A computing power node adjustable power resource cooperative scheduling method and device

Abstract

The present application relates to a kind of computing power node adjustable power resource collaborative scheduling method and device, its method includes: constructing computing power node multi-resource dynamic response model;Disassemble IT load to obtain multiple types of IT load model;Build computing power node adjustable resource model;Based on multi-resource dynamic response model, multiple types of IT load model and adjustable resource, establish the source-net-load-storage collaborative optimization model with the minimum operating cost and carbon emission as objective function;Using the improved Duqiangyu fish optimization algorithm that is integrated into Tent chaos mapping initialization, golden sinusoidal strategy optimization search path and elite reverse learning mechanism, the source-net-load-storage collaborative optimization model is solved, and the optimal scheduling strategy is obtained.The present application realizes the economic low-carbon operation of computing power node under the premise of meeting the safety constraint of power grid, and improves the flexible use of system adjustable resource.

Description

Technical Field

[0001] This invention relates to the field of adjustable power resource utilization and data center node scheduling, specifically to a method and apparatus for collaborative scheduling of adjustable power resources for computing nodes. Background Technology

[0002] In today's world, the digital economy is developing rapidly, and computing power has become an important resource for social development. The continuous expansion of computing power facilities has posed a huge challenge to the scheduling of power resources. The adjustable resources in the computing power nodes of urban power distribution networks are characterized by large fluctuations and strong randomness, making it difficult to coordinate and schedule with central air conditioning, electrochemical energy storage, diesel generators and water-cooled storage.

[0003] Current research largely categorizes IT load into core and non-core tasks. However, IT load exhibits complex dynamic characteristics, with different computing tasks (real-time data processing, offline computing, redundant data processing, etc.) showing significant differences across time scales. Furthermore, the scheduling value of "electricity-cooling" coupled resources such as water-based cooling systems has not been fully recognized. These resources can alleviate the cooling burden of central air conditioning by storing cooling during off-peak hours and releasing it during peak hours, achieving source-load synergy in conjunction with other adjustable resources. Meanwhile, existing traditional algorithms suffer from slow convergence speed, poor optimization accuracy, overly concentrated search range during the exploration phase, and simplistic logic for handling inferior solutions during the development process when solving complex optimization problems with multiple objectives, multiple constraints, and high decision variable dimensionality. Summary of the Invention

[0004] This invention provides a method and apparatus for the coordinated scheduling of power resources with adjustable computing nodes, in order to solve at least one of the above-mentioned technical problems.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for coordinated scheduling of power resources with adjustable computing nodes, comprising: Construct a multi-resource dynamic response model for computing power nodes; wherein, the multi-resource dynamic response model includes a photovoltaic model and a wind power model; The overall IT load model of computing nodes is deconstructed according to the type of computing task to obtain multiple types of IT load models; among them, the multiple types of IT load models include the basic load model, the shiftable load model, the interruptible load model, and the power adjustable load model. Construct an adjustable resource model for computing power nodes; wherein, the adjustable resource model includes a central air conditioning model, an electrochemical energy storage model, a diesel generator model, and a water-cooled storage model; Based on the multi-resource dynamic response model, the multi-type IT load model, and the adjustable resources, a source-grid-load-storage collaborative optimization model is established with the objective function of minimizing operating costs and carbon emissions. An improved Cuckoo Catfish optimization algorithm, incorporating Tent chaotic mapping initialization, golden sine strategy optimization of the search path, and elite reverse learning mechanism, is used to solve the source-grid-load-storage collaborative optimization model to obtain the optimal scheduling strategy.

[0006] Based on the above-mentioned method for coordinated scheduling of power resources with adjustable computing nodes, this invention also provides a device for coordinated scheduling of power resources with adjustable computing nodes.

[0007] A computing node adjustable power resource collaborative scheduling device includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the computing node adjustable power resource collaborative scheduling method as described above.

[0008] The beneficial effects of this invention are as follows: This invention provides a method and device for collaborative scheduling of adjustable power resources on computing power nodes. Addressing the characteristics of large fluctuations and strong randomness in adjustable power resources on computing power nodes in urban power grids, it first constructs a dynamic response model for multiple resources on computing power nodes, including photovoltaic and wind power, IT load models for computing power nodes, and other adjustable resource models. Then, it deconstructs the computing power node load; IT load includes four types: basic, movable, interruptible, and power adjustable; other adjustable resources include four types: central air conditioning, electrochemical energy storage, diesel generators, and water-based cooling. Next, it establishes a source-grid-load-storage collaborative optimization model with the objective function of minimizing operating costs and carbon emissions. Finally, to solve this optimization problem, an improved Cuckoo Catfish Optimizer (CCO) algorithm is proposed. This algorithm introduces three strategies based on the Cuckoo Catfish Optimizer (CCO): Tent chaotic mapping initialization, golden sine strategy optimization search path, and elite back-learning mechanism, effectively improving the convergence speed and optimization accuracy of the algorithm. This invention enables computing nodes to operate economically and with low carbon emissions while meeting power grid security constraints, thereby improving the flexible utilization of the system's adjustable resources. Attached Figure Description

[0009] Figure 1 This is a flowchart of a method for collaborative scheduling of adjustable power resources using computing nodes, as described in this invention. Figure 2 This is a schematic diagram of the topology of the IEEE 33-node system. Figure 3 A schematic diagram of the total output curves for wind power, photovoltaic power, and renewable energy. Figure 4 Flowchart for improving the optimization algorithm for cuckoo catfish; Figure 5 This is a schematic diagram of the convergence curves of various algorithms; Figure 6A diagram showing the comparison of IT load before and after optimization; Figure 7 This is a schematic diagram of the electrochemical energy storage operation strategy curve; Figure 8 Schematic diagram of water-based cooling operation strategy curve; Figure 9 This is a schematic diagram of energy distribution on the load side and the power supply side; Figure 10 A diagram showing the comparison between the total daily operating cost and total daily carbon emissions before and after optimization; Figure 11 A schematic diagram showing the changes in voltage at IT load nodes before and after optimization. Detailed Implementation

[0010] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0011] like Figure 1 As shown, a method for coordinated scheduling of power resources with adjustable computing nodes includes: S1, Construct a multi-resource dynamic response model for computing power nodes; wherein, the multi-resource dynamic response model includes a photovoltaic model and a wind power model; S2, the total IT basic load model of computing power nodes is deconstructed according to the computing power task type to obtain multiple types of IT load models; wherein, the multiple types of IT load models include basic load model, shiftable load model, interruptible load model and power adjustable load model; S3, construct an adjustable resource model for computing power nodes; wherein, the adjustable resource model includes a central air conditioning model, an electrochemical energy storage model, a diesel generator model, and a water-cooled storage model; S4. Based on the multi-resource dynamic response model, the multi-type IT load model, and the adjustable resources, establish a source-grid-load-storage collaborative optimization model with the objective function of minimizing operating costs and carbon emissions; S5 employs an improved Cuckoo Catfish optimization algorithm that incorporates Tent chaotic mapping initialization, golden sine strategy optimization of the search path, and elite reverse learning mechanism to solve the source-network-load-storage collaborative optimization model and obtain the optimal scheduling strategy.

[0012] The steps of the method of the present invention will be described in detail below.

[0013] S1, Construct a multi-resource dynamic response model.

[0014] Photovoltaic modeling: The photovoltaic model considers the effects of light intensity and temperature, taking into account randomness and fluctuations, and conforms to the diurnal variation pattern; therefore, the photovoltaic model is expressed as: (1) In the formula, for Solar power output at all times The rated installed capacity of photovoltaic modules. for Normalized light intensity at any given time The temperature coefficient of a photovoltaic module. for The ambient temperature at any given time This is the standard test temperature for photovoltaic modules; among which, Random factors were considered as a correction term for random fluctuations: (2) In the formula, This is a baseline curve for ideal intraday light intensity. The photovoltaic power output fluctuation coefficient. To obey Randomly distributed random variables; The range of values ​​is It simulates the intraday light variation pattern and the nighttime light variation pattern. ,in: (3) In the formula, At sunrise, It is sunset time; (4) In the formula, Based on ambient temperature, This represents the amplitude of temperature fluctuations. For temperature phase shift, To obey Gaussian random variables; Photovoltaic output satisfies the non-negativity constraint: (5) Wind power modeling: Wind power modeling is based on the power-wind speed curve of wind turbines, taking into account both randomness and volatility; therefore, the wind power model is expressed as: (6) In the formula, for Wind power output at any given moment; The rated installed capacity of the wind turbine unit. for The actual wind speed at that moment, Cut-in wind speed (i.e., the minimum wind speed at which the wind turbine starts generating electricity). To cut off the wind speed, Rated wind speed; where: (7) In the formula, This is the baseline trend curve for intraday wind speed. This is the wind speed fluctuation amplitude coefficient. To simulate wind speed disturbances A uniformly distributed random variable; where: (8) In the formula, The average wind speed, For wind speed amplitude, This refers to the wind speed phase shift.

[0015] Wind power output must meet rated capacity constraints: (9) S2, deconstructing the total IT infrastructure load model of computing power nodes.

[0016] Total IT infrastructure load model for computing nodes: The time series of the total IT infrastructure load of computing nodes before they participate in scheduling needs to reflect time volatility and randomness; the total IT infrastructure load model of computing nodes simulates the dual peaks of IT load, which is expressed as: (10) In the formula, for Total IT infrastructure load power of computing nodes at any given time. This is the load baseline value. and This represents the load factor for the intraday double peak period. and This is the peak load time of the day. and The distribution width during peak hours, It is a random fluctuation term that follows a uniform distribution, used to simulate the small fluctuations in real computing tasks.

[0017] Basic load model: Basic load is a rigid load that ensures core computing power tasks, including real-time data processing and critical service response. This type of load cannot be interrupted, moved, or adjusted, and does not participate in scheduling. Its modeling satisfies the requirement of being related to the total IT basic load power. A fixed proportional relationship is established; the basic load model is expressed as: (11) In the formula, for Base load power at any given time The fixed proportion coefficient of the base load; The constraints on the base load are: (12) Transferable load model: Sliding loads are non-core, flexible loads responsible for time-sensitive computing tasks, including offline computing and batch data backup. They adjust power during peak and off-peak periods through decision variables. The sliding load model is represented as follows: (13) In the formula, for The power of the load that can be moved at any time. This is the reference ratio coefficient for movable loads. After time period attribute correction Adjustment of decision variables at different times; among them, The adjustment rules are dynamically determined by the time-of-use pricing period: (14) In the formula, , and These are the trough period, peak period, and normal period, respectively. for The original translation adjustment decision variables at time t satisfy the following constraints: (15) In the formula, and They are respectively The upper and lower bounds; The load can be moved while ensuring that the total workload remains unchanged during task time migration. Therefore, the total daily electricity consumption must be equal to that in the original state. (16) In the formula, In its original state The power of the load that can be moved at any given time.

[0018] Interruptible load model: Interruptible loads are flexible loads that can be temporarily suspended. They are responsible for non-critical computing tasks, including non-urgent rendering and redundant data cleanup. By adjusting the interruption coefficient, they reduce the load on the power grid during peak periods. The interruptible load model is represented as follows: (17) In the formula, for Interruptible load power at any given time This is the baseline percentage coefficient for interruptible loads. for Adjusting decision variables during time-based interruptions, increasing them during peak periods. By reducing the normal and trough periods To adjust ;in, It can be shared with adjustable loads, simplifying the dimensionality of decision variables and optimizing computational complexity, while satisfying the following constraints: (18) In the formula, for The maximum value; To avoid data loss or restart costs due to complete task interruption, interruptible loads must maintain minimum operating power: (19) Power-adjustable load model: Adjustable power loads are flexible loads whose operating intensity can be dynamically adjusted. They are responsible for handling computational tasks with flexible computing power allocation or adjustable priorities. Their adjustment range is smaller than that of interruptible loads, and the adjustment is smoother. The adjustable power load model is represented as follows: (20) In the formula, for The power of the load can be adjusted at any time. This is the power adjustable load reference ratio coefficient. This is the power adjustment amplitude coefficient; by taking a reasonable value, smooth power adjustment can be achieved. To further achieve smooth power regulation, the time intervals are... arrive The active power difference fulfills the constraint: ;(twenty one) In the formula, The maximum allowable power adjustment rate is determined by task characteristics and equipment limitations.

[0019] The optimized total load model after incorporating decision variable adjustments: In the original total load, the base load, shiftable load, interruptible load, and power-adjustable load operate according to their respective baseline proportions: ;(twenty two) In the formula, , , and In their original states The baseline load power, shiftable load power, interruptible load power, and adjustable load power at any given time, and the respective baseline proportion coefficients satisfy the following: ;(twenty three) Based on equations (22) and (23), it is not difficult to derive another expression for equation (10): ;(twenty four) The IT load power, after incorporating decision variable regulation and optimization, satisfies: (25) In the formula, for IT load power at any given time.

[0020] S3 constructs an adjustable resource model for computing nodes.

[0021] Central air conditioning model: The central air conditioning system is modeled based on the equivalent thermal parameter ETP, which reflects the dynamic balance between indoor thermal inertia and cooling demand, and is coupled with IT load heat dissipation and water storage cooling. Total heat dissipation requirements consist of IT load heat dissipation and environmental heat dissipation: (26) In the formula, for Total heat dissipation requirements at all times. for The heat dissipation power of the IT load at any time. for Indoor temperature at any given time for The ambient temperature at any given time Thermal resistance characterizes the thermal insulation performance of a building envelope; where: (27) In the formula, for Total IT load power at any given time This is a unit conversion factor used to establish the relationship between power and heat flux. The coefficient of performance is the heat dissipation efficiency. The total cooling demand is: (28) In the formula, for Total cooling demand at any given time Heat capacity, characterizing the thermal inertia of the indoor environment. To set the temperature, The cooling capacity is the amount of heat released; the cooling demand is non-negative, and no cooling is needed when there is no heat dissipation. Total cooling demand at any time After a portion of the cooling capacity is offset by the cooling released from water storage, the remaining demand must be met by the central air conditioning compressor. (29) In the formula, for The cooling capacity of the central air conditioning compressor at any given time. for The amount of water stored and released for cooling at any given time; The power of a central air conditioning compressor during operation is expressed as: (30) In the formula, for The power of the central air conditioning compressor at any given time. for The coefficient of performance (COP) at any given time varies dynamically with the outdoor temperature. (31) In the formula, and for Fit coefficient, For reference temperature; By balancing various cooling and heat dissipation capacities, a dynamic constraint on indoor temperature can be obtained: (32) In the formula, for Total actual cooling capacity at any given moment; Central air conditioning systems need to meet temperature constraints. Non-negative constraint on air conditioning power ;in, To allow the central air conditioning to reach the highest possible indoor temperature.

[0022] Electrochemical energy storage model: Electrochemical energy storage achieves peak shaving and valley filling through charge and discharge regulation, while ensuring the safe operation of the energy storage system, satisfying both economic efficiency and safety. The state of charge (SOC) of electrochemical energy storage is as follows: (33) In the formula, for The electrochemical state of charge at any given time. for The electrochemical state of charge at any given time. For time step, For charging and discharging power, For electrochemical energy storage, for The charging and discharging power at any given moment; The state of charge of electrochemical energy storage should be constrained within a safe range: (34) In the formula, The lowest electrochemical state of charge. This represents the highest electrochemical state of charge. The charging and discharging power should not exceed the rated power. (35) In the formula, Maximum charge / discharge power; Furthermore, the initial electrochemical energy storage state of charge is .

[0023] Diesel generator model: Diesel generators (DGs) serve as emergency power sources during peak loads or when renewable energy is insufficient, and their output power must meet the following requirements: (36) In the formula, diesel generator Output power at any moment This is the maximum output power; Diesel generator in the initial period ( The output power must not exceed the ramp rate: (37) In the formula, This refers to the speed of the diesel generator when climbing a hill. At that time, the output power of the diesel generator changes in accordance with: (38) Water-cooled model: Water-based cooling utilizes off-peak electricity pricing to produce ice for storage, and releases this ice during peak pricing periods to assist air conditioning, thus achieving "electricity-cooling" synergy by coupling with central air conditioning. When in ice-making mode, electricity is consumed to make ice, increasing the cold storage capacity. (39) In the formula, for Ice-making / cooling power at any given time for The amount of cold storage at any given time. for The amount of cold storage at any given time. This is the coefficient for cooling loss. The energy efficiency ratio of the ice-making process; At this time, it is in cold release mode, releasing cold energy and reducing the cold storage capacity: (40) In the formula, Ensure that the released cooling capacity does not exceed the current stored cooling capacity, and couple this with the central air conditioning model: (41) It should not exceed the upper limit of cooling capacity. and upper limit of cooling power : (42) The cold storage capacity should also be within a safe range and should not exceed the cold storage capacity. To prevent overcharging / over-discharging: (43) S4. Establish a collaborative optimization model for source-grid-load-storage.

[0024] Objective function: The objective function contains two objectives: minimizing total operating cost and minimizing total carbon emissions. These objectives are constructed by weighted summation and the introduction of a penalty function. (44) In the formula, Let be the objective function. The total daily operating cost of the system (in yuan). This represents the total daily carbon emissions (kg). The weighting factor for carbon emissions (yuan / kg) is used. To constrain violations and penalties; Breakdown of daily total operating costs: Total daily operating costs include electricity purchase costs from the grid. fuel costs of diesel generators : (45) In the formula, For the cost of purchasing electricity from the power grid, The fuel cost of the diesel generator; of which: in accordance with Time-of-use electricity pricing and The power consumption at any given time is determined by: (46) In the formula, for Time-of-use electricity price (MW) for Power purchased at any time (RMB / MWh) This refers to the number of time periods throughout the day; According to diesel generator Calculation of output power and unit fuel price at any given time: (47) In the formula, diesel generator Output power (MW) at any given time Price per unit of fuel (RMB / kWh) For time step.

[0025] Calculation of daily carbon emissions: Total daily carbon emissions encompass both indirect carbon emissions from electricity purchases from the grid and direct carbon emissions from diesel generators. Total daily carbon emissions are expressed as follows: (48) In the formula, Carbon emission coefficient (kg / kWh) for electricity purchasers from the power grid. The unit carbon emission factor (kg / kWh) for diesel generators.

[0026] Power balance constraints: Output and load need to be kept in dynamic balance, taking into account factors such as IT load, adjustable resources, wind and solar power output, and electricity purchases. (49) In the formula, for Solar power output at all times for Wind power output at all times diesel generator Output power at any moment for Purchase power at any time for IT load power at any given time for The power of the central air conditioning compressor at any given time. for Ice-making / cooling power at any given time for The charging and discharging power at any given moment.

[0027] Equipment operating constraints: All types of equipment must meet the characteristic constraints defined in S2 and S3 to operate.

[0028] Power grid security constraints: The voltage at each node in the power grid must be maintained within a safe range: (50) In the formula, for Time of the first Node voltage (pu) This represents the total number of nodes in the power grid. , The upper and lower limits of the allowable voltage (pu).

[0029] Penalties for violating constraints: The penalties for violating the constraints are as follows: (51) In the formula, This is a penalty item for voltage exceeding the limit. For energy storage Penalties for exceeding limits Penalty for exceeding the limit for water storage capacity. This is a penalty for exceeding temperature limits. For diesel generators, there is a hill-climbing penalty. Reverse power transmission penalty from the power grid The strategy-guided penalty terms guide the correct actions of water cooling and energy storage. Each penalty term satisfies the following: (52) In the formula, The voltage penalty coefficient is (yuan / (pu)²). , for The system's lowest and highest node voltages (pu) at any given time. , These are the upper and lower limits of the allowable voltage. The electrochemical energy storage penalty coefficient (yuan) is given. for The electrochemical state of charge at any given time. The lowest electrochemical state of charge. This represents the highest electrochemical state of charge. Water storage cooling penalty coefficient (yuan / 2 ), for Water storage capacity at any time ( ), For cold storage capacity; The central air conditioning temperature penalty coefficient (yuan / ℃²). for Indoor temperature at any given time (°C). To allow the central air conditioning to reach the highest possible indoor temperature; The climbing penalty coefficient for diesel generators is (yuan / (MW)²). diesel generator Output power at any moment diesel generator Output power at any moment This refers to the maximum permissible ramp output threshold for a single dispatch period of a diesel generator; The reverse power supply penalty coefficient is (yuan / (MW)²). for Purchase power at any time Reverse power supply tolerance (MW); The strategy guidance penalty coefficient (RMB / MW) for The charging and discharging power at any given time for Ice-making / cooling power at any given time and They are respectively The off-peak and peak electricity price indicators for each time period (1 for off-peak periods and 0 for other periods; 1 for peak periods and 0 for other periods).

[0030] S5, an improved optimization algorithm for solving the cuckoo catfish problem.

[0031] This invention employs an improved Cuckoo Catfish Optimization (ICCO) algorithm to solve the source-grid-load-storage collaborative optimization model. This algorithm is based on the Cuckoo Catfish Optimization (CCO) algorithm, incorporating Tent chaotic mapping initialization, a golden sine strategy for optimizing the search path, and an elite back-learning mechanism to achieve a dynamic balance between exploration and development. The specific process of solving the source-grid-load-storage collaborative optimization model using the improved Cuckoo Catfish Optimization algorithm is as follows: Population initialization: The initialization introduces the Tent chaotic mapping mechanism to generate an initial population of cichlid catfish with uniform distribution and rich diversity, laying the foundation for global optimization.

[0032] Let the population size be The decision variable dimension is The lower bound of the decision variable is The upper boundary is During initialization, an initial chaotic sequence is first generated. , , Initial chaotic sequence It is OK A two-dimensional array of columns that satisfies the Tent chaotic mapping rule: (53) In the formula, Let be the index of the chaotic variable in the initial chaotic sequence, and , The first chaotic sequence in the initial chaotic sequence There are several chaotic variables; it should be noted that the Tent chaotic mapping rule in equation (53) is similar to... It is unrelated to, but related to Related, therefore use To indicate; An initial population is constructed based on the initial chaotic sequence; wherein, in the initial population, the first... The position of an individual The generating formula is: (54) In the formula, For the index of individuals in the population, and It should be noted that the individual position in equation (54) is related to... Irrelevant, but related to Related, therefore use To indicate; Introducing chaotic variables to generate an initial population can more comprehensively cover the solution space and avoid local optimization traps caused by the initial aggregation of individuals.

[0033] Fitness calculation: After the initial population is generated, the fitness value of each individual is calculated. fitness value The calculation formula is: (55) The initial optimal fitness value is selected through fitness evaluation. and the corresponding optimal individual position and initialize the convergence curve. To record the changes in the optimal fitness value during the iteration process.

[0034] Iterative optimization calculation: After entering the iterative optimization phase, let the maximum number of iterations be... , No. The next iteration ( First, the golden ratio and random guiding variables are introduced to construct the golden sine strategy to optimize the search path and enhance the algorithm's development capability. Define the golden ratio Two key guiding points were calculated based on the golden ratio principle. , : (56) (57) In the formula, , For interval parameters; For the first in the population For each individual, a hybrid search strategy is used to update the location: like The golden sine strategy is employed for local fine-grained search, utilizing the periodicity of the sine function to guide individuals to converge toward the optimal solution. The individual position update formula has two cases: (58) In the formula, and All numbers are uniformly distributed random numbers within the interval [0,1] (two random numbers were generated during the local fine-tuning search using the golden sine strategy). For the first The updated position of each individual , All are random variables, and , ; like The encirclement search mechanism based on Levy flight in the CCO algorithm is retained to maintain the algorithm's exploration capability. Simulating the random search behavior of the cuckoo catfish in finding a host, the update formula is: (59) In the formula, Levy's flight stride length is the step size coefficient for linear decay, and This ensures that the algorithm has a wide scope of exploration in the early stage and high accuracy in the later stage of development; After an individual's location is updated, a new individual is formed. Boundary checks are required to ensure that the decision variables are within the feasible region. The boundary handling formula is as follows: (60) The fitness value of the new individual is then calculated. ,like Then the original values ​​are replaced with the new individual's position and fitness values. , .

[0035] Elite Reverse Learning: CCO uses a death parasitic mechanism to eliminate inferior individuals and replenish them with new individuals to maintain population vitality, while ICCO uses an elite reverse learning mechanism to replace the death parasitic mechanism to maintain population diversity. During the iteration process, the population is sorted in descending order of fitness value, and the worst 20% of individuals are eliminated. For the th iteration... The worst individual (index is ) Generate mutant individuals based on the optimal solution. Small mutations are performed near the optimal individual position: (61) In the formula, For a random vector that follows a standard normal distribution, The coefficient of variation is 1. This is an adjustment factor that decays with iteration; Calculation after boundary check fitness value ,like If the worst individual is replaced by the candidate individual, then the candidate individual is used instead. , .

[0036] Update the optimal fitness value and optimal individual location: (62) In the formula, This represents the optimal fitness value for the current iteration. This represents the globally optimal fitness value. This is the index of the current best individual.

[0037] Algorithm termination and optimal scheduling strategy output: When the number of iterations reaches the maximum number of iterations When the time is up, the algorithm terminates and outputs the algorithm convergence result and the optimal scheduling strategy.

[0038] The method of the present invention will be illustrated below with specific examples.

[0039] This specific example is based on the simulation analysis of the IEEE 33-node system. The topology of the IEEE 33-node system is as follows: Figure 2 As shown; among them, IT load, central air conditioning, electrochemical energy storage, diesel generator and water-cooled storage are connected to node 30, wind power is connected to node 8, and photovoltaic is connected to nodes 2 and 3.

[0040] S1, Construct a dynamic response model for multiple resources of computing nodes: In photovoltaic modeling, , , , , , , , , .

[0041] In wind power modeling, , , , , , , .

[0042] Wind power output, solar power output, and total output of renewable energy (wind power + solar power) are as follows Figure 3 As shown.

[0043] S2, Deconstructing the Total IT Infrastructure Load Model of Computing Nodes: In the IT load dual-peak model (i.e., the total IT basic load model of computing power nodes), A =2.5MW B =0.5、 C =0.4、 , , , , It follows a uniform distribution in the range [0, 0.05].

[0044] base load middle, .

[0045] Transferable load middle, , , , , , .

[0046] Interruptible load middle, , .

[0047] Adjustable load middle, , , .

[0048] In the original total load, the four types of IT loads operate according to their respective baseline proportions. After optimization, the total load satisfies the sum of the various types of IT loads.

[0049] S3, constructing an adjustable resource model for computing nodes: In the central air conditioning model, , , , , , , , , .

[0050] In electrochemical energy storage models, , , , , , , .

[0051] In the diesel generator (DG) model, , .

[0052] In the water-cooled storage model, , , , , .

[0053] S4. Establish a source-grid-load-storage collaborative optimization model: In the objective function, .

[0054] Of the total daily operating costs, the time-of-use electricity price includes ; diesel generator .

[0055] Of the total daily carbon emissions , .

[0056] In power balance constraints, output and load must maintain a dynamic balance.

[0057] In the equipment operation constraints, the operation of various types of equipment must meet the characteristic constraints defined in S2 and S3.

[0058] In power grid security constraints, , .

[0059] To address violations of constraints, penalties are applied for voltage exceeding limits, energy storage SOC exceeding limits, water-cooled storage capacity exceeding limits, temperature exceeding limits, diesel generator ramping penalties, grid reverse power transmission penalties, and strategy guidance penalties. =10 9 Yuan / (pu)² =10 9 Yuan, =10 9 Yuan / 2 , =10 9 Yuan / ℃² =105 Yuan / (MW)² =10 7 Yuan / (MW)² =0.1MW =10 4 Yuan / MW.

[0060] S5, Solving with an improved optimization algorithm for the cuckoo catfish: The improved solution process for the Cuckoo Catfish Optimization (ICCO) algorithm is as follows: Figure 4 As shown: During population initialization, =150, =120, the decision variables comprehensively consider energy storage, diesel generators, water-cooled storage, interruptible, shiftable, and power-adjustable IT loads: .

[0061] Fitness calculation.

[0062] In iterative optimization calculations, , , , .

[0063] In elite reverse learning, .

[0064] Update the optimal fitness value and the optimal individual location.

[0065] In the algorithm termination and optimal scheduling strategy output, the algorithm terminates when the number of iterations reaches 300, and outputs the algorithm convergence result and the optimal scheduling strategy.

[0066] Figure 5 The diagram illustrates the convergence curves of various algorithms, showing the convergence curves of the ICCO algorithm, particle swarm optimization algorithm, dung beetle optimization algorithm, and standard cuckoo catfish algorithm; the convergence speed of the ICCO algorithm is the best among the four algorithms.

[0067] By adjusting the allocation of resources, the daily cost before optimization was 39,216.99 yuan, and the daily cost after optimization was 19,560.78 yuan, saving 50.12%; the carbon emissions before optimization were 39,715 kg, and the carbon emissions after optimization were 20,407.6 kg, reducing carbon emissions by 48.61%.

[0068] The comparison curve before and after IT load optimization is as follows: Figure 6 As shown, by changing the load in a way that allows for shifting, interruption, and power adjustment, the IT load can be shaving off peaks and filling valleys.

[0069] Electrochemical energy storage operation strategy curves are as follows Figure 7As shown, it charges during off-peak electricity prices and discharges during peak electricity prices.

[0070] Water storage cooling operation strategy curve as shown Figure 8 As shown, cold is stored during off-peak electricity prices and released during peak electricity prices to share the cooling load of the central air conditioning system.

[0071] Energy distribution on the load side and power supply side as follows Figure 9 As shown.

[0072] Comparison of total daily operating costs and total daily carbon emissions before and after optimization, for example Figure 10 As shown, cost and carbon emissions are reduced through optimization.

[0073] Changes in IT load node voltage before and after optimization, such as Figure 11 As shown, the optimized voltage is between the upper and lower limits.

[0074] Based on the above-mentioned method for coordinated scheduling of power resources with adjustable computing nodes, this invention also provides a device for coordinated scheduling of power resources with adjustable computing nodes.

[0075] A computing node adjustable power resource collaborative scheduling device includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the computing node adjustable power resource collaborative scheduling method as described above.

[0076] This invention discloses a method and apparatus for collaborative scheduling of adjustable power resources on computing power nodes. Addressing the characteristics of large fluctuations and strong randomness in adjustable power resources on computing power nodes in urban power grids, the invention first constructs a dynamic response model for multiple resources on computing power nodes, including photovoltaic and wind power, IT load models for computing power nodes, and models for other adjustable resources. Then, the computing power node load is deconstructed; IT load includes four types: basic, movable, interruptible, and power adjustable; other adjustable resources include four types: central air conditioning, electrochemical energy storage, diesel generators, and water-based cooling. Next, a source-grid-load-storage collaborative optimization model is established with the objective function of minimizing operating costs and carbon emissions. Finally, to solve this optimization problem, an improved Cuckoo Catfish Optimizer (CCO) algorithm is proposed. This algorithm, based on the Cuckoo Catfish Optimizer (CCO), introduces three strategies: Tent chaotic mapping initialization, golden sine strategy optimization search path, and elite back-learning mechanism, effectively improving the convergence speed and optimization accuracy of the algorithm. This invention enables computing nodes to operate economically and with low carbon emissions while meeting power grid security constraints, thereby improving the flexible utilization of the system's adjustable resources.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for coordinated scheduling of power resources with adjustable computing nodes, characterized in that, include: Construct a multi-resource dynamic response model for computing power nodes; wherein, the multi-resource dynamic response model includes a photovoltaic model and a wind power model; The overall IT load model of computing nodes is deconstructed according to the type of computing task to obtain multiple types of IT load models; among them, the multiple types of IT load models include the basic load model, the shiftable load model, the interruptible load model, and the power adjustable load model. Construct an adjustable resource model for computing power nodes; wherein, the adjustable resource model includes a central air conditioning model, an electrochemical energy storage model, a diesel generator model, and a water-cooled storage model; Based on the multi-resource dynamic response model, the multi-type IT load model, and the adjustable resources, a source-grid-load-storage collaborative optimization model is established with the objective function of minimizing operating costs and carbon emissions. An improved Cuckoo Catfish optimization algorithm, incorporating Tent chaotic mapping initialization, golden sine strategy optimization of the search path, and elite reverse learning mechanism, is used to solve the source-grid-load-storage collaborative optimization model to obtain the optimal scheduling strategy.

2. The method for coordinated scheduling of adjustable power resources by computing nodes according to claim 1, characterized in that, The photovoltaic model is represented as follows: ； In the formula, for Solar power output at all times The rated installed capacity of photovoltaic modules. for Normalized light intensity at any given time The temperature coefficient of a photovoltaic module. for The ambient temperature at any given time The standard test temperature for photovoltaic modules; where: ； In the formula, This is a baseline curve for ideal intraday light intensity. The photovoltaic power output fluctuation coefficient. To obey A random variable with a random distribution; where: ； In the formula, At sunrise, It is sunset time; ； In the formula, Based on ambient temperature, This represents the amplitude of temperature fluctuations. For temperature phase shift, To obey Gaussian random variables.

3. The method for coordinated scheduling of adjustable power resources by computing nodes according to claim 1, characterized in that, The wind power model is represented as follows: ； In the formula, for Wind power output at any given moment; The rated installed capacity of the wind turbine unit. for The actual wind speed at that moment, To cut into wind speed, To cut off the wind speed, Rated wind speed; where: ； In the formula, This is the baseline trend curve for intraday wind speed. This is the wind speed fluctuation amplitude coefficient. To simulate wind speed disturbances A uniformly distributed random variable; where: ； In the formula, The average wind speed, For wind speed amplitude, This refers to the wind speed phase shift.

4. The method for coordinated scheduling of adjustable power resources by computing nodes according to claim 1, characterized in that, The total IT infrastructure load model for the computing power nodes is represented as follows: ； In the formula, for Total IT infrastructure load power of computing nodes at any given time. This is the load baseline value. and This represents the load factor for the intraday double peak period. and This is the peak load time of the day. and The distribution width during peak hours, This is a random fluctuation term that follows a uniform distribution.

5. The method for coordinated scheduling of adjustable power resources by computing nodes according to claim 4, characterized in that, The basic load model is represented as follows: ； In the formula, for Base load power at any given time The fixed proportion factor of the base load; The transferable load model is represented as follows: ； In the formula, for The power of the load that can be moved at any time. This is the reference ratio coefficient for movable loads. After time period attribute correction Adjustment of decision variables at time; where: ； In the formula, , and These are the trough period, peak period, and normal period, respectively. for The original translation adjustment decision variables at each moment; The interruptible load model is represented as follows: ； In the formula, for Interruptible load power at any given time This is the baseline percentage coefficient for interruptible loads. for Interruptions at any time adjust decision variables; The power-adjustable load model is represented as follows: ； In the formula, for The power of the load can be adjusted at any time. This is the power adjustable load reference ratio coefficient. This is the power adjustment amplitude coefficient; The fixed proportion coefficient of the base load, the reference proportion coefficient of the movable load, and the reference proportion coefficient of the interruptible load satisfy the following: 。 6. The method for coordinated scheduling of adjustable power resources by computing nodes according to claim 1, characterized in that, The central air conditioning model is represented as follows: ； In the formula, for The power of the central air conditioning compressor at any given time. for Coefficient of performance at any given time For unit conversion factor, for The cooling capacity of the central air conditioning compressor at any given time; where: ； In the formula, and for Fit coefficient, For reference temperature, for The ambient temperature at that moment; ； In the formula, for The amount of cooling released by water at any given time. for Total cooling demand at any given time; of which: ； In the formula, For heat capacity, for Indoor temperature at any given time To set the temperature, for Total heat dissipation requirements at any given time; of which: ； In the formula, for The heat dissipation power of the IT load at any time. For thermal resistance; where: ； In the formula, for Total IT load power at any given time The coefficient of performance is the heat dissipation efficiency. The electrochemical energy storage model is expressed as follows: ； In the formula, for The electrochemical state of charge at any given time. for The electrochemical state of charge at any given time. For time step, For charging and discharging power, For electrochemical energy storage, for The charging and discharging power at any given moment; The diesel generator model is represented as follows: ； In the formula, diesel generator Output power at any moment This is the maximum output power of the diesel generator. This refers to the output power of the diesel generator during the initial period. This refers to the speed of the diesel generator when climbing a hill. The water-based cooling model is represented as follows: ； In the formula, for The amount of cooling released by water at any given time. for Ice-making / cooling power at any given time for The amount of cold storage at any given time; of which: ； In the formula, for The amount of cold storage at any given time. This is the coefficient for cooling loss. This refers to the energy efficiency ratio of the ice-making process.

7. The method for coordinated scheduling of adjustable power resources by computing nodes according to claim 1, characterized in that, The source-grid-load-storage collaborative optimization model is expressed as follows: ； In the formula, Let be the objective function. The total daily operating cost of the system. This refers to the total daily carbon emissions. Quantify the weighting coefficients for carbon emissions. To constrain the penalties for violations; among which: ； In the formula, For the cost of purchasing electricity from the power grid, The fuel cost of the diesel generator; of which: ； ； In the formula, for Purchase power at any time for Time-of-use electricity pricing This refers to the number of time periods throughout the day. For time step; diesel generator Output power at any moment Price per unit of fuel; ； In the formula, Carbon emission coefficient of the electricity purchaser from the power grid. The unit carbon emission coefficient of a diesel generator; ； In the formula, This is a penalty item for voltage exceeding the limit. For energy storage Penalties for exceeding limits Penalty for exceeding the limit for water storage capacity. This is a penalty for exceeding temperature limits. For diesel generators, there is a hill-climbing penalty. Reverse power transmission penalty from the power grid For strategy-guided penalty items; where: ； In the formula, This is the voltage penalty coefficient. , for The system's lowest and highest node voltages at specific times. , These are the upper and lower limits of the allowable voltage. This is the electrochemical energy storage penalty coefficient. for The electrochemical state of charge at any given time. The lowest electrochemical state of charge. This represents the highest electrochemical state of charge. The water storage cooling penalty coefficient, for The amount of water stored for cooling at any given time. For cold storage capacity; This refers to the temperature penalty coefficient for central air conditioning. for Indoor temperature at any given time To allow the central air conditioning to reach the highest possible indoor temperature; This represents the hill-climbing penalty coefficient for diesel generators. diesel generator Output power at any moment diesel generator Output power at any moment This refers to the maximum permissible ramp output threshold for a single dispatch period of a diesel generator; This is the penalty coefficient for reverse power transmission. for Purchase power at any time This is the reverse power supply tolerance value; The penalty coefficient is used to guide the strategy. for The charging and discharging power at any given time for Ice-making / cooling power at any given time and They are respectively The off-peak electricity price and peak electricity price are displayed at different times.

8. The method for coordinated scheduling of adjustable power resources by computing nodes according to claim 7, characterized in that, The source-grid-load-storage collaborative optimization model satisfies power balance constraints, equipment operation constraints, and grid security constraints; wherein: The power balance constraint is expressed as: ； In the formula, for Solar power output at all times for Wind power output at all times diesel generator Output power at any moment for Purchase power at any time for IT load power at any given time for The power of the central air conditioning compressor at any given time. for Ice-making / cooling power at any given time for The charging and discharging power at any given moment.

9. The method for coordinated scheduling of adjustable power resources by computing nodes according to claim 1, characterized in that, The improved cuckoo catfish optimization algorithm is used to solve the source-grid-load-storage collaborative optimization model, specifically including: An initial chaotic sequence is randomly generated using the Tent chaotic mapping rule, and an initial population is constructed based on the initial chaotic sequence. The fitness value of each individual in the initial population is calculated using the source-network-load-storage collaborative optimization model, and the optimal fitness value and the corresponding optimal individual position are selected. A hybrid search strategy, consisting of the golden sine strategy and the Levy flight-based encirclement search mechanism in the cuckoo catfish optimization algorithm, is used to iteratively update the positions of individuals in the population; where: like Then, the golden sine strategy is used to iteratively update the position of individuals in the population; where the formula for iteratively updating the position of individuals in the population using the golden sine strategy is: ； In the formula, For the first The position of each individual, For the first The updated position of each individual For the optimal individual position, , All are random variables, and , , , These are all key guiding points. and All are uniformly distributed random numbers within the interval [0,1]; where: ， ； In the formula, The golden ratio, , For interval parameters; like Then, the Levy-based encirclement search mechanism in the Cuckoo Catfish Optimization Algorithm is used to iteratively update the positions of individuals in the population; whereby the formula for iteratively updating the positions of individuals in the population using the Levy-based encirclement search mechanism in the Cuckoo Catfish Optimization Algorithm is: ； In the formula, Levy's flight stride length is the step size coefficient for linear decay, and , This represents the current iteration number. This represents the maximum number of iterations. After each iteration updates the individual position, inferior individuals are eliminated, and elite back-learning is used to generate mutated individuals based on the optimal solution to replace the eliminated inferior individuals; the formula for generating mutated individuals based on the optimal solution using elite back-learning is as follows: ； In the formula, As a mutated individual, For a random vector that follows a standard normal distribution, The coefficient of variation is 1. , These are the lower and upper bounds of the decision variables; After updating the individual position in each iteration, update the optimal fitness value and the optimal individual position; When the maximum number of iterations is reached, the optimal scheduling strategy is output.

10. A power resource collaborative scheduling device with adjustable computing nodes, characterized in that, It includes a processor, a memory, and a computer program stored in the memory, wherein the computer program, when executed by the processor, implements the adjustable power resource collaborative scheduling method for computing nodes as described in any one of claims 1 to 9.

Images