Method for energy efficiency optimization of ssd data center and energy efficiency optimization system performing the same

Abstract

The application relates to the technical field of data processing, in particular to an energy efficiency optimization method of an SSD data center and an energy efficiency optimization system thereof, wherein when an SSD device operation instruction is received, operation data of the SSD device is acquired and input into a preset target load prediction model to output a predicted load in a future period of time; when the predicted load is greater than an upper limit value of a preset load threshold interval, an APL reserved energy threshold value is reduced, and surplus reserved energy is converted and transmitted to a PCIe slot level energy storage module for storage; when the predicted load is less than a lower limit value of the preset load threshold interval, the APL reserved energy threshold value is increased, and the remaining electric energy is recovered to an SSD built-in capacitor. The application dynamically adjusts the APL reserved energy threshold value, reasonably allocates and stores energy according to the load condition, realizes full-dimensional energy efficiency optimization of the SSD data center, helps to improve the service life of the SSD, meets complex technical requirements and reduces energy consumption.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an energy efficiency optimization method for SSD data centers and an energy efficiency optimization system for implementing the same. Background Technology

[0002] In today's digital age, solid-state drives (SSDs), as core data storage devices, have garnered significant attention for their performance, reliability, and energy consumption. Higher SSD performance and larger capacity generally lead to higher energy consumption. With the exponential growth of data volume, consumer electronics, from smartphones and laptops to tablets, require large-capacity, high-performance SSDs to store data and improve device speed. Enterprise data centers rely heavily on SSDs to ensure rapid read and write of massive amounts of data, meeting real-time business requirements. Consequently, SSD technology is constantly iterating and upgrading, resulting in increasingly complex internal architectures. From the types of flash memory chips, such as TLC, MLC, and QLC, to the performance improvements of controller chips, and various advanced technologies, such as the application of protocols from SATA to NVMe, all contribute to the continuous optimization of SSD performance and functionality. However, this also increases the difficulty and complexity of testing. It requires compatibility with various server platforms, precise control of test conditions, monitoring of numerous parameters, and ensuring data consistency. Manual testing is not only inefficient but also prone to human error. Therefore, it necessitates building a comprehensive energy efficiency optimization system to create an automated testing laboratory, meeting complex technical requirements and extending the lifespan of SSDs. However, in current SSD data centers or testing labs, the existing energy efficiency optimization methods for SSD data centers simply rely on the SSD's own capacitors to save energy. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention provides an energy efficiency optimization method for SSD data centers and an energy efficiency optimization system for implementing it, thereby improving the overall performance of the SSD data center.

[0004] The first aspect of this application provides an energy efficiency optimization method for SSD data centers, the method comprising: When an SSD device's execution command is received, the execution data of the SSD device is obtained; The operating data is input into a preset target load prediction model so that the target load prediction model outputs the predicted load of the SSD device for a future period of time. When it is determined that the predicted load is greater than the upper limit of the preset load threshold range, the APL reserved energy threshold is reduced, and the excess reserved energy is converted by the circuit and transferred to the PCIe slot-level energy storage module for storage. When it is determined that the preset load is less than the lower limit of the preset load threshold range, the APL reserved energy threshold is increased, and the remaining electrical energy is recovered to the SSD built-in capacitor.

[0005] In an optional implementation, the operational data includes the SSD chip's chip temperature, and the method further includes: Obtain basic APL reservations; A risk assessment is performed on the SSD built-in capacitors of the SSD device based on the chip temperature and the predicted load, and the risk assessment result is obtained. The basic adjustment coefficient is determined based on the risk assessment results. The final adjustment factor is calculated based on the basic adjustment coefficient, temperature risk coefficient, load fluctuation coefficient, and health coefficient; the temperature risk coefficient is determined based on the chip temperature, the load fluctuation coefficient is determined based on the predicted load, and the health coefficient is determined based on the capacitor health status of the SSD's built-in capacitors. Calculate the final APL reserve based on the basic APL reserve and the final adjustment factor; Obtain the current total energy and calculate the recoverable excess energy based on the current total energy and the final APL reserve; The excess energy is stored.

[0006] In an optional implementation, the method further includes When it is determined that the chip temperature is greater than a first preset temperature threshold, the heat of the SSD chip is obtained; The heat from the chip is conducted to the thermoelectric conversion module through the graphene thermal conductive layer, so that the thermoelectric conversion module converts the heat from the chip into electrical energy output. The electrical energy is processed by a DC-DC converter so that the DC-DC converter outputs a stable DC power. The stable DC power is stored in the supercapacitor of the energy storage device.

[0007] In an optional implementation, the runtime data further includes SSD load, and the method further includes: When the SSD load exceeds a first preset load threshold, the APL reserved energy is reduced to release excess electrical energy. When the temperature difference is greater than the first preset temperature threshold, the thermoelectric conversion module uses the temperature difference to convert heat energy into electrical energy to obtain thermoelectric electrical energy; the temperature difference is determined by the chip temperature and the heat sink temperature of the SSD heat sink. The total recovered energy is determined based on the excess electrical energy and the thermoelectric conversion electrical energy. The total recovered energy is transferred to the energy storage device for storage.

[0008] In an optional implementation, the method further includes: When the SSD load is less than the second preset load threshold, the SSD device will be switched from its current operating state to a low-power mode. The operation of converting heat energy into electrical energy using the temperature difference through the thermoelectric conversion module is performed to obtain thermoelectric-converted electrical energy. The thermoelectric energy is transmitted to a rack-mounted supercapacitor array, and the thermoelectric energy is stored in the supercapacitors through the rack-mounted supercapacitor array.

[0009] In an optional implementation, the method further includes: Acquire real-time power data of the rack-mounted supercapacitor array; The real-time power data is compared with a first preset power threshold. When it is determined that the real-time power data is greater than the first preset power threshold, the power of the rack-level supercapacitor array is transmitted to the UPS system of the data center through the DC-DC conversion circuit; Acquire UPS power data and real-time power grid quality data for the UPS system; When the real-time power data meets the second preset power threshold, and / or the UPS power data meets the third preset power threshold, and / or when the real-time power grid power quality is detected to have returned to normal and stable, the UPS system that transmits power from the rack-level supercapacitor array to the data center through the DC-DC conversion circuit is stopped.

[0010] In an optional implementation, the method further includes: Based on the preset load, the corresponding target energy storage strategy is matched in the preset energy storage strategy decision matrix; The energy storage parameters are dynamically adjusted according to the target energy storage strategy. The adjusted energy storage parameters are sent to all energy storage devices so that all energy storage devices can store energy according to the adjusted energy storage parameters.

[0011] In an optional implementation, the method further includes: Acquire PCM temperature data and SSD chip junction temperature of the SSD chip; RT44HC paraffin-based PCM is embedded in the SSD heatsink, and the PCM temperature data is the temperature data corresponding to the RT44HC paraffin-based PCM. The temperature stage of the SSD heatsink is determined based on the PCM temperature data and the junction temperature of the SSD chip. When it is determined that the SSD heatsink is at its peak temperature, the active cooling fan of the thermoelectric conversion module is turned off. When it is determined that the SSD heatsink is at its lowest temperature, the liquid cooling system is activated.

[0012] In an optional implementation, the method further includes: Based on the chip temperature and the preset temperature threshold segmentation range, the fan speed and liquid cooling flow rate are adjusted according to the preset load. The preset temperature threshold segmentation range includes multiple temperature threshold ranges, and each temperature threshold range corresponds to an adjustment method.

[0013] A second aspect of this application provides an energy efficiency optimization system for SSD data centers, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned energy efficiency optimization method for SSD data centers.

[0014] In summary, the energy efficiency optimization method for SSD data centers and the energy efficiency optimization system implementing it provided in this application, upon receiving an operation command from an SSD device, acquires the operation data of the SSD device; inputs the operation data into a preset target load prediction model, so that the target load prediction model outputs the predicted load of the SSD device for a future period of time, enabling advance understanding of the load situation of the SSD device at different time periods. When it is determined that the predicted load is greater than the upper limit of the preset load threshold range, the APL reserved energy threshold is lowered, and the excess reserved energy is converted by the circuit and transferred to the PCIe slot-level energy storage module for storage. When it is determined that the preset load is less than the lower limit of the preset load threshold range, the APL reserved energy threshold is raised, and the remaining electrical energy is recovered to the SSD's built-in capacitor. By lowering the APL reserved energy threshold under high load conditions and storing excess energy in the PCIe slot-level energy storage module, reasonable energy allocation and storage are achieved, avoiding energy waste, improving energy efficiency, better adapting to energy demands under high load conditions, ensuring the performance of SSD devices under high load, and resolving the contradiction between energy consumption and performance under high load. Under low load conditions, increasing the APL reserved energy threshold recovers the remaining power to the SSD's built-in capacitor, achieving energy recovery and utilization, improving energy efficiency. It can dynamically adjust the reserved energy according to the load conditions to ensure effective energy utilization under different loads, further optimizing the SSD's energy efficiency. Attached Figure Description

[0015] Figure 1This is a schematic diagram of the physical hierarchy architecture of an energy efficiency optimization system for an SSD data center, as shown in an embodiment of this application. Figure 2 This is another schematic diagram of an energy efficiency optimization system for an SSD data center, as shown in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the multi-level energy storage collaboration between different levels in an SSD data center energy efficiency optimization system, as shown in an embodiment of this application. Figure 4 This is a flowchart illustrating an energy efficiency optimization method for an SSD data center, as shown in an embodiment of this application. Figure 5 This is a schematic diagram of the waste heat recovery path in an energy efficiency optimization method for an SSD data center, as shown in an embodiment of this application. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] The following will clearly and completely describe the concept, specific structure, and technical effects of the present invention in conjunction with embodiments and accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the scope of protection of the present invention. Furthermore, all connections / linkages involved in the patent do not simply refer to direct contact between components, but rather to the ability to form a better connection structure by adding or reducing connecting accessories according to specific implementation conditions. The various technical features in this invention can be combined interactively without contradicting each other.

[0018] Reference Figure 1 The diagram shown is a schematic diagram of the physical layer architecture of an energy efficiency optimization system for an SSD data center, as illustrated in an embodiment of this application.

[0019] The energy efficiency optimization system of the SSD data center includes SSD internal level devices / modules (including main control chip, NAND flash memory, APL dynamic management unit and satellite thermoelectric generator), PCIe slot level devices / modules (including thermoelectric conversion module, local energy storage module and smart energy router), rack level devices / modules (including supercapacitor array, DC / DC conversion matrix and liquid cooling distribution system) and data center level devices / modules (including UPS power system, energy management server and grid connection).

[0020] The thermoelectric conversion module consists of a heat sink (aluminum alloy), graphene thermally conductive adhesive, a Bi2Te3 thermopile, a ceramic substrate, and copper electrodes. It employs a micropillar array structure (50μm diameter, 200μm height) to increase the thermal contact area, and quantum well technology to reduce phonon thermal conductivity, achieving a ZT value of 1.2 (room temperature). The intelligent energy router enables multi-channel energy harvesting and aggregation (supporting 8 inputs), adaptive voltage conversion (2.5V-12V dynamic adjustment), fault isolation protection (fuse current <5A), and a digital communication interface (I2C / SMBus).

[0021] This application integrates dynamic APL energy recovery, heat-to-electricity conversion, and a multi-level energy storage network to construct a closed-loop energy cycle system, namely an energy efficiency optimization system for SSD data centers. It employs a three-level energy storage network design to improve the energy efficiency of the data center. The first level of energy storage utilizes the SSD's built-in capacitors; the second level utilizes energy storage modules integrated into the PCIe slots / backplane; and the third level utilizes a rack-level supercapacitor array.

[0022] (1) First level: SSD built-in capacitors.

[0023] The SSD's built-in capacitors utilize Advanced Power Loss (APL) to dynamically manage energy reserves and achieve energy recovery. Specifically, the capacitor charge (capacitor_charge) is obtained to determine the current energy storage status and available energy level. The `calculate_base_reserve()` function is then called to calculate the base APL reserve based on the current capacitor charge and a preset safety threshold (e.g., a minimum 30% reserve). For example, if the capacitor charge is 80%, the base reserve might be set to 80% × 20% = 16% (assuming a base reserve ratio of 20%). Next, the `lstm_predict_load()` function and a preset target predicted load model are called to predict the load trend (predicted_load) for the next 30 minutes based on historical load data from the past 30 minutes. For example, the target load prediction model predicts that the load will increase from the current 60% to 95% in the next 30 minutes. This allows for further multi-dimensional risk assessment based on the SSD chip temperature and the predicted load. If the current temperature exceeds 70°C, it is marked as a high temperature risk, and energy reserves need to be increased to cope with potential thermal runaway; if the predicted load exceeds 90%, it is marked as a high load risk, and more energy needs to be reserved to ensure peak performance; at the same time, the capacitor health status (SOH) is evaluated by the get_capacitor_soh() function. If the SOH is below 80%, it is marked as a health risk, and reserves need to be increased to compensate for aging losses.

[0024] Next, the basic adjustment coefficient is determined based on the risk assessment results. When there is a high temperature or high load risk, the adjustment coefficient is 1.2 (increasing the reserve by 20%); in low-risk scenarios (temperature <45℃ and predicted load <30%), the adjustment coefficient is 0.8 (reducing the reserve by 20%); and under normal conditions, the adjustment coefficient is 1.0. The basic adjustment coefficient is multiplied by the temperature risk coefficient, load fluctuation coefficient, and capacitor health coefficient to obtain the final adjustment factor. For example, if the basic adjustment coefficient is 1.2 (high load warning), the temperature risk coefficient is 1.3 (temperature 68℃), the load fluctuation coefficient is 1.2 (predicted load fluctuation >30%), and the health coefficient is 0.9 (SOH = 72%), then the final adjustment factor is 1.2 × 1.3 × 1.2 × 0.9 ≈ 1.68. Additionally, the basic APL reserve is multiplied by the adjustment coefficient to obtain a preliminary calculation result. For example, 16% × 1.68 ≈ 26.9%. The upper and lower limits are used to ensure that the reserve amount is within a safe range. The minimum value constraint is MIN_APL_RESERVE (e.g., 10%), and the maximum value constraint is MAX_APL_RESERVE (e.g., 30%), thus determining the final APL reserve amount. For example, if 26.9% is within the range of [10%, 30%], the final APL reserve amount is determined to be 26.9%.

[0025] Furthermore, the `get_total_energy()` function is called to obtain the current total energy, which is then subtracted from the final APL reserve to obtain the recoverable excess energy. For example, if the current total energy is 100W, the excess energy = 100W - 26.9W = 73.1W. The recovery efficiency is dynamically adjusted based on the current temperature (efficiency = 90% when temperature ≤ 40℃; efficiency = 90% - (temperature - 40) × 1% when temperature > 40℃) to calculate the actual usable energy for priority allocation. For example, at 55℃, the efficiency = 90% - (55 - 40) × 1% = 75%, and the actual usable energy = 73.1W × 75% ≈ 54.8W. In high-load scenarios (current load > 70%), 70% of the actual usable energy is used for performance enhancement (such as accelerating data read / write), and 30% is stored in the PCIe slot module; for example, 54.8W × 70% ≈ 38.4W is used for performance improvement, and 54.8W × 30% ≈ 16.4W is stored in the energy storage module. In low-load scenarios, all energy is stored in the PCIe slot module. Additionally, the system writes the calculated final APL reserve to the SSD controller to update the energy management strategy. The `allocate_to_performance_boost()` and `store_in_pcie_module()` functions are called to allocate excess energy to the performance optimization and energy storage systems. The deviation between the actual load and the predicted load is recorded. If the error exceeds 15%, the target load prediction model parameters are updated to improve subsequent prediction accuracy.

[0026] (2) Second level: PCIe slot / backplane integrated energy storage module.

[0027] Among them, PCIe slot energy storage modules, also known as PCIe backplane integrated energy storage modules, are built based on existing PCIe links. PCIe slots, as a hardware interface supporting high-speed data transmission, are widely used in data centers. With technological advancements, especially the progress of the PCIe 7.0 protocol, the PCIe 7.0 specification is establishing rules related to energy recovery. This specification aims to improve data transmission rates while optimizing energy efficiency. PCIe 7.0 continues to use and optimizes the four-level pulse amplitude modulation (PAM4) signaling technology introduced in PCIe 6.0, allowing two data bits to be encoded per clock cycle, improving data transmission efficiency and creating better conditions for hardware modules to achieve energy recovery.

[0028] In this embodiment, the PCIe slot / backplane integrated energy storage module includes a supercapacitor storage unit and a corresponding energy conversion and control circuit. The supercapacitor, as the primary energy storage component, boasts advantages such as fast charging and discharging speeds and long cycle life, enabling rapid response to energy fluctuations in the SSD device. The energy conversion and control circuit is responsible for efficiently and stably storing the electrical energy generated from the APL energy recovery and thermoelectric conversion modules into the supercapacitor, and precisely controlling the charging and discharging process of the supercapacitor according to the SSD device's load. For example, when the SSD is under high load reading and writing and the APL has excess reserved energy, the energy conversion and control circuit quickly converts the excess energy into a form suitable for supercapacitor storage and stores it in the supercapacitor; conversely, when the SSD load is too high and the built-in capacitor's charge is insufficient, the circuit can quickly release the energy from the supercapacitor to replenish the SSD's power and ensure its stable operation.

[0029] (3) Third level: rack-mount supercapacitor array.

[0030] Among them, rack-mounted supercapacitor arrays are used to centrally manage the energy surplus of multiple SSDs, enabling larger-scale energy storage and allocation.

[0031] In some embodiments, a Bi2Te3-based thermoelectric module (ZT value ≈ 1.0) can be deployed near the SSD controller chip and NAND flash memory to convert an 85°C temperature difference into 3-5V DC power (conversion efficiency of approximately 5-8%). The actual conversion efficiency is affected by multiple factors, including the magnitude of the temperature difference, material purity and fabrication process, and load matching. Specifically, the conversion efficiency of the Bi2Te3-based module increases with increasing temperature difference, but when the temperature difference exceeds a certain threshold (approximately 100°C), the material performance deteriorates due to thermal stress and other issues, leading to a slowdown or even a decrease in efficiency growth. Furthermore, high-purity Bi2Te3 materials with a complete crystal structure, combined with advanced thin-film fabrication processes (such as molecular beam epitaxy), can bring the ZT value close to the theoretical value, significantly improving conversion efficiency; conversely, high impurity levels and rough fabrication processes can lead to a significant reduction in efficiency. In addition, the electrical energy output from the thermoelectric module needs to be processed by a DC-DC boost circuit. If the circuit is mismatched with the module load, energy loss will occur, reducing the overall conversion efficiency. To steadily improve the conversion efficiency of Bi2Te3-based thermoelectric modules, an adaptive load adjustment circuit can be used to adjust the DC-DC boost circuit parameters in real time to ensure optimal matching between the module and the load. The heat dissipation structure design of the thermoelectric module can also be optimized to maintain the temperature difference between the chip and the heat sink within the efficient conversion range of 60-85℃. And / or the Bi2Te3 material can be regularly tested and maintained, and modules with degraded performance can be replaced in a timely manner.

[0032] In other embodiments, the Bi2Te3-based thermoelectric module can be replaced with a PbTe-based thermoelectric module. The lead telluride (PbTe) in the PbTe-based thermoelectric module has a high ZT value (up to 1.5-2.0) in a medium-temperature (300-600℃) environment. If the operating temperature of the data center SSD chip can be maintained in a high range, the PbTe-based module can be used to improve energy conversion efficiency. Alternatively, it can be replaced with a SnSe-based thermoelectric module. Tin selenide (SnSe) in the SnSe-based thermoelectric module is an emerging environmentally friendly thermoelectric material with a theoretical ZT value exceeding 2.0, exhibiting excellent performance in the room temperature to medium temperature range. It can also be replaced with a Si-Ge-based thermoelectric module. The silicon-germanium (Si-Ge) alloy in the Si-Ge-based thermoelectric module is commonly used in high-temperature (above 600℃) environments, with a ZT value reaching 1.0-1.2 at high temperatures. It is suitable for data centers equipped with special heat dissipation devices that can raise the waste heat from the chip to a high-temperature state, achieving efficient heat-to-electricity conversion.

[0033] It should be noted that in practical applications, alternative modules can be selected based on a comprehensive consideration of the data center's chip operating temperature range, environmental requirements, cost budget, and conversion efficiency requirements. For example, if the data center has extremely high environmental requirements and a sufficient budget, SnSe-based modules can be given priority; if the chip's waste heat temperature is high, PbTe-based or Si-Ge-based modules are more suitable.

[0034] The energy efficiency optimization system for SSD data centers utilizes a multi-level energy storage collaborative strategy, revolving around a four-layer structure: SSD built-in capacitors (also known as SSD internal capacitors), slot-level energy storage, rack-level supercapacitors, and data center UPS. Based on the energy storage capacity status of each layer and system load requirements, it operates according to a two-way logic: collecting excess energy from the inside out and distributing energy on demand from the outside in, achieving dynamic energy flow and efficient utilization. (See also...) Figure 3 Specifically, regarding energy harvesting, in the SSD's built-in capacitor layer, when the SSD's built-in capacitor detects that its remaining capacity is less than 80% (i.e., there is excess energy that can be released), it prioritizes actively transferring this excess energy to the slot-level energy storage, which serves as the nearest next-level energy storage unit. This allows for rapid energy absorption and localized energy recovery, preventing internal capacitors from affecting the SSD's core functions due to energy redundancy. For example, after the SSD completes a high-load read / write operation, the internal capacitors briefly accumulate 75% of their remaining capacity, triggering a transfer that pushes the excess energy to the slot-level energy storage. At the slot-level energy storage layer, its capacity is continuously monitored. When the remaining capacity is less than 90%, it receives energy from the SSD's internal capacitors. If subsequent monitoring reveals remaining capacity redundancy (having completed energy reception from the internal capacitors but not yet reaching full capacity), a "stable energy flow transfer" process is initiated to the rack-level supercapacitor. This energy flow is preliminarily integrated and relatively stable, used to supplement higher-level energy storage and improve the overall energy reserve of the data center. For example, after receiving energy from its internal capacitors, the slot-level energy storage reaches 85% of its capacity (still <90%), then transfers a stable energy flow to the rack-level supercapacitor. At the rack-level supercapacitor layer, it receives stable energy from the slot-level energy storage and serves as a centralized energy storage unit at the data center rack level, continuously accumulating energy. A "periodic replenishment" process is executed daily (e.g., during low-load periods in the early morning), transferring the accumulated energy to the data center UPS to replenish its long-term reserve energy and strengthen the data center's emergency power supply foundation. For example, at midnight each day, the rack-level supercapacitor transfers the stable energy accumulated that day to the UPS through a preset channel, completing the periodic replenishment.

[0035] Regarding energy distribution, at the data center UPS layer, as the highest-level energy storage and power supply unit, in emergency power supply scenarios (such as a mains power outage), an "emergency power supply" process is triggered to the rack-level supercapacitor, rapidly releasing stored energy to ensure basic power supply for critical rack-level equipment. For example, in the event of a sudden mains power outage, the UPS immediately outputs emergency power energy to the rack-level supercapacitor, maintaining the rack-level equipment's operation for a short period. At the rack-level supercapacitor layer, after receiving emergency power from the UPS or its own stored energy, during high-load periods (such as peak business periods), an "on-demand allocation" process is initiated, releasing energy to the slot-level energy storage. For localized high loads, energy is replenished from the rack level to the slot level, ensuring high-performance operation of the SSD slot area. For example, during peak business periods, if the rack-level supercapacitor detects a surge in load in the area where the slot-level energy storage is located, it allocates energy to the corresponding slot-level energy storage as needed. At the slot-level energy storage layer, after receiving energy allocated on demand from the rack-level supercapacitor, a "performance tuning energy injection" process is executed to precisely inject energy back into the SSD's internal capacitors. This replenishes the SSD's energy consumption under high load, ensuring continuous high-performance read and write operations and achieving a closed loop of "local load peak - precise energy replenishment." For example, after receiving rack-level energy, the slot-level energy storage detects that its SSD is experiencing internal capacitor energy shortage due to high load and immediately injects energy to maintain stable SSD operation. In other words, energy collection proceeds from SSD internal capacitors to slot-level energy storage, then to rack-level supercapacitors, and finally to data center UPS, maximizing the utilization of excess energy at each level by prioritizing local redundancy and replenishing reserves at each level. Energy allocation follows the order of data center UPS → rack-level supercapacitors → slot-level energy storage → SSD internal capacitors, prioritizing global emergencies and precisely replenishing local peaks to ensure stable system operation and performance optimization under different load scenarios.

[0036] This application achieves dynamic coordination of four levels of energy storage through a dual mechanism of "capacity threshold monitoring (triggering energy flow) + scenario-based demand (driving energy allocation)". It can both recover and utilize excess energy during SSD operation and accurately schedule energy under high load, power outage and other scenarios to support the efficient and stable operation of data centers.

[0037] Reference Figure 4 The diagram illustrates a flowchart of an energy efficiency optimization method for an SSD data center, as shown in this embodiment. The energy efficiency optimization method for the SSD data center includes the following steps.

[0038] S41, when the SSD device's operation command is received, the operation data of the SSD device is obtained.

[0039] In some embodiments, when the data center starts up and receives the running instruction of the SSD device, the system begins to collect the running data of the SSD device in real time. The running data may include SSD status parameters (current read / write load (IOPS, throughput), chip temperature, remaining power of built-in capacitors, health status (SOH), etc.), rack-level parameters (supercapacitor array power (RackCharge), temperature (TRack), power output fluctuation (ΔP), data center global indicators (PUE (Power Usage Effectiveness), WUE (Water Usage Effectiveness), remaining power of PCIe slot modules, real-time electricity price (Price), carbon emissions (Carbon)), APL reserved energy value, etc. The collection frequency can be set to once per minute to obtain the running data within a preset time period (e.g., 72 hours) to form a sequence of running data.

[0040] Once the sequence running data is acquired, the electronic device can preprocess the sequence running data, including mapping parameters of different dimensions to the [0, 1] interval (e.g., temperature 40-80℃→0.0-1.0); detecting jump values ​​through a sliding window (e.g., temperature jump >5℃ / second) and marking them as abnormal data; calculating the temperature gradient (dT / dt) and load change rate (dIOPS / dt), extracting the historical trend of PUE (mean of the last 7 days, standard deviation of fluctuation), and constructing a multi-dimensional feature vector as [Tchip, IOPS, CapCharge, RackCharge, PUE, Price].

[0041] S42, the running data is input into a preset target load prediction model so that the target load prediction model outputs the predicted load of the SSD device for a future period of time.

[0042] S43, when it is determined that the predicted load is greater than the upper limit of the preset load threshold range, the APL reserved energy threshold is reduced, and the excess reserved energy is converted by the circuit and transmitted to the PCIe slot-level energy storage module for storage.

[0043] S44, when it is determined that the preset load is less than the lower limit of the preset load threshold range, the APL reserved energy threshold is increased, and the remaining power is recovered to the SSD built-in capacitor.

[0044] When operational data for a period of time is acquired, this sequence of operational data can be input into a preset target load prediction model. This model predicts the load change trend of the SSD device within the next 30 minutes. The predicted load is compared with a preset load threshold range (e.g., [20%, 80%]). If the predicted load exceeds the upper limit of the preset load threshold range, i.e., 80%, the system automatically lowers the APL reserved energy threshold and transfers the excess reserved energy to the PCIe slot-level energy storage module for storage through circuit conversion. If the predicted load is lower than the lower limit of the preset load threshold range, i.e., 20%, the APL reserved energy threshold is appropriately increased to ensure device stability, and the remaining power is recovered to the SSD's built-in capacitor. The target load prediction model is built on an LSTM network. In addition to predicting the load trend for the next 30 minutes, it can also predict temperature changes, PUE fluctuations, etc., and can output the corresponding predicted load peak (LoadPeak), temperature extreme value (Tmax), and electricity price fluctuation (ΔPrice).

[0045] For example, assuming the sequence running data is Tchip=65℃, IOPS=4500, CapCharge=70%, RackCharge=85%, PUE=1.4, Price=0.8 yuan / kWh, the feature vector obtained after feature extraction is [0.625, 0.75, 0.7, 0.85, 1.4, 0.8]. Outputting the feature vector to the prediction model can output LoadPeak=92% and Tmax=72℃ for the next 30 minutes.

[0046] In some embodiments, when the SSD chip temperature is detected in real time, and the chip temperature exceeds 60°C, the built-in Bi2Te3 thermoelectric conversion module is activated. This module utilizes the temperature difference (ΔT) between the chip and the heatsink to generate electrical energy. The output 3-5V voltage is then used to directly charge the PCIe slot-level energy storage module via a DC-DC boost circuit. Simultaneously, if the slot-level module is fully charged (>90%), excess energy is transferred to the rack-level supercapacitor array for secondary storage. When the SSD device is under high load (>90%), the built-in SSD capacitors prioritize providing transient power compensation to maintain data write stability. If the built-in capacitor charge drops below 20%, the PCIe slot-level energy storage module is automatically triggered to discharge, replenishing the power gap. If the slot-level module is depleted and the load remains high, the rack-level supercapacitor array is activated to provide power. Furthermore, when the overall data center load is low (e.g., at night), the rack-level array can reverse-transfer the stored energy to the data center UPS system, reducing grid dependence. When the SSD device enters an idle state (load <10%), the system integrates unused electrical energy from various energy storage devices and prioritizes its use to preheat the SSD controller chip, reducing startup power consumption. The remaining energy is stored in the rack-level array to prepare for the next round of high-load scenarios. At the same time, the thermoelectric conversion module continues to run, converting the waste heat of the SSD chip into electrical energy, realizing closed-loop energy utilization.

[0047] In some embodiments, when the acquired chip temperature Tchip > 75°C, emergency cooling is immediately triggered, waste heat recovery is suspended, and chip safety is prioritized; and / or when the acquired load at the current and previous moments determines a sudden load change (dIOPS / dt > 20% / minute), a rapid energy storage release mode is activated to quickly inject rack-level energy into the slot-level module. The system can also automatically isolate faulty units and redistribute the load when a fault in the energy storage device is detected (such as a surge in capacitor internal resistance); and switch to energy storage priority mode when a mains power fluctuation exceeds ±10% to reduce dependence on the power grid.

[0048] In an optional implementation, the operational data further includes the SSD chip temperature, and the method further includes: Obtain basic APL reservations; A risk assessment is performed on the SSD built-in capacitors of the SSD device based on the chip temperature and the predicted load, and the risk assessment result is obtained. The basic adjustment coefficient is determined based on the risk assessment results. The final adjustment factor is calculated based on the basic adjustment coefficient, temperature risk coefficient, load fluctuation coefficient, and health coefficient; the temperature risk coefficient is determined based on the chip temperature, the load fluctuation coefficient is determined based on the predicted load, and the health coefficient is determined based on the capacitor health status of the SSD's built-in capacitors. Calculate the final APL reserve based on the basic APL reserve and the final adjustment factor; Obtain the current total energy and calculate the recoverable excess energy based on the current total energy and the final APL reserve; The excess energy is stored.

[0049] In some embodiments, the system can recover excess energy in the same manner as the first-stage energy storage described above, which uses APL dynamic management through SSD built-in capacitors to achieve energy recovery.

[0050] To facilitate understanding of the inventive concept of this application, the following is a code example of APL dynamic management for energy recovery: # Key Algorithm Logic def dynamic_apl_management(current_load, capacitor_charge,temperature): # Basic APL Reserve (Safety Baseline) base_apl_reserve = calculate_base_reserve(capacitor_charge) # Load forecasting (based on LSTM model) predicted_load = lstm_predict_load(historical_data, window=30) # Predict the load for the next 30 minutes # Dynamic adjustment coefficient if temperature>70: # High temperature scenario adjustment_factor = 0.8 # Increase reserve and reduce risk elif predicted_load>90: # High load warning adjustment_factor = 1.2 # Increase reserve to cope with peak demand else: adjustment_factor = 1.0 # Normal state # Calculate the final APL reservation final_apl_reserve = base_apl_reserve * adjustment_factor # Excess energy released is used for performance optimization excess_energy = get_total_energy() - final_apl_reserve if excess_energy>0: allocate_to_performance_boost(excess_energy) return final_apl_reserve By comprehensively considering multiple factors such as SSD chip temperature, predicted load, and capacitor health status through the above optional implementation methods, dynamic APL energy management is achieved, which effectively improves the energy management efficiency of SSD devices, ensures stable operation of devices, reduces energy consumption, and improves the overall system performance.

[0051] In an optional implementation, the method further includes: When it is determined that the chip temperature is greater than a first preset temperature threshold, the heat of the SSD chip is obtained; The heat from the chip is conducted to the thermoelectric conversion module through the graphene thermal conductive layer, so that the thermoelectric conversion module converts the heat from the chip into electrical energy output. The electrical energy is processed by a DC-DC converter so that the DC-DC converter outputs a stable DC power. The stable DC power is stored in the supercapacitor of the energy storage device.

[0052] Refer to together Figure 5 In some embodiments, a high-precision temperature sensor (accuracy ±0.1℃) directly mounted on the surface of the SSD chip can monitor the SSD chip temperature in real time. When the chip temperature meets a first preset temperature threshold (e.g., 55℃), a heat recovery mechanism is triggered to conduct the heat generated during the operation of the SSD chip, i.e., chip heat, to the graphene thermal conductive layer through the chip packaging material (such as a copper heat dissipation substrate). The graphene thermal conductive layer is prepared by chemical vapor deposition (CVD) with a thickness of 50μm, an in-plane thermal conductivity of up to 5300W / m・K, and a conduction efficiency of about 95%, ensuring rapid and efficient heat transfer.

[0053] The graphene thermally conductive layer is tightly bonded to the surface of the SSD chip, rapidly diffusing and guiding the heat conducted by the SSD chip to the thermoelectric conversion module, increasing the heat transfer speed by 3 times. After evenly distributing the chip heat, it is output to the thermoelectric conversion module in the form of a stable heat flow. In this embodiment, the thermoelectric conversion module can be a Bi2Te3-based thermoelectric conversion module. This module receives the chip heat transferred by the graphene thermally conductive layer and converts the temperature difference into electrical energy using the Seebeck effect. The hot and cold ends of the Bi2Te3-based thermoelectric conversion module are connected to the graphene thermally conductive layer (hot end) and the heat sink (cold end), respectively. When the temperature difference between the hot and cold ends reaches a certain threshold (e.g., 85°C), the Bi2Te3-based thermoelectric conversion module outputs 3-5V DC power with a conversion efficiency of approximately 5-8%. The electrical energy output by the Bi2Te3-based thermoelectric conversion module is an unstable DC signal and requires further processing.

[0054] Next, the electrical energy output from the Bi2Te3-based thermoelectric conversion module is processed and output via a DC-DC converter. The DC-DC converter employs an adaptive boost topology, receiving the electrical energy output from the thermoelectric conversion module, i.e., the unstable 3-5V DC power. Through internal inductors, capacitors, and a control chip, the input voltage is boosted and regulated, stabilizing it to 12V and eliminating voltage ripple, ultimately outputting a stable 12V DC power to meet the charging requirements of the energy storage device, achieving a conversion efficiency of up to 92%.

[0055] Furthermore, the stable 12V DC power output from the DC-DC converter is processed through an energy storage device. This energy storage device can be a supercapacitor module integrated into the PCIe slot / backplane, with a rated operating voltage of 12V. Upon receiving the stable 12V DC power from the DC-DC converter, the charging process is controlled by an intelligent charging management circuit, storing the energy in the supercapacitor. During high-load operation of the SSD or power outages, the energy storage device rapidly releases the energy, outputting a stable 12V DC power to provide transient power compensation or sustain data writing for the SSD.

[0056] Through the aforementioned optional implementation methods, a high-precision temperature sensor monitors the chip temperature in real time. When the temperature exceeds a threshold, a graphene thermal conductive layer efficiently conducts heat from the chip to the thermoelectric conversion module, converting the heat into electrical energy and improving energy utilization. A DC-DC converter boosts and regulates the unstable DC power, outputting stable DC power to meet the charging requirements of the energy storage device. The energy storage device stores electrical energy in a supercapacitor, which can be rapidly released during high loads or power outages in the SSD device to provide transient power compensation or sustain data writing. This solution achieves heat recovery and energy storage, optimizes the energy efficiency of the SSD data center, and improves the stability and reliability of device operation.

[0057] In an optional implementation, the runtime data further includes SSD load, and the method further includes: When the SSD load exceeds a first preset load threshold, the APL reserved energy is reduced to release excess electrical energy. When the temperature difference is greater than the first preset temperature threshold, the thermoelectric conversion module uses the temperature difference to convert heat energy into electrical energy to obtain thermoelectric electrical energy; the temperature difference is determined by the chip temperature and the heat sink temperature of the SSD heat sink. The total recovered energy is determined based on the excess electrical energy and the thermoelectric conversion electrical energy. The total recovered energy is transferred to the energy storage device for storage.

[0058] In some embodiments, the input / output operation rate (IOPS) and data throughput of the SSD device can be collected in real time. The maximum IOPS and maximum data throughput of the SSD device can be determined using the technical specifications provided by the SSD manufacturer, or by using professional performance testing tools (such as fio, CrystalDiskMark, etc.) to perform multiple tests on the SSD device to determine the maximum IOPS and maximum data throughput. When the current actual IOPS value is obtained, the IOPS load percentage is determined based on the ratio of the current actual IOPS value to the maximum IOPS value: IOPS load percentage = (current actual IOPS value / maximum IOPS value) × 100%. For example, if the current actual random read IOPS value is 60,000 times / second and the maximum random read IOPS value is 100,000 times / second, then the IOPS load percentage = (60,000 / 100,000) × 100% = 60%. Simultaneously, when the current actual data throughput is obtained, the data throughput load percentage is determined based on the ratio of the current actual data throughput to the maximum data throughput. Data throughput load percentage = (current actual data throughput / maximum data throughput) × 100%. For example, if the current actual sequential read throughput is 2100 MB / s and the maximum sequential read throughput is 3500 MB / s, then the data throughput load percentage = (2100 / 3500) × 100% = 60%.

[0059] Since IOPS and data throughput may have different impacts on SSD load, different weights can be assigned to them based on the actual situation, and then a weighted average method can be used to calculate the overall load percentage. SSD load percentage = (IOPS load percentage × IOPS weight + Data throughput load percentage × Data throughput weight), where IOPS weight + Data throughput weight = 1. Furthermore, the calculated SSD load is compared with a first preset load threshold (e.g., 80%). When the SSD load is determined to be greater than the preset load threshold, the SSD device can be identified as being in a high-performance read / write state, triggering the energy recovery mechanism. Specifically, the `apl_dynamic_budget()` function is called. This function dynamically adjusts the APL reserved energy threshold based on the load trend predicted by the target load prediction model for the next 30 minutes, i.e., the predicted load and the current operating data of the SSD device (such as chip temperature, capacitor charge, etc.). If the predicted load remains high, the APL reserved energy is reduced to release excess power. Next, by starting the `thermoelectric_conversion(temperature_delta)` function, the temperature sensor installed on the SSD chip monitors the chip temperature in real time and calculates the temperature difference (temperature_delta) between the chip temperature and the heatsink temperature. When the temperature difference reaches a predicted temperature threshold (e.g., 85°C), the thermoelectric conversion module, specifically the Bi2Te3-based thermoelectric conversion module, is activated. This module converts the temperature difference into electrical energy through the Seebeck effect, yielding thermoelectric energy. When the excess electrical energy (excess_power) and the thermoelectric energy (thermal_energy) are calculated, they can be combined to obtain the total recovered energy: total_recovered_energy = excess_power + thermal_energy.

[0060] Furthermore, the `store_energy(total_recovered_energy,target="slot_module")` function is called to transfer the total recovered energy to the energy storage device, namely the supercapacitor energy storage module integrated into the PCIe slot / backplane. Through an intelligent charging management circuit, the input electrical energy (an unstable DC signal) is rectified and regulated before being stored in the supercapacitor. When energy is successfully stored in the energy storage device, a feedback signal indicating successful storage can be output.

[0061] By comprehensively considering IOPS and data throughput and allocating weights to calculate the SSD load percentage through the above optional implementation methods, the SSD load status can be more accurately determined. When the load exceeds the threshold, the APL reserved energy is reduced to release excess electrical energy, achieving reasonable energy allocation. Utilizing the temperature difference between the chip and the heatsink, the thermoelectric conversion module converts heat energy into electrical energy, further expanding the energy recovery path. Excess electrical energy and thermoelectric conversion electrical energy are combined into total recovered energy and stored in the energy storage device, improving energy utilization. Effective energy recovery during high-performance SSD read / write operations optimizes energy efficiency and reduces energy waste. At the same time, the energy storage device provides electrical support when necessary, enhancing the stability and reliability of SSD operation.

[0062] In an optional implementation, the method further includes: When the SSD load is less than the second preset load threshold, the SSD device will be switched from its current operating state to a low-power mode. The operation of converting heat energy into electrical energy using the temperature difference through the thermoelectric conversion module is performed to obtain thermoelectric-converted electrical energy. The thermoelectric energy is transmitted to a rack-mounted supercapacitor array, and the thermoelectric energy is stored in the supercapacitors through the rack-mounted supercapacitor array.

[0063] In some embodiments, the calculated SSD load is also compared with a second preset threshold (e.g., 10%). If the SSD load is below 10%, the SSD device is determined to be in an idle state, triggering a deep recycling mechanism. The `activate_low_power_mode()` function is called to switch the SSD device to a low-power mode. Specifically, this involves reducing the operating frequency of the main controller chip, disabling unnecessary I / O interfaces, and reducing the refresh frequency of the DRAM cache. This reduces device power consumption while ensuring the chip generates stable heat, creating conditions for heat collection.

[0064] Next, the `thermoelectric_conversion(temperature_delta,enhance=True)` function is initiated to execute an enhanced heat harvesting strategy in the Idle state. A temperature sensor installed on the SSD chip monitors the chip temperature in real time, calculating the temperature difference (temperature_delta) between the chip and the heatsink. Simultaneously, additional cooling auxiliary devices (such as micro-fans and thermoelectric coolers for reverse heating) are activated to artificially increase the temperature difference to above 95°C, thereby improving the conversion efficiency of the Bi2Te3-based thermoelectric conversion module. The thermoelectric conversion module converts the temperature difference into electrical energy through the Seebeck effect. Further, the `store_energy(thermal_energy,target="rack_array")` function is called to transfer the electrical energy (thermal_energy) generated by the thermoelectric conversion to a rack-mount supercapacitor array. The rack-mount supercapacitor array, through intelligent charging management circuitry, rectifies and boosts the unstable DC power input, storing it stably in the supercapacitors for subsequent use in overall data center power distribution or to power other high-load equipment.

[0065] Through the above optional implementation methods, when the SSD load is low, it is switched to a low-power mode to reduce energy consumption. At the same time, an enhanced heat collection strategy is used to expand the temperature difference and improve the thermoelectric conversion efficiency. The thermoelectric converted electrical energy is stored in a rack-level supercapacitor array to realize energy recovery, which can be used for data center power allocation or to power other equipment, thus optimizing the overall energy efficiency.

[0066] In an optional implementation, the method further includes: Acquire real-time power data of the rack-mounted supercapacitor array; The real-time power data is compared with a first preset power threshold. When it is determined that the real-time power data is greater than the first preset power threshold, the power of the rack-level supercapacitor array is transmitted to the UPS system of the data center through the DC-DC conversion circuit; Acquire UPS power data and real-time power grid quality data for the UPS system; When the real-time power data meets the second preset power threshold, and / or the UPS power data meets the third preset power threshold, and / or when the real-time power grid power quality is detected to have returned to normal and stable, the UPS system that transmits power from the rack-level supercapacitor array to the data center through the DC-DC conversion circuit is stopped.

[0067] In some embodiments, the system can collect the state of charge (SOC) of the rack-level supercapacitor array in real time, i.e., real-time power data, through a power monitoring circuit, and compare the real-time power data with a first preset power threshold (e.g., 90%). The real-time power data collection frequency can be set to once per minute to ensure timely capture of power changes. When the real-time power data of the rack-level supercapacitor array exceeds 90%, the system further assesses the overall power status of the data center. This includes acquiring overall data center power status information, such as the UPS (Uninterruptible Power Supply) SOC, power grid quality data (e.g., voltage stability, frequency fluctuations), and data on the predicted power demand trend of the data center within a preset time period (e.g., 1 hour). By checking the UPS SOC, the system determines whether it needs additional power.

[0068] Furthermore, the system determines whether the UPS power level is below 80% based on the acquired UPS state of charge data, whether the power grid quality is unstable based on the power grid power quality data, and whether the predicted power demand is increasing based on the data center power demand trend forecast data. If all three conditions are met—that is, the UPS power level is below 80% or the power grid quality is unstable, and the predicted power demand is increasing—then the energy transfer process is triggered. Specifically, the `transfer_energy_to_ups()` function is called to transfer the power from the rack-level supercapacitor array to the data center's UPS system via a dedicated DC-DC conversion circuit. This involves stepping down the supercapacitor array's output voltage (typically 48V) to the UPS system's charging voltage (24V / 48V adaptive). A constant current-constant voltage (CC-CV) charging algorithm is used to control the current transferred to the UPS system. In the initial stage of charging, a constant current is used for rapid charging, and once the voltage reaches a certain threshold, it switches to constant voltage charging to ensure a safe and efficient charging process. During the energy transfer process, the rate of power loss of the supercapacitor array is continuously monitored, and the energy transfer parameters are dynamically adjusted based on the real-time monitoring results to avoid over-discharge. During energy transfer, real-time status information is acquired, including real-time power data of the rack-level supercapacitor array, UPS power data of the UPS system, and real-time power grid quality data. The system automatically terminates energy transfer when one of the following conditions occurs: 1) The real-time power data of the rack-level supercapacitor array drops to a second preset power threshold (e.g., 70%); 2) The UPS system is fully charged, i.e., the UPS power data reaches a third preset power threshold (e.g., 95%); 3) The power grid quality is detected to have returned to normal and stable. After energy transfer is terminated, the system can generate an energy transfer report, recording the transferred power, duration, and energy efficiency data, and update the status information of each energy storage device.

[0069] Through the aforementioned optional implementation methods, by acquiring rack-level supercapacitor array power data in real time and comparing it with preset thresholds, power can be transferred to the data center UPS system when power is sufficient, achieving rational energy allocation and improving energy utilization. During the energy transfer process, UPS power, grid power quality, and data center power demand trends are comprehensively considered to ensure a safe and efficient transfer process. Simultaneously, relevant data is continuously monitored, and energy transfer is automatically terminated when specific conditions are met to avoid over-discharge. Furthermore, an energy transfer report is generated to record key information, aiding in subsequent analysis and optimization, effectively improving the intelligence level of data center energy management, ensuring power supply stability, and reducing operating costs.

[0070] To facilitate understanding of the inventive concept of this application, the following is a code example of energy recovery in an SSD data center energy efficiency optimization system: def energy_management_system(): while True: # Mode 1: Energy recovery during high-performance read / write if ssd_load > 80%: excess_power = apl_dynamic_budget() # Dynamic APL reserve of energy thermal_energy = thermoelectric_conversion(temperature_delta) # Thermoelectric conversion store_energy(excess_power + thermal_energy, target="slot_module") # Mode 2: Deep recycling in Idle state elif ssd_load<10%: # Activate low-power mode to enhance heat harvesting activate_low_power_mode() thermal_energy = thermoelectric_conversion(temperature_delta, enhance=True) store_energy(thermal_energy, target="rack_array") # Intelligent scheduling of energy storage release if rack_array_charge>90%: transfer_energy_to_ups() In an optional implementation, the method further includes: Based on the preset load, the corresponding target energy storage strategy is matched in the preset energy storage strategy decision matrix; The energy storage parameters are dynamically adjusted according to the target energy storage strategy. The adjusted energy storage parameters are sent to all energy storage devices so that all energy storage devices can store energy according to the adjusted energy storage parameters.

[0071] In some embodiments, the energy storage strategy decision matrix can be preset, as shown in Table 1 below: Table 1:

[0072] When the target load prediction model outputs the predicted load, the predicted load can be input into a preset energy storage strategy decision matrix to match the corresponding energy storage strategy, i.e., the target energy storage strategy. When the target energy storage strategy is matched, the system can dynamically adjust the corresponding energy storage parameters, which may include the APL threshold, the charging and discharging rate of the energy storage device, and the energy transfer path. Regarding the APL threshold adjustment, during high load prediction, the APL reserved energy is reduced (e.g., from the default 15% to 10%) to release more energy for performance assurance; during low load prediction, the APL reserved energy is increased (e.g., to 20%) to enhance power outage protection capabilities. Regarding the charging and discharging rate control of the energy storage device, during high load prediction, the discharge rate of the PCIe slot-level energy storage module is increased (e.g., from 5A to 8A); during low load prediction, the charging rate is reduced (e.g., from 3A to 1.5A) to reduce energy loss. To optimize the power transfer path, if it is predicted that the rack-level array will reach full charge, the power transfer to the UPS system will be planned in advance; if it is predicted that the SSD devices will frequently switch operating modes, the coordination strategy between the built-in capacitors and the slot-level modules will be optimized.

[0073] When adjusting energy storage parameters based on the target energy storage strategy, the system always follows the preset energy recovery priority. For power outage protection, regardless of the forecast results, it ensures that the SSD's built-in capacitors retain at least 50% charge for power outage data protection; for performance optimization, under high load forecasts, priority is given to ensuring SSD read and write performance, allowing for a temporary reduction in system power recovery efficiency; for system power supply, energy is only recovered to the data center UPS when all device loads are below 10% and the energy storage devices have sufficient charge.

[0074] After adjusting the energy storage parameters based on the target storage strategy, the adjusted energy storage parameters can be sent to the management units of each level of energy storage equipment by issuing control commands; and the actual response data of the energy storage equipment (such as charging and discharging current and temperature changes) can be collected for real-time monitoring and feedback; at the same time, the weight parameters of the target load prediction model can be dynamically adjusted by comparing the actual load with the prediction results to correct prediction errors and improve the accuracy of subsequent predictions.

[0075] Furthermore, by using existing line multiplexing methods to transmit recovered electrical energy to the supercapacitor of the energy storage device via reserved pins (such as PRSNT1# / PRSNT2#) on the PCIe bus, no additional wiring is required; and 2.45GHz microwave energy transmission (efficiency ≈60%, transmission distance ≤5cm) can also be achieved in high-density racks.

[0076] Through the aforementioned optional implementation methods, intelligent energy storage management is achieved by pre-setting an energy storage strategy decision matrix, matching the target energy storage strategy based on predicted load, and dynamically adjusting energy storage parameters. For different load scenarios, the APL threshold, charge / discharge rate, and energy transfer path are flexibly adjusted to optimize energy utilization efficiency. Simultaneously, adhering to energy recovery priorities ensures power outage protection and performance optimization requirements. Real-time monitoring and feedback mechanisms dynamically correct prediction errors, improving prediction accuracy. Furthermore, utilizing existing line multiplexing and microwave energy transmission technologies reduces implementation costs and improves transmission efficiency.

[0077] In an optional implementation, the method further includes: Acquire PCM temperature data and SSD chip junction temperature of the SSD chip; RT44HC paraffin-based PCM is embedded in the SSD heatsink, and the PCM temperature data is the temperature data corresponding to the RT44HC paraffin-based PCM. The temperature stage of the SSD heatsink is determined based on the PCM temperature data and the junction temperature of the SSD chip. When it is determined that the SSD heatsink is at its peak temperature, the active cooling fan of the thermoelectric conversion module is turned off. When it is determined that the SSD heatsink is at its lowest temperature, the liquid cooling system is activated.

[0078] In some embodiments, RT44HC paraffin-based PCM (phase change material) can be embedded in the SSD heatsink. This absorbs heat during temperature peaks and releases it for thermoelectric conversion during troughs, thus smoothing temperature fluctuations. Specifically, a distributed temperature sensor array (accuracy ±0.5℃, sampling frequency 10Hz) is deployed in the RT44HC paraffin-based PCM area embedded in the SSD heatsink to collect the solid-liquid interface temperature of the PCM (i.e., phase change temperature 44℃±2℃) and the SSD chip junction temperature (Tj) in real time. Simultaneously, the input temperature (T1) and output power (Pout) of the thermoelectric conversion module are recorded, forming a multidimensional temperature dataset. The multidimensional temperature dataset includes PCM temperature data T_pcm(t), SSD chip junction temperature Tj(t), and ambient temperature data Ta(t). The PCM temperature data T_pcm(t) reflects the current state of the phase change material (solid, solid-liquid mixture, liquid). The chip junction temperature data Tj(t) characterizes the real-time heat generated by the SSD workload. The temperature gradient ΔT=Tj-Ta between the chip junction temperature data and the ambient temperature data can be calculated.

[0079] When the following multiple conditions are met in a combined triggering process, the temperature peak stage can be determined: I. Absolute temperature threshold triggering: (1) The chip junction temperature Tj is greater than the second preset temperature threshold (e.g., 55℃), which is more than the upper limit temperature of the liquid phase of RT44HC paraffin-based PCM, which is 50℃, with a safety margin of 5℃. (2) When the PCM temperature T_pcm reaches the third preset temperature threshold (e.g., 46℃), which is 2℃ above the phase change point, it is confirmed that the liquid endothermic stage has been entered.

[0080] II. Triggered by Temperature Change Rate: (3) dTj / dt > the first preset temperature change rate, for example 2℃ / s (the chip junction temperature rise rate exceeds the normal operating range). (4) dT_pcm / dt> the second preset temperature change rate, for example, 1℃ / s (the PCM temperature rise rate indicates that heat is being absorbed rapidly).

[0081] If the duration of the above conditions (1), (2), (3), and (4) exceeds the first preset time period (e.g., 30 seconds), transient pulse interference is eliminated, a "temperature peak" signal is generated, the PCM full-power heat absorption mode is triggered, and the thermoelectric conversion module is adjusted to "low-priority energy recovery" (prioritizing the heat absorption efficiency of the PCM).

[0082] When it is determined that the SSD heatsink has entered the peak temperature stage, the active cooling fan of the thermoelectric conversion module is turned off to reduce interference with the heat absorption path of the PCM; the power allocation strategy of the SSD firmware is adjusted to temporarily reduce the power consumption of non-critical components (such as the cache controller) to assist in cooling the SSD chip.

[0083] When the following multiple conditions are met in a combined triggering configuration, it can be determined that the temperature has entered a low-temperature phase: I. Absolute temperature threshold triggering: (5) The junction temperature Tj of the SSD chip is less than the fourth preset temperature threshold (e.g., 35°C), which is lower than the solid-state lower limit temperature of RT44HC paraffin-based PCM of 38°C, with a safety margin of 3°C. (6) PCM temperature T_pcm < fifth preset temperature threshold (e.g. 40℃), that is, 4℃ below the phase change point, confirm that it has entered the solid-state exothermic stage.

[0084] II. Triggered by Temperature Change Rate: (7) dTj / dt < third preset temperature change rate, for example -1℃ / s (the rate of decrease in chip junction temperature indicates a decrease in load); (8) dT_pcm / dt < fourth preset temperature change rate, for example -0.5℃ / s (PCM temperature drop rate indicates the start of releasing stored heat).

[0085] If the above conditions (5), (6), (7), and (8) last for more than the second preset time period (e.g., 60 seconds), it is confirmed that a stable low-load state has been entered, a "temperature trough" signal is generated, the PCM heat release process is activated, the stored heat energy is directed to the thermoelectric conversion module through the heat-conducting layer, and the thermoelectric conversion efficiency is increased to the highest level (e.g., Seebeck coefficient is increased by 15%).

[0086] When the SSD heatsink is determined to be in a low temperature phase, the micro-pump liquid cooling system (if applicable) in the thermal conductive layer is activated to accelerate heat conduction between the PCM and the thermoelectric conversion module; the heat release rate is dynamically adjusted according to the current solid phase ratio of the PCM (monitored by an impedance sensor) to avoid a sudden temperature rise.

[0087] The system can also perform temperature sensor calibration every 24 hours, combine historical data to statistically analyze the actual phase change range of the PCM (due to thermal aging that may occur with long-term use), and automatically correct threshold offsets (e.g., compensate 0.3℃ per year).

[0088] By acquiring PCM temperature data and SSD chip junction temperature through the above optional implementation methods, the temperature stage of the SSD heatsink can be accurately determined. When the temperature is at its peak, the thermoelectric conversion module fan is turned off to reduce interference, and the power consumption distribution is adjusted to assist in cooling. When the temperature is at its low point, the liquid cooling system is activated to accelerate the release of heat from the PCM, improve the thermoelectric conversion efficiency, effectively smooth out SSD temperature fluctuations, improve heat dissipation efficiency and energy utilization efficiency, ensure stable operation of the SSD, and extend the service life of the device.

[0089] In an optional implementation, the method further includes: Based on the chip temperature and the preset temperature threshold segmentation range, the fan speed and liquid cooling flow rate are adjusted according to the preset load. The preset temperature threshold segmentation range includes multiple temperature threshold ranges, and each temperature threshold range corresponds to an adjustment method.

[0090] In some embodiments, the system can dynamically adjust the fan speed and liquid cooling flow rate to maintain the optimal temperature gradient while ensuring that the SSD chip temperature is <70℃. The system collects the chip temperature in real time using a high-precision temperature sensor deployed on the SSD chip surface, obtains the liquid cooling flow rate and pipe pressure using flow and pressure sensors on the liquid cooling pipes, collects the real-time fan speed using a fan speed encoder, and simultaneously collects environmental data such as data center ambient temperature, humidity, and current data center load rate as raw data. The collection frequency can be set to once per second. The raw data is then normalized, mapping data of different dimensions to the [0, 1] interval. A moving average filtering method is used to remove data noise, with the sliding window size set to 10 sampling points. A time-series dataset containing the previous 10 minutes of historical data is constructed as the input feature of the AI ​​temperature control algorithm model.

[0091] An AI temperature control algorithm model is constructed by fusing deep learning's Long Short-Term Memory (LSTM) network with reinforcement learning (Q-Learning). The LSTM prediction model uses a time-series dataset to predict the chip temperature change trend (Tpred) over the next 5 minutes, with the prediction error controlled within ±2℃. Similarly, the Q-Learning decision module is used to divide the chip temperature into 5 intervals: interval S1: Tchip < 50℃, interval S2: 50℃ ≤ Tchip < 60℃, interval S3: 60℃ ≤ Tchip < 65℃, interval S4: 65℃ ≤ Tchip < 70℃, and interval S5: Tchip ≥ 70℃. Combined with the liquid cooling flow rate and fan speed, a multi-dimensional state space is formed. The fan speed is set to 5 levels: level 1 R1, level 2 R2, level 3 R3, level 4 R4, and level 5 R5, with the speed increasing sequentially. The liquid cooling flow rate is also set to 5 levels: level 1 V1, level 2 V2, level 3 V3, level 4 V4, and level 5 V5, with the flow rate increasing sequentially, forming an action combination space.

[0092] In addition, the system can construct a reward function. If the chip temperature is controlled at Tchip < 70℃ and the energy consumption is reduced, a positive reward (R+) is given; if the chip temperature exceeds 70℃ or the energy consumption increases significantly, a negative reward (R−) is given. At the same time, the corresponding reward value is calculated. The reward value calculation formula is R = α × (70 − Tchip) − β × (Ecurrent − Epreviious), where α and β are weighting coefficients, and Ecurrent and Epreviious are the cooling system energy consumption at the current time and the previous time, respectively.

[0093] Furthermore, the fan speed and liquid cooling flow rate are adjusted based on the chip temperature. Specifically, when Tchip < 50℃ (state S1), if the LSTM model predicts no significant temperature increase in the next 5 minutes, the fan speed is adjusted to the lowest setting R1, the liquid cooling flow rate is adjusted to the lowest setting V1, and the fan intelligent sleep mechanism is activated, performing a 30-second heat dissipation check every 10 minutes. When 50℃ ≤ Tchip < 60℃ (state S2), based on the optimal action output by the Q-Learning decision module, if increasing the fan speed is recommended, the fan speed is adjusted to R2; if increasing the liquid cooling flow rate is recommended, the liquid cooling flow rate is adjusted to V2. After maintaining the current state for 1 minute, the chip temperature is reassessed. When 60℃ ≤ Tchip < 65℃ (state S3), the liquid cooling flow rate adjustment strategy is prioritized, increasing the liquid cooling flow rate to V3. If the chip temperature still does not decrease after 1 minute, the fan speed is simultaneously increased to R3, and the auxiliary heat dissipation function of the thermoelectric conversion module is activated to convert some heat into electrical energy. When Tchip temperature is between 65℃ and 70℃ (state S4), the fan speed and liquid cooling flow rate are simultaneously set to their highest levels (R4 and V4) to trigger the phase change material (PCM) rapid heat dissipation mode, accelerating heat absorption. If the chip temperature does not drop below 65℃ within 3 minutes, a temperature warning is issued. When Tchip temperature is ≥ 70℃ (state S5), the fan speed and liquid cooling flow rate are immediately set to their extreme levels (R5 and V5), non-critical SSD devices are suspended to reduce the overall load, and the backup cooling system is activated to ensure the chip temperature drops rapidly.

[0094] After each adjustment, the actual chip temperature change, energy consumption change, and adjusted system state are recorded. The actual data is compared with the predicted data, the error is calculated, and the LSTM model parameters are updated. The reward value is calculated according to the reward function, and the Q-table of the Q-Learning decision module is updated to optimize subsequent control strategies. Simultaneously, a full model training is performed weekly, utilizing the latest historical data to improve the algorithm's accuracy and adaptability.

[0095] Through the aforementioned optional implementation methods, multi-dimensional data such as chip temperature are collected in real time. An AI temperature control algorithm model integrating deep learning and reinforcement learning is used to accurately predict chip temperature trends and classify temperature ranges. Based on different temperature ranges, fan speed and liquid cooling flow rate are dynamically adjusted to form a refined temperature control strategy. This effectively ensures that the SSD chip temperature remains within a safe range, maintains the optimal temperature gradient, reduces energy consumption, and improves heat dissipation efficiency. Simultaneously, an early warning mechanism enhances system stability and extends device lifespan.

[0096] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0097] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0098] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for optimizing energy efficiency in an SSD data center, characterized in that, The method includes: When an SSD device's execution command is received, the execution data of the SSD device is obtained; The operating data is input into a preset target load prediction model so that the target load prediction model outputs the predicted load of the SSD device for a future period of time. When it is determined that the predicted load is greater than the upper limit of the preset load threshold range, the APL reserved energy threshold is reduced, and the excess reserved energy is converted by the circuit and transferred to the PCIe slot-level energy storage module for storage. When it is determined that the preset load is less than the lower limit of the preset load threshold range, the APL reserved energy threshold is increased, and the remaining electrical energy is recovered to the SSD built-in capacitor.

2. The energy efficiency optimization method for SSD data centers according to claim 1, characterized in that, The operational data includes the SSD chip temperature, and the method further includes: Obtain basic APL reservations; A risk assessment is performed on the SSD built-in capacitors of the SSD device based on the chip temperature and the predicted load, and the risk assessment result is obtained. The basic adjustment coefficient is determined based on the risk assessment results. The final adjustment factor is calculated based on the basic adjustment coefficient, temperature risk coefficient, load fluctuation coefficient, and health coefficient; the temperature risk coefficient is determined based on the chip temperature, the load fluctuation coefficient is determined based on the predicted load, and the health coefficient is determined based on the capacitor health status of the SSD's built-in capacitors. Calculate the final APL reserve based on the basic APL reserve and the final adjustment factor; Obtain the current total energy and calculate the recoverable excess energy based on the current total energy and the final APL reserve; The excess energy is stored.

3. The energy efficiency optimization method for SSD data centers according to claim 2, characterized in that, The method also includes When it is determined that the chip temperature is greater than a first preset temperature threshold, the heat of the SSD chip is obtained; The heat from the chip is conducted to the thermoelectric conversion module through the graphene thermal conductive layer, so that the thermoelectric conversion module converts the heat from the chip into electrical energy output. The electrical energy is processed by a DC-DC converter so that the DC-DC converter outputs a stable DC power. The stable DC power is stored in the supercapacitor of the energy storage device.

4. The energy efficiency optimization method for SSD data centers according to claim 3, characterized in that, The operational data also includes SSD load, and the method further includes: When the SSD load exceeds a first preset load threshold, the APL reserved energy is reduced to release excess electrical energy. When the temperature difference is greater than the first preset temperature threshold, the thermoelectric conversion module uses the temperature difference to convert heat energy into electrical energy to obtain thermoelectric electrical energy; the temperature difference is determined by the chip temperature and the heat sink temperature of the SSD heat sink. The total recovered energy is determined based on the excess electrical energy and the thermoelectric conversion electrical energy. The total recovered energy is transferred to the energy storage device for storage.

5. The energy efficiency optimization method for SSD data centers according to claim 4, characterized in that, The method further includes: When the SSD load is less than the second preset load threshold, the SSD device will be switched from its current operating state to a low-power mode. The operation of converting heat energy into electrical energy using the temperature difference through the thermoelectric conversion module is performed to obtain thermoelectric-converted electrical energy. The thermoelectric energy is transmitted to a rack-mounted supercapacitor array, and the thermoelectric energy is stored in the supercapacitors through the rack-mounted supercapacitor array.

6. The energy efficiency optimization method for SSD data centers according to claim 5, characterized in that, The method further includes: Acquire real-time power data of the rack-mounted supercapacitor array; The real-time power data is compared with a first preset power threshold. When it is determined that the real-time power data is greater than the first preset power threshold, the power of the rack-level supercapacitor array is transmitted to the UPS system of the data center through the DC-DC conversion circuit; Acquire UPS power data and real-time power grid quality data for the UPS system; When the real-time power data meets the second preset power threshold, and / or the UPS power data meets the third preset power threshold, and / or when the real-time power grid power quality is detected to have returned to normal and stable, the UPS system that transmits power from the rack-level supercapacitor array to the data center through the DC-DC conversion circuit is stopped.

7. The energy efficiency optimization method for SSD data centers according to claim 5, characterized in that, The method further includes: Based on the preset load, the corresponding target energy storage strategy is matched in the preset energy storage strategy decision matrix; The energy storage parameters are dynamically adjusted according to the target energy storage strategy. The adjusted energy storage parameters are sent to all energy storage devices so that all energy storage devices can store energy according to the adjusted energy storage parameters.

8. The energy efficiency optimization method for SSD data centers according to claim 5, characterized in that, The method further includes: Acquire PCM temperature data and SSD chip junction temperature of the SSD chip; RT44HC paraffin-based PCM is embedded in the SSD heatsink, and the PCM temperature data is the temperature data corresponding to the RT44HC paraffin-based PCM. The temperature stage of the SSD heatsink is determined based on the PCM temperature data and the junction temperature of the SSD chip. When it is determined that the SSD heatsink is at its peak temperature, the active cooling fan of the thermoelectric conversion module is turned off. When it is determined that the SSD heatsink is at its lowest temperature, the liquid cooling system is activated.

9. The energy efficiency optimization method for SSD data centers according to any one of claims 2 to 8, characterized in that, The method further includes: Based on the chip temperature and the preset temperature threshold segmentation range, the fan speed and liquid cooling flow rate are adjusted according to the preset load. The preset temperature threshold segmentation range includes multiple temperature threshold ranges, and each temperature threshold range corresponds to an adjustment method.

10. An energy efficiency optimization system for SSD data centers, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the method described in any one of claims 1-9.

Images