WO2022057427A1 - 功率预测的方法、装置及设备 - Google Patents

功率预测的方法、装置及设备 Download PDF

Info

Publication number
WO2022057427A1
WO2022057427A1 PCT/CN2021/106964 CN2021106964W WO2022057427A1 WO 2022057427 A1 WO2022057427 A1 WO 2022057427A1 CN 2021106964 W CN2021106964 W CN 2021106964W WO 2022057427 A1 WO2022057427 A1 WO 2022057427A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
power
prediction
unit
power consumption
Prior art date
Application number
PCT/CN2021/106964
Other languages
English (en)
French (fr)
Inventor
王淑倩
柴峰
姜炜祥
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to EP21868256.5A priority Critical patent/EP4207528A4/en
Publication of WO2022057427A1 publication Critical patent/WO2022057427A1/zh
Priority to US18/183,473 priority patent/US20230216296A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/008—Circuit arrangements for ac mains or ac distribution networks involving trading of energy or energy transmission rights
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
    • H02J13/00001—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by the display of information or by user interaction, e.g. supervisory control and data acquisition systems [SCADA] or graphical user interfaces [GUI]
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
    • H02J13/00006—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by information or instructions transport means between the monitoring, controlling or managing units and monitored, controlled or operated power network element or electrical equipment
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
    • H02J13/00006—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by information or instructions transport means between the monitoring, controlling or managing units and monitored, controlled or operated power network element or electrical equipment
    • H02J13/00016—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by information or instructions transport means between the monitoring, controlling or managing units and monitored, controlled or operated power network element or electrical equipment using a wired telecommunication network or a data transmission bus
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
    • H02J13/00006—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by information or instructions transport means between the monitoring, controlling or managing units and monitored, controlled or operated power network element or electrical equipment
    • H02J13/00019—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by information or instructions transport means between the monitoring, controlling or managing units and monitored, controlled or operated power network element or electrical equipment using optical means
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/28—Arrangements for balancing of the load in a network by storage of energy
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/28—Arrangements for balancing of the load in a network by storage of energy
    • H02J3/32—Arrangements for balancing of the load in a network by storage of energy using batteries with converting means
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2310/00—The network for supplying or distributing electric power characterised by its spatial reach or by the load
    • H02J2310/10—The network having a local or delimited stationary reach
    • H02J2310/12—The local stationary network supplying a household or a building
    • H02J2310/16—The load or loads being an Information and Communication Technology [ICT] facility
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/14—Energy storage units
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S40/00—Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them
    • Y04S40/12—Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them characterised by data transport means between the monitoring, controlling or managing units and monitored, controlled or operated electrical equipment
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S40/00—Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them
    • Y04S40/12—Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them characterised by data transport means between the monitoring, controlling or managing units and monitored, controlled or operated electrical equipment
    • Y04S40/124—Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them characterised by data transport means between the monitoring, controlling or managing units and monitored, controlled or operated electrical equipment using wired telecommunication networks or data transmission busses

Definitions

  • the present application relates to the field of computer technology, and in particular, to a power prediction method, apparatus, and device.
  • a data center is a global collaborative network of specific devices used to transmit, accelerate, display, compute, and store data information on the Internet infrastructure. Data centers consume a lot of power during operation.
  • an energy storage system can be provided for the data center. The energy storage system can be charged when electricity prices are low and discharged when electricity prices are high, thereby powering the data center.
  • the dispatch management system In order to maximize the utilization of the energy storage system, the dispatch management system usually needs to provide a reasonable charging and discharging strategy according to the power of the data center in the future, combined with information such as electricity price curve and battery status. Based on this, how to accurately predict the power of the data center has become a key concern of the industry.
  • the present application provides a method of power prediction.
  • the method evaluates the models in the model pool online through the evaluation value of the model, and selects an appropriate model to predict the power of the first power consumption unit according to the evaluation result, thereby adapting to the power change of the first power consumption unit and improving the prediction accuracy .
  • the present application also provides apparatuses, devices, computer-readable storage media, and computer program products corresponding to the above methods.
  • the present application provides a method for power prediction.
  • the method can be applied to a dispatch management system, specifically a power prediction module in the dispatch management system.
  • the dispatch management system eg, a power prediction module
  • the dispatch management system performs power prediction for the data center with the granularity of one power consumption unit in the data center.
  • the power prediction method of the present application will be described below by taking the power prediction of any power consumption unit in the data center, that is, the first power consumption unit as an example.
  • the dispatching management system obtains the evaluation index value of the model in the model pool, and the evaluation index value is used to indicate the accuracy of the model, and then the dispatching management system selects the first model according to the evaluation index value to carry out the first power consumption unit. power prediction, and then the dispatch management system presents the results of the power prediction for the first powered unit.
  • the method uses the evaluation value of the model to evaluate the models in the model pool online, and selects an appropriate model to predict the power of the first power unit according to the evaluation results, instead of always using a single model to predict the power of the first unit.
  • the business types of different power consumption units such as different electrical equipment of the whole cabinet, may be different.
  • This method uses the power consumption unit as the granularity to perform power prediction, which can make the power prediction for each power consumption unit more accurate. High prediction accuracy improves the accuracy of power prediction for the data center as a whole.
  • the dispatch management system may select a first model from the model pool according to the evaluation index value, and then use the first model according to the historical power distribution of the first power consumption unit.
  • the model predicts the future power distribution of the first power consumption unit.
  • the historical power distribution indicates the power of at least one statistical period before the current moment
  • the future power distribution indicates the power of at least one statistical period after the current moment.
  • the statistical period is equal to the time interval over which the power distribution is predicted.
  • the statistical period may be a collection period for collecting the real power value.
  • the statistical period may also be an integer multiple of the acquisition period.
  • the power prediction of the first power consuming unit in different statistical periods can be realized, and the power change can be adapted to have better prediction accuracy.
  • the evaluation index value includes an error value.
  • the error value can be specifically determined according to the predicted value and the actual value of the power.
  • the scheduling management system may select a model whose error value meets a preset condition from the model pool as the first model according to the error value. Therefore, a model with higher prediction accuracy is selected to perform power prediction on the first power consuming unit, so that the power prediction of the first unit maintains a higher accuracy.
  • the error value satisfying a preset condition includes: the error value is smaller than a preset threshold; or, the error value is the smallest.
  • the scheduling management system may compare the error value of the model with a preset threshold, and select a model with an error value smaller than the preset threshold as the first model.
  • the dispatch management system may randomly select one model from the multiple models as the first model, or select the model with the smallest error value as the first model.
  • the dispatch management system can directly select the model with the smallest error value as the first model.
  • a suitable first model can be selected to predict the power of the first power consumption unit, so that the power prediction of the first power consumption unit has better accuracy.
  • the scheduling management system may update the model in the model pool according to the evaluation index value, and then select the first model from the updated model pool according to the updated evaluation index value of the model.
  • a model Specifically, when the evaluation index values (for example, the error value) of the models in the model pool are not less than the preset threshold, it indicates that the accuracy of the model cannot meet the requirements, and at least one model has exceeded the preset time since the last update time, such as more than M day, the historical data changes relatively large, and the model management system can update the model according to the updated historical data. In this way, the dispatch management system selects the first model from the updated model pool according to the evaluation index value of the updated model to perform power prediction on the first power consumption unit. On the one hand, better prediction accuracy can be obtained, and on the other hand, the balance between prediction accuracy and computing resource consumption can be achieved.
  • the evaluation index value is determined according to at least one of an interval error value and a single-point error value during power prediction of the first power consuming unit in the data center.
  • the dispatch management system may predict the power distribution of the first power consumption unit within a period of time, specifically the power of the first power consumption unit in at least one statistical period.
  • the statistical period may be a power collection period. For example, when power is collected every 10 minutes, the collection period is 10 minutes, and the corresponding statistical period may be 10 minutes.
  • the dispatch management system predicts the power of the first power consumption unit in multiple statistical periods, it can predict the deviation value of the power of the first power consumption unit in each statistical period according to the model (for example, the absolute value of the deviation value). value) to determine the error value of the model.
  • the error value is the above single-point error value.
  • the dispatching management system can assign the price to belong to the same time interval.
  • the deviation values are summed to determine the error value in the interval, and then the error value of the model is determined according to the error value of the model in each time interval. This error value is also referred to as an interval error value.
  • the interval error value is usually smaller than the single-point error value.
  • the number of model updates can be reduced, so as to avoid the consumption of a large amount of computing resources caused by frequent model updates.
  • the dispatch management system may also, according to the power distribution of the existing first power consumption unit in the data center and the first model for predicting the first power consumption unit, provide a new second power consumption unit for the newly added second power consumption unit.
  • the electric unit recommends a suitable second model for power prediction, thereby shortening the time for determining the second model, enabling the dispatch management system to achieve higher prediction accuracy in a relatively short period of time and reducing power costs.
  • the scheduling management system acquires the power distribution of the second power consumption unit newly added in the data center, and then determines from the first power consumption unit that the similarity with the power distribution of the second power consumption unit reaches At least one third power consuming unit with a preset similarity degree, and then determining a second model for performing power prediction on the second power consuming unit according to a model for performing power prediction on the third power consuming unit.
  • the dispatch management system may directly determine the model for power prediction of the third power consumption unit as the second model for power prediction of the second power consumption unit.
  • the time for determining the second model is greatly shortened, so that the power prediction of the second power consumption unit can achieve higher accuracy in a shorter time.
  • the dispatch management system may add a model for predicting the power of the at least one third power consumption unit to a model pool of the second power consumption unit, and obtain the model of the second power consumption unit
  • the evaluation index value of the in-battery model for predicting the power of the second power consumption unit, and according to the evaluation index value, it is selected from the model pool of the second power consumption unit for power prediction of the second power consumption unit.
  • the model in the model pool is a model that is affinity with the second power consumption unit. Selecting a model from the model pool according to the evaluation index value can make the power prediction of the second power consumption unit achieve high accuracy in a short time.
  • the dispatch management system may also generate training samples according to the historical power distribution of the first power consumption unit, and then use the training samples to train the initial model to obtain the model in the model pool, and then It is used to predict the power of the first unit to ensure the prediction accuracy.
  • the model pool includes two or more models.
  • the model pool may include one or more of tree models, neural network models, autoregressive models, and simple models.
  • a model pool may specifically include different models under this model.
  • the model pool includes a tree model, it can include an extreme gradient boosting model and a random forest model.
  • the model pool includes a neural network model, it may specifically include a deep neural network model and a long short-term memory network model.
  • the simple model is specifically a model that uses the mathematical statistics of historical power to predict the future power.
  • the mathematical statistical value may be any one of the weighted average value, arithmetic average value, median value, maximum value, minimum value, etc. of historical power.
  • the first power consuming unit is a set of power consuming units
  • the set of power consuming units includes at least one power consuming device, at least one power consuming device in an entire cabinet, or at least one power consuming device in a computer room equipment.
  • the method performs power prediction with one power consumption unit as the prediction granularity, which can make the power prediction for each power consumption unit have high prediction accuracy.
  • the power consumption unit can be an independent power consumption device (such as a server) , or all or part of the electrical equipment included in a single cabinet, or all or part of the electrical equipment in the entire equipment room.
  • Using a model pool including multiple models to predict different sets of electrical equipment enables the prediction granularity to be dynamically adjusted according to business requirements, thereby improving the accuracy of power prediction for the data center as a whole.
  • the dispatch management system may further present a result of the power prediction of the first power consumption unit by at least one model in the model pool through a graphical user interface. This can help users make decisions.
  • the present application provides an apparatus for power prediction.
  • the apparatus includes various units for performing the method for power prediction in the first aspect or any possible implementation manner of the first aspect.
  • the present application provides an apparatus including a processor and a memory.
  • the processor and the memory communicate with each other.
  • the processor is configured to execute instructions stored in the memory to cause an apparatus to perform a method of power prediction as in the first aspect or any implementation of the first aspect.
  • the present application provides a computer-readable storage medium, where instructions are stored in the computer-readable storage medium, and the instructions instruct a device to execute the first aspect or any implementation manner of the first aspect.
  • Methods of power prediction are stored in the computer-readable storage medium, and the instructions instruct a device to execute the first aspect or any implementation manner of the first aspect.
  • the present application provides a computer program product comprising instructions that, when executed on a device, cause the device to perform the power prediction method described in the first aspect or any one of the implementations of the first aspect.
  • the present application may further combine to provide more implementation manners.
  • FIG. 1 is a system architecture diagram for managing an energy storage system by a dispatch management system provided by an embodiment of the present application
  • FIG. 2 is a flowchart of a method for power prediction provided by an embodiment of the present application
  • FIG. 3 is a schematic diagram of presenting a result of power prediction provided by an embodiment of the present application.
  • 4A is a schematic diagram of determining a second model according to an embodiment of the present application.
  • 4B is a schematic diagram of determining a second model according to an embodiment of the present application.
  • FIG. 5 is a flow chart of generating sample data according to an embodiment of the present application.
  • FIG. 6 is a schematic structural diagram of an apparatus for power prediction provided by an embodiment of the present application.
  • FIG. 7 is a schematic structural diagram of a device provided by an embodiment of the present application.
  • a data center is a global collaborative network of specific devices used to transmit, accelerate, display, compute, and store data information on the Internet infrastructure.
  • Devices such as servers, switches, etc., used to transmit, accelerate, display, calculate, and store data information in the data center can be deployed in the entire cabinet.
  • a power supply system is usually required to supply power to various devices included in the data center.
  • the power supply system may include a mains power supply system.
  • the mains power supply system is used to directly transmit the electric energy generated by the power station to the electricity consumer (such as a data center) through the power grid, so as to supply power to the electricity consumer.
  • the price of electricity provided by the power company during different time periods can be different.
  • the owner of the data center can build an energy storage system, control the charging of the energy storage system during the low electricity price period, and control the discharge of the energy storage system during the high electricity price period to supply power to the data center. That is, the power supply system may also include an energy storage system.
  • the energy storage system includes at least one energy storage device.
  • the energy storage device may be a device that supports charging and discharging, for example, may be an energy storage battery.
  • the energy storage battery may include at least one or more of different types of batteries such as lithium batteries, nickel-hydrogen batteries, lead-acid batteries, and sodium-sulfur batteries.
  • the capacity of the energy storage system is limited, and the electricity price of the mains power supply system is different in different time periods, it is possible to combine the electricity price curve and the state of the energy storage equipment according to the power usage of the data center in the future period of time. information to determine the charging and discharging strategy.
  • the energy storage devices in the energy storage system are charged and discharged according to the charging and discharging strategy, so as to make full use of the capacity of the energy storage system and reduce the power cost of the data center.
  • the power usage of the data center in the future period of time can be predicted according to the power usage of the data center in the historical time.
  • the power prediction of the data center and the process of determining the charging and discharging strategy can be implemented by the dispatch management system of the energy storage system.
  • the energy storage system 100 includes at least one energy storage device 102 , and the energy storage device 102 may specifically be an energy storage battery.
  • the dispatching management system 200 is used for dispatching management of the energy storage system, for example, dispatching the energy storage system 100 to charge in some time periods and discharge in other time periods, so as to supply power to the data center.
  • the dispatch management system 200 includes a communication and control module 202 , a power prediction module 204 and a policy formulation module 206 .
  • the communication and control module 202 is used for communicating with the energy storage system 100 and the data center, and controlling the energy storage system 100 , so as to realize the scheduling management of the energy storage system 100 .
  • the power prediction module 204 is used to predict the power of the data center, for example, predict the power distribution of the data center in the future.
  • the strategy formulation module 206 is configured to formulate a charging and discharging strategy according to the predicted power, the state of the energy storage device 102 and the electricity price curve.
  • the communication and control module 202 can acquire the charging and discharging strategy and send it to the energy storage system 100 .
  • the scheduling management system 200 may further include a storage module 208 .
  • the storage module 208 may specifically be a database.
  • the storage module 208 may be used to store at least one of information such as historical power, the state of the energy storage device 102 in the energy storage system 100 and the like of the power of the data center.
  • the state of the energy storage device 102 includes one or more of a state of health (SoH), a state of charge (SoC), and the like.
  • SoH state of health
  • SoC state of charge
  • the power prediction module 204 can obtain the historical power of the data center from the above-mentioned storage module 208, so as to predict the future power of the data center.
  • the strategy formulation module 206 is configured to acquire the state information of the energy storage device 102 from the above-mentioned storage module 208, so as to be used to formulate a charging and discharging strategy.
  • the energy storage system 100 may further include an energy storage device management module 104 .
  • the communication and control module 202 of the dispatch management system 200 can communicate with the energy storage device 102 through the energy storage device management module 104 and control the energy storage device 102 .
  • the energy storage device 102 may be an energy storage battery
  • the energy storage device management module 104 may be a battery management unit (battery management system).
  • the scheduling management system 200 (for example, the power prediction module 204 ) performs power prediction for the data center, a single model is mainly used for prediction.
  • the power of the electrical equipment has a high probability of changing with time. If the original model cannot adapt to this change, the prediction accuracy will drop significantly.
  • the charging and discharging strategy formulated by the dispatch management system 200 (eg, the strategy formulation module 206 ) cannot meet the requirements.
  • an embodiment of the present application provides a method for power prediction.
  • the dispatch management system 200 eg, the power prediction module 204
  • the power consumption unit may be a set of power consumption devices
  • the set of power consumption devices includes at least one power consumption device, such as one or more servers.
  • the set of electrical devices may also be electrical devices of at least one entire cabinet, such as all or part of servers in one or more entire cabinets.
  • the set of power consumption units may also be power consumption devices of at least one computer room, such as all or part of servers in one or more computer rooms.
  • the embodiment of the present application uses an example to illustrate the prediction process of the first power consumption unit.
  • the first power consumption unit may be any power consumption unit in the data center.
  • the dispatch management system 200 can obtain the evaluation index value of the model in the model pool, and the evaluation index value can be used to indicate the accuracy of the model, and then the dispatch management system 200 selects the first model according to the evaluation index value to perform power prediction on the first power consumption unit, The results of the power prediction for the first electrical unit are then presented.
  • the method uses the evaluation value of the model to evaluate the models in the model pool online, and selects an appropriate model to predict the power of the first power unit according to the evaluation results, instead of always using a single model to predict the power of the first unit.
  • This method uses the power consumption unit as a granularity to perform power prediction, so that the power prediction for each power consumption unit has a higher power. improve the prediction accuracy of the overall data center power prediction.
  • the scheduling management system 200 provided in this embodiment of the present application may be a software module, and the software module may be deployed in a hardware device to provide external services.
  • the scheduling management system 200 has multiple deployment modes, and the various deployment modes will be described in detail below.
  • the scheduling management system 200 may be deployed in a cloud computing cluster. Wherein, each module of the scheduling management system 200 may be centrally deployed in one cloud computing device (eg, cloud server) of the cloud computing cluster, or may be distributed in different cloud computing devices of the cloud computing cluster.
  • the power prediction method provided in this embodiment of the present application may be provided to a user in the form of a cloud service.
  • the scheduling management system 200 may be deployed in a local computing device.
  • the local computing device means that the local device includes a computing device under the direct control of the user, for example, a terminal such as a desktop computer, a notebook computer, or a local server.
  • the energy storage device 102 in the energy storage system 100 may be a hardware device.
  • the hardware device is connected to the dispatch management system 200 .
  • the energy storage device 102 may be deployed in an information technology (information technology, IT) infrastructure cabinet, referred to as an IT cabinet for short.
  • IT cabinet information technology infrastructure cabinet
  • the energy storage device management module 104 may be deployed in the IT cabinet together with the energy storage device 102 .
  • FIG. 1 merely illustrates the deployment mode of the dispatch management system 200 and the energy storage system 100 .
  • the dispatch management system 200 and the energy storage system 100 may also adopt other deployment methods.
  • the method includes:
  • S202 The scheduling management system 200 obtains the evaluation index value of the model in the model pool.
  • the model pool specifically refers to a logical pool including at least one model.
  • the models in the model pool are used to make power predictions for the electrical units.
  • Each power consumption unit in the data center corresponds to a model pool.
  • the model pools of different power consumption units can be the same or different.
  • the model pool may include two or more models.
  • the pool of models may include one or more of different kinds of models, such as tree models, neural network models, autoregressive models, and simple models.
  • the model pool may specifically include different models under this category.
  • the model pool when the model pool includes a tree model, the model pool may specifically include different models in the tree model, such as an extreme gradient boosting (xgboost) model and a random forest (random forest) model.
  • the model pool when the model pool includes a neural network model, the model pool may specifically include different models in the neural network model, such as deep neural networks (DNN) and long short term memory networks (long short term memory). , LSTM).
  • DNN deep neural networks
  • long short term memory networks long short term memory
  • the simple model is specifically a model that uses the mathematical statistics of historical power to predict the future power.
  • the mathematical statistical value may be any one of the weighted average value, arithmetic average value, median value, maximum value, minimum value, etc. of historical power.
  • a simple model may take a weighted average of three historical powers as a prediction of power one day in the future.
  • Each model has an evaluation index value.
  • the evaluation index value is specifically used to indicate the accuracy of the model.
  • the evaluation index value may be an error value of the power prediction performed by the model.
  • the error value can be specifically determined according to the predicted value and the actual value of the power.
  • the dispatch management system 200 may predict the power distribution of the first power consumption unit within a period of time, specifically, the power of the first power consumption unit in at least one statistical period.
  • the statistical period may be a power collection period. For example, when power is collected every 10 minutes, the collection period is 10 minutes, and the corresponding statistical period is 10 minutes. Based on this, when the dispatch management system 200 predicts the power of the first power consumption unit in a plurality of statistical periods, it can predict the deviation value of the power of the first power consumption unit in each statistical period (for example, the deviation value of the deviation value) according to the model. absolute value) to determine the error value of the model.
  • the error value is also called the single-point error value, and the specific calculation formula is as follows:
  • error sin represents the single-point error value.
  • y prei refers to the model's predicted value of the power of the power consumption unit in the ith statistical period, and y i refers to the real power value (also called the actual power value) of the power consumption unit in the ith statistical period.
  • the dispatch management system 200 may The deviation values of the interval are summed to determine the error value in the interval, and then the error value of the model is determined according to the error value of the model in each time interval.
  • the error value is also called the interval error value, and the specific calculation formula is as follows:
  • the error range represents the interval error value.
  • M is the quantity of electricity price.
  • electricity prices in different time intervals can be 0.195, 0.58, and 0.885, and M can be 3.
  • y prej refers to the model's predicted value of the power of the power consumption unit in the jth statistical period in the time interval
  • yj refers to the real power value of the power consumption unit in the jth statistical period in the time interval.
  • Characterization refers to the average value of the real power value of the power consumption unit in the time interval.
  • Q characterizes the number of multiple statistical periods in the time interval.
  • an electricity price corresponds to multiple discontinuous time intervals
  • an electricity price corresponds to the following time intervals [0:00, 7:00), [13:00, 14:00), [23:00, 24:00)
  • the scheduling management system 200 may also combine multiple discontinuous time intervals when determining the error value.
  • Q may be the sum of the numbers of multiple statistical periods in multiple discrete time intervals.
  • the interval error value is usually smaller than the single-point error value.
  • the number of model updates can be reduced, and the frequent updating of the model can lead to consumption of a large amount of computing resources.
  • the scheduling management system 200 may also jointly determine the error value according to the single-point error value and the interval error value, as shown below:
  • error represents the error value
  • k 1 and k 2 represent the weight of the single-point error value and the interval error value, respectively.
  • the sum of k 1 and k 2 is 1.
  • the dispatch management system 200 selects a first model to perform power prediction on the first power consumption unit according to the evaluation index value.
  • the dispatch management system 200 can select a model whose accuracy meets the requirements from the model pool as the first model according to the evaluation index value, and use the first model to predict the future of the first power consumption unit according to the historical power distribution of the first power consumption unit. power distribution.
  • the historical power distribution indicates the power of at least one statistical period before the current moment
  • the future power distribution indicates the power of at least one statistical period after the current moment.
  • the number of statistical periods corresponding to the historical power distribution may be greater than or equal to the number of statistical periods corresponding to the future power distribution.
  • the dispatch management system 200 can predict the power distribution for the next day using the power distribution of the past three days.
  • the dispatch management system 200 may select a model whose error value satisfies a preset condition as the first model. Specifically, the scheduling management system 200 may compare the error value of the model with a preset threshold, so as to select a model whose error value satisfies the preset condition as the first model. For example, the dispatch management system 200 may select a model whose error value is less than a preset threshold as the first model.
  • the model with the error value satisfying the preset condition may also be the model with the smallest error value.
  • the dispatch management system 200 may select the model with the smallest error value as the first model.
  • the result of the power prediction of the first power consumption unit by the first model may be provided to the policy formulation module 206 in the dispatch management system 200 to formulate a charging and discharging policy.
  • the preset time may be set according to an empirical value, for example, the preset time may be set to L days. It should be noted that, for different models, the value of L may be the same or different, and the scheduling management system 200 may set it according to the actual situation.
  • the scheduling management system 200 may update the models in the model pool, and then select the first model from the updated model pool according to the evaluation index value of the updated model to perform the first power consuming unit Power forecast.
  • the model management system 200 can update the model according to the updated historical data.
  • model management system 200 may also store the model type, update time, hyperparameter values of the updated model, parameter values of the updated model, and evaluation index values (such as error values) of the updated model in the database. , for subsequent use.
  • the dispatch management system 200 presents the result of the power prediction for the first power consumption unit. Specifically, the dispatch management system 200 may present the power prediction result of at least one model in the model pool to the first power consumption unit to the user through a graphical user interface (graphical user interface, GUI).
  • the at least one model may include the first model determined in S202 above.
  • the dispatch management system 200 may also present the results of the power prediction of the first power consumption unit by each model in the model pool to the user through a GUI. For example, the dispatch management system 200 may present the power distribution of the first power consumption unit predicted by each model in the model pool to the user in the future one day through the GUI.
  • the interface 300 presents the first model (which may be a DNN model in this example) to the first power consumption unit (which may be a DNN model in this example)
  • the power distribution of the cabinet 4) in one day is specifically shown by the curve 301 in FIG. 3 .
  • the interface 300 also carries a power unit selection control 302, a time selection control 303, and a model selection control 304, and the user can select other power units, other times, and/ or other models, so as to present in the interface 300 the power distribution of other electrical consumers, the power distribution of electrical consumers at other times, and/or the power distribution of electrical consumers predicted by other models.
  • the user can select all models in the model pool through the model selection control 304 to display in the interface 300 the power distribution predicted by each model in the model pool.
  • the interface 300 can also carry at least one of the actual power display control 305 , the strategy display control 306 , the electricity display control 307 , the electricity price curve display control 308 , the actual income display control 309 , and the ideal income display control 310 .
  • the actual power display control 305 can also carry at least one of the actual power display control 305 , the strategy display control 306 , the electricity display control 307 , the electricity price curve display control 308 , the actual income display control 309 , and the ideal income display control 310 .
  • the dispatch management system 200 can display the actual power of the power consumption unit in the interface 300, that is, the actual power value, which can be specifically shown as the curve 311 in FIG. 3 . .
  • the scheduling management system 200 displays the charging and discharging strategy formulated by the scheduling management system 200 in the interface 300 , which can be specifically shown as the curve 312 in FIG. 3 .
  • the dispatch management system 200 can display the power of the energy storage system 100 on the interface 300 , which can be specifically shown by the curve 314 in FIG. 3 .
  • the dispatch management system 200 can display the electricity price curve in the interface 300, which can be specifically shown as the curve 313 in FIG. 3 .
  • the dispatch management system 200 displays the actual income obtained by charging and discharging the energy storage system 100 to supply power to the data center in the interface 300 , and the income is specifically shown by the curve 315 in FIG. 3 Show.
  • the dispatch management system 200 displays in the interface 300 the income that can theoretically be obtained by supplying power to the data center by charging and discharging the energy storage system 100 .
  • the income is specifically shown by the curve 316 in FIG. 3 .
  • an embodiment of the present application provides a method for power prediction.
  • the dispatch management system 200 can obtain the evaluation index value of the model in the model pool, select the first model from the model pool according to the model accuracy indicated by the evaluation index value, perform power prediction on the first power consumption unit, and then present the correct The result of the power prediction of the first power consumption unit.
  • the evaluation value of the model is used to evaluate the model in the model pool online, and the appropriate model is selected according to the evaluation result to predict the power of the first power consumption unit, instead of using a single model all the time to predict the power of the first power consumption unit.
  • the first unit performs power prediction, so it can adapt to the power change of the first power consuming unit, improve the prediction accuracy, and meet business requirements.
  • two or more model prediction results can also be presented in the interface 300 . Specifically, they can be displayed on different tabs or in different regions of the same page, respectively. Display the prediction results of each model, and display the data of each model on the interface according to different dimensions (for example, power consumption, power consumption, etc. related to cost data).
  • Maintenance personnel can manually select a model as the final model according to their needs.
  • the first unit performs power prediction.
  • the maintenance personnel may simultaneously select two or more models as the final model to perform power prediction on the first unit.
  • a data center usually includes more power consumption units.
  • the dispatch management system 200 may also perform power prediction for the new power consumption unit recommendation model.
  • the embodiment of the present application refers to the new power consumption unit as the second power consumption unit.
  • the database of the dispatch management system 200 may store the existing power distribution of the first power consumption unit and the first model (specifically including the type of the first model) for predicting the first power consumption unit and parameter values of the first model, and in some cases hyperparameter values of the first model).
  • the dispatch management system 200 further stores the evaluation index value of the first model, such as the error value of the first model.
  • the dispatch management system 200 can recommend an appropriate power consumption unit for the newly added second power consumption unit according to the power distribution of the existing first power consumption unit in the data center and the first model for predicting the first power consumption unit.
  • the second model performs power prediction, thereby shortening the time for determining the second model, so that the dispatch management system 200 can achieve higher prediction accuracy in a shorter period of time and reduce power costs.
  • the scheduling management system 200 acquires the power distribution of the second power consumption unit newly added in the data center, where the power distribution refers to a time power sequence formed by collecting real power values in at least one collection period. Then the dispatch management system 200 determines the similarity between the power distribution of the second power consumption unit and the power distribution of the existing first power consumption unit in the data center, and determines from the first power consumption unit the similarity with the second power consumption unit according to the similarity. At least one third power consuming unit whose power distribution similarity of the units reaches a preset similarity. Next, the dispatch management system 200 determines a second model for power prediction of the power distribution of the second power consumption unit according to the model for power prediction of the power distribution of the third power consumption unit.
  • the power distribution can be expressed as a dynamic time power sequence.
  • the scheduling management system 200 may determine the similarity of the power distributions by a dynamic time warping (DTW).
  • DTW dynamic time warping
  • the DTW algorithm provides a similarity function (also called a distance function) for time series data. By substituting the time power series into the above similarity function, the similarity of the power distribution can be obtained.
  • the scheduling management system 200 may also use other similarity functions or distance functions to determine the similarity of the power distribution.
  • the dispatch management system 200 may determine the similarity of power distributions according to any one or more of Euclidean distance, Chebyshev distance, Manhattan distance, and the like.
  • the scheduling management system 200 may determine the second model for power prediction of the power distribution of the second power consumption unit according to the model for power prediction of the power distribution of the third power consumption unit, and there may be various implementation manners.
  • the embodiments of the present application exemplarily provide two manners for determining the second model, which will be described in detail below with reference to the accompanying drawings.
  • the scheduling management system 200 determines from the existing first power consumption unit 402 of the data center 400 that the similarity of the power distribution with the newly added second power consumption unit 404 reaches a preset level
  • the at least one third power consuming unit of similarity determines a second model for performing power prediction on the second power consuming unit according to the model for performing power prediction on the third power consuming unit.
  • the dispatch management system 200 may randomly select a model from the models for predicting the power of the third power consuming unit as the second model for predicting the power of the second power consuming unit 404, or select an error value smaller than a preset threshold value
  • the model is used as the second model for power prediction of the second power consuming unit 404 .
  • the scheduling management system 200 determines from the existing first power consumption unit 402 of the data center 400 that the similarity of the power distribution with the newly added second power consumption unit 404 reaches a preset level. At least one third power consumption unit with a similarity degree, and then the model for performing power prediction on the at least one third power consumption unit is added to the model pool of the second power consumption unit 404 . Further, the dispatch management system 200 may also add the prediction model of the power consumption unit whose similarity is ranked as the top S in the at least one third power consumption unit into the model pool.
  • the dispatch management system 200 obtains the evaluation index value of the model in the model pool of the second power consumption unit 404 to predict the power of the second power consumption unit, and obtains the evaluation index value from the second power consumption unit 404 according to the evaluation index value.
  • a second model for power prediction of the second power consuming unit 404 is selected from the model pool of .
  • the power prediction method provided in this embodiment of the present application is implemented based on the models in the model pool.
  • Models in the model pool such as tree models, neural network models, etc., can be obtained through training.
  • the dispatch management system 200 may generate sample data according to the historical power of the first power consumption unit, and then use the sample data to train the initial model, thereby obtaining the model in the model pool of the first power consumption unit.
  • S502 The dispatch management system 200 collects the power distribution of the first power consumption unit in the previous N days.
  • N When the types of models being trained are different, the value of N can be different. For example, when training a DNN model, N can take a value of 10, and when training a random forest model, N can take a value of 5. In addition, the value of N may be different according to the prediction span of the model (ie, the length of the time period in which power prediction is performed). For example, when the model is used to predict the power of one day in the future, N can be 7, and when the model is used to predict the power of the next three days, N can be 15.
  • S504 The dispatch management system 200 resamples the power distribution of the first power consumption unit in the previous N days.
  • the dispatch management system 200 may collect the power (real power value) of the first power consumption unit according to the collection period.
  • the scheduling management system 200 may resample the above historical power distribution, for example, may resample the historical power distribution according to a prediction time interval (specifically, a statistical period).
  • the statistical period may be an integer multiple of the collection period. In some cases, the statistical period may be equal to the acquisition period.
  • the dispatch management system 200 may also determine the minimum electricity price span according to the electricity price curve.
  • the electricity price span refers to the shortest time for electricity price to change from one price to another.
  • the dispatch management system 200 can determine the minimum electricity price span.
  • the dispatch management system 200 may determine the resampling period according to the minimum electricity price span. For example, when the minimum electricity price span is 30 minutes, the dispatching management system may determine the resampling period as 30 minutes.
  • the noise in the original power distribution can be reduced.
  • selecting an appropriate resampling period to resample the original power distribution can effectively reduce the amount of data and improve the iteration speed of the model.
  • S506 The scheduling management system 200 fills in the missing power values in the power distribution.
  • the scheduling management system 200 may also fill in the missing power values in the power distribution. Specifically, the scheduling management system 200 may determine a time window, the time window starts from a time point of a preset duration before the time point when the power value is missing, and ends at the time when the power value is missing, and then according to the time window Mathematical statistics of the power values at different time points, such as mean, median, etc., fill in the missing power values at time points.
  • S508 The dispatch management system 200 performs abnormal point detection on the power distribution, and corrects the power value of the abnormal point.
  • An abnormal point is a point in the power distribution where the power exceeds the normal range.
  • the dispatch management system 200 may perform outlier detection on the power distribution by using an outlier (outlier) detection algorithm.
  • Outlier detection algorithms include statistical hypothesis testing algorithms, local outlier factor (LOF) algorithms, boxplot (interquartile range, IQR) algorithms, and so on.
  • the dispatch management system 200 may construct a corresponding box diagram.
  • the box plot defines 5 basic values, specifically the minimum (minimum, min)), the lower quartile or the first quartile (first quartile, Q1), the median, the median ( median) or second quartile (second quartile, Q2), upper quartile or third quartile (third quartile, Q3), maximum (maximum, max).
  • the interquartile range IQR represents the distance between the lower quartile Q1 and the upper quartile.
  • the lower quartile Q1 is the data sequence (such as the data sequence formed by the power values in the power distribution) sorted from small to large and ranked 25% of the values
  • the median Q2 is the data sequence sorted from small to large
  • the bottom rank is the 50% value
  • the upper quartile is the 75% value after sorting the data series from small to large.
  • the minimum and maximum values in the box plot are not necessarily equal to the minimum and maximum values of the power values in the power distribution, but are determined according to the IQR. Specifically, the minimum value is Q1-1.5IQR, and the maximum value is Q3+1.5IQR. Among them, the abnormal point is the point whose power value is less than Q1-1.5IQR, or the point whose power value is greater than Q3+1.5IQR.
  • the scheduling management system 200 may correct the power value of the abnormal point in a manner similar to filling missing values, for example, using a mathematical statistical value within a time window as the corrected power value.
  • S510 The dispatch management system 200 normalizes the power distribution.
  • the scheduling management system 200 may further normalize the power values at each time point in the power distribution.
  • the scheduling management system 200 performs feature extraction from the power distribution, and generates sample data according to the extracted features.
  • the scheduling management system 200 can extract features by means of feature engineering. Specifically, the scheduling management system 200 may determine a time window, which may be different from the time window for filling missing power values described above, and then obtain mathematical statistics of power values at multiple time points within the time window, such as At least one of extreme value, mean, variance, etc., and the time information corresponding to the obtained power, such as time, the corresponding day order (the day of the week), the month order (the month of the year) , and then generate sample data according to the above mathematical statistics and time information in the time window.
  • a time window which may be different from the time window for filling missing power values described above
  • the scheduling management system 200 may determine a time window, which may be different from the time window for filling missing power values described above, and then obtain mathematical statistics of power values at multiple time points within the time window, such as At least one of extreme value, mean, variance, etc., and the time information corresponding to the obtained power, such as time, the corresponding day order (the day of the week), the month order (
  • the sample data can be expressed as (X, Y), where X includes the time power series in the time window and the extracted features, such as the extreme value, mean, variance and time, day order, and month order of the above power values, and Y is Supervision information, which may be a time power sequence within a time window after the time window.
  • X includes the time power series in the time window and the extracted features, such as the extreme value, mean, variance and time, day order, and month order of the above power values
  • Y is Supervision information, which may be a time power sequence within a time window after the time window.
  • feature extraction may also be performed on the time power series within the time window, and the time power series and the extracted features are used as inputs.
  • the scheduling management system 200 may also directly generate sample data according to a time power sequence of a time window and a time power sequence after the time window. That is, X in the sample data (X, Y) represents the time power sequence of a time window, and Y is the supervision information, and the supervision information is the feature in a time window after the time window.
  • the lengths of the time windows corresponding to X and Y may be equal or unequal.
  • the length of the time window corresponding to X may be greater than the length of the actual window corresponding to Y.
  • the length of the time window corresponding to X may be 3 days, and the length of the time window corresponding to Y may be 1 day.
  • S504 to S510 are optional steps, and in other possible implementation manners of the embodiment of this application, the above S504 to S510 may not be performed.
  • the scheduling management system 200 may also be divided into a training set and a verification set respectively.
  • the scheduling management system 200 may divide the sample data into a training set and a validation set respectively according to a first preset ratio, such as 8:2.
  • the schedule management system 200 may divide the sample data into training sets, validation sets, and test sets, respectively.
  • the scheduling management system 200 may divide the sample data into a training set, a validation set, and a test set, respectively, according to a second preset ratio, eg, 7:2:1.
  • the training set is used to fit the model
  • the validation set is used to verify the model trained based on the sample data in the training set (ie, the training samples), so as to adjust the hyperparameters of the model and the ability of the model such as Preliminary evaluation of prediction accuracy.
  • the test set is used for the generalization ability of the validated model.
  • the scheduling management system 200 After the scheduling management system 200 obtains the data samples, the sample data in the training set, that is, the training samples, can be used as the input of the model pool to perform distributed model training.
  • the scheduling management system 200 can select hyperparameters by using automatic parameter tuning methods such as grid search, particle swarm optimization (PSO) and other optimization algorithms according to the search space of the hyperparameters of the model. and tuning, and select and tune the parameters of the model according to the loss function obtained after the training samples are input into the model, and stop training when the model converges.
  • the scheduling management system 200 can use the sample data in the verification set, that is, the verification samples, to verify the model obtained by training, so as to preliminarily evaluate the accuracy of the model. When the accuracy does not meet the requirements, the scheduling management system 200 can further tune the hyperparameters, and then retrain the model. When the trained model is validated, it can be used for power prediction.
  • the scheduling management system 200 may also first determine the abnormal points in the power distribution when determining the evaluation index value of the model.
  • the power value of the point is corrected.
  • the scheduling management system 200 can detect the abnormal point by the same or similar method as the abnormal point detection method in the model training process, such as using the interquartile range of the box plot to detect the abnormal point, and use the abnormal point as the end point to detect the abnormal point.
  • the mathematical statistics of the power values at multiple time points in the time window corrects the power values of the abnormal points, and then determines the evaluation index value of the model according to the corrected power values, so that the accuracy can be improved.
  • the apparatus 600 includes:
  • a communication unit 602 configured to obtain an evaluation index value of the model in the model pool, where the evaluation index value is used to indicate the accuracy of the model;
  • a prediction unit 604 configured to select a first model to perform power prediction on a first power consumption unit according to the evaluation index value, where the first power consumption unit is any power consumption unit in the data center;
  • a display unit 606, configured to present the result of the power prediction for the first power consumption unit.
  • the prediction unit 604 is specifically configured to:
  • the first model is used to predict the future power distribution of the first power consumption unit, the historical power distribution indicates the power of at least one statistical period before the current moment, and the The future power distribution indicates the power at least one statistical period after the current time.
  • the evaluation index value includes an error value
  • the predicting unit 604 is specifically configured to:
  • a model whose error value satisfies a preset condition is selected from the model pool as the first model.
  • the error value satisfying a preset condition includes:
  • the error value is smaller than a preset threshold; or, the error value is the smallest.
  • the prediction unit 604 is specifically configured to:
  • the first model is selected from the updated model pool according to the updated evaluation index value of the model.
  • the evaluation index value is determined according to at least one of an interval error value and a single-point error value during power prediction of the first power consuming unit in the data center.
  • the communication unit 602 is further configured to:
  • the apparatus 600 also includes:
  • a determining unit configured to determine, from the first power consuming unit, at least one third power consuming unit whose power distribution similarity with the second power consuming unit reaches a preset similarity
  • the determining unit is further configured to determine a second model for predicting the power of the second power unit according to the model for predicting the power of the third power unit.
  • the determining unit is specifically used for:
  • a model for power prediction of the third power consumption unit is determined as a second model for power prediction of the second power consumption unit.
  • the determining unit is specifically used for:
  • a second model for power prediction of the second power consumption unit is selected from the model pool of the second power consumption unit according to the evaluation index value.
  • the apparatus 600 further includes:
  • a generating unit configured to generate training samples according to the historical power distribution of the first power consumption unit
  • a training unit configured to use the training samples to train an initial model to obtain a model in the model pool.
  • the model pool includes two or more models.
  • the first power consuming unit is a collection of electrical equipment
  • the set of electrical equipment includes at least one electrical equipment, at least one electrical equipment in a whole cabinet, or at least one electrical equipment in a computer room. One or more of the electrical equipment.
  • the display unit is specifically used for:
  • a result of the power prediction of the first power consumption unit by at least one model in the model pool is presented through a graphical user interface.
  • the apparatus 600 for power prediction may correspond to executing the methods described in the embodiments of the present application, and the above-mentioned and other operations and/or functions of the respective modules/units of the apparatus 600 for power prediction are in order to implement the methods shown in FIG. 2 , respectively.
  • the corresponding flow of each method in the illustrated embodiment is not repeated here for brevity.
  • the device 700 may be an end-side device such as a notebook computer and a desktop computer, or a computer cluster in a cloud environment or an edge environment, or a combination of end-side devices and devices in a cloud environment and an edge environment.
  • the device 700 is specifically used to implement the function of the apparatus 600 for power prediction in the embodiment shown in FIG. 6 .
  • FIG. 7 provides a schematic structural diagram of an electronic device 700 .
  • the electronic device 70 includes a bus 701 , a processor 702 , a communication interface 703 , a memory 704 , and a display 705 .
  • the processor 702 , the memory 704 , the communication interface 703 , and the display 705 communicate through the bus 701 .
  • the bus 701 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus or the like.
  • PCI peripheral component interconnect
  • EISA extended industry standard architecture
  • the bus can be divided into address bus, data bus, control bus and so on. For ease of presentation, only one thick line is used in FIG. 7, but it does not mean that there is only one bus or one type of bus.
  • the processor 702 may be a central processing unit (CPU). In some embodiments, the processor 702 may also be any processor such as a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP). one or more.
  • CPU central processing unit
  • the processor 702 may also be any processor such as a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP). one or more.
  • GPU graphics processing unit
  • MP microprocessor
  • DSP digital signal processor
  • the communication interface 703 is used for external communication.
  • the communication interface 703 can be used to obtain the evaluation index value of the model in the model pool, or to obtain the power distribution of the second power consumption unit newly added in the data center, and so on.
  • Memory 704 may include volatile memory, such as random access memory (RAM).
  • RAM random access memory
  • the memory 704 may also include non-volatile memory (non-volatile memory), such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (solid state drive) , SSD).
  • non-volatile memory such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (solid state drive) , SSD).
  • Display 705 is an input/output (I/O) device.
  • the device can display electronic files such as images and text on the screen for users to view.
  • the display 705 can be classified into a liquid crystal display (LCD), an organic light emitting diode (OLED) display, and the like.
  • LCD liquid crystal display
  • OLED organic light emitting diode
  • Executable code is stored in the memory 704, and the processor 702 executes the executable code to perform the aforementioned method of power prediction.
  • each unit of the apparatus 600 for power prediction described in the embodiment in FIG. 6 is implemented by software, all the functions of the prediction unit 604 in FIG. 6 are executed.
  • the required software or program code is stored in memory 704 .
  • the communication unit 602 functions through the communication interface 703 .
  • the communication interface 703 is used to obtain the evaluation index value of the model in the model pool, and transmit the evaluation index value of the model to the processor 702 through the bus 701, and the processor 702 executes the program code corresponding to each unit stored in the memory 704, such as executing prediction.
  • the program code corresponding to the unit 604 is used to execute the step of selecting the first model according to the evaluation index value to perform power prediction on the first power consuming unit. Then the processor 702 transmits the result of the power prediction for the first power consumption unit to the display 705 through the bus.
  • the display 705 presents the result of the power prediction for the first electrical unit.
  • the electronic device 700 in the embodiment of the present application may correspond to the apparatus 600 for power prediction described in FIG. 6 in the embodiment of the present application, and the electronic device 700 is configured to implement the method executed by the corresponding subject in the method described in the foregoing FIG. 2 .
  • the operation steps are not repeated here for brevity.
  • the computer program product includes one or more computer instructions.
  • the computer may be a general purpose computer, special purpose computer, computer network, or other programmable device.
  • the computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be retrieved from a website, computer, training device, or data Transmission from the center to another website site, computer, training facility or data center via wired (eg coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (eg infrared, wireless, microwave, etc.) means.
  • wired eg coaxial cable, fiber optic, digital subscriber line (DSL)
  • wireless eg infrared, wireless, microwave, etc.
  • the computer-readable storage medium may be any available medium that can be stored by a computer, or a data storage device such as a training device, a data center, or the like that includes an integration of one or more available media.
  • the usable media may be magnetic media (eg, floppy disks, hard disks, magnetic tapes), optical media (eg, DVDs), or semiconductor media (eg, solid state disks (SSDs)), and the like.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Power Engineering (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Educational Administration (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Human Computer Interaction (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Supply And Distribution Of Alternating Current (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请提供了一种功率预测的方法，包括：获取模型池中模型的评价指标值，该评价指标值用于指示模型的精度，根据评价指标值选择第一模型对第一用电单元进行功率预测，呈现对第一用电单元的功率预测的结果。该方法采用模型的评价值对模型池中的模型进行在线评估，根据评估结果选择合适的模型对第一用电单元进行功率预测，而不是始终采用单一模型对第一单元进行功率预测，如此可以适应第一用电单元的功率变化，提高预测精度，满足业务需求。

Description

功率预测的方法、装置及设备 技术领域
本申请涉及计算机技术领域，尤其涉及一种功率预测的方法、装置及设备。
背景技术
数据中心(data center，DC)是全球协作的特定设备网络，用来在因特网基础设施上传递、加速、展示、计算、存储数据信息。数据中心在运行过程中会消耗大量电能。为了减少数据中心的电费开支，降低数据中心运营成本，可以为数据中心提供一储能系统。该储能系统可以在低电价时充电，在高电价时放电，从而为数据中心供电。
为了最大化利用储能系统，调度管理系统通常需要根据数据中心未来一段时间内的功率，结合电价曲线和电池状态等信息提供合理的充放电策略。基于此，如何准确地预测数据中心的功率成为业界重点关注的问题。
发明内容
本申请提供了一种功率预测的方法。该方法通过模型的评价值对模型池中的模型进行在线评估，根据评估结果选择合适的模型对第一用电单元进行功率预测，由此可以适应第一用电单元的功率变化，提高预测精度。本申请还提供了上述方法对应的装置、设备、计算机可读存储介质以及计算机程序产品。
第一方面，本申请提供了一种功率预测的方法。该方法可以应用于调度管理系统，具体是调度管理系统中的功率预测模块。调度管理系统(例如，功率预测模块)以数据中心中的一个用电单元为粒度对数据中心进行功率预测。为了便于描述，下面以对数据中心中的任意一个用电单元即第一用电单元进行功率预测为例，对本申请的功率预测的方法进行说明。
具体地，调度管理系统获取模型池中模型的评价指标值，该评价指标值用于指示所述模型的精度，然后调度管理系统根据所述评价指标值选择第一模型对第一用电单元进行功率预测，接着调度管理系统呈现对所述第一用电单元的所述功率预测的结果。
该方法采用模型的评价值对模型池中的模型进行在线评估，根据评估结果选择合适的模型对第一用电单元进行功率预测，而不是始终采用单一模型对第一单元进行功率预测，如此可以适应第一用电单元的功率变化，提高预测精度，满足业务需求。
进一步地，不同用电单元如不同整机柜用电设备运行的业务类型可以是不同的，该方法以用电单元为粒度进行功率预测，可以使得对每个用电单元的功率预测均具有较高的预测精度，提高对数据中心整体进行功率预测的精度。
在一些可能的实现方式中，所述调度管理系统可以根据所述评价指标值从所述模型池中选择第一模型，然后根据所述第一用电单元的历史功率分布，利用所述第一模型预测所述第一用电单元的未来功率分布。
其中，所述历史功率分布指示当前时刻之前至少一个统计周期的功率，所述未来功率分布指示当前时刻之后至少一个统计周期的功率。该统计周期等于对功率分布进行预测的时间间隔。具体地，统计周期可以为采集真实功率值的采集周期。当然，在一些实施例中，统计 周期也可以是采集周期的整数倍。
通过该方法，可以实现对第一用电单元在不同统计周期的功率预测，并且能够适应功率变化，具有较好的预测精度。
在一些可能的实现方式中，所述评价指标值包括误差值。误差值具体可以根据功率的预测值和真实值确定。调度管理系统在选择第一模型时，可以根据所述误差值从所述模型池中选择所述误差值满足预设条件的模型为第一模型。由此选择出预测精度较高的模型对第一用电单元进行功率预测，使得对第一单元的功率预测保持较高的精度。
在一些可能的实现方式中，所述误差值满足预设条件包括：所述误差值小于预设阈值；或者，所述误差值最小。
具体地，调度管理系统可以比较模型的误差值和预设阈值，选择误差值小于预设阈值的模型为第一模型。当误差值小于预设阈值的模型包括多个时，调度管理系统可以从多个模型中随机选择一个模型作为第一模型，或者选择误差值最小的模型作为第一模型。
当模型池中模型的误差值均不小于预设阈值时，若模型池中的模型距离上次更新的时间不超过预设时间，由于对模型进行频繁更新会造成大量计算资源的消耗，而且历史数据变化较小时，模型精度不会明显改善，此时，调度管理系统可以直接选择误差值最小的模型为第一模型。
通过上述方法可以选择出合适的第一模型对第一用电单元进行功率预测，从而使得对第一用电单元的功率预测具有较好的精度。
在一些可能的实现方式中，调度管理系统可以根据所述评价指标值更新所述模型池中的模型，然后根据更新后的所述模型的评价指标值从更新后的所述模型池中选择第一模型。具体地，模型池中模型的评价指标值(例如误差值)均不小于预设阈值时，表明模型的精度不能满足要求，并且至少有一个模型距离上次更新时间超过预设时间，如超过M天，则历史数据变化相对较大，模型管理系统可以根据更新的历史数据更新模型。如此，调度管理系统根据更新的模型的评价指标值从更新后的模型池中选择第一模型对第一用电单元进行功率预测。一方面可以获得较好的预测精度，另一方面可以实现预测精度和计算资源消耗的均衡。
在一些可能的实现方式中，所述评价指标值根据对所述数据中心中第一用电单元的功率预测时的区间误差值和单点误差值中的至少一种确定。
具体地，调度管理系统可以对第一用电单元在一段时间内的功率分布，具体是第一用电单元在至少一个统计周期的功率进行预测。其中，统计周期可以是功率的采集周期。例如，每隔10min采集一次功率时，采集周期即为10min，相应地统计周期可以为10min。基于此，调度管理系统对第一用电单元在多个统计周期的功率进行预测时，可以根据模型对第一用电单元在每个统计周期的功率预测的偏差值(例如是偏差值的绝对值)确定模型的误差值。该误差值即为上述单点误差值。
考虑到不同统计周期对应的电价可以是相同的，如果在同一电价的时间区间的多个统计周期中，模型的预测值存在偏高和偏低的情况，则调度管理系统可以将属于同一时间区间的偏差值进行求和，从而确定在该区间内的误差值，接着根据模型在各个时间区间的误差值确定该模型的误差值。该误差值也称作区间误差值。
其中，区间误差值通常小于单点误差值，评价指标值中包括区间误差值时可以减少模型更新次数，避免模型频繁更新导致消耗大量的计算资源。
在一些可能的实现方式中，调度管理系统还可以根据数据中心中已有的第一用电单元的功率分布和对该第一用电单元进行预测的第一模型，为新加入的第二用电单元推荐合适的第 二模型进行功率预测，由此缩短确定第二模型的时间，使得调度管理系统能够在较短时间内达到较高的预测精度，降低电力成本。
具体地，调度管理系统获取所述数据中心中新加入的第二用电单元的功率分布，然后从所述第一用电单元中确定与所述第二用电单元的功率分布的相似度达到预设相似度的至少一个第三用电单元，接着根据对所述第三用电单元进行功率预测的模型确定用于对所述第二用电单元进行功率预测的第二模型。
在一些可能的实现方式中，调度管理系统可以直接将对所述第三用电单元进行功率预测的模型确定为用于对所述第二用电单元进行功率预测的第二模型。由此大幅缩短确定第二模型的时间，使得对第二用电单元的功率预测能够在较短时间达到较高的精度。
在一些可能的实现方式中，调度管理系统可以将对所述至少一个第三用电单元进行功率预测的模型加入所述第二用电单元的模型池，获取所述第二用电单元的模型池中模型对所述第二用电单元进行功率预测的评价指标值，根据所述评价指标值从所述第二用电单元的模型池中选择用于对所述第二用电单元进行功率预测的第二模型。模型池中的模型是与第二用电单元亲和的模型，根据评价指标值从模型池中选择模型，可以使得对第二用电单元的功率预测能够在较短时间达到较高的精度。
在一些可能的实现方式中，调度管理系统还可以根据所述第一用电单元的历史功率分布生成训练样本，然后利用所述训练样本对初始模型进行训练得到所述模型池中的模型，进而用于对第一单元进行功率预测，保障预测精度。
在一些可能的实现方式中，所述模型池中包括两种或两种以上模型。例如，模型池中可以包括树模型、神经网络模型、自回归模型和简单模型中的一种或多种。当模型池中包括一种模型时，具体可以包括该种模型下的不同模型。例如模型池中包括树模型一种模型时，可以包括极端梯度提升模型、随机森林模型。又例如，模型池中包括神经网络模型一种模型时，具体可以包括深度神经网络模型和长短期记忆网络模型。
简单模型具体是一种以历史功率的数学统计值为未来功率预测值的模型。其中，数学统计值可以是历史功率的加权平均值、算术平均值、中位值、极大值、极小值等中的任意一种。
由于简单模型仅仅是进行数学统计，无需训练，因此，在模型池中增加简单模型，可以解决数据中心中添加新的用电单元时，由于历史数据较少无法训练树模型、神经网络模型、自回归模型等模型导致的冷启动问题。
在一些可能的实现方式中，所述第一用电单元为用电单元集合，该用电单元集合包括至少一个用电设备，至少一个整机柜的用电设备，或者至少一个机房的用电设备。该方法以一个用电单元为预测粒度进行功率预测，可以使得对每个用电单元的功率预测均具有较高的预测精度，例如，该用电单元可以是独立的用电设备(如服务器)、也可以是单个机柜包括的所有或部分用电设备，还可以是整个机房的所有或部分用电设备。利用包括多个模型的模型池对不同用电设备集合的预测，使得预测粒度可以根据业务需求进行动态调整，以此提高对数据中心整体进行功率预测的精度。
在一些可能的实现方式中，调度管理系统还可以通过图形用户界面呈现所述模型池中至少一个模型对所述第一用电单元的所述功率预测的结果。由此可以为用户进行决策提供帮助。
第二方面，本申请提供了一种功率预测的装置，该装置包括用于执行第一方面或第一方面任意一种可能实现方式中的功率预测的方法的各个单元。
第三方面，本申请提供一种设备，所述设备包括处理器和存储器。所述处理器、所述存储器进行相互的通信。所述处理器用于执行所述存储器中存储的指令，以使得设备执行如第 一方面或第一方面的任一种实现方式中的功率预测的方法。
第四方面，本申请提供一种计算机可读存储介质，所述计算机可读存储介质中存储有指令，所述指令指示设备执行上述第一方面或第一方面的任一种实现方式所述的功率预测的方法。
第五方面，本申请提供了一种包含指令的计算机程序产品，当其在设备上运行时，使得设备执行上述第一方面或第一方面的任一种实现方式所述的功率预测的方法。
本申请在上述各方面提供的实现方式的基础上，还可以进行进一步组合以提供更多实现方式。
附图说明
图1为本申请实施例提供的一种调度管理系统对储能系统进行管理的系统架构图；
图2为本申请实施例提供的一种功率预测的方法的流程图；
图3为本申请实施例提供的一种呈现功率预测的结果的示意图；
图4A为本申请实施例提供的一种确定第二模型的示意图；
图4B为本申请实施例提供的一种确定第二模型的示意图；
图5为本申请实施例提供的一种生成样本数据的流程图；
图6为本申请实施例提供的一种功率预测的装置的结构示意图；
图7为本申请实施例提供的一种设备的结构示意图。
具体实施方式
为了便于理解，首先对本申请实施例中所涉及到的一些技术术语进行介绍。
数据中心是全球协作的特定设备网络，用来在因特网基础设施上传递、加速、展示、计算、存储数据信息。数据中心中用于传递、加速、展示、计算、存储数据信息的设备如服务器、交换机等可以部署在整机柜中。为了保障数据中心的正常运行，通常需要供电系统为数据中心包括的各种设备供电。
供电系统可以包括市电供电系统。其中，市电供电系统用于将发电站产生的电能通过电网直接传输至用电方(如数据中心)，从而为用电方供电。然而，电力公司在不同时间段提供的电价可以是不同的。为了减少数据中心的电费开支，降低运营成本，数据中心所有方可以构建储能系统，通过低电价时间控制储能系统充电，在高电价时间段控制储能系统放电，以为数据中心供电。也即供电系统还可以包括储能系统。
储能系统包括至少一个储能设备。该储能设备可以是支持充电和放电的设备，例如，可以是储能电池。其中，储能电池可以包括锂电池、镍氢电池、铅酸电池、钠硫电池等不同类型电池中的至少一种或多种。
由于储能系统的容量是有限的，而且市电供电系统在不同时间段的电价是不同的，因此可以根据数据中心在未来一段时间内的功率使用情况，结合电价曲线和储能设备的状态等信息确定充放电策略。储能系统中的储能设备按照该充放电策略进行充电和放电，进而充分利用储能系统能力，减少数据中心电力成本。
其中，数据中心在未来一段时间的功率使用情况可以根据数据中心在历史时间的功率使用情况进行预测。对数据中心的功率预测以及确定充放电策略的过程可以由储能系统的调度管理系统实现。
如图1所示，储能系统100包括至少一个储能设备102，该储能设备102具体可以是储能电池。调度管理系统200用于对储能系统进行调度管理，如调度储能系统100在一些时间段充电，在另一些时间段放电，从而为数据中心供电。
具体地，调度管理系统200包括通信与控制模块202、功率预测模块204和策略制定模块206。通信与控制模块202用于与储能系统100、数据中心进行通信，对储能系统100进行控制，从而实现对储能系统100的调度管理。功率预测模块204用于对数据中心进行功率预测，例如预测数据中心在未来一段时间的功率分布。策略制定模块206用于根据预测的功率、储能设备102的状态、电价曲线制定充放电策略。通信与控制模块202可以获取该充放电策略下发至储能系统100。
在一些可能的实现方式中，调度管理系统200还可以包括存储模块208。该存储模块208具体可以是数据库。存储模块208可以用于存储数据中心的功率如历史功率、储能系统100中储能设备102的状态等信息中的至少一种。其中，储能设备102的状态包括健康状态(state of health，SoH)、荷电状态(state of charge，SoC)等中的一种或多种。
SoC也称作剩余电量，用于表征储能设备102使用一段时间或长期搁置不用后的剩余容量与其完全充电状态的容量的比值，该比值通常可以采用百分数表示。其取值范围为0～1，当SoC＝0时，表示电池放电完全，当SoC＝1时表示电池完全充满。SoH指的是储能设备102的健康状态，主要用于表征电池老化程度。
如此，功率预测模块204可以从上述存储模块208中获取数据中心的历史功率，以用于预测数据中心的未来功率。策略制定模块206用于从上述存储模块208获取储能设备102的状态信息，以用于制定充放电策略。
还需要说明的是，储能系统100中还可以包括储能设备管理模块104。调度管理系统200的通信与控制模块202可以通过储能设备管理模块104与储能设备102进行通信，并对储能设备102进行控制。其中，储能设备102可以是储能电池，储能设备管理模块104可以是电池管理单元(battery management system)。
目前，调度管理系统200(例如是功率预测模块204)在对数据中心进行功率预测时，主要是采用单一模型进行预测。然而，受到数据中心中用电设备运行业务的影响，用电设备的功率有较大概率随着时间变化发生较大的变化，原有模型如果不能适应该变化，就会导致预测精度大幅下降，进而导致调度管理系统200(例如是策略制定模块206)制定的充放电策略无法满足需求。
有鉴于此，本申请实施例提供了一种功率预测的方法。在该方法中，调度管理系统200(例如，功率预测模块204)以数据中心中的一个用电单元为粒度对数据中心进行功率预测。其中，用电单元可以是用电设备的集合，该用电设备的集合包括至少一个用电设备，如一个或多个服务器。在一些实施例中，该用电设备的集合也可以是至少一个整机柜的用电设备，如一个或多个整机柜中所有或部分服务器等。在另一些实施例中，该用电单元的集合还可以是至少一个机房的用电设备，如一个或多个机房的所有或部分服务器。
为了便于描述，本申请实施例以对第一用电单元的预测过程进行示例说明。第一用电单元可以是数据中心中任一用电单元。调度管理系统200可以获取模型池中模型的评价指标值，该评价指标值可以用于指示模型的精度，然后调度管理系统200根据评价指标值选择第一模型对第一用电单元进行功率预测，接着呈现对第一用电单元的功率预测的结果。
该方法采用模型的评价值对模型池中的模型进行在线评估，根据评估结果选择合适的模型对第一用电单元进行功率预测，而不是始终采用单一模型对第一单元进行功率预测，如此 可以适应第一用电单元的功率变化，提高预测精度，满足业务需求。
进一步地，不同用电单元如不同整机柜用电设备运行的业务类型可以是不同的，该方法以用电单元为粒度进行功率预测，使得对每个用电单元的功率预测均具有较高的预测精度，提高对数据中心整体进行功率预测的精度。
本申请实施例提供的调度管理系统200可以是软件模块，该软件模块可以部署在硬件设备中，从而对外提供服务。其中，调度管理系统200具有多种部署方式，下面分别对多种部署方式进行详细说明。
在一些可能的实现方式中，调度管理系统200可以部署在云计算集群中。其中，调度管理系统200的各个模块可以集中部署在云计算集群的一个云计算设备(如云服务器)中，也可以分布式地部署在云计算集群的不同云计算设备中。当调度管理系统200部署在云计算集群中时，本申请实施例提供的功率预测的方法可以以云服务的方式提供给用户。
在另一些可能的实现方式中，调度管理系统200可以部署在本地计算设备中。本地计算设备是指本地设备包括处于用于用户直接控制之下的计算设备，例如可以是台式机、笔记本电脑等终端，或者是本地服务器。
储能系统100中的储能设备102可以为硬件设备。该硬件设备与调度管理系统200连接。为了便于供电，储能设备102可以部署在信息技术(information technology，IT)基础设施柜，简称为IT柜中。当储能系统100包括储能设备管理模块104时，储能设备管理模块104可以随着储能设备102一同部署在IT柜中。
需要说明的是，图1仅仅是对调度管理系统200、储能系统100的部署方式进行示例说明，在本申请实施例其他可能的实现方式中，调度管理系统200、储能系统100也可以采用其他部署方式。
接下来，将从调度管理系统200的角度对本申请实施例提供的功率预测的方法进行介绍。
参见图2所示功率预测的方法的流程图，该方法包括：
S202：调度管理系统200获取模型池中模型的评价指标值。
模型池具体是指包括至少一个模型的逻辑池。模型池中的模型用于对用电单元进行功率预测。数据中心的每个用电单元对应一个模型池。不同用电单元的模型池可以相同，也可以是不同的。
在一些可能的实现方式中，模型池可以包括两种或两种以上的模型。例如，模型池可以包括树模型、神经网络模型、自回归模型和简单模型等不同种类模型中的一种或多种。当模型池包括上述模型中的一种模型时，模型池具体可以包括该种类下的不同模型。
例如，模型池包括树模型一种模型时，该模型池具体可以包括树模型中的不同模型，如极端梯度提升(extreme gradient boosting，xgboost)模型和随机森林(random forest)模型。又例如，模型池包括神经网络模型一种模型时，该模型池具体可以包括神经网络模型中的不同模型，如深度神经网络模型(deep neural networks，DNN)和长短期记忆网络(long short term memory，LSTM)。
简单模型具体是一种以历史功率的数学统计值为未来功率预测值的模型。其中，数学统计值可以是历史功率的加权平均值、算术平均值、中位值、极大值、极小值等中的任意一种。为了便于理解，下面结合具体示例进行说明。在一个示例中，简单模型可以将历史三天的功率的加权平均值作为未来一天的功率的预测值。
由于简单模型仅仅是进行数学统计，无需训练，因此，在模型池中增加简单模型，可以解决数据中心中添加新的用电单元时，由于历史数据较少无法训练树模型、神经网络模型、自回归模型等模型导致的冷启动问题。
每个模型具有评价指标值。该评价指标值具体用于指示所述模型的精度。在一些实施例中，评价指标值可以是模型进行功率预测的误差值。误差值具体可以根据功率的预测值和真实值确定。
调度管理系统200可以对第一用电单元在一段时间内的功率分布，具体是第一用电单元在至少一个统计周期的功率进行预测。其中，统计周期可以是功率的采集周期。例如，每隔10min采集一次功率时，采集周期即为10min，相应地统计周期即为10min。基于此，调度管理系统200对第一用电单元在多个统计周期的功率进行预测时，可以根据模型对第一用电单元在每个统计周期的功率预测的偏差值(例如是偏差值的绝对值)确定模型的误差值。该误差值也称作单点误差值，具体计算公式如下：
Figure PCTCN2021106964-appb-000001
其中，error sin表征单点误差值。N是指需要预测的统计周期的数量，例如需要预测一天的功率时，而统计周期为10分钟时，则N＝(60÷10)×24＝144。y prei是指模型对用电单元在第i个统计周期的功率的预测值，y i是指用电单元在第i个统计周期的真实功率值(也称作实际功率值)。
考虑到不同统计周期对应的电价可以是相同的，如果在同一电价的时间区间的多个统计周期中，模型的预测值存在偏高和偏低的情况，则调度管理系统200可以将属于同一时间区间的偏差值进行求和，从而确定在该区间内的误差值，接着根据模型在各个时间区间的误差值确定该模型的误差值。该误差值也称作区间误差值，具体计算公式如下：
Figure PCTCN2021106964-appb-000002
其中，error range表征区间误差值。M为电价数量，例如不同时间区间的电价可以为0.195、0.58和0.885，则M可以取值为3。y prej是指模型对用电单元在时间区间内第j个统计周期的功率的预测值，y j是指用电单元在时间区间内第j个统计周期的真实功率值。
Figure PCTCN2021106964-appb-000003
表征指用电单元在时间区间内真实功率值的平均值。Q表征时间区间内多个统计周期的数量。
需要说明的是，当一个电价对应多个不连续的时间区间，例如一个电价分别对应如下时间区间[0:00，7:00)、[13:00，14:00)、[23:00，24:00)时，调度管理系统200在确定误差值时还可以将多个不连续的时间区间合并。对应地，Q可以是多个不连续的时间区间内多个统计周期的数量之和。
基于上述公式(1)和公式(2)可知，区间误差值通常小于单点误差值，评价指标值中包括区间误差值时可以减少模型更新次数，避免模型频繁更新导致消耗大量的计算资源。
在一些可能的实现方式中，调度管理系统200还可以根据上述单点误差值和区间误差值，共同确定误差值，具体如下所示：
Figure PCTCN2021106964-appb-000004
其中，error表征误差值，k 1和k 2分别表征单点误差值和区间误差值的权重。k 1与k 2的和值为1。
S204：调度管理系统200根据所述评价指标值选择第一模型对第一用电单元进行功率预测。
具体地，调度管理系统200可以根据评价指标值从模型池中选择精度满足要求的模型为第一模型，根据第一用电单元的历史功率分布，利用第一模型预测第一用电单元的未来功率分布。其中，历史功率分布指示当前时刻之前至少一个统计周期的功率，未来功率分布指示当前时刻之后至少一个统计周期的功率。需要说明的是，历史功率分布对应的统计周期的数量可以大于或等于未来功率分布对应的统计周期的数量。例如，调度管理系统200可以利用历史三天的功率分布预测未来一天的功率分布。
由于精度可以通过评价指标值，例如误差值进行表征。因此，调度管理系统200可以选择误差值满足预设条件的模型作为第一模型。具体地，调度管理系统200可以将模型的误差值与预设阈值进行比较，从而选择出误差值满足预设条件的模型为第一模型。例如，调度管理系统200可以选择误差值小于预设阈值的模型为第一模型。
在一些可能的实现方式中，误差值满足预设条件的模型还可以是误差值最小的模型。具体地，当所有模型的误差值均不小于预设阈值，并且距离模型上次更新时间并未超过预设时间时，由于对模型进行频繁更新会造成大量计算资源的消耗，而且历史数据变化较小时，模型精度不会明显改善，因此，调度管理系统200可以选择误差值最小的模型作为第一模型。第一模型对第一用电单元进行功率预测的结果可以提供给调度管理系统200中的策略制定模块206进行充放电策略制定。
其中，预设时间可以根据经验值设置，例如预设时间可以设置为L天。需要说明的是，针对不同模型，L取值可以是相同的，也可以是不同的，调度管理系统200可以根据实际情况进行设置。
在一些可能的实现方式中，调度管理系统200可以对模型池中的模型进行更新，然后根据更新后的模型的评价指标值从更新后的模型池中选择第一模型对第一用电单元进行功率预测。
具体地，模型池中模型的评价指标值(例如误差值)均不小于预设阈值时，表明模型的精度不能满足要求，并且至少有一个模型距离上次更新时间超过预设时间，如超过M天，则历史数据变化相对较大，模型管理系统200可以根据更新的历史数据更新模型。
进一步地，模型管理系统200还可以将模型类型、更新时间、更新后的模型的超参值、更新后的模型的参数值、更新后的模型的评价指标值(如误差值)存储在数据库中，以便后续使用。
S206：调度管理系统200呈现对所述第一用电单元的所述功率预测的结果。具体地，调度管理系统200可以通过图形用户界面(graphical user interface，GUI)向用户呈现模型池中至少一个模型对第一用电单元的功率预测的结果。其中，至少一个模型可以包括上述S202中确定的第一模型。为了便于用户查看各个模型预测结果的差异，调度管理系统200也可以通过GUI向用户呈现模型池中各个模型对第一用电单元的功率预测的结果。例如，调度管理系统200可以通过GUI向用户呈现模型池中各个模型预测的第一用电单元在未来一天的功率分布。
下面结合附图对呈现功率预测的结果进行示例说明。
参见图3所示的呈现功率预测的结果的界面示意图，如图3所示，界面300中呈现有第一模型(该示例中可以是DNN模型)对第一用电单元(该示例中可以是机柜4)在一天中的功率分布，具体如图3中的曲线301所示。进一步地，界面300中还承载有用电单元选择控件302、时间选择控件303、模型选择控件304，用户可以通过上述控件以增减方式或者下拉选择方式选择其他用电单元，其他时间，和/或其他模型，以便在界面300中呈现其他用电单元的功率分布，用电单元在其他时间的功率分布，和/或其他模型预测的用电单元的功率分布。在一些实现方式中，用户可以通过模型选择控件304选择模型池中的所有模型，以便在界面300中显示模型池中的各模型预测的功率分布。
在一些可能的实现方式中，界面300中还可以承载实际功率显示控件305、策略显示控件306、电量显示控件307、电价曲线显示控件308、实际收益显示控件309、理想收益显示控件310中的至少一种。
其中，实际功率显示控件305被触发时，例如被用户选中时，调度管理系统200可以在界面300中显示用电单元的实际功率，即真实功率值，具体可以如图3中的曲线311所示。
策略显示控件306被触发时，调度管理系统200在界面300中显示该调度管理系统200制定的充放电策略，具体可以如图3中的曲线312所示。电量显示控件307倍触发时，调度管理系统200可以在界面300中显示储能系统100的电量，具体可以如图3中的曲线314所示。电价曲线显示控件308被触发时，调度管理系统200可以在界面300中显示电价曲线，具体可以如图3中的曲线313所示。
类似地，实际收益显示控件309被触发时，调度管理系统200在界面300中显示通过储能系统100充放电，为数据中心供电所获得的实际收益，该收益具体如图3中的曲线315所示。理想收益显示控件310被触发时，调度管理系统200在界面300中显示通过储能系统100充放电，为数据中心供电在理论上能够获得的收益，该收益具体如图3中的曲线316所示。
基于上述内容描述，本申请实施例提供了一种功率预测的方法。在该方法中，调度管理系统200可获取模型池中模型的评价指标值，根据评价指标值指示的模型精度从模型池中选择第一模型，对第一用电单元进行功率预测，接着呈现对第一用电单元的功率预测的结果。由于在第一用电单元运行过程中，采用模型的评价值对模型池中的模型进行在线评估，根据评估结果选择合适的模型对第一用电单元进行功率预测，而不是始终采用单一模型对第一单元进行功率预测，因而能够适应第一用电单元的功率变化，提高了预测精度，满足了业务需求。
作为一种可能的实施方式，除了图3所示的功率预测结果外，还可以将两种或两种以上模型预测结果呈现在界面300中，具体可以以不同页签或同一页面的不同区域分别显示每个模型预测结果，按照不同维度(例如，功率消耗、耗电等涉及成本数据)在界面显示各个模型的数据，维护人员可以根据其需求人为选择一个模型作为最终模型，由该最终模型对第一单元进行功率预测。可选地，维护人员也可以同时选择两个或多个模型作为最终模型对第一单元进行功率预测。
进一步地，数据中心中通常包括较多的用电单元。当有新的用电单元加入数据中心时，调度管理系统200还可以为该新的用电单元推荐模型进行功率预测。为了便于描述，本申请实施例将新的用电单元称为第二用电单元。
在一些可能的实现方式中，调度管理系统200的数据库中可以存储已有的第一用电单元的功率分布和对该第一用电单元进行预测的第一模型(具体包括第一模型的类型和第一模型 的参数值，在有些情况下还包括第一模型的超参值)。在一些实施例中，调度管理系统200还存储有上述第一模型的评价指标值，如第一模型的误差值。
基于此，调度管理系统200可以根据数据中心中已有的第一用电单元的功率分布和对该第一用电单元进行预测的第一模型，为新加入的第二用电单元推荐合适的第二模型进行功率预测，由此缩短确定第二模型的时间，使得调度管理系统200能够在较短时间内达到较高的预测精度，降低电力成本。
具体地，调度管理系统200获取数据中心中新加入的第二用电单元的功率分布，该功率分布是指在至少一个采集周期采集真实功率值所形成的时间功率序列。然后调度管理系统200确定第二用电单元的功率分布与数据中心中已有的第一用电单元的功率分布的相似度，根据该相似度从第一用电单元中确定与第二用电单元的功率分布的相似度达到预设相似度的至少一个第三用电单元。接着调度管理系统200根据对第三用电单元的功率分布进行功率预测的模型确定用于对第二用电单元的功率分布进行功率预测的第二模型。
其中，功率分布可以表示为动态的时间功率序列。为此，调度管理系统200可以动态时间规整算法(dynamic time warping，DTW)确定功率分布的相似度。DTW算法提供了一种针对时间序列数据的相似函数(也称作距离函数)，通过将时间功率序列代入上述相似函数，可以获得功率分布的相似度。
在一些可能的实现方式中，调度管理系统200还可以采用其他相似函数或距离函数确定功率分布的相似度。例如，调度管理系统200可以根据欧式距离、切比雪夫距离、曼哈顿距离等中的任意一种或多种确定功率分布的相似度。
其中，调度管理系统200根据对第三用电单元的功率分布进行功率预测的模型确定用于对第二用电单元的功率分布进行功率预测的第二模型可以有多种实现方式。本申请实施例示例性地提供了两种确定第二模型的方式，下面结合附图对这两种方式进行详细说明。
第一种实现方式，如图4A所示，调度管理系统200从数据中心400已有的第一用电单元402中确定与新加入的第二用电单元404的功率分布的相似度达到预设相似度的至少一个第三用电单元，根据对第三用电单元进行功率预测的模型确定对第二用电单元进行功率预测的第二模型。其中，调度管理系统200可以从对第三用电单元进行功率预测的模型中随机选择一个模型作为对第二用电单元404进行功率预测的第二模型，或者是选择一个误差值小于预设阈值的模型作为对第二用电单元404进行功率预测的第二模型。
第二种实现方式，如图4B所示，调度管理系统200从数据中心400已有的第一用电单元402中确定与新加入的第二用电单元404的功率分布的相似度达到预设相似度的至少一个第三用电单元，然后将对所述至少一个第三用电单元进行功率预测的模型加入所述第二用电单元404的模型池。进一步地，调度管理系统200也可以是将至少一个第三用电单元中相似度排序为前S的用电单元的预测模型加入模型池。接着调度管理系统200获取所述第二用电单元404的模型池中模型对所述第二用电单元进行功率预测的评价指标值，根据所述评价指标值从所述第二用电单元404的模型池中选择用于对所述第二用电单元404进行功率预测的第二模型。
本申请实施例提供的功率预测的方法是基于模型池中的模型实现的。模型池中的模型，例如树模型、神经网络模型等可以通过训练得到。具体地，调度管理系统200可以根据第一用电单元的历史功率生成样本数据，然后利用样本数据对初始模型进行训练，从而得到第一用电单元的模型池中的模型。
首先，结合附图对调度管理系统200生成样本数据的过程进行介绍。
参见图5所示的生成样本数据的流程图，具体包括如下步骤：
S502：调度管理系统200采集第一用电单元在前N天的功率分布。
当训练的模型类型不同时，N的取值可以是不同的。例如，训练DNN模型时，N可以取值为10，训练随机森林模型时，N可以取值为5。此外，根据模型的预测跨度(即进行功率预测的时间段的长度)不同，N的取值也可以是不同的。例如，模型用于预测未来一天的功率时，N可以取值为7，模型用于预测未来三天的功率时，N可以取值为15。
S504：调度管理系统200对第一用电单元在前N天的功率分布进行重采样。
调度管理系统200可以按照采集周期采集第一用电单元的功率(真实功率值)。当采集周期较小时，例如为5秒时，调度管理系统200获得大量的数据。为此，调度管理系统200可以对上述历史功率分布进行重采样，例如可以按照预测时间间隔(具体为统计周期)对历史功率分布进行重采样。其中，统计周期可以是采集周期的整数倍。在有些情况下，统计周期可以等于采集周期。
考虑到阶梯电价，调度管理系统200还可以根据电价曲线，确定最小电价跨度。电价跨度是指电价由一个价格变为另一个价格的最短时间。当电价曲线中存在三种或三种以上的电价时，调度管理系统200可以确定出最小电价跨度。调度管理系统200可以根据最小电价跨度确定重采样的周期。例如，最小电价跨度为30min时，调度管理系统可以将重采样的周期确定为30min。
通过对原始的功率分布(S502中采集的功率分布)进行重采样，可以降低原始的功率分布中的噪声。并且，选择合适的重采样的周期对原始的功率分布进行重采样可以有效减少数据量，提高模型的迭代速度。
S506：调度管理系统200填充功率分布中缺失的功率值。
考虑到功率分布中可能存在个别时间点的功率值缺失的情况，调度管理系统200还可以填充功率分布中缺失的功率值。具体地，调度管理系统200可以确定出一个时间窗口，该时间窗口以发生功率值缺失的时间点之前预设时长的时间点为起点，以发生功率值缺失的时间点为终点，然后根据时间窗口内不同时间点的功率值的数学统计值，如平均值、中位值等填充缺失的时间点的功率值。
S508：调度管理系统200对功率分布进行异常点检测，并对异常点的功率值进行修正。
异常点是指功率分布中功率超出正常范围的点。具体地，调度管理系统200可以利用异常点(离群点)检测算法对功率分布进行异常点检测。异常点检测算法包括统计假设检验算法、局部异常因子(local out factor，LOF)算法、箱型图(boxplot)的四分位距(interquartile range,IQR)算法等等。
为了便于理解，下面以箱型图的四分位距为例进行说明。针对一个功率分布，调度管理系统200可以构建对应的箱型图。该箱型图中定义了5个基本数值，具体为最小值(minimum，min))，下四分位数或第一个四分位数(first quartile，Q1)，中值、中位数(median)或第二个四分位数(second quartile,Q2)，上四分位数或第三个四分位数(third quartile,Q3)，最大值(maximum，max)。四分位距IQR表示下四分位数Q1和上四分位数的间距。
其中，下四分位数Q1为数据序列(如功率分布中的功率值所形成的数据序列)由小至大排序后排名为25％的数值，中位数Q2为数据序列由小至大排序后排名为50％的数值，上四分位数为数据序列由小至大排序后排名为75％的数值。
需要说明的是，箱型图中的最小值和最大值不一定等于功率分布中功率值的最小值和最大值，而是根据IQR确定。具体地，最小值为Q1-1.5IQR，最大值为Q3+1.5IQR。其中，异常点为功率值小于Q1-1.5IQR的点，或者功率值大于Q3+1.5IQR的点。
在检测出异常点后，调度管理系统200可以采用类似填充缺失值的方式纠正异常点的功率值，例如以时间窗口内的数学统计值为修正后的功率值。
S510：调度管理系统200对功率分布进行归一化。
在一些实施例中，为了加快训练模型的收敛速度，调度管理系统200还可以对功率分布中各时间点的功率值进行归一化。
S512：调度管理系统200从功率分布进行特征提取，根据提取的特征生成样本数据。
具体地，调度管理系统200可以通过特征工程的方式提取特征。具体地，调度管理系统200可以确定时间窗口，该时间窗口可以与上文所述的填充缺失的功率值的时间窗口不同，然后获取时间窗口内多个时间点的功率值的数学统计值，例如极值、均值、方差等中的至少一种，以及获取功率对应的时间信息，如时刻、对应的日排序(在一周中的第几日)、月排序(在一年中的第几月)，然后根据时间窗口内的上述数学统计值和时间信息生成样本数据。该样本数据可以表示为(X，Y)，其中，X包括时间窗口内的时间功率序列以及提取的特征，如上述功率值的极值、均值、方差和时刻、日排序、月排序，Y为监督信息，该监督信息可以是该时间窗口之后一个时间窗口内的时间功率序列。对应地，在利用模型进行预测时，也可以对时间窗口内的时间功率序列进行特征提取，将时间功率序列和提取的特征作为输入。
在另一些可能的实现方式中，调度管理系统200也可以直接根据一个时间窗口的时间功率序列以及该时间窗口之后的一个时间功率序列生成样本数据。即样本数据(X，Y)中的X表示一个时间窗口的时间功率序列，Y为监督信息，该监督信息为该时间窗口之后一个时间窗口内的特征。
需要说明的是，X和Y对应的时间窗口的长度可以是相等的，也可以是不等的。X对应的时间窗口的长度可以大于Y对应的实际窗口的长度。例如，X对应的时间窗口的长度可以为3天，Y对应的时间窗口的长度可以为1天。
在本实施例中，S504至S510为可选步骤，在本申请实施例其他可能的实现方式中，也可以不执行上述S504至S510。
下面结合附图进一步介绍调度管理系统200如何对各个模型进行训练和验证的过程。调度管理系统200生成多个样本数据后，还可以将样本数据分别划分值训练集和验证集。例如，调度管理系统200可以按照第一预设比例，例如8:2，将样本数据分别划分值训练集和验证集。在一些实施例中，调度管理系统200可以将样本数据分别划分至训练集、验证集和测试集。例如，调度管理系统200可以按照第二预设比例，例如7:2:1，将样本数据分别划分值训练集、验证集和测试集。
其中，训练集用于对模型进行拟合，验证集用于对基于训练集中的样本数据(即训练样本)训练所得的模型进行验证，以对模型的超参数进行调整，以及对模型的能力如预测精度进行初步评估。测试集用于对验证通过的模型的泛化能力。
调度管理系统200获得数据样本后，可以将训练集中的样本数据即训练样本作为模型池的输入，进行分布式模型训练。在模型训练阶段，调度管理系统200可以根据模型的超参的搜索空间，利用自动调参方法如网格搜索(grid search)、粒子群(particle swarm optimization，PSO)等优化算法进行超参的选择和调优，并根据训练样本输入模型后所得的损失函数，进行模型的参数的选择和调优，当模型收敛时，停止训练。进一步地，调度管理系统200可以 利用验证集中的样本数据即验证样本，对训练得到的模型进行验证，以初步评估模型的精度。当精度不满足要求时，调度管理系统200可以对超参进行进一步调优，然后对模型进行再训练。当训练的模型验证通过时，可以用于进行功率预测。
另外，在进行功率预测时，考虑到功率分布中可能存在异常点的情况，在图2所示实施例中，调度管理系统200还可以在确定模型的评价指标值时，先对功率分布中异常点的功率值进行修正。其中，调度管理系统200可以采用与模型训练过程中异常点检测方法相同或相似的方法检测出异常点，如采用箱型图的四分位距检测出异常点，并以该异常点为终点的时间窗口内多个时间点的功率值的数学统计值修正异常点的功率值，然后根据修正后的功率值确定模型的评价指标值，如此可以提高准确度。
上文结合图1至图5对本申请实施例提供的功率预测的方法进行了详细介绍，下面将结合附图对本申请实施例提供的装置、设备进行介绍。
参见图6所示的功率预测的装置的结构示意图，该装置600包括：
通信单元602，用于获取模型池中模型的评价指标值，所述评价指标值用于指示所述模型的精度；
预测单元604，用于根据所述评价指标值选择第一模型对第一用电单元进行功率预测，所述第一用电单元为数据中心中任一用电单元；
显示单元606，用于呈现对所述第一用电单元的所述功率预测的结果。
在一些可能的实现方式中，所述预测单元604具体用于：
根据所述评价指标值从所述模型池中选择第一模型；
根据所述第一用电单元的历史功率分布，利用所述第一模型预测所述第一用电单元的未来功率分布，所述历史功率分布指示当前时刻之前至少一个统计周期的功率，所述未来功率分布指示当前时刻之后至少一个统计周期的功率。
在一些可能的实现方式中，所述评价指标值包括误差值，所述预测单元604具体用于：
根据所述误差值从所述模型池中选择所述误差值满足预设条件的模型为第一模型。
在一些可能的实现方式中，所述误差值满足预设条件包括：
所述误差值小于预设阈值；或者，所述误差值最小。
在一些可能的实现方式中，所述预测单元604具体用于：
根据所述评价指标值更新所述模型池中的模型；
根据更新后的所述模型的评价指标值从更新后的所述模型池中选择第一模型。
在一些可能的实现方式中，所述评价指标值根据对所述数据中心中第一用电单元的功率预测时的区间误差值和单点误差值中的至少一种确定。
在一些可能的实现方式中，所述通信单元602还用于：
获取所述数据中心中新加入的第二用电单元的功率分布；
所述装置600还包括：
确定单元，用于从所述第一用电单元中确定与所述第二用电单元的功率分布的相似度达到预设相似度的至少一个第三用电单元；
所述确定单元，还用于根据对所述第三用电单元进行功率预测的模型确定用于对所述第二用电单元进行功率预测的第二模型。
在一些可能的实现方式中，所述确定单元具体用于：
将对所述第三用电单元进行功率预测的模型确定为用于对所述第二用电单元进行功率预测的第二模型。
在一些可能的实现方式中，所述确定单元具体用于：
将对所述至少一个第三用电单元进行功率预测的模型加入所述第二用电单元的模型池；
获取所述第二用电单元的模型池中模型对所述第二用电单元进行功率预测的评价指标值；
根据所述评价指标值从所述第二用电单元的模型池中选择用于对所述第二用电单元进行功率预测的第二模型。
在一些可能的实现方式中，所述装置600还包括：
生成单元，用于根据所述第一用电单元的历史功率分布生成训练样本；
训练单元，用于利用所述训练样本对初始模型进行训练得到所述模型池中的模型。
在一些可能的实现方式中，所述模型池中包括两种或两种以上模型。
在一些可能的实现方式中，所述第一用电单元为用电设备的集合，该用电设备的集合包括至少一个用电设备，至少一个整机柜的用电设备，或者至少一个机房的用电设备中一种或多种。
在一些可能的实现方式中，所述显示单元具体用于：
通过图形用户界面呈现所述模型池中至少一个模型对所述第一用电单元的所述功率预测的结果。
根据本申请实施例的功率预测的装置600可对应于执行本申请实施例中描述的方法，并且功率预测的装置600的各个模块/单元的上述和其它操作和/或功能分别为了实现图2所示实施例中的各个方法的相应流程，为了简洁，在此不再赘述。
本申请实施例还提供了一种设备700。该设备700可以是笔记本电脑、台式机等端侧设备，也可以是云环境或边缘环境中的计算机集群，或者端侧设备以及云环境、边缘环境中设备的组合。该设备700具体用于实现如图6所示实施例中功率预测的装置600的功能。
图7提供了一种电子设备700的结构示意图，如图7所示，电子设备70包括总线701、处理器702、通信接口703和存储器704、显示器705。处理器702、存储器704和通信接口703、显示器705之间通过总线701通信。
总线701可以是外设部件互连标准(peripheral component interconnect，PCI)总线或扩展工业标准结构(extended industry standard architecture，EISA)总线等。总线可以分为地址总线、数据总线、控制总线等。为便于表示，图7中仅用一条粗线表示，但并不表示仅有一根总线或一种类型的总线。
处理器702可以为中央处理器(central processing unit，CPU)。在一些实施例中，处理器702也可以为图形处理器(graphics processing unit，GPU)、微处理器(micro processor，MP)或者数字信号处理器(digital signal processor，DSP)等处理器中的任意一种或多种。
通信接口703用于与外部通信。例如，通信接口703可以用于获取模型池中模型的评价指标值，或者用于获取数据中心中新加入的第二用电单元的功率分布等等。
存储器704可以包括易失性存储器(volatile memory)，例如随机存取存储器(random access memory，RAM)。存储器704还可以包括非易失性存储器(non-volatile memory)，例如只读存储器(read-only memory，ROM)，快闪存储器，硬盘驱动器(hard disk drive，HDD)或固态驱动器(solid state drive，SSD)。
显示器705一种输入输出(input/output，I/O)设备。该设备可以将电子文件如图像、文字显示到屏幕上，以供用户查看。根据制造材料不同，显示器705可以分为液晶显示器(liquid crystal display，LCD)、有机电激光(organic light emitting diode，OLED)显示器等。
存储器704中存储有可执行代码，处理器702执行该可执行代码以执行前述功率预测的方法。
具体地，在实现图6所示实施例的情况下，且图6实施例中所描述的功率预测的装置600的各单元为通过软件实现的情况下，执行图6中的预测单元604功能所需的软件或程序代码存储在存储器704中。
通信单元602功能通过通信接口703实现。通信接口703用于获取模型池中模型的评价指标值，并将模型的评价指标值通过总线701传输至处理器702，处理器702执行存储器704中存储的各单元对应的程序代码，如执行预测单元604对应的程序代码，以执行根据所述评价指标值选择第一模型对第一用电单元进行功率预测的步骤。然后处理器702将对第一用电单元的功率预测的结果通过总线传输至显示器705。显示器705呈现对第一用电单元的功率预测的结果。
应理解，本申请实施例的电子设备700可对应于本申请实施例中的图6所述的功率预测的装置600，电子设备700用于实现上述图2所述方法中相应主体执行的方法的操作步骤，为了简洁，在此不再赘述。
在上述实施例中，可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。
所述计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行所述计算机程序指令时，全部或部分地产生按照本申请实施例所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一计算机可读存储介质传输，例如，所述计算机指令可以从一个网站站点、计算机、训练设备或数据中心通过有线(例如同轴电缆、光纤、数字用户线(DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、训练设备或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存储的任何可用介质或者是包含一个或多个可用介质集成的训练设备、数据中心等数据存储设备。所述可用介质可以是磁性介质，(例如，软盘、硬盘、磁带)、光介质(例如，DVD)、或者半导体介质(例如固态硬盘(solid state disk，SSD))等。
以上所述，仅为本申请的具体实施方式。熟悉本技术领域的技术人员根据本申请提供的具体实施方式，可想到变化或替换，都应涵盖在本申请的保护范围之内。

Claims (19)

  1. 一种功率预测的方法，其特征在于，所述方法包括：
    获取模型池中模型的评价指标值，所述评价指标值用于指示所述模型的精度；
    根据所述评价指标值选择第一模型对第一用电单元进行功率预测，所述第一用电单元为数据中心中任一用电单元；
    呈现对所述第一用电单元的所述功率预测的结果。
  2. 根据权利要求1所述的方法，其特征在于，所述根据所述评价指标值选择第一模型对第一用电单元进行功率预测，包括：
    根据所述评价指标值从所述模型池中选择第一模型；
    根据所述第一用电单元的历史功率分布，利用所述第一模型预测所述第一用电单元的未来功率分布，所述历史功率分布指示当前时刻之前至少一个统计周期的功率，所述未来功率分布指示当前时刻之后至少一个统计周期的功率。
  3. 根据权利要求1或2所述的方法，其特征在于，所述评价指标值包括误差值，所述根据所述评价指标值选择第一模型，包括：
    根据所述误差值从所述模型池中选择所述误差值满足预设条件的模型为第一模型。
  4. 根据权利要求3所述的方法，其特征在于，所述误差值满足预设条件包括：
    所述误差值小于预设阈值；或者，所述误差值最小。
  5. 根据权利要求1至4任一项所述的方法，其特征在于，所述根据所述评价指标值选择第一模型，包括：
    根据所述评价指标值更新所述模型池中的模型；
    根据更新后的所述模型的评价指标值从更新后的所述模型池中选择第一模型。
  6. 根据权利要求1至5任一项所述的方法，其特征在于，所述评价指标值根据对所述数据中心中第一用电单元的功率预测时的区间误差值和单点误差值中的至少一种确定。
  7. 根据权利要求1至6任一项所述的方法，其特征在于，所述模型池中包括两种或两种以上模型。
  8. 根据权利要求1至7任一项所述的方法，其特征在于，所述第一用电单元为用电设备的集合，所述用电设备的集合包括至少一个用电设备，至少一个整机柜的用电设备，或者至少一个机房的用电设备中一种或多种。
  9. 根据权利要求1至8任一项所述的方法，其特征在于，所述呈现对所述第一用电单元的所述功率预测的结果，包括：
    通过图形用户界面呈现所述模型池中至少一个模型对所述第一用电单元的所述功率预测的结果。
  10. 一种功率预测的装置，其特征在于，所述装置包括：
    通信单元，用于获取模型池中模型的评价指标值，所述评价指标值用于指示所述模型的精度；
    预测单元，用于根据所述评价指标值选择第一模型对第一用电单元进行功率预测，所述第一用电单元为数据中心中任一用电单元；
    显示单元，用于呈现对所述第一用电单元的所述功率预测的结果。
  11. 根据权利要求10所述的装置，其特征在于，所述预测单元具体用于：
    根据所述评价指标值从所述模型池中选择第一模型；
    根据所述第一用电单元的历史功率分布，利用所述第一模型预测所述第一用电单元的未 来功率分布，所述历史功率分布指示当前时刻之前至少一个统计周期的功率，所述未来功率分布指示当前时刻之后至少一个统计周期的功率。
  12. 根据权利要求10或11所述的装置，其特征在于，所述评价指标值包括误差值，所述预测单元具体用于：
    根据所述误差值从所述模型池中选择所述误差值满足预设条件的模型为第一模型。
  13. 根据权利要求12所述的装置，其特征在于，所述误差值满足预设条件包括：
    所述误差值小于预设阈值；或者，所述误差值最小。
  14. 根据权利要求10至13任一项所述的装置，其特征在于，所述预测单元具体用于：
    根据所述评价指标值更新所述模型池中的模型；
    根据更新后的所述模型的评价指标值从更新后的所述模型池中选择第一模型。
  15. 根据权利要求10至14任一项所述的装置，其特征在于，所述评价指标值根据对所述数据中心中第一用电单元的功率预测时的区间误差值和单点误差值中的至少一种确定。
  16. 根据权利要求10至15任一项所述的装置，其特征在于，所述模型池中包括两种或两种以上模型。
  17. 根据权利要求10至16任一项所述的装置，其特征在于，所述第一用电单元为用电单元集合，所述用电单元集合包括至少一个用电设备，至少一个整机柜的用电设备，或者至少一个机房的用电设备中一种或多种。
  18. 根据权利要求10至17任一项所述的装置，其特征在于，所述显示单元具体用于：
    通过图形用户界面呈现所述模型池中至少一个模型对所述第一用电单元的所述功率预测的结果。
  19. 一种设备，其特征在于，所述设备包括处理器和存储器；
    所述处理器用于执行所述存储器中存储的指令，以使得所述设备执行如权利要求1至9中任一项所述的方法。
PCT/CN2021/106964 2020-09-15 2021-07-17 功率预测的方法、装置及设备 WO2022057427A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP21868256.5A EP4207528A4 (en) 2020-09-15 2021-07-17 POWER PREDICTION METHOD, APPARATUS AND DEVICE
US18/183,473 US20230216296A1 (en) 2020-09-15 2023-03-14 Power Prediction Method and Apparatus, and Device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010966481.7A CN114188935A (zh) 2020-09-15 2020-09-15 功率预测的方法、装置及设备
CN202010966481.7 2020-09-15

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/183,473 Continuation US20230216296A1 (en) 2020-09-15 2023-03-14 Power Prediction Method and Apparatus, and Device

Publications (1)

Publication Number Publication Date
WO2022057427A1 true WO2022057427A1 (zh) 2022-03-24

Family

ID=80600792

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/106964 WO2022057427A1 (zh) 2020-09-15 2021-07-17 功率预测的方法、装置及设备

Country Status (4)

Country Link
US (1) US20230216296A1 (zh)
EP (1) EP4207528A4 (zh)
CN (1) CN114188935A (zh)
WO (1) WO2022057427A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103679298A (zh) * 2013-12-27 2014-03-26 中能电力科技开发有限公司 风电场短期功率预测精度的评价方法
CN106875033A (zh) * 2016-12-26 2017-06-20 华中科技大学 一种基于动态自适应的风电集群功率预测方法
CN109657856A (zh) * 2018-12-14 2019-04-19 中国科学院广州能源研究所 一种用于电功率预测模型评价的精度指标构建方法
CN109816196A (zh) * 2018-12-04 2019-05-28 平安科技（深圳）有限公司 预测模型的评价值计算方法、装置、设备及可读存储介质
CN110188932A (zh) * 2019-05-20 2019-08-30 国核电力规划设计研究院有限公司 基于评价优化的数据中心能耗预测方法
CN110619420A (zh) * 2019-07-31 2019-12-27 广东工业大学 一种基于Attention-GRU的短期住宅负荷预测方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1877950B1 (en) * 2005-05-02 2018-11-28 Schneider Electric IT Corporation Methods and systems for managing facility power and cooling
US10366346B2 (en) * 2014-05-23 2019-07-30 DataRobot, Inc. Systems and techniques for determining the predictive value of a feature
US10355913B2 (en) * 2017-05-04 2019-07-16 Servicenow, Inc. Operational analytics in managed networks
US11163271B2 (en) * 2018-08-28 2021-11-02 Johnson Controls Technology Company Cloud based building energy optimization system with a dynamically trained load prediction model

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103679298A (zh) * 2013-12-27 2014-03-26 中能电力科技开发有限公司 风电场短期功率预测精度的评价方法
CN106875033A (zh) * 2016-12-26 2017-06-20 华中科技大学 一种基于动态自适应的风电集群功率预测方法
CN109816196A (zh) * 2018-12-04 2019-05-28 平安科技（深圳）有限公司 预测模型的评价值计算方法、装置、设备及可读存储介质
CN109657856A (zh) * 2018-12-14 2019-04-19 中国科学院广州能源研究所 一种用于电功率预测模型评价的精度指标构建方法
CN110188932A (zh) * 2019-05-20 2019-08-30 国核电力规划设计研究院有限公司 基于评价优化的数据中心能耗预测方法
CN110619420A (zh) * 2019-07-31 2019-12-27 广东工业大学 一种基于Attention-GRU的短期住宅负荷预测方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4207528A4 *

Also Published As

Publication number Publication date
US20230216296A1 (en) 2023-07-06
EP4207528A4 (en) 2024-03-06
EP4207528A1 (en) 2023-07-05
CN114188935A (zh) 2022-03-15

Similar Documents

Publication Publication Date Title
US11876374B2 (en) System and method for optimal control of energy storage system
JP7482167B2 (ja) 動的エネルギーストレージシステム制御のためのシステムおよび方法
US11836599B2 (en) Optimizing data center controls using neural networks
JP6156576B2 (ja) 電力需給制御装置、電力需給制御方法、及び、プログラム
US20190258307A1 (en) Time varying power management within datacenters
US8761953B2 (en) Grid optimization resource dispatch scheduling
US20200088796A1 (en) Predictive rechargeable battery management system
CN106899660A (zh) 基于滚动灰色预测模型的云数据中心节能调度实现方法
JP7249155B2 (ja) 蓄電池管理装置および蓄電池管理方法
JP2019216552A (ja) 行動生成装置、蓄電素子評価装置、コンピュータプログラム、学習方法及び評価方法
CN115330275B (zh) 一种退役电池的梯次利用方法及装置
US20180166878A1 (en) Machine Learning Based Demand Charge
US10103575B2 (en) Power interchange management system and power interchange management method for maintaining a balance between power supply and demand
Zheng et al. Energy storage state-of-charge market model
CN109522120B (zh) 一种基于Hadoop的智能家居管理平台
Singhal et al. A reserve response set model for systems with stochastic resources
US10931107B2 (en) System and method for management of an electricity distribution grid
WO2022057427A1 (zh) 功率预测的方法、装置及设备
CN111106415B (zh) 一种电池管理方法、装置及云服务器
CN107292454B (zh) 面向共享的电力资源优化配置管理方法
Zhang et al. Dynamic energy storage control for reducing electricity cost in data centers
CN115237080B (zh) 基于虚拟电厂的设备调控方法、装置、设备和可读介质
CN117575113B (zh) 基于马尔科夫链的边端协同任务处理方法、装置和设备
CN117117907A (zh) 低压配电网负荷接入方法、装置、电子设备及存储介质
JP2024007073A (ja) 発電バランシンググループの組成システム、方法及びプログラム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21868256

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021868256

Country of ref document: EP

Effective date: 20230329

NENP Non-entry into the national phase

Ref country code: DE