Data center energy consumption optimization method based on electricity price and load prediction

Abstract

The invention discloses a data center energy consumption optimization method based on electricity price and load prediction, and relates to the technical field of scheduling optimization of a demand side, and the method comprises the steps: collecting parameters related to the energy consumption of a data center, constructing a data center energy consumption model based on workload transfer, proposing an electricity price and workload prediction model based on an LSTM neural network, and optimizing the energy consumption of the data center. The method comprises the following steps: constructing a data center load optimization scheduling decision model in a market environment by taking minimization of power purchase cost as an objective function and considering constraint conditions, and performing simulation verification on the model by utilizing an IEEE30 node system to obtain the electricity price of a prediction day, the arrived workload, the workload scheduling condition and the power purchase cost. According to the data center energy consumption optimization method based on the electricity price and load prediction, through introduction of the LSTM model, short-term and long-term fluctuation of the electricity price can be accurately predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of demand-side scheduling optimization, and in particular to a data center energy optimization method based on electricity price and load forecasting. Background Art

[0002] With the rapid development of artificial intelligence and information technology, data centers, as physical carriers for various digital technology applications, have steadily increased in size. By the end of 2021, the global data center market size exceeded US$67.9 billion, an increase of 9.8% over 2020. In March 2020, my country officially included data centers in the new infrastructure, unveiling the prelude to the construction of computing power infrastructure represented by data centers. By the end of 2021, my country's data centers had 5.2 million racks in use and 19 million servers in use, ranking second in the world in terms of computing power. As a high-energy-consuming industry, data centers consume a lot of electricity during operation. In 2022, global data centers consumed about 460 terawatt-hours of electricity, accounting for about 2% of the world's total electricity consumption, equivalent to the electricity consumption of 153 million households. According to estimates by the International Energy Agency, due to the increasing popularity of generative artificial intelligence that requires a lot of computing to operate, global data center power consumption will increase to 2.3 times that of 2022 in 2026. In my country, the total electricity consumption of data centers nationwide in 2020 was about 204.5 billion kWh, accounting for about 2.7% of the total electricity consumption in the society. Compared with 2018, it increased by about 27.2%, and it is still on an upward trend. It is expected to exceed 700 billion kWh in 2035. Therefore, data centers have become an important power load in my country, and it is of great significance to understand their energy consumption characteristics. The high energy consumption of data centers has brought high electricity expenses. For example, Google consumed 226 million kWh in 2010, and the corresponding electricity bill exceeded 135 million US dollars. China Unicom's electricity expenditure in 2012 was 1.7 billion US dollars, and its profit in the same year was only 1.2 billion US dollars. Therefore, for data centers, saving energy consumption costs, which account for about 70% of operating costs, is the main problem that needs to be solved in their operation process.

[0003] The workload of the data center is divided into two categories: offline load and online load, depending on whether it can be delayed. Among them, the offline load has the characteristic of being transferable in the time dimension. The data center can transfer part of the offline load to the off-peak period for processing, achieving the effect of "peak shaving and valley filling". In addition, in order to ensure the high reliability of the service, the hardware configuration of the data center is highly redundant, and the average utilization rate of typical servers is about 12%-18%. Therefore, the data center has a large adjustable space, which can achieve delayed processing of workloads without affecting the quality of service.

[0004] Research can be conducted on the energy optimization of data centers under market conditions to reduce their electricity purchase costs and improve economic benefits. Summary of the invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is: the existing data center energy consumption optimization method has low prediction accuracy, low load scheduling efficiency, inflexible energy consumption model, and how to dynamically adjust the load to make full use of low electricity price periods to achieve energy efficiency optimization.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: a data center energy optimization method based on electricity price and load forecasting, comprising collecting parameters related to data center energy consumption and constructing a data center energy consumption model based on workload transfer;

[0008] Propose an electricity price and workload prediction model based on LSTM neural network;

[0009] Taking minimizing the electricity purchase cost as the objective function and considering the constraints, a data center load optimization scheduling decision model is constructed under the market environment;

[0010] The model is simulated and verified using the IEEE30-node system to obtain the electricity price on the predicted day, the workload arriving, the workload scheduling, and the electricity purchase cost.

[0011] As a preferred scheme of the data center energy optimization method based on electricity price and load forecast described in the present invention, wherein: the construction of the data center energy consumption model includes that the energy consumption of the data center is highly correlated with the number and type of workloads, and based on this relationship, an energy consumption model based on data center workload processing can be constructed.

[0012] As a preferred solution of the data center energy optimization method based on electricity price and load forecasting described in the present invention, the LSTM neural network includes calculating the coefficients of three gates and calculating the current neuron candidate state quantity based on the current input value and the output value at the previous moment, and the forgetting gate and the input gate determine the proportion of the state value at the previous moment and the candidate state value at the current moment in the new state value and calculate the output at the current moment.

[0013] As a preferred solution of the data center energy optimization method based on electricity price and load forecasting described in the present invention, wherein: the construction of a data center load optimization scheduling decision model includes inputting the prediction results into the decision model for solving, and the data center schedules the workload between time periods according to the solution results. The decision model needs to meet offline load constraints, service quality requirements constraints, and data center capacity constraints.

[0014] As a preferred scheme of the data center energy optimization method based on electricity price and load forecast described in the present invention, the simulation verification includes taking the relevant parameters of the data center and the clearing electricity prices of the day-ahead market in each period of the area where the data center is located and the load conditions to be processed as input conditions of the decision model, obtaining the scheduling status of the data center load and calculating the electricity expenditure of the data center before and after the workload transfer based on the scheduling status.

[0015] As a preferred solution of the data center energy optimization method based on electricity price and load forecasting described in the present invention, the purpose of the LSTM model includes obtaining the electricity price and workload demand for each time period of the next day by constructing an electricity price and workload forecasting model, and using the obtained data as the basis for data center load optimization scheduling decisions.

[0016] As a preferred solution of the data center energy optimization method based on electricity price and load forecast described in the present invention, the data center scheduling includes, under the premise of meeting constraints, the data center arranges offline loads to be processed in time periods with low electricity prices as much as possible, and when the electricity price is high, the data center will postpone the processing of offline loads.

[0017] The optimized scheduling decision model includes adding a joint model of short-term and long-term electricity price fluctuation correlation in the prediction link, which not only refines the electricity price information of the time period but also provides general direction support for the overall scheduling, and adds a prediction error correction mechanism in the scheduling link to ensure that the error caused by short-term and long-term electricity price fluctuations to the prediction link is minimized.

[0018] Another object of the present invention is to provide a data center energy consumption optimization system based on electricity price and workload prediction in a market environment, which can solve the problems of low prediction accuracy, poor load scheduling efficiency and inflexible energy consumption model in current data center energy consumption optimization technology through electricity price and workload prediction model based on LSTM neural network and combined with data center load optimization scheduling decision model.

[0019] As a preferred solution of the data center energy optimization system based on electricity price and workload prediction in the market environment described in the present invention, it includes: a data acquisition and preprocessing module, a prediction and modeling module, and a load scheduling optimization and decision-making module.

[0020] The data acquisition and preprocessing module is used to obtain the number and type of online loads and offline loads, record the changes in data center workloads in different time periods, obtain historical electricity price data including market electricity price fluctuation patterns to facilitate electricity price forecasting, obtain power consumption data of IT equipment, refrigeration systems, UPS power supplies, power distribution systems, and lighting systems as input to the energy consumption model, and perform format conversion, cleaning, and denoising on the collected data to ensure data accuracy and consistency.

[0021] The prediction and modeling module is used to use historical electricity price data to predict the fluctuation trend of future electricity prices through the LSTM model. The LSTM model improves the accuracy of electricity price prediction by processing the relationship between short-term and long-term electricity price fluctuations. The historical workload data is used to train the LSTM model to predict future workload demand to ensure that the trend of load changes can be accurately reflected in future scheduling decisions. The electricity price and workload prediction results are used to build an energy consumption model of the data center, define load optimization scheduling goals and formulate constraints.

[0022] The load scheduling optimization and decision-making is used to input the electricity price fluctuations and workload demands predicted by the LSTM model into the load optimization scheduling decision model, optimize the load distribution between time periods according to the optimization scheduling model, electricity price fluctuations and load demands, postpone some offline loads to the period of low electricity prices to reduce overall energy consumption, and use simulation tools to verify the optimization scheduling results, calculate the electricity price, workload, scheduling status and electricity purchase cost on the predicted day, and further optimize the scheduling strategy, calculate the electricity bill according to the scheduling optimization results, and demonstrate the economic benefits of optimized scheduling.

[0023] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a method for optimizing energy consumption of a data center based on electricity price and load prediction.

[0024] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a data center energy optimization method based on electricity price and load prediction.

[0025] Beneficial effects of the present invention: The data center energy optimization method based on electricity price and load forecasting provided by the present invention can accurately predict short-term and long-term fluctuations in electricity prices through the introduction of the LSTM model, solves the gradient vanishing or explosion problem in traditional forecasting methods, and significantly improves the accuracy of electricity price forecasting. By combining the electricity price forecast results with workload requirements, dynamically adjusting the load distribution, and effectively utilizing low electricity price periods, the scheduling efficiency is significantly improved. The model flexibly handles the different characteristics of online and offline loads and optimizes their distribution in the time dimension, so that the data center can dynamically manage energy consumption according to actual conditions, thereby minimizing overall energy consumption and electricity expenses. The present invention achieves better results in terms of electricity price forecast accuracy, load scheduling efficiency, and energy consumption optimization flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0027] Figure 1 An overall flow chart of a data center energy optimization method based on electricity price and load forecasting provided in the first embodiment of the present invention.

[0028] Figure 2 A load transfer diagram of a data center energy optimization method based on electricity price and load forecasting is provided in the first embodiment of the present invention.

[0029] Figure 3 A specific structural diagram of an LSTM neuron of a data center energy optimization method based on electricity price and load forecasting provided in the first embodiment of the present invention.

[0030] Figure 4 A load condition diagram of a data center energy optimization method based on electricity price and load forecasting provided in the first embodiment of the present invention.

[0031] Figure 5 A scheduling diagram of a data center energy optimization method based on electricity price and load forecasting provided in the first embodiment of the present invention.

[0032] Figure 6 A data center related parameter diagram of a data center energy optimization system based on electricity price and workload prediction in a market environment is provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0034] Example 1, reference Figure 1-6 , which is an embodiment of the present invention, provides a data center energy optimization method based on electricity price and load forecasting, comprising:

[0035] S1: Collect parameters related to data center energy consumption and build a data center energy consumption model based on workload transfer.

[0036] Specifically, a data center is mainly composed of four parts: site layer, facility layer, IT hardware and IT software. The electrical equipment mainly includes IT equipment, refrigeration system, UPS power supply, power distribution system and lighting system. Among them, IT equipment and refrigeration system have the largest power consumption, accounting for about 82% of the total power consumption. When the workload increases, the power consumption of IT equipment increases, and the heat dissipation also increases. The working frequency of the refrigeration system increases, which increases the overall energy consumption of the data center. Therefore, the energy consumption of the data center is highly correlated with the number and type of workloads, and an energy consumption model based on the workload processing of the data center can be constructed.

[0037]

[0038] Among them, P et represents the energy consumption of the data center during period t, α PUE represents the ratio of the energy consumption of the data center to the energy consumption of its IT equipment, e1 and e2 represent the energy consumption of the IT equipment when processing the online load and offline load of the unit, and σ t , Indicates the number of online loads and offline loads processed during period t.

[0039] The workload of the data center can be divided into two categories: online load and offline load, depending on whether there is a delay. Among them, offline load is also called delay-tolerant load, such as computing services, storage backup, data processing, etc. This type of load has a lower priority, and the server can delay its processing. It only needs to be completed before the deadline, that is, when the node is in the peak power consumption period or the period with high electricity prices, part of it can be transferred offline to the low power consumption period or the period with low electricity prices for processing. The data center is adjustable in the time dimension. By optimizing the distribution of offline load in time, part of the load at time A can be delayed by Δt to time B while meeting the quality of service.

[0040] S2: Propose an electricity price and workload prediction model based on LSTM neural network.

[0041] One of the foundations of data center load optimization scheduling is to accurately predict future electricity prices and workloads. The prediction accuracy of the prediction model will greatly affect its optimization scheduling decision. The fluctuation pattern of electricity prices at a certain moment is not only related to recent electricity price data, but also closely related to electricity price data in the same period of historical years. Traditional workload and prediction models, time series models, and RNN prediction models are prone to task failure due to gradient disappearance or gradient explosion when processing such data. They cannot take into account the regularity of electricity prices in short and long time series at the same time. We choose to use LSTM neural networks to process electricity prices and workloads.

[0042] Furthermore, as a special recurrent neural network, the main improvement of LSTM over RNN is that it introduces cell states, using ct Indicates. LSTM adds three "gates" to the original RNN hidden layer neuron structure to control c t The state of input gate i t , forget gate f t , output gate o t The gate is a coefficient in the interval [0,1]. If it is 1, it means that all relevant information is remembered. If it is 0, it means that all relevant information is forgotten. Usually, it is within the interval, which means that only the information that needs to be saved is remembered, which can effectively reduce the possibility of gradient disappearance and gradient explosion. At time t, the neuron has 3 inputs and 2 outputs. The input value x at time t is t , output value y at time t-1 t-1 , the unit state c at time t-1 t-1 , output value y at time t t and the cell state c t .

[0043] The specific operation process of neurons in LSTM mainly includes calculating the coefficients of three gates based on the current input value x t and the output value y at the previous moment t Calculate the current neuron candidate state By the forget gate t and input gate i t Determine the state value c at the previous moment t-1 and the candidate state value at the current moment In the new state value c t and calculate the output at the current moment.

[0044] i t =σ(W i ·[y t-1 ,x t ]+b i )

[0045] f t =σ(W f ·[y t-1 ,x t ]+b f )

[0046] o t =σ(W o ·[y t-1 ,x t ]+b o )

[0047]

[0048] y t =ot ·tanh(c t )

[0049] Among them, σ represents the Sigmoid function, W represents the weight matrix, and b represents the bias.

[0050] S3: Taking minimizing the electricity purchase cost as the objective function and considering the constraints, a data center load optimization scheduling decision model is constructed under the market environment.

[0051] Collect historical data, build electricity price data sets and load data sets respectively, train and predict through LSTM model, obtain electricity price and workload demand for each time period of the next day, input the prediction results into decision model for solution, the data center schedules workload between time periods according to the solution results, and the objective function of data center load optimization scheduling decision model is to minimize the electricity cost of data center.

[0052]

[0053] Among them, λ t represents the node electricity price during period t. The model needs to satisfy the time flexibility of offline load, but is subject to the upper limit of offline load that can be stored in the data center.

[0054] Q t =Q t-1 -ζ t-1 +Ω t

[0055] 0≤Q t ≤Q max

[0056]

[0057] Among them, Q t Indicates the number of offline loads to be processed in the data center during period t. Ω t It represents the number of offline loads arriving at the data center in period t. All arriving offline loads should be processed before the scheduling deadline. The response time for processing workloads cannot exceed the delay limit D of the service level agreement. The service quality requirements of the data center are modeled using M / M / 1 queues.

[0058]

[0059] Among them, u represents the average service rate, n t represents the number of servers powered on in the data center during period t, N represents the total number of servers in the data center, and X t Indicates the number of online loads that the data center reaches during period t.

[0060] Cx t +dζt ≤M

[0061] Where M represents the memory of the data center, and c and d represent the data volume of online load and offline load.

[0062] Furthermore, short-term electricity price fluctuations are mainly affected by instantaneous factors such as market demand, weather changes, and equipment failures, and have strong randomness and rapid fluctuation characteristics. Long-term electricity price fluctuations reflect overall market trends, such as changes in energy policies, long-term supply and demand adjustments, and have strong stability. Joint modeling introduces a weight factor α to dynamically balance short-term fluctuations and long-term trends, which can capture short-term high-frequency characteristics while retaining the guiding role of long-term trends, avoiding the limitations of a single prediction method and achieving accurate characterization of multiple time scales.

[0063] p t =αp t 短期 +(1-α)p t 长期

[0064] Among them, α represents the weight factor, dynamically adjusting the impact of short-term and long-term electricity price forecasts, p t Represents the comprehensive forecast electricity price for time period t.

[0065] Furthermore, when there is a large error in the predicted electricity price, the correction mechanism can adjust the load processing period by increasing the weight of the predicted electricity price, avoiding excessive load operation during the high electricity price period and reducing electricity expenses. When the model predicts that the electricity price is low during the period and the actual electricity price is high, the correction mechanism dynamically postpones offline load processing and arranges other loads to reduce energy consumption during the high electricity price period, thereby reducing the impact of the prediction error on the scheduling results.

[0066] L t 调整 =L t 原 +δt

[0067] Among them, L t 调整 represents the load dispatching amount of the modified period t, L t 原 is the original load dispatch amount of time period t without correction, δt is the load correction amount of time period t, which represents the load adjustment introduced by the electricity price prediction error.

[0068] S4: The model is simulated and verified using the IEEE30-node system to obtain the electricity price, arriving workload, workload scheduling, and electricity purchase cost on the predicted day.

[0069] Considering that the data center participates in the electricity market as a price acceptor, it is assumed that the data volume of the unit online load and offline load is 1.25MB and 100MB respectively. Using the constructed LSTM-based electricity price and load forecasting model, the clearing electricity price and the load to be processed in each period of the day-ahead market in the area where the data center is located are obtained. The forecast results are used as the input conditions of the decision model to obtain the scheduling of the data center load. Under the premise of meeting the constraints, the data center will arrange the offline load to be processed in the period with low electricity price as much as possible. When the electricity price is high, the data center will postpone the processing of offline load.

[0070] The electricity expenditure of the data center before and after the workload transfer was calculated. If the load was not transferred, the electricity expenditure of the data center on the target day would be 369,189 yuan. If the load was transferred according to the results obtained by the optimized scheduling decision model, the electricity expenditure of the data center would be reduced to 328,192 yuan, a decrease of 11.1% compared with before.

[0071] Example 2 is an embodiment of the present invention, which provides a method for optimizing energy consumption in a data center based on electricity price and load forecasting. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0072] First, we collected parameters related to data center energy consumption, including power load, current workload, historical electricity price, ambient temperature and other factors. During the data collection process, we selected several representative data centers, covering computer rooms of different sizes and covering operation data of different time periods. The frequency of data collection for each data center is once every 15 minutes to ensure real-time monitoring and reflection of load fluctuations. Based on these collected data, we built a data center energy consumption model. Specifically, the model is trained with characteristic data such as historical load, temperature, and workload, and uses multiple regression methods to predict future energy consumption. The key goal of this model is to reasonably infer the future energy consumption changes of the data center based on the changes in workload, so as to provide data support for subsequent scheduling optimization. Next, we proposed an electricity price and workload prediction model based on LSTM neural network. Through this model, we can accurately predict the trend of electricity prices and workload changes in the future. LSTM neural network is a deep learning model that can process time series data, especially suitable for highly time-series data such as electricity price and workload prediction. Through training with historical electricity price and load data, the LSTM model can accurately capture the laws of electricity price fluctuations and load changes. In order to achieve the optimal scheduling of data center load, this experiment further constructed a scheduling model with minimizing the cost of electricity purchase as the objective function. The model takes into account constraints such as electricity price fluctuations, workload forecasts, and the load carrying capacity of the data center itself, aiming to reduce unnecessary electricity expenditures by optimizing the scheduling strategy. During the scheduling process, the model will dynamically adjust the workload of each node according to the real-time electricity price and load forecast to avoid excessive electricity consumption during high electricity price periods, thereby effectively reducing the overall operating cost. The constructed load optimization scheduling model was simulated and verified using the IEEE30-node system. Through simulation, the electricity price trend, the workload of each node, and the corresponding scheduling results within the forecast day can be obtained. The core indicators of the simulation results are the daily electricity purchase cost, the load distribution in each time period, and the response efficiency of the system under different electricity price fluctuations.

[0073] Table 1 Experimental data table

[0074]

[0075]

[0076] The electricity price and workload prediction model based on LSTM neural network successfully optimized the load scheduling of data centers in different time periods and effectively reduced the cost of purchasing electricity. During the implementation process, each data center dynamically adjusted its load scheduling according to the real-time electricity price prediction and workload changes.

[0077] The electricity price is generally low during 00:00-01:00, but rises during 01:00-02:00. Through the prediction of the model, the data center can adjust its load scheduling in advance to adapt to the optimal energy consumption configuration under different electricity price levels. For example, the electricity price of data center 1 in the first period is 0.45 yuan / kWh, the workload is 200kW, the scheduling volume is 150kW, and the electricity purchase cost is 90 yuan; in the second period, the electricity price rises to 0.48 yuan / kWh, the workload is 220kW, the scheduling volume increases to 170kW, and the electricity purchase cost increases to 99 yuan accordingly. This shows that through load scheduling optimization, the data center adopts a relatively energy-saving strategy during the period of high electricity prices, avoiding excessive electricity purchases during peak hours, thereby reducing operating costs. By comparing the electricity purchase costs of different data centers, it can be found that although their workloads and electricity price levels are different, all data centers have achieved lower electricity purchase costs through the scheduling optimization of the model.

[0078] When the electricity price is low, Data Center 3 can further reduce the load, and the electricity cost is reduced to 75.6 yuan. When the electricity price rises, the electricity cost is 94.4 yuan. This trend is in line with the expectations of electricity price fluctuations and load scheduling. By comparing with traditional load scheduling methods, this experiment shows the advantages of LSTM model in dealing with electricity price and workload prediction. Traditional methods may not be able to fully capture the complex relationship between electricity price and load changes, resulting in failure to adjust the load in time during peak electricity price periods. However, by learning from historical data, LSTM neural network can more accurately predict future electricity price trends, and dynamically adjust workloads based on these prediction results, achieving a more energy-saving and economical scheduling solution.

[0079] The use of the LSTM neural network-based model shows obvious innovation and novelty in data center load optimization scheduling, which not only improves scheduling efficiency but also effectively reduces operating costs.

[0080] Embodiment 3 is an embodiment of the present invention, which provides a data center energy optimization system based on electricity price and workload prediction in a market environment, including a data collection and preprocessing module, a prediction and modeling module, and a load scheduling optimization and decision-making module.

[0081] The data acquisition and preprocessing module is used to obtain the number and type of online loads and offline loads, record the changes in data center workloads in different time periods, obtain historical electricity price data including market electricity price fluctuations to facilitate electricity price forecasts, obtain power consumption data of IT equipment, refrigeration systems, UPS power supplies, power distribution systems, and lighting systems as inputs to energy consumption models, convert the format of the collected data, clean it, and denoise it to ensure data accuracy and consistency.

[0082] The prediction and modeling module is used to use historical electricity price data to predict the fluctuation trend of future electricity prices through the LSTM model. The LSTM model improves the accuracy of electricity price prediction by processing the relationship between short-term and long-term electricity price fluctuations. The historical workload data is used to train the LSTM model to predict future workload demand to ensure that the trend of load changes can be accurately reflected in future scheduling decisions. The electricity price and workload prediction results are used to build an energy consumption model of the data center, define load optimization scheduling goals and formulate constraints.

[0083] The load scheduling optimization and decision-making is used to input the electricity price fluctuations and workload demands predicted by the LSTM model into the load optimization scheduling decision model, optimize the load distribution between time periods according to the optimization scheduling model, electricity price fluctuations and load demands, postpone some offline loads to the period of low electricity prices to reduce overall energy consumption, and use simulation tools to verify the optimization scheduling results, calculate the electricity price, workload, scheduling status and electricity purchase cost on the predicted day, and further optimize the scheduling strategy, calculate the electricity bill according to the scheduling optimization results, and demonstrate the economic benefits of optimized scheduling.

[0084] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0085] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0086] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0087] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A data center energy optimization method based on electricity price and load forecasting, characterized in that: include: Collect parameters related to data center energy consumption and build a data center energy consumption model based on workload transfer; Propose an electricity price and workload prediction model based on LSTM neural network; Taking minimizing the electricity purchase cost as the objective function and considering the constraints, a data center load optimization scheduling decision model is constructed under the market environment; The model is simulated and verified using the IEEE30-node system to obtain the electricity price on the predicted day, the arriving workload, the workload scheduling, and the electricity purchase cost.

2. The data center energy optimization method based on electricity price and load forecasting according to claim 1, characterized in that: The construction of the data center energy consumption model includes that the energy consumption of the data center is highly correlated with the number and type of workloads, and based on this relationship, an energy consumption model based on data center workload processing can be constructed.

3. The data center energy optimization method based on electricity price and load forecasting according to claim 2, characterized in that: The LSTM neural network includes calculating the coefficients of three gates and calculating the current neuron candidate state quantity based on the current input value and the output value at the previous moment, and the forget gate and the input gate determine the proportion of the state value at the previous moment and the candidate state value at the current moment in the new state value and calculate the output at the current moment.

4. The data center energy optimization method based on electricity price and load forecasting according to claim 3, characterized in that: The construction of the data center load optimization scheduling decision model includes inputting the prediction results into the decision model for solving, and the data center schedules the workload between time periods according to the solution results. The decision model needs to meet offline load constraints, service quality requirement constraints, and data center capacity constraints.

5. The data center energy optimization method based on electricity price and load forecasting according to claim 4, characterized in that: The simulation verification includes taking the relevant parameters of the data center, the clearing electricity prices in the day-ahead market in each period of the area where the data center is located, and the load conditions that need to be processed as input conditions of the decision model, obtaining the scheduling status of the data center load and calculating the electricity expenditure of the data center before and after the workload transfer based on the scheduling status.

6. The data center energy optimization method based on electricity price and load forecasting according to claim 5, characterized in that: The purpose of the LSTM model includes obtaining the electricity price and workload demand for each time period of the next day by constructing an electricity price and workload prediction model, and using the obtained data as the basis for data center load optimization scheduling decisions.

7. The data center energy optimization method based on electricity price and load forecasting according to claim 6, characterized in that: The data center scheduling includes that, under the premise of satisfying the constraints, the data center arranges offline loads to be processed in the period with low electricity price as much as possible, and when the electricity price is high, the data center will postpone the processing of offline loads; The optimized scheduling decision model includes adding a joint model of short-term and long-term electricity price fluctuation correlation in the prediction link, which not only refines the electricity price information of the time period but also provides general direction support for the overall scheduling, and adds a prediction error correction mechanism in the scheduling link to ensure that the error caused by short-term and long-term electricity price fluctuations to the prediction link is minimized.

8. A system using the data center energy optimization method based on electricity price and load forecasting as claimed in any one of claims 1 to 7, characterized in that: Including data acquisition and preprocessing module, prediction and modeling module, load scheduling optimization and decision-making module; The data acquisition and preprocessing module is used to obtain the number and type of online loads and offline loads, record the changes in data center workloads at different times, obtain historical electricity price data including the fluctuation patterns of market electricity prices, so as to make electricity price forecasts, obtain power consumption data of IT equipment, refrigeration systems, UPS power supplies, power distribution systems, and lighting systems as inputs to energy consumption models, and perform format conversion, cleaning, and denoising on the collected data to ensure data accuracy and consistency; The prediction and modeling module is used to use historical electricity price data to predict the fluctuation trend of future electricity prices through the LSTM model. The LSTM model improves the accuracy of electricity price prediction by processing the relationship between short-term and long-term electricity price fluctuations. The historical workload data is used to train the LSTM model to predict future workload demand to ensure that the trend of load changes can be accurately reflected in future scheduling decisions. The electricity price and workload prediction results are used to build an energy consumption model of the data center, define load optimization scheduling goals and formulate constraints; The load scheduling optimization and decision-making is used to input the electricity price fluctuations and workload demands predicted by the LSTM model into the load optimization scheduling decision model, optimize the load distribution between time periods according to the optimization scheduling model, electricity price fluctuations and load demands, postpone some offline loads to the period of low electricity prices to reduce overall energy consumption, and use simulation tools to verify the optimization scheduling results, calculate the electricity price, workload, scheduling status and electricity purchase cost on the predicted day, and further optimize the scheduling strategy, calculate the electricity bill according to the scheduling optimization results, and demonstrate the economic benefits of optimized scheduling.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the data center energy optimization method based on electricity price and load forecasting described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the data center energy optimization method based on electricity price and load forecasting described in any one of claims 1 to 7 are implemented.

Images