Calculation load power demand data acquisition method based on flywheel energy storage system

Abstract

The invention relates to the technical field of data acquisition, and discloses a computing power load power demand data acquisition method based on a flywheel energy storage system. The method comprises the following steps: firstly, acquiring related data of the flywheel energy storage system, and performing classification, database construction and dimension reduction processing; secondly, extracting a load fluctuation mode, calculating power loads in a classified manner, generating power demand parameter characteristics, and automatically matching and acquiring parameter configuration; adjusting a parameter range according to a power grid dispatching demand to generate an acquisition instruction; updating the database, adjusting the load prediction model, judging the applicability of a newly added load mode, and expanding a configuration list; and finally, constructing a visual monitoring interface, encrypting storage data and setting authority control. The method can accurately collect computing power load power demand data, adapts to load dynamic changes, meets the power grid dispatching demand, improves the intelligence, safety and efficiency of data collection, and assists an electric power system in optimized operation.

Description

Technical Field

[0001] The present invention relates to the technical field of data acquisition, and specifically to a method for acquiring power demand data of computing power load based on a flywheel energy storage system. Background Art

[0002] With the acceleration of the digitalization process, computing power infrastructure has become increasingly important in modern society. The scale of computing power facilities such as data centers and cloud computing platforms is constantly expanding, and their power consumption is also continuously rising, becoming an important load in the power system. Accurately collecting and mastering the power demand data of computing power load is crucial for the rational planning, dispatching, and optimal operation of the power system.

[0003] Traditional data acquisition methods have many limitations when dealing with the power demand of computing power load. The computing power load has significant dynamic change characteristics. Its power demand is affected by various factors such as business volume fluctuations and user usage habits, showing complex change laws. For example, during promotional activities on Internet e-commerce platforms, the data processing volume increases explosively, the computing power load increases instantaneously, and the power demand also fluctuates greatly. Traditional data acquisition methods are difficult to capture this rapid and complex change in real time and accurately, resulting in the collected data being unable to truly reflect the actual power demand of the computing power load.

[0004] In terms of combination with the flywheel energy storage system, traditional methods have deficiencies. As a new type of energy storage technology, the flywheel energy storage system has advantages such as fast response speed and high energy density, and can be used in the power system to regulate load fluctuations and improve power quality. However, traditional data acquisition methods do not fully consider the characteristics of the flywheel energy storage system and cannot effectively match the acquisition parameters with the operating state of the flywheel energy storage system. For example, during the charge and discharge process of the flywheel energy storage system, parameters such as its rotation speed and energy storage capacity will change, while traditional acquisition methods cannot dynamically adjust the acquisition parameters according to these changes, resulting in the accuracy and effectiveness of the collected data being affected.

[0005] The requirements of the power grid dispatching for the power demand data of computing power load are getting higher and higher. The power grid needs to perform accurate load forecasting and dispatching based on these data to ensure the safe and stable operation of the power system. However, traditional data acquisition methods cannot meet the strict requirements of the power grid dispatching for data accuracy, real-time performance, and integrity in different scenarios. During peak electricity consumption periods, the power grid needs to quickly obtain accurate power demand data of computing power load in order to reasonably allocate power resources and avoid problems such as power shortages or overloads. However, the data collected by traditional methods often have problems such as delays and large errors, making it difficult to meet the actual needs of the power grid dispatching.

[0006] With the continuous expansion of the scale and the increase in complexity of the power system, the amount of data has grown exponentially. Traditional data acquisition methods also face challenges in data processing and storage. They lack effective means of data dimensionality reduction, classification, and feature extraction, resulting in high costs for data storage and transmission, and it is difficult to quickly extract valuable information from the vast amount of data, affecting the overall operation efficiency of the power system and the scientific nature of decision-making. Summary of the Invention

[0007] The purpose of the present invention is to provide a data acquisition method for computing power load power demand based on a flywheel energy storage system to solve the problems presented in the above-mentioned background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solution: A data acquisition method for computing power load power demand based on a flywheel energy storage system, the method comprising: Obtain the real-time rotational speed, energy storage capacity, and power output data of the flywheel energy storage system, classify the power demand data, construct a power feature database, and perform data dimensionality reduction processing; Extract the load fluctuation pattern according to the power feature database, and classify the computing power load to be collected; according to the dynamic change characteristics of the computing power load, generate features for the power demand parameters under each load classification; according to the usage preferences and configuration conditions of the system operator, automatically match and adapt the appropriate acquisition parameter configuration for the real-time demands of different loads. Dynamically adjust the parameter range in the configuration list according to different grid dispatching requirements to generate an acquisition instruction that accurately describes the power demand. Update the power feature database, adjust the corresponding load forecasting model according to the database update result, and determine whether the new load pattern exceeds the applicable range of the current acquisition parameters. Expand the acquisition parameter configuration list according to the new load pattern and apply it to real-time data acquisition.

[0009] Preferably, the obtaining the real-time rotational speed, energy storage capacity, and power output data of the flywheel energy storage system, classifying the power demand data, constructing a power feature database, and performing data dimensionality reduction processing includes: Deploy multi-type sensor networks to collect data on the rotational speed, temperature of the flywheel rotor, and the power change of the energy storage unit; obtain the real-time power output and load demand curve through the power grid dispatching platform; integrate historical operation logs and fault records to verify the integrity and consistency of the data; divide the power demand data into base load, peak load, and fluctuating load according to the load type, and add timestamp tags to each type of data; use a distributed storage system to establish the table structure of the power feature database, and store information such as rotational speed, capacity, power, and load classification; based on the principal component analysis method, transform the power demand data into low-dimensional feature vectors; the features after dimensionality reduction include the amplitude of load fluctuation, duration, periodic law, and the state of associated devices; The method of transforming the power demand data into low-dimensional feature vectors based on the principal component analysis method specifically includes: Perform normalization processing on the original power data to eliminate the dimensional difference; extract the local features of the time series data through a sliding window mechanism to construct a multi-dimensional feature matrix; calculate the eigenvectors and eigenvalues using the covariance matrix, and select the principal components whose cumulative contribution rate exceeds the preset threshold; input the selected principal components as low-dimensional feature vectors into the load prediction model.

[0010] Preferably, the method of extracting the load fluctuation pattern according to the power feature database and classifying the computing power load to be collected includes: Define the acquisition priority and response time constraint of the computing power load; input the key parameters of the load fluctuation pattern into the power feature database for matching and retrieval, and screen the historical data related to the target computing power load; for each historical data, use wavelet transform to extract the frequency domain features and mutation point information of the load waveform; cluster the extracted features according to the load type, fluctuation intensity, and duration to form a computing power load classification rule base.

[0011] Preferably, the method of generating features for the power demand parameters under each load classification according to the dynamic change characteristics of the computing power load includes: Based on the load classification rule base, determine the dynamic change interval of the computing power load; combine the power grid operation standards and the technical indicators of the flywheel energy storage system to set the threshold of the power demand parameters under each load classification; collect the device operation state data and external environment variables to construct a multi-dimensional parameter correlation matrix; use the long short-term memory network model to extract features and predict trends for the time series parameters; generate a load feature coding table according to the prediction results for the parameter configuration of the real-time acquisition instruction.

[0012] Preferably, the method of automatically matching and adapting the acquisition parameter configuration according to the usage preferences and configuration conditions of the system operator for the real-time requirements of different loads includes: Analyze the historical configuration records and parameter adjustment frequencies of the operator, extract their common parameter combinations and adjustment strategies; limit the range and step size of configurable parameters according to the operator's technical level and equipment permissions; combine the current grid load status and the remaining capacity of the flywheel energy storage system to dynamically generate a candidate set of parameter configurations; use the analytic hierarchy process to rank the candidate set by priority and recommend the optimal parameter configuration plan; Based on the Euclidean distance similarity calculation method, recommend the acquisition parameter configuration list in similar scenarios in historical data, specifically including: Compare the load feature vectors in the historical scenario with the current scenario item by item, calculate the Euclidean distance and generate a similarity score; screen the historical scenarios with a similarity higher than the threshold according to the score, and extract their parameter configuration plans; fuse the configuration plans of multiple similar scenarios to generate a recommended list after weighted averaging.

[0013] Preferably, dynamically adjust the parameter range in the configuration list according to different grid dispatching requirements to generate an acquisition instruction that accurately describes the power demand, including: Analyze the load regulation target and time window requirements in the grid dispatching instruction; divide multiple regulation stages and set stage parameter constraints according to the output results of the load prediction model; use the particle swarm optimization algorithm to iteratively adjust the parameter range to ensure meeting the dynamic response requirements of grid dispatching; encode the adjusted parameter range into a standardized acquisition instruction and send it to the flywheel energy storage control system; Group the grid dispatching requirements based on the density clustering algorithm to determine the parameter thresholds for different groups, specifically including: Extract the load regulation amplitude, response speed and duration characteristics in the dispatching requirements; calculate the data point density in the feature space, identify the high-density areas as the core clustering clusters; merge adjacent density-reachable clustering clusters and remove the noise data; divide the dispatching requirement groups according to the clustering results and set the upper and lower limits of the parameter thresholds for each group.

[0014] Preferably, update the power feature database, adjust the corresponding load prediction model according to the database update result, and judge whether the new load pattern exceeds the applicable range of the current acquisition parameters, including: Access the real-time grid monitoring data and the operation log of the flywheel energy storage system to update the load pattern samples in the power feature database; extract the features of the new load pattern and compare them with the existing patterns for similarity; based on the random forest regression model, predict the impact degree of the new pattern on the grid frequency stability; construct a gradient boosting decision tree model to judge whether the new pattern can be adapted by adjusting the flywheel speed and energy storage capacity; if it exceeds the current parameter range, trigger the parameter expansion process; Based on the random forest regression model, predict the impact of the new load pattern on the grid stability, specifically including: Collect data on grid frequency deviation, voltage fluctuations, and load mutation events as training samples; construct multiple regression decision trees and output a stability score through a voting mechanism; input the characteristics of the new load pattern into the model to obtain its predicted score for grid stability, and judge the risk level according to the score threshold.

[0015] Preferably, the step of expanding the acquisition parameter configuration list according to the new load pattern and applying it to real-time data acquisition includes: Extract the key characteristics of the new load pattern, including the fluctuation period, peak power, and duration; calculate the feasible range of parameter expansion according to the maximum charge and discharge rate and capacity limit of the flywheel energy storage system; use the genetic algorithm to perform multi-objective optimization on the parameter combination and screen the optimal solution set that meets the system constraints; add the optimized parameters to the configuration list and synchronously update them to the real-time acquisition system; Based on the genetic algorithm optimization strategy, determine the constraint conditions of parameter configuration and perform dynamic matching with the system operation state, specifically including: Define the objective function of parameter optimization, including minimizing the load response time and equalizing the system loss; initialize the parameter population, generate the offspring population through crossover and mutation operations; calculate the fitness value of each individual, and retain the individuals with high fitness values for the next generation of iteration; finally, output the optimal parameter configuration scheme that meets multiple constraints.

[0016] Preferably, the method further includes: Construct a visual monitoring interface for power demand data acquisition, integrating load curves, parameter configuration status, and system alarm information; allow operators to manually adjust the acquisition frequency and data accuracy through interactive controls; use a time-series database to store the acquired data, supporting millisecond-level data query and retrospective analysis.

[0017] Preferably, the method further includes: Perform encrypted transmission and storage on the acquired data, and use national encryption algorithms to protect sensitive information end-to-end; set up a multi-level access permission control mechanism to limit the operation permissions of unauthorized users.

[0018] Compared with the prior art, the beneficial effects of the present invention are: At the data collection and processing level, a multi-type sensor network is deployed to collect data on the rotational speed, temperature of the flywheel rotor, and the change in the power of the energy storage unit. Combined with the power grid dispatching platform, the real-time power output power and load demand curve are obtained, and the historical operation logs and fault records are integrated to verify the integrity and consistency of the data. This multi-source data fusion method ensures the comprehensiveness and reliability of the collected data. The power demand data is divided by load type and timestamp tags are added, facilitating subsequent classification management and analysis. The principal component analysis method is used for data dimensionality reduction, which reduces the complexity of data processing, improves data processing efficiency, and reduces storage and transmission costs while retaining key information.

[0019] Regarding the load fluctuation pattern and computing power load classification, the present invention defines the collection priority and response time constraints, uses the power feature database to match and retrieve historical data, and adopts wavelet transform to extract frequency domain features and mutation point information and cluster to form a classification rule library. This makes the classification of computing power loads more scientific and accurate, can better adapt to the characteristics of different types of computing power loads, and provides a solid foundation for the subsequent generation of power demand parameter features and the configuration of collection parameters.

[0020] In terms of generating power demand parameter features, based on the load classification rule library, the dynamic change interval is determined, the parameter threshold is set in combination with the power grid operation standard and the technical indicators of the flywheel energy storage system, a multi-dimensional data correlation matrix is constructed, and the long short-term memory network model is used for feature extraction and trend prediction to generate a load feature coding table, which can grasp the dynamic changes of the power demand of the computing power load in real time and accurately, providing strong support for precise collection.

[0021] Automatically match the collection parameter configuration according to the system operator's usage preferences and configuration conditions. Generate a candidate set by analyzing the historical configuration records and adjustment frequency, combining the current power grid and energy storage system status, and sort it using the analytic hierarchy process. At the same time, recommend a list of historical similar scenario configurations based on the Euclidean distance similarity, which improves the intelligence and personalization level of parameter configuration, reduces the complexity and error of manual configuration, and enhances the usability and operation efficiency of the system.

[0022] When responding to the power grid dispatching requirements, parse the dispatching instructions, divide the adjustment stage to set parameter constraints, use the particle swarm optimization algorithm to adjust the parameter range and encode it into a collection instruction, and at the same time determine the parameter threshold by grouping based on the density clustering algorithm, which can quickly and accurately respond to the dynamic requirements of the power grid dispatching, ensure the stable operation of the power system, improve the allocation efficiency of power resources, and reduce the operation risk of the power grid.

[0023] Update the power feature database and adjust the load forecasting model. Use the random forest regression model and the gradient boosting decision tree model to evaluate the impact of the new load pattern and determine whether it is suitable, and trigger the parameter expansion process in a timely manner, ensuring the timeliness of the data and the accuracy of the model, enabling the system to continuously adapt to the changes of the new load pattern.

[0024] Expand the acquisition parameter configuration list based on the new load pattern, and use the genetic algorithm to optimize the parameters to ensure that the parameter configuration meets the system constraints, further improving the adaptability and accuracy of data acquisition. In addition, construct a visual monitoring interface to facilitate the operator to monitor and manually adjust, use a time series database to store data to support efficient query and retrospective analysis; encrypt the data during transmission and storage and set up a multi-level access control mechanism to ensure the security and privacy of the data. In summary, the present invention improves the quality and efficiency of the data acquisition of the computing power load power demand from multiple aspects, providing strong technical support for the optimal operation, precise scheduling and safety management of the power system. Brief Description of the Drawings

[0025] Figure 1 It is the working principle diagram of the method for collecting computing power load power demand data based on the flywheel energy storage system described in the present invention; Figure 2 It is the flow chart for obtaining flywheel energy storage system data and data processing; Figure 3 It is the flow chart for updating the power feature database and judging the new load pattern; Figure 4 It is the flow chart for expanding and applying the acquisition parameter configuration based on the new load pattern. Detailed Embodiments

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the 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. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0027] Please refer to Figures 1-4 , the present invention provides a method for collecting computing power load power demand data based on a flywheel energy storage system, and the specific implementation steps are as follows: Obtain the real-time speed, energy storage capacity and power output data of the flywheel energy storage system, classify the power demand data, construct a power feature database, and perform data dimensionality reduction processing; Extract the load fluctuation pattern according to the power feature database and classify the computing power load to be collected; generate features for the power demand parameters under each load classification according to the dynamic change characteristics of the computing power load; automatically match and adapt the acquisition parameter configuration for the real-time demands of different loads according to the usage preferences and configuration conditions of the system operator. Dynamically adjust the parameter range in the configuration list according to different grid dispatching requirements to generate an acquisition instruction that accurately describes the power demand. Update the power feature database, adjust the corresponding load forecasting model according to the database update result, and judge whether the new load pattern exceeds the applicable range of the current acquisition parameters. Expand the acquisition parameter configuration list according to the new load pattern and apply it to real-time data acquisition.

[0028] Embodiment 1: The system obtains the real-time rotation speed, energy storage capacity and power output data of the flywheel energy storage system, classifies the power demand data, constructs a power feature database, and performs data dimensionality reduction processing. The specific process is as follows.

[0029] In the data acquisition stage, to accurately obtain the operation state data of the flywheel energy storage system, it is necessary to deploy multi-type sensor networks. For the flywheel rotor, install a high-precision rotation speed sensor, which monitors the rotation speed of the flywheel rotor in real time through the principle of photoelectric induction or magnetoelectric induction, converts the mechanical rotation signal into an electrical signal or a digital signal, and acquires data at a sampling frequency of milliseconds to ensure that the subtle changes in the rotation speed can be captured; at the same time, configure a temperature sensor, using a thermocouple or a thermal resistance temperature sensor, installed close to the surface or key parts of the flywheel rotor to collect the temperature data during the operation of the rotor in real time, and avoid affecting the performance of the flywheel or causing safety problems due to excessive temperature. For the energy storage unit, deploy a power sensor, which calculates the real-time power change data of the energy storage unit by measuring parameters such as battery voltage and current, combined with specific algorithms such as the ampere-hour integration method or the coulomb counting method, to accurately reflect the charge and discharge state of the energy storage unit.

[0030] The grid dispatching platform is the key source for obtaining the real-time power output power and load demand curve. The system establishes a stable and reliable communication connection, such as using industrial Ethernet, fiber optic communication, etc. to interact data with the grid dispatching platform. Using standardized communication protocols, such as IEC61850 or Modbus protocol, receive the real-time power output power data sent by the grid dispatching platform, and these data contain the power magnitude output by the flywheel energy storage system to the grid at different time nodes; at the same time, obtain the load demand curve, which intuitively presents the change trend of the power demand of the grid within a certain period of time, providing a basic basis for subsequent data processing.

[0031] To ensure data quality, the system integrates historical operation logs and fault records. The historical operation logs detail information such as the past operation parameters and operation records of the flywheel energy storage system; the fault records include various fault phenomena, occurrence times, handling measures, etc. that occur during the system operation. The system uses data verification algorithms to compare and analyze newly collected data with historical data, check the integrity of the data to ensure that there is no data loss or omission; verify the consistency of the data to prevent the occurrence of conflicting data and ensure the accuracy of subsequent data processing and analysis.

[0032] After completing data collection and verification, the power demand data is classified. It is divided into base load, peak load, and fluctuating load according to the load type. The base load refers to the relatively stable and less fluctuating power demand in the power grid under normal operation. This type of load is usually generated by some continuously operating equipment with relatively stable power consumption, such as some continuously producing industrial equipment, long-term lighting facilities, etc.; the peak load shows a significant increase in power within a short period of time, commonly seen during peak electricity consumption periods, such as when air conditioners are turned on intensively in summer, and when there is intensive electricity consumption in large event venues; the fluctuating load is the power demand with irregular power changes, which may be generated by some intermittently working equipment or equipment greatly affected by external factors, such as new energy power generation equipment like wind power and photovoltaic power generation, whose power output fluctuates with natural conditions. During the classification process, a time stamp label accurate to the second level is added to each type of data to facilitate subsequent analysis of the changes and interrelationships of various loads at different time points.

[0033] When constructing the power characteristic database, a distributed storage system is used to establish the table structure. The distributed storage system combines the advantages of a distributed file system and a distributed database, and has characteristics such as high scalability, high reliability, and high performance. In the table structure design, fields are respectively set to store the rotational speed data of the flywheel rotor, the capacity data of the energy storage unit, the power output data, and the load classification information, etc. Through reasonable table structure design, efficient data storage and rapid retrieval are achieved, facilitating subsequent data analysis and processing.

[0034] Dimensionality reduction processing is performed on the power demand data based on the principal component analysis method. First, the original power data is normalized. Since there are differences in the dimensions and value ranges of different parameters in the original data, such as the unit of rotational speed being revolutions per minute and the unit of power being kilowatts, through normalization, data with different dimensions is mapped to the same interval to eliminate the influence of dimensional differences on data analysis. Normalization methods can use min-max normalization or Z-score normalization, etc. Then, the sliding window mechanism is used to extract the local features of the time series data at fixed time intervals to construct a multi-dimensional feature matrix. For example, with a 1-minute sliding window, statistical features such as the mean, variance, maximum value, and minimum value of data such as rotational speed and power within this time period are extracted as a row of data in the multi-dimensional feature matrix. Then, the eigenvectors and eigenvalues are calculated using the covariance matrix, and the covariance matrix reflects the correlation between the features. By performing eigen-decomposition on the covariance matrix, the eigenvectors and eigenvalues are obtained. A cumulative contribution rate threshold is set, such as 85% or 90%, and the principal components with a cumulative contribution rate exceeding this threshold are selected. These principal components contain the main information of the original data. The selected principal components are used as low-dimensional feature vectors, which contain key information such as the load fluctuation amplitude, duration, periodic law, and associated equipment status, and are input into the load prediction model to provide data support for subsequent load analysis and prediction.

[0035] After completing the data dimensionality reduction processing, the system continues to extract the load fluctuation patterns based on the power feature database and classify the computing power loads to be collected; according to the dynamic change characteristics of the computing power loads, generate features for the power demand parameters under each load classification; according to the usage preferences and configuration conditions of the system operator, automatically match the appropriate collection parameter configurations for the real-time demands of different loads; according to different grid dispatching requirements, dynamically adjust the parameter ranges in the configuration list to generate collection instructions that accurately describe the power demands; update the power feature database, adjust the corresponding load prediction model according to the database update results, and judge whether the new load pattern exceeds the applicable range of the current collection parameters; expand the collection parameter configuration list according to the new load pattern and apply it to real-time data collection. In subsequent links, based on the low-dimensional feature vectors after data dimensionality reduction processing, accurate collection and processing of the computing power load power demand data are realized through a series of algorithms and models to ensure that the system can efficiently and accurately meet the requirements of power grid operation and management.

[0036] Embodiment 2: When classifying the power demand data, constructing the power feature database, and performing data dimensionality reduction processing, the specific operations are as follows.

[0037] The deployment of a multi-type sensor network is the basis for obtaining accurate data. Among them, the rotational speed sensor uses a high-precision Hall effect sensor. By detecting the pulse signal generated by the rotation of the magnet on the flywheel rotor, it converts mechanical rotation into an electrical signal. This sensor has the characteristics of high resolution and fast response, and can accurately capture the rotational speed changes of the flywheel rotor under different working conditions. The sampling frequency is set to 1000Hz to ensure accurate acquisition of rotational speed data even under high-speed rotation. The temperature sensor selects a platinum resistance temperature sensor, which has the characteristics of high measurement accuracy and good stability. It uses a contact measurement method and is closely attached to the key parts of the flywheel rotor surface to monitor the temperature changes of the rotor in real time. To avoid interference from environmental factors, the sensor is wrapped with heat-insulating materials and is also equipped with a signal conditioning circuit to amplify and filter the collected temperature signal to improve the signal quality. For the energy storage unit, an intelligent power monitoring module is deployed. This module integrates a voltage sensor, a current sensor, and a microprocessor. By measuring the voltage and current of the battery in real time and combining the charge and discharge characteristic curves of the battery, it accurately calculates the remaining power and charge and discharge status of the energy storage unit.

[0038] The power grid dispatching platform conducts data interaction with the system through an industrial Ethernet interface. It adopts the IEC61850 communication protocol to ensure the reliability and real-time performance of data transmission. The system sends data requests to the power grid dispatching platform at regular intervals and receives real-time power output power data and load demand curves. The power output power data includes parameters such as active power, reactive power, and power factor. These parameters are stored in the form of a time series for subsequent analysis. The load demand curve reflects the changing trend of the power grid's power demand at different time periods. The system analyzes and processes it to extract key feature points, such as peak demand, valley demand, and demand change rate.

[0039] To ensure data quality, the system integrates historical operation logs and fault records. The historical operation logs are stored in a distributed file system and contain all operation parameters and operation records since the system was put into use. The fault records are stored in a relational database in a structured manner, recording the occurrence time, fault phenomenon, handling process, and final result of each fault. The system has developed a special data integration module. This module first cleans the historical data to remove duplicate data and invalid data, and then performs format standardization processing to convert data from different sources and different formats into a unified data structure. Finally, it uses a data verification algorithm to verify the integrity and consistency of the integrated data to ensure the accuracy and reliability of the data.

[0040] In terms of load type classification, the system classifies according to the changing characteristics of power demand. The identification of base load is based on long-term data analysis. By calculating the mean and standard deviation of the load, the load with a fluctuation range within ±10% of the mean is defined as the base load. This type of load mainly comes from continuously operating industrial equipment, lighting systems, and basic electrical equipment in residential life. The determination of peak load combines time factors and the power change rate. When the power of the load rises by more than 50% of the mean within a short period (such as within 15 minutes), it is determined as the peak load. The identification of fluctuating load is relatively complex. The system uses the frequency domain analysis method to perform Fourier transform on the load data, extract the high-frequency components, and defines the load with a relatively large proportion of high-frequency components as the fluctuating load. Add timestamp tags accurate to the millisecond level to each type of data. The timestamp uses Coordinated Universal Time (UTC) to ensure the consistency of data time in the distributed system.

[0041] The distributed storage system adopts the combination of Hadoop Distributed File System (HDFS) and Apache HBase distributed database. HDFS is responsible for storing the original sensor data and historical operation logs. It has the characteristics of high fault tolerance and high throughput, and can handle the storage and read / write operations of massive data. HBase is used to store structured power characteristic data, such as rotational speed, capacity, power, and load classification information, etc. In the table structure design, the combination of partition table and index table is adopted. Partition according to time and load type to improve the data query efficiency. At the same time, establish multi-level indexes, including time index, load type index, and parameter value index, to support fast data retrieval and statistical analysis.

[0042] In data dimensionality reduction processing, the original power data is processed based on the principal component analysis method. First, perform the normalization operation. Adopt the Z-score normalization method to convert each parameter value into a standard normal distribution, eliminating the influence of different parameter dimensions. The sliding window mechanism divides the time series data at a fixed time interval (such as 1 minute). Extract various statistical features within each window, including mean, variance, maximum value, minimum value, median, and quartiles, etc. These features constitute a multi-dimensional feature matrix, and each row of the matrix represents the power data features within a time window. By calculating the covariance matrix of the feature matrix, analyze the correlation between each feature. Perform eigenvalue decomposition on the covariance matrix to obtain eigenvectors and eigenvalues. According to the size of the eigenvalues, select the principal components with a cumulative contribution rate exceeding 90%. These principal components contain the main information of the original data. Use them as low-dimensional feature vectors to obtain low-dimensional feature vectors containing key information such as load fluctuation amplitude, duration, periodic law, and associated equipment status. The dimension of this low-dimensional feature vector is significantly reduced compared to the original data, reducing the complexity of data storage and processing, while retaining the main features of the data, providing an efficient data representation for subsequent load analysis and prediction.

[0043] After the data dimensionality reduction is completed, the low-dimensional feature vectors are input into the subsequent analysis module. The system uses these feature vectors for load pattern recognition and classification. By comparing historical data with the current features, it can quickly and accurately identify the type and characteristics of the load. At the same time, these feature vectors also provide inputs for the load prediction model. Based on historical data and current features, the model predicts the future trend of power demand changes, providing decision-making support for power grid scheduling and energy storage system control. In addition, the system also uses low-dimensional feature vectors for anomaly detection. By setting the normal feature range, it can timely detect abnormal situations in the power system, such as equipment failures and load mutations, and issue warning signals to ensure the safe and stable operation of the power system.

[0044] Example 3: In the process of classifying and generating features of computing power load, it is necessary to clarify the acquisition priority and response time constraints of computing power load first. The acquisition priority is set according to the importance of computing power devices in the business process. For key computing power devices that ensure the operation of core services, a high priority is given, and the system is required to collect their power demand data first; for computing power devices of auxiliary or non-critical services, a relatively low priority is set. The response time constraint is determined according to the requirements of data real-time in different business scenarios. For example, for services with extremely high requirements for data real-time, such as real-time transaction processing and online games, the response time is required to be controlled within milliseconds; for services such as data processing and background analysis, the response time can be relaxed to seconds or minutes.

[0045] The key parameters of the load fluctuation pattern are input into the power feature database for matching and retrieval. These key parameters cover the load change frequency (unit: times / minute, indicating the number of times the load changes per unit time), amplitude (unit: kilowatt, indicating the difference in power before and after the load change), duration (unit: minute, indicating the duration of the load in a certain specific state), etc. The system searches for historical data related to the target computing power load in the power feature database based on these parameters. During the retrieval process, an efficient data indexing mechanism is adopted. By pre-classifying and constructing indexes for historical data according to the key parameters, the data records that meet the conditions can be quickly located.

[0046] For each retrieved historical data, the load waveform is analyzed using wavelet transform. Wavelet transform is a mathematical tool that converts time domain signals into a joint representation of frequency domain and time domain. Its basic principle is to obtain the characteristics of the signal at different frequencies and time resolutions by performing inner product operations with wavelet functions of different scales and positions. Through wavelet transform, it is possible to extract the changing patterns of the load waveform at different frequency components, capture the fast-changing parts of the load corresponding to the high-frequency components, and the slow-changing trends of the load corresponding to the low-frequency components. At the same time, the information of the mutation points in the waveform can also be accurately determined. These mutation points often correspond to events such as the startup, shutdown, or switching of the operating status of the computing power equipment.

[0047] The extracted features are clustered according to load type, fluctuation intensity and duration, and a clustering algorithm (such as K-means clustering algorithm) is used to form a computing load classification rule base. The core idea of ​​the K-means clustering algorithm is to divide the data into clusters, so that the sum of the squares of the distances from each data point to the center of the cluster to which it belongs is minimized. In the clustering process, we first randomly initialize The center of each cluster is calculated, and the distance from each characteristic data point to each cluster center is calculated, and the data point is assigned to the cluster with the closest distance. Next, the center position of each cluster is recalculated, and the above distribution and center calculation process is repeated until the cluster center no longer changes significantly or the preset number of iterations is reached. Through clustering, computing loads with similar characteristics are classified into one category to form different classification rules, which describe in detail the characteristic range and typical patterns of each type of computing load.

[0048] In the feature generation phase, the dynamic change range of computing load is determined based on the load classification rule base. Combined with the grid operation standards and the technical indicators of the flywheel energy storage system, the power demand parameter thresholds under each load classification are set. For example, for power parameters, according to the maximum output power of the flywheel energy storage system and the safe operating power range of the power grid, setting power caps for different types of computing loads and lower limit ; For current parameters, determine the current threshold based on the rated current of the device and the carrying capacity of the line. Collect device operating status data, including device temperature ,Voltage , Current etc., as well as external environmental variables such as ambient temperature ,humidity Etc., construct a multidimensional parameter correlation matrix. Each row in the matrix represents a sampling moment, and each column corresponds to a parameter. The relationship between the parameters is integrated and represented in matrix form.

[0049] The long short-term memory network (LSTM) model is used to extract features and predict trends of time series parameters. The LSTM model is a special type of recurrent neural network (RNN) that can effectively handle the long-term dependence problem in long sequence data by introducing a gating mechanism. The LSTM model includes an input gate, a forget gate, and an output gate. The input gate controls the degree to which the current input information enters the cell state. The forget gate determines which information in the previous cell state is retained. The output gate determines the output based on the cell state and the current input. During the training process, the time series data in the multi-dimensional parameter correlation matrix is used as the input. After being processed by multiple LSTM units, the complex features and changing trends contained in the data are extracted. The model adjusts its internal parameters through the backpropagation algorithm to minimize the error between the predicted value and the actual value. According to the prediction results of the LSTM model, a load feature coding table is generated. Different load features are uniformly encoded in the coding table, which is used for the parameter configuration of real-time acquisition instructions, enabling the system to automatically adjust the data acquisition parameters according to the predicted load features, improving the pertinence and effectiveness of the acquired data.

[0050] Embodiment 4: The process of automatically matching and adapting the acquisition parameter configuration and adjusting the parameters based on the power grid dispatching requirements is as follows.

[0051] The system first analyzes the operator's historical configuration records and parameter adjustment frequencies. Suppose there is an operator A who, in the past month, for a certain type of common basic load computing power equipment, has set the acquisition frequency to 1 time per minute and the data accuracy to retain two decimal places multiple times. The system uses data mining algorithms to deeply analyze these historical operation records of operator A, extracts the commonly used parameter combination as (acquisition frequency: 1 time / minute, data accuracy: two decimal places), and at the same time summarizes that operator A, when facing basic load computing power equipment, is accustomed to giving priority to ensuring the stability of data and rarely making frequent adjustments to parameters. Based on this, according to the technical level and equipment permissions of operator A, if his technical level is intermediate, the range of configurable parameters is set: the acquisition frequency can be adjusted between 0.5 - 2 times per minute, with a step size of 0.1 time per minute; the data accuracy can be selected between retaining one decimal place and three decimal places, with a step size of one decimal place.

[0052] Combined with the current power grid load status and the remaining capacity of the flywheel energy storage system, the system dynamically generates a candidate set of parameter configurations. For example, when the power grid is in a low electricity consumption period, the remaining capacity of the flywheel energy storage system is 80%, and the computing power load for data collection currently is the non-core computing task of a certain data center (belonging to fluctuating load), the system uses the state evaluation algorithm, comprehensively considering factors such as light power grid load, sufficient energy storage capacity, and load fluctuation characteristics, to generate a candidate set of parameter configurations. The candidate set may include: Option 1 (collection frequency: 0.8 times / minute, data accuracy: one decimal place), Option 2 (collection frequency: 1 time / minute, data accuracy: two decimal places), Option 3 (collection frequency: 1.2 times / minute, data accuracy: one decimal place).

[0053] Using the analytic hierarchy process, the system constructs a hierarchical structure model to rank the priority of the candidate set. In this model, parameters such as collection frequency and data accuracy are used as the criterion layer, different parameter configuration schemes are used as the scheme layer, and the goal layer is to select the optimal parameter configuration to meet the current load demand. The system comprehensively considers the importance of each parameter in different scenarios and assigns corresponding weights to each parameter. For the above scenario of fluctuating load, light power grid load, and sufficient energy storage capacity, assume the weight of the collection frequency is 0.6 and the weight of the data accuracy is 0.4. Through the comprehensive evaluation of the parameters in each candidate scheme, it is calculated that the priority of Option 2 is the highest, so Option 2 is presented to the operator as the recommended optimal parameter configuration scheme.

[0054] Based on the Euclidean distance similarity calculation method, the system compares the load feature vectors in the historical scenario with the current scenario item by item. Assume that in the historical scenario, there is Scenario B, whose load feature vector is (load change frequency: 0.5 times / minute, load change amplitude: 5 kW, duration: 30 minutes), and the load feature vector of the current scenario is (load change frequency: 0.6 times / minute, load change amplitude: 6 kW, duration: 25 minutes). The system compares the corresponding elements of these two groups of vectors, calculates the Euclidean distance, and generates a similarity score. Set the similarity threshold to 0.8, filter out the historical scenarios with similarity higher than this threshold, and extract their parameter configuration schemes. If the similarity score of Scenario B reaches 0.85 and meets the condition, extract the parameter configuration scheme of Scenario B (collection frequency: 0.7 times / minute, data accuracy: two decimal places). If there are multiple similar scenarios, the system fuses the configuration schemes of these scenarios to generate a weighted average recommended list.

[0055] In terms of adjusting the demand parameters of power grid dispatching, the system analyzes the load regulation target and time window requirements in the power grid dispatching instructions. For example, the power grid dispatching instruction requires that within the next 2 hours, the load in a certain area be adjusted from the current 500 kW to 600 kW. The system divides multiple adjustment stages according to the output results of the load prediction model. Suppose it is divided into three stages: the first stage (0 - 30 minutes), adjust the load to 530 kW; the second stage (30 - 90 minutes), adjust the load to 570 kW; the third stage (90 - 120 minutes), adjust the load to 600 kW, and set parameter constraints for each stage. For example, the allowed power adjustment rate in the first stage does not exceed 1 kW / minute.

[0056] Using the particle swarm optimization algorithm, the system sets the position and velocity of the particles. Each particle represents a set of parameter configurations. The position of the particle corresponds to the value of the parameter, and the velocity represents the change trend of the parameter. By iteratively adjusting the parameter range, the parameters are made to meet the dynamic response requirements of power grid dispatching. During the iteration process, the particles continuously update their positions and velocities according to their own optimal positions and the optimal position of the group, gradually searching for a better parameter configuration. Finally, the adjusted parameter range is encoded into a standardized acquisition instruction and sent to the flywheel energy storage control system through the communication protocol.

[0057] Based on the density clustering algorithm, the system extracts the load regulation amplitude, response speed, and duration characteristics in the dispatching requirements. For example, for a series of dispatching requirements, some require a load regulation amplitude of 100 kW, a response speed of completing the regulation within 5 minutes, and a duration of 60 minutes; some require a load regulation amplitude of 50 kW, a response speed of completing the regulation within 3 minutes, and a duration of 30 minutes, etc. The system calculates the data point density of these characteristics in the feature space, identifies the high-density area as the core clustering cluster. Merges adjacent density-reachable clustering clusters, eliminates noise data, and divides the dispatching requirement groups according to the clustering results. For example, the dispatching requirements are divided into groups with high regulation amplitude and long duration, medium regulation amplitude and medium duration, etc., and upper and lower limits of parameter thresholds are set for each group. For the group with high regulation amplitude and long duration, the lower limit of the power regulation amplitude is set to 80 kW, and the upper limit is unlimited; the lower limit of the response speed is 5 minutes, and the upper limit is unlimited, etc.

[0058] Example 5: In terms of updating the database, judging the load pattern, and expanding the parameter configuration list, the system continuously accesses real-time power grid monitoring data and the operation logs of the flywheel energy storage system. The real-time power grid monitoring data includes information such as the voltage and frequency of each node in the power grid. These data are collected by dedicated monitoring devices at fixed time intervals and transmitted to the system through a stable communication network. The operation logs of the flywheel energy storage system detail the operating states of the internal components of the system, such as the rotational speed changes of the flywheel, the charge and discharge processes of the energy storage unit, and the operation instructions of the relevant control modules. These log data are generated and stored in the local storage device in real time during the system operation and then transmitted to the system for integration, adding the new data to the load pattern samples in the power feature database.

[0059] When extracting the features of the newly added load pattern, the same method as the existing patterns is adopted. For example, wavelet transform is used to analyze the waveform of the newly added load pattern, capture its variation rules under different frequency components, and at the same time determine the mutation points in the waveform. These mutation points may correspond to situations such as the access of new computing power devices and the switching of device operating states. Then, using the similarity measurement algorithm, the features of the newly added load pattern are compared with the existing patterns to calculate the similarity between the two. If the similarity is high, it indicates that the newly added load pattern has a certain similarity with the existing patterns and the processing method of the existing patterns can be referred to; if the similarity is low, further in-depth analysis is required.

[0060] Based on the random forest regression model, the system collects data on power grid frequency deviation, voltage fluctuation, and load mutation events as training samples. The power grid frequency deviation data reflects the difference between the actual operating frequency of the power grid and the standard frequency, the voltage fluctuation data reflects the voltage changes of each node in the power grid, and the load mutation event data records the significant changes in the load within a short period of time. By constructing multiple regression decision trees, each decision tree is constructed based on different subsets of training samples and features, and then the stability score is output through a voting mechanism. After inputting the features of the newly added load pattern into the model, the model will comprehensively consider the results of multiple decision trees to obtain its prediction score for the power grid stability and judge the risk level according to the preset score threshold.

[0061] The system constructs a gradient boosting decision tree model to judge whether the newly added pattern can be adapted by adjusting the flywheel rotational speed and energy storage capacity through continuous iterative learning. This model will gradually optimize the decision-making process according to historical data and the current load pattern. Each iteration will be adjusted based on the previous prediction error to improve the prediction accuracy. If the judgment result shows that the newly added load pattern exceeds the applicable range of the current collected parameters, the parameter expansion process will be triggered.

[0062] In the parameter expansion phase, the system extracts the key features of the new load pattern, including the fluctuation period, peak power, and duration, etc. For example, if the new load pattern shows periodic power fluctuations, then the fluctuation period is an important feature; the peak power reflects the maximum power value that the load pattern may reach during operation; the duration represents the length of time that the load pattern maintains a certain state. According to the maximum charge-discharge rate and capacity limit of the flywheel energy storage system, a feasible range for parameter expansion is determined through a series of calculations. The maximum charge-discharge rate determines the maximum energy that the system can release or absorb per unit time, and the capacity limit stipulates the maximum amount of electricity that the system can store. These factors will all impose constraints on the range of parameter expansion.

[0063] Using the genetic algorithm, the system defines the objective function for parameter optimization, including minimizing the load response time and equalizing the system losses. The load response time refers to the time required for the system to respond to load changes and make adjustments. Minimizing this time can improve the response speed and stability of the system; equalizing the system losses aims to reasonably distribute the energy losses during system operation, avoid excessive losses of some components, and extend the service life of the system. Initialize the parameter population, where each individual represents a set of possible parameter configurations. Generate the offspring population through crossover and mutation operations. The crossover operation is to exchange some genes of two individuals to generate new individuals; the mutation operation is to randomly change the genes of an individual to increase the diversity of the population. Calculate the fitness value of each individual. The higher the fitness value, the more the parameter configuration corresponding to the individual meets the requirements of the objective function. Retain the individuals with high fitness values for the next generation of iteration. After multiple rounds of iteration, finally output the optimal parameter configuration plan that meets multiple constraints, add the optimized parameters to the configuration list, and synchronously update them to the real-time acquisition system.

[0064] In addition, the system constructs a visual monitoring interface for power demand data acquisition. This interface integrates the load curve, parameter configuration status, and system alarm information. The load curve graphically shows the power load changes in different time periods, facilitating the operator to intuitively understand the load fluctuation trend; the parameter configuration status displays the current acquisition parameters adopted by the system in real time, as well as the setting basis and applicable scenarios of these parameters; the system alarm information will issue a prompt in a timely manner when an abnormal situation is detected, such as the load exceeding the preset range, the abnormal operation status of the equipment, etc. Through interactive controls, the operator can manually adjust the acquisition frequency and data accuracy to meet the personalized needs in different scenarios. The system uses a time-series database to store the acquired data. This database is specifically optimized for time-series data and can efficiently store and query data arranged in chronological order, supporting millisecond-level data query and retrospective analysis, which is convenient for the operator to conduct in-depth research and analysis on historical data.

[0065] Meanwhile, the system encrypts the collected data during transmission and storage, and uses national cryptography algorithms to protect sensitive information end-to-end. During data transmission, the data is encrypted to ensure that it cannot be stolen or tampered with during network transmission; during the storage process, the stored data is also encrypted to prevent illegal access to the data in the storage device. A multi-level access permission control mechanism is set up. According to the identity and responsibilities of users, different access permissions are assigned to restrict the operation permissions of unauthorized users. Only users with corresponding permissions can view, modify, etc. the data, ensuring the security and privacy of the data.

[0066] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0067] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for collecting computing power load power demand data based on a flywheel energy storage system, characterized in that, The method includes: Obtaining the real-time speed, energy storage capacity, and power output data of the flywheel energy storage system, classifying the power demand data, constructing a power feature database, and performing data dimensionality reduction processing; Extracting the load fluctuation pattern based on the power feature database and classifying the computing power load to be collected; generating features for the power demand parameters under each load classification according to the dynamic change characteristics of the computing power load; automatically matching and adapting the collected parameter configuration according to the usage preferences and configuration conditions of the system operator for the real-time demands of different loads; Dynamically adjusting the parameter range in the configuration list according to different grid dispatching requirements to generate a collection instruction that accurately describes the power demand; Updating the power feature database, adjusting the corresponding load prediction model according to the database update result, and judging whether the newly added load pattern exceeds the applicable range of the current collection parameters; Expanding the collection parameter configuration list according to the newly added load pattern and applying it to real-time data collection.

2. The method for collecting computing power load power demand data based on a flywheel energy storage system according to claim 1, wherein The obtaining the real-time speed, energy storage capacity, and power output data of the flywheel energy storage system, classifying the power demand data, constructing a power feature database, and performing data dimensionality reduction processing includes: Deploying a multi-type sensor network to collect the speed, temperature of the flywheel rotor, and the power change data of the energy storage unit; obtaining the real-time power output power and load demand curve through the grid dispatching platform; integrating the historical operation logs and fault records to verify the integrity and consistency of the data; dividing the power demand data into base load, peak load, and fluctuating load according to the load type, and adding a timestamp label to each type of data; establishing the table structure of the power feature database using a distributed storage system to store the speed, capacity, power, and load classification information; based on the principal component analysis method, transforming the power demand data into a low-dimensional feature vector; the features after dimensionality reduction include the load fluctuation amplitude, duration, periodic law, and the state of associated equipment; The transforming the power demand data into a low-dimensional feature vector based on the principal component analysis method specifically includes: Performing normalization processing on the original power data to eliminate the dimension difference; extracting the local features of the time series data through a sliding window mechanism to construct a multi-dimensional feature matrix; calculating the eigenvectors and eigenvalues using the covariance matrix, and screening the principal components with a cumulative contribution rate exceeding a preset threshold; inputting the screened principal components as the low-dimensional feature vector into the load prediction model.

3. The method for collecting computing power load power demand data based on a flywheel energy storage system according to claim 1, wherein The extracting the load fluctuation pattern based on the power feature database and classifying the computing power load to be collected includes: Defining the collection priority and response time constraint of the computing power load; inputting the key parameters of the load fluctuation pattern into the power feature database for matching and retrieval, and screening the historical data related to the target computing power load; for each historical data, using wavelet transform to extract the frequency domain features and mutation point information of the load waveform; clustering the extracted features according to the load type, fluctuation intensity, and duration to form a computing power load classification rule library.

4. The method for collecting computing power load power demand data based on a flywheel energy storage system according to claim 1, wherein The generating features for the power demand parameters under each load classification according to the dynamic change characteristics of the computing power load includes: Based on the load classification rule base, determine the dynamic change range of the computing power load; combine the grid operation standards and the technical indicators of the flywheel energy storage system to set the power demand parameter thresholds for each load classification; collect the device operation status data and external environmental variables to construct a multi-dimensional parameter correlation matrix; use the long short-term memory network model to extract features and predict trends of the time series parameters; generate a load feature coding table according to the prediction results for the parameter configuration of the real-time acquisition instructions.

5. The method for collecting computing power load power demand data based on a flywheel energy storage system according to claim 1, wherein Automatically match the appropriate acquisition parameter configuration according to the usage preferences and configuration conditions of the system operator for the real-time requirements of different loads, including: Analyze the operator's historical configuration records and parameter adjustment frequencies, and extract their common parameter combinations and adjustment strategies; limit the range and step size of the configurable parameters according to the operator's technical level and device permissions; combine the current grid load status and the remaining capacity of the flywheel energy storage system to dynamically generate a candidate set of parameter configurations; use the analytic hierarchy process to rank the candidate set by priority and recommend the optimal parameter configuration scheme; Recommend a list of acquisition parameter configurations in similar scenarios in the historical data based on the Euclidean distance similarity calculation method, specifically including: Compare the load feature vectors in the historical scenario with the current scenario item by item, calculate the Euclidean distance and generate a similarity score; screen the historical scenarios with similarity higher than the threshold according to the score, and extract their parameter configuration schemes; fuse the configuration schemes of multiple similar scenarios to generate a weighted average recommended list.

6. The method for collecting computing power load power demand data based on a flywheel energy storage system according to claim 1, characterized in that, Dynamically adjust the parameter range in the configuration list according to different grid dispatching requirements to generate an acquisition instruction that accurately describes the power demand, including: Parse the load regulation target and time window requirements in the grid dispatching instruction; divide multiple regulation stages and set stage parameter constraints according to the output results of the load prediction model; use the particle swarm optimization algorithm to iteratively adjust the parameter range to ensure meeting the dynamic response requirements of the grid dispatching; encode the adjusted parameter range into a standardized acquisition instruction and send it to the flywheel energy storage control system; Group the grid dispatching requirements based on the density clustering algorithm to determine the parameter thresholds for different groups, specifically including: Extract the load regulation amplitude, response speed and duration characteristics in the dispatching requirements; calculate the data point density in the feature space, identify the high-density area as the core clustering cluster; merge adjacent density-reachable clustering clusters and eliminate noise data; divide the dispatching requirement groups according to the clustering results and set the upper and lower limits of the parameter thresholds for each group.

7. The method for collecting computing power load power demand data based on a flywheel energy storage system according to claim 1, wherein Update the power feature database, adjust the corresponding load prediction model according to the database update results, and judge whether the newly added load pattern exceeds the applicable range of the current acquisition parameters, including: Access the real-time grid monitoring data and the operation logs of the flywheel energy storage system to update the load pattern samples in the power feature database; extract the features of the new load patterns and compare their similarities with the existing patterns; based on the random forest regression model, predict the impact of the new patterns on the grid frequency stability; construct a gradient boosting decision tree model to determine whether the new patterns can be adapted by adjusting the flywheel speed and energy storage capacity; if it exceeds the current parameter range, trigger the parameter expansion process. Based on the random forest regression model, predict the impact of the new load patterns on the grid stability, specifically including: Collect data on grid frequency deviation, voltage fluctuation, and load mutation events as training samples; construct multiple regression decision trees and output the stability score through a voting mechanism; input the features of the new load patterns into the model to obtain the predicted score of their impact on the grid stability, and judge the risk level according to the score threshold.

8. The method for collecting computing power load power demand data based on a flywheel energy storage system according to claim 1, wherein The expansion of the acquisition parameter configuration list according to the new load pattern and its application to real-time data acquisition includes: Extract the key features of the new load pattern, including the fluctuation period, peak power, and duration; calculate the feasible interval for parameter expansion according to the maximum charge and discharge rate and capacity limit of the flywheel energy storage system; use the genetic algorithm to perform multi-objective optimization on the parameter combination and screen the optimal solution set that meets the system constraints; add the optimized parameters to the configuration list and synchronously update them to the real-time acquisition system. Based on the genetic algorithm optimization strategy, determine the constraint conditions for parameter configuration and dynamically match them with the system operation status, specifically including: Define the objective function for parameter optimization, including minimizing the load response time and equalizing the system losses; initialize the parameter population and generate the offspring population through crossover and mutation operations; calculate the fitness value of each individual and retain the individuals with high fitness values for the next generation of iteration; finally, output the optimal parameter configuration plan that meets multiple constraints.

9. The method for collecting computing power load power demand data based on a flywheel energy storage system according to claim 1, wherein, The method further includes: Construct a visual monitoring interface for power demand data acquisition, integrating load curves, parameter configuration status, and system alarm information; allow operators to manually adjust the acquisition frequency and data accuracy through interactive controls; use a time series database to store the acquired data, supporting millisecond-level data query and retrospective analysis.

10. The method for collecting computing power load power demand data based on a flywheel energy storage system according to claim 1, wherein The method further includes: Encrypt the transmission and storage of the acquired data, and use national cryptographic algorithms to provide end-to-end protection for sensitive information; set up a multi-level access permission control mechanism to limit the operation permissions of unauthorized users.

Images