Calculation power optimization method for green energy driving of data center

Abstract

The invention provides a computing power optimization method for green energy driving of a data center, and the method comprises the steps: obtaining real-time tidal energy supply data and load demand calculation data, analyzing the matching degree of periodic supply of tidal energy and load fluctuation demand calculation, and determining the peak dislocation time period of tidal energy supply and load demand calculation; evaluating an execution window of a high-energy-consumption calculation task of the data center according to a peak dislocation period of tidal energy supply and calculation load demand, and obtaining an available resource list in an energy sufficient period; if the migrated computing power resource allocation balance degree does not reach a preset balance threshold value, obtaining a current load of a server cluster of the data center, and determining an energy valley period activation condition; and analyzing the mitigation degree of resource mismatching from the available capacity index of the buffer energy, reallocating idle computing power resources according to the mitigation degree, and determining the supplement demand of the tidal energy shortage time period.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for optimizing computing power in data centers driven by green energy. Background Technology

[0002] Ocean tidal energy, as a clean and renewable energy source, has garnered significant attention in the energy sector due to its stability and predictability, particularly its potential in promoting the construction of green data centers. Tidal energy is converted into electricity through tidal movements, providing sustainable power to data centers and directly impacting the stable supply of computing resources. However, the mismatch between the periodic supply of tidal energy and the dynamic computing demands of data centers can lead to insufficient or excessive computing power allocation, affecting operational efficiency. This mismatch not only reduces energy utilization efficiency but may also increase data center operating costs and stability risks due to wasted or scarce computing resources. Existing methods for addressing the mismatch between tidal energy and data center demand primarily rely on static energy scheduling strategies or simple backup power switching. These methods typically cannot dynamically adapt to the periodic fluctuations in tidal energy supply or flexibly adjust according to the real-time demands of computing tasks. For example, some schemes allocate energy through fixed time periods, ignoring the changing computing demands of data centers at different times, resulting in low energy utilization efficiency. Furthermore, the use of backup power often relies on traditional energy sources, weakening the green and environmentally friendly advantages of tidal energy and failing to fundamentally solve the dynamic matching problem between energy and computing power. Tidal power generation is influenced by the tidal cycle, exhibiting fixed peaks and troughs daily. However, the computing demands of data centers fluctuate due to factors such as user access and task type. For example, in video streaming scenarios, high user activity, surges in traffic, and frequent content updates can cause evening peak user access periods to overlap with tidal power generation troughs. This is because, in some geographical locations, tidal power generation relies on the potential energy difference created by the rise and fall of seawater. Since the tidal cycle is not perfectly synchronized with the diurnal cycle, nighttime tides may be at low tide or slack tide, resulting in the evening typically corresponding to the lowest tide level or smallest tidal range. Consequently, tidal power generation drops significantly, failing to meet peak demand and resulting in insufficient energy to support high-load tasks. Conversely, during peak tidal periods, there may be fewer computing tasks, leading to energy waste. This resource matching problem caused by time misalignment stems from the lack of a mechanism capable of dynamically adjusting the time distribution of computing tasks and server operating status. Therefore, how to dynamically adjust the allocation of computing tasks and the scale of server operation in data centers based on the periodic supply characteristics of tidal energy, so as to achieve real-time matching between energy supply and computing power demand, has become a key issue for data centers to utilize tidal energy. Summary of the Invention

[0003] This invention provides a green energy-driven computing power optimization method for data centers, mainly comprising: The process involves analyzing the matching degree between tidal energy supply data and computing load demand data to determine the peak misalignment periods between tidal energy supply and computing load demand. Based on these peak misalignment periods, the execution window for high-energy-consuming computing tasks is evaluated, generating a list of available resources during periods of sufficient energy. The priority sequence of high-energy-consuming tasks is extracted from this list, and tasks matching the tidal energy supply periods are dynamically migrated to the corresponding periods for execution, determining the balance of computing power resource allocation after migration. Based on the balance of computing power resource allocation and the current load of the server cluster, the server cluster size is adjusted to the target number of active nodes, determining the activation conditions for off-peak periods. Based on the activation conditions for off-peak periods, a buffer energy availability capacity index is generated. Based on the buffer energy availability capacity index, idle computing power resources are reallocated to determine the supplementary demand during periods of insufficient tidal energy. Finally, the task execution efficiency and energy consumption data in the dynamically migrated task execution logs are analyzed, and combined with the supplementary demand during periods of insufficient tidal energy, a computing power optimization scheme is generated.

[0004] Furthermore, the analysis of the matching degree between tidal energy supply data and computational load demand data, and the determination of peak misalignment periods between tidal energy supply and computational load demand, includes: Data on tidal height, current velocity, and power generation from tidal power plants are collected, along with data on CPU utilization, memory usage, and network bandwidth consumption from data center servers. The tidal power plant data is decomposed into a time series to extract periodic characteristic curves. Based on these curves, a sliding window method is used to process and calculate load demand data, determining the load mean and variance, and identifying peak load times. A supply-demand time series deviation matrix is ​​generated by comparing the peak tidal energy supply times with the peak load times. Periods with time offsets exceeding a threshold are extracted from the deviation matrix and marked as misaligned periods. A misaligned period record table is generated, containing the misalignment start time, duration, and energy difference.

[0005] Furthermore, the step of evaluating the execution window of high-energy-consuming computing tasks based on the peak-off period and generating a list of available resources during periods of sufficient energy includes: Based on the peak-off time periods, identify the power requirements and runtime of video transcoding, batch data processing, and machine learning training tasks; compare tidal energy supply data with power requirements to determine the time interval during which energy supply exceeds task requirements as the execution window; calculate the net power value after subtracting infrastructure power consumption from tidal power generation within the execution window to determine the maximum number of supported nodes and total storage capacity; based on the maximum number of nodes, query the network device configuration table, accumulate the node network card speed as the bandwidth requirement value, adjust the bandwidth quota, and generate an available resource list containing the number of computing nodes, storage capacity, and bandwidth allocation.

[0006] Furthermore, the step of extracting the priority sequence of high-energy-consuming tasks from the available resource list, dynamically migrating tasks matching the tidal energy supply period to the corresponding period for execution, and determining the balance of computing power resource allocation after migration includes: The number of computing nodes, storage capacity, and bandwidth allocation data are read from the available resource list. A priority sequence of high-energy-consuming tasks is generated by combining the difference between the task deadline and the current time. The execution duration of tasks in the priority sequence is compared with the tidal energy supply period, matching tasks are marked, and a task migration candidate set is constructed. Tasks in the task migration candidate set are migrated to computing nodes in the corresponding time period, and the CPU utilization, memory usage, and bandwidth occupancy rate are recorded for each time period after migration. The variance of the CPU utilization rate is used to evaluate the balance of computing resource allocation.

[0007] Furthermore, after generating the list of available resources during periods of sufficient energy, the process includes: Extract the execution requirement parameters of tasks in the task migration candidate set, including minimum runtime, maximum allowable delay, resource consumption, and interruptibility; calculate the overlap ratio between the time window of the tasks in the task migration candidate set and the tidal energy supply period to determine the fit; adjust the task scheduling plan according to the fit, and migrate the task execution time to the start time of the tidal energy supply period; set the task end time according to the minimum runtime, and generate the migrated and adjusted task execution schedule.

[0008] Furthermore, the step of adjusting the server cluster size to the target number of active nodes based on the computing power resource allocation balance and the current load of the server cluster, and determining the activation conditions during off-peak periods, includes: Based on the balance of computing resources allocation, obtain the CPU utilization, memory usage, and task queue length of the server cluster, and calculate the current load value; if the current load value exceeds the threshold, sort the node load values ​​and execute a hibernation operation until the target number of active nodes is reached; based on the target number of active nodes, set the minimum load threshold, node operating efficiency, and energy supply stability, and generate off-peak activation conditions.

[0009] Furthermore, the buffer energy availability capacity index generated based on the low-peak activation condition includes: Based on the off-peak activation conditions, the difference between tidal power generation and data center power consumption is detected, and excess energy is allocated to electrochemical or mechanical energy storage units; a charging or mechanical energy storage process is executed, and the energy storage growth rate and conversion efficiency are recorded; when the tidal energy supply data is lower than the power consumption, the energy storage units are controlled to release energy, and the remaining capacity, conversion efficiency and release rate are statistically analyzed to generate a buffer energy available capacity index.

[0010] Furthermore, the generated buffer energy available capacity index includes: The tidal energy supply data is compared with the data center power consumption to mark the energy gap period and power difference; based on the power difference, the energy storage unit outputs electrical energy and the release rate is adjusted to match the gap; the remaining power of the energy storage unit, the ratio of output power to consumed stored energy, and the instantaneous maximum output power are collected; the average conversion efficiency and peak release rate within the release cycle are statistically analyzed to generate a buffer energy available capacity index.

[0011] Furthermore, the step of reallocating idle computing resources based on the available buffer energy capacity index to determine the replenishment demand during periods of tidal energy shortage includes: Extract energy storage capacity and conversion efficiency from the available capacity index of the buffer energy, and calculate the actual available energy; compare the actual available energy with the required energy to determine the resource mismatch mitigation level; allocate idle computing power to low priority tasks according to the resource mismatch mitigation level; and generate supplementary requirements including computing power replenishment amount, energy supply priority, and task scheduling latency by statistically analyzing the number of additional startup nodes, task urgency, and average waiting time.

[0012] Furthermore, the analysis of task execution efficiency and energy consumption data in the dynamically migrated task execution logs, combined with the replenishment needs during periods of insufficient tidal energy, generates a computing power optimization scheme, including: Extract task identifiers, execution durations, and resource utilization rates from the task execution logs to generate a set of task execution efficiency indicators; calculate unit task energy consumption based on the set of task execution efficiency indicators; schedule high-energy-consuming tasks to energy-sufficient periods based on the replenishment needs during periods of tidal energy shortage, and formulate power quotas and energy storage release plans; determine the number of nodes to activate or hibernate based on the task load, and generate a computing power optimization scheme.

[0013] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a green energy-driven computing power optimization method for data centers. Addressing the resource mismatch problem caused by insufficient matching between the periodic supply of tidal energy and the fluctuating demand of computing loads, the method analyzes the peak and off-peak periods of tidal energy supply and load demand, dynamically adjusts the execution window of high-energy-consuming tasks, and migrates tasks to periods with sufficient energy. It also optimizes the balance of computing power allocation by combining an available resource list. During periods of low energy, this invention stores excess energy through electrochemical or mechanical energy storage and dynamically releases it to fill the gap. Simultaneously, it monitors the energy storage capacity and conversion efficiency in real time to alleviate resource mismatch, ultimately generating a green energy-driven computing power optimization scheme that includes task scheduling, energy allocation, and node adjustment. This invention significantly improves the energy utilization efficiency and the rationality of computing resource allocation in data centers, achieving green and sustainable operation. Attached Figure Description

[0014] Figure 1 This is a flowchart of a data center green energy-driven computing power optimization method according to the present invention.

[0015] Figure 2 This is a schematic diagram of step S101 in a data center green energy-driven computing power optimization method of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figures 1-2 This embodiment of a data center green energy-driven computing power optimization method may specifically include: Step S101: Obtain real-time tidal energy supply data and computational load demand data, analyze the matching degree between the periodic supply of tidal energy and the fluctuating demand of computational load, and determine the peak misalignment period between tidal energy supply and computational load demand.

[0018] Data on tidal height, flow velocity, and power generation are collected using a sensor array deployed at the tidal power station. Simultaneously, server CPU utilization, memory usage, and network bandwidth consumption records are obtained from the data center management platform. The collected tidal energy data is decomposed into a time series, extracting the main periodic components for 24-hour and 12.42-hour periods to obtain the periodic characteristic curve of tidal energy supply. Based on this periodic characteristic curve, a sliding window method is used to segment the calculated load data, calculating the load mean and variance for each time period. The peak time and duration of load fluctuations are identified. By calculating the time difference between the peak time of tidal energy supply and the peak time of calculated load, the time offset between the two is determined, resulting in a supply-demand time series deviation matrix. The time periods with time offsets exceeding a preset offset threshold are extracted from the supply-demand time series deviation matrix. The preset offset threshold is one-quarter of the tidal cycle length. These periods are marked as misaligned periods. The energy gap and energy surplus values ​​within each misaligned period are calculated. The severity of the misalignment is determined based on the ratio of the gap or surplus value to the average energy supply for that period, resulting in a misaligned period record table containing the misalignment start time, duration, and energy difference. If a misaligned period with a duration exceeding a preset duration threshold exists in the misaligned period record table, K-means clustering is performed on the historical load data within that period. Based on the clustering results, periodically recurring high-load task types and their occurrence patterns are identified. Combined with the energy difference data in the misaligned period record table, the peak misalignment periods between tidal energy supply and calculated load demand are determined.

[0019] Specifically, in one implementation, the sensor array of the tidal power station adopts a layered deployment architecture, with pressure sensors, flow velocity sensors, and power monitoring sensors respectively installed at the seawater inlet, turbine unit, and power output end. The pressure sensor acquires tidal height data by measuring changes in seawater pressure, sampling once per minute to record real-time changes in tidal level. The flow velocity sensor is installed at the front end of the turbine blades to monitor the speed of seawater flowing through the turbine and calculates the kinetic energy of the water flow based on Bernoulli's principle. The power monitoring sensor is directly connected to the generator output end to collect power generation data in real time. The time series decomposition uses the Fourier transform method to perform frequency domain analysis on tidal energy data for 30 consecutive days, extracting the dominant frequency components. The 24-hour period corresponds to the diurnal tidal variation caused by the Earth's rotation, and the 12.42-hour period corresponds to the semi-diurnal tidal variation caused by the Moon's gravity; these two periodic components constitute the main characteristics of tidal energy supply.

[0020] Specifically, the data center management platform collects real-time resource usage data for each server node through deployed monitoring agents. CPU utilization is obtained by reading processor performance counters, memory usage is obtained by querying the operating system's memory management interface, and network bandwidth consumption is recorded by the network interface card (NIC) traffic statistics module. This data is sampled at 5-minute intervals to form a continuous time series of computing load.

[0021] For example, in the implementation of the sliding window method, the window width is set to 1 hour and the sliding step is 15 minutes. For the load data within each window, the arithmetic mean is calculated as the load average for that period, and the standard deviation is calculated as a measure of load fluctuation. When the load value of a certain period exceeds 1.5 times the average of the preceding and following periods, it is identified as the load peak moment. The duration of the peak is determined by recording the start and end times of the peak. The calculation of the time offset involves performing a difference operation between the timestamp of the tidal energy supply peak and the timestamp of the calculated load peak to form a two-dimensional matrix, where the rows of the matrix represent different dates and the columns represent different time periods within a day.

[0022] In one possible implementation, the preset deviation threshold is determined based on tidal cycle characteristics. Since the semi-diurnal tidal cycle is 12.42 hours, and one-quarter of it is approximately 3.1 hours, this means that when the time offset between energy supply and load demand exceeds 3.1 hours, they are essentially out of phase, resulting in the worst matching degree. The severity of the misalignment is quantified using a ratio method, calculated as follows: the degree of misalignment equals the energy deficit or surplus value divided by the average energy supply during that period. A ratio exceeding 0.3 indicates severe misalignment; a ratio between 0.1 and 0.3 indicates moderate misalignment; and a ratio less than 0.1 indicates mild misalignment.

[0023] It's important to note that the specific application of K-means clustering in load pattern recognition involves constructing feature vectors from historical load data according to the time and load intensity dimensions. Each feature vector includes attributes such as load peak value, duration, and peak occurrence time. The number of clusters K is determined using the elbow rule, typically set to 3 to 5 clusters, corresponding to different types of computational loads such as video transcoding tasks, data backup tasks, and user interaction tasks. Cluster centers represent the typical load pattern of each task type; the task type is determined by calculating the Euclidean distance between a new load and each cluster center.

[0024] Preferably, when identifying high-load task types, special attention should be paid to specific patterns in video streaming processing scenarios. During peak user viewing hours from 8 PM to 11 PM, video transcoding and content distribution tasks surge, with server CPU utilization reaching over 85% and memory usage exceeding 70%. However, tidal energy is in its low-tide period at this time, with power generation only around 30% of peak capacity. This typical misalignment scenario requires a collaborative solution involving task scheduling and energy storage systems.

[0025] In one embodiment, the data structure of the misalignment period record table includes multiple fields. The misalignment start time is accurate to the minute, recording the specific moment the misalignment begins. The duration is in hours, representing the time span during which the misalignment state persists. The energy difference is divided into two categories: deficit value and surplus value. The deficit value represents the amount of electricity demanded exceeds the energy supply, and the surplus value represents the amount of electricity supplyed exceeds the load demand; both are measured in kilowatt-hours. This structured record facilitates subsequent task scheduling optimization and energy storage system allocation.

[0026] For example, when a continuous misalignment period is detected to exceed a preset duration threshold of 4 hours, a deep analysis mechanism is triggered. This not only analyzes the current load pattern but also traces back load data from the same period over the past 7 days to identify periodic patterns. If a certain type of high-load task is found to recur at a fixed time each day and has a stable phase difference with the tidal energy supply cycle, this period is marked as a critical misalignment period requiring focused energy scheduling optimization.

[0027] Understandably, the final determination of peak-hour shift periods took into account multiple dimensions of energy supply and demand. This included not only the matching degree of energy quantity, but also factors such as the stability of energy quality, the urgency of task execution, and the economics of system operation. By establishing a multi-dimensional evaluation system, it was ensured that the identified shift periods truly impacted the data center's operational efficiency, providing accurate decision-making basis for subsequent optimization scheduling.

[0028] Step S102: Based on the peak misalignment period between tidal energy supply and computing load demand, assess the execution window of high-energy-consuming computing tasks in the data center to obtain a list of available resources during periods of sufficient energy.

[0029] Based on the start time and duration of peak-shifting periods, the execution characteristics of video transcoding, batch data processing, and machine learning training tasks within the data center are identified. The power demand, runtime, and resource utilization of each type of high-energy-consuming task are extracted. By comparing the power generation value of the tidal energy supply curve during the same period, the time interval where energy supply exceeds the task's power demand is determined as the task execution window. The net power available for computing tasks is obtained by subtracting the data center infrastructure power consumption from the hourly tidal power generation within the task execution window. Based on the power consumption record of a single computing node running at full load, the maximum number of nodes supported by the net power value is calculated. Simultaneously, the memory size and hard disk capacity of each node are counted, and the total storage capacity for that period is accumulated. Based on the maximum number of nodes and the total storage capacity, the network card speed of each computing node is queried from the data center network equipment configuration table. The sum of the network card speeds of all nodes is used as the bandwidth demand value. If the bandwidth demand value exceeds the core switch port bandwidth limit, the bandwidth quota of each node is reallocated according to task priority, generating a list of available resources during energy-sufficient periods that includes the number of computing nodes, storage capacity, and bandwidth allocation.

[0030] Specifically, in one implementation, the identification process for high-energy-consuming computing tasks is based on classification using historical power consumption records. Video transcoding tasks, when performing H.265 encoding, require 200 to 300 watts of power per task, with a typical runtime of 2 to 4 hours. Batch data processing tasks involve parsing and statistically analyzing massive amounts of log files, requiring 150 to 200 watts of power, with runtime fluctuating between 1 and 3 hours depending on the data volume. Machine learning training tasks, when training deep neural networks, require 500 to 800 watts of power from GPU clusters, with training cycles ranging from several hours to several days.

[0031] Specifically, calculating net power requires an accurate assessment of infrastructure power consumption. Data center infrastructure power consumption includes cooling system power consumption, uninterruptible power supply (UPS) losses, lighting power consumption, and basic network equipment power consumption. Cooling systems dynamically adjust based on server heat dissipation and typically account for 30% to 40% of total power consumption. UPS systems incur approximately 10% energy loss during charge-discharge transitions. These basic power consumptions are subtracted from the total tidal power output; the remaining amount is the net power available for computing tasks.

[0032] It should be noted that the power consumption of a single computing node is recorded in real time through intelligent power distribution unit monitoring. The power consumption of each node varies significantly under idle, half-load, and full-load conditions: approximately 100 watts under idle, 180 watts under half-load, and reaching 250 watts under full load. The calculation of the maximum number of nodes uses the full-load power consumption value as a benchmark to ensure sufficient energy supply even under extreme load conditions.

[0033] Preferably, bandwidth allocation employs a tiered management mechanism. The core switch has multiple 10 Gigabit ports with a total bandwidth capacity of 160 Gbps. When the cumulative bandwidth demand exceeds this limit, it is reallocated according to task priority. Real-time video streaming tasks receive the highest priority, allocating 40% of the bandwidth resources; batch data transmission tasks are allocated the next highest priority, receiving 35% of the bandwidth; the remaining tasks share the remaining 25% of the bandwidth. This dynamic allocation mechanism ensures that network transmission for critical tasks is not affected.

[0034] In one embodiment, the list of available resources during periods of sufficient energy is presented in the form of a structured table, with each period corresponding to a row of records, including four fields: start and end time of the period, number of available computing nodes, total storage capacity, and allocated bandwidth.

[0035] Step S103: Extract the priority sequence of high-energy-consuming tasks from the available resource list. If there are tasks in the priority sequence that match the tidal energy supply period, dynamically migrate the tasks that match the tidal energy supply period to the corresponding period for execution, and determine the balance of computing power resource allocation after migration based on the available resource list.

[0036] The system reads the number of compute nodes, storage capacity, and bandwidth allocation data for each time period from the available resource list. Simultaneously, it acquires the difference between the deadline and the current time for video transcoding, data backup, and machine learning training tasks, assigning this value as an urgency index. These urgency indexes are then sorted in descending order based on their product with the task power consumption value, resulting in a priority sequence for high-energy-consuming tasks. The estimated execution time of each task in this priority sequence is compared to the duration of the tidal energy supply period. If the task's execution time is less than a preset threshold proportion of the supply period's duration, it is marked as a matchable task. The original execution time and resource requirements of all matchable tasks are extracted to construct a task migration candidate set. For each task in the task migration candidate set, its current running status and memory image data are read. The task execution environment is migrated to the compute node corresponding to the tidal energy-sufficient period using a container orchestration tool. After the migration operation, the CPU utilization, memory usage, and network bandwidth utilization values ​​for each time period are recorded, resulting in a resource utilization record for the migrated time period. The variance of CPU utilization in all time periods in the resource utilization record is used to assess the degree of difference in the allocation of computing resources in the time dimension. If the variance exceeds the preset balance threshold, the percentage difference between the time period with the highest and lowest resource utilization is calculated, and the balance of computing resource allocation after migration is judged based on the percentage difference.

[0037] Specifically, in one implementation, the construction of the high-energy-consuming task priority sequence adopts a two-dimensional evaluation mechanism. The urgency index is obtained by calculating the difference between the task deadline and the current system time; the smaller the difference, the more urgent the task. Optionally, the urgency index can be specifically set using a segmented scoring method based on the task type and remaining time.

[0038] For example, for live streaming tasks in video transcoding, if the remaining time is less than 2 hours, it is considered extremely urgent, with an urgency index of 100; if the remaining time is between 2 and 12 hours, it is considered relatively urgent, with an index of 60; and if the remaining time exceeds 12 hours, it is considered not urgent, with an index of 20. For on-demand video transcoding tasks, if the remaining time is less than 6 hours, it is considered relatively urgent, with an urgency index of 40; if the remaining time is between 6 and 24 hours, it is considered normal, with an index of 10; and if the remaining time exceeds 24 hours, the time is relatively flexible, with an index of 5. For data backup tasks, if the remaining time is less than 1 hour, it is considered urgent, with an index of 80; if the remaining time is between 1 and 8 hours, it is considered moderately urgent, with an index of 30; and if the remaining time exceeds 8 hours, it is considered low priority, with an index of 10. For machine learning training tasks, which typically involve high power consumption and long computation cycles, tasks with less than one hour remaining should be completed as quickly as possible, with an urgency value of 90. Tasks with one to four hours remaining require more attention to execution timing, with an urgency value of 50. Tasks exceeding four hours can be appropriately delayed, with an urgency value of 20. For video transcoding tasks, if a live streaming platform's transcoding task has only two hours remaining before its broadcast time, its urgency value is set to 100. For offline transcoding tasks of on-demand videos with a deadline of 24 hours, the urgency value is set to 10. Task power consumption values ​​are extracted from historical execution records. The average power consumption for video transcoding tasks is 250 watts, for data backup tasks it's 180 watts, and for machine learning training tasks it's 650 watts. The urgency values ​​are multiplied by the power consumption values ​​and then sorted in descending order to ensure that both urgent and energy-intensive tasks receive higher priority.

[0039] Specifically, task execution time prediction is based on statistical analysis of historical execution data. The system records the execution time of each type of task under different data scales and constructs a task duration prediction table. A 4K video transcoding task takes 30 minutes to process one hour of footage, batch data backup takes 15 minutes per 100GB, and deep learning model training takes 45 minutes per epoch. The duration of tidal energy supply periods is extracted from the tidal cycle curve, typically with high tide lasting 3 to 4 hours. A preset ratio threshold is set to 0.8, meaning that tasks whose execution time does not exceed 80% of the supply period are marked as matchable tasks, reserving a 20% buffer time to handle execution delays.

[0040] For example, the task migration process of a container orchestration tool involves multiple technical steps. A Kubernetes orchestration platform is used, with dynamic task migration achieved through a Pod scheduler. Before migration, the system creates checkpoints for the tasks, saving the current execution state, memory data, and intermediate results. Checkpoint data is synchronized to the target node via a distributed storage system to ensure data consistency during migration. The migration operation is completed by modifying the node affinity tag of the Pod, and the scheduler automatically migrates the Pod from the source node to the target node. The downtime of the entire migration process is controlled within seconds, which has almost no impact on overall execution efficiency for interruptible tasks such as video transcoding.

[0041] It should be noted that the resource utilization data collected for each time period covers three dimensions of monitoring metrics. CPU utilization is obtained by reading the usage time slice of each core from the / proc / stat file and calculating the ratio of busy time to total time. Memory utilization is obtained by reading used memory and total memory from the / proc / meminfo file and calculating the percentage of memory used. Network bandwidth utilization is obtained by obtaining the real-time transmission rate through the network card driver interface and comparing it with the maximum speed of the network card. These metrics are collected every 5 minutes to form continuous time-series data. Each time period contains 12 sampling points, and the average value is taken as the resource utilization rate for that time period.

[0042] In one possible implementation, variance is used to quantify the degree of uneven resource allocation. CPU utilization data for all time periods is collected, and the average is calculated as a baseline. The squared differences between the utilization rate of each time period and the average are summed, and then divided by the total number of time periods to obtain the variance. The larger the variance, the more uneven the distribution of resources over time. A preset equalization threshold is set based on data center operational experience, typically 0.15. When the variance exceeds this threshold, it indicates the existence of significant resource utilization peaks and troughs, requiring further optimization.

[0043] Preferably, calculating the percentage difference provides a more intuitive evaluation of resource balance. The system identifies the period with the highest resource utilization, where CPU utilization may reach 90%; simultaneously, it identifies the period with the lowest utilization, where utilization may be as low as 20%. The difference between the two is 70%, and this percentage directly reflects the extreme degree of resource allocation. Based on the percentage difference, the balance is divided into three levels: a difference of less than 30% is considered balanced, 30% to 60% is considered slightly unbalanced, and more than 60% is considered severely unbalanced. For example, during the peak user access period from 8 PM to 11 PM, when tidal energy is at its lowest, a task migration mechanism can be used to migrate batch data processing and offline analysis tasks originally scheduled for the evening to the peak tidal energy period to balance resource allocation.

[0044] In one embodiment, the construction of the task migration candidate set considers multiple constraints. In addition to execution duration matching, task interruptibility, data dependencies, and service level agreement (SLA) requirements are also evaluated. Tasks with high real-time requirements are excluded from the migration candidates to ensure that service quality is not affected. For task chains with data dependencies, the feasibility of migration needs to be assessed holistically to avoid disrupting the dependencies between tasks.

[0045] Understandably, judging the balance of computing resource allocation not only focuses on resource utilization at a single moment, but more importantly, assesses the rationality of resource allocation throughout the entire operating cycle. Dynamic task migration allows for a better match between the supply curve of tidal energy and the demand curve of computing load, reducing energy waste and resource idleness, and improving the overall energy efficiency and operational stability of the data center.

[0046] Identify the execution requirements and time windows of tasks that match the tidal energy supply period, match the original start time, end time and interruptibility of the tasks with the tidal energy supply period to obtain the task-energy supply period fit, adjust the task scheduling plan according to the task-energy supply period fit, and migrate the task execution time from the original time period to the peak time period of tidal energy supply to obtain the migrated and adjusted task execution schedule, and determine the start and end time of each matched task in the new time period.

[0047] Tasks matching the tidal energy supply period are identified, and execution requirement parameters for each task are extracted, including minimum runtime, maximum allowable delay, and resource consumption. The original start and end times of the tasks are also obtained. The interruptibility attribute is determined based on the task type: video transcoding tasks are marked as interruptible, while real-time data processing tasks are marked as non-interruptable, resulting in a task execution requirement feature set. Using the original start time and minimum runtime in the task execution requirement feature set, the start and end points of the task time window are determined. The overlap duration between this window and the peak tidal energy period is calculated, and the overlap duration is divided by the task's minimum runtime to obtain the time overlap ratio. Simultaneously, it is determined whether the task's end time is later than the peak energy period's end time. Combining these two indicators yields the task's fit with the energy supply period. Based on the fit degree, if the value exceeds a preset threshold, the task's scheduling plan is adjusted. The original execution record is deleted from the task scheduling table, the task's start time is modified to the start time of the peak tidal energy period, and the expected end time of the task in the new period is set based on the minimum runtime in the task execution requirement feature set, resulting in the adjusted task time arrangement. Using the adjusted task time schedule, the specific start time of each matching task during the peak period of tidal energy is determined. If the expected end time exceeds the range of the peak period, the task execution time is shortened by increasing the number of parallel processing nodes, the end time is recalculated, and the new start time and end time of each task are recorded to obtain the adjusted task execution schedule.

[0048] Specifically, in one implementation, the task execution requirement parameters are extracted based on historical runtime data and task configuration files. The minimum runtime is calculated from the number of basic processing units and the unit processing speed of the task. A video transcoding task requires a fixed encoding / decoding time to process one group of gigabytes (GOPs), and the cumulative processing time for all GOPs is the minimum runtime. The maximum allowable latency is determined according to the service level agreement (SLA). The maximum allowable latency for live transcoding tasks is typically 30 minutes, for on-demand video it can be delayed by 24 hours, and for batch data backup it can be delayed by 48 hours. Resource consumption includes three dimensions: CPU core count, memory capacity, and storage space. 4K video transcoding requires 8 CPU cores, 16GB of memory, and 100GB of temporary storage space.

[0049] Specifically, the determination of task interruptibility follows strict classification criteria. Video transcoding tasks can be safely interrupted based on keyframe positions; after an interruption, processing can resume from the most recent keyframe without affecting the final output quality. Batch data processing tasks support chunked processing; each chunk completes independently, saving intermediate results, and after an interruption, unfinished chunks continue execution. Machine learning training tasks save model parameters at the end of each training cycle and can be interrupted and resumed at any cycle boundary. Real-time data stream processing tasks cannot be interrupted because it would cause data loss and accumulated processing latency. Transaction processing tasks have transactional requirements and must be executed completely or rolled back entirely; they are also marked as uninterruptible.

[0050] For example, the calculation of the time overlap ratio involves the intersection operation of time intervals. The task time window is determined by the original start time and the minimum running time, forming a time interval. The peak period of tidal energy is extracted from the tidal power generation curve, typically occurring in the middle of high and low tides, lasting 3 to 4 hours. The intersection of two time intervals is determined by comparing the start and end times; the start point of the intersection is the larger of the two interval start points, and the end point is the smaller of the two interval end points. If the end point is greater than the start point, a valid intersection exists, and the intersection duration is equal to the difference between the end point and the start point. The time overlap ratio is equal to the intersection duration divided by the minimum task running time; a higher ratio indicates a more abundant energy supply during task execution.

[0051] It should be noted that the comprehensive assessment of suitability considers not only the proportion of time overlap but also the degree to which deadline constraints are met. If the task deadline is earlier than the end of the energy peak, it means the task can be completed during a period of sufficient energy, and the suitability weighting coefficient is set to 1.2. If the deadline is later than the end of the peak but earlier than the start of the next peak, the weighting coefficient is 1.0. If the deadline is very tight, even if there is overlap with the energy peak, it is not suitable for relocation, and the weighting coefficient drops to 0.5. Suitability equals the proportion of time overlap multiplied by the weighting coefficient, comprehensively reflecting the feasibility and benefits of task relocation.

[0052] Preferably, the adjustment of the task scheduling plan adopts a transactional update mechanism to ensure the consistency of the scheduling table. The system first locks the scheduling records of the tasks to be adjusted to prevent concurrent modifications. When deleting the original execution record from the scheduling table, the computing resources reserved for the original time period are released simultaneously, including CPU quotas, memory allocations, and network bandwidth. The task execution start time is changed to the start time of the tidal energy peak period, and corresponding resources are reserved for the new time period. After the scheduling table is updated, a scheduling change notification is sent to the task executor to ensure that the task is executed according to the new time plan.

[0053] In one possible implementation, when the expected completion time exceeds the peak time range, the system compresses the execution time by adding parallel processing nodes. Parallelization relies on the decomposability of the task; video transcoding tasks can be divided by scene, with each scene assigned to different nodes for parallel processing. Data backup tasks are divided by file directories, with multiple nodes simultaneously backing up different directories. Machine learning training employs a data-parallel strategy, splitting the training dataset and distributing it across multiple GPU nodes. Each node independently calculates gradients, and the results are then aggregated to update the model parameters. The upper limit of parallelism is constrained by the parallelization efficiency of the task itself, typically set to 2 to 4 times the original number of nodes.

[0054] In one embodiment, the migrated and adjusted task execution schedule uses a Gantt chart to record the time arrangement of each task. The horizontal axis represents time, and the vertical axis represents different computing nodes. Each task is represented by a rectangle, with the start point corresponding to the start time and the end point corresponding to the end time. The height of the rectangle represents the number of nodes occupied. Through visualization, operations and maintenance personnel can intuitively see the task distribution and promptly detect resource conflicts and scheduling anomalies.

[0055] Step S104: If the distribution balance of computing power resources after migration does not reach the preset balance threshold, obtain the current load of the server cluster in the data center and determine the activation conditions for the low energy period.

[0056] If the balanced distribution of computing resources after migration does not reach the preset balance threshold, the CPU utilization, memory usage, and task queue length of all nodes in the cluster are obtained from the server management platform. The CPU utilization is multiplied by a preset weighting coefficient, the memory usage is multiplied by the corresponding weighting coefficient, and the normalized value of the task queue length is added to calculate the current load value of the server cluster. Based on the current load value of the server cluster, if it exceeds the preset load threshold, the number of nodes with load values ​​lower than the preset low load standard is counted. A node list is generated by sorting the load values ​​from low to high. Starting from the first node in the list, nodes are selected to perform a hibernation operation until the number of remaining active nodes reaches the target value. The list of hibernated node numbers is recorded. Based on the list of hibernated node numbers and the target number of active nodes, the node activation conditions during the energy off-peak period are set. The minimum load threshold is set as a preset lower limit of the average load value of the cluster. The node operating efficiency is required to reach a preset efficiency threshold or higher. The fluctuation range of tidal energy supply power is less than a preset stability threshold and the duration exceeds a preset monitoring duration. The energy off-peak period activation conditions, including the minimum load threshold, node operating efficiency, and energy supply stability, are determined.

[0057] Specifically, in one implementation, the current load value of the server cluster is calculated using a three-dimensional weighted evaluation method. CPU utilization is weighted at 0.5, as processor resources are the core element determining computing power. Memory usage is weighted at 0.3, reflecting the need for data caching and processing space. The task queue length needs to be normalized by dividing the queue length by the maximum queue capacity; its weight is set at 0.2. The weighted sum of the three dimensions forms the overall load value, ranging from 0 to 1.

[0058] Specifically, the node hibernation operation follows a gradual shutdown process. First, a preparation signal is sent to the node to be hibernated. Upon receiving the signal, the node stops accepting new tasks and completes the currently executing task. After the task is completed, the node synchronizes critical data in memory to shared storage to ensure no data loss. The node then enters hibernation mode, shutting down the CPU clock speed, reducing the memory refresh rate, and stopping network card data transmission, retaining only the wake-up listening function. The entire hibernation process typically takes less than 30 seconds.

[0059] It should be noted that the low load standard is set based on statistical analysis of historical operating data. Load values ​​for each node at different times over the past 30 days are collected, and the 25th percentile of the load distribution is calculated as the low load standard. When a node's load value is lower than this standard, it indicates that the node is in a light-load state and is a preferred candidate for hibernation.

[0060] For example, the three dimensions of the activation conditions during energy off-peak periods are interrelated. The minimum load threshold ensures that dormant nodes are activated only when the overall cluster load drops to a low level, avoiding frequent dormant-to-wake switching. Node operating efficiency requires evaluating the resource utilization efficiency of activated nodes; when the efficiency value reaches or exceeds a preset threshold, it indicates that existing nodes can efficiently handle the current task without needing to wake up more nodes. Energy supply stability is judged by monitoring fluctuations in tidal power generation; if the power fluctuation amplitude remains within a preset range during continuous monitoring, it indicates that the energy supply has entered a stable period, suitable for adjusting the node operating scale.

[0061] Preferably, in practical applications, data centers reduce energy consumption during tidal energy low periods through a node hibernation mechanism. When the tide is low at 2 AM, power generation drops to 40% of its peak, and user access is also relatively low, with cluster load values ​​falling below 0.3. 30% of the nodes are selected for hibernation based on load ranking, reducing the number of active nodes from 100 to 70.

[0062] Step S105: Based on the determined off-peak activation conditions, excess tidal energy is converted into electrochemical or mechanical energy storage and preferentially allocated to high-efficiency energy storage units. It is then dynamically released during off-peak periods to supplement the energy gap, thereby storing excess tidal energy and obtaining the available capacity index of buffer energy.

[0063] Based on the established off-peak activation conditions, the difference between tidal power generation and the real-time power consumption of the data center is detected. If the difference is positive, it is identified as excess energy. The excess tidal energy is allocated according to the rated input power ratio of the electrochemical energy storage units and the mechanical energy storage units, resulting in the input power allocation value for each energy storage channel. Using the input power allocation value, the electrochemical energy storage units are charged, the state of charge of the battery pack is monitored, and the flywheel of the mechanical energy storage unit is accelerated or compressed air is added to the compressed air tank. The energy storage growth rate and conversion efficiency of each energy storage unit are recorded, and more energy is allocated to energy storage units with conversion efficiency exceeding a preset threshold. By continuously storing excess energy, when the tidal energy enters an off-peak period and the data center power consumption exceeds the tidal power generation, the energy gap value is calculated. The release order is determined based on the remaining stored energy and discharge efficiency of each energy storage unit, and the output power of the energy storage units is controlled to dynamically release stored energy to fill the gap. The current remaining capacity of each energy storage unit, the overall conversion efficiency from storage to release, and the peak instantaneous release rate are statistically analyzed. These three parameters are combined to form the available capacity index of buffer energy, resulting in an available capacity index of buffer energy that includes energy storage capacity, conversion efficiency, and release rate.

[0064] Specifically, in one implementation, excess tidal energy is detected based on real-time power difference calculations. The output power of the tidal power station is collected once per second by a power transmitter, while the total power consumption of the data center is monitored in real time by the main meter of the smart distribution cabinet. When the tidal power generation is 15 MW and the data center only consumes 10 MW, the difference of 5 MW is considered excess energy. This excess energy needs to be stored promptly to avoid energy loss caused by directly feeding it back into the grid.

[0065] Specifically, the power allocation of the energy storage channels follows the capacity matching principle. The electrochemical energy storage units use lithium iron phosphate battery packs, with a rated input power of 500 kW for a single battery cabinet. The entire energy storage station is equipped with 10 battery cabinets, for a total rated input power of 5 MW. Mechanical energy storage employs both flywheel and compressed air energy storage methods. The rated input power of the flywheel energy storage system is 2 MW, and that of the compressed air energy storage system is 3 MW. Based on the rated input power ratio of each energy storage unit, excess energy is allocated to the three energy storage channels in a 5:2:3 ratio. When the excess energy is 5 MW, 2.5 MW is allocated to electrochemical energy storage, 1 MW to flywheel energy storage, and 1.5 MW to compressed air energy storage.

[0066] For example, the electrochemical energy storage charging process employs a three-stage charging strategy. The first stage is constant current charging, where the charging current is set at 0.5 times the battery capacity. During this stage, the battery rapidly absorbs energy, and the state of charge (SOC) increases from 20% to 80%. The second stage is constant voltage charging. Once the battery voltage reaches its rated value, the voltage remains constant, and the charging current gradually decreases, increasing the SOC from 80% to 95%. The third stage is trickle charging, which maintains the battery at a very low current to ensure full charge. Throughout the charging process, the battery management system monitors the voltage, temperature, and internal resistance of each cell in real time. When the temperature exceeds 45 degrees Celsius, the charging power is reduced to ensure charging safety.

[0067] It's important to note that different mechanical energy storage systems operate on different principles. Flywheel energy storage uses an electric motor to drive a flywheel to rotate at high speed within a vacuum chamber, converting electrical energy into rotational kinetic energy. The flywheel is made of carbon fiber composite material, with a maximum speed of 20,000 revolutions per minute, and a single flywheel can store 50 kilowatt-hours of energy. Compressed air energy storage uses a compressor to compress air into a high-pressure storage tank, reaching pressures up to 30 MPa. The heat generated during compression is recovered through a heat exchanger and stored in the heat storage material, improving overall energy conversion efficiency. When energy needs to be released, the high-pressure air expands through a turbine generator, performing work while simultaneously using the stored heat to heat the air, further improving power generation efficiency.

[0068] Preferably, real-time monitoring and optimization of conversion efficiency are key to improving the performance of energy storage systems. The charge-discharge efficiency of electrochemical energy storage is calculated by comparing input and output energy; the round-trip efficiency of lithium iron phosphate batteries is typically between 90% and 95%. The efficiency of flywheel energy storage is mainly affected by bearing friction and wind resistance, and can reach 85% to 90% in a vacuum environment. Compressed air energy storage has a relatively low efficiency, but it can be improved to 70% to 75% through heat recovery technology. Power allocation is dynamically adjusted based on the real-time efficiency of each energy storage unit. When the electrochemical energy storage efficiency is higher than a preset threshold of 92%, its allocation ratio is increased to 60%; when the flywheel energy storage efficiency is lower than 85%, its allocation ratio is reduced to 15%.

[0069] In one possible implementation, the dynamic replenishment of the energy gap employs a tiered release strategy. When tidal energy enters a low-voltage period, generating power drops to 8 MW, while the data center load is 12 MW, creating a 4 MW energy gap. First, the remaining capacity and discharge capability of each energy storage unit are assessed. Electrochemical energy storage, due to its fast response and stable output, is chosen as the first priority release source. Flywheel energy storage, with a response time in the millisecond range, is suitable for handling instantaneous power fluctuations and is chosen as the second priority. Compressed air energy storage, with its longer start-up time but longer continuous power supply time, is chosen as the third priority. Based on the size and duration of the gap, the system automatically combines different energy storage units to achieve smooth energy replenishment.

[0070] For example, during the low tide period in the early morning, when tidal power generation drops to its lowest point, data centers still have batch computing tasks running. Upon detecting a 3 MW energy shortfall, electrochemical energy storage is immediately activated to release 2 MW, while the compressed air energy storage system is preheated. Five minutes later, the compressed air energy storage begins releasing 1 MW, and the electrochemical energy storage correspondingly reduces its output, maintaining a total output of 3 MW. This coordinated release strategy both meets immediate energy demand and extends the power supply time of the energy storage system.

[0071] In one embodiment, the statistical analysis of available buffer energy capacity provides a basis for system operation and maintenance decisions. Energy storage capacity is obtained by summing the remaining power of each energy storage unit: 2000 kWh remaining for electrochemical energy storage, 100 kWh remaining for flywheel energy storage, and 1500 kWh remaining for compressed air energy storage, for a total capacity of 3600 kWh. Conversion efficiency is calculated as the weighted average of all energy storage units, with weights determined according to the proportion of actual released power. Release rate records the system's maximum released power over the past 24 hours, reflecting the system's instantaneous power supply capability.

[0072] Understandably, by establishing a comprehensive energy storage system and dynamic control mechanism, data centers have achieved time-shifted utilization of tidal energy. Excess energy is stored during periods of energy abundance, and released to fill gaps during periods of energy scarcity, forming a stable and reliable green energy supply system. Three-dimensional capacity indicators reflect the real-time operating status of the energy storage system, providing quantitative support for energy dispatch decisions and ensuring the stable operation of data centers even when fully reliant on tidal energy.

[0073] During off-peak periods, real-time energy demand data and tidal energy supply data are first received to identify the time period and scale of the energy gap. Then, excess tidal energy stored in the energy storage unit is dynamically released, and the release rate is adjusted to match the gap demand. During the adjustment process, the remaining capacity, energy conversion efficiency, and release rate of the energy storage unit are monitored in real time and converted into an available capacity index of buffer energy. The conversion efficiency is evaluated, and combined with the release rate record, a comprehensive index including energy storage capacity, conversion efficiency, and release rate is obtained.

[0074] During off-peak periods, real-time energy demand data from the data center and tidal energy supply data from the tidal power station are received. By comparing the time-series curves of the two sets of data, the difference between the demanded power and the supplied power at each time point is calculated. If the demand exceeds the supply, it is marked as a gap period. The start time, duration, and power difference of the gap are recorded to obtain the time-period distribution and scale of the energy gap. Based on the time-period distribution and scale of the energy gap, the current storage capacity of electrochemical and mechanical energy storage in the energy storage unit is read. The gap power value is used as the release target, and the energy storage inverter and generator set are controlled to output electrical energy according to the target power. By adjusting the pulse width modulation duty cycle of the inverter and the speed of the generator set, the actual release rate is matched with the gap demand to obtain a dynamic release power curve. Energy release is executed through the dynamic release power curve. The remaining electricity of the energy storage unit is collected in real time as the remaining capacity, the ratio of output electricity to consumed stored energy is calculated as the conversion efficiency, and the instantaneous maximum output power is recorded as the release rate. These three parameters are updated in real time and the change process is recorded to obtain an energy storage operation status dataset. Using the energy storage operation status dataset, the average conversion efficiency over the entire release cycle is statistically analyzed, and the peak and average values ​​of the release rate are extracted. Combined with the current remaining capacity value, the data of the three dimensions of energy storage capacity, conversion efficiency, and release rate are combined to obtain a comprehensive index that includes energy storage capacity, conversion efficiency, and release rate.

[0075] Specifically, in one implementation, energy gap identification is based on precise comparative analysis of time-series curves. Energy demand data from the data center is collected via a distributed sensor network, with each server rack equipped with a power monitoring module that reports instantaneous power values ​​at a rate of seconds. Supply data from the tidal power station comes from the output of the generator sets; three-phase electrical parameters are measured using current and voltage transformers to calculate real-time power generation. These two data streams converge on the energy management platform, forming a time-aligned power curve. When the demand is 12 MW and the supply is only 8 MW at a given moment, the 4 MW difference represents the energy gap at that moment.

[0076] Specifically, the pulse width modulation (PWM) duty cycle adjustment enables precise control of the energy storage power release. The energy storage inverter employs a full-bridge topology, adjusting the effective value of the output voltage by controlling the on-time ratio of the switching transistors. The duty cycle is continuously adjustable from 0 to 100%, corresponding to a linear change in output power from zero to rated power. When 2 MW of power needs to be released, the controller calculates the required duty cycle value based on the current DC bus voltage and load impedance, typically between 60% and 80%. The modulation frequency is set to 20 kHz, above the range of human hearing, to avoid noise interference. The generator set's speed control uses variable frequency drive (VFD) technology, adjusting the speed by changing the supply frequency of the stator windings, thereby controlling the output power.

[0077] For example, the formation of the dynamic power release curve involves the coordination of multiple control links. First, the total power release target is determined based on the deficit size. Then, release tasks are allocated according to the current capacity and efficiency of each energy storage unit. Electrochemical energy storage, due to its fast response speed, undertakes the base load portion, maintaining a relatively stable output. Mechanical energy storage is responsible for handling the fluctuating portion, dynamically adjusting its output according to real-time deficit changes. The controller updates the power allocation command every 100 milliseconds to ensure that the total output power always matches the deficit demand. During the release process, if an energy storage unit is detected to be nearing depletion, the system automatically transfers its load to other units, achieving seamless switching. This dynamic coordination mechanism ensures the continuity and stability of energy supply.

[0078] It should be noted that the construction of the energy storage operation status dataset relies on multi-dimensional real-time monitoring. For electrochemical energy storage, remaining capacity monitoring involves accumulating charge and discharge quantities using coulometric measurement, combined with open-circuit voltage correction, to obtain an accurate state of charge. For mechanical energy storage, flywheel systems calculate kinetic energy storage by measuring rotational speed, while compressed air systems calculate air storage capacity using pressure and temperature sensors. Conversion efficiency calculation requires simultaneous measurement of input and output energy. Input energy is obtained by integrating charging power, and output energy is obtained by integrating discharging power; the ratio of the two is the conversion efficiency. The release rate is measured directly using a power transmitter to measure instantaneous output power and record the peak value.

[0079] Preferably, the data sampling and processing during real-time monitoring adopts a hierarchical architecture. The bottom-level acquisition devices sample the raw signal at millisecond-level frequencies and remove high-frequency noise through digital filtering. The intermediate-level aggregation nodes perform preliminary data processing, calculating the moving average and standard deviation. The top-level management platform receives the processed data and performs trend analysis and anomaly detection. When a sudden drop in conversion efficiency exceeding 5% is detected, the system triggers a diagnostic procedure to check the health status of the energy storage unit.

[0080] In one possible implementation, statistical analysis of comprehensive indicators provides a quantitative basis for system optimization. The entire release cycle typically lasts 3 to 4 hours, corresponding to the duration of the tidal low period. The system records a set of operating parameters every minute, including the output power, remaining capacity, and internal temperature of each energy storage unit. After the release cycle ends, all data are statistically processed. The average conversion efficiency is obtained by arithmetically averaging the efficiency values ​​at each moment, reflecting the overall performance of the energy storage system. The peak and average release rates reflect the system's maximum power supply capacity and average power supply level, respectively.

[0081] For example, during a typical off-peak energy replenishment process, the system detected a 3 MW energy gap. Electrochemical energy storage immediately released 2 MW, flywheel energy storage released 0.5 MW, and compressed air energy storage released 0.5 MW after 2 minutes of preheating. Over the next 3 hours, the gap fluctuated between 2 and 4 MW, and the system maintained a supply-demand balance by dynamically adjusting the output of each energy storage unit. Throughout the process, the conversion efficiency of electrochemical energy storage remained at 92%, the overall efficiency of mechanical energy storage was 78%, and the overall system efficiency reached 87%.

[0082] In one embodiment, the visualization of three-dimensional comprehensive indicators helps maintenance personnel intuitively understand the system status. Energy storage capacity is displayed as a bar chart, with different colors representing different types of energy storage units. Conversion efficiency is displayed as a line graph to show historical trends, making it easy to identify performance degradation. Release rate is presented as a heatmap, with color intensity indicating power output.

[0083] Step S106: Analyze the degree of mitigation of resource mismatch from the available capacity index of buffer energy, reallocate idle computing resources according to the degree of mitigation, and determine the supplementary demand during periods of insufficient tidal energy.

[0084] The energy storage capacity, conversion efficiency, and release rate are extracted from the available capacity indicators of buffer energy. The energy storage capacity is multiplied by the conversion efficiency to obtain the actual available energy. This is compared with the theoretical energy demand of the data center during the corresponding period. If the ratio exceeds a preset threshold, the resource mismatch is considered mitigated, resulting in a resource mismatch mitigation degree value. Based on this mitigation degree value, idle computing nodes within the data center are identified. Idle computing power corresponding to periods with high mitigation degrees is reallocated to low-priority tasks in the task queue, resulting in a computing power resource reallocation table. Using this table, the number of additional computing nodes required to be started during periods of tidal energy shortage is calculated as the computing power replenishment amount. Energy replenishment priorities are determined based on task urgency, and the average task waiting time is calculated as the task scheduling latency. This determines the replenishment demand during periods of tidal energy shortage, including computing power replenishment amount, energy replenishment priority, and task scheduling latency.

[0085] Specifically, in one implementation, the assessment of resource mismatch mitigation is based on the actual supply capacity of the energy storage system. The actual available energy is calculated by multiplying the energy storage capacity by the conversion efficiency; for example, if the energy storage capacity is 1000 kWh and the conversion efficiency is 85%, then the actual available energy is 850 kWh. The theoretical energy demand of a data center is calculated based on historical operating data, typically ranging from 800 to 1200 kWh during off-peak periods. When the ratio of actual available energy to theoretical energy demand exceeds 0.8, the resource mismatch is considered to be effectively mitigated.

[0086] Specifically, the identification and reallocation of idle computing power follows the principle of load balancing. The resource usage of all computing nodes is scanned, and nodes with CPU utilization below 30% and memory usage below 40% are marked as idle. Based on the mitigation level, the system manages this idle computing power in a tiered manner. During periods with a mitigation level higher than 0.9, idle computing power can be allocated to low-priority tasks such as batch data processing and log analysis; during periods with a mitigation level between 0.7 and 0.9, it is only allocated to interruptible background tasks. This differentiated allocation ensures resource availability for critical tasks.

[0087] It should be noted that the computing power replenishment statistics take into account the real-time status of the task queue. The resource requirements of each task in the queue are monitored and summed to arrive at the total requirement. The difference between this total and the total computing power of currently active nodes represents the number of additional nodes that need to be started. For example, if there are 50 video transcoding tasks in the queue, each requiring 2 CPU cores, and only 80 cores are currently available, then the computing power replenishment requirement is 20 cores, corresponding to the need to start 5 quad-core nodes.

[0088] For example, energy replenishment priority is determined based on the business attributes of the tasks. Real-time live streaming transcoding tasks, due to their high timeliness requirements, are set to the highest priority; user interactive queries are next; batch processing and data backup tasks have the lowest priority. Task scheduling latency is calculated by recording the time interval from task submission to execution start. The system maintains a sliding time window and calculates the average waiting time of all tasks over the past hour. In practical applications, when tidal energy enters a period of shortage, the system allocates resources according to the determined replenishment needs. If the computing power replenishment amount is 10 nodes, newer servers with high energy efficiency are prioritized for wake-up; energy replenishment is allocated according to priority order to ensure that high-priority tasks receive sufficient power; for tasks with acceptable delays, their scheduling latency is appropriately extended until energy is sufficient before execution.

[0089] Step S107: Obtain the task execution logs that match the periods when tidal energy supply is sufficient after dynamic migration, analyze the task execution efficiency and energy consumption data in the task execution logs, and generate a computing power optimization scheme for the data center driven by green energy in combination with the supplementary needs during periods when tidal energy is insufficient.

[0090] After dynamic migration, task execution logs matching periods with sufficient tidal energy supply are obtained. Task identifiers, start times, end times, CPU utilization, and memory usage data are extracted from the log files. The execution duration is obtained by subtracting the start time from the end time. The average of CPU utilization and memory usage is used as the resource utilization rate, resulting in a set of task execution efficiency indicators. Based on this set of indicators, the corresponding time-period power consumption data recorded by the data center power distribution cabinet is read. The total power consumption is divided by the number of completed computing tasks to obtain the unit task energy consumption. Simultaneously, the computing power replenishment amount, energy supply priority, and task scheduling latency parameters from the supplementary demand during periods of insufficient tidal energy are extracted to obtain energy consumption assessment results. Based on these energy consumption assessment results, high-energy-consuming tasks are scheduled for execution during periods of sufficient tidal energy, while low-energy-consuming tasks are distributed across various time periods. Hourly power quotas and energy storage release plans are developed based on the tidal energy supply curve, resulting in time-based scheduling and energy allocation rules. Using the aforementioned scheduling and energy allocation rules, the number of server nodes that need to be activated or hibernated is determined based on the task load in each time period. The three parts of task execution time arrangement, energy quota for each time period, and node addition / reduction plan are integrated to generate a green energy-driven data center computing power optimization scheme that includes task scheduling, energy allocation plan, and dynamic node adjustment.

[0091] Specifically, in one implementation, the acquisition and analysis of task execution logs form the data foundation of the entire optimization scheme. The data center's task management system records detailed logs for each executed task, including multi-dimensional information such as the task's unique identifier, submission time, start time, completion time, allocated CPU cores, actual CPU utilization, memory allocation, actual memory usage, and disk I / O throughput. These log files are stored in a distributed file system in timestamp order, with one log file generated every hour, facilitating time-based retrieval and analysis.

[0092] Specifically, calculating task execution efficiency involves a comprehensive evaluation of multiple performance metrics. Execution time is calculated by subtracting the start time from the completion time, reflecting the task's time efficiency. Resource utilization requires separate calculations for CPU and memory. CPU utilization equals the actual CPU time used divided by the allocated CPU time, and memory utilization equals the peak memory usage divided by the allocated memory capacity. For video transcoding tasks, typical execution times range from 30 to 120 minutes, with CPU utilization typically between 70% and 90%. Data backup tasks have longer execution times, reaching 3 to 5 hours, but with relatively lower CPU utilization, approximately 40% to 60%. Statistical analysis of these metrics can identify the execution characteristics and resource consumption patterns of different task types.

[0093] For example, the collection and analysis of energy consumption data needs to be accurate to the task level. A data center's intelligent power distribution system installs smart meters on each server rack, recording power consumption with second-level precision. When a task begins execution, the system records the initial meter reading on the server hosting the task; when the task completes, the meter reading is read again, and the difference between the two readings represents the total energy consumption of the task. Calculating energy consumption per unit of task requires considering the computational workload, typically measured in floating-point operations or the amount of data processed.

[0094] For example, if a transcoding task processing 1TB of video data consumes 500 kWh of electricity, then the energy consumption per GB is 0.5 kWh. This refined energy consumption analysis provides a quantitative basis for subsequent task scheduling optimization.

[0095] It's important to note that the three dimensions of energy replenishment requirements during periods of tidal energy shortage are interconnected. The amount of computing power replenishment reflects the additional computing resources needed to be activated, calculated by the difference between the total resource demands of the pending tasks and the currently available resources. Energy replenishment priority is determined based on the task's business importance and timeliness requirements, with real-time streaming media processing receiving the highest priority and batch processing tasks receiving the lowest. Task scheduling latency represents the maximum tolerable waiting time for a task; for real-time tasks, latency requirements are in the millisecond range; for batch processing tasks, delays of several hours are acceptable.

[0096] Preferably, the time-based scheduling rules are formulated based on the periodic characteristics of the energy supply curve. The day is divided into 24 time periods, each corresponding to one hour. Based on historical data analysis, tidal energy peaks between 2 PM and 5 PM, and this period is designated as the energy-sufficient period. High-energy-consuming tasks, such as large-scale data analysis and machine learning model training, are prioritized for execution during these periods. Low-energy-consuming tasks, such as log archiving and configuration backup, are distributed across various time periods to fully utilize the basic energy supply.

[0097] In one possible implementation, the energy allocation plan needs to balance multiple constraints. Power quotas for each time period are determined based on tidal power generation forecasts, typically with a 10% safety margin. Energy storage release plans are formulated based on the current capacity of the storage systems and the anticipated energy gap. When the workload for a given time period exceeds the tidal energy supply, electrochemical energy storage is prioritized for supplementation due to its rapid response; if the gap is prolonged, mechanical energy storage systems are activated. Quota allocation follows a priority principle to ensure that critical tasks receive sufficient energy supply.

[0098] For example, in a typical operating cycle, a data center schedules 50 video transcoding tasks and 20 data backup tasks during the early morning off-peak period. Based on task execution efficiency analysis, these tasks are expected to consume 8 MWh of electricity. Meanwhile, the predicted tidal power generation for the same period is 10 MWh, leaving a 2 MWh reserve. Based on this, an energy allocation plan is formulated: 6 MWh are allocated to the video transcoding tasks, 2 MWh to the backup tasks, and the remaining 2 MWh is used to charge the energy storage system to prepare for subsequent periods of energy shortage.

[0099] In one embodiment, the decision to dynamically adjust nodes is based on real-time load monitoring and prediction. A node status table is maintained, recording the operating status, current load, and energy consumption level of each server. When the task load increases during a certain period, the system activates dormant nodes in descending order of energy efficiency; when the load decreases, older servers with lower energy efficiency are prioritized for dormancy. This dynamic adjustment mechanism ensures that energy consumption is minimized while meeting computing needs.

[0100] Understandably, the computing power optimization solution is a comprehensive one, comprising three core components: task scheduling strategy, energy allocation plan, and node management. The task scheduling strategy clarifies the execution schedule of various tasks at different times; the energy allocation plan specifies the power quota and energy storage usage plan for each time period; and the node dynamic adjustment plan determines the start-up and shutdown timing and number of servers. These three components work together to achieve efficient utilization of data center computing resources under tidal energy constraints, achieving the dual goals of green energy saving and stable operation.

[0101] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for optimizing computing power in data centers driven by green energy, characterized in that, include: Analyze the matching degree between tidal energy supply data and computational load demand data to determine the peak misalignment period between tidal energy supply and computational load demand; Based on the peak-off period, the execution window of high-energy-consuming computing tasks is evaluated, and a list of available resources during periods of sufficient energy is generated. The priority sequence of high-energy-consuming tasks is extracted from the list of available resources, and tasks that match the tidal energy supply period are dynamically migrated to the corresponding period for execution. The balance of computing power resource allocation after migration is then determined. Based on the balanced distribution of computing resources and the current load of the server cluster, adjust the server cluster size to the target number of active nodes and determine the activation conditions during off-peak periods; based on the activation conditions during off-peak periods, generate a buffer energy availability capacity index; based on the buffer energy availability capacity index, reallocate idle computing resources and determine the replenishment needs during periods of tidal energy shortage; analyze the task execution efficiency and energy consumption data in the task execution logs after dynamic migration, and combine them with the replenishment needs during periods of tidal energy shortage to generate a computing power optimization scheme.

2. The computing power optimization method according to claim 1, characterized in that, The analysis of the matching degree between tidal energy supply data and computational load demand data, and the determination of peak misalignment periods between tidal energy supply and computational load demand, includes: Data on tidal height, current velocity, and power generation from tidal power plants are collected, along with data on CPU utilization, memory usage, and network bandwidth consumption from data center servers. The tidal power plant data is decomposed into a time series to extract periodic characteristic curves. Based on these curves, a sliding window method is used to process and calculate load demand data, determining the load mean and variance, and identifying peak load times. A supply-demand time series deviation matrix is ​​generated by comparing the peak tidal energy supply times with the peak load times. Periods with time offsets exceeding a threshold are extracted from the deviation matrix and marked as misaligned periods. A misaligned period record table is generated, containing the misalignment start time, duration, and energy difference.

3. The computing power optimization method according to claim 1, characterized in that, The step involves evaluating the execution window of high-energy-consuming computing tasks based on the peak-off periods and generating a list of available resources during periods of sufficient energy, including: Based on the peak-off time periods, identify the power requirements and runtime of video transcoding, batch data processing, and machine learning training tasks; compare tidal energy supply data with power requirements to determine the time interval during which energy supply exceeds task requirements as the execution window; calculate the net power value after subtracting infrastructure power consumption from tidal power generation within the execution window to determine the maximum number of supported nodes and total storage capacity; based on the maximum number of nodes, query the network device configuration table, accumulate the node network card speed as the bandwidth requirement value, adjust the bandwidth quota, and generate an available resource list containing the number of computing nodes, storage capacity, and bandwidth allocation.

4. The computing power optimization method according to claim 1, characterized in that, The step of extracting the priority sequence of high-energy-consuming tasks from the available resource list, dynamically migrating tasks matching the tidal energy supply period to the corresponding period for execution, and determining the balance of computing power resource allocation after migration includes: The number of computing nodes, storage capacity, and bandwidth allocation data are read from the available resource list. A priority sequence of high-energy-consuming tasks is generated by combining the difference between the task deadline and the current time. The execution duration of tasks in the priority sequence is compared with the tidal energy supply period, matching tasks are marked, and a task migration candidate set is constructed. Tasks in the task migration candidate set are migrated to computing nodes in the corresponding time period, and the CPU utilization, memory usage, and bandwidth occupancy rate are recorded for each time period after migration. The variance of the CPU utilization rate is used to evaluate the balance of computing resource allocation.

5. The computing power optimization method according to claim 1, characterized in that, After generating the list of available resources during periods of sufficient energy, it includes: Extract the execution requirement parameters of tasks in the task migration candidate set, including minimum runtime, maximum allowable delay, resource consumption, and interruptibility; calculate the overlap ratio between the time window of the tasks in the task migration candidate set and the tidal energy supply period to determine the fit; adjust the task scheduling plan according to the fit, and migrate the task execution time to the start time of the tidal energy supply period; set the task end time according to the minimum runtime, and generate the migrated and adjusted task execution schedule.

6. The computing power optimization method according to claim 1, characterized in that, The step of adjusting the server cluster size to the target number of active nodes based on the computing power resource allocation balance and the current load of the server cluster, and determining the activation conditions during off-peak periods, includes: Based on the balance of computing resources allocation, obtain the CPU utilization, memory usage, and task queue length of the server cluster, and calculate the current load value; if the current load value exceeds the threshold, sort the node load values ​​and execute a hibernation operation until the target number of active nodes is reached; based on the target number of active nodes, set the minimum load threshold, node operating efficiency, and energy supply stability, and generate off-peak activation conditions.

7. The computing power optimization method according to claim 1, characterized in that, The buffer energy availability capacity index generated based on the low-peak activation conditions includes: Based on the off-peak activation conditions, the difference between tidal power generation and data center power consumption is detected, and excess energy is allocated to electrochemical or mechanical energy storage units; a charging or mechanical energy storage process is executed, and the energy storage growth rate and conversion efficiency are recorded; when the tidal energy supply data is lower than the power consumption, the energy storage units are controlled to release energy, and the remaining capacity, conversion efficiency and release rate are statistically analyzed to generate a buffer energy available capacity index.

8. The computing power optimization method according to claim 1, characterized in that, The generated buffer energy available capacity index includes: The tidal energy supply data is compared with the data center power consumption to mark the energy gap period and power difference; based on the power difference, the energy storage unit outputs electrical energy and the release rate is adjusted to match the gap; the remaining power of the energy storage unit, the ratio of output power to consumed stored energy, and the instantaneous maximum output power are collected; the average conversion efficiency and peak release rate within the release cycle are statistically analyzed to generate a buffer energy available capacity index.

9. The computing power optimization method according to claim 1, characterized in that, The step of reallocating idle computing resources based on the available capacity index of buffer energy and determining the replenishment demand during periods of tidal energy shortage includes: Extract energy storage capacity and conversion efficiency from the available capacity index of the buffer energy, and calculate the actual available energy; compare the actual available energy with the required energy to determine the resource mismatch mitigation level; allocate idle computing power to low priority tasks according to the resource mismatch mitigation level; and generate supplementary requirements including computing power replenishment amount, energy supply priority, and task scheduling latency by statistically analyzing the number of additional startup nodes, task urgency, and average waiting time.

10. The computing power optimization method according to claim 1, characterized in that, The analysis of task execution efficiency and energy consumption data in the dynamically migrated task execution logs, combined with the replenishment needs during periods of tidal energy shortage, generates a computing power optimization scheme, including: Extract task identifiers, execution durations, and resource utilization rates from the task execution logs to generate a set of task execution efficiency indicators; calculate unit task energy consumption based on the set of task execution efficiency indicators; schedule high-energy-consuming tasks to energy-sufficient periods based on the replenishment needs during periods of tidal energy shortage, and formulate power quotas and energy storage release plans; determine the number of nodes to activate or hibernate based on the task load, and generate a computing power optimization scheme.

Images