Distributed optimization method for integrated demand response considering waste heat of data center

Abstract

The present application relates to a kind of data center participation integrated demand response distributed optimization method considering waste heat, belong to power system economic dispatch field, with the following beneficial effects: 1) the data center distributed optimization method proposed, detailed data center internal energy flow, data flow model is constructed, not only improve data center waste heat utilization rate, and in DR scheduling process, meet the demand of data center for data privacy protection.2) on the basis of considering that data center is equipped with back pressure turbine as energy conversion device, the present application makes full use of the heat and power conversion capacity of data center, constructs the integrated demand response model of data center to reduce total purchase energy cost.3) by using the distributed optimization algorithm based on alternating direction multiplier method, since data center only exchanges non-sensitive data, and each data center uses the proposed distributed algorithm to optimize independently, the protection of data center sensitive data can be realized.

Description

Technical Field

[0001] This invention belongs to the field of economic dispatch of power systems, and specifically relates to a distributed optimization method for integrated demand response that takes into account waste heat and involves data centers. Background Technology

[0002] To meet the ever-increasing demand for online computing, data centers (DCs) providing information technology services have developed rapidly in recent years. Modern DCs typically have high occupancy rates, resulting in extremely high costs. As a new type of demand response (DR) resource, DCs possess powerful DR capabilities.

[0003] There is a correlation between data center energy consumption and network load. In terms of time, data center network load includes both real-time user requests and tasks that allow for latency, such as large-scale data analysis; therefore, data centers possess potential for time-based load balancing. In terms of space, data network load balancing is fast and highly feasible. Load balancing can be used to replace long-distance power transmission. Large telecommunications operators and internet companies typically deploy data centers in different regions; the interconnectedness of data links and their geographical dispersion offer potential for cross-time zone and cross-regional load balancing.

[0004] Data center servers generate a significant amount of waste heat during operation. This waste heat is easily extracted and has abundant heat sources, making its recovery and reuse a promising application in data centers. The recovered heat can be used for domestic hot water and heating, thus reducing carbon emissions from heating. Alternatively, it can be sold to offset data center operating costs.

[0005] Furthermore, an increasing number of studies are focusing on privacy protection issues in areas such as multi-agent interaction and collaboration. Data sharing plays a crucial role in the Internet of Things (IoT), but due to the massive volume and diverse transmission methods, the leakage of private information poses new security threats to users. Therefore, distributed optimization theory and applications are gradually permeating all aspects of scientific research, engineering applications, and social life. Distributed optimization effectively achieves optimization tasks through cooperation and coordination among multiple agents, and can be used to solve many large-scale optimization problems and privacy protection issues that centralized algorithms struggle to address. Summary of the Invention

[0006] In view of the above-mentioned prior art, the technical problem solved by the present invention is to provide a distributed optimization method for data center participation in comprehensive demand response that takes into account waste heat utilization, which is used to solve at least: 1) the problem of insufficient utilization of waste heat of data center coolant 2) the problem that traditional centralized data center DR scheduling algorithm cannot meet the data privacy protection problem.

[0007] To address the aforementioned technical problems, this invention proposes a distributed optimization method for data centers participating in integrated demand response, which takes waste heat into account. The method obtains the electricity and heat purchase demands of the data center at different times by finding the optimal value of the distributed optimization integrated demand response model of the data center under constraints.

[0008] The distributed optimization comprehensive demand response model for the data center is as follows:

[0009]

[0010] The constraints that this model must satisfy are:

[0011]

[0012]

[0013]

[0014]

[0015]

[0016]

[0017]

[0018]

[0019]

[0020]

[0021]

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034] In the formula:

[0035] A collection of back-pressure turbines;

[0036] A collection of electric refrigeration units;

[0037] A collection of energy storage devices;

[0038] It is a collection of double-effect absorption chillers;

[0039] For data centers d device c At any moment t The cooling power required;

[0040] A collection of thermal storage devices;

[0041] A collection of waste heat collection devices;

[0042] Celec d,t For data centers d At any moment t Electricity costs;

[0043] Cheat d,t For data centers d At any moment t The cost of heating;

[0044] Coam d,t For data centers d At any moment t Maintenance costs;

[0045] For data centers d device c Coefficient of performance (COP)

[0046] For data centers d electric refrigeration unit c1 Secondary heating performance coefficient;

[0047] For data centers d Dual-effect absorption chiller c2 Secondary heating performance coefficient;

[0048] For capacity redundancy;

[0049] CR d For data centers d Capacity redundancy requirements;

[0050] For data center collection;

[0051] In addition to data centers d A collection of data centers other than those mentioned above;

[0052] f d,n For data centers d server n Operating frequency;

[0053] f j For computational tasks j CPU usage frequency;

[0054] H abs This represents the heat absorption power vector of the thermal storage device.

[0055] For data centers d device c At any moment t The heat absorption power;

[0056] For data centers d At any moment t Other cooling power requirements;

[0057] H DAC,in The heat input power vector of a double-effect absorption chiller;

[0058] For data centers d devicec At any moment t Input thermal power;

[0059] For data centers d At any moment t The heat power consumed after the participation of integrated demand response;

[0060] For data centers d At any moment t The heat dissipation capacity of the memory module;

[0061] For data centers d At any moment t ;

[0062] For data centers d device c At any moment t ;

[0063] For data centers d At any moment t Heat dissipation power of other electronic components in the server;

[0064] For data centers d At any moment t Processor's heat dissipation power;

[0065] For data centers d At any moment t Purchase heat capacity;

[0066] For data centers d device c At any moment t The heat dissipation power;

[0067] H rel This is the heat release power vector of the thermal storage device;

[0068] H WHRD,in This is the inlet heat power vector of the waste heat recovery device;

[0069] For data centers d server n At any moment t The power-on indicator value;

[0070] For data centersd server n At any moment t The power-off indicator value;

[0071] For computational tasks j At any moment t The start processing instruction quantity;

[0072] To compute the task at time... t The amount of processing completed;

[0073] For the set of computational tasks;

[0074] A set of non-delayable computation tasks;

[0075] L d For data centers d The Lagrange objective function;

[0076] For a set of servers;

[0077] Nj It is a decoupled decision vector within the data center;

[0078] = × × × ;

[0079] For data centers d The total number of computational tasks;

[0080] For data centers d The initial number of computational tasks;

[0081] For data centers d The number of tasks to be transferred out;

[0082] For data centers d The number of tasks transferred to the calculation;

[0083] For data centers d The total number of deferred computing tasks;

[0084] For data centers Number of deferred computing tasks transferred in;

[0085] For data centers d Number of deferred computing tasks transferred out;

[0086] NjDd For data centers d A vector of deferred computational tasks;

[0087] For data centers d to other data centers Number of deferred computing tasks transferred out;

[0088] For data centers A vector of deferred computational tasks;

[0089] For other data centers From data center d Number of deferred computing tasks transferred in;

[0090] For data centers d The total number of non-delayable computation tasks;

[0091] For data centers d Number of non-delayable computation tasks transferred out;

[0092] For data centers The number of non-delayable computation tasks transferred in;

[0093] NjNd For data centers d The vector of non-delayable computational tasks;

[0094] For data centers d to other data centers Number of non-delayable computation tasks transferred out;

[0095] For data centers The vector of non-delayable computational tasks;

[0096] For other data centers From data center d The number of non-delayable computation tasks transferred in;

[0097] P cha The charging power vector for the energy storage device;

[0098] For data centersd device c At any moment t The charging power;

[0099] For data centers d device c At any moment t The discharge power;

[0100] P dis This is the discharge power vector of the energy storage device;

[0101] P EC,in The input power vector of the electric chiller;

[0102] For data centers d device c At any moment t The incoming power;

[0103] For data centers d At any moment t The power consumption after the participation of integrated demand response;

[0104] For data centers d At any moment t The power consumption of the lighting system;

[0105] For data centers d device c At any moment t ; output electrical power;

[0106] For data centers d At any moment t The power consumption of the water pump;

[0107] For data centers d At any moment t The power consumption of electricity purchased;

[0108] P redu This is the power reduction vector for electrical loads;

[0109] For data centers d At any moment t The power consumption of the power supply system;

[0110] For data centers d At any momentt The power consumption of the server;

[0111] Rcomp d,t For data centers d At any moment t Demand response compensation revenue; For computational tasks j At any moment t The execution status;

[0112] For computational tasks j At any moment t-1 The execution status;

[0113] For computational tasks j time t In data center d server n The execution status;

[0114] For data centers d server n At any moment t The running status;

[0115] For data centers d server n At any moment t-1 The running status;

[0116] NT This is the last scheduling period of the day;

[0117] For the set of scheduling periods;

[0118] To calculate the arrival time of the task;

[0119] This is to calculate the completion time of the task;

[0120] For computational tasks j The computational processing time;

[0121] For computational tasks j The task start processing time;

[0122] For non-delayed computing tasks j N The task start processing time;

[0123] For data centers d server n At any moment t Utilization rate;

[0124] X It is the decision vector within the data center;

[0125] X = P redu × P EC,in × H DAC,in × H WHRD,in × P cha × P dis × H abs × H rel ;

[0126] λ It is a Lagrange multiplier vector;

[0127] ρ It is a punishment factor;

[0128] λ = × ;

[0129] λ NJ For non-delayable computation tasks, the Lagrange multiplier vectors are used.

[0130] λ DJ For a deferred computation task, the Lagrange multiplier vector is used.

[0131] For data centers d and Data Center The non-delayable computation task of Lagrange multiplier vectors;

[0132] For data centers d and Data Center The deferred computation task of Lagrange multiplier vectors.

[0133] In the above technical solution, the present invention has the following beneficial effects:

[0134] 1) The distributed optimization method for data centers proposed in this invention constructs a detailed model of energy flow and data flow within the data center, which not only improves the utilization rate of waste heat in the data center, but also meets the data privacy protection requirements of the data center during the DR scheduling process.

[0135] 2) Taking into account that data centers are equipped with back-pressure turbines as energy conversion devices, this invention makes full use of the thermoelectric conversion capabilities of data centers to construct a comprehensive demand response model for data centers to reduce total energy purchase costs.

[0136] 3) This invention employs a distributed optimization algorithm based on the alternating direction multiplier method. Since data centers only exchange non-sensitive data, and each data center independently optimizes using the proposed distributed algorithm, the invention can protect sensitive data in the data centers. Attached Figure Description

[0137] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0138] Figure 1 A schematic diagram of energy flow and data flow in a data center;

[0139] Figure 2 This is a schematic diagram of a hybrid cooling system for a data center.

[0140] Figure 3 A chart showing the day-ahead electricity prices for the three regions;

[0141] Figure 4 This diagram illustrates the relationship between average server utilization in a data center and electricity prices.

[0142] Figure 5a This is a power balance diagram for data center DC1.

[0143] Figure 5b This is a power balance diagram for data center DC2;

[0144] Figure 5c This is a power balance diagram for data center DC3;

[0145] Figure 6a A schematic diagram comparing the proportion of back-pressure turbine output power to total power supply under different electro-thermal conversion coefficients in data center DC1;

[0146] Figure 6bA schematic diagram comparing the proportion of back-pressure turbine output power to total power supply under different electro-thermal conversion coefficients in data center DC2;

[0147] Figure 6c A schematic diagram comparing the proportion of back-pressure turbine output power to total power supply under different electro-thermal conversion coefficients in data center DC3;

[0148] Figure 7a This is a schematic diagram of the thermal balance in data center DC1;

[0149] Figure 7b This is a schematic diagram of the thermal balance in data center DC2;

[0150] Figure 7c This is a schematic diagram of the thermal balance in data center DC3. Detailed Implementation

[0151] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0152] The parameters involved in one embodiment of the present invention are explained in Table 1.

[0153] Table 1

[0154]

[0155] Continued from Table 1

[0156]

[0157] Continued from Table 1

[0158]

[0159] Continued from Table 1

[0160]

[0161] Continued from Table 1

[0162]

[0163] Continued from Table 1

[0164]

[0165] Continued from Table 1

[0166]

[0167] Continued from Table 1

[0168]

[0169] Figure 1 This diagram illustrates the energy and data flows within a data center. Data flow within a data center includes day-ahead electricity prices, heat prices, demand response compensation prices from the system operator, DR plans submitted to the system operator, and processing information for computational tasks. The data center contains three types of energy flows: current, heat, and cooling.

[0170] Figure 2 This is a diagram of a hybrid cooling system for a data center. The closed-loop water system between the liquid-cooled plate dual-effect absorption chiller and the server is designed to eliminate heat dissipation from the processor and memory modules. In the closed-loop water system, after the cooled water is pumped to absorb heat from the processor and memory, the hot water is treated in a waste heat recovery unit. The waste heat recovery unit collects the waste heat retained in the hot water through further treatment. The hot water is then sent to the dual-effect absorption chiller, and the collected reusable waste heat is used to heat the dual-effect absorption chiller. Compared to the processor and memory, other electronic components in the server with lower power density (such as disks and chipsets) generate less heat during server operation. For these components, air cooling is used for heat dissipation.

[0171] Power-consuming components in a data center include servers, electric chillers, and water pumps. Heat-consuming components are double-effect absorption chillers. In a hybrid cooling system, double-effect absorption chillers and water pumps are the main components for liquid cooling, while electric chillers are used for air cooling. Double-effect absorption chillers and electric chillers provide cooling energy for server heat dissipation. Considering that heat dissipated using liquid cooling remains in the coolant, waste heat collection devices are used for waste heat collection and treatment. Energy storage components include electrical storage devices and thermal storage devices.

[0172] Data center model

[0173] The server model is expressed as follows:

[0174] (1)

[0175] (2)

[0176] (3)

[0177] (4)

[0178] Since a large portion of a server's power consumption is ultimately converted into heat, effectively recovering and utilizing waste heat from data centers holds significant promise for energy optimization. The total heat generated in a server includes heat from the processor, memory modules, and other electronic components. Air cooling and water cooling are two typical cooling methods in modern hybrid cooling systems. Electric chillers use electrically driven air cooling, while dual-effect absorption chillers use thermally driven liquid cooling. In modern data centers, for server heat dissipation, water cooling is typically used for cooling the processor and memory, while air cooling is typically used for cooling other electronic components.

[0179] The hybrid cooling system model is expressed as follows:

[0180] (5)

[0181] (6)

[0182] (7)

[0183] (8)

[0184] (9)

[0185] (10)

[0186] (11)

[0187] (12)

[0188] (13)

[0189] (14)

[0190] (15)

[0191] (16)

[0192] Considering that the heat generated by liquid cooling remains in the coolant, this invention designs a waste heat recovery process to reuse the waste heat. Specifically, this invention uses a waste heat recovery device to further process the waste heat in the water.

[0193] The waste heat recovery device model is expressed as follows:

[0194] (17)

[0195] (18)

[0196] (19)

[0197] The back-pressure turbine model is as follows:

[0198] (20)

[0199] (twenty one)

[0200] (twenty two)

[0201] The thermal storage device model is expressed as follows:

[0202] (twenty three)

[0203] (twenty four)

[0204] (25)

[0205] (26)

[0206] (27)

[0207] The energy storage device model is expressed as follows:

[0208] (28)

[0209] (29)

[0210] (30)

[0211] (31)

[0212] (32)

[0213] Data Center Integrated Demand Response Model

[0214] The comprehensive demand response capability of a data center is defined as the ability to transfer electricity and heat across time and space. A data center with comprehensive demand response capability can respond sensitively to changes in energy prices over time and space. For collaborative data centers, when one data center handles too many CPU-intensive computing tasks, some of these tasks can be distributed to other data centers to meet service quality requirements. In response to reduction requests from system operators, collaborative data centers will rearrange computing tasks to minimize costs. The comprehensive demand response model for data centers is expressed as follows:

[0215] (33)

[0216] (34)

[0217] (35)

[0218] (36)

[0219] (37)

[0220] (38)

[0221] (39)

[0222] (40)

[0223] (41)

[0224] (42)

[0225] (43)

[0226] (44)

[0227] (45)

[0228] (46)

[0229] (47)

[0230] (48)

[0231] (49)

[0232] (50)

[0233] (51)

[0234] In this invention's model, multiple collaborative data centers are located on different nodes and have different energy prices. Based on the day-ahead electricity price, the data centers rearrange their computing job allocation and energy consumption plans to minimize their daily costs. The objective function formula for the data centers is as follows:

[0235] (52)

[0236] in:

[0237] (53)

[0238] (54)

[0239] (55)

[0240] (56)

[0241] (57)

[0242] (58)

[0243] (59)

[0244] (60)

[0245] In addition to the power consumption mentioned above, the power consumption of air conditioning, power supply, and lighting systems also accounts for a significant portion of the total power consumption of a data center. The power balance constraints for each data center are as follows:

[0246] (61)

[0247] (62)

[0248] (63)

[0249] Distributed optimization algorithm based on alternating direction multiplier method

[0250] This invention employs a distributed algorithm based on the alternating direction multiplier method. In the distributed scheduling mode, participants do not need to submit internal data to the scheduling center. Data centers only need to exchange non-sensitive data. Therefore, distributed optimization can effectively protect sensitive data within data centers. Each data center makes independent optimization decisions and adjusts its demand response scheduling scheme based on current optimization results and historical data exchange.

[0251] In the centralized scheduling mode, the computational task transfer variables need to be decoupled in the distributed scheduling mode. In the distributed mode, the transfer of computational jobs between two data centers is determined by the amount of computational tasks moved in from one data center and the amount of computational tasks moved out from the other data center. Therefore, the computational task transfer variables are decoupled into replication variables within the two data centers. The two replication variables should be equal. Therefore, equations (36)-(39) are rewritten as (64)-(67). The constraint relationship between replication variables is expressed as (68)-(69).

[0252] (64)

[0253] (65)

[0254] (66)

[0255] (67)

[0256] (68)

[0257] (69)

[0258] The distributed optimization integrated demand response model for data centers is rewritten as follows:

[0259] (70)

[0260] (71)

[0261] (72)

[0262] (73)

[0263] To solve the distributed problem for each data center, the original distributed optimization model is further decomposed into several sub-problems, and each sub-problem is optimized separately. For each sub-problem, the alternating direction multiplier algorithm is used to transform the constraints connecting the variables of other sub-problems into penalty terms of the objective function. Therefore, the augmented Lagrangian method is applied to rewrite the comprehensive demand response model for each data center as follows:

[0264] (74)

[0265] (75)

[0266] In the formula:

[0267] X = P redu × P EC,in × H DAC,in × H WHRD,in × P cha × P dis × H abs × H rel ;

[0268] = × × × ;

[0269] λ = × .

[0270] The key rule for implementing the alternating direction multiplier method is that in each step, only one variable needs to be optimized, while the other variables remain fixed with their latest updated values. Let... r This is the index for the iteration. The iterative update process of the alternating direction multiplier method is as follows:

[0271] (76)

[0272] (77)

[0273] Utilize the updated X r+1 and Nj r+1 Lagrange multipliers λ Updated using the subgradient method:

[0274] (78)

[0275] Repeat the above update process until the error factor is reached. e 1 and e Both met the convergence criteria.

[0276] (79)

[0277] (80)

[0278] In the formula: The current objective function value is updated by equations (76)-(78).

[0279] Through the above description of the embodiments, those skilled in the art can clearly understand that the method of this disclosure can be implemented by means of software plus necessary general-purpose hardware, and of course, it can also be implemented by special-purpose hardware, including application-specific integrated circuits, dedicated CPUs, dedicated memory, dedicated components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or dedicated circuits. However, for the purposes of this disclosure, software program implementation is more often a preferred implementation method.

[0280] Case Analysis

[0281] To verify the effectiveness of the distributed optimization algorithm proposed in this invention, the solution was obtained using the distributed optimization method for data center participation in comprehensive demand response, considering waste heat utilization, designed in this invention, under the MATLAB 2020a environment. The system hardware configuration was an i7-9700 CPU (3.00GHz), 16.00 GB of RAM, and a Win10 64-bit operating system. The scenario of this invention includes one system operator and three data centers (DCs). The three DCs are denoted as DC1, DC2, and DC3. The day is divided into 24 time periods. The daytime demand response (DR) period published by the system operator is from 1:00 PM to 9:00 PM, during which the DR compensation price for load reduction is $5 / MWh.

[0282] DC1, DC2, and DC3 are located in three different regions of the US PJM market: BC, COMED, ​​and DEOK. Figure 3 This table shows the day-ahead electricity prices for three regions from 00:00 to 23:00 Central Time on July 9, 2021. Heat prices for the three regions are considered constant at $20 / MWh, $15 / MWh, and $18 / MWh, respectively. Each data center (DC) contains 100 batch servers, and each batch server aggregates 10,000 sub-servers. The CPU frequency of the sub-servers is 3.00 GHz. Details of the sub-servers are shown in Table 1. Server utilization is capped at 90%. Information on computing jobs allocated to the data center is shown in Table 2. Details of each energy component in the DC are shown in Table 3. The electrothermal conversion efficiency of the electronic components is 0.7.

[0283] Table 1 Server Information

[0284]

[0285] Table 2 Data Center Computing Job Information

[0286]

[0287] Table 3 Information on various energy components in the data center

[0288]

[0289] To analyze the impact of DC participation in DR and whether it has IDR capability on its economic benefits, Table 4 shows the energy purchase cost and energy purchased by DCs under four scenarios. The scheduling results of scenario 4 are used as a baseline. Clearly, when a DC participates in DR and has IDR capability, its daily energy purchase cost is the lowest. Compared to scenario 4, the total energy purchase cost in scenarios 1 and 2 increases by 24.19% and 5.64%, respectively. When a DC does not participate in DR, it cannot rearrange the allocation of calculation tasks among the DCs based on electricity price information, leading to increased energy purchase costs. Furthermore, DCs without IDR capability cannot respond sensitively to electricity and heat price information. Specifically, when the heat price is much higher than the electricity price, DCs with IDR capability prefer to purchase heat and convert it into electricity rather than directly purchase electricity. As mentioned above, Table 4 verifies the crucial role of participating in DR and having IDR capability in reducing energy purchase costs. Compared to the scheduling results of scenario 4, the rate of change in scenario 3 is less than 0.4%. The similarity between cases 3 and 4 verifies the accuracy of the results of the distributed algorithm, indicating that the proposed distributed algorithm can minimize the total energy cost while effectively protecting data privacy.

[0290] Table 4. Daily energy purchase cost and energy purchase for data centers under four scenarios.

[0291]

[0292] Continued from Table 4

[0293]

[0294] Scenario 1: Not involved in DR, no IDR capability.

[0295] Scenario 2: Not participating in DR, but having IDR capability.

[0296] Scenario 3: Distributed scheduling: Participates in DR and has IDR capability.

[0297] Scenario 4: Centralized scheduling: Participates in DR and has IDR capability.

[0298] Figure 4This chart shows the relationship between average server utilization rate (ASUR) and electricity price for three data centers. To minimize daily costs for the data centers, ASUR is relatively low during peak electricity price periods. Conversely, ASUR is quite high during off-peak electricity price periods. Specifically, from 0:00 AM to 8:00 AM, ASUR in all three data centers approaches the upper limit of server utilization. Furthermore, due to the lowest electricity price in DC2, servers in DC2 undertook the most computing tasks after rescheduling. The lowest electricity price occurred after 9:00 AM, at 10:00 AM. To fully utilize server computing power, ASUR in DC2 remained high at 10:00 AM. Figure 4 It has been demonstrated that the proposed model can reasonably allocate computational tasks based on price signals.

[0299] Figure 5a , Figure 5b , Figure 5c A power balance diagram for three data centers is shown. Power supply and demand are maintained in balance at all times. Clearly, DC1 consumes the least power among the three DCs. Combined with... Figure 5a The server in DC1 consumes the least power after rescheduling because the electricity price in DC1 is the highest among the three regions. Electricity prices show a similar trend across the three regions: they reach their lowest point at 4:00 AM and their highest point at 5:00 PM. Therefore, by charging during off-peak hours and discharging during peak hours, electricity costs can be effectively reduced. Consequently, energy storage devices in different DCs are charged at 4:00 AM and 5:00 AM, and discharged at 5:00 PM and 6:00 PM. The total number of charging and discharging cycles is constrained by the maximum charge / discharge cycle of the energy storage device.

[0300] To analyze the influence of the thermal power coefficient of BPT (back pressure turbine), Figure 6a , Figure 6b , Figure 6c This depicts the proportion of BPT output power to total power supply under different electro-thermal conversion coefficients. When the electro-thermal conversion coefficient is small (i.e., γ HtP =0.25), when the demand for electrical energy output from the BPT is the same, the BPT consumes more heat energy. To minimize the total energy purchase cost, the DC only chooses to purchase heat and convert it into electrical energy when the price of electricity is much higher than the price of heat energy (i.e., DC1 from 1:00 PM to 7:00 PM, DC2 from 12:00 AM to 9:00 PM, and DC3 from 2:00 PM to 7:00 PM). Otherwise, the DC prefers to purchase electricity directly from the grid. When the electro-thermal conversion coefficient is large (i.e., ... γ HtPWhen the electro-thermal conversion coefficient is 0.35, obtaining electricity through heat-to-electric conversion via BPT becomes more attractive. In this case, the period during the day when DC obtains electricity through heat-to-electric conversion increases significantly. Furthermore, the higher the electro-thermal conversion coefficient, the higher the proportion of BPT output power in the total power supply.

[0301] Figure 7a , Figure 7b , Figure 7c A schematic diagram illustrating the thermal balance of three data centers is presented. Heat supply and demand remain in balance at all times. Among the three DCs, DC1 has the lowest thermal power consumption, consistent with the electrical power consumption of the three DCs. Since heat prices are not time-varying, DCs cannot improve their economic efficiency by absorbing heat during periods of low prices and releasing heat during periods of high prices. Therefore, it is meaningless for DCs to schedule the release of stored heat energy from the thermal storage devices for economic reasons after the initial heat energy storage has been released. Figures 7a-7c As shown, unlike energy storage devices, thermal energy storage devices only release heat energy during the first period of the day. Since the amount of waste heat recovered by a waste heat recovery device depends on the heat dissipation of the double-effect absorption chiller, more waste heat will be recovered and utilized during periods of high heat dissipation from the double-effect absorption chiller. Therefore, when the heat demand of the double-effect absorption chiller accounts for a high proportion of the total heat demand (i.e., DC1 from 0:00 AM to 11:00 AM and 11:00 PM, DC2 from 2:00 AM to 7:00 AM, and DC3 from 1:00 AM to 9:00 AM), the reused waste heat plays a significant role in the total heat supply.

[0302] Although embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the specific embodiments and application fields described above. The specific embodiments described above are merely illustrative and instructive, and not restrictive. Those skilled in the art can make many other forms based on the guidance of this specification and without departing from the scope of protection of the claims of the present invention, and all of these are within the scope of protection of the present invention.

Claims

1. A distributed optimization method for data centers participating in integrated demand response, considering waste heat, characterized in that, The method obtains the electricity and heat purchase demands of the data center at different times by finding the optimal value of the distributed optimization integrated demand response model of the data center under constraints. The distributed optimization comprehensive demand response model for the data center is as follows: ； in: The constraints that this model must satisfy are: In the formula: A collection of back-pressure turbines; A collection of electric refrigeration units; A collection of energy storage devices; It is a collection of double-effect absorption chillers; For data centers d At any moment t Server maintenance costs; Let the cooling power of device c in data center d be measured at time t. A collection of thermal storage devices; A collection of waste heat collection devices; Celec d,t For data centers d At any moment t Electricity costs, For data centers d At any moment t Electricity prices; Cheat d,t For data centers d At any moment t heating costs For data centers d The price of heat; Coam d,t For data centers d At any moment t The operation and maintenance costs This represents the operation and maintenance cost coefficient of device c in data center d. For data centers d device c Coefficient of performance (COP) For data centers d electric refrigeration unit c1 Secondary heating performance coefficient; For data centers d Dual-effect absorption chiller c2 Secondary heating performance coefficient; For capacity redundancy; CR d For data centers d Capacity redundancy requirements; For data center collection; In addition to data centers d A collection of data centers other than those mentioned above; f d,n For data centers d server n Operating frequency; f j For computational tasks j CPU usage frequency; H abs This represents the heat absorption power vector of the thermal storage device. For data centers d device c At any moment t The heat absorption power; For data centers d At any moment t Other cooling power requirements; H DAC,in The heat input power vector of a double-effect absorption chiller; For data centers d device c At any moment t Input thermal power; For data centers d At any moment t The heat power consumed after the participation of integrated demand response; For data centers d At any moment t The heat dissipation capacity of the memory module; For data centers d At any moment t ; For data centers d device c At any moment t ; For data centers d At any moment t Heat dissipation power of other electronic components in the server; For data centers d At any moment t Processor's heat dissipation power; For data centers d At any moment t Purchase heat capacity; For data centers d device c At any moment t The heat dissipation power; H rel This is the heat release power vector of the thermal storage device; H WHRD,in This is the inlet heat power vector of the waste heat recovery device; For data centers d server n At any moment t The power-on indicator value; For data centers d server n At any moment t The power-off indicator value; For computational tasks j At any moment t The start processing instruction quantity; To compute the task at time... t The amount of processing completed; For the set of computational tasks; A set of non-delayable computation tasks; L d For data centers d The Lagrange objective function; For a set of servers; The number of servers in data center d; Nj It is a decoupled decision vector within the data center; = × × × ； For data centers d The total number of computational tasks; For data centers d The initial number of computational tasks; For data centers d The number of tasks to be transferred out; For data centers d The number of tasks transferred to the calculation; For data centers d The total number of deferred computing tasks; For data centers Number of deferred computing tasks transferred in; For data centers d Number of deferred computing tasks transferred out; NjDd For data centers d A vector of deferred computational tasks; For data centers d to other data centers Number of deferred computing tasks transferred out; For data centers A vector of deferred computational tasks; For other data centers From data center d Number of deferred computing tasks transferred in; For data centers d The total number of non-delayable computation tasks; For data centers d Number of non-delayable computation tasks transferred out; For data centers The number of non-delayable computation tasks transferred in; NjNd For data centers d The vector of non-delayable computational tasks; For data centers d to other data centers Number of non-delayable computation tasks transferred out; For data centers The vector of non-delayable computational tasks; For other data centers From data center d The number of non-delayable computation tasks transferred in; P cha The charging power vector for the energy storage device; For data centers d device c At any moment t The charging power; For data centers d device c At any moment t The discharge power; P dis This represents the discharge power vector of the energy storage device. P EC,in The input power vector of the electric chiller; For data centers d device c At any moment t The incoming power; For data centers d At any moment t The power consumption after the participation of integrated demand response; For data centers d At any moment t The power consumption of the lighting system; For data centers d device c At any moment t ; output electrical power; For data centers d At any moment t The power consumption of the water pump; For data centers d At any moment t The power consumption of electricity purchased; P redu This is the power reduction vector for electrical loads; For data centers d At any moment t The power consumption of the power supply system; For data centers d At any moment t Server power consumption; Rcomp d,t For data centers d At any moment t Demand response compensation revenue, This represents the demand response compensation price at time t. Let be the power reduction amount of data center d at time t. This refers to the set of data center node locations that have demand response compensation. For computational tasks j At any moment t The execution status; For computational tasks j At any moment t-1 The execution status; For computational tasks j time t In data center d server n The execution status; For data centers d server n At any moment t The running status; For data centers d server n At any moment t-1 The running status; NT This is the last scheduling period of the day; For the set of scheduling periods; For the set of demand response periods; To calculate the arrival time of the task; This is to calculate the completion time of the task; For computational tasks j The computational processing time; For adjacent time intervals; For computational tasks j The task start processing time; For non-delayed computing tasks j N The task start processing time; For data centers d server n At any moment t Utilization rate; Let n be the server in data center d at time t. X It is the decision vector within the data center; X = P redu × P EC,in × H DAC,in × H WHRD,in × P cha × P dis × H abs × H rel ； λ It is a Lagrange multiplier vector; ρ It is a punishment factor; λ = × ； λ NJ For non-delayable computation tasks, the Lagrange multiplier vectors are used. λ DJ For a deferred computation task, the Lagrange multiplier vector is used. For data centers d and Data Center The non-delayable computation task of Lagrange multiplier vectors; For data centers d and Data Center The deferred computation task of Lagrange multiplier vectors.

Images