Flywheel energy storage and storage battery hybrid UPS system for data center

Abstract

The invention relates to the technical field of power management, in particular to a flywheel energy storage and storage battery hybrid UPS system for a data center, which comprises a load identification module, a disturbance correction module, a scheduling screening module, a time reconstruction module and an instruction updating module. According to the method, the current rotating speed and the output current of the flywheel are obtained, slope change trend analysis is executed, the impact characteristic generated when the load is started can be recognized in advance, and the transient instability risk caused by large-current impact is avoided. And carrying out cross analysis on the original access time sequence of the load and the power disturbance response of the power supply side, so that the load access behavior with concentrated original time sequence and violent current fluctuation is orderly expanded in the time domain. The power supply parameters of the air conditioner group are extracted by the flywheel in the steady-state operation period to serve as the reference, interval adaptation judgment is conducted on the power supply parameters and the operation characteristics of the operation and maintenance equipment to be connected, the equipment with high stability is scheduled preferentially, and the balance and controllability of overall scheduling response are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power management, and in particular to a flywheel energy storage and battery hybrid UPS system for a data center. Background Art

[0002] The field of power management technology encompasses the research and application of the distribution, control, conversion, and regulation of power resources, primarily to ensure the continuous and stable operation of electrical equipment under various power supply conditions. This includes the development of emergency response mechanisms for unstable utility power, measures to ensure power continuity, and the coordinated control of multiple backup power supply methods.

[0003] The flywheel energy storage and battery hybrid UPS system for data centers utilizes a multi-energy storage combined power supply architecture, created by connecting flywheel energy storage devices in parallel with battery banks, to achieve uninterruptible power supply switching control. This system primarily addresses issues such as inconsistent power supply switching response times, limited power redundancy, and the insufficient transient load support capabilities of traditional battery UPS systems in data centers during sudden power outages or voltage fluctuations.

[0004] Traditional UPS systems rely on batteries for instantaneous response to sudden power outages or voltage fluctuations. This often results in delayed response to the high current surges during peak load startup, making them unable to match the instantaneous power demands of sudden load increases. Load connection sequencing often relies on preset strategies or static judgments, failing to consider the dynamic changes in current power supply capacity and load behavior. This leads to uneven current distribution during power switching and frequent relay mis-triggering. The lack of a load connection adjustment mechanism based on real-time power disturbance trajectories causes current spikes when multiple loads are connected at similar times, exacerbating system instability and potentially causing a sudden drop in flywheel speed or battery overload discharge. The scheduling of stable equipment, such as air conditioners, is not differentiated from highly fluctuating operation and maintenance equipment. The scheduling strategy lacks a basis for adapting the operating characteristics of different equipment, making it impossible to effectively filter out incompatible targets, which can disrupt the scheduling order of other loads. The original control triggering logic fails to dynamically adjust to the real-time state of load connection, resulting in control commands being unable to promptly respond to changes in load demand. Delays in relay logic in the power supply chain affect overall response speed, reducing the reliability of continuous power supply in data centers during emergencies. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a flywheel energy storage and battery hybrid UPS system for data centers.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A flywheel energy storage and battery hybrid UPS system for a data center includes: The load identification module obtains the current speed and output current of the flywheel and performs synchronous slope analysis to extract the starting current threshold of the current load to be connected. It then filters out devices whose starting current threshold is greater than the flywheel power supply limit and marks the load delay item set. The disturbance correction module obtains the change trajectory of the flywheel output power and the battery voltage response and performs cluster division, and performs a backshift processing on the original access time point of the device in the load delay item set according to the division result to generate a load delay adjustment sequence; The scheduling and screening module obtains the current mean and apparent power output standard deviation of the data center air conditioning group during the steady-state operation cycle of the flywheel power supply, compares it with the operation and maintenance equipment currently to be connected, and selects the advance scheduling list of operation and maintenance equipment; The time reconstruction module reorganizes the time axis of the load in the main power supply channel of the data center based on the load delay adjustment sequence and the advance scheduling list of the operation and maintenance equipment to form a load regulation time mapping index table; The instruction update module updates the trigger control process in the main current channel of the data center combined with the flywheel and the battery through the load regulation time mapping index table to form a trigger control scheduling instruction flow.

[0007] As a further solution of the present invention, the load delay item set includes devices whose starting current threshold exceeds the flywheel power supply limit, loads to be connected that need to be delayed, and load targets identified by synchronous slope analysis. The load delay adjustment sequence includes the offset value of the original access time point, the disturbance group identifier corresponding to the clustering division, and the load mapping item corresponding to the power disturbance peak. The operation and maintenance equipment advance scheduling list includes devices that meet the current mean requirements, equipment that meet the apparent power output standard deviation conditions, and operation and maintenance equipment consistent with the operating characteristics of the air-conditioning group. The load control time mapping index table includes the access time update items in the operation and maintenance equipment advance scheduling list, the access time adjustment items in the load delay adjustment sequence, and the time axis arrangement items after unified mapping. The trigger control scheduling instruction flow includes the load access sequence update instruction, the trigger control signal in the main current channel, and the scheduling trigger action corresponding to the time mapping index.

[0008] As a further solution of the present invention, the load identification module includes: The speed and current extraction submodule obtains the continuous sampling data of the current speed and current output of the flywheel, calculates the speed change rate and current change rate within each sampling period, and generates a set of joint speed and current change rates; The slope mutation determination submodule uses a local weighted regression algorithm to perform trend fitting based on the speed and current joint change rate set, performs continuous period slope gradient calculation on the fitting residual sequence, and determines whether the residual gradient rises continuously and exceeds the average background fluctuation amplitude of the trend segment, thereby generating a trend slope mutation identification interval; The power supply limit screening submodule extracts the starting current threshold of the load to be connected according to the load identified in the trend slope mutation identification interval, compares it with the current power supply limit value of the flywheel, screens all target loads whose starting current threshold is greater than the power supply limit value, and generates a set of load delay items.

[0009] As a further solution of the present invention, the disturbance correction module includes: The power feature extraction submodule obtains the change trajectory of flywheel output power and battery voltage response, extracts the flywheel power change rate, battery voltage mutation point time, and the response delay of the intersection of the two, and performs normalization processing to generate a set of disturbance interval feature parameters; The interleaved disturbance clustering submodule uses a spectral clustering algorithm to cluster the voltage and current mutation points generated during the alternating power supply process according to the disturbance interval characteristic parameter set, extracts the overlapping sections in the clustering results where the disturbance intensity is greater than the average disturbance threshold, and obtains the disturbance peak cluster distribution information; The access time adjustment submodule calls the original access time point of each device in the load delay item set based on the disturbance peak cluster distribution information, maps it with the disturbance duration of the corresponding overlapping segment, performs time point backward offset processing, and generates a load delay adjustment sequence.

[0010] As a further solution of the present invention, the scheduling screening module includes: The power characteristic calculation submodule obtains the current data and apparent power data of the data center air conditioning group during the steady-state operation cycle of the flywheel power supply, calculates the current mean and apparent power output standard deviation within the time period, and forms the power supply side benchmark comparison items to generate the air conditioning group operation characteristic parameters; The device matching and comparison submodule extracts the starting current requirement and operating power factor of the currently connected operation and maintenance equipment based on the operating characteristic parameters of the air conditioning group, calculates the deviation difference with the power supply side benchmark comparison item, and performs interval intersection judgment on the deviation difference set of all devices to obtain the equipment operation adaptation interval; The access load screening submodule extracts the device access attributes for the operation and maintenance devices in the device operation adaptation range, organizes the unique identification and operation parameters of each device, and generates an advance scheduling list for the operation and maintenance devices.

[0011] As a further solution of the present invention, the time reconstruction module includes: The time offset identification submodule extracts the time offset value and original access time of each load based on the load delay adjustment sequence and the advance scheduling list of the operation and maintenance equipment, classifies the forward offset and reverse offset respectively, and generates load access offset structure data; The access time rearrangement submodule replaces the original access timeline of the current main power supply channel as a whole according to the load access offset structure data, shifts the delayed items backward and the advanced items forward, rearranges the access positions according to the adjusted time sequence, and obtains the time axis sequence rearrangement result; The time index generation submodule calls the time axis sequence rearrangement result, extracts the new access time point corresponding to each load and the relative position number in the rearranged sequence, integrates the load unique identifier and time coordinate information, and generates a load control time mapping index table.

[0012] As a further solution of the present invention, the instruction update module includes: The time node extraction submodule calls all load identifiers and corresponding access time information in the load control time mapping index table, extracts the adjusted start time field, constructs an update sequence in chronological order and binds the relay action sequence number to generate a control instruction time sorting result; The control logic rearrangement submodule completely replaces the relay triggering logic sequence in the main current channel shared by the flywheel and the battery based on the control instruction time sequence, maps the original relay instructions with the updated time sequence, and rearranges the control triggering sequence according to the time sequence to generate a relay response rearrangement queue; The trigger process generation submodule calls the relay response reordering queue, synchronously writes the relay actions and the corresponding time sequence into the execution list of the control channel contacts, and merges all the adjusted control trigger behaviors into a unified task flow to generate a trigger control scheduling instruction flow for coordinating the load access timing of the flywheel energy storage and the battery under the hybrid UPS structure, thereby realizing the power supply trigger sequence control in the main current channel of the data center.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by obtaining the current speed and output current of the flywheel and performing slope change trend analysis, the impact characteristics generated when the load is started can be identified in advance, and load targets that are difficult to cover with the power supply capacity can be screened accordingly, avoiding the risk of transient instability caused by large current impact. The original access timing of the load is cross-analyzed with the power disturbance response on the power supply side to identify the disturbance intensity distribution in the alternating power supply and match the load delayed access node, so that the load access behavior that was originally concentrated in time and had severe current fluctuations can be orderly unfolded in the time domain. The power supply parameters of the air-conditioning group are extracted as a benchmark by the flywheel during the steady-state operation cycle, and the interval adaptation judgment is performed with the operating characteristics of the operation and maintenance equipment to be connected. Targets with excessive power deviation are eliminated, and equipment with high stability is scheduled first, thereby improving the balance and controllability of the overall scheduling response. The access time of all loads is processed with positive and negative offsets and the time axis structure is rearranged to achieve timing reconstruction of different types of equipment according to dynamic load intensity, reduce the current peak at the moment of load access, and improve the instruction control accuracy of the relay action in the power supply path. Mapping the new access time of each load to a unified scheduling trigger flow, and synchronizing control actions through relay contacts, reduces trigger response jitter caused by control delays in the power supply chain. This enables fine-grained time control for the coordinated access of multiple types of devices under dynamic load conditions, improving the stability and flexibility of the main power supply channel in high-frequency disturbance scenarios, and enhancing the synergistic efficiency of flywheel and battery hybrid power supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a system flow chart of the present invention; Figure 2 This is a flow chart of the load identification module of the present invention; Figure 3 This is a flow chart of the disturbance correction module of the present invention; Figure 4 A flowchart of the scheduling screening module of the present invention; Figure 5 This is a flow chart of the time reconstruction module of the present invention; Figure 6 This is a flow chart of the instruction update module of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0016] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0017] See also Figure 1 A flywheel energy storage and battery hybrid UPS system for a data center includes: The load identification module obtains the current speed and output current of the flywheel and performs synchronous slope analysis to extract the starting current threshold of the current load to be connected. It then filters out devices whose starting current threshold is greater than the flywheel power supply limit and marks the load delay item set. The disturbance correction module obtains the change trajectory of the flywheel output power and the battery voltage response and performs clustering. Based on the clustering results, the original access time point of the equipment in the load delay item set is shifted back to generate a load delay adjustment sequence. The scheduling and screening module obtains the current mean and apparent power output standard deviation of the data center air conditioning group during the steady-state operation cycle of the flywheel power supply, compares it with the operation and maintenance equipment currently to be connected, and selects the advance scheduling list of operation and maintenance equipment; The time reconstruction module maps and reorganizes the time axis of the load in the main power supply channel of the data center based on the load delay adjustment sequence and the advance scheduling list of operation and maintenance equipment to form a load control time mapping index table; The instruction update module updates the trigger control process in the main current channel of the data center combined with the flywheel and battery through the load control time mapping index table to form a trigger control scheduling instruction flow; The load delay item set includes devices whose starting current threshold exceeds the flywheel power supply limit, loads to be connected that need to be delayed, and load targets identified by synchronous slope analysis. The load delay adjustment sequence includes the offset value of the original access time point, the disturbance group identifier corresponding to the clustering division, and the load mapping item corresponding to the power disturbance peak. The advance scheduling list of operation and maintenance equipment includes equipment that meets the current mean requirements, equipment that meets the apparent power output standard deviation conditions, and operation and maintenance equipment consistent with the operating characteristics of the air-conditioning group. The load control time mapping index table includes the access time update items in the advance scheduling list of operation and maintenance equipment, the access time adjustment items in the load delay adjustment sequence, and the time axis arrangement items after unified mapping. The trigger control scheduling instruction flow includes the load access sequence update instruction, the trigger control signal in the main current channel, and the scheduling trigger action corresponding to the time mapping index.

[0018] See also Figure 2 , the load identification module includes: The speed and current extraction submodule obtains the continuous sampling data of the current speed and current output of the flywheel, calculates the speed change rate and current change rate within each sampling period, and generates a set of joint speed and current change rates; To obtain the continuous sampling data of the current speed and output current of the flywheel, it is necessary to set a 1-second detection cycle during the operation of the flywheel energy storage system in the data center, and perform isochronous sampling at a fixed time interval of 1 millisecond within this cycle, collecting a total of 20 sets of instantaneous data of the flywheel angular velocity and output current. The flywheel angular velocity recorded during the sampling process is ,in Indicates the flywheel angular velocity in radians per minute (rad / min), with the subscript Indicates the sampling time; the output current is recorded as ,in Indicates the instantaneous output current value in amperes (A). and Calculate the changes between to obtain the rate of change per unit time. The index representing the speed change rate, corresponding to the sampling time point , and the next moment is , the speed change rate calculation formula is: ; in Indicates the The rate of change per unit time of the flywheel angular velocity in the group data, Indicates the The angular velocity value of the sampling point, It represents the angular velocity value at the next moment. is the sampling time interval, which is constant at 1ms.

[0019] set up The index representing the current change rate, corresponding to the sampling time point , and the next moment is , the current change rate calculation formula is: ; in Indicates the The rate of change per unit time of the output current in the group data, Indicates the The current value of the sampling point, Indicates the current value at the next moment.

[0020] Taking the actual sampling data as an example, if at the 6th sampling moment, the flywheel angular velocity , Group 7 is ,but: ; If the current at the corresponding moment is ,but: ; Calculate the change rate for all 19 groups of adjacent sampling points in turn, and form The data set composed of This set is used to describe the instantaneous synergistic relationship between the flywheel speed fluctuation and the current response, and serves as the input data source for subsequent trend modeling and anomaly identification, ultimately generating a set of joint speed and current change rates.

[0021] The slope mutation judgment submodule uses a local weighted regression algorithm to perform trend fitting based on the set of joint speed and current change rates. It calculates the continuous period slope gradient of the fitted residual sequence and determines whether the residual gradient rises continuously and exceeds the average background fluctuation amplitude of the trend segment, thereby generating a trend slope mutation identification interval. Based on the speed and current joint change rate set ,in, Indicates the The rate of change of the flywheel speed within a sampling interval, in rad / min / ms, Indicates the The current change rate within a sampling interval is in A / ms. ,by is the independent variable, As the dependent variable, a local weighted regression model is constructed, and the regression function form is: ; in, Indicates the fitting result The predicted current change rate of the sample point is Represents the bias term, that is, The predicted current change rate when is the trend slope coefficient, which indicates the current response change amplitude corresponding to the unit speed change rate. Indicates the The speed change rate of each sample point.

[0022] The sample weighting in the regression process is determined by the Gaussian kernel function, and the weight is expressed as: ; in, For the Sample pairs fitting center point The weight of For the The speed change rate of samples, is the speed change rate of the current fitting center point, is the bandwidth coefficient, which controls the weight attenuation rate. The setting basis is: within the flywheel change response cycle, according to the actual sampling data The variance value setting Its standard deviation , for example, in the sampled data The sample standard deviation of , then suppose .

[0023] Regression coefficient 、 It is derived from the weighted least squares criterion to minimize the weighted residual sum of squares: ; in, is the objective function, is the number of regression samples, For the The current change rate of each sample is obtained by solving the normal equations. 、 make Minimum.

[0024] Then calculate the residual value of each fitting point: ; in, Indicates the The residual value of each sample point measures the fitting error, and the unit is A / ms.

[0025] Divide the entire residual sequence into continuous sliding detection windows, and set The corresponding residual value of the sliding detection window is , the residual slope is calculated as: ; in, It is The first The residual slope of the residual point, Indicates the The first The global index position of the residual point, express The next residual point of is used to construct the two data points in the residual slope calculation for this pair, It is The first The residual point and the The difference in the rate of change of angular velocity between the residual points.

[0026] Then the average value of all residual slopes in the sliding detection window is calculated as the residual increase rate of the sliding detection window. : ; in, Indicates the The average residual growth rate within a sliding detection window reflects the strength of the fitting residual change trend. is the number of residual slopes that can be calculated in the sliding detection window.

[0027] Setting the mutation determination threshold The threshold is based on all the historical steady-state operation periods. average value Setting, expressed as: ; in, Indicates the amplification factor, which is used to improve the sensitivity of mutation judgment. The setting basis is the upper limit of the system's fluctuation tolerance under non-mutation conditions. If the empirical evaluation , and the system design allows the mutation recognition threshold to be 1.5 times of this value, so ,have to .

[0028] Pick , suppose a set of actual sampling data is: , residual value .

[0029] The corresponding residual slope is calculated as follows: ; ; ; .

[0030] Then the residual increase rate of the sliding detection window is: ; because , reaches the mutation determination threshold, the system records the sliding detection window number , the sample time interval within it, and the load number in the access queue during this time period form the trend slope mutation identification interval.

[0031] The power supply limit screening submodule identifies the load to be connected in the interval based on the trend slope mutation, extracts the starting current threshold of the load, compares it with the current power supply limit of the flywheel, screens all target loads whose starting current threshold is greater than the power supply limit, and generates a set of load delay items; Based on the number of the load to be connected identified during the trend slope mutation identification interval, the corresponding load's starting current threshold information is first retrieved from the data center load management database. This information is typically clearly defined in the equipment's factory configuration parameters and recorded in the load management table. For example, for a flywheel power supply segment, three loads (Load A, Load B, and Load C) are detected corresponding to the mutation interval. Their corresponding starting current thresholds are 220A, 180A, and 250A, respectively, all in amperes (A). Simultaneously, the flywheel control unit's operating status monitoring data is retrieved to obtain the upper limit of the flywheel's output capacity during the mutation identification period, known as the power supply limit. This value is affected by the flywheel's remaining energy storage, current speed, and voltage steady state, and can fluctuate in real time. For example, at the time of detection, the power supply limit is 200A. The starting current threshold of each load to be connected is then compared against this power supply limit to determine whether any limit is exceeded. Loads with starting current thresholds exceeding the power supply limit are deemed inaccessible for the current cycle to prevent transient voltage drops caused by excessive flywheel output load. For example, the 220A starting current of load A is higher than the 200A power supply limit, so it needs to be connected later; load B is 180A, which is lower than the limit value and can be left unchanged; load C is 250A, which is also beyond the power supply range and needs to be started later. All loads that meet the starting current threshold value higher than the current flywheel power supply limit value are marked according to their original start time in the connection sequence and included in a unified delay processing list. This list not only retains the load unique identifier and starting current value, but also includes its original set connection time point and its associated number corresponding to the mutation identification sliding detection window, which serves as the basic data for subsequent disturbance avoidance and connection sorting reconstruction processing, and finally summarizes and generates a set of load delay items.

[0032] See also Figure 3 , the disturbance correction module includes: The power feature extraction submodule obtains the change trajectory of flywheel output power and battery voltage response, extracts the flywheel power change rate, battery voltage mutation point time, and the response delay of the intersection of the two, and performs normalization processing to generate a set of disturbance interval feature parameters; Based on the set of load delay items, the delayed load device number and original access period recorded in it are retrieved one by one, and the power supply channel corresponding to each delayed load is located in the flywheel-battery hybrid power supply architecture. The power output and battery voltage response trajectory within 0.5 seconds before and after the access moment are included in the joint analysis window. The flywheel output power data is obtained by multiplying the bus voltage and current measurement values. For example, if the delay access time of a delayed load L1 is 2.65 seconds, the starting point of the observation joint analysis window is set to 2.15 seconds, and the sampling time is 2.65 seconds. The interval is 2ms. The flywheel power value sequence measured in the joint analysis window is 5.6kW, 5.1kW, 4.4kW, 3.9kW, 3.2kW, and 2.6kW. According to the difference between adjacent points, the power change rate is calculated, and the sudden drop point occurs at the fifth sampling point, that is, 2.23 seconds. The battery voltage response value sequence is 232V, 229V, 227V, 220V, 211V, and 204V. Comparative analysis shows that the maximum voltage drop occurs at the sixth point, that is, 2.24 seconds. This point is extracted as the voltage mutation reference point. Then, The time difference between the power mutation point and the voltage mutation point is calculated to obtain the response delay time. In this example, the response delay is one sampling period, that is, 2ms. This delay value will be used as one of the subsequent disturbance characteristics. Then, the maximum flywheel power change rate and the voltage drop amplitude in this section are calculated respectively. For example, the flywheel power drops by 2.4kW from 2.21 seconds to 2.23 seconds, and the voltage drops by 25V from 2.22 seconds to 2.24 seconds. Subsequently, the three index values ​​are normalized, where the normalized reference value is set to the flywheel rated maximum. The power variation amplitude is 8.0kW, the maximum voltage variation drop amplitude is 60V, and the maximum observation tolerance of the response delay is 20ms. After normalization, the power variation rate is 0.30, the voltage drop is 0.42, and the response delay is 0.10. These three values ​​are combined to form the feature vector of a single disturbance sample. For all delayed loads, the power-voltage trajectory information in each power supply channel is collected according to this processing logic. The corresponding feature vectors are cyclically extracted and arranged in sequence to form a normalized feature data sequence of the disturbance interval, which is finally integrated to obtain the disturbance interval feature parameter set.

[0033] The staggered disturbance clustering submodule uses a spectral clustering algorithm to cluster the voltage and current mutation points generated during the alternating power supply process based on the set of disturbance interval characteristic parameters. It extracts the overlapping sections in the clustering results where the disturbance intensity is greater than the average disturbance threshold and obtains the cluster distribution information of the disturbance peak. Based on the set of disturbance interval characteristic parameters, each disturbance sample in the set consists of three-dimensional normalized values, namely the flywheel power change rate, battery voltage drop amplitude, and response delay time. These values ​​together form the sample's feature vector. To identify the correlation trend between these samples, the similarity between any two samples is first calculated, using the Euclidean distance to measure the degree of difference between the two in the three-dimensional disturbance feature space. The distance value is then substituted into the similarity calculation formula to indicate the degree of closeness between the two samples in terms of disturbance behavior. The similarity calculation is expressed as follows: ; in, 、 Respectively represent Hedi The perturbation feature vector of samples, Represents the Euclidean distance between them, and the unit is a dimensionless normalized value. It is the similarity distance scale coefficient, which is used to control the sensitivity of distance difference to similarity. The value is set according to the mean of the standard deviation of the distance between all sample pairs. For example, if the statistical mean of the standard deviation of the distance between all sample pairs is 0.48, then set , It is Hedi The similarity of samples.

[0034] Take sample 1: Sample 2: Take the example to calculate, assuming sample 1 is the samples, sample 2 is Samples: ; .

[0035] The remaining sample pairs are calculated in the same way to obtain the similarity values ​​between all perturbation samples. After completion, the classification of the samples is determined according to their association strength.

[0036] Subsequently, to identify different types of disturbance trends, spectral clustering is performed on all samples. Specifically, a connection graph is constructed based on the similarity information between all disturbance feature samples. Eigenvalue decomposition is performed on this graph, and the first few eigenvectors are extracted to form a new representation. This representation is then clustered using mean values. The number of clusters is determined based on the system's operational disturbance classification principles, typically three categories, corresponding to mild, moderate, and strong disturbance regions.

[0037] Suppose there are 5 perturbation samples, sample 1: , sample 2: , sample 3: , sample 4: , sample 5: .

[0038] The square of the Euclidean distance between sample 1 and sample 2 is calculated as: ; Substitute into the similarity formula: .

[0039] After calculating the similarity of all samples in turn, the connection relationship between samples is constructed. After feature decomposition, they are mapped to a low-dimensional space and divided into three groups of disturbance types using the K-means algorithm. After clustering, the mean of the disturbance intensity is calculated for each group of samples. If the mean of the normalized value of the flywheel power change rate in a certain type of sample is greater than the overall mean of this item for all samples, 1.3 times, then the class is defined as a high-intensity disturbance class. For example, if , then the threshold is If the mean value of a certain class of samples is 0.72, then this class is selected as the strong disturbance segment. Finally, the original timestamps, sequence positions, and associated load numbers of all samples belonging to this strong disturbance category are extracted, and their start and end positions in the time series interval, concentrated distribution segments, and corresponding power supply periods are recorded together to form the disturbance peak cluster distribution information.

[0040] The access time adjustment submodule uses the disturbance peak cluster distribution information to call the original access time point of each device in the load delay item set, maps it with the disturbance duration of the corresponding overlapping segment, performs backward time point offset processing, and generates a load delay adjustment sequence; Based on the perturbation peak cluster distribution information, first retrieve all the perturbation samples marked as strong perturbation categories, each of which contains the original timestamp, the perturbation sequence number , and the corresponding segment start time and end time , both in seconds (s), the above two time values ​​together define the actual duration of the disturbance segment. Subsequently, a time matching operation is performed on each load to be connected in the identified load delay item set, and the load numbers are read in turn. , original access time ,in Indicates the first item in the delayed item collection Load objects, unit is seconds (s). Compare with the time boundaries of all high-disturbance sections and use closed interval matching logic to determine whether a certain section exists The access time satisfies the following conditions: If the condition is met, it means that the load hits a high-intensity disturbance segment and the access time needs to be adjusted. Otherwise, the original planned access time remains unchanged. The adjustment range of the access time depends on the disturbance segment hit. The duration of the disturbance is recorded as , unit is seconds (s), the system presets a fixed buffer delay value , used to ensure that the load avoids the impact of the disturbance tail, unit is seconds (s), so the new access time is set to: ,in, Indicates the access time after delay adjustment. For example, if the load Original access time , the time range of hitting segment D1 , then the disturbance duration , the buffer value is set to , the final offset is 0.17 seconds, and the new access time is If the load hits multiple disturbance sections at the same time, there are multiple Make Fall into multiple In the interval, the last segment (i.e. The largest one) corresponds to and The sum of the two is used as the offset basis to avoid interference conflicts between continuous disturbances. All access times after adjustment It is necessary to verify again whether there is overlap between it and other loads, and set the minimum access time interval to If multiple loads are found, 、 All adjusted to the same access time , the subsequent loads are added in sequence according to the original plan priority order. The offset of In this way, the access times of all conflicting loads are iteratively adjusted to ensure that the power supply system scheduling sequence is reasonable and without overlap. Ultimately, each delayed load adjustment record should contain the following fields: device number, original access time, hit disturbance segment number, adjusted access time, and access time offset. All of the above data are arranged according to the original order of the loads in the delay item set to form the final output result, namely the load delay adjustment sequence.

[0041] See also Figure 4 , the scheduling screening module includes: The power characteristic calculation submodule obtains the current data and apparent power data of the data center air conditioning group during the steady-state operation cycle of the flywheel power supply, calculates the current mean and apparent power output standard deviation within the time period, and forms the power supply side benchmark comparison items to generate the air conditioning group operation characteristic parameters; In the process of collecting the operating information of the air-conditioning group in the flywheel power supply channel, the system first determines the continuous operating cycle of the data center in the steady-state power supply state. The detection period is generally set to 5 seconds, and isochronous sampling is performed at intervals of every 200 milliseconds. A total of 25 groups of sample data are obtained. For each group of samples, the instantaneous current value and apparent power value of the air-conditioning group are recorded respectively in amperes (A) and kilovolt-amperes (kVA). The current sequence obtained by sampling is such as 13.2A, 13.5A, 13.3A, 13.7A... After performing an average operation on this sequence, the average value of the air-conditioning current in this period is obtained. For example, the sum is 33 3A, the mean is 13.32A. At the same time, the standard deviation calculation is performed on the apparent power data series such as 3.9kVA, 4.0kVA, 3.8kVA, 4.1kVA... If the standard deviation is 0.12kVA, it means that the degree of power fluctuation is this value. The combination of these two values ​​is the typical power supply behavior characteristics of the air-conditioning group in the current period, which is used as the power supply side benchmark comparison item when subsequent equipment is connected. It is correspondingly recorded as the "power supply current benchmark mean" and "power output standard deviation" parameters, which are used to measure the consistency between the power supply load connection conditions and the current operating state, and finally generate the air-conditioning group operation characteristic parameters.

[0042] The device matching and comparison submodule extracts the starting current requirement and operating power factor of the currently connected operation and maintenance equipment based on the operating characteristic parameters of the air conditioning group. It calculates the deviation difference with the power supply side benchmark comparison items, and performs interval intersection judgment on the deviation difference set of all devices to obtain the equipment operation adaptation interval; Based on the operating characteristic parameters of the air-conditioning group, the starting current demand value and the power factor value in the operating state of all the operation and maintenance equipment to be connected are extracted, with the units being amperes (A) and the dimensionless power factor ratio value respectively. Then, the numerical difference between the starting current of each device and the above-mentioned power supply current benchmark average is calculated to obtain the current deviation between the current state of the current device in the startup phase and the power supply side, which is defined as the "starting current deviation". Similarly, the difference between the operating power factor and the average power factor of the air-conditioning group on the power supply side (estimated by the apparent power standard deviation and the equivalent real power) is processed and defined as the "power factor deviation". For example, if the starting current of device X is 15.0A and the power supply side benchmark is 13.32A, then the deviation is 15.0A. The deviation is 1.68A; if the device power factor is 0.94 and the power supply side benchmark is 0.88, the deviation is 0.06. Subsequently, the two deviations of all devices are summarized separately to form two sets. A mapping spectrum is established according to the numerical interval distribution, and an interval intersection judgment is performed to determine which devices have both current deviation and power factor deviation within the compatible range allowed by the system. For example, the system setting allows a current deviation of no more than ±2.0A and a power factor deviation of no more than ±0.08. In the above example, device X falls within the adaptation range. Continue to judge all devices in this way, filter out all devices within the double parameter intersection range, output their index and parameter identifier, and complete the identification of the device operation adaptation range.

[0043] The access load screening submodule extracts the device access attributes for the operation and maintenance devices in the device operation adaptation range, organizes the unique identification and operation parameters of each device, and generates an advance scheduling list for the operation and maintenance devices; For all devices selected within the obtained device operation adaptation range, the access attribute record for each device is further retrieved. This includes five basic pieces of information: device number, access priority, installation location identifier, access mode category (e.g., grid-connected / standalone), average operating time, and historical current peak value. By matching and extracting the device ID field from the database of the linked system operation and maintenance management module, a separate access attribute list record is constructed for each device. The previously calculated starting current deviation and power factor deviation parameter fields are attached to this record to form the final operating parameter group. This data collation is then performed on all devices in sequence to generate a list of devices for pre-scheduling preparation. A preliminary sorting is then performed by device priority, followed by detailed adjustments based on operating parameter stability (e.g., the sum of the absolute values ​​of the deviations, from small to large). Finally, a data structure sequence with unique identifiers, access parameters, and operating evaluation fields is generated. The output serves as the identification input for the scheduling system, forming the pre-scheduling list of operation and maintenance devices for the current scheduling cycle.

[0044] See also Figure 5 , the time reconstruction module includes: The time offset identification submodule extracts the time offset value and original access time of each load based on the load delay adjustment sequence and the advance scheduling list of operation and maintenance equipment, classifies the forward offset and reverse offset separately, and generates load access offset structure data; Based on the load delay adjustment sequence and the advance scheduling list of operation and maintenance equipment, each load item is traversed in turn, and the time difference between the original connection time and the adjusted connection time is extracted. The offset direction is determined based on the sign of the difference. If the adjusted time is later than the original connection time, it is classified as a positive offset, indicating a delayed load connection; if the adjusted time is earlier than the original connection time, it is classified as a negative offset, indicating an early load connection. For example, load A has an original connection time of 5.00 seconds and an adjusted time of 5.18 seconds, resulting in an offset of +0.18 seconds, placing it in the positive offset set. Load B has an original time of 6.50 seconds and an adjusted time of 6.42 seconds, resulting in an offset of -0.08 seconds, placing it in the negative offset set. After completing this classification for all loads, the positive offset loads are unified into a set of structured data, where each item contains the device number, original time, offset value, and direction identifier. The negative offsets also form a corresponding set, maintaining a consistent structure. These two sets together constitute the load connection offset structure data.

[0045] The access time rearrangement submodule replaces the original access timeline of the current main power supply channel as a whole based on the load access offset structure data, shifts the delayed items backward and the advanced items forward, rearranges the access positions according to the adjusted time sequence, and obtains the timeline sequence rearrangement result; Based on the load connection offset structure data, a comprehensive reconstruction of the original connection time series on the current main power supply channel is performed. The process is as follows: First, all loads with positive offsets are rearranged according to their offset time points, removed from their original positions in the original time series, and inserted into a new time series with their new connection times. Next, loads with negative offsets are processed, removed from their original time points, and inserted into the new time series with their advanced connection times. For example, load C's original time was 7.10 seconds, offset by +0.15 seconds, and updated to 7.25 seconds; load D's original time was 7.40 seconds, offset by -0.20 seconds, and updated to 7.20 seconds; and load E's original time was 7.30 seconds, remaining unchanged. After the update, the connection order becomes: D (7.20 seconds), C (7.25 seconds), E (7.30 seconds). The new sequence, adjusted based on the time size, is: D (7.20 seconds), C (7.25 seconds), E (7.30 seconds). Therefore, after the sorting is completed, the insertion position of each load is rearranged according to the actual update time, overwriting the original access table, and realizing the order rearrangement of the access scheduling sequence on the time axis.

[0046] The time index generation submodule calls the time axis sequence rearrangement result, extracts the new access time point corresponding to each load and its relative position number in the rearranged sequence, integrates the load unique identifier and time coordinate information, and generates a load control time mapping index table; After completing the order re-arrangement of the time series, the system further indexes the access time of each load and generates a corresponding time mapping structure. Each load records its updated access time point and calculates its relative position number in the new time series. For example, in the updated sequence, load X is the first access item, load Y is the second, and load Z is the third, corresponding to their positions in the sorted timeline. The unique identifier of each load is then combined with its updated access time point and sequence number to form a complete time-number mapping information. Taking load F as an example, its updated access time is 8.75 seconds, and it is ranked 5th in the sorted list. The record item is: load F, access time 8.75 seconds, position number 5. This step is performed for all loads, and the system finally outputs three structural information items including the number, time and position number of all valid access loads, forming a complete load control time mapping index table.

[0047] See also Figure 6 , the instruction update module includes: The time node extraction submodule calls all load identifiers and corresponding access time information in the load control time mapping index table, extracts the adjusted start time field, constructs an update sequence in chronological order, binds the relay action sequence number, and generates a control instruction time sorting result; In the load control time mapping index table, the unique load identifiers and corresponding new connection time information are sequentially read from all entries. The adjusted start-up time field of each load is used as the reference point for the next stage of the control system scheduling. These entries are sorted in ascending order by connection time value to construct a time update sequence. For example, if there are five loads with connection times of 7.20 seconds, 7.35 seconds, 7.60 seconds, 7.90 seconds, and 8.10 seconds, the system will sort them as follows: the first load has a connection time of 7.20 seconds, the second load has a connection time of 7.35 seconds, and so on to the fifth load. After sorting, a relay action sequence number is generated for each load based on its ranking position. For example, the first load binding number is 001, the second is 002, and so on to the fifth load, 005. Each instruction records three key information fields: load number, connection time, and relay instruction number. These form a standardized relay operation configuration item. These are accumulated to form a complete control instruction time sequence. This sequence is used to guide the control logic for the connection of various loads in the shared power supply path between the flywheel and battery.

[0048] The control logic reordering submodule completely replaces the relay triggering logic sequence in the main current channel shared by the flywheel and battery based on the control instruction time sequence results, maps the original relay instructions with the updated time sequence, and rearranges the control triggering order according to time sequence to generate a relay response reordering queue; Based on the control command time sorting results, the relay control logic originally deployed on the flywheel-battery shared main current channel is completely reordered. First, all existing control commands and their corresponding execution times are extracted from the original relay logic and compared item by item with the newly sorted command times. If the original load relay command differs from the updated time, the original control entry is marked invalid. The corresponding relay logic queue is then re-established according to the updated command sequence, ensuring that each relay action command is executed one by one according to the new activation time on the timeline. For example, if the original relay triggering time for load A is 7.50 seconds and is now adjusted to 7.20 seconds, the system will cancel the control action at the original 7.50-second period and create a new relay command at 7.20 seconds, with the action number 001. The same operation is performed for load B if the triggering time is changed from 8.00 seconds to 7.35 seconds. All the adjusted relay logic is recombined into a new time-based triggering sequence with no overlap, no jumps, and a fully updated sequence, ultimately generating a complete relay response reordering queue.

[0049] The trigger process generation submodule calls the relay response reordering queue, synchronously writes the relay actions and corresponding time sequence into the execution list of the control channel contacts, and merges all the adjusted control trigger behaviors into a unified task flow to generate a trigger control scheduling instruction flow, which is used to coordinate the load connection timing of the flywheel energy storage and battery in the hybrid UPS structure, and realize the power supply trigger sequence control in the main current channel of the data center; After the relay action rescheduling is complete, the system enters the trigger control generation phase. This phase calls all relay action items in the entire relay response rescheduling queue and their associated time sequence. Each specific action is written into the relay contact execution list of the main control channel. This list structure is periodically read by the central control unit and control commands are issued. During this writing process, the system ensures that the action time field is synchronized with the clock triggering the actual control signal with millisecond accuracy. For example, if the first action is triggered at 7:20 seconds, the system will begin preparing for signal conversion at 7:199 seconds and officially close the relay at 7:200 seconds. Subsequently, all relay operations and their corresponding execution time sequences are organized into a standardized format and encapsulated into a unified scheduling task flow. This task flow records the logical sequence of all control events, the execution object, the command number, and the trigger time field. It serves as the primary scheduling vehicle for the power supply control system during the operation phase, ultimately forming a trigger control scheduling instruction flow that can be issued to the execution layer. This instruction stream structure can be directly mapped to the dual-channel power supply control logic of flywheel energy storage and battery in the hybrid UPS structure, triggering the control nodes one by one in the main current channel according to the adjustment order, driving the load access action to be carried out in an orderly manner within a unified time frame, and fully covering the power supply sequence control process when multiple loads are switched concurrently in the data center.

[0050] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A flywheel energy storage and battery hybrid UPS system for a data center, characterized in that: The system comprises: The load identification module obtains the current speed and output current of the flywheel and performs synchronous slope analysis to extract the starting current threshold of the current load to be connected. It then filters out devices whose starting current threshold is greater than the flywheel power supply limit and marks the load delay item set. The disturbance correction module obtains the change trajectory of the flywheel output power and the battery voltage response and performs cluster division, and performs a backshift processing on the original access time point of the device in the load delay item set according to the division result to generate a load delay adjustment sequence; The scheduling and screening module obtains the current mean and apparent power output standard deviation of the data center air conditioning group during the steady-state operation cycle of the flywheel power supply, compares it with the operation and maintenance equipment currently to be connected, and selects the advance scheduling list of operation and maintenance equipment; The time reconstruction module reorganizes the time axis of the load in the main power supply channel of the data center based on the load delay adjustment sequence and the advance scheduling list of the operation and maintenance equipment to form a load regulation time mapping index table; The instruction update module updates the trigger control process in the main current channel of the data center combined with the flywheel and the battery through the load regulation time mapping index table to form a trigger control scheduling instruction flow.

2. The flywheel energy storage and battery hybrid UPS system for a data center according to claim 1, characterized in that: The load delay item set includes devices whose starting current threshold exceeds the flywheel power supply limit, loads to be connected that need to be delayed, and load targets identified by synchronous slope analysis. The load delay adjustment sequence includes the offset value of the original access time point, the disturbance group identifier corresponding to the clustering division, and the load mapping item corresponding to the power disturbance peak. The operation and maintenance equipment advance scheduling list includes equipment that meets the current mean requirements, equipment that meets the apparent power output standard deviation conditions, and operation and maintenance equipment consistent with the operating characteristics of the air-conditioning group. The load control time mapping index table includes the access time update items in the operation and maintenance equipment advance scheduling list, the access time adjustment items in the load delay adjustment sequence, and the time axis arrangement items after unified mapping. The trigger control scheduling instruction flow includes the load access sequence update instruction, the trigger control signal in the main current channel, and the scheduling trigger action corresponding to the time mapping index.

3. The flywheel energy storage and battery hybrid UPS system for a data center according to claim 1, characterized in that: The load identification module includes: The speed and current extraction submodule obtains the continuous sampling data of the current speed and current output of the flywheel, calculates the speed change rate and current change rate within each sampling period, and generates a set of joint speed and current change rates; The slope mutation determination submodule uses a local weighted regression algorithm to perform trend fitting based on the speed and current joint change rate set, performs continuous period slope gradient calculation on the fitting residual sequence, and determines whether the residual gradient rises continuously and exceeds the average background fluctuation amplitude of the trend segment, thereby generating a trend slope mutation identification interval; The power supply limit screening submodule extracts the starting current threshold of the load to be connected according to the load identified in the trend slope mutation identification interval, compares it with the current power supply limit value of the flywheel, screens all target loads whose starting current threshold is greater than the power supply limit value, and generates a set of load delay items.

4. The flywheel energy storage and battery hybrid UPS system for a data center according to claim 3, characterized in that: The disturbance correction module includes: The power feature extraction submodule obtains the change trajectory of flywheel output power and battery voltage response, extracts the flywheel power change rate, battery voltage mutation point time, and the response delay of the intersection of the two, and performs normalization processing to generate a set of disturbance interval feature parameters; The interleaved disturbance clustering submodule uses a spectral clustering algorithm to cluster the voltage and current mutation points generated during the alternating power supply process according to the disturbance interval characteristic parameter set, extracts the overlapping sections in the clustering results where the disturbance intensity is greater than the average disturbance threshold, and obtains the disturbance peak cluster distribution information; The access time adjustment submodule calls the original access time point of each device in the load delay item set based on the disturbance peak cluster distribution information, maps it with the disturbance duration of the corresponding overlapping segment, performs time point backward offset processing, and generates a load delay adjustment sequence.

5. The flywheel energy storage and battery hybrid UPS system for a data center according to claim 4, characterized in that: The scheduling screening module includes: The power characteristic calculation submodule obtains the current data and apparent power data of the data center air conditioning group during the steady-state operation cycle of the flywheel power supply, calculates the current mean and apparent power output standard deviation within the time period, and forms the power supply side benchmark comparison items to generate the air conditioning group operation characteristic parameters; The device matching and comparison submodule extracts the starting current requirement and operating power factor of the currently connected operation and maintenance equipment based on the operating characteristic parameters of the air conditioning group, calculates the deviation difference with the power supply side benchmark comparison item, and performs interval intersection judgment on the deviation difference set of all devices to obtain the equipment operation adaptation interval; The access load screening submodule extracts the device access attributes for the operation and maintenance devices in the device operation adaptation range, organizes the unique identification and operation parameters of each device, and generates an advance scheduling list for the operation and maintenance devices.

6. The flywheel energy storage and battery hybrid UPS system for a data center according to claim 5, characterized in that: The time reconstruction module includes: The time offset identification submodule extracts the time offset value and original access time of each load based on the load delay adjustment sequence and the advance scheduling list of the operation and maintenance equipment, classifies the forward offset and reverse offset respectively, and generates load access offset structure data; The access time rearrangement submodule replaces the original access timeline of the current main power supply channel as a whole according to the load access offset structure data, shifts the delayed items backward and the advanced items forward, rearranges the access positions according to the adjusted time sequence, and obtains the time axis sequence rearrangement result; The time index generation submodule calls the time axis sequence rearrangement result, extracts the new access time point corresponding to each load and the relative position number in the rearranged sequence, integrates the load unique identifier and time coordinate information, and generates a load control time mapping index table.

7. The flywheel energy storage and battery hybrid UPS system for a data center according to claim 6, characterized in that: The instruction update module includes: The time node extraction submodule calls all load identifiers and corresponding access time information in the load control time mapping index table, extracts the adjusted start time field, constructs an update sequence in chronological order and binds the relay action sequence number to generate a control instruction time sorting result; The control logic rearrangement submodule completely replaces the relay triggering logic sequence in the main current channel shared by the flywheel and the battery based on the control instruction time sequence, maps the original relay instructions with the updated time sequence, and rearranges the control triggering sequence according to the time sequence to generate a relay response rearrangement queue; The trigger process generation submodule calls the relay response reordering queue, synchronously writes the relay actions and the corresponding time sequence into the execution list of the control channel contacts, and merges all the adjusted control trigger behaviors into a unified task flow to generate a trigger control scheduling instruction flow for coordinating the load access timing of the flywheel energy storage and the battery under the hybrid UPS structure, thereby realizing the power supply trigger sequence control in the main current channel of the data center.

Images