A computing power optimization method and system in power data maintenance

Abstract

The application discloses a computing power optimization method and system in power data maintenance, and relates to the technical field of data center and power grid cooperation, which comprises the following steps: collecting and standardizing the hardware, software, energy consumption and power supply data of the computing power center through a unified interface protocol, and forming a data stream after security processing; constructing a comprehensive resource model of the integrated computing power supply and power demand in the region based on the data stream; then, establishing a digital twin model of the coupling of the computing power center and the power system, and realizing real-time mapping of the physical entity and the virtual model; generating a dynamic twin body supporting multi-dimensional visual display through synchronous real-time monitoring data; simulating different computing power scheduling strategies based on the twin body, and automatically generating a decision scheme of the cooperation optimization of the computing power resource and the power supply. The application realizes joint modeling and cooperative simulation optimization of the computing power and the power system.

Description

Technical Field

[0001] This invention belongs to the field of data center and power grid collaboration technology, specifically a computing power optimization method and system for power data maintenance. Background Technology

[0002] Currently, the operation and management of computing centers and the dispatching and operation of regional power systems are typically optimized independently under different systems. The resource scheduling of computing centers mainly focuses on hardware utilization, software performance, and internal energy efficiency, with their optimization models rarely incorporating key external conditions such as grid power supply capacity, electricity price fluctuations, and network constraints in real time. Power system dispatching, on the other hand, focuses on the balance and security of power generation and transmission, usually treating computing centers as a single, static, or simply modeled load, failing to perceive the elasticity, portability, and corresponding fine-grained power consumption changes of their internal computing tasks. This separation leads to potential conflicts between the optimization objectives of the two systems. Sudden loads on computing centers may impact the local stability of the power grid, while grid dispatching instructions may affect the quality and cost of computing services.

[0003] In existing technologies, even those employing digital twin technology mostly establish one-way, isolated virtual mappings for the internal infrastructure of computing centers or the power grid itself. These twin models can achieve state monitoring and simulation within their own system, but lack the ability to couple and co-simulate with another related system. Therefore, it is impossible to assess the specific impact of a computing power scheduling strategy on the power grid load before making decisions, and it is also difficult to dynamically generate the optimal adjustment plan for computing power load based on real-time power supply conditions. Management typically can only discover discrepancies afterward through energy consumption bills or independent monitoring alarms, and take passive, non-coordinated countermeasures. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes a computing power optimization method for power data maintenance, comprising: Data from multiple sources is collected based on a unified data interface protocol to generate a standardized fused data set. The multi-source data includes hardware operating parameters of the computing center, software load data, energy consumption data, and power supply data from the power company. The standardized fused data set is subjected to data encryption and identity authentication processing to generate a securely transmitted computing power-electricity fused data stream; Based on the securely transmitted computing power and power fusion data stream, a computing power resource model within the region is built, and a comprehensive resource model including computing power supply curves and power demand curves is generated. Based on the comprehensive resource model, a digital twin model of the computing center and the power system is constructed to generate a real-time mapping relationship between the physical entity and the virtual model; The real-time monitoring data is synchronized with the digital twin model to generate a dynamically updated multi-dimensional twin, which supports the visualization of energy consumption distribution, equipment status and environmental parameters. Based on the multi-dimensional twin simulation of different computing power scheduling strategies, a collaborative optimization decision scheme for computing power resources and power supply is generated.

[0005] Preferably, the step of collecting multi-source data based on a unified data interface protocol to generate a standardized fused data set includes: The hardware operating parameters of the computing center are collected through IoT terminal devices. These hardware operating parameters include the CPU utilization, memory usage, and network bandwidth usage of the server cluster. The software load data is collected by a software probe, and the software load data includes application process resource consumption, number of virtual machine instances, and containerized service deployment status. The energy consumption data is collected by smart meters, and the energy consumption data includes the total power consumption of the computing center, rack-level power consumption and cooling equipment energy consumption. The power supply data of the power company is collected through the power dispatching system interface. The power supply data includes real-time grid load, substation output and regional power supply reliability indicators. The hardware operating parameters, software load data, energy consumption data, and power supply data are converted and aligned in structure according to the unified data interface protocol to generate the standardized fused data set.

[0006] Preferably, the step of performing data encryption and identity authentication processing on the standardized fused data set to generate a securely transmitted computing power-electricity fused data stream includes: The standardized fused data set is encrypted using an asymmetric encryption algorithm to generate an encrypted data packet; Perform two-way authentication of the data acquisition terminal and the receiving server to generate an authentication token; The authentication token is bound to the encrypted data packet to generate an encrypted transmission stream with authentication information; The encrypted transmission stream is transmitted to the central data processing platform through a dedicated communication channel. The central data processing platform performs decryption and integrity verification to generate the securely transmitted computing power-electricity fusion data stream.

[0007] Preferably, the step of building a computing power resource model within the region based on the securely transmitted computing power-power fusion data stream, and generating a comprehensive resource model including computing power supply curves and power demand curves, includes: Historical computing power consumption sequences and corresponding timestamps are extracted from the securely transmitted computing power-power fusion data stream to construct a computing power demand time series; Power supply capacity data and electricity price information are extracted from the securely transmitted computing power-power fusion data stream to construct a power supply capacity time series. A correlation function between computing power consumption and electricity consumption is established, which is obtained based on regression analysis of the computing power demand time series and electricity consumption data. The computing power demand time series is converted into an equivalent electricity demand series based on the correlation function. The equivalent power demand sequence and the power supply capacity time series are merged and standardized to generate a comprehensive resource model that includes a computing power supply curve and a power demand curve. The computing power supply curve represents the total available computing power over time, and the power demand curve represents the predicted power consumption over time.

[0008] Preferably, the construction of a digital twin model of the computing center and the power system, and the generation of a real-time mapping relationship between the physical entity and the virtual model, includes: Based on the comprehensive resource model as the basic data architecture, a computing power center digital twin is constructed, which includes a virtual server cluster model, a virtual network topology, and a virtual cooling system model. Based on the power grid geographic information system data and the power supply capacity time series, a digital twin of the power system is constructed, which includes a virtual transmission line model, a virtual substation model and a virtual distribution network model. Establish an energy flow and information flow connection channel between the computing center digital twin and the power system digital twin, and generate a real-time mapping relationship between the physical entity and the virtual model. The real-time mapping relationship ensures that the state changes of the virtual model are synchronized with the actual state of the physical entity.

[0009] Preferably, the step of synchronizing the real-time monitoring data with the digital twin model to generate a dynamically updated multi-dimensional twin includes: The system continuously collects real-time hardware operating parameters, real-time software load data, and real-time energy consumption data of the computing center to generate a real-time computing power monitoring data stream. The system continuously collects real-time grid load data, substation operation status data, and power supply quality data of the power system to generate a real-time power monitoring data stream. The real-time computing power monitoring data stream is injected into the computing power center digital twin to drive the status update of the virtual server cluster model, virtual network topology, and virtual cooling system model; The real-time power monitoring data stream is injected into the power system digital twin to drive the state updates of the virtual transmission line model, virtual substation model, and virtual distribution network model. The updated computing center digital twin and the power system digital twin are merged to generate the dynamically updated multi-dimensional twin, which supports the visualization of energy consumption distribution heatmap, equipment status topology map and environmental parameter change curves.

[0010] Preferably, the step of generating a collaborative optimization decision scheme for computing resources and power supply based on the multi-dimensional twin simulation of different computing power scheduling strategies includes: A load migration strategy simulation scenario is set in the multi-dimensional twin, and the load migration strategy simulation scenario includes the operation of migrating virtual machine instances from a high-power physical server to a low-power physical server. A task scheduling strategy simulation scenario is set in the multi-dimensional twin, which includes the operation of delaying computationally intensive tasks to the period of low electricity price. A device start-stop policy simulation scenario is set in the multi-dimensional twin, which includes the operation of dynamically shutting down or starting some servers and cooling equipment according to the predicted load. Run the load migration strategy simulation scenario, the task scheduling strategy simulation scenario, and the device start-stop strategy simulation scenario to obtain the corresponding simulated power consumption curve, simulated computing power supply curve, and simulated system stability index, respectively. Based on a preset optimization objective function, the simulated power consumption curve, the simulated computing power supply curve, and the stability index of the simulated system are evaluated. The optimal strategy combination based on the comprehensive evaluation results is selected to generate a collaborative optimization decision scheme for computing power resources and power supply. The evaluation of the simulated power consumption curve, the simulated computing power supply curve, and the stability index of the simulated system based on a preset optimization objective function includes: An optimization objective function is constructed with the dual objectives of minimizing total operating cost and maximizing the reliability of computing power services. The total operating cost includes electricity cost and equipment depreciation cost. Substitute the simulated power consumption curve into the optimization objective function to calculate the predicted power cost under the corresponding strategy; The simulated computing power supply curve is compared with the preset computing power demand baseline to calculate the computing power demand satisfaction rate index. The stability indicators of the simulated system, including the mean time between failures (MTBF) and the power supply voltage qualification rate, are quantified into a reliability score. The predicted electricity cost, the computing power demand satisfaction rate, and the reliability score are weighted and summed based on the weighting coefficients to generate a comprehensive evaluation score for each strategy combination. Compare the overall evaluation scores of all simulated strategy combinations and select the strategy combination with the highest score as the evaluation result.

[0011] Preferably, the step of encrypting the standardized fused data set using an asymmetric encryption algorithm to generate an encrypted data packet includes: The central data processing platform generates a pair of public and private keys for an asymmetric encryption algorithm and distributes the public key to each data acquisition terminal. Each of the data acquisition terminals uses the received public key to perform encryption operations on the standardized fused data set to generate ciphertext data segments; Each of the data acquisition terminals adds a timestamp and a data source identifier to each encrypted data segment, generating an encrypted data unit with time sequence and source information; All encrypted data units belonging to the same transmission batch are encapsulated and sorted to generate a complete encrypted data packet.

[0012] Preferably, establishing the energy flow and information flow connection channel between the computing center digital twin and the power system digital twin, and generating a real-time mapping relationship between the physical entity and the virtual model, includes: A virtual energy output interface is defined in the digital twin of the computing center, and the virtual energy output interface corresponds one-to-one with the actual power consumption metering point of the computing center; A virtual energy input interface is defined in the power system digital twin, and the virtual energy input interface corresponds one-to-one with the substation outgoing port that supplies power to the computing center; A virtual energy flow connection channel is established from the energy output virtual interface to the energy input virtual interface for bidirectional transmission of real-time data on power, electricity, and energy efficiency. A virtual information flow connection channel is established between the computing center digital twin and the power system digital twin for bidirectional transmission of load forecast information, scheduling instructions and fault alarm information; Through the virtual energy flow connection channel and the virtual information flow connection channel, the state changes of the computing center digital twin trigger the real-time update of the associated parameters of the power system digital twin, generating a real-time mapping relationship between the physical entity and the virtual model.

[0013] Preferably, the present invention also includes a computing power optimization system for power data maintenance, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the computing power optimization method for power data maintenance as described above.

[0014] Compared with the prior art, the beneficial effects of the present invention are: By constructing a comprehensive resource model that integrates computing power supply curves and electricity demand curves, two originally independent systems are mapped into a unified digital twin. This technology establishes a quantifiable, real-time coupled mapping relationship between computing power load dynamics and power system state. This enables the system to accurately reflect how adjustments to a computing task will affect the real-time load of the distribution network, and how grid fluctuations or dispatch instructions will constrain the availability and operating costs of computing resources. This breaks down information barriers between systems, providing a precise, consistent, and realistic state assessment basis that reflects actual interaction constraints for any subsequent optimization.

[0015] The simulation and derivation of computing power scheduling strategies are based on the aforementioned dynamically coupled twin. When a computing power allocation or migration scheme is input, the model can automatically and continuously deduce the power consumption change time series of each computing device under the scheme, and apply this change as a load input to the virtual power grid model in real time. The system synchronously calculates the power flow distribution, node voltage, and line capacity margin of the power grid, and feeds back the stability and economic evaluation results. This closed-loop simulation process can directly output a collaborative optimization scheme that integrates computing efficiency, quality of service, and power grid safety and economic operation indicators, realizing a paradigm shift from local decision-making in a single system to global optimal decision-making across systems. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of the computing power optimization method in power data maintenance as described in this invention. Figure 2 A flowchart for multi-source data acquisition; Figure 3 Flowchart for building a comprehensive resource model; Figure 4 Evaluation graph for collaborative optimization of computing power scheduling strategies; Figure 5 A radar chart for multi-dimensional evaluation of computing power scheduling strategies. Detailed Implementation

[0017] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1 A computing power optimization method for power data maintenance is proposed. This method collects multi-source data based on a unified data interface protocol, generating a standardized fused data set. The multi-source data includes hardware operating parameters of the computing power center, software load data, energy consumption data, and power supply data from the power company. The standardized fused data set undergoes data encryption and authentication to generate a secure computing power-power fused data stream. Based on this secure data stream, a regional computing power resource model is built, generating a comprehensive resource model including computing power supply curves and power demand curves. Based on this comprehensive resource model, a digital twin model of the computing power center and the power system is constructed, generating a real-time mapping relationship between the physical entity and the virtual model. Real-time monitoring data and the digital twin model are synchronized to generate a dynamically updated multi-dimensional twin, which supports visualization of energy consumption distribution, equipment status, and environmental parameters. Based on the multi-dimensional twin, different computing power scheduling strategies are simulated to generate a collaborative optimization decision-making scheme for computing power resources and power supply.

[0019] In one embodiment of the present invention, see [reference] Figure 2 This document describes the specific process of collecting multi-source data based on a unified data interface protocol to generate a standardized fused data set. The process includes: collecting hardware operating parameters of the computing center through IoT terminal devices, including CPU utilization, memory usage, and network bandwidth consumption of the server cluster; collecting software load data through software probes, including application process resource consumption, number of virtual machine instances, and containerized service deployment status; collecting energy consumption data through smart meters, including total power consumption of the computing center, rack-level power consumption, and cooling equipment energy consumption; and collecting power supply data from the power dispatching system interface, including real-time grid load, substation output, and regional power supply reliability indicators. Finally, the hardware operating parameters, software load data, energy consumption data, and power supply data are formatted and structurally aligned according to the unified data interface protocol to generate a standardized fused data set.

[0020] In practice, multi-source data is collected based on a unified data interface protocol to generate a standardized fused data set. This process involves collecting hardware operating parameters of the computing center through IoT terminal devices, including CPU utilization, memory usage, and network bandwidth usage of the server cluster; collecting software load data through software probes, including application process resource consumption, number of virtual machine instances, and containerized service deployment status; collecting energy consumption data through smart meters, including total power consumption of the computing center, rack-level power consumption, and cooling equipment energy consumption; and collecting power supply data from the power company through the power dispatching system interface, including real-time grid load, substation output, and regional power supply reliability indicators. Subsequently, the hardware operating parameters, software load data, energy consumption data, and power supply data are converted and aligned according to the unified data interface protocol to generate a standardized fused data set.

[0021] In some embodiments, IoT terminal devices are deployed in server racks of a computing center to collect real-time data on CPU utilization, memory usage, and network bandwidth usage. CPU utilization is expressed as a percentage, such as 85.5%, memory usage is expressed in gigabytes, such as 72.3%, and network bandwidth usage is expressed in megabits per second, such as 120. A software probe is embedded in the operating system of the computing center to monitor application process resource consumption, the number of virtual machine instances, and the deployment status of containerized services. Application process resource consumption records CPU time and memory usage values, the number of virtual machine instances is expressed as an integer, such as 50, and the deployment status of containerized services is identified by a string, such as "running". The smart meter is installed at the power inlet of the computing center to measure the total power consumption of the computing center, rack-level power consumption, and cooling equipment energy consumption. The total power consumption of the computing center is expressed in kilowatts, such as 950.5. The rack-level power consumption is recorded in array form, such as [45.2, 48.1]. The cooling equipment energy consumption is expressed in kilowatts, such as 200.3. The power dispatch system interface connects to the power company network through the application programming interface to obtain the real-time load of the power grid, the output of substations, and the regional power supply reliability index. The real-time load of the power grid is expressed in megawatts, such as 500.2. The output of substations is expressed in megavolt-amperes, such as 300.1. The regional power supply reliability index is expressed as a percentage, such as 99.8.

[0022] It is understandable that hardware operating parameters, software load data, energy consumption data, and power supply data originate from different systems and have varying original formats. For example, hardware operating parameters are transmitted in key-value pairs, software load data is stored in Extended Markup Language (ESL) format, energy consumption data is recorded in comma-separated value format, and power supply data is provided in relational database tables. In specific implementation, a unified data interface protocol is defined as a data schema based on JavaScript object representation. All collected data must be converted into JavaScript object representation documents that conform to this schema. The format conversion process is performed by a data preprocessing module, which parses the original data fields and maps them to... The protocol defines fields such as the original field "CPU_Usage" for CPU utilization, which is mapped to "cpu_utilization", and the original field "TotalPower_kW" for total power consumption of the computing center, which is mapped to "total_power". The structure alignment ensures timestamp synchronization. The timestamp field is uniformly named "timestamp" and the value is in the international standard time format such as "2023-10-01T00:00:00Z". Data comparison shows that the data structure of hardware operating parameters before conversion is a loose key-value pair, and after conversion, it becomes a structured JavaScript object representation document with consistent field names and types.

[0023] Optionally, numerical normalization is performed when generating the standardized fused dataset to eliminate the influence of different units. In practice, the original numerical values ​​are scaled using the following formula:

[0024] in: This represents the normalized value. Represents the original value. This indicates the lower limit of the field in historical data. This indicates the upper limit of the field in historical data. For example, the original value of CPU utilization is 85.5, the historical lower limit is 0, and the historical upper limit is 100. After normalization, it becomes 0.855. The original value of real-time grid load is 500.2, the historical lower limit is 200, and the historical upper limit is 800. After normalization, it becomes 0.50025. The normalization operation is applied to all numerical fields in hardware operating parameters, software load data, energy consumption data, and power supply data, but it is only triggered when needed for subsequent analysis.

[0025] In one embodiment of the present invention, the central data processing platform generates a pair of public and private keys for an asymmetric encryption algorithm and distributes the public key to each data acquisition terminal. Each data acquisition terminal uses the received public key to perform encryption operations on a standardized fused data set, generating ciphertext data segments. Each data acquisition terminal adds a timestamp and data source identifier to each ciphertext data segment, generating an encrypted data unit with time sequence and source information. All encrypted data units belonging to the same transmission batch are encapsulated and sorted to generate a complete encrypted data packet. Two-way authentication is performed on the identities of the data acquisition terminals and the receiving server to generate an authentication token. The authentication token is bound to the encrypted data packet to generate an encrypted transmission stream with authentication information. The encrypted transmission stream is transmitted to the central data processing platform through a dedicated communication channel, where it is decrypted and its integrity is verified, generating a securely transmitted computing power-electricity fusion data stream.

[0026] In practice, standardized fused data sets undergo data encryption and identity authentication to generate securely transmitted computing power-electricity fused data streams. This process begins with the central data processing platform generating a pair of public and private keys for an asymmetric encryption algorithm. The central data processing platform distributes the public key to each data acquisition terminal. The data acquisition terminal uses the received public key to perform encryption operations on the standardized fused data sets to generate ciphertext data segments. The data acquisition terminal adds a timestamp and data source identifier to each ciphertext data segment to generate encrypted data units with time sequence and source information. All encrypted data units belonging to the same transmission batch are encapsulated and sorted to generate complete encrypted data packets. At the same time, the identities of the data acquisition terminals and receiving servers are mutually authenticated to generate identity authentication tokens. The identity authentication tokens are bound to the encrypted data packets to generate encrypted transmission streams with identity verification information. The encrypted transmission streams are transmitted to the central data processing platform through a dedicated communication channel, where they are decrypted and their integrity is verified, ultimately generating securely transmitted computing power-electricity fused data streams.

[0027] In some embodiments, the central data processing platform uses the RSA algorithm to generate a pair of 2048-bit public and private keys. The public key is encapsulated in X.509 certificate format and distributed through a secure channel to various data acquisition terminals, such as server monitoring probes deployed in computing centers, smart meters deployed in power cabinets, and software acquisition probes deployed at the network edge. The standardized fused data set exists in the form of a JavaScript object representation string. The data acquisition terminal uses the received public key to encrypt this JavaScript object representation string using the RSA encryption algorithm in combination with the optimal asymmetric encryption padding mode. The output result is a binary ciphertext data segment with a length of 256 bytes. The data acquisition terminal then appends a Unix timestamp and a unique data source identifier to this ciphertext data segment to form a structured encrypted data unit. The internal format of the encrypted data unit includes a ciphertext length field, a ciphertext content field, a timestamp field, and a data source identifier field.

[0028] It is understandable that there is a significant difference in data format and content before and after encryption. The standardized fused data set before encryption is structured plaintext data, the content of which can be directly read and parsed. The encrypted data segment after encryption is an unreadable binary data stream, and no business information can be directly obtained. Data comparison shows that the original JavaScript object representation document size is about 150 bytes. After public key encryption and the addition of metadata, the size of the encrypted data unit increases to about 300 bytes. Integrity verification is performed on the central data processing platform. After the central data processing platform uses the private key to decrypt and obtain the original JavaScript object representation string, it calculates its secure hash algorithm -256 hash value, and compares this hash value with the hash value calculated by the data acquisition terminal and implicit in the identity authentication token before the encrypted data unit is transmitted. If the hash values ​​match, the integrity verification is passed; otherwise, it is marked as transmission corruption.

[0029] Optionally, the generation of the authentication token is based on a challenge-response mechanism and a digital certificate. In practice, the central data processing platform sends a random challenge code to the data acquisition terminal that initiated the connection. The data acquisition terminal signs the challenge code using its own client digital certificate private key and sends the signature along with the client digital certificate to the central data processing platform. The central data processing platform verifies the validity of the client digital certificate and verifies the signature using the public key in the certificate. After successful verification, the central data processing platform generates a session authentication token. The generation of the authentication token conforms to the following formula:

[0030] in: This represents the digest value of the generated authentication token. This indicates the Secure Hash Algorithm -256 hash function. This indicates the symmetric session key negotiated in this session. Indicates the current timestamp. A unique identifier representing the data acquisition terminal, an authentication token, and a session key. The validity period information is sent to the data acquisition terminal along with the encrypted data packet. The encrypted transmission stream consists of a plaintext header and an encrypted body. The plaintext header contains the digest value of the authentication token. The encryption algorithm identifier and the encrypted body contain the encryption key used in the session. The actual encrypted data packet.

[0031] In some embodiments, the dedicated communication channel employs a wired fiber optic network based on Transmission Control Protocol (TCP) and is configured with an Internet Protocol Security (IPS) tunnel. The encrypted transport stream is encapsulated within IPS data packets for transmission. Upon receiving the data packets, the central data processing platform first strips the IPS encapsulation, then parses the plaintext header to extract the authentication token digest value for identity verification. After confirming the session's validity, it uses the corresponding session key stored locally. The encrypted body is decrypted to obtain a complete encrypted data packet. The central data processing platform then unpacks and reassembles each encrypted data unit according to the sequence encapsulated within the encrypted data packet. It then uses the RSA private key held by the central data processing platform to decrypt the ciphertext data segment in each encrypted data unit one by one, restoring the original standardized JavaScript object representation string of the fused data set. The data that has completed decryption and integrity verification is marked as verified and output as a securely transmitted computing power fusion data stream to the next processing module.

[0032] In one embodiment of the present invention, see [reference] Figure 3This process involves extracting historical computing power consumption sequences and corresponding timestamps from securely transmitted computing power and power fusion data streams to construct a computing power demand time series. It also extracts power supply capacity data and electricity price information from the same data streams to construct a power supply capacity time series. A correlation function between computing power consumption and power consumption is established, derived from regression analysis of the computing power demand time series and power consumption data. The computing power demand time series is then converted into an equivalent power demand sequence based on the correlation function. The equivalent power demand sequence and the power supply capacity time series are merged and standardized to generate a comprehensive resource model containing both computing power supply curves and power demand curves. The computing power supply curve represents the total available computing power over time, while the power demand curve represents the predicted power consumption over time. Based on the comprehensive resource model as the basic data architecture, a digital twin of the computing power center is constructed, comprising a virtual server cluster model, a virtual network topology, and a virtual cooling system model. Finally, based on power grid geographic information system data and the power supply capacity time series, a digital twin of the power system is constructed, comprising a virtual transmission line model, a virtual substation model, and a virtual distribution network model. A virtual energy output interface is defined in the computing center's digital twin, with each virtual interface corresponding one-to-one with an actual power consumption metering point within the computing center. A virtual energy input interface is defined in the power system's digital twin, with each virtual interface corresponding one-to-one with an outgoing line port of the substation supplying power to the computing center. A virtual energy flow connection channel is established from the energy output virtual interface to the energy input virtual interface for bidirectional transmission of real-time data on power, energy consumption, and energy efficiency. A virtual information flow connection channel is established between the computing center's digital twin and the power system's digital twin for bidirectional transmission of load forecast information, dispatch instructions, and fault alarm information. Through the virtual energy flow connection channel and the virtual information flow connection channel, changes in the state of the computing center's digital twin trigger real-time updates of associated parameters in the power system's digital twin, generating a real-time mapping relationship between the physical entity and the virtual model.

[0033] In practical implementation, a computing power resource model within the region is built based on the securely transmitted computing power and power fusion data stream, and a comprehensive resource model including computing power supply curves and power demand curves is generated. Subsequently, a digital twin model is constructed to generate real-time mapping relationships. This process first extracts historical computing power consumption sequences and corresponding timestamps from the securely transmitted computing power and power fusion data stream to construct a computing power demand time series. The data points of the computing power demand time series include timestamp fields and computing power consumption value fields. The timestamp field follows the international standard time format. The computing power consumption value is obtained by aggregating fields such as CPU utilization, memory occupancy, and network bandwidth usage from the securely transmitted computing power and power fusion data stream and calculating a comprehensive computing power index through weighting. Among them, 152.7 is a dimensionless comprehensive computing power index. Power supply capacity data and electricity price information are extracted from the securely transmitted computing power and power fusion data stream to construct a power supply capacity time series. The data points of the power supply capacity time series include timestamp fields, the maximum power supply capacity of the power grid, and real-time electricity price fields.

[0034] In some embodiments, a correlation function is established between computing power consumption and power consumption. This correlation function is obtained based on regression analysis of computing power demand time series and power consumption data. Power consumption data is obtained from the "total_power" field in the securely transmitted computing power-power fusion data stream, representing the total power consumption of the computing center. In specific implementations, historical periods, such as the past thirty days, are collected as computing power demand time series data points, and power consumption data points at the same time are used to form sample pairs. The least squares method is used for linear regression fitting. The expression for the correlation function is:

[0035] in: This represents the predicted electricity consumption value, in kilowatts. This represents the comprehensive computing power index obtained from the computing power demand time series. The regression coefficient represents the electricity consumed per unit of comprehensive computing power index. The threshold represents the regression intercept, which represents the static power consumption of the computing center's infrastructure.

[0036] It is understandable that merging and standardizing the equivalent electricity demand sequence and the electricity supply capacity time series to generate a comprehensive resource model is a common approach. The merging operation aligns the data points of the two sequences based on a common timestamp field, forming a multi-dimensional data point that includes timestamps, predicted electricity consumption values, the maximum available power supply of the grid, and real-time electricity prices. The standardization process uses a min-max normalization method to scale the predicted electricity consumption value and the maximum available power supply of the grid to between 0 and 1, generating a comprehensive resource model that includes a computing power supply curve and an electricity demand curve. The computing power supply curve is plotted on the horizontal axis with time as the horizontal axis and the standardized maximum available power supply sequence of the grid as the vertical axis, while the electricity demand curve is plotted on the horizontal axis with time as the horizontal axis and the standardized predicted electricity consumption value sequence as the vertical axis. Data comparison shows that the original values ​​before standardization are difficult to compare intuitively on the same coordinate system due to differences in dimensions and orders of magnitude. The two curves after standardization are within the same numerical range, making it easier to observe the degree and trend of the matching relationship between electricity supply and demand.

[0037] In one embodiment of the present invention, real-time hardware operating parameters, real-time software load data, and real-time energy consumption data of the computing center are continuously collected to generate a real-time computing power monitoring data stream. Real-time grid load data, substation operating status data, and power supply quality data of the power system are continuously collected to generate a real-time power monitoring data stream. The real-time computing power monitoring data stream is injected into the computing center's digital twin to drive state updates of the virtual server cluster model, virtual network topology, and virtual cooling system model. The real-time power monitoring data stream is injected into the power system's digital twin to drive state updates of the virtual transmission line model, virtual substation model, and virtual distribution network model. The updated computing center digital twin and the power system digital twin are merged to generate a dynamically updated multi-dimensional twin, which supports the visualization of energy consumption distribution heatmaps, equipment status topology diagrams, and environmental parameter change curves.

[0038] In practice, real-time monitoring data and digital twin models are synchronized to generate a dynamically updated multi-dimensional twin. This process involves continuously collecting real-time hardware operating parameters, real-time software load data, and real-time energy consumption data from the computing center to generate a real-time computing power monitoring data stream. Simultaneously, real-time grid load data, substation operating status data, and power supply quality data from the power system are continuously collected to generate a real-time power monitoring data stream. Subsequently, the real-time computing power monitoring data stream is injected into the computing center's digital twin to drive the state updates of the virtual server cluster model, virtual network topology, and virtual cooling system model. The real-time power monitoring data stream is injected into the power system's digital twin to drive the state updates of the virtual transmission line model, virtual substation model, and virtual distribution network model. Finally, the updated computing center digital twin and power system digital twin are merged to generate a dynamically updated multi-dimensional twin. The multi-dimensional twin supports the visualization of energy consumption distribution heatmaps, equipment status topology diagrams, and environmental parameter change curves.

[0039] In some embodiments, the continuous collection of real-time hardware operating parameters, real-time software load data, and real-time energy consumption data of the computing center is accomplished through sensors and software agents deployed on physical devices. The collection frequency is set to once every ten seconds. Real-time hardware operating parameters include the CPU utilization, memory usage, and network bandwidth usage of each server in the server cluster. For example, the real-time data collected from server "SVR-01" shows a CPU utilization of 72%, a memory usage of 64%, and a network bandwidth usage of 125 megabits per second. Real-time software load data includes data collected from the container orchestration platform. The number of virtual machine instances and the deployment status of containerized services are obtained. For example, the number of virtual machine instances is 85, and the status of the containerized service "App-DB" is "running". Real-time energy consumption data includes the total power consumption of the computing center read from smart meters, the power consumption of each rack level, and the energy consumption of the precision air conditioning cooling equipment. For example, the total power consumption is 1025.3 kilowatts, the power consumption of rack 1 is 52.1 kilowatts, and the energy consumption of the cooling equipment is 215.6 kilowatts. This data is encapsulated into JavaScript object representation messages with a unified timestamp, forming a continuous real-time computing power monitoring data stream.

[0040] It is understandable that the continuous collection of real-time grid load data, substation operation status data, and power supply quality data of the power system is accomplished through the data acquisition and monitoring system interface of the power dispatching system. The acquisition frequency is also once every ten seconds. Real-time grid load data refers to the total active power value on the bus obtained from the regional grid dispatching center, for example, the current value is 523.7 MW. Substation operation status data refers to the load rate of the main transformer, the status of each outgoing switch, and the bus voltage value obtained from the "110kV Yunggu Substation" that supplies power to the computing center. For example, the load rate of the No. 1 main transformer is 78%, the status of the outgoing switch "YGL-101" is "closed", and the bus voltage is 10.52 kV. Power supply quality data refers to the voltage deviation, frequency deviation, and harmonic distortion rate indicators obtained from the power quality monitoring device. For example, the voltage deviation is +0.5% and the frequency deviation is -0.02Hz. These data are also encapsulated into JavaScript object representation messages with a unified timestamp, forming a continuous real-time power monitoring data stream.

[0041] In some embodiments, injecting real-time power monitoring data streams into the power system digital twin follows a similar injection and mapping mechanism. The data injection engine parses real-time data messages from the power system and updates the attributes of the power system digital twin according to another set of mapping rules. For example, the real-time grid load data of 523.7 MW is mapped and updated to the "load_rate" attribute of the linear element representing the regional trunk line "TL-500" in the virtual transmission line model. This attribute value is calculated to be 84.5% based on the line's rated capacity. The load rate of the No. 1 main transformer in the substation operation status data of 78% is mapped and updated to the "transformer_load" attribute of the polygonal element representing the "110kV Yung Valley Substation" in the virtual substation model. The outgoing switch status "closed" is mapped and updated to the "switch_status" attribute of the corresponding node "Node_YGL-101" in the virtual distribution network model. Driving the status updates of the virtual transmission line model, virtual substation model, and virtual distribution network model means that the graphical representation and numerical attributes of these virtual objects will be dynamically refreshed with the injection of real-time data.

[0042] Optionally, the state synchronization process includes a data consistency verification step to ensure the consistency between the virtual model and the physical entity state. In specific implementations, a hash verification function is designed for periodic verification. The hash verification function is defined as follows:

[0043] in: Represents a consistency check code. This represents a message digest algorithm function based on MD5. This represents a string generated at a certain moment after serializing all key virtual model attributes from the digital twin at the computing center. This represents the string generated at the same moment after serializing all key virtual model attributes from the power system digital twin. The data injection engine calculates the current multi-dimensional twin every five minutes. The value is compared with the baseline hash value calculated directly from the physical monitoring system. If they are inconsistent, an alarm is triggered and the data resynchronization process is started. This ensures that the dynamically updated multidimensional twin is a highly reliable mirror of the physical world state.

[0044] In some embodiments, the process of generating a dynamically updated multi-dimensional twin involves correlation analysis of real-time data streams and historical data to enrich the visualization dimensions. When updating virtual model attributes, the data injection engine not only assigns the current value to the attribute but also appends that value as a new data point to the time-series database associated with that attribute. This allows the multi-dimensional twin to support visualizations that not only present the current energy consumption distribution heatmap and equipment status topology map but also draw environmental parameter change curves for any time range using historical data, such as the total power consumption change curve of a server cluster or the fluctuation curve of the bus voltage of a substation in the past hour. The fused, dynamically updated multi-dimensional twin serves as a unified data service interface, providing query and subscription services for upper-layer applications. See Table 1.

[0045] Table 1: Device Status Mapping Table

[0046] See Figure 4 This is a collaborative optimization evaluation chart for computing power scheduling strategies. The bar chart compares the performance of four computing power scheduling strategies across three dimensions: power cost, reliability score, and system stability. It clarifies that the "combined strategy" is the optimal solution, avoiding decision-making risks arising from subjective judgment and directly guiding strategy selection in production environments. It reveals the advantages and disadvantages of individual strategies: for example, "task scheduling" is highly effective in cost reduction but has slightly lower stability; "load migration" ensures high reliability but has the highest cost. It provides data anchors for subsequent iterative optimization of the computing power scheduling system, allowing for further algorithm tuning and balancing of multiple objectives. The clear comparisons and quantifiable indicators are ideal for use in project presentations and technical reviews, enabling managers without technical backgrounds to quickly understand the value of the technical solution.

[0047] In one embodiment of the present invention, a load migration strategy simulation scenario is set up in a multi-dimensional twin, which includes the operation of migrating virtual machine instances from high-power physical servers to low-power physical servers. A task scheduling strategy simulation scenario is also set up in the multi-dimensional twin, which includes the operation of delaying computationally intensive tasks to periods of low electricity prices. A device start-up / stop strategy simulation scenario is also set up in the multi-dimensional twin, which includes the operation of dynamically shutting down or starting some servers and cooling equipment based on predicted load. Running the load migration strategy simulation scenario, task scheduling strategy simulation scenario, and device start-up / stop strategy simulation scenario yields corresponding simulated power consumption curves, simulated computing power supply curves, and simulated system stability indices. An optimization objective function is constructed with the dual objectives of minimizing total operating cost and maximizing computing power service reliability, where total operating cost includes electricity cost and equipment loss cost. Substituting the simulated power consumption curves into the optimization objective function, the predicted power cost under the corresponding strategy is calculated. The simulated computing power supply curve is compared with a preset computing power demand baseline to calculate the computing power demand fulfillment rate index. The stability indicators of the simulated system, including the mean time between failures (MTBF) and the power supply voltage qualification rate, are quantified into a reliability score. Based on weighted coefficients, the predicted power cost, computing power demand fulfillment rate, and reliability score are weighted and summed to generate a comprehensive evaluation score for each strategy combination. The comprehensive evaluation scores of all simulated strategy combinations are compared, and the strategy combination with the highest score is selected as the evaluation result, generating a collaborative optimization decision scheme for computing power resources and power supply.

[0048] In practical implementation, different computing power scheduling strategies are simulated based on a multi-dimensional twin to generate a collaborative optimization decision scheme for computing resources and power supply. This process first sets up a load migration strategy simulation scenario in the multi-dimensional twin. The load migration strategy simulation scenario includes the operation of migrating virtual machine instances from high-power physical servers to low-power physical servers. For example, in the virtual server cluster model contained in the dynamically updated multi-dimensional twin, the virtual server identified as "VSVR-H01" has a real-time power consumption attribute of 850 watts and currently hosts 12 virtual machine instances. The virtual server identified as "VSVR-L05" has a real-time power consumption attribute of 520 watts and currently hosts 8 virtual machine instances. The load migration strategy simulation scenario is defined as migrating 4 virtual machine instances on the "VSVR-H01" virtual server to the "VSVR-L05" virtual server. The task is set up in the multi-dimensional twin. The scheduling strategy simulation scenario includes operations such as delaying computationally intensive tasks to low-price periods. For example, based on the real-time electricity price time series in the power system digital twin, the period when the electricity price is below 0.5 yuan per kilowatt-hour is defined as the low-price period. The task scheduling strategy simulation scenario is defined as delaying a batch of computationally intensive tasks expected to start at 14:00 to start at 22:00 during the low-price period. The device start-stop strategy simulation scenario in the multi-dimensional twin includes operations such as dynamically shutting down or starting some servers and cooling equipment based on predicted load. For example, based on the historical load prediction model of the computing center digital twin, it is predicted that the computing power demand will decrease by 30% in the next two hours. The device start-stop strategy simulation scenario is defined as shutting down 10 virtual servers identified as "VSVR-Pool02" in the virtual server cluster model and some virtual air conditioning units in the associated virtual cooling system model.

[0049] In some embodiments, running load migration strategy simulation scenarios, task scheduling strategy simulation scenarios, and device start-up / stop strategy simulation scenarios yields corresponding simulated power consumption curves, simulated computing power supply curves, and simulated system stability indicators, respectively. When running the load migration strategy simulation scenario, the simulation engine in the multi-dimensional twin modifies the virtual machine instance quantity attribute of the relevant virtual servers in the virtual server cluster model according to the definition of the migration operation, and recalculates the real-time power consumption attributes of these virtual servers based on a preset virtual machine instance power consumption model. The simulation engine then drives the virtual energy flow connection channel to transmit the updated total power consumption value to the power system digital twin, thereby generating a model within the entire simulation time span. The simulated power consumption curve is represented in time series form. Each data point includes a timestamp and the simulated total power consumption value. For example, at the time point "2023-10-01T14:00:00Z", the simulated total power consumption after implementing the load migration strategy is 980.5 kW, while the baseline total power consumption without implementing the strategy is 1025.3 kW. At the same time, the simulation engine calculates and outputs a simulated computing power supply curve based on the changes in the CPU utilization of each virtual server after migration. The simulated computing power supply curve is represented as a time series of the comprehensive computing power index. The system stability index is quantified by the change in the standard deviation of CPU utilization of virtual server nodes caused by migration operations during the simulation period.

[0050] It's understandable that when running task scheduling strategies to simulate scenarios, the simulation engine redistributes the computational load of computationally intensive tasks from the original planned time period to off-peak electricity periods. This alters the shape of the simulated computing power supply curve at different times of the day. For example, the original plan had a peak in the computing power supply curve between 14:00 and 15:00; after the delay, this peak is smoothed out, and the peak computing power supply drops from 1800 units of comprehensive computing power index to 1500 units of comprehensive computing power index, while the computing power supply between 22:00 and 23:00 increases accordingly. The simulated electricity consumption curve changes accordingly, reflecting the increase in electricity consumption during off-peak electricity periods and the decrease in electricity consumption during the original period. During peak hours, when power consumption decreases and the simulation engine generates signals based on the predicted load model, it triggers events to shut down some virtual servers and virtual air conditioning units in the simulation timeline. The virtual server shutdown event sets its associated real-time power consumption attribute to zero, and the virtual air conditioning unit shutdown event reduces the total cooling equipment energy consumption of the virtual cooling system model. As a result, the simulated power consumption curve shows a step-like decrease, and the simulated computing power supply curve also decreases accordingly due to the reduction in available servers. The system stability index simulates the change in mean time between failures during the equipment startup and shutdown process and the risk of local overheating that may be caused by concentrated load.

[0051] Optionally, based on a preset optimization objective function, the simulated power consumption curve, simulated computing power supply curve, and simulated system stability index are evaluated to select the optimal strategy combination. An optimization objective function is constructed with the dual objectives of minimizing total operating cost and maximizing computing power service reliability. Total operating cost includes power cost and equipment depreciation cost. Power cost is calculated based on the simulated power consumption curve and real-time electricity price data for the corresponding time period. Equipment depreciation cost is calculated using a linear model based on server uptime and start / stop counts. Computing power service reliability is jointly measured by the computing power demand fulfillment rate and system stability index. The specific form of the optimization objective function is expressed as a weighted comprehensive scoring function for multiple evaluation dimensions.

[0052] in: This represents the overall evaluation score; a lower score indicates a better strategy combination. This represents the predicted electricity cost calculated based on simulated electricity consumption curves and electricity prices. This represents an indicator of the computing power demand fulfillment rate.

[0053] Indicates reliability score, This represents the weighting coefficient of electricity costs in the overall assessment, used to measure the predicted electricity costs. Comprehensive evaluation score The extent of the impact The inverse weighting coefficient represents the computing power demand fulfillment rate, used to measure... Items for comprehensive evaluation score The extent of the impact The inverse weighting coefficient represents the reliability score, used to measure... Items for comprehensive evaluation score The extent of the impact.

[0054] In some embodiments, the stability indicators of the simulated system are quantified into a reliability score. The mean time between failures (MTBF) of the equipment is estimated based on historical server failure rate data and the cumulative uptime of the servers in the simulation. The power supply voltage qualification rate is statistically derived from the virtual bus voltage data recorded by the power system digital twin during the simulation. For example, in a load migration strategy simulation scenario, due to the potential for increased target server temperature caused by the concentration of virtual machines, the simulated MTBF slightly decreases from the baseline of 10,000 hours to 9,800 hours, quantified as 0.98 points. The power supply voltage qualification rate remains at 100%, quantified as 1.0 point. The reliability score is obtained by weighted averaging of the two scores. The value is 0.99, based on the weighting coefficient for predicting electricity costs. Computing power demand satisfaction rate index and reliability rating A weighted summation is performed to generate a comprehensive evaluation score for each strategy combination. For example, setting For load migration strategies, Then the comprehensive evaluation score Compare the overall evaluation scores of all simulated strategy combinations. The lowest-scoring strategy combination is selected as the evaluation result, thereby generating a collaborative optimization decision scheme for computing resources and power supply. The content of the decision scheme is clearly defined as the specific strategy combination to be adopted and its simulated execution schedule.

[0055] See Figure 5 This is a radar chart evaluating computing power scheduling strategies from multiple dimensions. It compares the overall performance of four computing power scheduling strategies across five dimensions: computing power demand fulfillment rate, power cost, total operating cost, equipment depreciation cost, and system stability. It clarifies that the combined strategy (purple) is the optimal overall solution, providing data-driven selection criteria for technical and business decision-makers and reducing the risk of subjective judgment. It reveals the "shortcomings" of a single strategy and the "balanced advantages" of the combined strategy, avoiding the one-sidedness of decisions based on single indicators. It provides clear directions for subsequent algorithm optimization, such as how to improve the computing power fulfillment rate of "task-only scheduling" while maintaining its cost advantage. It also provides a basis for scenario-based selection, such as "load migration only" being suitable for high-reliability scenarios and "task scheduling only" being suitable for cost-sensitive scenarios, directly guiding strategy configuration in production environments.

[0056] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for optimizing computing power in power data maintenance, characterized in that, include: Data from multiple sources is collected based on a unified data interface protocol to generate a standardized fused data set. The multi-source data includes hardware operating parameters of the computing center, software load data, energy consumption data, and power supply data from the power company. The standardized fused data set is subjected to data encryption and identity authentication processing to generate a securely transmitted computing power-electricity fused data stream; Based on the securely transmitted computing power and power fusion data stream, a computing power resource model within the region is built, and a comprehensive resource model including computing power supply curves and power demand curves is generated. Based on the comprehensive resource model, a digital twin model of the computing center and the power system is constructed to generate a real-time mapping relationship between the physical entity and the virtual model; The real-time monitoring data is synchronized with the digital twin model to generate a dynamically updated multi-dimensional twin, which supports the visualization of energy consumption distribution, equipment status and environmental parameters. Based on the multi-dimensional twin simulation of different computing power scheduling strategies, a collaborative optimization decision scheme for computing power resources and power supply is generated.

2. The computing power optimization method for power data maintenance according to claim 1, characterized in that, The process of collecting multi-source data based on a unified data interface protocol and generating a standardized fused data set includes: The hardware operating parameters of the computing center are collected through IoT terminal devices. These hardware operating parameters include the CPU utilization, memory usage, and network bandwidth usage of the server cluster. The software load data is collected by a software probe, and the software load data includes application process resource consumption, number of virtual machine instances, and containerized service deployment status. The energy consumption data is collected by smart meters, and the energy consumption data includes the total power consumption of the computing center, rack-level power consumption and cooling equipment energy consumption. The power supply data of the power company is collected through the power dispatching system interface. The power supply data includes real-time grid load, substation output and regional power supply reliability indicators. The hardware operating parameters, software load data, energy consumption data, and power supply data are converted and aligned in structure according to the unified data interface protocol to generate the standardized fused data set.

3. The computing power optimization method for power data maintenance according to claim 1, characterized in that, The process of encrypting and authenticating the standardized fused data set to generate a securely transmitted computing power-electricity fused data stream includes: The standardized fused data set is encrypted using an asymmetric encryption algorithm to generate an encrypted data packet; Perform two-way authentication of the data acquisition terminal and the receiving server to generate an authentication token; The authentication token is bound to the encrypted data packet to generate an encrypted transmission stream with authentication information; The encrypted transmission stream is transmitted to the central data processing platform through a dedicated communication channel. The central data processing platform performs decryption and integrity verification to generate the securely transmitted computing power-electricity fusion data stream.

4. The computing power optimization method for power data maintenance according to claim 1, characterized in that, The process of building a computing power resource model within the region based on the securely transmitted computing power-power fusion data stream, and generating a comprehensive resource model including computing power supply curves and power demand curves, includes: Historical computing power consumption sequences and corresponding timestamps are extracted from the securely transmitted computing power-power fusion data stream to construct a computing power demand time series; Power supply capacity data and electricity price information are extracted from the securely transmitted computing power-power fusion data stream to construct a power supply capacity time series. A correlation function between computing power consumption and electricity consumption is established, which is obtained based on regression analysis of the computing power demand time series and electricity consumption data. The computing power demand time series is converted into an equivalent electricity demand series based on the correlation function. The equivalent power demand sequence and the power supply capacity time series are merged and standardized to generate a comprehensive resource model that includes a computing power supply curve and a power demand curve. The computing power supply curve represents the total available computing power over time, and the power demand curve represents the predicted power consumption over time.

5. The computing power optimization method for power data maintenance according to claim 4, characterized in that, The construction of a digital twin model of the computing center and the power system, generating a real-time mapping relationship between the physical entity and the virtual model, includes: Based on the comprehensive resource model as the basic data architecture, a computing power center digital twin is constructed, which includes a virtual server cluster model, a virtual network topology, and a virtual cooling system model. Based on the power grid geographic information system data and the power supply capacity time series, a digital twin of the power system is constructed, which includes a virtual transmission line model, a virtual substation model and a virtual distribution network model. Establish an energy flow and information flow connection channel between the computing center digital twin and the power system digital twin, and generate a real-time mapping relationship between the physical entity and the virtual model. The real-time mapping relationship ensures that the state changes of the virtual model are synchronized with the actual state of the physical entity.

6. The computing power optimization method for power data maintenance according to claim 1, characterized in that, The step of synchronizing real-time monitoring data with the digital twin model to generate a dynamically updated multi-dimensional twin includes: The system continuously collects real-time hardware operating parameters, real-time software load data, and real-time energy consumption data of the computing center to generate a real-time computing power monitoring data stream. The system continuously collects real-time grid load data, substation operation status data, and power supply quality data of the power system to generate a real-time power monitoring data stream. The real-time computing power monitoring data stream is injected into the computing power center digital twin to drive the status update of the virtual server cluster model, virtual network topology, and virtual cooling system model; The real-time power monitoring data stream is injected into the power system digital twin to drive the state updates of the virtual transmission line model, virtual substation model, and virtual distribution network model. The updated computing center digital twin and the power system digital twin are merged to generate the dynamically updated multi-dimensional twin, which supports the visualization of energy consumption distribution heatmap, equipment status topology map and environmental parameter change curves.

7. The computing power optimization method for power data maintenance according to claim 6, characterized in that, The process of generating a collaborative optimization decision scheme for computing resources and power supply based on the simulation of different computing power scheduling strategies using the multi-dimensional twin includes: A load migration strategy simulation scenario is set in the multi-dimensional twin, and the load migration strategy simulation scenario includes the operation of migrating virtual machine instances from a high-power physical server to a low-power physical server. A task scheduling strategy simulation scenario is set in the multi-dimensional twin, which includes the operation of delaying computationally intensive tasks to the period of low electricity price. A device start-stop policy simulation scenario is set in the multi-dimensional twin, which includes the operation of dynamically shutting down or starting some servers and cooling equipment according to the predicted load. Run the load migration strategy simulation scenario, the task scheduling strategy simulation scenario, and the device start-stop strategy simulation scenario to obtain the corresponding simulated power consumption curve, simulated computing power supply curve, and simulated system stability index, respectively. Based on a preset optimization objective function, the simulated power consumption curve, the simulated computing power supply curve, and the stability index of the simulated system are evaluated. The optimal strategy combination based on the comprehensive evaluation results is selected to generate a collaborative optimization decision scheme for computing power resources and power supply. The evaluation of the simulated power consumption curve, the simulated computing power supply curve, and the stability index of the simulated system based on a preset optimization objective function includes: An optimization objective function is constructed with the dual objectives of minimizing total operating cost and maximizing the reliability of computing power services. The total operating cost includes electricity cost and equipment depreciation cost. Substitute the simulated power consumption curve into the optimization objective function to calculate the predicted power cost under the corresponding strategy; The simulated computing power supply curve is compared with the preset computing power demand baseline to calculate the computing power demand satisfaction rate index. The stability indicators of the simulated system, including the mean time between failures (MTBF) and the power supply voltage qualification rate, are quantified into a reliability score. The predicted electricity cost, the computing power demand satisfaction rate, and the reliability score are weighted and summed based on the weighting coefficients to generate a comprehensive evaluation score for each strategy combination. Compare the overall evaluation scores of all simulated strategy combinations and select the strategy combination with the highest score as the evaluation result.

8. The computing power optimization method for power data maintenance according to claim 3, characterized in that, The step of encrypting the standardized fused data set using an asymmetric encryption algorithm to generate an encrypted data packet includes: The central data processing platform generates a pair of public and private keys for an asymmetric encryption algorithm and distributes the public key to each data acquisition terminal. Each of the data acquisition terminals uses the received public key to perform encryption operations on the standardized fused data set to generate ciphertext data segments; Each of the data acquisition terminals adds a timestamp and a data source identifier to each encrypted data segment, generating an encrypted data unit with time sequence and source information; All encrypted data units belonging to the same transmission batch are encapsulated and sorted to generate a complete encrypted data packet.

9. The computing power optimization method for power data maintenance according to claim 5, characterized in that, The process of establishing an energy flow and information flow connection channel between the computing center digital twin and the power system digital twin, and generating a real-time mapping relationship between the physical entity and the virtual model, includes: A virtual energy output interface is defined in the digital twin of the computing center, and the virtual energy output interface corresponds one-to-one with the actual power consumption metering point of the computing center; A virtual energy input interface is defined in the power system digital twin, and the virtual energy input interface corresponds one-to-one with the substation outgoing port that supplies power to the computing center; A virtual energy flow connection channel is established from the energy output virtual interface to the energy input virtual interface for bidirectional transmission of real-time data on power, electricity, and energy efficiency. A virtual information flow connection channel is established between the computing center digital twin and the power system digital twin for bidirectional transmission of load forecast information, scheduling instructions and fault alarm information; Through the virtual energy flow connection channel and the virtual information flow connection channel, the state changes of the computing center digital twin trigger the real-time update of the associated parameters of the power system digital twin, generating a real-time mapping relationship between the physical entity and the virtual model.

10. A computing power optimization system for power data maintenance, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the computing power optimization method in power data maintenance as described in any one of claims 1 to 9.

Images