Calculation force load stabilizing and energy-saving management system based on flywheel energy storage system

Abstract

The invention discloses a computing power load stabilizing and energy-saving management system based on a flywheel energy storage system, and belongs to the technical field of energy storage and management, and the system comprises a data collection module which is used for collecting and calculating a real-time current harmonic spectrum of a cluster and server rack vibration data to generate an electric energy quality characteristic spectrum; the prediction modeling module is used for establishing a prediction model output computing power demand dynamic distribution matrix based on multi-dimensional space-time correlation; the optimization control module is used for analyzing a peak load phase in the power demand dynamic distribution matrix to generate a dynamic coordination control strategy; and the power execution module is used for executing the dynamic coordination control strategy to adjust the phase parameter of the reverse electromotive force of the flywheel so as to offset the voltage harmonic component of the server bus. Harmonic phase compensation is realized through dynamic reverse electromotive force modulation, the calculation load prediction precision can be effectively improved, and the energy utilization efficiency and the equipment operation reliability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage and management, and particularly to a computing power load smoothing and energy-saving management system based on a flywheel energy storage system. Background Art

[0002] A flywheel energy storage device consists of a high-speed rotating wheel and an electric motor or generator, stores energy by converting mechanical energy into electrical energy, and converts it back into mechanical energy when needed, and is widely used in power systems, aerospace, transportation and other fields.

[0003] Current mainstream load management schemes usually adopt a one-way energy buffering strategy, lacking a phase coupling mechanism for the harmonic components of the server bus and the back electromotive force of the flywheel energy storage system, and the power quality regulation effect is difficult to meet the power supply requirements of high-precision computing devices. At the same time, existing optimization algorithms are prone to falling into local optimal solutions when solving multi-objective constraints, and do not fully consider safety parameters such as the stress distribution of the flywheel rotor and the fatigue evolution of materials, resulting in insufficient long-term operation reliability of the system.

[0004] In the integration of the flywheel energy storage system and the computing power cluster, existing technical solutions mostly control energy release through a fixed threshold mechanism, unable to dynamically analyze the phase characteristics of peak loads, and it is difficult to achieve precise matching between the flywheel kinetic energy curve and the computing power fluctuations. In addition, traditional harmonic suppression methods rely on passive filtering devices and do not effectively utilize the electromagnetic inertia characteristics of the flywheel for active compensation, especially prone to harmonic resonance in transient overload scenarios, exacerbating equipment losses.

[0005] In summary, there are generally problems in the existing computing power load smoothing systems such as slow dynamic response, insufficient multi-objective optimization, and single harmonic compensation mechanism. There is an urgent need for a collaborative management method that deeply integrates the characteristics of flywheel energy storage and computing power demand prediction to achieve synchronous improvement of energy efficiency and power supply quality. Summary of the Invention

[0006] To solve the above problems, the present invention provides a computing power load smoothing and energy-saving management system based on a flywheel energy storage system, which realizes harmonic phase compensation through dynamic back electromotive force modulation, can effectively improve the prediction accuracy of computing power load, and significantly improve energy utilization efficiency and equipment operation reliability.

[0007] The above object can be achieved by the following solutions: A computing power load smoothing and energy-saving management system based on a flywheel energy storage system, including a data acquisition module, which is used to collect the real-time current harmonic spectrum of a computing cluster and the server rack vibration data, and generate an electric energy quality feature map; a prediction and modeling module, which is used to establish a prediction model based on multi-dimensional spatio-temporal correlation, perform spatio-temporal convolution operation on the electric energy quality feature map, grid load fluctuation data, and historical power consumption curve, and output a dynamic distribution matrix of computing power demand; an optimization control module, which is used to analyze the peak load phase in the dynamic distribution matrix of computing power demand, extract the rotational speed fluctuation characteristics of the flywheel energy storage unit and the energy consumption time series correlation parameters of the server rack, and generate a dynamic coordination control strategy; a power execution module, which is used to execute the dynamic coordination control strategy and adjust the phase parameter of the flywheel back electromotive force to cancel the voltage harmonic component of the server bus.

[0008] Optionally, the prediction and modeling module includes: an input parameter acquisition unit, which is used to obtain the temperature data of the server room, the air pressure monitoring data of the flywheel energy storage cavity, the electric energy quality feature map, and a preset set of constraint conditions, and construct a multi-dimensional input parameter set, where the set of constraint conditions includes a temperature rise gradient threshold, an air pressure safety threshold, and a safe rotational speed threshold; a prediction model construction unit, which is used to establish a spatio-temporal correlation convolutional neural network model and fuse the multi-dimensional input parameter set with the grid real-time frequency modulation data; a data processing unit, which is used to apply a sliding time window algorithm to dynamically segment the fused data and generate a spatio-temporal feature tensor containing spatial correlation, where the spatio-temporal feature tensor includes the spatial correlation of server energy consumption distribution and load fluctuation trend; a computing power demand output unit, which is used to calibrate the time series deviation component in the spatio-temporal feature tensor by using a long short-term memory network and output a corrected dynamic distribution matrix of computing power demand.

[0009] Optionally, the optimization control module includes: an optimization function establishment unit, which is used to construct a multi-objective optimization function including energy storage capacity constraint, heat dissipation efficiency equation, and power supply cost model. For the multi-objective optimization function , there is ; In the formula, is the decoupling coefficient, is the weight coefficient of flywheel mechanical loss, is the weight coefficient of task delay time, is the flywheel mechanical loss power, is the flywheel mechanical rated power, is the task delay time, is the maximum time of the task; a solution set calculation unit that calls a quantum-inspired annealing algorithm to solve the multi-objective optimization function, calculates the optimization balance point between the flywheel mechanical loss power and the task delay time, and outputs a set of Pareto optimal solutions; a solution set screening unit that inputs the set of Pareto optimal solutions into a preset quantum genetic algorithm, and performs iterative optimization based on the computing power demand dynamic distribution matrix to generate an energy scheduling optimization solution vector; a control strategy generation unit that maps the energy scheduling optimization solution vector to physical control instructions, generates a flywheel speed adjustment curve and a server power allocation sequence, and obtains a dynamic coordination control strategy.

[0010] Optionally, the optimization process of the quantum genetic algorithm includes: designing a hybrid coding mechanism based on quantum phase rotation parameters, correlating the stress monitoring data of the flywheel rotor with the server chip temperature, where the stress monitoring data is extracted from the vibration data of the power quality feature map; introducing a dynamic penalty factor weight, and adjusting the weight according to the flywheel bearing wear coefficient in the previous cycle, where the bearing wear coefficient is calculated based on the air pressure monitoring data; by amplifying the amplitude gradient distribution of the quantum state, generating a constraint boundary in combination with the heat dissipation efficiency equation, and using Monte Carlo sampling to screen the energy scheduling optimization solution vectors whose energy consumption difference meets the preset difference.

[0011] Optionally, the execution of the dynamic coordination control strategy includes: controlling the pulse width modulation signal of the electromagnetic eddy current brake to convert the flywheel kinetic energy into a controllable back electromotive force, where the phase of the controllable back electromotive force is determined by the energy storage capacity constraint; based on the heat dissipation efficiency equation, adjusting the flow rate parameter of the liquid metal cooling loop to synchronously match the braking heat distribution of the flywheel group; coupling the back electromotive force with the harmonic components of the server power supply bus, and performing dynamic voltage compensation through a three-phase transformer, where the harmonic components are correlated with the current harmonic spectrum in the power quality feature map.

[0012] Optionally, the modulation process of the back electromotive force includes: detecting the harmonic distortion rate of the server power bus, generating a real-time waveform matching template based on the current harmonic spectrum, and calculating the equivalent load impedance using the waveform matching template; absorbing the excessive components of the transient current pulse through a fast-response energy storage inductor, combining with the equivalent load impedance, calculating the overload energy characteristic parameter, and recording it in the power quality feature map; analyzing the overload energy characteristic parameter according to the amplitude gradient distribution of the quantum state to adjust the excitation parameter of the flywheel electromagnet and output the back electromotive force.

[0013] Optionally, after generating the power quality characteristic map, the following steps are further included: identifying abnormal harmonic components in the power quality characteristic map, and combining the spatio-temporal feature tensor to construct an energy consumption characteristic coding tree for the server device; based on the node weight distribution of the coding tree, positioning the physical server cluster coordinates corresponding to the high-energy-consuming computing processes; mapping the coordinate information to the load gradient parameters of the multi-dimensional input parameter set, and inputting them into the multi-dimensional spatio-temporal correlation prediction model to generate an energy efficiency optimization instruction based on fuzzy control to adjust the weight coefficient of the task delay time.

[0014] Optionally, the system further includes a safety monitoring module, and the safety monitoring module is used for: collecting the residual strain distribution data of the flywheel rotor, and reconstructing the three-dimensional stress field distribution cloud map according to the stress monitoring data; inputting the stress field distribution cloud map into a preset phase change material life prediction model to calculate the material fatigue limit; calculating the safety speed threshold for the next cycle according to the energy storage capacity constraint and the material fatigue limit, and updating the constraint condition set.

[0015] Optionally, the method for updating the phase change material life prediction model includes: detecting the microscopic crack distribution of the flywheel composite material, obtaining the damage evolution direction through an ultrasonic guided wave array, and synchronizing the data to the stress field distribution cloud map; establishing a multi-scale finite element model based on the residual strain distribution data, and correlating the stress concentration coefficient with the cycle life of the energy storage unit; calculating the reliability index for different load scenarios according to the heat dissipation efficiency equation, and generating a confidence interval using a preset modified Weibull distribution model; obtaining the material fatigue limit according to the confidence interval.

[0016] Based on the same inventive concept, the present invention also provides a method for suppressing computing power load and energy-saving management based on a flywheel energy storage system. The method includes collecting the real-time current harmonic spectrum and server rack vibration data of a computing cluster to generate a power quality characteristic map; establishing a prediction model based on multi-dimensional spatio-temporal correlation, performing spatio-temporal convolution operations on the power quality characteristic map, grid load fluctuation data, and historical power consumption curves, and outputting a dynamic distribution matrix of computing power demand; analyzing the peak load phase in the dynamic distribution matrix of computing power demand, and extracting the rotational speed fluctuation characteristics of the flywheel energy storage unit and the energy consumption time series correlation parameters of the server rack; generating a dynamic coordination control strategy based on quantum-inspired optimization, where the control strategy includes the flywheel energy release rate constraint and the computing task scheduling priority topology; executing the dynamic coordination strategy to make the kinetic energy release curve of the flywheel group form a phase compensation effect with the computing load fluctuation.

[0017] Compared with the prior art, the present invention has the following advantages: 1. The present invention collects the current harmonic spectrum and server rack vibration data of the computing cluster in real time through the data acquisition module to generate a power quality characteristic map; combined with the multi-dimensional spatio-temporal correlation prediction model in the prediction modeling module, which integrates power grid load fluctuation data, historical power consumption curves, and multi-dimensional environmental data, it realizes high-precision dynamic prediction of computing power demand; this prediction method not only considers the spatio-temporal correlation, but also calibrates the time series deviation component through the sliding time window algorithm and the long short-term memory network, thus significantly improving the accuracy of computing power load prediction; at the same time, the system can dynamically respond to load changes, and cancel the voltage harmonic component of the server bus by adjusting the phase parameter of the flywheel back electromotive force, realizing fast and accurate load suppression; 2. The present invention constructs a multi-objective optimization function including energy storage capacity constraint, heat dissipation efficiency equation, and power supply cost model in the optimization control module; through the combined use of the quantum-inspired annealing algorithm and the quantum genetic algorithm, the system can find the optimal balance point between the mechanical loss of the flywheel and the task delay time, avoiding the problem that traditional multi-objective optimization algorithms are prone to falling into local optimal solutions; this optimization strategy not only considers the energy efficiency and economy of the system, but also dynamically adjusts the optimization weight to cope with actual operating conditions such as equipment wear, thus achieving a significant improvement in energy utilization efficiency and equipment operation reliability; 3. The present invention realizes the active modulation of the flywheel back electromotive force through the power execution module to compensate for the voltage harmonic component of the server bus; this method has higher efficiency and flexibility compared with traditional passive filtering devices; the system can detect the harmonic distortion rate of the server power bus in real time, generate a waveform matching template based on the current harmonic spectrum, and output the back electromotive force by adjusting the excitation parameters of the flywheel electromagnet to achieve precise cancellation of the harmonic component; in addition, the system also combines the flow rate adjustment of the liquid metal heat dissipation loop and the dynamic voltage compensation of the three-phase transformer to further improve the power quality and the heat dissipation efficiency of the system; this active harmonic compensation mechanism is particularly suitable for high-density computing clusters, and can effectively reduce equipment loss and extend the service life of equipment.

[0018] Other features and advantages of the present invention will be described in the following specification, and in part will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures pointed out in the specification, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0020] Figure 1 is a framework diagram of the computing power load smoothing and energy-saving management system based on the flywheel energy storage system according to an embodiment of the present invention.

[0021] Figure 2 is a schematic structural diagram of the computing power load smoothing and energy-saving management system based on the flywheel energy storage system according to an embodiment of the present invention.

[0022] Figure 3 is a three-dimensional effect diagram of 24-hour load dynamic matching according to an embodiment of the present invention.

[0023] Figure 4 is a comparison diagram of load fluctuation suppression according to an embodiment of the present invention.

[0024] Figure 5 is a schematic flow diagram of the computing power load smoothing and energy-saving management method based on the flywheel energy storage system according to an embodiment of the present invention. Specific Embodiments

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0026] Refer to Figure 1 , an embodiment of the present invention proposes a computing power load smoothing and energy-saving management system based on the flywheel energy storage system, which can achieve harmonic phase compensation through dynamic back electromotive force modulation, effectively improve the accuracy of computing power load prediction, and significantly improve the energy utilization efficiency and equipment operation reliability.

[0027] The system of this embodiment specifically includes: A data acquisition module, which is used to collect the real-time current harmonic spectrum and server rack vibration data of the computing cluster, and generate an electrical energy quality feature map; Specifically, the data acquisition module is the perception layer of the system. It collects the current harmonic spectrum of the computing cluster and the vibration signal of the server rack by deploying high-frequency sensors and vibration detectors. The current harmonic spectrum is the amplitude-frequency characteristic distribution data of each harmonic current in the power system, reflecting the distortion component in the power supply quality; the server rack vibration data is measured by an accelerometer, mainly identifying the mechanical resonance frequency caused by equipment overload. After performing fast Fourier transform and root mean square operation on both, a power quality characteristic map is generated. This map is a two-dimensional matrix model, and the matrix elements represent the coupling strength between specific harmonic frequency points and vibration energy, which is used to quantify the abnormal load area.

[0028] Specifically, the real-time current harmonic spectrum of the computing cluster is collected through fast Fourier transform, with the spectrum resolution ≤ 1Hz, and the root mean square value of the server rack vibration data is calculated within a 2-second sliding window. The power quality characteristic map is generated according to the following formula: ; In the formula, is the element in the th row and th column of the map, is the amplitude of the th harmonic current, is the root mean square value of the vibration at the corresponding frequency point, and

[0029] is the preset safe vibration threshold, with a value of 3mm / s. The prediction modeling module is used to establish a prediction model based on multi-dimensional spatio-temporal correlation, perform spatio-temporal convolution operation on the power quality characteristic map, grid load fluctuation data, and historical power consumption curve, and output the dynamic distribution matrix of computing power demand; The optimization control module is used to analyze the peak load phase in the dynamic distribution matrix of computing power demand, extract the rotational speed fluctuation characteristics of the flywheel energy storage unit and the energy consumption time series correlation parameters of the server rack, and generate a dynamic coordination control strategy; The power execution module is used to execute the dynamic coordination control strategy and adjust the phase parameter of the flywheel back electromotive force to cancel the voltage harmonic component of the server bus.

[0030] Specifically, the power execution module realizes harmonic suppression by adjusting the flywheel back electromotive force. The back electromotive force is the reverse voltage generated when the flywheel rotor cuts the magnetic induction line, and its phase parameter is dynamically determined by the spectral characteristics of the harmonic component. The controller uses a phase-locked loop technology to track the harmonic phase of the server bus in real time and generate a reverse compensation electromotive force waveform. When the 5th harmonic is detected as the dominant component, the module accurately offsets the phase of the back electromotive force by 180 degrees to achieve cancellation.

[0031] This method establishes a dynamic phase compensation mechanism based on the mechanical energy stored in the flywheel and the fluctuations of the computing power load. In the load prediction stage, a spatio-temporal convolutional model captures the coupling relationship between the power grid, servers, and the environment; in the optimization control stage, a quantum algorithm is used to balance equipment losses and computing efficiency; finally, real-time harmonic suppression is achieved through the active modulation of the back electromotive force. Its beneficial effects are reflected in three aspects: First, the phase matching technology between the flywheel energy storage release curve and the load peak and valley reduces the impact on the power grid; second, the quantum optimization algorithm breaks through the convergence speed limit of traditional multi-objective optimization; finally, the active harmonic compensation based on electromagnetic induction avoids the energy loss of passive filters and is especially suitable for high-density computing clusters.

[0032] Optionally, the prediction and modeling module includes: An input parameter acquisition unit for obtaining temperature data of the server room, air pressure monitoring data of the flywheel energy storage cavity, the power quality characteristic map, and a preset set of constraint conditions, and constructing a multi-dimensional input parameter set, where the set of constraint conditions includes a temperature rise gradient threshold, an air pressure safety threshold, and a safe rotation speed threshold; Specifically, this unit provides a multi-dimensional input data source for the prediction model. The temperature data of the server room is a direct reflection of the rack heat dissipation efficiency, and the temperature rise gradient of each area is collected every 30 seconds through a distributed thermocouple array; the air pressure monitoring data of the flywheel energy storage cavity is a key parameter to ensure the aerobic sealing of the energy storage, and a differential pressure sensor is used to measure the airtight state of the cavity in real time with an accuracy of 0.1 kPa; the power quality characteristic map is a fusion matrix of current harmonic spectra and vibration data, which is used to quantify the electromagnetic-mechanical coupling effect during equipment operation; the preset set of constraint conditions includes a temperature rise gradient threshold of the maximum allowable temperature rise rate of the facility heat dissipation system, such as ≤1.5 °C / min, an air pressure safety threshold of the cavity pressure holding limit value, such as 200 kPa, and a safe rotation speed threshold of the critical value of the stress bearing of the flywheel rotor material, such as 45,000 rpm.

[0033] A prediction model construction unit for establishing a spatio-temporal associated convolutional neural network model and fusing and processing the multi-dimensional input parameter set with the real-time frequency modulation data of the power grid; Specifically, the spatio-temporal associated convolutional neural network model adopts a three-dimensional convolutional kernel structure: the time dimension window corresponds to the power grid frequency modulation rate, for example, a 5-minute granularity, the space dimension corresponds to the topological layout of the server racks, such as the cabinet row and column coordinates), and the feature dimension fuses the power spectrum, temperature, and pressure data. The real-time frequency modulation data of the power grid is obtained through a synchronized phasor measurement unit (PMU) to obtain the dynamic values of voltage and frequency, and is spliced with the multi-dimensional input parameter set into a fourth-order tensor for input. The design of the convolutional kernel parameters of the model follows the principle of spatio-temporal feature decoupling, that is, the first two layers of convolution focus on extracting the periodic law of load fluctuations, and the last three layers of convolution capture the heat conduction correlation characteristics between server groups.

[0034] A data processing unit for dynamically segmenting the fused data by applying a sliding time window algorithm to generate a spatio-temporal feature tensor containing spatial correlation, where the spatio-temporal feature tensor includes the spatially correlated server energy consumption distribution and the load fluctuation trend; A computing power demand output unit for calibrating the temporal deviation component in the spatio-temporal feature tensor by using a long short-term memory network and outputting a corrected dynamic distribution matrix of computing power demand.

[0035] Specifically, the spatio-temporal feature tensor is a three-dimensional matrix, and its dimensions include a time dimension, a space dimension, and a feature dimension; the time dimension is a 60-minute sliding time window, segmented at a 5-minute step, with a total of 12 sub-windows; the space dimension is that the server computer room is divided into grids, each grid is 2m×2m, and the real-time power consumption of each grid is recorded; the feature dimension contains three parameters: current harmonic distortion rate, vibration RMS value, and temperature rise value. The current harmonic distortion rate is the proportion of total harmonic distortion, the vibration RMS value is the effective value intensity of vibration pulses, and the temperature rise value is the change in the center of the grid relative to the reference temperature; the number of input layer nodes of the long short-term memory network is 36, 12 time slices × 3 features, and the number of hidden layer nodes is 24.

[0036] Specifically, the prediction and modeling module realizes high-precision dynamic prediction of computing power demand through hierarchical feature extraction of spatio-temporal convolution and temporal calibration of LSTM. The multi-dimensional data fusion mechanism breaks through the limitations of single-sensor data. Especially in high-density server clusters, it can accurately capture the load coupling phenomenon caused by heat conduction and electromagnetic interference; the dynamic segmentation of the sliding window ensures the sensitive response of the model to transient events.

[0037] Optionally, the optimization control module includes: An optimization function establishment unit for constructing a multi-objective optimization function including energy storage capacity constraints, heat dissipation efficiency equations, and power supply cost models. For the multi-objective optimization function , there is ; In the formula, is the decoupling coefficient, is the weight coefficient of flywheel mechanical loss, is the weight coefficient of task delay time, is the flywheel mechanical loss power, is the flywheel mechanical rated power, is the task delay time, is the maximum task time; Specifically, the multi-objective optimization function is used to minimize the comprehensive cost of mechanical loss and task delay while preventing suboptimal solutions caused by their coupling. For the energy storage capacity constraint, the ratio of the mechanical loss power of the flywheel to its rated power directly reflects the energy efficiency level of the flywheel energy storage system; when the mechanical loss power of the flywheel approaches or exceeds its mechanical rated power, it means that the energy storage capacity of the flywheel energy storage system is about to be exhausted or cannot effectively support the current load; therefore, minimizing this ratio in the optimization function actually ensures that the flywheel energy storage system will not exceed its energy storage capacity limit due to overload.

[0038] Among them, the heat dissipation efficiency parameter cross-influences the quantitative calculation of the mechanical loss power of the flywheel. For the heat dissipation efficiency equation, there is ; In the formula, is the heat dissipation efficiency parameter, is the mechanical friction coefficient, is the angular velocity of the flywheel, is the thermal conductivity, is the temperature difference of the cavity. In the high-temperature mode, the decrease of the heat dissipation efficiency parameter leads to an increase in the mechanical loss power of the flywheel, forcing the optimization result to tend to the low-speed scheme, such as preferentially releasing the energy of the small flywheel group.

[0039] For the power supply cost model, an explicit economic penalty model is established between the task delay time and the power supply cost. In the formula, the ratio of the task delay time to the maximum task time is the normalized delay ratio, and the exponential parameter ( ) of the coupling term actually implies the non-linear effect of the power supply cost. For example, when the electricity price is at a peak, set to focus on the delay time index to reduce additional costs.

[0040] Among them, the weight coefficient of the flywheel mechanical loss and the weight coefficient of the task delay time are dynamically adjusted based on the loss-delay ratio of historical tasks. For the weight coefficient of the flywheel mechanical loss, there is ; and it satisfies ; In the formula, is the statistical period, with a default of 30 days.

[0041] The solution set calculation unit calls the quantum-inspired annealing algorithm to solve the multi-objective optimization function, calculates the optimization balance point of the flywheel mechanical loss power and the task delay time, and outputs the set of Pareto-optimal solutions; Specifically, the quantum-inspired annealing algorithm uses the probability of quantum tunneling effect to determine whether to accept a new solution. Among them, the probability of accepting a new solution depends on temperature, breaking through the local optimum limit of traditional annealing. The probability formula is ; In the formula, is the energy difference, is the Boltzmann constant, is the operating temperature of the superconducting coil, which is obtained by back-calculating the air pressure monitoring data, breaking through the trap of local optimum solutions.

[0042] Among them, for the energy difference , there is ; In the formula, is the optimization target value of the current solution, is the objective function value of the candidate new solution generated by the algorithm; when the energy difference is greater than 0, it means that the new solution is worse and the cost is higher, and it needs to be accepted with a certain probability; when the energy difference is less than or equal to 0, it means that the new solution is better, the cost is lower or the same, and it is forced to be accepted.

[0043] Exemplarily, the current solution is , the target value ; the candidate solution after quantum perturbation is , calculate . The calculated energy difference is , indicating optimization improvement. At this time, calculate the acceptance probability is 1, indicating that the new solution must be accepted. Therefore, use the new solution to replace the current solution, then . The optimization target value of the current solution and the objective function value of the candidate new solution are the key comparison parameters for the quantum annealing algorithm to search for the global optimum solution. By dynamically accepting inferior solutions, it avoids falling into local optima and improves the optimization robustness of multi-constraint complex systems; in the high-dimensional parameter space of the flywheel energy storage system, this method efficiently balances the collaborative optimization of loss, delay, and cost.

[0044] The solution set screening unit is used to input the set of frontier solutions into a preset quantum genetic algorithm, and perform iterative optimization based on the dynamic distribution matrix of computing power requirements to generate an optimized solution vector for energy scheduling; Specifically, the frontier solution set is a set composed of a series of non-dominated solutions. In the solution set, if a solution is not inferior to other solutions in all optimization objectives and is strictly better in at least one objective, then this solution is a "non-dominated solution". For example, assume two objectives: minimizing the mechanical loss power of the flywheel and minimizing the task delay time. Solution A has a flywheel mechanical loss power of 150 kW and a task delay time of 30 s; Solution B has a flywheel mechanical loss power of 160 kW and a task delay time of 25 s. If there is no Solution C that satisfies a flywheel mechanical loss power less than or equal to 150 and a task delay time less than or equal to 25, then both Solution A and Solution B are members of the frontier solution set.

[0045] The control strategy generation unit is used to map the energy scheduling optimization solution vector to physical control instructions, generate the flywheel speed regulation curve and the server power allocation sequence, and obtain the dynamic coordination control strategy.

[0046] Specifically, convert the abstract solution vector into physical control instructions. For example, the energy scheduling optimization solution vector is , the corresponding required power is 155 kW, calculate the target speed according to the flywheel kinetic energy formula to achieve the control of the flywheel; allocate 40% of the load to server cluster A and 60% to B, corresponding to the priority topology, such as assigning high priority to critical tasks to reduce the task delay time.

[0047] Exemplarily, an IDC center encounters a sudden surge in instantaneous computing power demand and needs to suppress the power grid impact and ensure task delay. The data acquisition module monitors that the 5th harmonic amplitude is 20 A and the RMS value of the rack vibration is 2.8 mm / s, and generates an electrical energy quality characteristic map , highlighting the abnormal area in the 5th row and 5th column. Then, the spatio-temporal convolutional model of the prediction modeling module inputs the power grid load, the computer room temperature, and the map data, and outputs the computing power demand matrix for the next 15 minutes. The computer room temperature is 10 °C, and the peak phase of the computing power demand matrix reaches 850 kW at t = 8 min. In the optimization control module, the solution set calculation unit calls the quantum annealing algorithm to generate a solution set ( ), and the solution set screening unit uses the quantum genetic algorithm to screen the solution with the lowest loss as ; the control strategy generation unit generates control instructions, that is, the flywheel group outputs 780 kW for 8 seconds, and the server allocates tasks according to the priority, and the high-priority task delay ≤ 5 s. The power execution module adjusts the flywheel PWM duty cycle to 65% to match the 5th harmonic phase, and the back electromotive force phase shifts by 180°. At the same time, the liquid metal heat pump speeds up to 20 L / min to offset the braking heat. The dynamic coordination control strategy generates the flywheel energy release rate constraint and calculates the task priority topology through the quantum optimization algorithm, so that the kinetic energy release curve of the flywheel group forms a compensation effect with the phase of the load fluctuation. As Figure 3As shown, within a 24-hour cycle, the output curve of the flywheel and the computing power load exhibit a dynamic inverse correlation coefficient, and the fluctuation amplitude of the grid demand is significantly reduced, verifying the effectiveness of the phase compensation mechanism.

[0048] Optionally, the optimization process of the quantum genetic algorithm includes: Design a hybrid coding mechanism based on quantum phase rotation parameters to correlate the stress monitoring data of the flywheel rotor with the temperature of the server chip, where the stress monitoring data is extracted from the vibration data of the power quality characteristic map; Specifically, the design of the quantum phase rotation parameters includes decomposing the stress monitoring data into high-frequency vibration components and low-frequency deformation components through frequency-domain wavelet transform; normalizing the amplitude of the high-frequency components; for the initial angle of the quantum phase , there is ; In the formula, is the energy of the high-frequency component of the stress data, is the energy of the low-frequency component of the stress data; the corresponding quantum state is: ; Among them, is the phase offset mapped by the temperature of the server chip, , is the temperature of the server chip, in °C.

[0049] Introduce a dynamic penalty factor weight, and adjust the weight according to the flywheel bearing wear coefficient of the previous cycle, where the bearing wear coefficient is calculated based on the air pressure monitoring data; Specifically, input the air pressure monitoring data of the flywheel cavity, establish a linear model of the air pressure change rate and bearing wear, assuming that the air pressure drops per unit time, then the bearing wear coefficient satisfies: ; In the formula, is the experimental calibration coefficient, is the initial wear value. For the dynamic penalty factor weight , there is ; In the formula, is the bearing wear coefficient, and the bearing wear coefficient usually ranges from 0 to 100; when increases, it represents that the bearing wear is aggravated, the penalty weight decays, and the algorithm is allowed to relax the constraint on the loss to avoid suboptimal solutions; for example, if , it represents medium wear, then the dynamic penalty factor weight , indicating that more attention is paid to task delay rather than flywheel loss during optimization.

[0050] By amplifying the amplitude gradient distribution of the quantum state, combining the heat dissipation efficiency equation to generate a constraint boundary, and using Monte Carlo sampling to screen for an optimal solution vector of energy scheduling whose energy consumption difference meets a preset difference.

[0051] Specifically, by calculating the amplitude gradient distribution of the quantum state, identifying the parameter sensitivity direction in the solution space, such as the sensitivity degree of high-frequency vibration to temperature change, for the amplitude gradient of the amplitude gradient distribution of the quantum state, there is : ; Here is the imaginary unit; through the heat dissipation efficiency equation, the equation is converted into a constraint condition, for example, setting ; then using Monte Carlo sampling to screen, that is, generating random points in the solution space of the flywheel speed and task delay time, and retaining the solutions that satisfy the heat dissipation equation and the penalty condition; calculating the energy consumption difference between the candidate solution and the current optimal solution, and selecting an optimal solution vector of energy scheduling that satisfies the energy consumption difference less than or equal to the preset threshold.

[0052] Exemplarily, as Figure 4 shown, a certain data center suddenly encounters a sharp increase in computing power demand, and an optimal solution needs to be generated when the flywheel wear intensifies and the chip temperature rises. The bearing wear coefficient , the server chip temperature . After the stress data is wavelet decomposed, is obtained, , calculating the initial angle , the chip temperature causes a phase shift of , the quantum state is encoded as , emphasizing the coupling effect of medium and high frequency vibrations and high temperature. At this time, the weight of the dynamic penalty factor is , and the algorithm will reduce the attention to flywheel loss and give priority to ensuring task delay. At this time, Monte Carlo generates candidate solutions, solution A is ; solution B is ; after verification through constraints, for solution A, , it just passes the critical value, but because , the attention to flywheel loss is reduced, so the solution A with lower delay is retained. Finally, solution A is selected, and the flywheel runs at high speed to quickly suppress the load, sacrificing part of the loss in exchange for task response. By integrating flywheel stress and temperature parameters through quantum hybrid coding, dynamically adjusting the optimization weight to cope with equipment wear, and using Monte Carlo to screen for efficient solutions under heat dissipation constraints. In the example, sacrificing a certain amount of loss to achieve low-latency response reflects the fine balance between computing power load and energy efficiency.

[0053] Optionally, the execution of the dynamic coordination control strategy includes: The pulse-width modulation signal for controlling the electromagnetic eddy current brake converts the kinetic energy of the flywheel into a controllable back electromotive force, where the phase of the controllable back electromotive force is determined by the energy storage capacity constraint; Specifically, a SiC MOSFET module is used to generate a variable-duty-cycle PWM signal, and the duty-cycle adjustment step is less than or equal to 1%; the phase of the voltage harmonics of the server bus is tracked in real time through a phase-locked loop (PLL) to generate a synchronous compensation signal; based on the detection of the actual output of the back electromotive force by a Hall current sensor, the amplitude error of the PWM signal is corrected.

[0054] Among them, the duty-cycle adjustment step of the pulse-width modulation signal satisfies: ; In the formula, , are PI control parameters, and the duty-cycle update period is less than or equal to 10 ms.

[0055] Based on the heat dissipation efficiency equation, adjust the flow rate parameter of the liquid metal heat dissipation loop to synchronously match the braking heat distribution of the flywheel group; Specifically, based on the heat dissipation efficiency equation, calculate the instantaneous heat generation. For the instantaneous heat generation , there is ; Adopt a PID controller, combine the instantaneous heat generation, and adjust the pump speed according to the temperature sensor data, such as 20–200 L / min; when the cavity temperature difference increases, such as ≥10°C, the flow rate is automatically increased to a preset threshold, such as 150 L / min, to ensure that the heat dissipation efficiency satisfies: ; The heat dissipation efficiency equation directly affects the calculation of the mechanical loss of the flywheel. The flow rate adjustment is dynamically bound to the flywheel speed. For example, when the flywheel runs at 40000 rpm, the heat generation , by increasing the flow rate of the liquid metal to 120 L / min, is stabilized within 5°C.

[0056] Couple the back electromotive force with the harmonic components of the server power supply bus, and perform dynamic voltage compensation through a three-phase transformer. The harmonic components are associated with the current harmonic spectrum in the power quality characteristic map.

[0057] Specifically, inject the back electromotive force into the power supply bus through a multi-winding three-phase transformer to compensate for the harmonic voltage; among them, adjust the transformer turns ratio to match the load impedance, that is, the equivalent load impedance . The vector control technology is adopted to adjust the amplitude and phase of the back electromotive force in real time, and the compensation error rate is ≤ 2%. If the resonance risk is detected and the harmonic distortion rate ≥ 8%, the fast energy storage inductor is triggered to absorb the transient current pulse. The harmonic spectra in the power quality characteristic map, such as the 5th and 7th harmonics, directly generate waveform matching templates to guide the compensation strategy. For example, for the 5th harmonic, such as 20 A, the output voltage of the transformer is 50 V, and the phase is 180°. After canceling the harmonic, the bus distortion rate drops from 8% to 3%.

[0058] Exemplarily, a data center suddenly encounters a sharp increase in computing power demand, with a peak load of 850 kW, resulting in the 5th harmonic exceeding the standard on the server bus, with an amplitude of 30 A, and local overheating, with a temperature of 12 °C. The power quality characteristic map detects the 5th harmonic and vibration anomalies, that is = 0.7; the prediction modeling module outputs the computing power demand matrix for the next 10 minutes, with the peak at t = 5 min; the optimization control module uses the quantum annealing algorithm to calculate the optimal solution as the flywheel output power of 780 kW, the delay time ≤ 8 s, and the liquid metal flow rate of 180 L / min; the dynamic coordination control strategy generates a PWM duty cycle of 70%, the flow rate parameter, and a safety speed threshold of 42,000 rpm. The power execution module controls the electromagnetic brake according to the dynamic coordination control strategy to generate a back electromotive force to cancel the 5th harmonic, with the back electromotive force phase of 180° and amplitude of 75 V; the liquid metal pump speeds up to 180 L / min, and controls the cavity temperature difference within 8 °C; the three-phase transformer injects a compensation voltage to reduce the bus harmonic distortion rate from 10% to 2.5% and reduce the energy consumption by 18%. The flywheel energy storage system levels the load peak, reduces the server task delay by 30%, significantly improves the power quality, and maintains the heat dissipation efficiency at 92%. Through dynamic harmonic compensation, thermal energy collaborative management, and electromagnetic coupling control, the computing power load is leveled and energy is saved.

[0059] Optionally, the modulation process of the back electromotive force includes: Detect the harmonic distortion rate of the server power bus, generate a real-time waveform matching template based on the current harmonic spectrum, and calculate the equivalent load impedance using the waveform matching template; Specifically, a high-speed power quality analyzer, such as Fluke 435-II, is used to collect the server bus voltage and current signals in real time, with a frequency resolution less than or equal to 1 Hz; the harmonic components are decomposed by the fast Fourier transform (FFT) to calculate the total harmonic distortion rate. For the total harmonic distortion rate , there is ; In the formula, is the fundamental wave current; extract the harmonic dominant components, such as the 5th, 7th, and 11th harmonics, generate the reference waveform of the compensation signal, and dynamically adjust the template amplitude to meet the phase matching condition: ; In the formula, is the correction coefficient, is the th harmonic corresponding equivalent load impedance, is the harmonic angular frequency, the th harmonic phase offset. Absorb the excessive components of the transient current pulse through the fast-response energy storage inductor, combine with the said equivalent load impedance, calculate the overload energy characteristic parameter, and record it into the power quality characteristic map; Specifically, use a fast-response energy storage inductor, and the saturation current satisfies: ; and the inductance value is 200 μH ± 5; the inductor quickly absorbs transient energy through a core reset circuit, such as a flyback topology, and the time constant is less than or equal to 10 μs; the overload energy characteristic parameter is extracted based on the peak value of the inductor current and the equivalent impedance, where the overload energy is ; In the formula, is the inductance value of the fast-response energy storage inductor, is the peak current of the transient current, is the rms value of the th harmonic current, is the th harmonic corresponding equivalent load impedance, is the harmonic duration; the overload energy threshold is set to 20% of the flywheel's rated energy storage. For example, in a 200 kWh energy storage system, the upper limit of the energy storage capacity constraint is 40 kWh.

[0060] Analyze the overload energy characteristic parameter according to the amplitude gradient distribution of the quantum state to adjust the excitation parameter of the flywheel electromagnet and output the back electromotive force.

[0061] Specifically, the excitation current of the flywheel electromagnet is adjusted in the gradient direction. For the excitation current , there is ; In the formula, is the rated excitation current, is the gain coefficient, is the overload energy, is the amplitude gradient; therefore, for the back electromotive force , there is ; In the formula, is the motor constant, Determined by waveform matching template phase; the quantum state encoding integrates the overload energy and device states, such as flywheel stress and temperature, and dynamically optimizes the excitation strategy. For example, an overload event triggers the excitation current to change from 120A to 138A, and the back electromotive force amplitude increases to 85V.

[0062] Exemplarily, when the data center encounters a distributed denial-of-service attack, the computing power load soars instantaneously by 120%, resulting in the harmonic distortion rate of the server bus rising to 12%, with the 7th harmonic being the main one, triggering an emergency response of the flywheel energy storage system. The 7th harmonic distortion rate is detected to be 7.5%, and the total harmonic distortion rate , a compensation template is generated. The compensation template has an amplitude of 28A, a phase of 180°, and an equivalent load impedance of 0.6Ω; the fast-response inductor absorbs transient pulses. The inductance value of the fast-response inductor is 200μH, , and the overload energy is calculated ; the quantum algorithm generates an amplitude gradient based on the overload energy , increases the excitation current to 145A, and outputs the back electromotive force . Through the three-stage closed-loop control of active harmonic suppression, transient energy absorption, and quantum-optimized excitation, dynamic energy management and power quality improvement under sudden computing power loads are realized. Its core is to suppress transient impacts through a fast inductor, break through the limitations of traditional PI control using a quantum algorithm, and finally achieve the collaborative optimization of computing power load and flywheel response through precise modulation of the back electromotive force.

[0063] Optionally, after generating the power quality feature map, it further includes: Identifying abnormal harmonic components in the power quality feature map, and combining with the spatio-temporal feature tensor to construct an energy consumption feature encoding tree for server devices; Specifically, a high-frequency broadband power quality analyzer is used to collect the real-time current signal of the server bus, decompose each harmonic component through fast Fourier transform, and calculate the total harmonic distortion rate; the harmonic amplitude and vibration data are fused into a power quality feature map matrix, and the threshold determination method is used to identify abnormal regions. For example, when the matrix element corresponding to a certain harmonic frequency point exceeds a preset threshold, such as is greater than 0.5, it is determined that this frequency point is an abnormal harmonic component; a hash mapping table is established to associate the abnormal harmonic frequency with the physical coordinates of the server cabinet. For example, the 5th harmonic anomaly corresponds to the third row of cabinets in the computer room and is weighted through a weight formula; for the node weights of the encoding tree, there is ; In the formula, is the effective value of the th harmonic current, is the The equivalent load impedance corresponding to the subharmonic is the harmonic duration; Based on the node weight distribution of the encoding tree, locate the physical server cluster coordinates corresponding to the high-energy-consuming computing processes; Specifically, adopt the ultra-wideband (UWB) positioning technology with a positioning accuracy of ±0.5 m, map the node weights of the encoding tree to the physical coordinates of the server rack, such as the node with the largest weight corresponding to the cabinet number B4; combine the server chip temperature data collected by distributed thermocouples with the vibration abnormal area to construct a three-dimensional thermal map, which includes temperature, vibration intensity, and harmonic intensity, and determine the high-energy-consuming core area, such as the area where the chip temperature is greater than or equal to 85 °C and the vibration RMS value is greater than or equal to 3 mm / s.

[0064] Map the coordinate information to the load gradient parameter of the multi-dimensional input parameter set, input it into the multi-dimensional spatio-temporal correlation prediction model, and generate the weight coefficient for adjusting the task delay time based on the energy efficiency optimization instruction of fuzzy control.

[0065] Specifically, convert the physical coordinates into a computer room grid, such as each 2 m × 2 m as a grid, extract the load gradient parameters within the grid, including real-time power consumption, harmonic distortion rate, and temperature gradient; use a 60-minute sliding window with a step size of 5 minutes to splice the load gradient parameters to generate a three-dimensional tensor, where the three-dimensional tensor is the time dimension × space dimension × feature dimension, for example, the tensor dimension = (12 time slices, 25 grids, 3 features).

[0066] Specifically, define the membership function of the input variable for fuzzy control, and the input variables include load gradient, temperature, and harmonic intensity; define the membership function of the output variable, and the output variables include task delay tolerance and flywheel speed adjustment amount; the fuzzy rule can be "Rule 1: If the harmonic intensity is 'high' and the temperature gradient is 'rapidly rising', then the task delay penalty coefficient is reduced by 30%."; call the quantum genetic algorithm of the solution set screening unit to iteratively optimize the candidate solutions output by the fuzzy rules, and use Monte Carlo sampling to screen the solution vectors that meet the constraints, where the constraint can be that the flywheel speed is less than or equal to 42,000 rpm. Map the optimal solution to the physical control instruction, that is, the energy efficiency optimization instruction. For example, the energy efficiency optimization instruction can be "The speed of flywheel group B3 is increased to 38,000 rpm and maintained for 8 seconds; 70% of the load is allocated to server cluster A, and the delay upper limit is set to 10 seconds."

[0067] Exemplarily, the data acquisition module detects that the 5th harmonic current suddenly increases to 25 A, where the normal value is less than 15 A, and the vibration RMS value reaches 4.2 mm / s, and generates an electrical power quality characteristic map = 0.78, and the threshold is 0.5. The encoding tree shows that the abnormal node has the highest weight. UWB positioning locates the 4th cabinet in row B of the computer room, where the temperature in this area reaches 88°C and the vibration is abnormal. The load gradient parameters of the 4th cabinet in row B of the computer room are power consumption of 120 kW, harmonic distortion rate of 8%, and temperature rise rate of 2°C / min. After inputting into the multi-dimensional spatio-temporal model, the predicted peak computing power demand in the next 5 minutes is 900 kW. The fuzzy logic output reduces the weight of the task delay penalty, and the weight coefficient of the task delay time decreases from 0.5 to 0.3, allowing a moderate increase in flywheel loss. The quantum genetic algorithm selects the optimal solution as the flywheel group output of 800 kW and the delay time of 9 seconds. The power execution module adjusts the excitation current of the flywheel group of the 4th cabinet in row B of the computer room to 135 A, and the phase shift of the back electromotive force is 180°, canceling the 5th harmonic. The task delay is reduced from 15 seconds to 8 seconds, and the system energy efficiency is improved by 22%. By deeply integrating the dynamic response of the flywheel energy storage system with the spatio-temporal coupling characteristics of the computing power load, the limitations of traditional fixed-threshold control are broken through, and the collaborative optimization of energy efficiency and computing performance is achieved.

[0068] Optionally, the system further includes a safety monitoring module, and the safety monitoring module is used for: Collecting the residual strain distribution data of the flywheel rotor and reconstructing the three-dimensional stress field distribution cloud map according to the stress monitoring data; Specifically, a distributed fiber optic grating strain sensor array, such as the HBM K-OR type, is embedded in the flywheel composite layer at a spacing of 5 mm; the residual strain in the circumferential and radial directions of the rotor is collected in real time at a frequency of 10 kHz, and the resolution is less than 1 με; local stress concentration points are identified through the strain mutation algorithm, and a threshold is set to trigger an alarm, for example, the strain gradient is greater than 50 με / mm. The strain mutation algorithm can be the sliding window standard deviation method; a multi-scale model of the flywheel rotor is established based on the finite element method, and the stress distribution is inversed by combining the measured strain data. The input material parameters are the elastic modulus, Poisson's ratio, and thermal expansion coefficient of the composite material, and the boundary conditions are the flywheel speed, temperature gradient, and centrifugal force distribution. The inversion result is mapped into a three-dimensional color-coded cloud map, such as red for the high stress area greater than 80% of the fatigue limit; green for the safe area, stored in the VTK (Visualization Toolkit) format, and supporting visualization by the ANSYS or COMSOL post-processing module; among them, for the actual stress , there is ; In the formula, is the elastic modulus of the material, is the residual strain, is the thermal expansion coefficient of the material, is the difference between the current ambient temperature and the reference temperature, such as the temperature rise during flywheel operation .

[0069] Input the stress field distribution nephogram into a preset phase change material life prediction model to calculate the fatigue limit of the material. Specifically, the input data of the phase change material life prediction model are the node stress values of the stress field nephogram. The maximum stress and average stress are extracted for each node according to its position. The input data also include the microscopic crack parameters detected by an ultrasonic guided wave array with a sampling rate of 100 MHz. The phase change material life prediction model includes a macroscopic layer, a mesoscopic layer, and a microscopic layer. The macroscopic layer is based on the stress-life (S-N) curve fitting, and the Manson-Coffin equation is used for the material. The mesoscopic layer is based on the lattice Boltzmann method (LBM) to simulate the crack propagation path. The microscopic layer is a fatigue damage evolution model calibrated by a scanning electron microscope (SEM). Through life confidence calculation, the lower limit of the confidence interval is calculated, which is the fatigue limit of the material.

[0070] Calculate the safety speed threshold for the next cycle according to the energy storage capacity constraint and the fatigue limit of the material, and update the set of constraint conditions.

[0071] Specifically, for the safety speed threshold , there is ; In the formula, is the width of the confidence interval, is the lower limit of the confidence interval, that is, the fatigue limit of the material, is the moment of inertia of the flywheel, is the upper limit of the energy storage capacity constraint; the moment of inertia of the flywheel is obtained through experiments. The experimental method is to record the driving torque and angular acceleration at a sampling frequency of 1000 Hz when the flywheel is idling and accelerating to 50% of the rated speed, and calculate it through the formula. The calculation of the moment of inertia covers the average value of 3 independent experiments. For the moment of inertia of the flywheel , there is ; In the formula, is the driving torque, is the angular acceleration; the value of the energy storage capacity constraint is based on the fact that under a 20% redundancy, there is ; In the formula, is the rated energy storage capacity, is the energy storage efficiency, is greater than 92%.

[0072] Optionally, the update method of the phase change material life prediction model includes: Detect the microscopic crack distribution of the flywheel composite material, obtain the damage evolution direction through an ultrasonic guided wave array, and synchronize the data to the stress field distribution nephogram. Specifically, a piezoelectric ceramic sensor array is embedded in the flywheel composite material, such as with a frequency of 100 MHz and a sampling rate of 1 kHz. By exciting low-frequency ultrasonic guided waves such as Lamb waves, the internal reflection signals of the material are detected. When the guided wave encounters a crack, the waveform amplitude and phase are distorted, and when the reflection coefficient is greater than 0.7, it is determined as a crack. The crack position is mapped to the stress field distribution cloud map, and the centrifugal force and thermal stress field data are superimposed to generate a four-dimensional dynamic damage evolution map, where the four-dimensional dynamic damage evolution map is three-dimensional in space + time series.

[0073] Based on the residual strain distribution data, a multi-scale finite element model is established, correlating the stress concentration coefficient with the cycle life of the energy storage unit; Specifically, a three-dimensional overall model of the flywheel is established at the macroscopic layer, with a mesh size less than 5 mm, and the boundary conditions are rotational speed and temperature gradient; at the mesoscopic layer, the local grid is refined, with a key area of 0.1 mm, and the lattice Boltzmann method is used to simulate the crack propagation path at the micro-nano scale; at the microscopic layer, the stress field at the crack tip is fitted through SEM electron microscope data, and when the J integral is greater than 100 N / mm², accelerated fatigue is triggered; the stress concentration coefficient is inserted into the S-N curve to match the on-site load cycle number database.

[0074] Calculate the reliability index for different load scenarios according to the heat dissipation efficiency equation, and generate a confidence interval using a preset modified Weibull distribution model; Obtain the fatigue limit of the material based on the confidence interval.

[0075] Specifically, the rotational speed of the flywheel is collected based on the heat dissipation efficiency equation , the temperature difference between the cavities , the heat dissipation efficiency parameter , the task duration , the mechanical loss power , the instantaneous heat generation , the actual temperature of the material , the actual temperature of the material is monitored in real time through a distributed thermocouple. The heat generation is converted into the impact of the heat load on the material through the Arrhenius correction model, and the equivalent temperature stress is calculated , there is ; In the formula, is the heat capacity of the material, is the mass of the flywheel; the failure rate per unit time is defined according to the Weibull distribution, and the life index is generated in combination with the heat dissipation efficiency, that is, the reliability index , for the reliability index , there is ; In the formula, is the material damage index, is a thermodynamic constant, is a shape parameter, is the characteristic life; the shape parameter and the characteristic life are dynamically adjusted according to the heat dissipation efficiency. For the shape parameter, there is ; In the formula, is the initial value of the shape parameter, is the correction coefficient; For the life characteristic, there is ; In the formula, is the initial value of the life characteristic, is the mechanical loss power under rated operating conditions, for example, it can be ; Monte Carlo simulation generates sub - samples, and probability distribution modeling is performed on , , to be normal or log - normal; the simulation data is fitted by maximum likelihood estimation, and the lower limit of the confidence interval is calculated, and there is ; In the formula, is the Gamma function, covering a 95% survival probability.

[0076] Exemplarily, when the heat dissipation efficiency drops to , the flow rate is insufficient, the cavity temperature difference is , , the mechanical loss power is , , by correcting the Weibull parameters, we get , , then the lower limit of the confidence interval can be calculated as , so the updated safety speed threshold is obtained as , with the original value being 45000 rpm. Through the dynamic correction of the Weibull parameters and the generation of the confidence interval, the system can still ensure reliability under high loads. When the heat dissipation efficiency drops, it automatically triggers the degradation of the life prediction model, adjusts the flywheel output strategy, and the lower confidence limit is directly fed back to the optimization control module to avoid over - limit operation.

[0077] Based on the same inventive concept, as shown in Figure 5 , the present invention also provides a method for suppressing computing power load and energy - saving management based on a flywheel energy storage system, and the method includes: Collecting the real - time current harmonic spectrum and server rack vibration data of the computing cluster to generate an electrical energy quality characteristic map; Build a prediction model based on multi-dimensional spatio-temporal association, perform spatio-temporal convolution operations on the power quality feature map, grid load fluctuation data, and historical power consumption curve, and output a dynamic distribution matrix of computing power requirements; Analyze the peak load phase in the dynamic distribution matrix of computing power requirements, and extract the rotational speed fluctuation characteristics of the flywheel energy storage unit and the energy consumption time series correlation parameters of the server rack; Generate a dynamic coordination control strategy based on quantum-inspired optimization, and the control strategy includes the flywheel energy release rate constraint and the computing task scheduling priority topology; Execute the dynamic coordination strategy to make the kinetic energy release curve of the flywheel group form a phase compensation effect with the computing load fluctuation.

[0078] It should be noted that the electrical connections between the above-mentioned various units do not necessarily represent direct connections of the lines. Indirect connection methods, as long as they achieve the purpose of the present invention, can be applied to the embodiments of the present invention. The above are only exemplary embodiments of the present invention, and the scope of the present invention cannot be limited thereby.

[0079] That is, any equivalent changes and modifications made in accordance with the teachings of the present invention still fall within the scope covered by the present invention. After considering the specification and the disclosure of the practical truth, those skilled in the art will easily think of other implementation schemes of the present invention. This application aims to cover any variations, uses, or adaptive changes of the present invention, and these variations, uses, or adaptive changes follow the general principles of the present invention and include common general knowledge or conventional technical means in the technical field not recorded in the present invention.

Claims

1. A computing load balancing and energy-saving management system based on a flywheel energy storage system, characterized in that: The system comprises: The data acquisition module is used to collect the real-time current harmonic spectrum of the computing cluster and the vibration data of the server rack to generate the power quality characteristic map; A prediction modeling module is used to establish a prediction model based on multi-dimensional spatiotemporal association, perform spatiotemporal convolution operations on the power quality characteristic map, grid load fluctuation data, and historical power consumption curves, and output a dynamic distribution matrix of computing power demand; An optimization control module is used to analyze the peak load phase in the computing power demand dynamic distribution matrix, extract the speed fluctuation characteristics of the flywheel energy storage unit and the energy consumption timing correlation parameters of the server rack, and generate a dynamic coordination control strategy; The power execution module is used to execute the dynamic coordination control strategy and adjust the phase parameter of the flywheel back electromotive force to offset the voltage harmonic component of the server bus.

2. The computing load balancing and energy-saving management system based on the flywheel energy storage system according to claim 1 is characterized in that: The prediction modeling module includes: An input parameter acquisition unit is used to obtain the temperature data of the server room, the air pressure monitoring data of the flywheel energy storage cavity, the power quality characteristic map, and a preset constraint condition set to construct a multi-dimensional input parameter set, wherein the constraint condition set includes a temperature rise gradient threshold, an air pressure safety threshold, and a safety speed threshold; A prediction model building unit, used to establish a spatiotemporal convolutional neural network model, and to fuse the multi-dimensional input parameter set with the real-time frequency modulation data of the power grid; A data processing unit, configured to dynamically segment the fused data using a sliding time window algorithm to generate a spatiotemporal feature tensor containing spatial correlation, wherein the spatiotemporal feature tensor contains a spatially correlated server energy consumption distribution and a load fluctuation trend; The computing power demand output unit is used to calibrate the timing deviation component in the spatiotemporal feature tensor using a long short-term memory network, and output a corrected computing power demand dynamic distribution matrix.

3. The computing load balancing and energy-saving management system based on the flywheel energy storage system according to claim 2 is characterized in that: The optimization control module includes: The optimization function establishment unit is used to construct a multi-objective optimization function including energy storage capacity constraints, heat dissipation efficiency equations and power supply cost models. ,have ， In the formula, is the uncoupled coefficient, is the weight coefficient of flywheel mechanical loss, is the weight coefficient of task delay time, is the mechanical power loss of the flywheel, is the mechanical rated power of the flywheel, is the task delay time, is the maximum time of the task; A solution set calculation unit, calling a quantum-inspired annealing algorithm to solve the multi-objective optimization function, calculates the optimal balance point between the flywheel mechanical power loss and the task delay time, and outputs a frontier solution set; A solution set screening unit, used to input the frontier solution set into a preset quantum genetic algorithm, perform iterative optimization based on the computing power demand dynamic distribution matrix, and generate an energy scheduling optimization solution vector; The control strategy generation unit is used to map the energy scheduling optimization solution vector to the physical control instruction, generate the flywheel speed adjustment curve and the server power allocation sequence, and obtain the dynamic coordination control strategy.

4. The computing load balancing and energy-saving management system based on the flywheel energy storage system according to claim 3 is characterized in that: The quantum genetic algorithm optimization process includes: Designing a hybrid coding mechanism based on quantum phase rotation parameters to feature-associate the stress monitoring data of the flywheel rotor with the temperature of the server chip, wherein the stress monitoring data is extracted through the vibration data of the power quality feature map; Introducing a dynamic penalty factor weight, adjusting the weight according to the flywheel bearing wear coefficient of the previous cycle, wherein the bearing wear coefficient is calculated based on the air pressure monitoring data; By amplifying the amplitude gradient distribution of the quantum state and combining it with the heat dissipation efficiency equation, a constraint boundary is generated, and Monte Carlo sampling is used to screen the energy scheduling optimization solution vector whose energy consumption difference meets the preset difference.

5. The computing load balancing and energy-saving management system based on the flywheel energy storage system according to claim 4 is characterized in that: The execution of the dynamic coordination control strategy includes: Controlling a pulse width modulation signal of an electromagnetic eddy current brake to convert flywheel kinetic energy into a controllable back electromotive force, wherein the phase of the controllable back electromotive force is determined by the energy storage capacity constraint; Based on the heat dissipation efficiency equation, the flow rate parameters of the liquid metal heat dissipation circuit are adjusted to synchronously match the brake heat distribution of the flywheel assembly; The back electromotive force is coupled with the harmonic component of the server power supply bus, and dynamic voltage compensation is performed through a three-phase transformer, wherein the harmonic component is associated with the current harmonic spectrum in the power quality characteristic spectrum.

6. The computing load balancing and energy-saving management system based on the flywheel energy storage system according to claim 5 is characterized in that: The modulation process of the back electromotive force includes: Detecting the harmonic distortion rate of the server power bus, generating a real-time waveform matching template based on the current harmonic spectrum, and calculating the equivalent load impedance using the waveform matching template; By fast-responding to the energy storage inductor, the excessive component of the transient current pulse is absorbed, and the overload energy characteristic parameters are calculated in combination with the equivalent load impedance, and recorded in the power quality characteristic map; The overload energy characteristic parameters are analyzed according to the amplitude gradient distribution of the quantum state to adjust the excitation parameters of the flywheel electromagnet and output the reverse electromotive force.

7. The computing load balancing and energy-saving management system based on the flywheel energy storage system according to claim 2 is characterized in that: After generating the power quality characteristic map, the method further includes: Identify abnormal harmonic components in the power quality characteristic spectrum, and construct an energy consumption characteristic coding tree of the server device in combination with the spatiotemporal characteristic tensor; Based on the node weight distribution of the coding tree, locating the physical server cluster coordinates corresponding to the high energy consumption computing process; The coordinate information is mapped into the load gradient parameter of the multidimensional input parameter set, and input into the multidimensional spatiotemporal correlation prediction model to generate a weight coefficient for adjusting the task delay time based on the fuzzy control energy efficiency optimization instruction.

8. The computing load balancing and energy-saving management system based on the flywheel energy storage system according to claim 4 is characterized in that: The system further comprises a safety monitoring module, which is used to: Collecting residual strain distribution data of the flywheel rotor, and reconstructing a three-dimensional stress field distribution cloud map based on the stress monitoring data; Inputting the stress field distribution cloud map into a preset phase change material life prediction model to calculate the material fatigue limit; The safe rotation speed threshold of the next cycle is calculated according to the energy storage capacity constraint and the material fatigue limit, and the constraint condition set is updated.

9. The computing load balancing and energy-saving management system based on the flywheel energy storage system according to claim 8, characterized in that: The updating method of the phase change material life prediction model comprises: Detecting the micro crack distribution of the flywheel composite material, obtaining the damage evolution direction through the ultrasonic guided wave array, and synchronizing the data to the stress field distribution cloud map; Based on the residual strain distribution data, a multi-scale finite element model is established to correlate the stress concentration factor with the cycle life of the energy storage unit; The reliability index of different load scenarios is calculated according to the heat dissipation efficiency equation, and the confidence interval is generated using a preset modified Weibull distribution model; The material fatigue limit is obtained according to the confidence interval.

10. A method for managing computing load balancing and energy saving based on a flywheel energy storage system, which uses a computing load balancing and energy saving management system based on a flywheel energy storage system as described in any one of claims 1 to 9, characterized in that: The method comprises: Collect the real-time current harmonic spectrum of the computing cluster and the vibration data of the server rack to generate the power quality characteristic map; Establish a prediction model based on multi-dimensional spatiotemporal association, perform spatiotemporal convolution operations on the power quality characteristic map, grid load fluctuation data, and historical power consumption curves, and output a dynamic distribution matrix of computing power demand; Analyze the peak load phase in the computing power demand dynamic distribution matrix, and extract the speed fluctuation characteristics of the flywheel energy storage unit and the energy consumption time series correlation parameters of the server rack; generating a dynamic coordinated control strategy based on quantum-inspired optimization, the control strategy including flywheel energy release rate constraints and computing task scheduling priority topology; The dynamic coordination strategy is executed to form a phase compensation effect between the kinetic energy release curve of the flywheel group and the calculated load fluctuation.

Images