasks are completed, the interconnected network is constantly evolving, ensuring the scheduling system's ability to perceive changes in business needs.

[0065] The demand graph construction submodule is responsible for the final information fusion and hierarchical organization. It uses the standardized feature descriptions generated by the business feature extraction submodule as entity nodes in the graph and the association network generated by the demand association analysis submodule as the relationship edges between entities, performing an integration operation. The integration process is not a simple superposition, but requires entity alignment and relationship disambiguation, such as merging tasks submitted by different business systems that point to the same physical computing resource. The graph construction adopts a bottom-up approach, with the bottom layer being specific computing task instances, the middle layer being task types, and the top layer being business domains, thus forming a hierarchical structure with semantic information. This graph is implemented using knowledge graph technology, typically based on RDF triples or attribute graph models, and utilizes graph query languages ​​to support complex queries, such as "find all fault handling related tasks that require GPU acceleration and must be completed within the next ten minutes". The graph is updated incrementally, adjusting in real time as new tasks arrive and old tasks are completed, ensuring that it always reflects the latest overall picture of computing power demand.

[0066] Throughout the data flow of this module, data acquisition is accomplished through lightweight agents deployed within the power business system. These agents listen for task submission events or parse system logs, encapsulating the raw request information into a standard message format before sending it to the message middleware. The machine learning model deployed in the feature extraction phase is periodically retrained using historical task data to maintain its recognition accuracy. The graph computing engine in the correlation analysis phase operates in micro-batch or continuous processing modes to balance real-time computation and resource overhead. The resulting cross-domain computing power demand feature map is provided for querying through a highly available graph service API, offering comprehensive and structured demand-side information input for downstream scheduling decisions. The entire module's design fully considers the high throughput and timeliness of power business data, employing a distributed stream processing architecture to ensure processing capacity.

[0067] Example 2: See Figure 3 The resource status monitoring module relies on a distributed data acquisition, processing, and persistence architecture. Its core task is to continuously acquire and maintain real-time resource views of each node in the computing cluster. The node performance acquisition submodule, acting as a data entry point, deploys lightweight monitoring agents on each physical server or virtual machine in the cluster. These agents proactively collect local node performance metrics at fixed time intervals. The metric acquisition unit obtains raw data through various means. For operating system-level metrics such as CPU utilization and memory usage, the agent directly reads relevant files under the / proc file system or calls the sysstat library function to obtain precise values. For storage I / O performance, the agent parses the output of the iostat command or interacts directly with the storage device's driver interface to obtain read / write throughput and IOPS data. Network bandwidth usage is obtained by monitoring network interface card statistics or using tools such as netstat and ethtool. All acquisition actions are encapsulated into a unified metric acquisition plugin, supporting dynamic loading and starting / stopping. The acquired raw data stream is immediately sent to a central message queue for buffering to handle instantaneous data spikes and decouple production from consumption.

[0068] The data cleaning unit, as the next processing stage, subscribes to the raw metric stream from the message queue. It is responsible for data quality control and incorporates multi-level filtering and correction logic. This unit performs basic validity checks, removes values ​​significantly exceeding reasonable ranges, and applies a sliding window algorithm to detect short-term abnormal fluctuations. For example, it uses the Z-Score algorithm to identify data points that significantly deviate from recent historical patterns. For consecutive missing values ​​caused by network jitter or temporary proxy disconnection, the cleaning unit uses time-series-based interpolation methods to fill in the gaps. Unrepairable dirty data or data missing over long periods is marked as invalid and logged. The cleaned data is appended with timestamps, node identifiers, and data quality tags, and converted into structured JSON records. The format is optimized to reduce serialization overhead and facilitate rapid subsequent parsing. The metric storage unit is responsible for persisting the processed data to a dedicated time-series database. The database schema is highly optimized for monitoring scenarios, using node IDs as the primary index tags, different performance metrics as measurement values, and timestamps as the primary keys. Data writing employs batch processing to reduce database connection pressure, while a reasonable data retention strategy is implemented. This time-series database not only supports efficient point and range queries to provide data for real-time dashboards, but also automatically calculates aggregate metrics such as rolling averages and maximum values ​​of resources through its built-in continuous query function.

[0069] The state matrix update submodule periodically extracts the latest and historical performance data from the time-series database. It maintains an in-memory distributed computing resource state matrix, where rows represent computing nodes in the cluster and columns represent different resource dimensions (CPU, memory, storage I / O, network I / O) and their derived metrics. The update process is not a simple replacement of old values, but includes a state analysis phase. This module compares the real-time performance metrics of a node with its own historical operating baseline, using statistical process control methods to determine whether the current load is within a normal range. When abnormal fluctuations in resource usage are detected, the update logic not only modifies the state value of that node in the matrix but also triggers an alarm event and records possible cause tags. The matrix itself is backed up using a scalable distributed key-value store to ensure rapid recovery of the cluster state after a module restart. Matrix update events are notified to other modules subscribed to the data via a publish-subscribe pattern.

[0070] The resource prediction submodule operates on a foundation of rich historical data. It utilizes long-term performance metrics accumulated in storage units to train a resource load prediction model for each node. The prediction process typically employs classic time series analysis methods, such as the Seasonal Autoregressive Integral Moving Average (SARIMA) model, to capture the periodic patterns of load, while incorporating current resource usage trends as short-term correction factors. For metrics with strong non-linear relationships, the module also supports integrating machine learning models such as gradient boosting decision trees or Long Short-Term Memory (LSTM) networks for prediction. These models are periodically retrained offline to maintain accuracy. The prediction submodule outputs the expected availability of key resources (CPU, memory) for each node over a future period, along with the probability distribution of their confidence intervals. These predictions are appended to the resource state matrix, providing crucial information for forward-looking scheduling decisions. This allows the system to proactively remove tasks from nodes about to become overloaded or wake up low-power nodes in advance to cope with anticipated load increases. The entire module, through a closed-loop feedback mechanism, enables the perception, recording, analysis, and prediction of resource state to form a continuously self-improving dynamic system.

[0071] Example 3: The scheduling strategy generation module relies on a complex decision logic chain. This module receives the cross-domain computing power demand feature map from the computing power demand perception module and the distributed computing power resource status matrix from the resource status monitoring module as core inputs. Its internal data processing flow begins with the activation of the demand matching analysis submodule. This submodule parses the resource demand vector of each computing task in the feature map, including dimensions such as the number of CPU cores, memory size, storage space, and network bandwidth. Simultaneously, it extracts the real-time available resources and performance indicators such as current load and response latency of each computing node from the resource status matrix. The matching process employs a multi-attribute decision method, assigning a quantified fit score to each task-node pair. This score is derived through a comprehensive formula, specifically expressed as:

[0072]

[0073] Where: characters Representative task With computing nodes The compatibility score between them, the higher the value, the higher the match; characters This indicates the total number of resource dimensions considered, typically including CPU, memory, storage, and network; character It's a resource-level index that iterates through all monitored resource types; characters Represents a node Resource type at the current moment Available quantities, such as the number of available CPU cores or the size of free memory; characters Indicates task For resource types The demand is extracted from the business computing power demand characteristic description; characters It is a resource dimension The weighting coefficients are used to calculate the fit score. The fit score is calculated in real time, and a re-evaluation is triggered for each newly arrived task or resource status update event. The score results are stored in a distributed cache for subsequent optimization.

[0074] The output of the demand matching analysis submodule is a large fitness score matrix, where rows represent tasks to be scheduled and columns represent available computing nodes. This matrix serves as the input to the strategy optimization submodule. After the strategy optimization submodule starts, its internal target weight allocation unit adjusts the weight ratios of optimization targets based on the dynamic characteristics of the power business scenario. For example, during peak electricity consumption periods, the weight of response delay is increased to ensure real-time performance, while during off-peak periods, energy efficiency may be prioritized to reduce operating costs. The weight adjustment logic is based on a predefined business rule base, which stores typical weight configurations for different scenarios (such as fault handling, load forecasting, and bill generation). It also supports automatic fine-tuning based on historical scheduling performance using machine learning models. After the weights are determined, the strategy generation unit begins its work. This unit encapsulates multi-objective optimization algorithms such as the Non-Dominated Sorting Genetic Algorithm (NSGA-II). This algorithm models the scheduling problem as a search problem, where the decision variables are the mapping relationship between tasks and nodes. The objective function includes several conflicting metrics such as maximizing the overall fitness score, minimizing the maximum cluster resource utilization, minimizing total energy consumption, and minimizing average response delay. During algorithm initialization, a set of possible task allocation schemes is randomly generated as the initial population. Each scheme represents a complete scheduling strategy. The solution space is explored through iterative selection, crossover and mutation operations. In each generation, the algorithm uses fast non-dominated sorting to divide the solutions into different Pareto fronts and calculates the crowding degree to maintain the diversity of solutions. Finally, it outputs a set of solutions that is approximately Pareto optimal, which is the set of candidate computing power scheduling strategies.

[0075] The strategy screening unit then post-processes the generated candidate strategy set. This unit applies a series of filtering rules to eliminate obviously unreasonable solutions, such as those that assign high-priority tasks to high-load nodes or cause the resource utilization of some nodes to exceed a safe threshold (e.g., 95%). The screening process is based on hard constraint checks, such as meeting the core resource requirements of the task, while also considering soft constraints such as geographical affinity. The remaining strategy set after screening typically contains dozens of feasible solutions with different trade-offs. These solutions are then passed to the strategy evaluation submodule for in-depth analysis. The strategy evaluation submodule operates in a discrete event simulation environment that simulates the actual operating conditions of a distributed computing cluster, including network transmission latency, task startup overhead, and resource contention effects. For each candidate strategy, the simulator virtually executes its defined task allocation scheme and records key performance indicators such as task completion time, overall cluster resource utilization variance, and total energy consumption estimates. Simulation results are compiled into an evaluation report, which calculates a comprehensive score for each strategy. This score is typically achieved by combining multiple simulation metrics into a single value using a weighted summation model, with weights consistent with those used in the optimization phase. Finally, the strategy with the highest comprehensive score is selected as the multi-objective optimization computing power scheduling strategy and output to the task assignment and execution module. The entire scheduling strategy generation module is designed with scalability and real-time performance in mind. Its core algorithm is implemented using a distributed computing framework to handle large-scale cluster scheduling problems, and the module's internal state is persisted to a distributed database to support fault recovery.

[0076] See Figure 4 The heatmap above clearly shows the distribution of adaptability between various power business tasks and different computing nodes through fine grayscale gradients. Some tasks exhibit extremely high matching degrees on specific nodes, reflecting the differentiated needs of power businesses for computing resources, while other tasks maintain a relatively balanced level of adaptability across multiple nodes, demonstrating the strong adaptability of batch processing tasks to resource environments. The multi-objective optimization strategy comparison bar chart below systematically compares the comprehensive performance of various candidate schemes in key performance dimensions. The optimal strategy achieves the best balance among multiple mutually constraining objectives such as overall adaptability, resource utilization, energy consumption control, and response latency. Other strategies show their respective advantages and disadvantages in different indicators. This comprehensive comparative analysis fully verifies the practical value of multi-objective optimization algorithms in solving complex scheduling problems in the power industry, providing important technical support and practical basis for building an efficient and reliable smart grid computing power scheduling system.

[0077] Example 4: Referring to Table 1, the task assignment and execution module demonstrates its workflow through a specific scenario. Imagine a provincial power grid company encountering a sudden local line fault. Its dispatch center needs to quickly execute a series of related calculation tasks to perform fault analysis, isolation, and power restoration. At this time, the optimized scheme from the upstream dispatch strategy generation module has been generated, and this module begins operation. The task priority sorting submodule receives a set of tasks to be scheduled, which contains multiple urgent tasks.

[0078] Table 1: Computation Task Queue in Fault Handling Scenarios

[0079]

[0080] The sorting submodule maintains a priority-based blocking queue. It sorts tasks based on the "business priority" and "maximum allowable delay" fields in the table. For tasks with the same priority, the order is adjusted according to the topological order formed by their data dependencies. The sorting algorithm identifies task T001, which has no prior dependencies, and places it at the front of the queue, followed by the core task T002. Although T003 and T004 both depend on T002, T003 is placed before T004 because of its higher priority and stricter delay requirements. T005, as a non-real-time task, is placed at the back of the queue. The final sequence of tasks to be scheduled is: T002->T001->T003->T004->T005.

[0081] The resource allocation submodule then begins its work. Based on the optimization scheme provided by the scheduling strategy generation module, it allocates suitable computing nodes to the ordered task sequence. The node selection unit of this submodule references a suitability score matrix calculated by the upstream module. Assume there are three currently available computing nodes: node N-21 (high-performance server, currently lightly loaded), node N-33 (general-purpose server, currently moderately loaded), and node N-47 (edge ​​computing gateway, limited resources but close to the fault data source). For the highest priority fault location task T002, the node selection unit finds that it has the highest suitability score with node N-21 from the score matrix. This is because N-21's sufficient CPU and memory resources can meet the intensive computing needs of T002, and its current low load ensures low latency. Therefore, T002 is preferentially assigned to N-21. The load balancing unit plays a crucial role in the allocation process. It detects that if the subsequent high-resource task T003 is also assigned to N-21, it may cause resource strain on that node in the future. Therefore, when selecting a node for T003, although N-21's suitability score is still high, the load balancing logic intervenes, guiding it to node N-33, which also meets resource requirements and has a lower load. The fault-tolerant allocation unit configures a backup plan for the critical task T002. While allocating T002 to the primary node N-21, it designates node N-33 as its backup node and synchronously warms up the task replica and related data on N-33. Once the system detects a heartbeat loss or performance drop on N-21, the task execution context will seamlessly switch to N-33.

[0082] The task triggering submodule is the endpoint of action execution. It receives the final task assignment instruction list from the resource allocation submodule, which specifies the target node for each task. For task T002 assigned to node N-21, the triggering submodule sends a task execution instruction packet to the task execution agent running on N-21 via Remote Procedure Call (RPC) or message queue. This instruction packet uses a standardized protocol format and includes the task ID, the storage path of the executable program or image, input parameters, environment variable configuration, and the expected storage location of the output results. After receiving the instruction, the task execution agent reserves local resources, and after confirming resource availability, pulls the necessary computing programs and data, and starts an independent process or container to execute the task. For tasks with data dependencies, such as T003 depending on the output of T002, the triggering submodule's instruction includes a waiting condition. After starting, the task agent of T003 continuously listens for the ready signal of T002's output data, and only begins formal calculation after the signal arrives, thus ensuring the correct execution order between tasks. The entire dispatch and triggering process is accompanied by detailed status log recording. These logs are sent to the dynamic feedback adjustment module in real time, providing a data foundation for subsequent monitoring and evaluation. Through this refined pipeline operation, the module transforms static scheduling strategies into dynamic and reliable task execution flows.

[0083] Example 5: The dynamic feedback adjustment module optimizes the computing power scheduling process through a continuous closed-loop control mechanism. This module immediately initiates its monitoring and adjustment functions after a task is assigned to a computing node. Taking a fault handling scenario as an example, when the task assignment and execution module assigns the fault location task T002 to node N-21 and starts execution, the execution monitoring submodule begins to track the task's running status on the node in real time. The execution monitoring submodule deploys lightweight monitoring agents on each computing node. These agents are embedded into the task execution environment through hook programs, periodically collecting fine-grained metrics at the task level. For example, for task T002, the monitoring agent records its CPU time usage, memory working set size, disk I / O throughput, and network connection status. Simultaneously, it captures execution progress information, such as the percentage of completed computation steps or the number of iterations, by parsing task logs or standard output streams. The collected raw performance data is encapsulated into structured event messages and transmitted in real time to the central data processing unit through a high-throughput stream processing platform. This unit performs preliminary data aggregation, such as calculating the average resource utilization or instantaneous peak value of the task within the sliding time window. The processed data stream is persisted to a time-series database for subsequent in-depth analysis. At the same time, some key indicators are sent to the real-time alarm engine to detect task freezes or abnormal termination.

[0084] The performance evaluation submodule periodically extracts a batch of task execution data from the time-series database and compares it with the expected targets preset by the scheduling strategy generation module. These expected targets are derived from optimization metrics in the scheduling strategy. For example, for task T002, expected targets might include completing computation within 3 seconds and maintaining CPU utilization between 70% and 80%. The comparison process uses a difference calculation algorithm to calculate the deviation between the actual and expected values ​​for each key performance indicator (KPI), such as the difference between the actual task completion time and the maximum allowable delay, or the deviation of resource utilization efficiency from the ideal value. Deviation analysis not only focuses on absolute values ​​but also incorporates historical patterns for contextual awareness. For example, during peak power grid fault periods, node loads are generally high, and a slight increase in task response latency might be considered an acceptable deviation, while at other times it might trigger an alarm. The analysis results are summarized in a computing power scheduling performance evaluation report. This report uses a structured format and includes a summary (overall deviation score), a detailed section (deviation decomposition for each task), and root cause analysis recommendations. The report is distributed to relevant modules and operations personnel via message bus or API interface.

[0085] The strategy adjustment submodule is the execution endpoint of the feedback loop. It receives the evaluation report generated by the effect evaluation submodule and parses the deviation analysis results. When the report indicates a systematic deviation, this submodule triggers the optimization process of the scheduling strategy generation module. The triggering mechanism can be event-driven, such as setting a deviation threshold and automatically initiating a strategy re-optimization request when the overall deviation score exceeds 0.1 (normalized value). Alternatively, it can be periodic, such as checking for adjustments every 5 minutes in conjunction with the evaluation cycle. When triggering optimization, the strategy adjustment submodule carries contextual information, such as the current resource status snapshot, deviation analysis details, and suggested weight adjustment directions. This information is passed as input parameters to the optimization algorithm of the scheduling strategy generation module. Based on this, the scheduling strategy generation module reruns its multi-objective optimization process to generate an updated scheduling strategy. The new strategy is gradually applied to subsequent task assignments, thereby achieving dynamic and adaptive adjustment of the computing power scheduling strategy, forming a complete feedback loop from monitoring to evaluation to optimization.

[0086] In the continuation of the fault handling scenario, assuming that task T002 executes slower than expected on node N-21, the monitoring submodule detects that its CPU utilization hovers around 50% while memory access latency is high. The performance evaluation submodule, after comparing with the expected target, determines that there is a performance deviation. The report indicates that the root cause of the deviation may be memory bandwidth contention on node N-21. Based on this, the policy adjustment submodule triggers optimization. In the next optimization, the scheduling policy generation module may lower the suitability score for node N-21 or adjust the task allocation logic to avoid similar contention, thereby achieving a more balanced load distribution in subsequent tasks. The entire dynamic feedback adjustment module is designed with low latency and high reliability in mind. Its components are deployed using a microservice architecture, supporting horizontal scaling to meet the monitoring needs of large-scale clusters and ensuring that the scheduling system can quickly respond to changes in the operating environment.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0088] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cross-domain computing power dynamic scheduling system for the power industry based on super-intelligent fusion, characterized in that, The system includes: The computing power demand perception module collects computing power demand data from various business domains in the power industry in real time, identifies the characteristics of computing resource demand under different business scenarios, and generates a cross-domain computing power demand characteristic map. The resource status monitoring module continuously tracks the real-time load status and available resources of each computing node, obtains the resource utilization, response latency and energy consumption indicators of the computing nodes, and constructs a distributed computing power resource status matrix. The scheduling strategy generation module analyzes the matching relationship between computing resource requirements and available resources in different business scenarios based on the cross-domain computing power demand feature map and the distributed computing power resource status matrix, and generates a multi-objective optimized computing power scheduling strategy scheme. The task assignment and execution module is based on a multi-objective optimization computing power scheduling strategy. It allocates power business computing tasks to the optimal computing nodes according to priority and resource demand characteristics, triggering the cross-domain execution process of computing tasks. The dynamic feedback adjustment module monitors the execution status and resource usage of computing tasks in real time, collects task execution progress and resource consumption data, generates a computing power scheduling effect evaluation report, and feeds it back to the scheduling strategy generation module. The computing power demand sensing module includes: The business feature extraction submodule parses the computing task requests from various business domains in the power industry, extracts the characteristics of task type, data scale, processing timeliness and computing complexity, and forms a description of business computing power demand characteristics. The demand correlation analysis submodule identifies data dependencies and execution order constraints between computing tasks in different business domains, and establishes a cross-business domain computing power demand correlation network. The demand graph construction submodule integrates the description of business computing power demand characteristics with the cross-business domain computing power demand association network to generate a cross-domain computing power demand characteristic graph with a hierarchical structure. The resource status monitoring module includes: The node performance acquisition submodule periodically polls the CPU utilization, memory usage, storage I / O performance, and network bandwidth usage of each computing node, and records the real-time performance metrics of the computing nodes. The state matrix update submodule compares and analyzes the real-time performance indicators of computing nodes with historical operating data, identifies abnormal fluctuations in resource usage, and updates the node state information in the distributed computing power resource state matrix. The resource prediction submodule predicts the available resources and performance changes of each node in the future period based on the historical load change patterns and current resource usage trends of computing nodes.

2. The power industry cross-domain computing power dynamic scheduling system based on super-intelligent fusion as described in claim 1, characterized in that, The scheduling policy generation module includes: The demand matching analysis submodule performs multi-dimensional matching between the business computing needs in the cross-domain computing power demand feature map and the node capabilities in the distributed computing power resource status matrix, and calculates the adaptability score of each node for a specific task. The strategy optimization submodule comprehensively considers multiple objectives such as task priority, resource utilization balance, energy efficiency and response latency to generate a set of candidate computing power scheduling strategies that meet the needs of power business. The strategy evaluation submodule simulates the execution effect of different candidate computing power scheduling strategies in a distributed environment and selects the scheme with the highest comprehensive score as the final multi-objective optimized computing power scheduling strategy.

3. The power industry cross-domain computing power dynamic scheduling system based on super-intelligent fusion as described in claim 1, characterized in that, The task assignment and execution module includes: The task priority sorting submodule prioritizes the queue of tasks waiting to be scheduled based on the urgency and importance of the computing tasks in the power business. The resource allocation submodule, following a multi-objective optimized computing power scheduling strategy, prioritizes the allocation of high-priority tasks to computing nodes that meet resource requirements, ensuring the timeliness of critical task execution. The task triggering submodule sends task execution instructions to the target computing node, initiating the distributed processing flow of the cross-domain computing task.

4. The power industry cross-domain computing power dynamic scheduling system based on super-intelligent fusion according to claim 1, characterized in that, The dynamic feedback adjustment module includes: The execution monitoring submodule tracks the execution progress and resource consumption of tasks on each computing node in real time, and collects data on task processing rate and resource utilization efficiency. The performance evaluation submodule compares the actual execution of tasks with the expected scheduling goals, analyzes the execution deviation of the computing power scheduling strategy, and generates a computing power scheduling performance evaluation report. The strategy adjustment submodule triggers the optimization process of the scheduling strategy generation module based on the deviation analysis results in the computing power scheduling effect evaluation report, thereby realizing the dynamic adjustment of the computing power scheduling strategy.

5. The power industry cross-domain computing power dynamic scheduling system based on super-intelligent fusion according to claim 1, characterized in that, The business feature extraction submodule includes: The task parsing unit breaks down the workflow of power business calculation tasks and identifies the calculation characteristics and data processing requirements of each stage of the task. The feature quantization unit transforms the computational complexity, data throughput, and timeliness requirements of a task into quantifiable computing power demand indicators. The feature integration unit normalizes computing power demand indicators from different dimensions to form a standardized description of business computing power demand characteristics.

6. The power industry cross-domain computing power dynamic scheduling system based on super-intelligent fusion according to claim 1, characterized in that, The node performance acquisition submodule includes: The metrics acquisition unit obtains real-time performance data of computing nodes, including processor load, memory usage, and network latency, through remote API calls. The data cleaning unit filters out outliers and noise from the collected data to ensure the accuracy and reliability of performance indicators. The indicator storage unit stores the cleaned performance data in a distributed database according to time series, supporting historical data backtracking and analysis.

7. The power industry cross-domain computing power dynamic scheduling system based on super-intelligent fusion according to claim 2, characterized in that, The strategy optimization submodule includes: The target weight allocation unit dynamically adjusts the weight ratios of resource utilization, energy efficiency, and response delay optimization targets based on the characteristics of the power business scenario. The strategy generation unit, based on a multi-objective optimization algorithm, explores the space of feasible computing power scheduling strategies that satisfy the constraints of each objective. The strategy screening unit removes obviously inferior solutions from the feasible strategy space and retains a set of candidate computing power scheduling strategies with optimization potential.

8. The power industry cross-domain computing power dynamic scheduling system based on super-intelligent fusion according to claim 3, characterized in that, The resource allocation submodule includes: The node selection unit determines the set of target nodes most suitable for executing the current task based on the characteristics of task resource requirements and the computing node suitability score. The load balancing unit takes into account the current load of the target node and avoids excessive concentration of resources on a few high-performance nodes; The fault-tolerant allocation unit configures backup execution nodes for critical tasks, automatically switching to the backup node to continue execution when the primary node fails.

Images