A computer system intelligent backup method based on thermal power plants

Abstract

The application provides a computer system intelligent backup method based on a thermal power plant, relates to the fields of power system automation and computer data storage technology, and comprises the following steps: S1, acquiring scene characteristic parameters, medium state parameters and data demand parameters according to power plant system signals, and outputting a parameter collection set; S2, constructing a data demand portrait according to the parameter collection set, calculating and outputting a data demand weight; and S3, calculating and outputting an adaptation degree matrix of each medium node in a distributed storage node cluster according to the data demand weight. Compared with the prior art, the application has the following beneficial effects: 1. The relationship between core data backup demand and medium load under a dynamic coordination starting scene can be coordinated; 2. The medium overload risk caused by the explosive growth of data under a maintenance scene can be effectively coped with; and 3. The resource utilization efficiency is continuously improved in long-term operation, and high backup performance is stably maintained.

Description

Technical Field

[0001] This invention relates to the fields of power system automation and computer data storage technology, and in particular to an intelligent backup method for computer systems based on thermal power plants. Background Technology

[0002] With the continuous improvement of the intelligence level of thermal power plants, computer systems are playing an increasingly important role in key aspects such as unit control, operation monitoring, and fault diagnosis. The control parameters, operating data, equipment status and other information generated during the power generation process have become core assets to ensure the safe and stable operation of power plants. The power industry has put forward higher requirements for data backup systems, which need to achieve real-time protection, rapid recovery and long-term traceability of key data.

[0003] Against this backdrop, CN119396629A, a cloud-based data storage and backup method, was proposed. Through hierarchical data storage and cross-regional synchronization mechanisms, it provides a basic solution for power system data protection. The technical solution includes: classifying power data according to importance level, adopting storage configuration strategies such as RAID-1 and RAID-5, realizing distributed storage through data block and hash mapping, and using multiple synchronization mechanisms for cross-regional mirror backup. This solution improves the reliability of data storage and cross-regional availability to a certain extent.

[0004] While this technical solution optimizes data availability through importance grading and synchronous backup, it fails to address the issues of media failure prediction and dynamic scenario adaptation. Firstly, the solution employs a static data grading standard, which cannot adapt to the real-time adjustment requirements of backup strategies during dynamic changes in power plant operations, such as unit startup, equipment maintenance, and peak grid periods. For instance, during unit startup in November 2023, boiler ignition parameters generated by the DCS control system should have been prioritized for backup to low-latency media. However, because the system still graded data according to the standard for regular time periods, this critical data was assigned to high-latency tape media for backup. As a result, during a sudden network fluctuation at the critical startup stage, the backup latency of core parameters reached 180ms, far exceeding the ≤50ms standard required for startup scenarios. This caused control commands to become asynchronous, ultimately extending the unit startup time by 35 minutes and directly resulting in significant economic losses. Secondly, the solution lacks a predictive monitoring mechanism for the health status of backup media, failing to provide early warnings and mitigate risks when potential failures occur in media such as hard drives and tapes. Most importantly, the solution lacks a closed loop for backup performance evaluation and strategy self-optimization, which prevents the backup system from dynamically optimizing based on actual operating results. This makes it difficult to guarantee backup success rate and service quality in complex and ever-changing power plant environments. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, the technical problem to be solved by this invention is to provide an intelligent backup method for computer systems in thermal power plants, which can dynamically resolve issues such as priority and load conflicts in startup scenarios and data volume and media overload in maintenance scenarios, thereby achieving precise allocation and closed-loop optimization of backup resources.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: The present invention provides an intelligent backup method for a computer system based on a thermal power plant, comprising the following steps: S1. Obtain scene characteristic parameters based on power plant system signals. Medium state parameters and data requirement parameters Output parameter collection set ; S2, Collect data based on parameters Build a data requirement profile, calculate and output the data requirement weights. ; S3, Based on data demand weights Calculate and output the fit degree of each media node in the distributed storage node cluster. matrix; S4, according to The matrix is ​​used to solve the multi-objective scheduling objective function using the MPC algorithm. Determine the optimal media allocation scheme and output the scheduling instruction set. ; S5. According to the scheduling instruction set Perform a backup operation and obtain the actual backup completion rate. and actual recovery time Based on the actual backup completion rate and actual recovery time Determine the optimization coefficients According to the optimization coefficient Feedback updates steps S2 and S4 form a closed loop.

[0007] In the preferred embodiment, the specific steps of step S1 are as follows: S11, Scene feature parameters S include scene type Scene duration and scenario priority ; Among them, scene type Including startup Q1, maintenance Q2, and peak period Q3; scenario priority. include class, class, class; S12, Medium State Parameters Including media loading rate Media health and medium response delay ; S13, Data Requirement Parameters Including data real-time requirements Data redundancy requirements and data compliance retention period ; Among them, the requirement for real-time data Including millisecond-level t1, minute-level t2, and offline t3; data redundancy requirements. Including dual-medium r2 and single-medium r1; S14, Parameter Acquisition Interval Dynamically set according to scene type: Q1 Q2 Q3 .

[0008] In the preferred embodiment, the specific steps of step S2 are as follows: S21. Based on data requirement parameters The need for real-time data Data redundancy requirements and data compliance retention period Perform normalization processing to determine the normalized value. , and Data requirement parameters The calculation formula for normalization is as follows: Real-time normalized value: ,in, As a real-time benchmark value, Pick hour , hour ; Redundancy normalized value: ; Compliance normalized value: ,in, The year serves as the compliance benchmark. S22. Based on scene feature parameters Set weight coefficients , , and : Q1 , , ; Q2 , , ; Q3 , , ; S23. Based on the normalized value , and and weighting coefficients , , Determine the calculation , The formula is: ,and The value range is [0,1].

[0009] In the preferred embodiment, step S3 specifically involves the following steps: S31. Based on the medium state parameters in step S1 Extracting normalized values , and : Normalized load rate: ; Health normalized value: ; Normalized response latency: ,in This is the baseline value for delay; S32. Determine the adaptation coefficient based on the scene feature parameter S: real-time correlation coefficient Redundancy correlation coefficient Compliance correlation coefficient ; S33. Based on the normalized value , and and fit coefficient , and Sure Matrix, calculation The formula is: ; Output A matrix, where each row corresponds to a medium node and the column values ​​are... , The value range is [0,1].

[0010] In the preferred embodiment, the multi-objective scheduling objective function in step S4 The solution process is as follows: S41, according to Matrix and medium response delay Determine the objective function objective function The calculation formula is as follows: ; in, This refers to the delay penalty coefficient. According to the scene type Setting, Q1 Q2 Q3 ; S42. Set constraints, including: Medium loading rate constraint: Media health constraints: point; Backup time constraints: And in Q1 Q2 Q3 ; S43. According to the objective function Given the constraints, the MPC algorithm is used to solve the problem. The maximum value is used to determine the optimal media node ID and backup priority, and a scheduling instruction set is output. .

[0011] In the preferred embodiment, step S4 further includes step S44 for dynamic adjustment of scene media conflicts, and the specific steps of S44 are as follows: S441. According to scene type Data demand weight and media loading rate Determine the type of conflict; when , And the optimal medium If so, it is determined to be a startup conflict; when And the amount of maintenance-related data If so, it is determined to be a maintenance conflict; The formula for calculating the available capacity of the medium is as follows: , This represents the total capacity of the medium. S442. Execute the adjustment strategy according to the conflict type, and obtain the execution result as follows: If a startup conflict is identified, the priority scheduling mechanism is executed: pause. Non-core data backup tasks, according to the formula Adjust the media loading rate to ; in, For non-core data volumes that are paused, This represents the number of tasks that have been paused. If a maintenance conflict is identified, the media expansion and data offloading mechanism will be implemented: priority will be given to calling... and Backup media, for the replacement cycle Low-frequency maintenance data according to formula Perform cloud-based traffic distribution and control ; S443. Based on the execution result of the adjustment strategy, output the adjusted scheduling instruction set. Replace the original scheduling instruction set .

[0012] In the preferred scheme, the optimization coefficient in step S5 The calculation and feedback process is as follows: S51. Set KPI target value: Core data backup success rate and recovery time minute; S52. Based on the actual backup completion rate and actual recovery time Sure ,calculate The formula is: ; S53, according to Value feedback update weight coefficient , , and delay penalty coefficient : when Then adjust the weighting coefficients of step S2: , ; when Then adjust the penalty coefficient in step S4. .

[0013] In a preferred embodiment, the present invention also provides an intelligent backup device for a computer system in a thermal power plant, characterized in that it comprises: The parameter acquisition module is used to perform step S1 above and acquire scene feature parameters. Medium state parameters and data requirement parameters Output parameter collection set ; The data requirement profiling module is used to perform step S2 above, collecting data based on the parameters. Calculate data demand weights ; The media compatibility calculation module is used to perform step S3 above, based on the data requirement weight. Calculate media compatibility matrix; The MPC scheduling decision module is used to execute steps S4 and S44 above, and to solve the multi-objective function using the MPC algorithm. It also handles scenario media conflicts and outputs a set of scheduling instructions. or ; The closed-loop optimization module is used to execute step S5 above, according to the scheduling instruction set. Perform a backup operation and obtain the backup KPIs. Determine the optimization coefficients based on the KPIs. ,according to It also provides feedback to update the weighting and penalty coefficients.

[0014] In the preferred embodiment, the parameter acquisition module includes: Scene sensors are used to interface with the power plant's SIS (System-In-Service) and MIS (Management-In-Service) systems to acquire scene feature parameters based on the outputs of the SIS and MIS systems. ; The media monitoring unit incorporates a hard drive SMART chip and a magnetic tape wear sensor. Based on the outputs of the hard drive SMART chip and the magnetic tape wear sensor, it obtains the media load rate. Health Response delay and total capacity ; The data requirement acquisition unit is used to interface with the DCS control system and, based on the output of the DCS control system, obtain data real-time requirements, redundancy, compliance requirements, and data volume. Output .

[0015] In the preferred embodiment, the MPC scheduling decision module includes: Objective function construction unit, based on and Build The expression; The constraint configuration unit is used to generate load, health, and backup time constraints based on preset media. MPC solver, used to solve problems based on the objective function Given the constraints, the gradient descent method is used to solve the problem. The maximum value is used to determine the optimal medium allocation scheme; The conflict adjustment unit is used to perform the above step S44, based on scene feature parameters. and medium state parameters Detect conflicts and implement adjustment strategies to generate an adjusted set of scheduling instructions. .

[0016] This invention provides an intelligent backup method for computer systems in thermal power plants. Through the coordination of the aforementioned structures, it offers the following advantages compared to existing technologies: First, by constructing a decision-making mechanism that coordinates scenario characteristics, media status, and data requirements, and introducing model predictive control algorithms for dynamic optimization scheduling, the relationship between core data backup requirements and media load in the startup scenario can be dynamically coordinated. This avoids the problem of core data backup interruption caused by the failure to handle this conflict in existing technologies. Since non-core data backup can be suspended through a priority scheduling mechanism to release media load, the stable backup of critical data during the startup process can be guaranteed, ensuring data security during the startup phase. Secondly, it can effectively address the risk of media overload caused by the surge in data volume during maintenance scenarios. By adding a dynamic detection and adjustment mechanism for scenario and media conflicts, it can proactively identify and resolve core conflicts such as high-priority data and media overload during startup, and the surge in data volume and insufficient media capacity during maintenance. This balances media capacity pressure and data backup integrity, meeting the reliable storage needs of a large amount of related data in maintenance scenarios. Third, by establishing a closed-loop optimization loop based on feedback from key performance indicators, the system's key configurations, such as weight coefficients and penalty parameters, can be adaptively adjusted according to the actual backup effect. This enables the backup system to automatically adjust its strategy based on the actual operating effect, thereby continuously improving resource utilization efficiency and maintaining high backup performance in long-term operation. Attached Figure Description

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a main view diagram of the process structure of this invention; Figure 2 This is a statistical chart showing the scene distribution and data volume of this invention; Figure 3 This is a comparison chart of the backup success rate of this invention; Figure 4 This is a comparison chart of the recovery time of this invention; Figure 5 This is a comparison chart of the success rate of conflict resolution in this invention; Figure 6 This is a comparison chart of the medium load stability rate of the present invention; Figure 7 This is a diagram illustrating the actual operation of the present invention; Figure 8 This invention presents a heatmap of KPI indicators and a curve showing the model training loss. Figure 9This is a system architecture diagram of the present invention; Figure 10 This is a resource usage curve diagram of the present invention. Detailed Implementation

[0018] To better understand the purpose, structure, and function of this invention, the embodiments and features described herein can be combined with each other without conflict. Exemplary embodiments of this disclosure will be described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding; these should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures (such as the basic storage principles of conventional computer data backup, and the general data interaction protocols of SIS / MIS / DCS systems in thermal power plants) are omitted in the following description.

[0019] In the fields of power system automation and computer data storage technology, data backup systems are crucial infrastructure for ensuring the stable operation of computer systems in thermal power plants. Traditional backup methods save critical data through periodic or triggered mechanisms, enabling basic recovery in the event of system failures or data loss. However, with the advancement of smart power plant construction, data backup systems need to cope with more complex operating scenarios and higher reliability requirements.

[0020] In recent years, data backup technology has evolved from single-media, fixed-strategy models to new systems supporting multi-media collaboration and basic strategy scheduling to adapt to the diverse business needs of power plants. Multi-media backup systems can uniformly schedule heterogeneous storage resources such as hard drives, tapes, and cloud storage, typically improving the utilization efficiency of backup resources. In typical scenarios such as start-up and shutdown of thermal power plants, equipment maintenance, and peak grid periods, efficient backup systems need to dynamically adjust strategies based on real-time scenario characteristics, media status, and data attributes. However, common multi-media backup systems have not yet fully achieved closed-loop integration of scenario awareness, media health prediction, and strategy self-optimization, making it difficult for the system to continuously maintain optimal backup performance under complex operating environments.

[0021] As thermal power plants transform towards intelligent operation, data backup technology is gradually developing towards "multi-media collaborative scheduling," "dynamic scenario adaptation," and "closed-loop intelligent optimization." On the one hand, backup media have expanded from single hard drives to combinations of multiple media such as NVMe hard drives, tape drives, and cloud storage, requiring pooling technology to break physical boundaries and achieve elastic scheduling. On the other hand, the dynamic nature of power plant production scenarios requires backup strategies to be adjusted according to scenario characteristics. Against this backdrop, intelligent backup solutions that integrate multi-dimensional parameter acquisition, model predictive control (MPC), and dynamic adjustment of scenario-media conflicts have become the focus of technological development. Traditional static backup technologies are no longer sufficient to meet these needs. For example, existing technology 1 only achieves basic automatic backup without media load conflict handling; existing technology 4 optimizes cross-regional synchronization delay but does not link the collaborative relationship between thermal power plant-specific scenarios and media status; existing technology 3 is only based on file access behavior classification and lacks adaptation logic for media capacity and data volume. None of these can meet the dynamic backup needs of intelligent thermal power plants.

[0022] The existing technologies have significant limitations in building backup systems for thermal power plants: In terms of media management, they are mostly limited to single media or fixed allocation modes, failing to achieve multi-media pooling and dynamic load balancing. For example, existing technology 2 is geared towards IT container resources and lacks industrial-grade multi-media collaborative design. In terms of scenario adaptation, they only cover general backup scenarios and lack adjustment mechanisms designed for specific scenarios such as thermal power plant start-up conflicts and maintenance overloads. In terms of optimization mechanisms, they lack pre-trained models based on historical data and closed-loop feedback. For example, existing technologies 1-4 lack parameter iterative optimization logic, making it difficult to continuously adapt to data characteristics and scenario changes. These limitations prevent these technologies from meeting the core requirements of intelligent thermal power plants for data backup: "precise resource utilization, early fault avoidance, and intelligent scenario adaptation." Therefore, a dedicated backup solution integrating multi-dimensional perception, intelligent scheduling, and dynamic optimization is urgently needed.

[0023] Example 1 In this embodiment, as shown... Figure 2 As shown, the core training objects are: the gradient descent solution model for MPC scheduling and the threshold and coefficient model for adjusting medium-level conflicts in the mid-scene. The training objective is to determine optimal parameters such as the delay penalty coefficient. Capacity adaptation coefficient Maximum diversion ratio Etc., ensure that the model outputs scheduling strategies that meet KPI targets in the Q1 / Q2 / Q3 scenarios of thermal power plants; In this embodiment, the specific steps for training data preparation and preprocessing are as follows: First, historical scenario data from January 2022 to December 2023 was collected from a 300MW thermal power plant, covering startup Q1 (2 hours after boiler ignition, 120 records in total), maintenance Q2 (during equipment maintenance work orders, 80 records in total), and peak period Q3 (unit load). (90% of the rated value, 200 records in total) Scene feature data: (Called Q1 / Q2 / Q3) (Actual duration) (Priority is marked according to power plant specifications); Media status data: (Real-time load rate of hard drive / tape / cloud) (Health score calculated based on SMART parameters) (Measured response delay) (Total capacity of the medium); Data requirements: (Data real-time annotation) (Redundancy requirements) (Compliance retention period) (Actual measured data volume); Labeling: Based on the experience of power plant operation and maintenance experts, label the "optimal scheduling result" corresponding to each historical data (such as non-core data that should be suspended during Q1 conflicts, and the expansion / diversion ratio during Q2 overload).

[0024] In this embodiment, preprocessing is performed on the collected data: 1. Outlier cleaning: removal (Sensor malfunction) (Invalid delays and other abnormal data) resulted in 450 valid records being retained, which were then divided into training sets in a 7:3 ratio. (315 items) and the validation set (135 items); 2. Data normalization: For , Continuous variables are treated according to the normalization formula (e.g.) This ensures that the magnitudes of the parameters in the input model are consistent. 3. Feature Alignment: The preprocessed data is associated with the labeled "optimal scheduling result" to form an "input feature output label" pair for each record (e.g., input "Q1 scenario, ...). , The tag "Pause 5GB+3GB non-core data" (”).

[0025] This step outputs: a standardized training set. With the validation set (Including input features and output labels), which serve as the basic input for subsequent model initialization and iterative training.

[0026] In this embodiment, the specific steps for initializing the model structure are as follows: Based on media compatibility calculation, MPC scheduling, and conflict adjustment, initialize the core parameters and structure of the model: 1. MPC scheduling model initialization: Prediction Time Domain Based on scenario type settings (Q1) Each acquisition cycle = 2.5s, Q2 Each acquisition cycle = 10s, Q3 Each acquisition cycle = 5s), control time domain One acquisition cycle (basic parameters for MPC solution); Delay penalty coefficient Initial values ​​are set according to the scenario (Q1=0.8, Q2=0.5, Q3=0.7), and the capacity adaptation coefficient $\mu=0.3$ (the initial value of the objective function for scenario Q2). Constraint threshold: , , (Initial threshold for conflict adjustment).

[0027] 2. Initialization of conflict adjustment strategy: Startup conflict target load (Leave a buffer if it is 5% below the constraint value); Maintenance conflict expansion ratio limit (To avoid overloading of the backup medium); Data priority threshold: core data Non-core data (The basic threshold for collision detection).

[0028] This step outputs: the initialized MPC scheduling model and conflict adjustment strategy model, and their parameters ( , , (etc.) are used as initial values ​​for iterative training.

[0029] In this embodiment, as shown... Figure 8 As shown, the specific steps of the iterative training process are as follows: Will Each record in the dataset is input sequentially to initialize the model. The model parameters are adjusted using gradient descent to minimize the deviation between the model's output scheduling result and the labeled data. Specific steps include: 1. Single-sample training (with) (Taking a Q1 scenario record as an example) Step 1: Input the preprocessed sample features (Scenario Q1, , point, , Core data , ); Step 2: The model outputs the initial scheduling results (before adjustment, because...). The output "Core data backup interrupted" deviates significantly from the label "8GB of non-core data paused, core data backup proceeded normally" (the output shows "Core data backup interrupted"). Step 3: Calculate the loss function (Based on KPI deviation:) , , , , ), This calculation yields (The deviation is relatively large); Step 4: Gradient Descent Parameter Adjustment: Adjust parameters along the negative gradient direction of the loss function. (Adjusted from 80% to 82%), Non-core data pause threshold ( (Keep it unchanged), recalculate the model output, and this time the output will show "Pause 8GB of non-core data, "Core data backup successful," loss It dropped to 0.02.

[0030] 2. Batch Iterative Training: Will 315 records were input into the model in batches (batch size = 30), and the average loss was calculated after training for each batch. ; like (If a preset loss threshold is set, then adjust the model parameters:) according to Fine-tuning according to Fine-tuning according to Fine-tuning; Repeat batch training until three consecutive batches are completed. Preliminary assessment indicates that the training has converged.

[0031] This step's output: (The output is...) The trained model parameters (e.g., Q1 scenario) , Q2 scenario , The average loss between the model output and the labeled training set. .

[0032] In this embodiment, the specific steps for model validation and convergence determination are as follows: Verification set (135 records) Input the trained model and evaluate its generalization ability on unseen data: 1. Verification Metric: Core Data Backup Success Rate Recovery time Conflict resolution success rate Medium load stability (KPI indicators); 2. Verification process: For The metric deviations for the records in Q1 / Q2 / Q3 scenarios are calculated separately. If the deviations for all scenarios are... , minute, , If so, the model validation is successful; 3. If the verification fails (e.g., in scenario Q2) If so, return to step 1.3 and supplement. Samples that did not pass the test Then, iterate the training again until all validation metrics are met.

[0033] This step outputs: a converged pre-trained model, including the final parameters (Q1 scenario). , Q2 scenario , Q3 scenario This can be used for subsequent experimental verification.

[0034] Example 2 like Figure 1 , 3 As shown in Figures 7 and 10, this embodiment illustrates an intelligent backup method for a computer system based on a thermal power plant, including the following steps: selecting 30 sets of data from historical data, as detailed in the table below:

[0035] In this embodiment, S1: Multi-dimensional parameter acquisition, the input parameters are as follows: Scene feature parameters Scene type (Start-up Q1 / Maintenance Q2 / Peak Period Q3), Scenario Duration (Unit: h), Scene Priority (Levels 1-5); Medium state parameters Medium loading rate (0~100%) Media health (0~100 points, calculation method:) bad track number Tape wear count, media response delay (Unit: ms) Total media capacity (Unit: GB) Data requirement parameters Data real-time requirements ( / / ), data redundancy requirements ( Dual-media / Single-media), data compliance retention period (Unit: Year), Data Volume (Unit: GB)

[0036] Specifically, the scene sensor interfaces with the SIS system to detect when the "boiler ignition command" triggers Q1 ( , ), the MIS system detects the "maintenance work order" and triggers Q2 ( , ), , Connecting unit load monitoring triggers Q3 (unit load ≤ 90% of rated value, , ); Each media monitoring unit collection Q1 (Capturing sudden changes in media load during startup), Q2 (During maintenance, the load is stable but capacity needs to be monitored), Q3 (Peak load fluctuations are moderate); The data acquisition unit connects to the DCS system to extract the associated device IDs and data volume. ,Sure (e.g., DCS control logic:) , , Year, Vibration curve during maintenance: , , Year, ).

[0037] Output parameters: Parameter collection set .

[0038] In this embodiment, S2: The specific steps for constructing the data requirement profile are as follows: S21: Normalization processing data requirement parameter D: Real-time normalized value: , (Benchmark value), for example: hour, ; Redundancy normalized value: ; Compliance normalized value: , Year (baseline value), for example: In that year, ; Normalized data volume (Q2 scenario only): , (Maintenance data volume baseline value), for example: hour, .

[0039] S22: Based on scenario type Set weighting coefficients ( Q2 scenario additional ,Adjustment , , ,make sure ): Q1 (Power-on): (Prioritizing real-time performance) (Secondary redundancy) (Secondary compliance); Q2 (Maintenance): (Sub-real-time) (Priority Redundancy) (Prioritize compliance) (Data volume correlation); Q3 (peak period): (Prioritizing real-time performance) (Secondary redundancy) (Sub-compliance).

[0040] S23: Calculate the weight of data demand : Scene: ; Q2 scenario: ; Example: DCS control logic in scenario Q1 ; Vibration curve during maintenance in scenario Q2: .

[0041] Output parameters: ( The value ranges from 0 to 1, and Q2 ranges from 0 to 2. The larger the value, the higher the comprehensive requirements of the data in terms of "real-time performance + redundancy + compliance + data volume".

[0042] In this embodiment, S3: The specific steps for calculating the media compatibility are as follows: S31: Normalized processing of medium state parameters : Normalized load rate: ,example: hour, ; Health normalized value: ,example: Time-sharing ; Normalized response latency: , (Benchmark value), for example: hour, ; Normalized capacity values ​​(Q2 scenario only): , (Available capacity of the medium) (Useable capacity reference value for maintenance), for example: hour, .

[0043] S32: Set the adaptation coefficient: (Real-time correlation coefficient) (Redundancy correlation coefficient) (Compliance correlation coefficient) (Capacity correlation coefficient, only for Q2 scenario), the coefficient value is based on fitting historical data of the power plant (if maintenance scenario, the weight of the capacity coefficient needs to be increased).

[0044] S33: Calculate the fit of a single media node : Scene: ; Q2 scenario: ; Example: In scenario Q1, a certain tape drive ( , point, )of In the Q2 scenario, a certain cloud node ( , point, , )of .

[0045] Output parameters: Matrix (each row corresponds to a medium node, and the column values ​​are...) , The value ranges from 0 to 1, and Q2 ranges from 0 to 2. A larger value indicates that the medium is more suitable for the current data requirements.

[0046] In this embodiment, S4: The specific steps of the MPC multi-objective scheduling decision are as follows: S41: Constructing a multi-objective scheduling objective function (With the goal of "maximum compatibility + minimum latency", capacity adaptation is added to the Q2 scenario): Scene: ; Q2 scenario: ,in (Capacity adaptation factor); in Delay penalty coefficient: Q1 (Startup is sensitive to latency), in Q2 (High tolerance for delays during maintenance), Q3 (Peak periods are more sensitive to delays).

[0047] S42: Set constraints (based on power plant safety regulations, with additional capacity constraints added for the Q2 scenario): Medium loading rate constraint: (Avoid overload); Media health constraints: (to avoid faulty media); Backup time constraints: (Q1) Q2 Q3 ); Capacity constraints (Q2 scenario): , (Maximum diversion ratio).

[0048] S43: Solved using the MPC algorithm Maximum value: Prediction time domain: Take the next 5 acquisition cycles (2.5s for Q1, 10s for Q2, and 5s for Q3). Control time domain: 2 acquisition cycles; Solution method: Gradient descent, iterating to... convergence( The system outputs the optimal media node ID, backup priority, and initial capacity allocation, forming a scheduling instruction set. .

[0049] S44: Dynamic adjustment of scene media conflicts: S441: Detecting the type of conflict between the scene and the medium (based on...) and (Matching results) Startup conflict determination: In the Q1 scenario, core data ( Optimal medium for adaptation (Load exceeds constraints), and there are no other... Alternative media; Maintenance conflict determination: In the Q2 scenario, the amount of maintenance-related data. (Available capacity of the medium), and (Overload exceeds 20%, adjustment is required); in , This represents the optimal total capacity of the medium.

[0050] S442: Implement adjustment strategies for different conflict types: Startup conflict adjustment: 1. Task of extracting non-core data from the current medium: filtering Backup tasks, total data volume statistics ; 2. Calculate the number of tasks that need to be paused: based on the target load. (Leave a buffer if the value is below the constraint value by 5%), the formula is: ; 3. Press The mission was paused from childhood to adulthood, until... To meet the requirements, update the media load: ; 4. Output the adjusted backup sequence: Core data is executed first, and paused non-core data is marked as "restart after peak period".

[0051] Maintenance conflict adjustment: 1. Call up the backup media pool: Filter and Backup media, total available capacity. ; 2. Calculate the expansion ratio: (Expansion should not exceed 50% to avoid overloading of backup media); 3. Calculate the diversion ratio: ,Target ; 4. Implement traffic diversion: Transfer low-frequency maintenance data (update cycle) )according to Data is offloaded to cloud storage; high-frequency data (update cycle) Distribute to backup media; 5. Output adjusted capacity allocation: Optimal media storage Backup media storage Cloud storage .

[0052] S443: Output the adjusted scheduling instruction set Includes adjusted media IDs, backup priorities, task pause / split flags, and capacity allocation, replacing the original scheduling instruction set. .

[0053] Output parameters: Scheduling instruction set (When there is no conflict) or (In case of conflict) the format is Data ID, Primary Media ID, Backup Media ID (if any), Cloud Distribution Marker (if any), Backup Priority, Task Status (Executing / Paused) .

[0054] In this embodiment, as shown... Figures 3 to 6 As shown, the specific operation of S5: closed-loop feedback optimization is as follows: S51: Collect and back up KPI data: Backup completion rate Core Data Importance level data ; Recovery time Core Data Minutes, Importance Level Data minute; Conflict resolution success rate After resolving startup conflicts, the success rate of core data backup is ≥99.9%, and after resolving maintenance conflicts, the data integrity rate is ≥99.8%. Medium load stability After adjustment, the media load fluctuation is ≤5% (Q1 scenario), and the capacity utilization rate is ≥70% (Q2 scenario).

[0055] S52: Calculate optimization coefficients (Added conflict adjustment effect weighting):

[0056] in (Conflict resolution success rate target). (Load stability target); Example: Core Data ( ), minute( minute), ( ), ( )hour,

[0057] .

[0058] S53: Feedback Update (Added Parameter Updates Related to Conflict Adjustment): like (Significant deviation): 1. Adjust the weighting coefficients of step S2: , (make sure or ); 2. Adjust the target load / distribution ratio in step S44: (Startup scenario) (Maintenance scenarios, not exceeding 60%) like (Extremely large deviation): 1. Adjust the penalty coefficient in step S4. ; 2. Adjust the capacity adaptation coefficient in step S44. (Maintenance scenario); 3. Trigger media maintenance reminders (e.g.) Points or hour).

[0059] Output parameters: Updated , , , , , and , The feedback is sent to steps S2, S4, and S44 to form a closed loop.

[0060] Example 3: Unit start-up scenario, i.e., Q1: 2 hours after boiler ignition. , , Startup conflict scenarios; 1. Algorithm parameter settings S1 parameters: , , , Core data Non-core data ; S2 weighting coefficient: , , ; S3 fit coefficient: , , ; S4 parameters: ,constraint , ; Conflict adjustment objectives ; S5 KPI Targets: , minute, , .

[0061] 2. Demonstration of execution steps Input scenario parameters: Q1 (boiler ignition command triggered). ; Media parameters: Optimal media (NVMe hard drive): (Over-constraint) point, , , ;none Alternative media; Data parameters: Core data (DCS control logic): , Non-core data (Daily Report 1): , Non-core data (Daily Report 2): , .

[0062] S1: Parameter acquisition: Acquired once every 0.5 seconds, output: .

[0063] S2: Data Requirement Profile Calculation: Core Data Report 1 Report 2 .

[0064] S3: Medium compatibility calculation: Optimal medium: (High load leads to) However, it is the only compatible medium.

[0065] S43: MPC Scheduling Decision: Output Initialization DCS Logical ID (Priority 1), Report 1 ID (Priority 3), Report 2 ID (Priority 4), Optimal Media ID, No Backup Media .

[0066] S44: Startup conflict adjustment: S441: Conflict Detection: Core Data Adaptation Medium No alternative media was found, indicating a startup conflict. S442: Adjust strategy: Tasks that need to be paused: ; Filtering non-core data: Report 1 (5GB) + Report 2 (3GB) = 8GB < 10GB, it is necessary to pause some low-priority core data (none, only 1 core data item), actually pausing Report 1 + Report 2, releasing 8GB. (Meets the standard); S443: Output DCS Logical ID (Priority 1, Execute), Report 1 ID (Priority 3, Pause), Report 2 ID (Priority 4, Pause), Optimal Media ID, No Backup Media .

[0067] S5: Closed-loop optimization: Actual KPIs: (Core data completed) minute, , ( Fluctuation of 2%) calculate :

[0068] (No adjustment required).

[0069] Specifically, this embodiment addresses startup conflict scenarios by "pausing non-core data to release load" to resolve "core data priority media load conflicts." This embodiment achieves a core data backup success rate of ≥99.98% and load fluctuation of ≤2%, meeting the requirements of startup scenarios.

[0070] Example 4: Equipment maintenance scenario, i.e., Q2: 8 hours after the maintenance work order is triggered. , , Overload inspection scenario: 1. Algorithm parameter settings S2 parameters: , , , Maintenance data ; S2 weighting coefficient: , , , ; S3 fit coefficient: , , , ; S4 parameters: , ,constraint , , Conflict adjustment ; S5 KPI Targets: , minute, , .

[0071] 2. Demonstration of execution steps Input data: Scenario parameters: Q2 (triggering a maintenance work order). ; Media parameters: Optimal media (tape drive): , point, , , Backup media (hard drive): , , , Cloud: , ; Data parameters: Maintenance vibration curve (high frequency, update cycle 1 hour): , Maintenance log (low frequency, update cycle 48 hours): , .

[0072] S1: Parameter Acquisition: Data is collected every 2 seconds, and the output is: .

[0073] S2: Data Requirement Profile Vibration curve ,log .

[0074] S3: Media compatibility calculation, the optimal media is as follows:

[0075] ; Backup medium:

[0076] .

[0077] S43: MPC Scheduling Decision: Output initial Vibration curve ID (priority 1), Log ID (priority 2), Optimal medium ID, Backup medium ID .

[0078] S44: Maintenance Conflict Adjustment: S441: Conflict determination: Overload It was determined to be a maintenance conflict; S442: Adjust strategy: Backup media capacity expansion: , (35.7%), capacity increased by 50GB; Cloud-based traffic offloading: Traffic splitting logs (low frequency) ; Capacity allocation: 60GB for optimal media (vibration curve), 50GB for backup media (vibration curve), 36GB for cloud (log), and 44GB of remaining logs for backup media; S443: Output Vibration Curve ID (60GB+50GB, executed), Log ID (44GB+36GB, executed), Optimal Media ID, Backup Media ID, Cloud Distributor Marker .

[0079] S5: Closed-loop optimization: Actual KPIs: , minute, , (Medium capacity utilization rate); calculate :

[0080] (No adjustment required).

[0081] This embodiment addresses the overload scenario during maintenance by using "spare media expansion + low-frequency data diversion" to solve the problem of "data surge and media overload". If backup is only performed according to a fixed configuration, this scenario will fail directly. This embodiment has a backup success rate of ≥99.92% and a capacity utilization rate of 75%, which meets the needs of the maintenance scenario.

[0082] Example 5: Peak grid scenario, i.e., Q3: Unit load ≥ 90% of rated value. , , : 1. Algorithm parameter settings Q3 parameters: , , , ; S2 weighting coefficient: , , ; S3 fit coefficient: , , ; S4 parameters: ,constraint , ; S5 KPI Targets: , minute.

[0083] 2. Demonstration of execution steps Input data: Scenario parameters: Q3 (unit load = 95% of rated value). ; Media parameters: Media 1 (NVMe hard drive): , point, Media 2 (Hard Disk): , point, Media 3 (Magnetic Tape Drive): , point, ; Data parameters: Real-time load data of the unit ( , , Year, ).

[0084] S1: Parameter Acquisition: Data is collected once per second, and the output is: .

[0085] S2: Data Requirement Profile .

[0086] S3: Medium compatibility calculation: Medium 1: ; Medium 2: ; Medium 3: ( ,invalid).

[0087] S43: MPC Scheduling Decision: Output Real-time load data ID, medium 1 ID, priority 1 (No conflict, no adjustment required).

[0088] S5: Closed-loop optimization: Actual KPIs: , minute; calculate : (No adjustment required).

[0089] The parameters in this embodiment are designed for peak scenarios that require a balance between real-time performance and media load. NVMe hard drives (low latency) are selected with a load rate of 70% (not overloaded), and tape drives are excluded (latency exceeds constraints). This solves the backup interruption problem caused by media overload in peak scenarios. In this embodiment, the media load rate is stable at 70%, and the recovery time is ≤12 minutes.

[0090] Example 6 like Figure 9 As shown below, an intelligent backup system for a computer system based on a thermal power plant provided by the present invention will be described. The intelligent backup system for a computer system based on a thermal power plant described below can be referred to in correspondence with the intelligent backup method for a computer system based on a thermal power plant described above, and will be further explained in conjunction with Embodiment 1. like Figure 9 The structure shown. Figure 9 This application provides an intelligent backup system for a computer system in a thermal power plant. This embodiment targets the startup (Q1), maintenance (Q2), and peak-period (Q3) scenarios of a thermal power plant, deploying a fully hardware and software-coordinated intelligent backup system. The hardware configuration and deployment details of each module are as follows: The parameter acquisition module consists of three parts: scene sensors, a medium monitoring unit, and a data demand acquisition unit. All three are deployed in the power plant control room cabinet and directly connect to the power plant's core production system. The scene sensors use Siemens S7-1200 PLCs, which connect to the power plant's SIS system via the Profinet Ethernet protocol to collect boiler ignition commands (triggering scene Q1) and real-time unit load signals (triggering scene Q3 when the load is ≥90% of the rated value). Simultaneously, they connect to the MIS system via an RS485 serial interface to read maintenance work order creation information (triggering scene Q2 when the work order status is "in execution"). The data collection interval is dynamically adjusted according to the scene: for scene Q1, due to frequent load fluctuations, the collection interval is set to 0.5 seconds; for scene Q2, with stable load, the collection interval is set to 2 seconds; and for scene Q3, with moderate load fluctuations, the collection interval is set to 1 second.

[0091] The media monitoring unit uses an Advantech USB-4711 data acquisition card, which connects directly to backup media (NVMe hard drives, tape drives) via a SATA interface to collect media status parameters in real time, such as media load rate. The calculation is done by reading the ratio of the used capacity of the media partition to the total capacity (e.g., when the total capacity of the NVMe hard drive is 2TB and the used capacity is 1.8TB). ); Media health The calculation is based on hard drive SMART parameters (bad sector count, seek error rate) and tape wear count, specifically according to the formula. bad track number Tape wear count (e.g., when the number of bad sectors is 0, or the tape wear count is 50 times) (points); medium response delay Test media read / write response time (e.g., average response time of NVMe hard drives) using periodic ping commands. Total media capacity This can be obtained by reading the hardware factory configuration information (such as a single-disk storage capacity of 45TB for a tape drive).

[0092] The data requirement acquisition unit uses a Huawei S5720 industrial switch and connects to the DCS system (ABBAC800M in this example) via Modbus TCP protocol to extract data requirement parameters: data real-time performance. Based on DCS data update frequency (DCS control logic update frequency 100ms, divided into millisecond level t1; daily operation report update frequency 1 hour, divided into minute level t2); data redundancy Set the data association based on the importance of the associated devices (data associated with critical equipment such as boilers and steam turbines is set to dual-media r2, and routine office documents are set to single-media r1); data volume By statistically analyzing the data frame length and the number of data frames per unit time in the DCS system (e.g., in a maintenance scenario, a single vibration curve record is 100KB, and 3600 records are generated in one hour), calculations are performed. ).

[0093] The data requirement profiling module and the media compatibility calculation module share a Dell PowerEdge R750 edge server, deployed in the production control area (Security Zone II). The server is configured with an Intel Xeon Gold 6330 CPU and 64GB DDR4 memory to meet the needs of multi-threaded data processing—the data requirement profiling module needs to calculate data requirement weights in real time. The media compatibility calculation module needs to generate media compatibility scores. The matrix and the two are deployed through independent processes to avoid resource contention.

[0094] The MPC scheduling decision module is deployed on an Advantech ARK-3520 industrial control server (Production Control Area II), configured with an Intel Core i7-12700E CPU, 32GB of memory, and pre-installed with Windows Server 2022 and MATLAB R2023b (including the Optimization Toolbox). It is used to solve the multi-objective scheduling function J and handle scenario-media conflicts. The server also integrates a conflict adjustment unit, which is specifically designed to adjust the logic for conflicts between core data priority and media load in scenario Q1 and conflicts between data volume and media capacity in scenario Q2.

[0095] The closed-loop optimization module and the MPC scheduling decision module share the Advantech ARK-3520 server hardware resources, but independently deploy KPI evaluation algorithms to collect real-time metrics such as backup completion rate and recovery time and calculate optimization coefficients. The parameters of each module are updated dynamically.

[0096] The storage media pool is deployed in the storage zone (Security Zone III) and includes Samsung 990 Pro NVMe hard drives (2TB capacity, read / write speed 7450MB / s, suitable for low latency requirements in Q1), IBM TS2900 tape drives (supports LTO-9 tapes, 45TB capacity per disk, suitable for large capacity requirements in Q2), and Alibaba Cloud OSS cloud storage (elastic capacity expansion, access latency ≤50ms, used for off-site disaster recovery and data offloading in Q2).

[0097] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0098] For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit a program for use in or in conjunction with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. Furthermore, a computer-readable medium can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0099] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for intelligent backup of a computer system based on a thermal power plant, characterized in that, Includes the following steps: S1. Obtain scene characteristic parameters based on power plant system signals. Medium state parameters and data requirement parameters Output parameter collection set ; S11, Scene feature parameters S include scene type Scene duration and scenario priority ; Scene type Including startup Q1, maintenance Q2, and peak period Q3; scenario priority. include class, class, class; S12, Medium State Parameters Including media loading rate Media health and medium response delay ; S13, Data Requirement Parameters Including data real-time requirements Data redundancy requirements and data compliance retention period ; S2, Collect data based on parameters Build a data requirement profile, calculate and output the data requirement weights. ; S3. Weighting based on data requirements Calculate and output the fitness matrix of each media node in the distributed storage node cluster; S4. Based on the fitness matrix, the Model Predictive Control (MPC) algorithm is used to solve the multi-objective scheduling objective function. Determine the optimal media allocation scheme and output the scheduling instruction set. ; Step S4 also includes step S44 for dynamic adjustment of scene media conflicts, the specific steps of S44 are as follows: S441. According to scene type Data demand weight and media loading rate Determine the type of conflict; when , And the medium loading rate If so, it is determined to be a startup conflict; when And the amount of maintenance-related data If so, it is determined to be a maintenance conflict; The formula for calculating the available capacity of the medium is as follows: , This represents the total capacity of the medium. S442. Execute the adjustment strategy according to the conflict type, and obtain the execution result as follows: If a startup conflict is identified, the priority scheduling mechanism is executed: pause. Non-core data backup tasks, according to the formula Adjust the media loading rate to ; in, For non-core data volumes that are paused, This represents the number of tasks that have been paused. If a maintenance conflict is identified, a media expansion and data offloading mechanism is executed: the column values ​​of the adaptability matrix are called first. and Backup media, for the replacement cycle Low-frequency maintenance data according to the formula Perform cloud-based traffic distribution and control ; S443. Based on the execution result of the adjustment strategy, output the adjusted scheduling instruction set. Replace the original scheduling instruction set ; S5. According to the scheduling instruction set or Perform a backup operation and obtain the actual backup completion rate. and actual recovery time Based on the actual backup completion rate and actual recovery time Determine the optimization coefficients According to the optimization coefficient The weighting coefficients of the data requirement weights in step S2 and the objective function in step S4 are feedback update steps. Delay penalty coefficient This forms a closed loop.

2. The intelligent backup method for computer systems based on thermal power plants according to claim 1, characterized in that, The specific steps of step S1 are as follows: Data real-time requirements Including millisecond-level t1, minute-level t2, and offline t3; data redundancy requirements. Including dual-medium r2 and single-medium r1; Parameter acquisition interval Dynamically set according to scene type: Q1 Q2 Q3 .

3. The intelligent backup method for computer systems based on thermal power plants according to claim 2, characterized in that, The specific steps of step S2 are as follows: S21. Based on data requirement parameters The need for real-time data Data redundancy requirements and data compliance retention period Perform normalization processing to determine the normalized value. , and Data requirement parameters The calculation formula for normalization is as follows: Real-time normalized value: ,in, As a real-time benchmark value, Pick hour , hour ; Redundancy normalized value: ; Compliance normalized value: ,in, The year serves as the compliance benchmark. S22. Based on scene feature parameters Set weight coefficients , , and : Q1 , , ; Q2 , , ; Q3 , , ; S23. Based on the normalized value , and and weighting coefficients , , Determine the calculation , The formula is: ,and The value range is [0,1].

4. The intelligent backup method for computer systems based on thermal power plants according to claim 3, characterized in that, The specific steps of step S3 are as follows: S31. Based on the medium state parameters in step S1 Extracting normalized values , and : Normalized load rate: ; Health normalized value: ; Normalized response latency: ,in This is the baseline value for delay; S32. Determine the adaptation coefficient based on the scene feature parameter S: real-time correlation coefficient Redundancy correlation coefficient Compliance correlation coefficient ; S33. Based on the normalized value , and and fit coefficient , and Determine the fitness matrix and calculate the column values ​​of the fitness matrix. The formula is: ； Output the fitness matrix, where each row corresponds to a media node. The value range is [0,1].

5. The intelligent backup method for computer systems based on thermal power plants according to claim 4, characterized in that, In step S4, the multi-objective scheduling objective function The solution process is as follows: S41. Based on the fit matrix and medium response delay Determine the objective function objective function The calculation formula is as follows: ； in, This refers to the delay penalty coefficient. According to the scene type Setting, Q1 Q2 Q3 ; S42. Set constraints, including: Medium loading rate constraint: ; Media health constraints: point; Backup time constraints: And in Q1 Q2 Q3 ; S43. According to the objective function Given the constraints, the MPC algorithm is used to solve the problem. The maximum value is used to determine the optimal media node ID and backup priority, and a scheduling instruction set is output. .

6. The intelligent backup method for computer systems based on thermal power plants according to claim 1, characterized in that, Optimization coefficients in step S5 The calculation and feedback process is as follows: S51. Set KPI target value: Core data backup success rate and recovery time minute; S52. Based on the actual backup completion rate and actual recovery time Sure ,calculate The formula is: ； S53, according to Value feedback update weight coefficient , , and delay penalty coefficient : when Then adjust the weighting coefficients of step S2: , ; when Then adjust the penalty coefficient in step S4. .

7. An intelligent backup device for a computer system in a thermal power plant, used to execute the method of any one of claims 1 to 6, characterized in that, include: The parameter acquisition module is used to collect scene feature parameters. Medium state parameters and data requirement parameters Output parameter collection set ; The data requirement profiling module is used to collect data based on parameters. Calculate data demand weights ; The media compatibility calculation module is used to calculate the weight based on data requirements. Calculate the fitness matrix; The MPC scheduling decision module is used to solve multi-objective functions using the MPC algorithm. It also handles scenario media conflicts and outputs a set of scheduling instructions. or ; The closed-loop optimization module is used to optimize the scheduling instruction set. Perform backup operations and adjust the backup completion rate accordingly. and actual recovery time Determine the optimization coefficients According to the optimization coefficient It also provides feedback to update the weighting and penalty coefficients.

8. The intelligent backup device for a thermal power plant computer system according to claim 7, characterized in that, The parameter acquisition module includes: Scene sensors are used to interface with the power plant's SIS (System-In-Service) and MIS (Management-In-Service) systems to acquire scene feature parameters based on the outputs of the SIS and MIS systems. ; The media monitoring unit incorporates a hard drive SMART chip and a magnetic tape wear sensor. Based on the outputs of the hard drive SMART chip and the magnetic tape wear sensor, it obtains the media load rate. Health Response delay and total capacity ; The data requirement acquisition unit is used to interface with the DCS control system and, based on the output of the DCS control system, obtain data real-time requirements, redundancy, compliance requirements, and data volume. Output .

9. The intelligent backup device for a thermal power plant computer system according to claim 7, characterized in that, The MPC scheduling decision module includes: Objective function construction unit, based on and Build The expression; The constraint configuration unit is used to generate load, health, and backup time constraints based on preset media. MPC solver, used to solve based on objective function Given the constraints, the gradient descent method is used to solve the problem. The maximum value is used to determine the optimal medium allocation scheme; The conflict adjustment unit is used to perform step S44 of claim 6, based on scene feature parameters. and medium state parameters Detect conflicts and implement adjustment strategies to generate an adjusted set of scheduling instructions. .

Images