New energy full direct current data center multi-stream mutual fusion method based on multi-scale entropy

Abstract

The invention discloses a new energy full direct current data center multi-stream mutual fusion method based on multi-scale entropy, and belongs to the technical field of energy internet and intelligent data center operation control. The method comprises four steps of multi-scale information acquisition, adaptive decomposition and calculation, system quantitative identification and multi-flow optimization and regulation, and is a collaborative regulation method based on the deep fusion of the computing power flow, the power flow and the cold flow of the full-direct-current power supply architecture. And multi-scale entropy analysis is combined to realize multi-resource flexible potential identification, system complexity evaluation and active response optimization under different time scales. According to the method, aiming at the problem of lack of unified description and dynamic coordination among computing power flow, electric power flow and cold flow, multi-scale sample entropy is introduced to a time sequence signal level, and four stages of state modeling, complexity quantification, collaborative optimization and active regulation and control are formed to form a complete closed loop; and multi-scale identification, multi-level flexible potential regulation and control and robust optimization control of system complexity under multi-energy flow are realized.

Description

Technical Field

[0001] This invention relates to the field of energy internet and intelligent data center operation control technology, and in particular to a multi-stream integration method for new energy all-DC data centers based on multi-scale entropy. Background Technology

[0002] Currently, data centers generally adopt AC power supply architectures or partial DC retrofit solutions. Most solutions focus on improving the efficiency of a single subsystem, while the integration of computing, power, and cooling flows, as well as complexity measurement and joint control methods based on multiple time scales, are still in the exploratory or preliminary experimental stages. Furthermore, existing single-subsystem solutions have many problems: the computing load system, power system, and environmental control system operate independently, lacking cross-energy flow coordination mechanisms; with the increase in the proportion of renewable energy access, data center loads are rigid, and load-side response is lagging, unable to adapt to the fluctuations in wind and solar power output; scheduling algorithms are insufficient in terms of cross-time scale response and robustness; energy management systems mainly use PUE as the core indicator, lacking real-time quantitative methods to characterize the dynamic complexity and adjustability of the system across multiple time scales. Therefore, there is an urgent need to develop a key technology to characterize the multi-energy flow coupling laws and system dynamic complexity of a 100% renewable, all-DC data center at different time scales. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a multi-flow integration method for new energy all-DC data centers based on multi-scale entropy, aiming to solve the problem that current methods cannot achieve multi-scale identification of system complexity, multi-level flexible potential regulation and robust optimization control under multi-energy flow.

[0004] Technical solution: A multi-stream integration method for new energy all-DC data centers based on multi-scale entropy, comprising: S1. Multi-scale information acquisition: Specific information includes computing power data, power supply data, and environmental control data; S2. Adaptive Decomposition and Computation: The time-scale adaptive partitioning algorithm based on variational mode decomposition and empirical mode decomposition is adopted to decompose the original time-series signal into multiple time scales; the sample entropy of the reconstructed vector sequence is calculated at each scale, the embedding dimension and similarity threshold are automatically adjusted and unified, and a data-model dual-driven multidimensional source-load prediction system is established by combining deep learning and other methods. S3. System Quantitative Identification: Based on unified modeling, the collaborative coupling characteristics between energy flows are further characterized by multi-dimensional and multi-scale sample entropy, so as to realize the quantitative identification of the system's flexibility potential and adjustment boundary. S4. Multi-flow optimization and regulation: Entropy-driven multi-energy flow collaborative optimization and robust regulation: By continuously monitoring the multi-scale entropy change trend, an entropy-driven multi-objective optimization model is established to actively identify and adaptively correct system stability, coupling imbalance and new energy fluctuation impact, so as to achieve optimal joint energy efficiency regulation of the three energy flows of electricity, computing and cooling.

[0005] Furthermore, the computing power data in step S1 includes CPU utilization, load change rate, and task migration frequency.

[0006] Furthermore, in step S1, the power-side data includes voltage, current, and power.

[0007] Furthermore, in step S1, the environmental control side data includes real-time cooling power, cooling output, cooling water / refrigerant temperature, chiller start / stop, and long-term cold storage device status such as cold storage capacity / electricity storage and daily-scale release / charge plan.

[0008] Beneficial effects: (1) The problem of “series response” caused by the independent operation of power, computing power and environmental control systems in the prior art can be solved by unified modeling of multi-energy flow driven by adaptive multi-scale sample entropy. This invention constructs a unified energy flow model of computing power flow, power flow and cooling flow, incorporates multi-level rack-level computing power migration and dynamic coupling of energy flow into the source load prediction model, and combines sample entropy to quantitatively characterize the coupling degree of different physical quantities, breaking the control logic barrier between traditional systems, so that power flow, heat flow and computing flow form a complementary and mutually supportive synergistic relationship, significantly improving the overall operating energy efficiency of the data center.

[0009] (2) To address the issues of rigid load and low renewable energy absorption rate in existing data centers, this invention identifies the complexity of energy flow coordination and the boundary of flexible potential through multi-dimensional, multi-scale sample entropy. It establishes a dynamic mapping relationship between computing power task migration, environmental control load adjustment, and energy storage scheduling, enabling dynamic evaluation and in-depth exploration of the flexible adjustment potential of data centers in multiple scenarios. Based on the feedback constraints of entropy change rate and energy flow adjustment rate, the system can achieve flexible adjustment at the second / minute / hour level, reducing passive dependence on energy storage equipment and mitigating the impact of power supply fluctuations on system stability.

[0010] (3) To address the limitations of existing technologies that rely on static indicators such as PUE, this invention uses multi-scale entropy as the core parameter to perform multi-scale coarse-grained analysis of the temporal fluctuations of electrical, computational, and cooling energy flows, which can accurately identify the full-domain complexity from second-level disturbances to hour-level slow-changing trends. Combining an entropy-driven multi-objective optimization model and feedback adjustment mechanism, a continuous power supply guarantee strategy for key loads adapted to 100% renewable energy scenarios is developed, and the system can actively correct coupling imbalances and fluctuation impacts.

[0011] (4) This invention enables all-DC data centers to adapt to 100% renewable energy power supply through a closed-loop mechanism of "state modeling - complexity quantification - collaborative optimization - active regulation". Its core lies in integrating multi-scale entropy analysis throughout the entire process of energy flow integration, which not only solves the technical bottleneck of cross-domain collaboration, but also provides a quantifiable and controllable dynamic evaluation tool, laying the foundation for data centers to participate in grid peak shaving and realize the integrated operation of "source-grid-load-storage-computing", and helping the technological upgrade of energy internet and intelligent data centers. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the method. Detailed Implementation

[0013] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0014] like Figure 1 As shown in the figure, the multi-stream integration method for new energy all-DC data centers based on multi-scale entropy provided in this embodiment specifically includes: Step 1: Multi-scale information acquisition Collect multi-scale information on computing power, power and environmental control system operation data, specifically including: computing power data (CPU utilization, load change rate, task migration frequency, etc.), power data (voltage, current, power, etc.), and environmental control data (real-time cooling power, cooling output, cooling water / refrigerant temperature, chiller start-up and shutdown, and long-term cold storage device status such as cold storage capacity / electricity storage and daily scale release / charge plan, etc.).

[0015] Step 2: Adaptive Decomposition and Computation Subsequently, adaptive time-scale decomposition and adaptive multi-scale sample entropy calculation are performed. The time-scale adaptive partitioning algorithm based on variational mode decomposition (VMD) and empirical mode decomposition (EMD) is used to decompose the original time-series signal into multiple time scales. The sample entropy of the reconstructed vector sequence is calculated at each scale, and the embedding dimension and similarity threshold are automatically adjusted and unified. Combined with deep learning and other methods, a data-model dual-driven multi-dimensional source-load prediction system is established to achieve unified representation and dynamic coupling of computing, electricity and cooling energy flows.

[0016] Step 3: System Quantitative Recognition Based on the identification of collaborative complexity and extraction of flexible potential using multivariate and multiscale sample entropy, and building upon unified modeling, this paper further characterizes the collaborative coupling features between energy flows through multivariate and multiscale sample entropy, thereby achieving quantitative identification of the system's flexible potential and adjustment boundaries. Specifically, this includes: introducing an adaptive training mechanism and uncertainty modeling method to quantify the flexible potential at different levels in real time, assessing the system's adjustability and dynamic response capabilities under typical operating scenarios; constructing and calculating a multivariate joint sample entropy state matrix to obtain the collaborative complexity matrix between energy flows, reflecting the coupling strength between them; and identifying coupling bottlenecks and flexible boundaries from the perspectives of load characteristic evolution, energy efficiency elasticity, and scheduling plasticity, establishing a quantitative index of flexible potential to evaluate the flexible adjustability potential of each energy flow.

[0017] Step 4: Multi-flow optimization and control Entropy-Driven Multi-Energy Flow Cooperative Optimization and Robust Regulation: By continuously monitoring multi-scale entropy change trends, an entropy-driven multi-objective optimization model is established to proactively identify and adaptively correct system stability, coupling imbalances, and new energy fluctuation impacts, achieving optimal joint energy efficiency regulation of the three energy flows: electricity, computing, and cooling. This includes: researching the system's safety and stability boundary characteristics based on anomaly scenario feature extraction to establish a continuous power supply guarantee optimization scheduling method; constructing objective functions based on multi-scale entropy values, using computing power, power, and load as constraints to achieve multi-scale entropy-driven dynamic resource adjustment, reaching second / minute / hour-level proactive response optimization; and developing an entropy feedback adjustment mechanism based on real-time entropy monitoring and anomaly detection to achieve robust system optimization.

[0018] This method addresses the lack of a unified description and dynamic coordination among computing power flow, power flow, and cooling flow. By introducing multi-scale sample entropy at the time-series signal level, it forms a complete closed loop consisting of four stages: "state modeling, complexity quantification, collaborative optimization, and active regulation." This enables multi-scale identification of system complexity under multi-energy flow, multi-level flexible potential regulation, and robust optimization control.

[0019] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for multi-stream integration in a new energy all-DC data center based on multi-scale entropy, characterized in that, include: S1. Multi-scale information acquisition: Specific information includes computing power data, power supply data, and environmental control data; S2. Adaptive Decomposition and Computation: The time-scale adaptive partitioning algorithm based on variational mode decomposition and empirical mode decomposition is adopted to decompose the original time-series signal into multiple time scales; the sample entropy of the reconstructed vector sequence is calculated at each scale, the embedding dimension and similarity threshold are automatically adjusted and unified, and a data-model dual-driven multidimensional source-load prediction system is established by combining deep learning and other methods. S3. System Quantitative Identification: Based on unified modeling, the collaborative coupling characteristics between energy flows are further characterized by multi-dimensional and multi-scale sample entropy, so as to realize the quantitative identification of the system's flexibility potential and adjustment boundary. S4. Multi-flow optimization and regulation: Entropy-driven multi-energy flow collaborative optimization and robust regulation: By continuously monitoring the multi-scale entropy change trend, an entropy-driven multi-objective optimization model is established to actively identify and adaptively correct system stability, coupling imbalance and new energy fluctuation impact, so as to achieve optimal joint energy efficiency regulation of the three energy flows of electricity, computing and cooling.

2. The method for multi-stream integration of new energy all-DC data centers based on multi-scale entropy according to claim 1, characterized in that, The computing power data in step S1 includes CPU utilization, load change rate, and task migration frequency.

3. The method for multi-stream integration of new energy all-DC data centers based on multi-scale entropy according to claim 1, characterized in that, In step S1, the power-side data includes voltage, current, and power.

4. The method for multi-stream integration of new energy all-DC data centers based on multi-scale entropy according to claim 1, characterized in that, In step S1, the environmental control side data includes real-time cooling power, cooling output, cooling water / refrigerant temperature, chiller start / stop, and long-term cold storage device status such as cold storage capacity / electricity storage and daily-scale release / charge plan.

Images