Power supply method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of computers, and provides a power supply method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining the current load data of electric equipment in real time in a current period; on the basis of the current load data, predicting target power of the electric equipment in the next period by adopting a mode of combining a load prediction model and statistical analysis; adjusting power output of a power supply system to the electric equipment based on the target power of the electric equipment; wherein the load prediction model is obtained through training by taking historical load data of a first period in historical data of the electric equipment as a sample and taking historical detection power of a second period in the historical data as a label based on a machine learning model, and the second period is the next period of the first period. According to the invention, the target power of the electric equipment can be accurately and dynamically predicted according to the current load data, so that the power supply output of the power supply system is dynamically adjusted.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a power supply method, device, electronic equipment and storage medium. Background Art

[0002] At present, the common server power supply systems on the market mainly use static power distribution to power the server. Specifically, the power supply system includes multiple power modules, each of which uses a fixed output power to power the corresponding server. The fixed output power is usually determined according to the maximum power consumption design of the server. Although this static power distribution method is simple and easy to implement, it will cause many problems in the complex and changeable environment of modern data centers. For example, the real-time load of the server varies greatly with different working hours, and the maximum power consumption will only be reached during the peak load period. If the server is always powered by the maximum power consumption designed by the server, it will lead to energy waste and low energy utilization. In addition, the server will be under great power pressure for a long time, which will accelerate the aging and wear of the server.

[0003] Therefore, how to dynamically adjust the power output of the corresponding power module according to the real-time load of the server (or other power-consuming equipment) is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The present invention provides a power supply method, device, electronic device and storage medium, which are used to solve the problem in the prior art that the power output of the corresponding power module cannot be dynamically adjusted according to the real-time load of the server (or other power-consuming equipment).

[0005] The invention provides a power supply method, which comprises the following steps.

[0006] During the current cycle, the current load data of the electrical equipment is obtained in real time.

[0007] Based on the current load data, a load prediction model and statistical analysis are combined to predict the target power of the electrical equipment in the next cycle.

[0008] Based on the target power of the electric device, the power output of the power supply system to the electric device is adjusted.

[0009] Among them, the load prediction model is a model based on machine learning, which takes the historical load data of the first cycle in the historical data of the electrical equipment as a sample, the historical detection power of the second cycle in the historical data is obtained by label training, and the second cycle is the next cycle of the first cycle.

[0010] According to a power supply method provided by the present invention, based on the current load data, a load prediction model and a combination of statistical analysis are used to predict the target power of the electrical equipment in the next cycle, including the following steps.

[0011] The current load data is input into the load prediction model to obtain the first power of the electrical equipment in the next cycle output by the load prediction model.

[0012] The current load data is statistically analyzed to predict the predicted load data of the electrical equipment in the next cycle, and the second power is determined based on the predicted load data.

[0013] Based on the first power and the second power, a target power of the electric device in a next cycle is determined.

[0014] According to a power supply method provided by the present invention, based on the first power and the second power, the target power of the power-consuming equipment in the next cycle is determined, including: obtaining the target power by weighted summing the first power, the second power and their respective weights, and the sum of the weights of the first power and the second power is 1.

[0015] According to a power supply method provided by the present invention, when the next cycle belongs to a time period in which the load of the electrical equipment changes periodically, the weight of the second power is adjusted to be greater than the weight of the first power; otherwise, the weight of the first power is adjusted to be greater than the weight of the second power.

[0016] According to a power supply method provided by the present invention, based on the first power and the second power, the target power of the power-consuming equipment in the next cycle is determined, including: inputting the first power and the second power into an integrated learning model, and determining the predicted power output by the integrated learning model as the target power.

[0017] The integrated learning model is trained using the historical first power corresponding to the first power and the historical second power corresponding to the second power as samples, and the historical detection power within the historical prediction period where the historical first power and the historical second power are located as labels.

[0018] According to a power supply method provided by the present invention, based on the target power of the power consuming device, the power output of the power supply system to the power consuming device is adjusted, and the method includes the following steps.

[0019] The target power of each power supply device is obtained, and based on each target power and the redundancy requirement corresponding to each power supply device, the required power of each power supply device is calculated.

[0020] Based on the power requirements of each power supply device, the total power requirements are calculated.

[0021] The maximum output power of each power module in the current working state of the power system is obtained, and the maximum total output power is calculated based on each maximum output power.

[0022] When the required total power is greater than the maximum output total power, at least one backup power module is controlled to enter a working state.

[0023] According to the power requirements of each electrical device, the distribution of the output power of the power modules in the working state to each electrical device is adjusted, and the constraint condition that one electrical device obtains electrical energy from at least two power modules and the real-time total output power of each power module does not exceed the corresponding maximum output power is met.

[0024] A power supply method provided according to the present invention further includes: when at least one of the power modules in a working state fails, adjusting the power output of the power supply system to the power consuming device based on the most recently predicted target power of each power consuming device.

[0025] The present invention also provides a power supply device, comprising the following functional modules.

[0026] The current load acquisition module is used to obtain the current load data of the electrical equipment in real time during the current cycle.

[0027] The target power prediction module is used to predict the target power of the electrical equipment in the next cycle based on the current load data by combining a load prediction model with statistical analysis.

[0028] The power output adjustment module is used to adjust the power output of the power supply system to the power-consuming device based on the target power of the power-consuming device.

[0029] Among them, the load prediction model is a model based on machine learning, which takes the historical load data of the first cycle in the historical data of the electrical equipment as a sample, the historical detection power of the second cycle in the historical data is obtained by label training, and the second cycle is the next cycle of the first cycle.

[0030] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above-described power supply methods when executing the computer program.

[0031] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the power supply method as described above is implemented.

[0032] The power supply method, device, electronic device and storage medium provided by the present invention predict the target power of the power-consuming equipment in the next cycle based on the current load data in the current cycle in combination with the load prediction model and the statistical analysis method, that is, it combines the advantages of the prediction ability of the machine learning model and the trend capture ability of the statistical analysis method to ensure that in a complex and changeable load environment, the target power of the power-consuming equipment can be accurately and dynamically predicted according to the current load data, thereby dynamically adjusting the power supply output of the power supply system, avoiding energy waste, improving low energy utilization, and avoiding the problem that the power-consuming equipment is subjected to large power pressure for a long time, which will accelerate the aging and wear of the server. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 This is one of the flow charts of the power supply method provided by the present invention.

[0035] Figure 2 This is the second flow chart of the power supply method provided by the present invention.

[0036] Figure 3 This is the third flow chart of the power supply method provided by the present invention.

[0037] Figure 4 It is a structural schematic diagram of the power supply device provided by the present invention.

[0038] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] In view of the above problems existing in the related art, an embodiment of the present invention provides a power supply method, such as Figure 1 As shown, the method includes the following steps S110 to S130.

[0041] Step S110: In the current cycle, the current load data of the electrical equipment is obtained in real time. Taking the scenario of a data center, where the power supply system supplies power to the servers in the data center as an example, for each server, the current load data includes: CPU utilization, memory occupancy, number of I / O requests, and network bandwidth utilization, etc. The server has a corresponding dedicated monitoring program for monitoring data such as CPU utilization, memory occupancy, number of I / O requests, and network bandwidth utilization. In this step, the dedicated monitoring program is accessed according to a preset cycle to periodically collect the current load data. The collection period can be 1 minute to 60 minutes. For example, the collection period is 10 minutes, and the current load data is collected every 10 minutes. In the subsequent steps, the target power of the server within the next 10 minutes is predicted.

[0042] It should be noted that: since the collection cycle is a time period, the current load data in the current cycle can be the average value of the load data monitored by the dedicated monitoring program in the corresponding time period. Of course, the current load data can also be the load data collected at the last moment of the current cycle.

[0043] Step S120: Based on the current load data, a load prediction model and statistical analysis are combined to predict the target power of the electrical equipment in the next cycle.

[0044] Among them, the load prediction model is a model based on machine learning, which is obtained by training with the historical load data of the first cycle in the historical data of the electrical equipment as a sample and the historical detection power of the second cycle in the historical data as a label, and the second cycle is the next cycle of the first cycle. That is, the load prediction model is trained with the historical load data of the previous cycle as a sample and the historical detection power of the next cycle as a label. It should be noted that: in step S110, it also includes obtaining the real-time power of the electrical equipment in the current cycle, which can be collected by high-precision sensors or monitoring equipment installed on the electrical equipment. As time goes by, the real-time power in the current cycle will become historical data, that is, the historical detection power of the second cycle in the historical data. Since the historical detection power is collected by high-precision sensors or monitoring equipment, it has high accuracy. This is used as a label of the load prediction model, so that the trained load prediction model can predict more accurately when applied.

[0045] Among them, statistical analysis can summarize the changing rules based on the changing trends of the historical load data of the power supply equipment, predict the predicted load data of the power equipment in the next cycle based on the current load data and the changing rules, and then obtain the power of the power equipment in the next cycle based on the relationship between the load data and power in the next cycle.

[0046] Step S130: Based on the target power of the power-consuming device, adjust the power output of the power system to the power-consuming device. Specifically, when the target power is reduced, the power output of the power system to the power-consuming device is reduced to prevent power waste and improve energy efficiency; conversely, the power output of the power system to the power-consuming device is increased to avoid voltage instability or system performance degradation caused by increased load.

[0047] The power supply method of the embodiment of the present invention predicts the target power of the power-consuming equipment in the next cycle based on the current load data in the current cycle, combined with the load prediction model and the statistical analysis method, that is, it combines the advantages of the prediction ability of the machine learning model and the trend capture ability of the statistical analysis method, ensuring that in a complex and changeable load environment, the target power of the power-consuming equipment can be accurately and dynamically predicted according to the current load data, thereby dynamically adjusting the power supply output of the power supply system, avoiding energy waste, improving low energy utilization, and avoiding the problem that the power-consuming equipment is subjected to large power pressure for a long time, which will accelerate the aging and wear of the server.

[0048] In some embodiments, Figure 2 As shown, step S120 specifically includes steps S210 to S230.

[0049] Step S210: input the current load data into the load prediction model to obtain the first power of the electrical equipment in the next cycle output by the load prediction model.

[0050] Specifically, the load prediction model can be a mature machine learning model such as a long short-term memory network (LSTM) or a random forest regression model. In this embodiment, for example, the power of the server in the next cycle is predicted using the LSTM model. When the LSTM model is trained, the historical load data of the first cycle of the historical data of the server collected periodically is used as a sample, that is, the input sample is ( x 1 , x 2 ,…, x m ),in, x i Represents a data item in historical load data (such as CPU utilization). i =1,2,…, m , based on j Prediction results for input samples y j To predict the first power of the second cycle, y j and the historical detection power of the second cycle as the tag Substitute the following mean squared error loss function (MSE), and the model training is completed when the loss function converges, where: n is the number of samples.

[0051] .

[0052] In this step, the current load data is input into the above-mentioned trained load prediction model to obtain the first power of the electrical equipment in the next cycle.

[0053] In this embodiment, in order to ensure the accuracy of the load prediction model, the load prediction model is trained once every predetermined time period using the latest historical data. Since the server load change trend may be different every long time period due to changes in server business, the power predicted by the retrained load prediction model is more in line with the business scenario of the current time period and the prediction is more accurate.

[0054] Step S220: Statistically analyzing the current load data to predict the predicted load data of the electrical equipment in the next cycle, and determining the second power based on the predicted load data.

[0055] Still taking the server power supply as an example, the current load data collection cycle is 10 minutes. The simplest statistical analysis method is to analyze the load data change trend of the same period (for example: 18:00-18:10 in the afternoon) in the previous few days (3-5 days). For example: for a certain game server, the CPU utilization rate in the 18:00-18:10 period a few days ago was about 15% higher than the CPU utilization rate in the 17:50-18:00 period. Then, today, the CPU utilization rate in the 17:50-18:00 period is increased by 15% as the prediction result of the CPU utilization rate in the 18:00-18:10 period. The CPU utilization rate corresponds to a sub-power, that is, the power consumed by the CPU. Other load data items correspond to corresponding sub-powers. The sum of the sub-powers corresponding to the prediction results of all load data items is the second power in the 18:00-18:10 period today. It should be noted that each load data item is directly proportional to the corresponding sub-power. The larger the load data value, the larger the sub-power. The proportional relationship can be fitted through the historical load data items and the corresponding sub-power (the sub-power can be obtained through the corresponding monitoring tools, for example, the sub-power of the CPU can be obtained through the monitoring tools of the CPU).

[0056] ARIMA statistical model can also be used for statistical analysis. ARIMA statistical model uses historical load data as samples, analyzes the time series characteristics in the samples, such as periodic changes in load, long-term trends and other characteristics, so as to predict the future trend of load data. During the analysis process, the ARIMA model automatically adjusts the autoregressive and moving average parameters. The ARIMA model can continuously optimize its prediction ability, and is particularly suitable for the prediction of long-term trends. In this embodiment, the predicted load data of the electrical equipment in the next cycle can be predicted by combining the current load data with the trend predicted by the ARIMA model, and then the second power of the electrical equipment in the next cycle is obtained by summing each data item in the predicted load data with the corresponding sub-power.

[0057] Step S230: Based on the first power and the second power, determine the target power of the electrical device in the next cycle, for example, the target power is the average value of the first power and the second power.

[0058] In some embodiments, step S230 specifically includes: obtaining the target power by weighted summing the first power, the second power and their respective weights, where the sum of the weights of the first power and the second power is 1, and the specific weighted summation formula is as follows: PFinal=α·PML+β·PStat.

[0059] PFinal is the target power, PML is the first power, PStat is the second power, α is the weight of the first power, β is the weight of the second power, and α+β=1. The weights α and β can be dynamically adjusted according to the prediction performance of machine learning and statistical analysis methods in historical data, so that the predicted target power can be more accurate.

[0060] Furthermore, if the next cycle belongs to a time period when the load of the electrical equipment changes periodically (for example, the time period after get off work on weekdays), it means that the statistical analysis prediction result will be better than the model prediction result, and the weight of the second power is adjusted to be greater than the weight of the first power, for example: β=0.7, α=0.3; otherwise, the weight of the first power is adjusted to be greater than the weight of the second power, that is, the machine learning model performs better than the statistical analysis in the time period of complex load changes, and the weight of the first power is adjusted to be greater than the weight of the second power. Specifically, the time period when the load changes periodically can be set according to past experience, and other time periods are time periods when the load changes are complex or random.

[0061] In this embodiment, the weights α and β are adjusted according to the time period of whether the load changes periodically, so that the target power can be accurately predicted in different load change scenarios.

[0062] In some embodiments, the step S230 specifically includes: inputting the first power and the second power into an integrated learning model, and determining the predicted power output by the integrated learning model as the target power. The integrated learning model can be a model based on voting method or weighted majority method.

[0063] The ensemble learning model is trained using the historical first power corresponding to the first power and the historical second power corresponding to the second power as samples, and the historical detection power in the historical prediction period where the historical first power and the historical second power are located as labels. The training method varies according to different ensemble learning models. For example, the training process of the ensemble learning model based on the voting method is a process of voting according to the difference between the historical first power and the historical second power and the historical detection power.

[0064] In some embodiments, Figure 3 As shown, the step S130 specifically includes the following steps S310 to S360.

[0065] Step S310: Obtain the target power of each power supply device, and calculate the required power of each power supply device based on each target power and the corresponding redundancy requirement of each power supply device. Specifically, for some power supply devices with frequent load changes, in order to ensure that they can continue to operate stably when the load suddenly increases, a certain amount of redundancy requirement is added when allocating power, for example: 20% power redundancy. Of course, for some power supply devices with relatively stable loads, the redundancy requirement can also be 0.

[0066] Step S320: based on the power requirements of each power supply device, the total power requirements are calculated, that is, the power requirements of each power supply device are summed up to obtain the total power requirements.

[0067] Step S330: obtaining the maximum output power of each power module in the power system in the current working state, and calculating the maximum total output power based on each maximum output power.

[0068] Step S340: Determine whether the required total power is greater than the maximum output total power, if so, execute step S350, otherwise execute step S360.

[0069] Step S350: Control at least one backup power module to enter a working state. Specifically, send an instruction to add a power module to the power system, and the redundant power management unit on the power system side starts at least one power module to enter a working state. Each power module has an independent control loop and protection mechanism, and can independently perform state control.

[0070] Step S360: According to the power requirements of each power-consuming device, the power modules in the working state are adjusted to distribute the output power of each power-consuming device, and the constraints that one power-consuming device obtains power from at least two power modules and the real-time total output power of each power module does not exceed the corresponding maximum output power are met. Specifically, the power allocated to the power-consuming device is not less than the required power. In order to avoid power-consuming device downtime caused by power module failure, it is necessary to ensure that the power-consuming device obtains power from at least two power modules. Even if one power module fails, the power-consuming device can be guaranteed to operate in a low-power state.

[0071] In this embodiment, based on the target power of the electrical equipment (the required power when there is a redundancy requirement), the power output of the power supply system to the electrical equipment is adjusted in real time, which not only avoids energy waste and improves energy utilization, but also takes into account the power redundancy requirement, so that the electrical equipment with frequent load fluctuations can continue to operate stably when the load suddenly increases.

[0072] In some embodiments, for any electrical device, its redundancy requirement is determined based on the power fluctuation size of the any electrical device in the same historical cycle period, that is, the redundancy requirement of any electrical device is different in different cycle periods, which further improves the accuracy of the output power distribution of the power module and avoids waste of electricity.

[0073] In some embodiments, the power supply method further includes: in the event that at least one of the power modules in the working state fails, based on the most recently predicted target power of each power device, adjusting the power output of the power supply system to the power device. Since the power module fails and cannot continue to supply power, it is necessary to reallocate the output power of the power module that is not failed to the power device, thereby implementing emergency fault handling to ensure the normal operation of each power device. Specifically, readjustment can be performed according to the above steps S310 to S360. Of course, the failed power module is used as a backup power module after the fault is eliminated.

[0074] In some embodiments, the power supply method further includes: real-time monitoring of the real-time power of the power-consuming device, where the real-time power may increase unexpectedly, and when the real-time power is greater than the power demand of the power-consuming device, and the current total power demand is not greater than the maximum total output power, more power is allocated to the power-consuming device from the power module currently in the working state, so that the power obtained by the power-consuming device meets the current real-time power. When the real-time power is greater than the power demand of the power-consuming device, and the current total power demand is greater than the maximum total output power, at least one backup power module is controlled to enter the working state, and more power is allocated to the power-consuming device from the backup power module, so that the power obtained by the power-consuming device meets the current real-time power.

[0075] Similarly, the real-time power may decrease unexpectedly. When the real-time power is less than the required power of the electrical device and the difference between the two reaches a preset percentage of the required power (for example, 20%), the output of the power module supplying power to the electrical device is reduced.

[0076] In this embodiment, after allocating corresponding power to the electrical equipment according to the required power, the real-time power of the electrical equipment is monitored in real time, and the unexpected increase or decrease in the real-time power of the electrical equipment is responded to according to the change in the real-time power. When the real-time power of the electrical equipment increases unexpectedly, the normal operation of the electrical equipment is guaranteed. When the real-time power of the electrical equipment decreases unexpectedly, the power output is reduced in time to further avoid waste of electricity.

[0077] In some embodiments, the power supply method also includes: real-time monitoring of abnormal conditions of electrical equipment (for example, abnormal current or excessive temperature), and when an abnormal condition is detected, reducing the output power of the power supply system to the electrical equipment or cutting off the power output to protect the electrical equipment and the power supply system.

[0078] In some embodiments, the historical data of the output power distribution of the power supply system and the historical load data and historical power data of the power-consuming equipment are sent to the central management system through the communication interface for storage. The current output power distribution results of the power supply system and the real-time power of the currently monitored power-consuming equipment are also sent to the central management system through the communication interface for the administrator to monitor and manage the power supply system and user equipment.

[0079] The following uses the predicted target power of four servers (A, B, C and D) and the power supply of three power modules in the power system as an example to illustrate how to adjust the power output of the power system to the power-consuming devices based on the target power of the power-consuming devices.

[0080] Through prediction, the target power of each server in the next hour is obtained as follows.

[0081] Server A: Target power is 800W (expected to increase to 960W during peak hours).

[0082] Server B: Target power is 500W (relatively stable).

[0083] Server C: The target power is 700W (subject to large fluctuations and may drop to 600W in the future).

[0084] Server D: Target power is 300W (redundant and backup node).

[0085] Among them, servers A and C fluctuate, so the redundancy requirements of servers A and C are set to 20%, then the required power of server A is 960W, the required power of server C is 714W, and the required power of servers B and D remains unchanged, which is the target power. The total required power is 960+714+500+300=2474W.

[0086] The three power modules in the power system are in working state, and their respective maximum output powers are as follows.

[0087] Power module 1: Maximum output power 1000 W.

[0088] Power module 2: Maximum output power 800 W.

[0089] Power module 3: Maximum output power 800 W.

[0090] The maximum total output power is 1000+800+800=2600W.

[0091] During the power allocation optimization process, it is ensured that each server obtains power from at least two power modules, and the output power of each power module must not exceed its maximum output power. Before power allocation, the maximum output total power and the required total power are compared. The maximum output total power is greater than the required total power. Based on the above three power modules, the power allocation of each server is as follows.

[0092] Server A: 600W from PSU 1, 360W from PSU 2, total: 960W (800W requirement + 20% redundancy).

[0093] Server B: Gets 500W from power module 2.

[0094] Server C: 300W from PSU 1, 414W from PSU 3, Total: 714W (700W requirement + 20% redundancy).

[0095] Server D: 100W from power module 1, 200W from power module 3, total: 300W.

[0096] In the above allocation method, the allocation of server B does not meet the constraint conditions, and the power to be output by power module 2 has exceeded 800W. Therefore, part of the power of server B can be adjusted to power module 3 to avoid overloading of power module 2. For example, server B obtains 60W from power module 3 and 440W from power module 2. The final allocation result is as follows.

[0097] Power module 1: 600 W (Server A) + 300 W (Server C) + 100 W (Server D) = 1000 W (does not exceed the maximum output of 1000 W).

[0098] Power module 2: 360 W (Server A) + 440 W (Server B) = 800 W (does not exceed the maximum output of 800 W).

[0099] Power module 3: 414 W (Server C) + 200 W (Server D) + 60 W (Server B) = 674 W (within the maximum output of 800 W).

[0100] It can be seen that the final power allocation satisfies the above constraints. The above allocation result is sent to the dynamic adjustment unit on the power system side, and the dynamic adjustment unit controls each power module to output the corresponding power.

[0101] In the above power distribution case, if power module 3 fails, it is obvious that the maximum total output power of the remaining two power modules 1 and 2 meets the requirement of the required total power. Therefore, it is necessary to control at least one backup power module (whose maximum output power cannot be less than 674W) to enter the working state, and then perform power distribution again.

[0102] For unexpected load changes of the server, when monitoring the load of server A during operation, due to sudden computing tasks, it instantly soared from the predicted power of 800W to 1000W (also exceeding the 20% redundancy of 960W), this change was not expected in the prediction. At this time, the dynamic adjustment unit in the power supply system that feeds back the real-time power will respond immediately and increase the power output to server A. Since the power modules 1 and 2 that originally powered server A have reached their maximum output power, only power module 3 can be adjusted to provide 40W of power to server A, thereby meeting the power demand of server A.

[0103] On the contrary, if server C suddenly stops a task, the load demand may drop to 500 W instantly, which is far lower than the predicted target power of 700 W. The dynamic adjustment module of the power system will reduce the power output to server C in time to avoid power waste.

[0104] The power supply device provided by the present invention is described below. The power supply device described below and the power supply method described above can be referred to each other.

[0105] The power supply device of the embodiment of the present invention is as follows: Figure 4 As shown, it includes the following functional modules.

[0106] The current load acquisition module 410 is used to acquire the current load data of the electrical equipment in real time during the current cycle.

[0107] The target power prediction module 420 is used to predict the target power of the electrical equipment in the next cycle based on the current load data by combining a load prediction model with statistical analysis.

[0108] The power output adjustment module 430 is used to adjust the power output of the power supply system to the power-consuming device based on the target power of the power-consuming device.

[0109] Among them, the load prediction model is a model based on machine learning, which takes the historical load data of the first cycle in the historical data of the electrical equipment as a sample, the historical detection power of the second cycle in the historical data is obtained by label training, and the second cycle is the next cycle of the first cycle.

[0110] The power supply device of the embodiment of the present invention predicts the target power of the electrical equipment in the next cycle based on the current load data in the current cycle in combination with the load prediction model and the statistical analysis method, that is, it combines the advantages of the prediction ability of the machine learning model and the trend capture ability of the statistical analysis method to ensure that in a complex and changeable load environment, the target power of the electrical equipment can be accurately and dynamically predicted according to the current load data, thereby dynamically adjusting the power supply output of the power supply system, avoiding energy waste, improving low energy utilization, and avoiding the problem that the electrical equipment is subjected to large power pressure for a long time, which will accelerate the aging and wear of the server.

[0111] In some embodiments, the target power prediction module 420 specifically includes the following modules.

[0112] The first power prediction module is used to input the current load data into the load prediction model to obtain the first power of the electrical equipment in the next cycle output by the load prediction model.

[0113] The first power determination module is used to perform statistical analysis on the current load data to predict the predicted load data of the electrical equipment in the next cycle, and determine the second power based on the predicted load data.

[0114] The target power determination module determines the target power of the electrical device in the next cycle based on the first power and the second power.

[0115] In some embodiments, the target power determination module is specifically used to obtain the target power by weighted summing the first power, the second power and their respective weights, and the sum of the weights of the first power and the second power is 1.

[0116] In some embodiments, the target power determination module is also used to adjust the weight of the second power to be greater than the weight of the first power when the next cycle belongs to a time period in which the load of the electrical equipment changes periodically; otherwise, adjust the weight of the first power to be greater than the weight of the second power.

[0117] In some embodiments, the target power determination module is specifically used to input the first power and the second power into an integrated learning model, and determine the predicted power output by the integrated learning model as the target power.

[0118] The integrated learning model is trained using the historical first power corresponding to the first power and the historical second power corresponding to the second power as samples, and the historical detection power within the historical prediction period where the historical first power and the historical second power are located as labels.

[0119] In some embodiments, the power output adjustment module 430 specifically includes the following modules.

[0120] The required power calculation module is used to obtain the target power of each power supply device, and calculate the required power of each power supply device based on each target power and the redundancy requirement corresponding to each power supply device.

[0121] The required total power calculation module is used to calculate the required total power based on the required power of each power supply device.

[0122] The maximum total output power calculation module is used to obtain the maximum output power of each power module in the current working state of the power supply system, and calculate the maximum total output power based on each maximum output power.

[0123] The backup power supply increasing module is used to control at least one backup power supply module to enter a working state when the required total power is greater than the maximum output total power.

[0124] The power distribution adjustment module is used to adjust the distribution of output power of the power supply module in the working state to each power-consuming device according to the required power of each power-consuming device, and to meet the constraint conditions that one power-consuming device obtains electrical energy from at least two power supply modules and the real-time total output power of each of the power supply modules does not exceed the corresponding maximum output power.

[0125] In some embodiments, the power supply device also includes: a fault handling module, which is used to adjust the power output of the power supply system to the power-consuming devices based on the most recently predicted target power of each power-consuming device when at least one of the power modules in a working state fails.

[0126] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the power supply method, which includes the following steps.

[0127] During the current cycle, the current load data of the electrical equipment is obtained in real time.

[0128] Based on the current load data, a load prediction model and statistical analysis are combined to predict the target power of the electrical equipment in the next cycle.

[0129] Based on the target power of the electric device, the power output of the power supply system to the electric device is adjusted.

[0130] Among them, the load prediction model is a model based on machine learning, which takes the historical load data of the first cycle in the historical data of the electrical equipment as a sample, the historical detection power of the second cycle in the historical data is obtained by label training, and the second cycle is the next cycle of the first cycle.

[0131] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0132] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the power supply method provided by the above methods, which includes the following steps.

[0133] During the current cycle, the current load data of the electrical equipment is obtained in real time.

[0134] Based on the current load data, a load prediction model and statistical analysis are combined to predict the target power of the electrical equipment in the next cycle.

[0135] Based on the target power of the electric device, the power output of the power supply system to the electric device is adjusted.

[0136] Among them, the load prediction model is a model based on machine learning, which takes the historical load data of the first cycle in the historical data of the electrical equipment as a sample, the historical detection power of the second cycle in the historical data is obtained by label training, and the second cycle is the next cycle of the first cycle.

[0137] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the power supply method provided by the above methods is implemented, and the method includes the following steps.

[0138] During the current cycle, the current load data of the electrical equipment is obtained in real time.

[0139] Based on the current load data, a load prediction model and statistical analysis are combined to predict the target power of the electrical equipment in the next cycle.

[0140] Based on the target power of the electric device, the power output of the power supply system to the electric device is adjusted.

[0141] Among them, the load prediction model is a model based on machine learning, which takes the historical load data of the first cycle in the historical data of the electrical equipment as a sample, the historical detection power of the second cycle in the historical data is obtained by label training, and the second cycle is the next cycle of the first cycle.

[0142] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0143] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power supply method, characterized in that: include: In the current cycle, the current load data of the electrical equipment is obtained in real time; Based on the current load data, a load prediction model and statistical analysis are combined to predict the target power of the electrical equipment in the next cycle; Based on the target power of the electrical equipment, adjusting the power output of the power supply system to the electrical equipment; Among them, the load prediction model is a model based on machine learning, which takes the historical load data of the first cycle in the historical data of the electrical equipment as a sample, the historical detection power of the second cycle in the historical data is obtained by label training, and the second cycle is the next cycle of the first cycle.

2. The power supply method according to claim 1, characterized in that: Based on the current load data, a load prediction model and statistical analysis are combined to predict the target power of the electrical equipment in the next cycle, including: Inputting the current load data into the load prediction model to obtain the first power of the electrical equipment in the next cycle output by the load prediction model; Performing statistical analysis on the current load data to predict the predicted load data of the electrical equipment in the next cycle, and determining the second power based on the predicted load data; Based on the first power and the second power, a target power of the electric device in a next cycle is determined.

3. The power supply method according to claim 2, characterized in that: Determining a target power of the electric device in a next cycle based on the first power and the second power includes: The target power is obtained by weighted summing the first power, the second power and their respective weights, and the sum of the weights of the first power and the second power is 1.

4. The power supply method according to claim 3, characterized in that: When the next cycle belongs to a time period in which the load of the electrical device changes periodically, the weight of the second power is adjusted to be greater than the weight of the first power; otherwise, the weight of the first power is adjusted to be greater than the weight of the second power.

5. The power supply method according to claim 2, characterized in that: Determining a target power of the electric device in a next cycle based on the first power and the second power includes: Inputting the first power and the second power into an integrated learning model, and determining the predicted power output by the integrated learning model as the target power; The integrated learning model is trained using the historical first power corresponding to the first power and the historical second power corresponding to the second power as samples, and the historical detection power within the historical prediction period where the historical first power and the historical second power are located as labels.

6. The power supply method according to claim 1, characterized in that: Adjusting the power output of the power supply system to the power consumption device based on the target power of the power consumption device includes: Obtaining the target power of each power supply device, and calculating the required power of each power supply device based on each target power and the redundancy requirement corresponding to each power supply device; Calculate the total power required based on the required power of each power supply device; Obtaining the maximum output power of each power module in the current working state of the power supply system, and calculating the maximum total output power based on each maximum output power; When the required total power is greater than the maximum output total power, controlling at least one backup power module to enter a working state; According to the power requirements of each electrical device, the distribution of the output power of the power modules in the working state to each electrical device is adjusted, and the constraint condition that one electrical device obtains electrical energy from at least two power modules and the real-time total output power of each power module does not exceed the corresponding maximum output power is met.

7. The power supply method according to any one of claims 1 to 6, characterized in that: Also includes: When at least one of the working power modules fails, the power output of the power supply system to the power consuming devices is adjusted based on the most recently predicted target power of each power consuming device.

8. A power supply device, characterized in that: include: The current load acquisition module is used to obtain the current load data of the electrical equipment in real time during the current cycle; A target power prediction module, used to predict the target power of the electrical equipment in the next cycle based on the current load data by combining a load prediction model with statistical analysis; A power output adjustment module, used to adjust the power output of the power supply system to the power-consuming device based on the target power of the power-consuming device; Among them, the load prediction model is a model based on machine learning, which takes the historical load data of the first cycle in the historical data of the electrical equipment as a sample, the historical detection power of the second cycle in the historical data is obtained by label training, and the second cycle is the next cycle of the first cycle.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the power supply method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the power supply method according to any one of claims 1 to 7 is implemented.

Images