AC Access Flywheel UPS Control Method for Leveling Dynamic Load in Data Center

Abstract

The present invention discloses an AC access flywheel UPS control method for suppressing dynamic load in a data center, which relates to the field of power supply management and includes the following steps: S1: Predict and analyze the load fluctuation of the data center, collect the operation data of the flywheel through sensors, and monitor and model the state of the flywheel UPS; The present invention can perform data fitting and pattern recognition more efficiently, reduce the training process that requires a large amount of time and computing resources in traditional methods, shorten the system response time, avoid over-reliance on preset models and static rules, effectively provide more flexible strategy options, and enhance the system's ability to cope with changes; It can quickly respond to load changes, enhance the robustness and stability of the system, avoid overloading and inefficient operation of a single device, and at the same time can efficiently evaluate the state and performance of the flywheel UPS system, avoid excessive consumption or waste of energy, improve energy utilization rate, extend the equipment life, and reduce maintenance costs.

Description

Technical Field

[0001] The present invention relates to the field of power supply management, and particularly to an AC access flywheel UPS control method for suppressing dynamic load in a data center. Background Art

[0002] With the continuous development of information technology and the advancement of digital transformation, data centers have become key infrastructure to support various information services in society. As the core infrastructure to support technologies such as cloud computing, big data analysis, and artificial intelligence, data centers have become an important part of modern society. The operation and maintenance of data centers require ensuring stable and efficient power supply. Especially when facing dynamic load fluctuations, how to ensure continuous power supply has become a key issue. The surging global data traffic and large-scale computing demands have led to a year-on-year increase in the power consumption of data centers. The instability of power supply and dynamic load fluctuations have brought great challenges to the efficient operation of data centers. Traditional power supply systems often face challenges such as grid load fluctuations, uneven equipment loads, and power supply failures, resulting in problems such as service interruptions and equipment overloads in data centers. To address this challenge, more and more data centers are beginning to introduce advanced energy storage systems, especially flywheel UPS systems, to balance power demand fluctuations and ensure the stability of power supply. As a physical energy storage device, the flywheel UPS stores energy by rotating the flywheel and provides emergency power when the power demand suddenly increases, with advantages of fast response, long life, and high-efficiency energy conversion. However, in practical applications, how the flywheel UPS accurately adjusts under dynamic load fluctuations and how to achieve coordinated operation between different energy storage units remain technical problems to be solved urgently. Summary of the Invention

[0003] The object of the present invention is to address the above-mentioned problems and provide an AC access flywheel UPS control method for suppressing dynamic load in a data center.

[0004] The present invention proposes an AC access flywheel UPS control method for suppressing dynamic load in a data center. The technical solution adopted to solve the technical problem is as follows:

[0005] S1: Predict and analyze the load fluctuations of the data center, collect the operation data of the flywheel through sensors, and monitor and model the state of the flywheel UPS;

[0006] S2: Based on the predicted load fluctuations, simulate the coordinated operation between different energy storage systems to create a load regulation plan and feedback it to the staff for viewing and adjustment;

[0007] S3: Dynamically optimize the charge-discharge cycle and power adjustment strategy of the flywheel UPS according to the monitoring and modeling results and the actual load conditions collected by the sensors;

[0008] S4: Monitor the operating status of the flywheel and the changes in external load through sensors and the data acquisition system, and adaptively adjust the flywheel UPS according to the real-time monitoring data;

[0009] S5: Establish a digital twin model of the flywheel UPS, synchronize all data and control schemes of the flywheel UPS in real time, evaluate the energy conversion efficiency of the flywheel UPS, and adjust the control parameters;

[0010] S6: Collect and analyze the performance data of the flywheel UPS in real time, evaluate the effect of the current control strategy, and readjust the control strategy of the flywheel UPS according to the evaluation results.

[0011] As a further solution of the present invention, the specific steps of predicting and analyzing the load fluctuation of the data center in S1 are as follows:

[0012] P1.1: The flywheel UPS control platform collects multiple groups of load data collected by the data center, removes the noise and abnormal data in each group of collected load data, fills the missing data by interpolation method, and then through Z-score normalization processing, unifies each group of load data within a specified range. Then, use the variance screening method to extract the required feature information from each group of load data, extract the historical load data from the data center, and divide it into a training set, a test set, and a validation set, and divide the training set and the validation set into multiple training subsets and validation subsets respectively;

[0013] P1.2: Create a prediction and analysis model based on the quantum support vector machine architecture. At the beginning of each round of training, input the training subset data into the prediction and analysis model in turn. The model maps each training set data into the quantum space through the angle encoding method and generates it on the corresponding quantum bits. Then, through the quantum superposition effect, the quantum bits of multiple groups of feature information exist in a quantum state at the same time. Then, the prediction and analysis model calculates the inner product between each quantum state in the quantum space through the quantum kernel function, and evaluates the similarity between each quantum state based on the calculated inner product;

[0014] P1.3: According to the quantum kernel function values between each group of training subset data calculated, construct the corresponding kernel matrix, and construct the corresponding objective function with the purpose of maximizing the interval between each input data, set the constraint conditions of the optimization objective, and then based on the constraint conditions, solve the objective function through the gradient descent algorithm to obtain the optimized parameters of the prediction and analysis model;

[0015] P1.4: After each round of optimization of the model parameters, the validation subset is sequentially input into the predictive analysis model, and the cross-validation error of the model is calculated to evaluate the average performance of the predictive analysis model. If the average performance of the model does not reach the preset threshold, the hyperparameters such as the regularization parameter, the number of qubits, and the inner product calculation method of the predictive analysis model are adjusted, and the training and validation of the predictive analysis model are repeated until the average performance of the model converges within the preset threshold. Then, the performance of the trained predictive analysis model on unknown data is detected using the test set data. If the detection result reaches the preset expected value, the predictive analysis model is deployed to the flywheel UPS control platform in the data center; otherwise, the predictive analysis model is retrained.

[0016] P1.5: The preprocessed new load data is input into the predictive analysis model. The predictive analysis model performs forward propagation on the new load data. The predictive analysis model converts the load data into a quantum state through quantum feature mapping, calculates the similarity between the new data and the training data, and makes a prediction on the load data through a decision function. Then, classification decisions are made based on the decision function values to generate corresponding class labels, where the class label values are 0 and 1. 0 indicates that the load fluctuation exceeds the preset threshold, and 1 indicates that the load fluctuation is within the normal range.

[0017] It should be further noted that the specific calculation formula of the angle encoding method described in P1.2 is as follows:

[0018]

[0019] In the formula, represents the th input data ; represents the quantum operation of rotation around the axis; represents the angle parameter of the qubit mapping, where , represents a constant used to adjust the amplitude of the data; represents the initial state of the qubit;

[0020] The specific calculation formula of the quantum kernel function described in P1.2 is as follows:

[0021]

[0022] In the formula, represents the value of the quantum kernel function, that is, the similarity between the quantum state where the th input data is located and the quantum state where the th input data is located; and respectively represent the One input data and the input data mapped quantum state; represents the inner product between two sets of quantum states.

[0023] The specific calculation formula of the decision function described in P1.5 is as follows:

[0024]

[0025] In the formula, represents the new load data; represents the total number of training data; represents the Lagrange multiplier; represents the training data label; represents the new load data and the training data kernel function value between; represents the bias term.

[0026] As a further solution of the present invention, the specific steps for monitoring and modeling the flywheel UPS state described in S1 are as follows:

[0027] P2.1: After the load fluctuation prediction is completed, collect various operation data of the flywheel UPS system in real time through each sensor, including rotational speed, voltage, power, charge state, torque, and temperature, and record the time when each operation parameter is collected. Denoise and normalize the collected data, and then construct a flywheel dynamic system model based on the processed flywheel operation data;

[0028] P2.2: Take the current flywheel state as the root node of the operation tree. According to the flywheel dynamic system model and the current flywheel state, randomly execute various operations of the current flywheel, and generate new states based on the executed groups of operations to generate the lower-level child nodes of the root node of the current operation tree. At the same time, initialize the access times and return values of each node of the current operation tree;

[0029] P2.3: Starting from the root node of the current operation tree, calculate the UCB values of each group of child nodes, and then select the child node with the highest UCB value layer by layer according to the upper confidence bound selection strategy until an unvisited child node is selected. Then, stop the selection, and according to the flywheel state of the current child node and the flywheel dynamic system model, randomly execute various operations of the current flywheel to expand a new flywheel state, and add it as a new child node to the operation tree;

[0030] P2.4: Randomly select an operation path for simulation. During the simulation, update the flywheel state through the flywheel dynamic system model until the set simulation duration is reached, then stop the simulation. Calculate the final simulation return value based on the power consumption, energy storage, and efficiency of the flywheel during the simulation. Then, backpropagate the return value obtained from the simulation from the current simulation node to the root node, and update the return values and visit counts of each child node in the same path.

[0031] P2.5: Through multiple rounds of selection, expansion, simulation, and backtracking until the preset number of iteration rounds is reached. Then, starting from the root node of the operation tree, traverse each child node layer by layer downward. According to the return values of each child node, select the operation path with the maximum return. After finding the operation path, make real-time decisions based on this path and control the operation of the flywheel UPS, continuously monitor and adjust the operation path according to the real-time flywheel data.

[0032] It should be further noted that the specific manifestation form of the flywheel dynamic system model described in P2.1 is as follows:

[0033]

[0034]

[0035]

[0036] In the formula, represents the angular momentum of the flywheel at time ; represents the moment of inertia of the flywheel; represents the rotational speed of the flywheel at time ; represents the energy storage of the flywheel at time ; represents the input power at time ; represents the external torque on the flywheel at time ; represents the energy storage of the flywheel at time ; represents the discrete time step.

[0037] As a further solution of the present invention, the specific steps for the simulation in S2 of the collaborative work between different energy storage systems to create a load regulation scheme are as follows:

[0038] P3.1: According to the design requirements of the flywheel UPS system and the initial load fluctuation conditions, set the energy states of each energy storage unit of the flywheel UPS, and take the generated optimal operation path as the initial collaborative working strategy. At the same time, with the goal of maximizing energy utilization efficiency, minimizing power loss, and extending the service life of the equipment, construct the corresponding fitness function, and generate the corresponding virtual ecological model based on the dynamic energy changes of each storage unit of the flywheel UPS system and the interaction between different energy storage units;

[0039] P3.2: Initialize a group of populations according to the number of each storage unit in the flywheel UPS system. Based on the virtual ecological model, simulate the dynamic energy changes and interactions of each storage unit under the collaborative working strategy, calculate the fitness values of each storage unit through the fitness function, and then screen out the storage units below the preset selection threshold;

[0040] P3.3: Copy the remaining storage units, and according to the energy states of the copied storage units, adjust the collaborative working strategies of the copied storage units by adjusting the charge-discharge power and energy flow paths. Then, combine the adjusted storage units with the remaining storage units to form a new population, and use the newly generated population as the population for the next iteration;

[0041] P3.4: Repeatedly update the population until the charge-discharge strategies of each storage unit no longer change and the fitness values converge to the preset optimization goal. Then, the virtual ecological model outputs the final charge-discharge strategies, energy flow paths, and the energy states of each storage unit, and implements them in the actual flywheel UPS system, while performing real-time control.

[0042] As a further solution of the present invention, the specific calculation formula of the fitness function described in P3.1 is as follows:

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] In the formula, represents the energy loss of the th storage unit; represents the Charging efficiency of each storage unit; Representative The input energy of each storage unit; Representative The discharge efficiency of each storage unit; Representative The output energy of each storage unit; Representative Energy utilization of each storage unit; Represents the overall energy loss of the flywheel UPS system and; Representative Energy loss per storage unit; Represents the overall energy utilization of the flywheel UPS system; Representative Energy utilization of each storage unit; Representative The service life of each storage unit; Representative The number of charge and discharge cycles of each storage unit; Representative Energy loss during each charge and discharge process of a storage unit; Represents the overall service life of the flywheel UPS system; Representative The service life of each storage unit; 、 as well as Both represent the total number of storage units in the flywheel UPS system; Represents the overall fitness value of the flywheel UPS system; represents the weight of energy utilization; represents the weight of the total power loss; The weight representing the service life of the equipment.

[0051] As a further solution of the present invention, the specific steps of adaptively adjusting the flywheel UPS according to the real-time monitoring data in S4 are as follows:

[0052] P4.1: Real-time monitoring of the operating status of the flywheel UPS system and external load changes, extracting the various states of the flywheel UPS system and the corresponding operations in each state from each set of data received, and constructing a state set based on the collected state information and operation information. and action sets ,Then calculate the probability of transferring to the next random state after collecting any action in the action set under each state in the state set;

[0053] P4.2: All state-action pairs The initial value of is set to 0, where , , collect all state-action pairs, build a corresponding data repository, and store each state-action pair and its corresponding value. Then, at each moment , based on the current state and the information in the data repository, select an action to execute;

[0054] P4.3: When selecting an action , generate a random number. If the random number is higher than the preset selection threshold , randomly select an action from the action set , otherwise select the action with the maximum value. After the selection is completed, simulate the flywheel UPS system to execute the action and enter the next state ; ;

[0055] P4.4: Calculate the immediate reward in the corresponding state after executing any action based on energy efficiency, power loss, and equipment life. According to the obtained immediate reward and the next state , use the Q-value update formula to update the current state-action pair , and update the data repository. Then repeat the action selection and data repository update until the value change of the state-action pair converges to the preset range;

[0056] P4.5: After the iteration ends, traverse the values of each state-action pair in the data repository, and select the state-action pair with the highest value as the optimal response strategy. At the same time, based on the feedback after each simulated state transition and the value update, preset the future operation strategy of the flywheel UPS system.

[0057] Advantages of the present invention:

[0058] 1. The present invention can perform data fitting and pattern recognition more efficiently, reduce the training process that requires a large amount of time and computing resources in traditional methods, shorten the system response time. At the same time, through quantum machine learning, the flywheel UPS system can adjust the control strategy in real time according to load fluctuations, avoid over-reliance on preset models and static rules, effectively provide more flexible strategy selection, and enhance the system's ability to cope with changes.

[0059] 2. By means of dynamic system modeling, the flywheel UPS system of the present invention can quickly respond to load changes, enhance the robustness and stability of the system, and can efficiently evaluate the state and performance of the flywheel UPS system. At the same time, by precisely controlling the charge-discharge strategy of the flywheel UPS, it can avoid excessive consumption or waste of energy, improve energy utilization efficiency, and ensure load balance and optimized resource allocation among the various units within the flywheel UPS system by simulating the interaction between various energy units and intelligent optimization strategies, avoid overloading and inefficient operation of a single device, extend the equipment life, and reduce the maintenance cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The present invention will be further described below in conjunction with the accompanying drawings.

[0061] Figure 1 It is a framework diagram of the AC access flywheel UPS control method for suppressing dynamic load in a data center. SPECIFIC EMBODIMENTS

[0062] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0063] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0064] Embodiment 1

[0065] The embodiment of the present invention provides an AC access flywheel UPS control method for suppressing dynamic load in a data center. Refer to Figure 1 , Figure 1 which is a framework diagram of the AC access flywheel UPS control method for suppressing dynamic load in a data center provided by the embodiment of the present invention. The method includes the following steps:

[0066] S1: Predict and analyze the load fluctuation of the data center, and collect the operation data of the flywheel through sensors to monitor and model the state of the flywheel UPS.

[0067] Specifically, the flywheel UPS control platform collects multiple sets of load data collected by the data center, removes the noise and abnormal data in each set of collected load data, fills in the missing data by interpolation method, and then through Z-score normalization, unifies each set of load data within the specified range. Then, the variance screening method is used to extract the required feature information from each set of load data, extracts the historical load data from the data center, and divides it into a training set, a test set, and a validation set. The training set and the validation set are respectively divided into multiple training subsets and validation subsets. A predictive analysis model is created based on the quantum support vector machine architecture. At the beginning of each round of training, the training subset data is sequentially input into the predictive analysis model. The model maps each training set data to the quantum space through the angle encoding method and generates it on the corresponding qubits. Then, through the quantum superposition effect, the qubits of multiple sets of feature information exist in a quantum state at the same time. After that, the predictive analysis model calculates the inner product between each quantum state in the quantum space through the quantum kernel function and evaluates the similarity between each quantum state based on the calculated inner product. According to the calculated quantum kernel function values between each set of training subset data, a corresponding kernel matrix is constructed, and a corresponding objective function is constructed with the aim of maximizing the interval between each input data. The constraint conditions of the optimization objective are set. Then, based on the constraint conditions, the objective function is solved through the gradient descent algorithm to obtain the optimized parameters of the predictive analysis model. After each round of model parameter optimization is completed, the validation subset is sequentially input into the predictive analysis model, and the cross-validation error of the model is calculated to evaluate the average performance of the predictive analysis model. If the average performance of the model does not reach the preset threshold, the regularization parameter, the number of qubits, and the inner product calculation method of the predictive analysis model are adjusted, and the predictive analysis model is repeatedly trained and verified until the average performance of the model converges within the preset threshold. Then, the performance of the trained predictive analysis model on unknown data is detected using the test set data. If the detection result reaches the preset expected value, the predictive analysis model is deployed to the flywheel UPS control platform of the data center. Otherwise, the predictive analysis model is retrained, and the preprocessed new load data is input into the predictive analysis model. The predictive analysis model performs forward propagation on the new load data. The predictive analysis model converts the load data into a quantum state through quantum feature mapping, calculates the similarity between the new data and the training data, and predicts the load data through the decision function. Then, classification decisions are made based on the decision function values to generate corresponding class labels, where the class label values are 0 and 1. 0 indicates that the load fluctuation exceeds the preset threshold, and 1 indicates that the load fluctuation is within the normal range.

[0068] Specifically, after the load fluctuation prediction is completed, various operating data of the flywheel UPS system are collected in real time through each sensor, including rotational speed, voltage, power, charge state, torque, and temperature, and the time when each operating parameter is collected is recorded. The collected data is denoised and normalized, and then based on the processed flywheel operating data, a flywheel dynamic system model is constructed. The current flywheel state is used as the root node of the operation tree. According to the flywheel dynamic system model and the current flywheel state, various operations of the current flywheel are randomly executed, and new states are generated based on the executed groups of operations, generating the lower-level child nodes of the root node of the current operation tree. At the same time, the access times and return values of each node of the current operation tree are initialized. Starting from the root node of the current operation tree, the UCB values of each group of child nodes are calculated, and then according to the upper confidence bound selection strategy, the child node with the highest UCB value is selected layer by layer until an unvisited child node is selected, and then the selection stops. According to the flywheel state of the current child node and the flywheel dynamic system model, various operations of the current flywheel are randomly executed to expand new flywheel states, and they are added to the operation tree as new child nodes. A random operation path is selected for simulation. During the simulation process, the flywheel state is updated through the flywheel dynamic system model until the set simulation duration is reached, and then the simulation stops. According to the power consumption, energy storage, and efficiency of the flywheel during the simulation process, the final simulation return value is calculated, and then the return value obtained from the simulation is backpropagated from the current simulation node to the root node, and the return values and access times of each child node in the same path are updated. Through multiple rounds of selection, expansion, simulation, and backtracking, until the preset iteration number is reached, and then starting from the root node of the operation tree, each child node is traversed layer by layer downward, and according to the return values of each child node, the operation path with the maximum return is selected. After the operation path is found, real-time decisions are made according to this path, and the operation of the flywheel UPS is controlled, and the operation path is continuously monitored and adjusted according to the real-time flywheel data.

[0069] In addition, it should be noted that the specific calculation formula of the angle encoding method is as follows:

[0070]

[0071] In the formula, represents the th input data ; represents the quantum operation rotating around the axis; represents the angle parameter of the qubit mapping, where , represents a constant used to adjust the amplitude of the data; represents the initial state of the qubit;

[0072] The specific calculation formula of the quantum kernel function is as follows:

[0073]

[0074] In the formula, represents the value of the quantum kernel function, that is, the th input data in the quantum state and the th input data in the quantum state, the similarity between them; and respectively represent the th input data and the th input data after mapping to the quantum state; represents the inner product between two groups of quantum states.

[0075] The specific calculation formula of the decision function is as follows:

[0076]

[0077] In the formula, represents the th new load data; represents the total number of training data; represents the th Lagrange multiplier; represents the th training data label; represents the new load data and the training data the kernel function value between them; represents the bias term;

[0078] The specific form of the flywheel dynamic system model is as follows:

[0079]

[0080]

[0081]

[0082] In the formula, represents the angular momentum of the flywheel at time ; represents the moment of inertia of the flywheel; represents the rotational speed of the flywheel at time ; represents the energy storage of the flywheel at time ; represents the input power at time ; represents the time The external torque on the flywheel at time represents time The energy storage of the flywheel at time represents the discrete time step.

[0083] S2: Based on the predicted load fluctuations, simulate the collaborative work between different energy storage systems to create a load regulation plan and feedback it to the staff for viewing and adjustment.

[0084] Specifically, according to the design requirements of the flywheel UPS system and the initial load fluctuation situation, set the energy states of each energy storage unit of the flywheel UPS. Take the generated optimal operation path as the initial collaborative work strategy. At the same time, with the goals of maximizing energy utilization efficiency, minimizing power loss, and extending the service life of the equipment, construct the corresponding fitness function. And based on the dynamic energy changes of each storage unit in the flywheel UPS system and the interaction between different energy storage units, generate the corresponding virtual ecological model. Initialize a group of populations according to the number of each storage unit in the flywheel UPS system. Based on the virtual ecological model, simulate the dynamic energy changes and interactions of each storage unit under the collaborative work strategy, and calculate the fitness values of each storage unit through the fitness function. Then screen out the storage units with fitness values lower than the preset selection threshold, copy the remaining storage units, and adjust the collaborative work strategy of the copied storage units by adjusting the charge-discharge power and energy flow path according to the energy states of the copied storage units. After that, combine the adjusted storage units with the remaining storage units to form a new population, and use the newly generated population as the population for the next iteration. Repeatedly update the population until the charge-discharge strategies of each storage unit no longer change and the fitness values converge to the preset optimization goal. Then the virtual ecological model outputs the final charge-discharge strategies, energy flow paths, and the energy states of each storage unit, and implements them in the actual flywheel UPS system while performing real-time control.

[0085] The specific calculation formula of the fitness function is as follows:

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] Wherein, represents the energy loss of the th storage unit; represents the charging efficiency of the th storage unit; represents the input energy of the th storage unit; represents the discharging efficiency of the th storage unit; represents the output energy of the th storage unit; represents the energy utilization rate of the th storage unit; represents the sum of the overall energy loss of the flywheel UPS system; represents the th storage unit's energy loss; represents the overall energy utilization rate of the flywheel UPS system; represents the th storage unit's energy utilization rate; represents the th storage unit's service life; represents the th storage unit's charge-discharge cycle count; represents the th storage unit's energy loss during each charge-discharge process; represents the overall service life of the flywheel UPS system; represents the th storage unit's service life; , and both represent the total number of storage units in the flywheel UPS system; represents the overall fitness value of the flywheel UPS system; represents the weight of the energy utilization rate; represents the weight of the total power loss; represents the weight of the equipment service life.

[0094] Example 2

[0095] The embodiment of the present invention provides an AC access flywheel UPS control method for suppressing dynamic load in a data center. Refer to Figure 1 , Figure 1 which is the framework diagram of the AC access flywheel UPS control method for suppressing dynamic load in the data center provided by the embodiment of the present invention. The method includes the following steps:

[0096] According to the monitoring and modeling results and the actual load conditions collected by the sensors, dynamically optimize the charge-discharge cycle and power adjustment strategy of the flywheel UPS.

[0097] Monitor the operating status of the flywheel and the change of external load through the sensor and data acquisition system, and adaptively adjust the flywheel UPS according to the real-time monitoring data.

[0098] Specifically, monitor the operating status of the flywheel UPS system and the change of external load in real time, extract various states of the flywheel UPS system and the corresponding operations under each state from the received groups of data, and construct a state set based on the collected state information and operation information and an action set , then calculate the probability of transferring to the next random state after collecting any action in the action set under each state in the state set, and set the initial value of all state-action pairs to 0, where , , collect all state-action pairs, and construct a corresponding data repository to store each state-action pair and its corresponding value. Then, at each moment , select an action according to the current state and the information in the data repository, and execute it. When selecting an action , generate a random number. If the random number is higher than the preset selection threshold , randomly select an action from the action set , otherwise select the action with the largest value. After the selection is completed, simulate that the flywheel UPS system enters the next state after executing this action , calculate the immediate reward in the corresponding state after executing any action according to the energy efficiency, power loss and equipment life, and update the current state-action pair using the Q-value update formula according to the obtained immediate reward and the next state , and update the data repository. Then repeat the action selection and data repository update until the value change of the state-action pair converges within the preset range. After the iteration ends, traverse the values of each state-action pair in the data repository, and select the state-action pair with the highest value as the optimal response strategy. At the same time, update according to the feedback after each simulated state transition and value, and preset the future operation strategy of the flywheel UPS system.

[0099] Establish a digital twin model of the flywheel UPS, synchronize all data and control schemes of the flywheel UPS in real time, evaluate the energy conversion efficiency of the flywheel UPS, and adjust the control parameters.

[0100] Collect and analyze the performance data of the flywheel UPS in real time, evaluate the effectiveness of the current control strategy, and readjust the control strategy of the flywheel UPS according to the evaluation results.

[0101] The above has described an embodiment of the present invention in detail, but the described content is only a preferred embodiment of the present invention and cannot be considered as defining the scope of implementation of the present invention. All equal changes and improvements made according to the scope of the application of the present invention should still fall within the scope covered by the patent of the present invention.

Claims

1. An AC-connected flywheel UPS control method for stabilizing dynamic loads in a data center, characterized in that: The following steps are involved: S1: Predict and analyze data center load fluctuations, collect flywheel operating data through sensors, and monitor and model the flywheel UPS status; S2: Based on the predicted load fluctuations, simulate the collaborative work between different energy storage systems to create a load adjustment plan and provide feedback to staff for review and adjustment; S3: Dynamically optimize the charge and discharge cycle and power adjustment strategy of the flywheel UPS based on the monitoring modeling results and the actual load conditions collected by the sensors; S4: Monitor the operating status of the flywheel and external load changes through sensors and data acquisition systems, and adaptively adjust the flywheel UPS based on real-time monitoring data; S5: Build a digital twin model of the flywheel UPS to synchronize various flywheel UPS data and control schemes in real time, evaluate the energy conversion efficiency of the flywheel UPS, and adjust control parameters; S6: Collect and analyze flywheel UPS performance data in real time, evaluate the effectiveness of the current control strategy, and readjust the flywheel UPS control strategy based on the evaluation results; The specific steps for predicting and analyzing the load fluctuation of the data center described in S1 are as follows: P1.1: The Flywheel UPS control platform collects multiple sets of load data from the data center, removes noise and outliers from each set, and fills in missing data using interpolation. It then normalizes each set of load data using the Z-score method to bring it within a specified range. It then uses variance screening to extract the required feature information from each set of load data. It also extracts historical load data from the data center and divides it into training, test, and validation sets. The training and validation sets are then divided into multiple training and validation subsets, respectively. P1.2: Create a predictive analysis model based on the quantum support vector machine architecture. At the beginning of each training round, the training subset data is input into the predictive analysis model in sequence. The model maps each training set data into the quantum space through the angle encoding method and generates corresponding quantum bits. Then, through the quantum superposition effect, the quantum bits of multiple sets of characteristic information exist simultaneously in the same quantum state. The predictive analysis model then calculates the inner product between each quantum state in the quantum space through the quantum kernel function and evaluates the similarity between each quantum state based on the calculated inner product. The specific calculation formula of the angle encoding method is as follows: Where, Representative Input data ; Represents around Quantum manipulation of axis rotation; represents the angular parameter of the quantum bit mapping, where , Represents a constant used to adjust the amplitude of the data; represents the initial state of the quantum bit; The specific calculation formula of the quantum kernel function is as follows: Where, Represents the value of the quantum kernel function, that is, Input data The quantum state and the Input data The similarity between the quantum states; and Representing the Input data With the Input data The quantum state after mapping; represents the inner product between two sets of quantum states; P1.3: Based on the calculated quantum kernel function values between each set of training subset data, a corresponding kernel matrix is constructed. The corresponding objective function is constructed with the goal of maximizing the interval between each input data. Constraints for the optimization objective are set. Based on the constraints, the objective function is solved using a gradient descent algorithm to obtain the optimized predictive analysis model parameters. P1.4: After each round of model parameter optimization, the validation subset is sequentially input into the predictive analysis model, and the cross-validation error of the model is calculated to evaluate the average performance of the predictive analysis model. If the average performance of the model does not reach the preset threshold, the regularization parameter, number of quantum bits, and inner product calculation method of the predictive analysis model are adjusted, and the predictive analysis model is repeatedly trained and verified until the average performance of the model converges to the preset threshold. The test set data is then used to test the performance of the trained predictive analysis model on unknown data. If the test result meets the preset expected value, the predictive analysis model is deployed to the flywheel UPS control platform of the data center. Otherwise, the predictive analysis model is retrained. P1.5: The pre-processed new load data is input into the prediction analysis model. The prediction analysis model performs forward propagation on the new load data. The prediction analysis model converts the load data into a quantum state through quantum feature mapping, calculates the similarity between the new data and the training data, and predicts the load data through the decision function. The decision function value is then used to make a classification decision to generate a corresponding category label. The category label value is 0 or 1. 0 indicates that the load fluctuation exceeds the preset threshold, and 1 indicates that the load fluctuation is within the normal range. The specific calculation formula of the decision function is as follows: Where, Representative New load data; Represents the total number of training data; Representative Lagrange multipliers; Representative training data 's label; Represents new load data and training data The kernel function value between ; represents the bias term.

2. The AC access flywheel UPS control method for stabilizing dynamic loads in a data center according to claim 1, characterized in that: The specific steps for monitoring and modeling the flywheel UPS status described in S1 are as follows: P2.1: After the load fluctuation prediction is completed, various sensors are used to collect various operating data of the flywheel UPS system in real time, including speed, voltage, power, charge status, torque, and temperature. The time of each operating parameter collection is recorded, the collected data is denoised and normalized, and the flywheel dynamic system model is constructed based on the processed flywheel operating data. P2.2: Use the current flywheel state as the root node of the operation tree. Based on the flywheel dynamic system model and the current flywheel state, randomly execute the current flywheel operations. Generate a new state based on each set of operations executed, generate the child nodes below the root node of the current operation tree, and initialize the visit count and reward value of each node in the current operation tree. P2.3: Starting from the root node of the current operation tree, calculate the UCB value of each group of child nodes. Then, according to the upper confidence bound selection strategy, select the child node with the highest UCB value layer by layer until an unvisited child node is selected. Then, stop selecting and randomly execute various operations of the current flywheel based on the flywheel state of the current child node and the flywheel dynamic system model to expand the new flywheel state and add it as a new child node to the operation tree. P2.4: Randomly select an operation path for simulation. During the simulation, the flywheel state is updated using the flywheel dynamic system model. After the set simulation duration is reached, the simulation is stopped. The final simulation reward value is calculated based on the flywheel's power consumption, energy storage, and efficiency during the simulation. The simulated reward value is then propagated back from the current simulation node to the root node, and the reward value and visit count of each child node in the same path are updated. P2.5: Through multiple rounds of selection, expansion, simulation, and backtracking until the preset number of iterations is reached, the system starts from the root node of the operation tree and traverses each child node layer by layer. Based on the reward value of each child node, the operation path with the maximum reward is selected. After the operation path is found, real-time decisions are made based on the path to control the operation of the flywheel UPS system, and the operation path is continuously monitored and adjusted based on real-time flywheel data.

3. The AC access flywheel UPS control method for stabilizing dynamic loads in a data center according to claim 2, characterized in that: The specific steps for simulating the collaborative work between different energy storage systems to create a load regulation solution as described in S2 are as follows: P3.1: Based on the flywheel UPS system design requirements and initial load fluctuations, the energy state of each energy storage unit in the flywheel UPS is set. The resulting optimal operation path is used as the initial collaborative working strategy. With the goals of maximizing energy utilization, minimizing power loss, and extending equipment life, a corresponding fitness function is constructed. Based on the dynamic energy changes of each storage unit in the flywheel UPS system and the interactions between different energy storage units, a corresponding virtual ecological model is generated. P3.2: Initialize a population based on the number of storage units in the flywheel UPS system. Simulate the dynamic energy changes and interactions of each storage unit under the collaborative working strategy using a virtual ecological model. Calculate the fitness value of each storage unit using a fitness function, and then filter out storage units that fall below a preset selection threshold. P3.3: Duplicate the remaining storage cells and adjust the collaborative working strategy of the copied storage cells by adjusting the charge and discharge power and energy flow path based on the energy state of the copied storage cells. Then, combine the adjusted storage cells with the remaining storage cells to form a new population, and use the newly generated population as the population for the next iteration; P3.4: The population is updated repeatedly until the charging and discharging strategies of each storage unit are no longer adjusted and the fitness value converges to the preset optimization target. The virtual ecological model then outputs the final charging and discharging strategy, energy flow path, and energy status of each storage unit, which are implemented in the actual flywheel UPS system and controlled in real time.

4. The AC access flywheel UPS control method for stabilizing dynamic loads in a data center according to claim 3, characterized in that: The specific calculation formula of the fitness function described in P3.1 is as follows: Where, Representative Energy loss per storage unit; Representative Charging efficiency of each storage unit; Representative The input energy of each storage unit; Representative The discharge efficiency of each storage unit; Representative Output energy of each storage unit; Representative Energy utilization of each storage unit; Represents the overall energy loss of the flywheel UPS system; Representative Energy loss per storage unit; Represents the overall energy utilization of the flywheel UPS system; Representative Energy utilization of each storage unit; Representative The service life of each storage unit; Representative The number of charge and discharge cycles of each storage unit; Representative Energy loss during each charge and discharge process of a storage unit; Represents the overall service life of the flywheel UPS system; Representative The service life of each storage unit; 、 as well as Both represent the total number of storage units in the flywheel UPS system; Represents the overall fitness value of the flywheel UPS system; represents the weight of energy utilization; represents the weight of the total power loss; The weight representing the service life of the equipment.

5. The AC access flywheel UPS control method for stabilizing dynamic loads in a data center according to claim 3, characterized in that: The specific steps of adaptively adjusting the flywheel UPS according to real-time monitoring data described in S4 are as follows: P4.1: Real-time monitoring of the operating status of the flywheel UPS system and external load changes, extracting the various states of the flywheel UPS system and the corresponding operations in each state from each set of data received, and constructing a state set based on the collected state information and operation information. and action sets ,Then calculate the probability of transferring to the next random state after collecting any action in the action set under each state in the state set; P4.2: All state-action pairs The initial value of is set to 0, where , , collect all state-action pairs and build a corresponding data repository to store each state-action pair and its corresponding value, and then at each moment , according to the current state and the information in the data repository, select Action implement; P4.3: Selecting an action When a random number is generated, if the random number is higher than the preset selection threshold , then from the action set Randomly select an action, otherwise select The action with the largest value, After the selection is completed, the simulated flywheel UPS system will enter the next state after executing the action. ; P4.4: Calculate the immediate reward for executing any action in the corresponding state based on energy efficiency, power loss, and device life. , use the Q value update formula to update the current state-action pair , and update the data repository, and then repeat the action selection and data repository update until the state-action pair The value changes converge to the preset range; P4.5: After the iteration, traverse the state-action pairs in the data repository. value, and select the state-action pair with the highest value as the optimal response strategy, and at the same time, based on the feedback and The value is updated to pre-set the future operation strategy of the flywheel UPS system.

Images