An energy scheduling management method and system for a container data center

Abstract

This invention discloses an energy dispatching and management method and system for container data centers, belonging to the field of energy dispatching technology. The method includes prioritizing the electricity demand of containers based on the importance of containers with different functions; collecting multi-dimensional power data from different functional containers in real time; determining the current power consumption status based on the current multi-dimensional power data; and executing corresponding energy dispatching strategies based on the current power consumption status and the priority division results. This invention, using the above method and system, ensures the reliability of power supply and business continuity for containers under normal and abnormal operating conditions. By predicting future short-term power demand and performing power dispatching based on the prediction results and the hierarchical classification of each container, it avoids energy waste or insufficient power supply redundancy.

Description

Technical Field

[0001] This invention relates to the field of energy dispatching technology, and in particular to an energy dispatching management method and system for container data centers. Background Technology

[0002] Containerized data centers, due to their flexible deployment, convenient migration, and strong scalability, have been widely used in edge computing, emergency command, and temporary meeting venues. In actual operation, the power demand characteristics of each functional container differ significantly: IT containers experience frequent load fluctuations, network containers have extremely high requirements for power supply continuity, power distribution containers handle power allocation, and diesel generator containers serve as backup power sources. How to rationally and efficiently schedule energy for each container under different power conditions (normal or abnormal mains power) is a crucial issue for ensuring business continuity and energy efficiency optimization in data centers.

[0003] Currently, existing technologies for energy dispatch and management of container data centers have the following main shortcomings: 1. Most solutions adopt a uniform power supply strategy for all containers, without fully considering the differences between containers with different functions, resulting in critical business operations not being given priority when power is scarce; 2. Traditional methods often employ real-time response or rule-based threshold control, failing to utilize power demand forecasts for advance scheduling. This makes it difficult to cope with rapid load changes, resulting in energy waste or insufficient power supply redundancy. Summary of the Invention

[0004] The purpose of this invention is to provide an energy dispatching and management method and system for container data centers to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an energy dispatching and management method for a containerized data center, the containerized data center comprising an IT container, a network container, a power distribution container, and a diesel generator container, comprising the following steps: S1. Prioritize the electricity demand of containers based on the importance of containers with different functions; S2. Collect multi-dimensional power data of containers with different functions in real time, and determine the current power consumption status based on the current multi-dimensional power data. The current power consumption status includes normal mains power status and abnormal mains power status. S3. Based on the current power consumption status and the priority division results, execute the corresponding energy dispatch strategy; S31. If the mains power is in normal condition, based on the current multi-dimensional power data, use the pre-trained power demand prediction model to predict the power demand change trend of each container in the near future, and schedule the power of each container according to the priority division results and the power demand change trend. S32. If the mains power is abnormal, diesel generator containers will be used to generate electricity, and a tiered emergency dispatch strategy will be implemented.

[0006] Preferably, step S1 specifically includes: based on the importance of containers with different functions, using the analytic hierarchy process (AHP) to comprehensively evaluate IT containers, network containers, power distribution containers, and diesel generator containers, and dividing each container into three levels, where IT containers and network containers are high priority sets, power distribution containers are medium priority sets, and diesel generator containers are low priority sets, and the core computing nodes in IT containers have higher priority than auxiliary computing nodes.

[0007] Preferably, the evaluation parameters include business continuity impact, load dynamic sensitivity, redundancy capability, and recovery time requirements.

[0008] Preferably, step S2 specifically includes: Real-time data collection includes mains voltage, mains current, active power, reactive power, power factor, mains frequency, actual energy consumption of each container, diesel generator start / stop status, and real-time load rate of each container. The collected data is preprocessed, including handling missing values, handling outliers, normalization, and timestamp alignment. Based on the preprocessed power data, thresholds for judging normal and abnormal mains power are set, and the current power consumption status is comprehensively judged.

[0009] Preferably, the electricity demand forecasting model uses an improved LSTM model, comprising a sequentially connected one-dimensional convolutional layer, two unidirectional LSTM layers, a self-attention layer, a fully connected layer, a Dropout layer, and an output layer, with the loss function being: ; In the formula, Indicates priority weight. The time decay weight is represented using an exponential decay form. This represents the Huber robust loss function. Indicates L2 regularization, N Indicates batch size, The model predicts the first The first sample The first time step The power of a container This represents the corresponding actual power value; ; In the formula, Represents the true value of a single sample Compared with the predicted value The cost of error between them This represents a positive real number threshold.

[0010] Preferably, based on the priority allocation results and the trend of electricity demand changes, the power of each container is dispatched, and the dispatching objectives include: ; In the formula, Indicates the scheduling time. Indicates a time step. The model number is represented by the first... The predicted values ​​for each time period, i=1,2,3,4 represent IT containers, network containers, power distribution containers, and diesel generator containers, respectively. For the first Dynamic reference power of each container To smooth out the penalty coefficient, , For time step.

[0011] Preferably, the scheduling target constraints include: The sum of the power allocated at any given time shall not exceed the available capacity of the mains power. The formula is: ; The upper and lower limits of the power consumption of a single container are given by the following formula: ; In the formula, To maintain the minimum operating requirements for containers, Indicates the maximum permissible power of the container; Power variation constraint, the formula is: ; In the formula, This indicates a power variation constraint.

[0012] Preferably, step S32 specifically includes: S321. Upon confirmation of an abnormal mains power signal, the diesel generator container is immediately triggered to generate electricity. S322. Based on the priority division results and the real-time load data of each container, determine the initial power generation of the diesel generator. The initial power generation shall not be less than the sum of the real-time loads of the high-grade and medium-grade containers, and shall not exceed 90% of the rated power of the diesel generator. The load of the high-grade container shall account for no less than 60% of the initial power generation. S323. Differentiated power supply allocation shall be implemented according to priority level. The power supply power of high-level containers shall not be less than 1.05 times their real-time load, and the power supply power of medium-level containers shall not be less than 0.95 times their real-time load.

[0013] Preferably, step S3 further includes: real-time monitoring of the mains power recovery status, and gradually switching the power supply source when the mains power output power is stable.

[0014] The present invention also provides an energy dispatch and management system for a container data center, comprising: The priority allocation module is used to prioritize the electricity demand of containers based on the importance of containers with different functions, and output the priority allocation results. The data acquisition and status judgment module is used to collect multi-dimensional power data of containers with different functions in real time, and judge the current power consumption status based on the current multi-dimensional power data. The current power consumption status includes normal mains power status and abnormal mains power status. The energy dispatch execution module is connected to the priority division module and the data acquisition and status judgment module, respectively, and is used to execute the corresponding energy dispatch strategy according to the current power consumption status and priority division results.

[0015] Therefore, the present invention employs the above-described energy dispatch management method and system for container data centers, which has the following beneficial effects: (1) By taking into account the impact of business continuity, load dynamic sensitivity, redundancy capability and recovery time requirements through the analytic hierarchy process, each container is classified to ensure that high priority loads receive the highest priority power supply guarantee when the mains power is abnormal or the capacity is limited, so as to minimize the risk of business interruption. (2) Use the improved LSTM model to predict the short-term power demand in the future, and carry out power dispatch based on the prediction results and the classification of each container to avoid power waste or insufficient power supply redundancy.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of an embodiment of an energy dispatch management method and system for container data centers according to the present invention; Figure 2 This is a system framework diagram of the regulating component of an embodiment of an energy dispatching and management system for a container data center according to the present invention. Detailed Implementation

[0018] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0020] Example like Figure 1-2 As shown, this invention provides an energy dispatching and management method for a containerized data center, which includes an IT container, a network container, a power distribution container, and a diesel generator container, comprising the following steps: S1. Based on the importance of containers with different functions, the priority of container power demand is divided. Specifically, the analytic hierarchy process (AHP) is used to comprehensively evaluate IT containers, network containers, power distribution containers, and diesel generator containers. Evaluation parameters include business continuity impact, load dynamic sensitivity, redundancy capacity, and recovery time requirements. Each container is divided into three levels. After comprehensive evaluation, IT containers and network containers are assigned to the high-priority set, power distribution containers to the medium-priority set, and diesel generator containers to the low-priority set. Within IT containers, core computing nodes have a higher priority than auxiliary computing nodes.

[0021] S2. Real-time collection of multi-dimensional power data from containers with different functions; based on the current multi-dimensional power data, determining the current power consumption status, which includes normal mains power status and abnormal mains power status. Specifically, this includes: Real-time data collection includes mains voltage, mains current, active power, reactive power, power factor, mains frequency, actual energy consumption of each container, diesel generator start / stop status, and real-time load rate of each container. The collected data is preprocessed, including handling missing values, handling outliers, normalization, and timestamp alignment. Based on the preprocessed power data, thresholds for judging normal and abnormal mains power are set to comprehensively determine the current power consumption status. A normal mains power condition satisfies the following conditions: mains voltage 0.38kV ≤ U ≤ 0.42kV, mains frequency 49.5Hz ≤ U. f ≤50.5Hz, power factor cos ≥0.85, and the duration of all the above conditions is ≥3s; An abnormal mains power condition is defined as follows, and the condition persists for the corresponding duration: U < 0.38kV or U > 0.42kV, duration ≥ 0.5s; f < 49.5Hz or f > 50.5Hz, duration ≥ 3s; cos <0.85, duration ≥10s; mains power interruption (U=0), duration ≥0.1s.

[0022] S3. Based on the current power consumption status and the priority allocation results, execute the corresponding energy dispatch strategy. Specifically, this includes: S31. If the mains power is in normal condition, based on the current multi-dimensional power data, use a pre-trained power demand forecasting model to predict the power demand change trend of each container in the near future, and schedule the power of each container according to the priority division results and the power demand change trend.

[0023] The electricity demand forecasting model uses an improved LSTM model, consisting of a sequentially connected one-dimensional convolutional layer, two unidirectional LSTM layers, a self-attention layer, a fully connected layer, a Dropout layer, and an output layer. The loss function is: ; In the formula, This indicates priority weights, with higher priority weights having greater weights than medium priority weights, and medium priority weights having greater weights than lower priority weights. The time decay weight is represented using an exponential decay form. This represents the Huber robust loss function. Indicates L2 regularization, N Indicates batch size, The model predicts the first The first sample The first time step The power of a container This represents the corresponding actual power value; ; In the formula, Represents the true value of a single sample Compared with the predicted value The cost of error between them This represents a positive real number threshold.

[0024] The training process is as follows: Multi-dimensional power data of containers with different functions under normal mains power conditions over the past three months were selected as the training dataset. The data acquisition frequency was 1 time / second. The raw data underwent preprocessing: outlier removal, using the 3σ criterion to remove abnormal data caused by sensor failure and transmission interference; missing value handling, using linear interpolation to fill in a small number of missing data; data normalization, normalizing all feature data to the [0,1] interval to eliminate the influence of units; and timestamp alignment to eliminate the influence of time. The preprocessed training sample set was divided into training set, validation set, and test set in a 7:2:1 ratio.

[0025] Set the model training hyperparameters as follows: initial learning rate of 0.001, batch size of 32, training epochs of 100, and early stopping patience value of 10.

[0026] The improved LSTM model is trained, and the validation set loss is monitored in real time during the training process. If the validation set loss does not decrease for 10 consecutive rounds, the training is stopped, the current optimal model parameters are saved, and the model is prevented from overfitting.

[0027] After each training round, the validation set samples are input into the model, the validation set loss and the validation set prediction error are calculated, and the model hyperparameters are adjusted according to the validation results. If both the validation set loss and the training set loss are high, it indicates that the model is underfitting. The number of LSTM hidden units and the output dimension of the fully connected layer are increased appropriately. If the model training convergence speed is too slow, the initial learning rate is increased appropriately and the batch size is adjusted.

[0028] Specifically, based on the priority allocation results and the trend of electricity demand changes, the power supply to each container is dispatched, and the dispatching objectives include: ; In the formula, Indicates the scheduling time. Indicates a time step. The model number is represented by the first... The predicted values ​​for each time period, i=1,2,3,4 represent IT containers, network containers, power distribution containers, and diesel generator containers, respectively. For the first The dynamic reference power of each container is taken as the smaller of the average actual power or the rated power of the container over the past M time steps. To smooth out the penalty coefficient, , For time step.

[0029] The scheduling target constraints include: The sum of the power allocated at any given time shall not exceed the available capacity of the mains power. The formula is: ; The upper and lower limits of the power consumption of a single container are given by the following formula: ; In the formula, To maintain the minimum operating requirements for containers, Indicates the maximum permissible power of the container; Power variation constraint, the formula is: ; In the formula, This indicates a power variation constraint.

[0030] S32. In the event of an abnormal mains power supply, diesel generator containers will be used for power generation, and a tiered emergency dispatch strategy will be implemented. Specifically, this includes: S321. Upon confirmation of an abnormal mains power signal, immediately trigger the diesel generator container start command with a start delay of ≤5s. After start-up, quickly complete speed, voltage, and frequency calibration (calibration time ≤10s) to ensure that the output power meets the container power supply standards (frequency 50Hz±0.5Hz, voltage 0.4kV±2%). After the generator output stabilizes, adopt a gradual load increase strategy: the initial load rate does not exceed 30% of the rated power, and then the load is increased at a rate not exceeding 10% of the rated power per minute until the initial power generation determined in step S322 is reached to avoid sudden power surges impacting the generator and power supply system. If the main diesel generator fails to start, immediately start the backup generator set and issue a first-level alarm, prioritizing emergency power supply to the first-priority container.

[0031] S322. Based on the priority division results and the real-time load data of each container, determine the initial power generation of the diesel generator. The initial power generation shall not be less than the sum of the real-time loads of the high-grade and medium-grade containers, and shall not exceed 90% of the rated power of the diesel generator. The load of the high-grade container shall account for no less than 60% of the initial power generation.

[0032] S323. Implement differentiated power supply allocation according to priority levels. The power supply power of the high-level container shall not be less than 1.05 times its real-time load and shall not exceed the maximum allowable power of the container. The power supply power of the medium-level container shall not be less than 0.95 times its real-time load and shall not exceed its maximum allowable power. The load of the low-level container shall be interrupted when necessary. If the 1.05 times load demand of the high-level container exceeds its maximum allowable power, the maximum allowable power shall be supplied and a level three alarm shall be issued to indicate that the container is approaching its power supply limit.

[0033] S324. Real-time monitoring of load changes in containers of each priority level and the operating status of diesel generators. When the load of a high-priority container increases, its power supply is increased first, which can reduce the non-core load of low-priority containers.

[0034] S325. Real-time monitoring of diesel generator fuel reserves. When the fuel reserve is below 30%, a level 3 alarm is issued, prioritizing power supply to high-priority containers. If the diesel generator malfunctions, a level 2 alarm is immediately issued, the backup generator set is started, and power supply is prioritized for first-priority containers during power switching. Normal hierarchical dispatching is restored after the fault is cleared.

[0035] S33. Monitor the mains power recovery status in real time, and gradually switch the power supply source when the mains power output power is stable.

[0036] like Figure 2 As shown, the present invention also provides an energy dispatch and management system for a container data center, comprising: The priority allocation module is used to prioritize the electricity demand of containers based on the importance of containers with different functions, and output the priority allocation results. The data acquisition and status judgment module is used to collect multi-dimensional power data of containers with different functions in real time, and judge the current power consumption status based on the current multi-dimensional power data. The current power consumption status includes normal mains power status and abnormal mains power status; the power supply to the lower-level container is suspended and its load is completely disconnected.

[0037] The energy dispatch execution module is connected to the priority division module and the data acquisition and status judgment module, respectively, and is used to execute the corresponding energy dispatch strategy according to the current power consumption status and priority division results.

[0038] Therefore, the present invention adopts the above-mentioned energy dispatch management method and system for container data centers, which ensures the power supply reliability and business continuity of containers under normal and abnormal operating conditions, avoids the interruption of critical loads due to prediction deviations or delayed emergency response, and avoids power waste or insufficient power supply redundancy by predicting future short-term power demand and dispatching power based on the prediction results and the classification of each container.

[0039] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for energy scheduling management of a container data center, the container data center comprising an IT container, a network container, a power distribution container, and a diesel generator container, the method comprising: Includes the following steps: ​ S1. Prioritize the electricity demand of containers based on the importance of containers with different functions; S2. Collect multi-dimensional power data of containers with different functions in real time, and determine the current power consumption status based on the current multi-dimensional power data. The current power consumption status includes normal mains power status and abnormal mains power status. S3. Based on the current power consumption status and the priority division results, execute the corresponding energy dispatch strategy; S31. If the mains power is in normal condition, based on the current multi-dimensional power data, use the pre-trained power demand prediction model to predict the power demand change trend of each container in the near future, and schedule the power of each container according to the priority division results and the power demand change trend. S32. If the mains power is abnormal, diesel generator containers will be used to generate electricity, and a tiered emergency dispatch strategy will be implemented.

2. The energy scheduling management method for a container data center of claim 1, wherein, Step S1 specifically includes: Based on the importance of containers with different functions, the analytic hierarchy process (AHP) is used to comprehensively evaluate IT containers, network containers, power distribution containers, and diesel generator containers, and each container is divided into three levels. Among them, IT containers and network containers are high priority sets, power distribution containers are medium priority sets, and diesel generator containers are low priority sets. In IT containers, the core computing nodes have a higher priority than the auxiliary computing nodes.

3. The method for energy scheduling management of a container data center of claim 2, wherein: Evaluation parameters include business continuity impact, load dynamic sensitivity, redundancy capability, and recovery time requirements.

4. The energy dispatching and management method for a container data center according to claim 1, characterized in that, Step S2 specifically includes: Real-time data collection includes mains voltage, mains current, active power, reactive power, power factor, mains frequency, actual energy consumption of each container, diesel generator start / stop status, and real-time load rate of each container. The collected data is preprocessed, including handling missing values, handling outliers, normalization, and timestamp alignment. Based on the preprocessed power data, thresholds for judging normal and abnormal mains power are set, and the current power consumption status is comprehensively judged.

5. The energy dispatching and management method for a container data center according to claim 4, characterized in that, The electricity demand forecasting model uses an improved LSTM model, consisting of a sequentially connected one-dimensional convolutional layer, two unidirectional LSTM layers, a self-attention layer, a fully connected layer, a Dropout layer, and an output layer. The loss function is: ； In the formula, Indicates priority weight. The time decay weight is represented using an exponential decay form. This represents the Huber robust loss function. Indicates L2 regularization, N Indicates batch size, The model predicts the first The first sample The first time step The power of a container This represents the corresponding actual power value; ； In the formula, Represents the true value of a single sample Compared with the predicted value The cost of error between them This represents a positive real number threshold.

6. The energy dispatching and management method for a container data center according to claim 5, characterized in that, Based on the priority allocation results and the trend of electricity demand changes, the power supply to each container is dispatched. The dispatching objectives include: ； In the formula, Indicates the scheduling time. Indicates a time step. The model number is represented by the first... The predicted values ​​for each time period, i=1,2,3,4 represent IT containers, network containers, power distribution containers, and diesel generator containers, respectively. For the first Dynamic reference power of each container To smooth out the penalty coefficient, , For time step.

7. The energy dispatching and management method for a container data center according to claim 6, characterized in that, The scheduling target constraints include: The sum of the power allocated at any given time shall not exceed the available capacity of the mains power. The formula is: ； The upper and lower limits of the power consumption of a single container are given by the following formula: ； In the formula, To maintain the minimum operating requirements for containers, Indicates the maximum permissible power of the container; Power variation constraint, the formula is: ； In the formula, This indicates a power variation constraint.

8. The energy dispatching and management method for a container data center according to claim 7, characterized in that, Step S32 specifically includes: S321. Upon confirmation of an abnormal mains power signal, the diesel generator container is immediately triggered to generate electricity. S322. Based on the priority division results and the real-time load data of each container, determine the initial power generation of the diesel generator. The initial power generation shall not be less than the sum of the real-time loads of the high-grade and medium-grade containers, and shall not exceed 90% of the rated power of the diesel generator. The load of the high-grade container shall account for no less than 60% of the initial power generation. S323. Differentiated power supply allocation shall be implemented according to priority level. The power supply power of high-level containers shall not be less than 1.05 times their real-time load, and the power supply power of medium-level containers shall not be less than 0.95 times their real-time load.

9. The energy dispatching and management method for a container data center according to claim 8, characterized in that, Step S3 also includes: real-time monitoring of the mains power recovery status, and gradually switching the power supply source when the mains power output power is stable.

10. An energy dispatching and management system for a container data center, executing the energy dispatching and management method for a container data center as described in any one of claims 1-9, characterized in that, include: The priority allocation module is used to prioritize the electricity demand of containers based on the importance of containers with different functions, and output the priority allocation results. The data acquisition and status judgment module is used to collect multi-dimensional power data of containers with different functions in real time, and judge the current power consumption status based on the current multi-dimensional power data. The current power consumption status includes normal mains power status and abnormal mains power status. The energy dispatch execution module is connected to the priority division module and the data acquisition and status judgment module, respectively, and is used to execute the corresponding energy dispatch strategy according to the current power consumption status and priority division results.

Images