Cloud data center power distribution method, device, equipment and medium

Abstract

The invention discloses a cloud data center power distribution method and device, equipment and a medium, and the method comprises the steps: carrying out the clustering of obtained power utilization data based on a similarity index of user attributes, and dividing the power utilization data of different users into different grades; classifying the users with different adjustment capabilities, and calculating the reference power of the server according to the classification result; establishing a power distribution algorithm based on a non-cooperative game model, and determining an optimal response function of the server according to the reference power; and performing iterative solution on the optimal response function based on a preset server power distribution algorithm, and determining power distribution of the server. The scheme can improve the energy efficiency and performance of the data center.

Description

Technical Field

[0001] The present invention relates to the technical field of data centers, and in particular to a cloud data center power allocation method, device, equipment and medium. Background Art

[0002] A data center is a set of complex facilities that integrates servers, storage devices, switches, routers, firewalls and other equipment to achieve centralized processing, storage, transmission, exchange and management of data information. It is the core infrastructure at the bottom of cloud computing, supporting a wide range of applications and services such as SaaS (Software as a Service), IaaS (Infrastructure as a Service) and PaaS (Platform as a Service), enabling enterprises to flexibly expand resources and access computing power on demand, thereby accelerating innovation and responding to market changes.

[0003] Existing data centers can only process electricity consumption data according to preset rules, and the energy efficiency and performance of data centers are low. Summary of the invention

[0004] In view of the above-mentioned defects, the present invention provides a cloud data center power allocation method, device, equipment and medium, which can improve the energy efficiency and performance of the data center.

[0005] An embodiment of the present invention provides a cloud data center power allocation method, the method comprising: The acquired electricity consumption data is clustered based on the similarity index of user attributes, and the electricity consumption data of different users are divided into different levels; Classify users with different adjustment capabilities and calculate the baseline power of the server based on the classification results; Establishing a power allocation algorithm based on a non-cooperative game model, and determining the best response function of the server according to the reference power; The optimal response function is iteratively solved based on a preset server power allocation algorithm to determine the power allocation of the server.

[0006] Preferably, the acquired power consumption data is clustered based on the similarity index of the user attributes, and the power consumption data of different users are divided into different levels, including: Performing standardization processing on the user data of the user to obtain a load standardization sequence; Fitting the probability distribution of the electric quantity amplitude in the load standardization sequence to obtain a cumulative probability distribution function; Divide the user's daily load data into several equal-length periods according to the time domain, calculate the average load of each period, and obtain the load sequence; Performing segmented amplitude division according to the quantiles of the cumulative probability distribution function, and converting the load mean in the load sequence into a corresponding integer level; The integer level corresponding to the load mean in the load sequence is output as a character string.

[0007] Furthermore, users with different adjustment capabilities are classified, including: Clustering the character strings based on the user attribute similarity index, selecting characters corresponding to the users of the data processing load at the same power usage time, and outputting a new character string; According to the new character string, a state transition probability matrix is ​​determined by using a Markov method, thereby calculating an attribute similarity index; Taking the calculated attribute similarity index as input, K-means is used for clustering to divide users into users with strong attribute similarity and weak attribute similarity.

[0008] Furthermore, the method further comprises: The demand parameter sequences of different users are used as input vectors of the self-organizing map neural network; Normalizing the weight vector of the competition layer in the self-organizing map neural network and the input vector; Comparing the similarity of all normalized weight vectors in the competition layer with each normalized input vector; The normalized weight vector with the highest similarity is determined as the winning neuron; The output of the winning neuron is determined to be 1, and the outputs of the remaining neurons are determined to be 0, and the electricity consumption data of different users are secondary clustered according to the adjusted weight vector of the winning neuron.

[0009] Preferably, calculating the baseline power of the server according to the classification result includes: According to the classification results, the data processing load of the electricity consumption data of users of different classifications is aggregated into different servers; Merging intermediate response states in a multi-unit state model of a server into equivalent response states; When the server operation state changes, determine the state switching probability of the server switching between different states; Determine the time characteristic distribution of state switching transfer according to the calculated state switching probability; Determine the average residence time of different states based on the calculated time characteristic distribution; Calculate the steady-state probability that the response state of the server reaches a steady state according to the calculated average stay time; The steady-state expected power of each server is calculated according to the expected value formula of instantaneous power, thereby obtaining the total expected value of steady-state response power; The reference power of the server is calculated according to the steady-state probability and the total expected value of the steady-state response power.

[0010] Preferably, a power allocation algorithm is established based on a non-cooperative game model, and the best response function of the server is determined according to the reference power, including: Construct the network utility function of each server based on the non-cooperative game model; Calculating a Nash equilibrium solution of the non-cooperative game model in the cloud data center according to the constructed network utility function; The optimal response functions of different servers in the cloud data center are calculated according to the Nash equilibrium solution.

[0011] Further, the best response function is iteratively solved based on a preset server power allocation algorithm to determine the power allocation of the server, including: Initialize the number of iterations and particle swarm optimization algorithm parameters; Update inertia factor; Calculating fitness values ​​for all particles using the network utility function; Update the local best position of each server in each particle and the global best position of all particles; Update the number of iterations, and use the preset particle update model to update the particle step size and particle value; Determine whether the preset convergence condition is met; If not, update the inertia factor, recalculate the fitness value, and update the local best position, global best position particle step size and particle value; If so, the global best position is output as the optimal power value for power allocation.

[0012] An embodiment of the present invention further provides a cloud data center power distribution device, the device comprising: A classification module is used to cluster the acquired power consumption data based on the similarity index of user attributes, and classify the power consumption data of different users into different levels; A power calculation module is used to classify users with different adjustment capabilities and calculate the baseline power of the server according to the classification results; A response module, used to establish a power allocation algorithm based on a non-cooperative game model, and determine the best response function of the server according to the reference power; The allocation module iteratively solves the best response function based on a preset server power allocation algorithm to determine the power allocation of the server.

[0013] An embodiment of the present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the cloud data center power allocation method as described in any one of the above embodiments is implemented.

[0014] An embodiment of the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the cloud data center power allocation method as described in any one of the above embodiments.

[0015] The cloud data center power allocation method, device, equipment and medium provided by the present invention cluster the acquired power consumption data based on the similarity index of user attributes, and divide the power consumption data of different users into different levels; classify users with different adjustment capabilities, and calculate the base power of the server according to the classification results; establish a power allocation algorithm based on a non-cooperative game model, and determine the best response function of the server according to the base power; iteratively solve the best response function based on a preset server power allocation algorithm to determine the power allocation of the server. This solution can improve the energy efficiency and performance of the data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of a cloud data center power allocation method provided by an embodiment of the present invention; Figure 2 is a schematic diagram of the structure of a cloud data center power distribution device provided by an embodiment of the present invention; Figure 3 It is a structural diagram of a terminal device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] See also Figure 1 , is a flow chart of a cloud data center power allocation method provided by an embodiment of the present invention, the method comprising steps S1 to S4: Step S1, clustering the acquired power consumption data based on the similarity index of user attributes, and dividing the power consumption data of different users into different levels; Step S2, classifying users with different adjustable capabilities, and calculating the baseline power of the server according to the classification results; Step S3, establishing a power allocation algorithm based on a non-cooperative game model, and determining the best response function of the server according to the reference power; Step S4, iteratively solving the best response function based on a preset server power allocation algorithm to determine the power allocation of the server.

[0019] In the specific implementation of this embodiment, the power consumption data is processed according to the user's data, and clustering is performed based on the user attribute similarity index to divide the users into different levels with strong attribute similarity and weak attribute similarity; and users with different adjustable capabilities are classified to obtain the baseline power of the server; A power allocation algorithm is established to optimize the power of the server based on a non-cooperative game model to ensure that each server achieves a balance in overall power allocation while pursuing its own interests. Graph coloring theory is used to allocate frequencies for data processing tasks, and a server power allocation algorithm based on interference pricing is used to determine the power allocation of the server.

[0020] This application scheme implements a cluster analysis based on the user attribute similarity index for the user's data processing power consumption data. This step converts complex data into a concise string form through dimensionality reduction technology. Classify users with different adjustable capabilities to help aggregators more effectively manage and utilize the response load resources under their control. In a short time range (such as half an hour), set the benchmark power conservation and calculate the benchmark power of the server. Based on the non-cooperative game model, the power of the server is optimized to ensure that each server achieves a balance in overall power distribution while pursuing its own interests. Subsequently, the graph coloring theory is used to allocate appropriate frequencies to data processing tasks to avoid frequency conflicts and improve the parallelism and efficiency of data processing. Finally, a server power allocation algorithm based on interference pricing is designed. By introducing a price mechanism, the power allocation of the server is dynamically adjusted to further optimize the power use of the cloud data center and achieve a dual improvement in energy efficiency and performance. This algorithm not only improves the energy efficiency of data processing, but also significantly enhances the overall performance of the cloud data center.

[0021] In another embodiment provided by the present invention, the above step S1 specifically includes the following steps: When preprocessing the power consumption data, the total daily load power consumption data of each user is processed based on the monthly data. is the input, where , the superscript in the series is the load sampling node of the day, the subscript is the month date, and the data is arranged in chronological order. The specific dimensionality reduction steps are as follows: After standardization, the processed value ; In the formula, for The mean of for The variance of the data processing load after standardization is expressed as in is a sequence with variance 1 and mean 0.

[0022] After completing the data processing load sequence standardization After that, the probability distribution of its electric quantity amplitude is fitted, and its cumulative probability distribution function is obtained.

[0023] Process the data into daily load data Divide the time domain into several equal-length segments, obtain the average load of each period, and obtain the sequence .

[0024] ; Where q means that the time axis is divided into q segments, and each data represents the data collected during this period. The elements in it are calculated according to the above standardization process, and M a Convert to .

[0025] According to the quantile of the cumulative probability distribution function, the power amplitude is divided into paragraph. Each load mean in is converted into an integer .

[0026] The integer level corresponding to the load mean in the load sequence is output as a string, and the output length is , by integer The string composed of .

[0027] In another embodiment provided by the present invention, the step S2 specifically includes the following steps: Based on the clustering of user attribute similarity index, the character string output after dimension reduction is used to select the characters corresponding to the same data processing power consumption time of the data processing load user every day in a certain month, and output it again Next, the Markov method is used to calculate the state transition probability matrix, and the attribute similarity index is combined , which measures the similarity of user attributes.

[0028] When the system leaves the state When entering the state Probability satisfy ; Assuming that there are a total of A days in a month, after completing the dimensionality reduction using the SAX method, we can obtain Transition probability matrix , calculate its average daily load curve, and construct the state transition probability matrix of the average daily load .

[0029] The data processing users are processed using the above steps and the attribute similarity index is calculated , which is used to measure the similarity of user attributes. The calculation formula of attribute similarity index is as follows: ; In the formula, Indicated in The element in the i-th row and j-th column of the transition probability matrix of the day, f represents the number of rows and columns, The larger the value of is, the higher the similarity of the power consumption attributes of the data processing users in this period is, and the stronger the regularity of the data processing power consumption is. The element in the i-th row and j-th column in the state transition probability matrix representing the average daily load.

[0030] By indicator As input, a clustering is performed through K-means to divide users into different levels of users with strong attribute similarity and weak attribute similarity. Aggregators tend to choose users with strong electricity consumption regularity and allocate more response capacity to them. Users with weak attribute similarity have weak electricity consumption regularity and are poor demand response resources, which is not conducive to the implementation of demand response projects. Therefore, no or a small amount of response capacity is allocated to users with weak electricity consumption regularity to avoid economic losses caused by user response uncertainty. Then, users with different adjustable capabilities are classified through secondary clustering so that aggregators can better grasp the response load resources under their control.

[0031] In another embodiment provided by the present invention, after performing the first clustering, the present application solution further provides a second clustering solution, including: Considering the secondary clustering of the user-adjustable characteristics of data processing, this application uses a self-organizing competitive neural network (SOM) to perform secondary clustering on data processing. This method has the advantages of fast calculation speed and good clustering effect when processing residential user data with large scale, high complexity and small differences. Compared with the shortcomings of traditional clustering algorithms, which have slow calculation speed and difficulty in meeting clustering accuracy requirements, SOM is more suitable for clustering analysis of residential data processing load data. Its algorithm steps can be divided into 4 parts: Sequence the required parameters of different users As the input of the self-organizing map neural network, the elements Indicates user Basic processing speed setting value; Indicates user The best experience; Indicates user The maximum adjustment speed that users are willing to accept for load control. If the speed exceeds this speed, the probability of users exiting load control will increase greatly. Indicates user Data processing load reduction rate after receiving demand response.

[0032] Normalizing the input vector of the self-organizing map neural network And the weight vector in the competition layer, get the normalized element And the normalized weight vector , .

[0033] ; y is an element in the input vector of SOM. When a self-organizing map neural network is obtained, When all the competition layers Both Compare similarities and The most similar Determined as the winning neuron, denoted by . Calculate the dot product , select the node j with the largest dot product .

[0034] The dot product calculation formula is: ; According to the competition rules of the self-organizing competitive neural network, only the winning neuron has the right to adjust its weight vector , and only the winning neuron outputs 1, and the outputs of the remaining neurons are 0. The adjusted weight vector is: ; In the formula, is a variable learning rate that increases and decreases over time, . When the , the weights of the corresponding neurons are not adjusted accordingly.

[0035] Data processing server aggregation model Due to the influence of user data processing equipment performance parameters and user electricity usage habits, each type of user with similar response capabilities may have large differences in data processing load characteristics. This application obtains user data processing devices with similar attributes and adjustable characteristics through secondary clustering, and aggregates all user devices with similar attributes and adjustable characteristics in the area and regards them as data processing servers to establish a data processing server output model.

[0036] In another embodiment of the present invention, the process of calculating the reference power in step S2 specifically includes the following steps: Based on clustering preprocessing, the data processing load is aggregated into different servers. The response states in each unit are similar, so the intermediate response states in the multi-unit state model are merged into equivalent response states. When the server operation state changes, the state Enter the state Probability satisfy: ; Based on the server status change, it changes from status To status The time characteristics of the transfer can be arbitrarily distributed .

[0037] ; state Under this premise, the average residence time is : ; Set the server from state arrive The mobility of When the execution cycle of the process If the time is small enough, then: ; The Markov process has no memory, so if the server stay time in each state satisfies the exponential distribution, then The parameter is The exponential distribution is: ; The average residence time can be obtained from the above formula ; After a certain period of time after the control command is issued, the response state of the server can converge to a certain constant value independent of the value before the response. This stable state probability is called the steady-state probability of the server response, and the state probability before reaching a stable response is called the transient probability. According to the memoryless principle of the Markov chain, the steady-state probability of the server for: ; In the formula, Indicates status The stationary distribution satisfies: ; Assuming that the server can adjust the speed according to the instructions issued by the load aggregator, the speed setting value after its response is: ; In the formula, Indicated in Period, data processing in unit m The set value of the data processing speed after the speed regulation of the participating aggregators; Indicates the initial speed setting value of the time period data processing; Indicates the change in speed setting value.

[0038] When the time interval is small enough, the instantaneous power is a discrete random variable ( or 0), according to the expected value formula of instantaneous power, the steady-state expected power can be obtained as: ; The expected value of the total steady-state response power of the data processing unit is: ; For the crew Data processing load in The processing speed is at the reference power under normal operation, and N is the total amount of data processing load. Assume that before the demand response signal starts, the change of solid speed is ignored, that is, when the following requirements are met: ; in, is the temperature of data processing load n in unit m under normal operating conditions, T a is the ambient temperature, T m is the temperature of unit m, t is the time; The expected reference power can be obtained as: ; in, It is the reference power of data processing load n in unit m under normal operating conditions.

[0039] is the efficiency coefficient of the data processing load n in unit m.

[0040] It is the voltage or some utility parameter of the data processing load n in unit m.

[0041] is the set temperature of data processing load n in unit m.

[0042] is the heat or energy loss of data processing load n in unit m.

[0043] is a temperature-dependent coefficient or parameter.

[0044] In a shorter time frame (such as half an hour), assuming that the baseline power is conserved, the server The base power is: ; According to the state transition probability of data processing, the response power of the entire server can be obtained. is the reference power of unit m under normal operating conditions, N m is the total amount of data processing load of unit m. Deviation from the total data processing load base power To achieve the flexibility of data processing server scheduling, the adjustable part of data processing load power is called peak-shaving power.

[0045] In another embodiment provided by the present invention, the step S3 specifically includes the following steps: Based on microeconomic theory, the server power is optimized through non-cooperative game modeling. In this model, it is considered that Each server is a rational and selfish game player, trying to optimize its utility without considering the utility of other players. The non-cooperative game framework can be expressed as: ; in, represents the set of game participants; represents the strategy space of the participants, namely, the power of each server on all cloud data centers, and the power on the cloud data centers Should satisfy , is a set of all cloud data center IDs. Indicates the maximum power of the server in each cloud data center; represents the utility function of server m, represents the net utility of the server under power P, , is the utility set of the participants.

[0046] Considering the gains brought by server power and the interference caused to users, the utility function of each server is constructed, which consists of two parts: reward function, which is the throughput gain generated by server power; penalty function, which is the interference caused by server power to data processing. The interference price is introduced to represent the pricing of data processing for server unit power. Therefore, for the server , and its network utility function is constructed as follows: ; Among them, U(P M m ) is the reward function, which represents the throughput benefit generated by the server power PM. M m ) is a penalty function, which represents the interference cost caused by the server power P to data processing. The exponential function is chosen as the penalty function so that the server can use less power to ensure its own performance and greatly reduce the interference to data processing.

[0047] In this game framework, each server optimizes its power to maximize its utility function. The key to solving the problem is to obtain the Nash Equilibrium Point (NEP) in the network.

[0048] Non-cooperative game model In the cloud data center The Nash equilibrium solution on , if it satisfies the following formula: ; in In the cloud data center In addition to the server The optimal power vector of other servers except at the Nash equilibrium point.

[0049] When NEP is reached, no player in the game will change their strategy to improve their utility alone, and it will not affect other players. Therefore, the final stable state of the game is the Nash equilibrium.

[0050] In this application, a participant is a collection of servers.

[0051] Setting up a non-cooperative game There exists a Nash equilibrium solution, where G is a non-cooperative game model. The power allocation of the server is optimized through the Nash equilibrium solution.

[0052] The optimal response functions of different servers in the cloud data center are calculated according to the Nash equilibrium solution.

[0053] Given other servers in the cloud data center Optimal power on , server m is in the cloud data center The best response function on is as follows: .

[0054] in, is the optimal power of server m in cloud data center k.

[0055] is the best response function of server m in cloud data center k.

[0056] is the optimal power of other servers in cloud data center k.

[0057] is the utility function of server m, which depends on its own power and other server power .

[0058] represents the power of selecting the one that maximizes the utility function U among all possible P.

[0059] represents the power upper limit of server m in cloud data center k.

[0060] In the cloud data center The Nash equilibrium solution on ; in, represents the optimal power vector of server k under Nash equilibrium.

[0061] represents the optimal response function of server k, which depends on the power vector .

[0062] represents the optimal response function of server i to server k, which depends on the power vector , i=1,2,…,M.

[0063] Non-cooperative game The Nash equilibrium solution is unique.

[0064] In another embodiment provided by the present invention, the step S4 specifically includes the following steps: Although this application has analyzed and discussed that there is a unique optimal power for each server, it is very difficult to obtain its closed-form solution. Therefore, in this application, the server power is solved by a power optimization algorithm. In the power optimization algorithm, each particle corresponds to the power value of the server. Through iterative updates, the particle with the maximum utility is searched, that is, the optimal power for each server. The utility function is selected as the fitness function.

[0065] At each iteration, in the cloud data center The particle velocity and its value are updated as follows: Particle velocity after iteration t+1: ; Particle value after t+1th iteration ; in, is the inertia factor, and is the learning factor, and A random number between 0 and 1. For particles The local best position of (i.e., the optimal power during the current number of iterations), For cloud data centers The global best position among all particles (i.e. the optimal particle value). Update the step size for the particle. For each server, the specific power optimization algorithm steps are as follows: When a particle is updated, the following steps are performed: Initialize the number of iterations And particle swarm optimization algorithm parameters: including the number of particle groups , then the particle group is Each particle corresponds to dimensional search space, i.e. the i-th particle and , maximum inertia factor and minimum inertia factor , the maximum number of iterations allowed ; Update inertia factor For all particles from arrive , calculate the fitness function value according to the network utility function; Update each server's local optimal position for each particle and the global best position among all particles ; Update iterations , respectively update the particle step size With particle value ; If it does not converge and the number of iterations , repeat the above steps; otherwise, the algorithm ends and outputs the optimal power value Through the power optimization algorithm, each server can obtain the optimal power in the cloud data center occupied by the central user. Assume that the power optimization algorithm used is After convergence, the complexity of the server power allocation algorithm in the network is , where represents the size of the particle.

[0066] This application scheme implements a cluster analysis based on the user attribute similarity index for the user's data processing power consumption data. This step converts complex data into a concise string form through dimensionality reduction technology. Classify users with different adjustable capabilities to help aggregators more effectively manage and utilize the response load resources under their control. In a short time range, set the baseline power conservation and calculate the baseline power of the server. Based on the non-cooperative game model, the power of the server is optimized to ensure that each server achieves a balance in overall power distribution while pursuing its own interests. Subsequently, the graph coloring theory is used to allocate appropriate frequencies to data processing tasks to avoid frequency conflicts and improve the parallelism and efficiency of data processing. Finally, a server power allocation algorithm based on interference pricing is designed. By introducing a price mechanism, the power allocation of the server is dynamically adjusted to further optimize the power use of the cloud data center and achieve a dual improvement in energy efficiency and performance. This algorithm not only improves the energy efficiency of data processing, but also significantly enhances the overall performance of the cloud data center.

[0067] The present invention also provides a cloud data center power distribution device, see Figure 2 , is a schematic diagram of the structure of a cloud data center power distribution device provided by an embodiment of the present invention, the device comprising: A classification module is used to cluster the acquired power consumption data based on the similarity index of user attributes, and classify the power consumption data of different users into different levels; A power calculation module is used to classify users with different adjustment capabilities and calculate the baseline power of the server according to the classification results; A response module, used to establish a power allocation algorithm based on a non-cooperative game model, and determine the best response function of the server according to the reference power; The allocation module iteratively solves the best response function based on a preset server power allocation algorithm to determine the power allocation of the server.

[0068] It should be noted that the cloud data center power allocation device provided in the embodiment of the present invention can execute the cloud data center power allocation method described in any of the above embodiments, and the specific functions of the cloud data center power allocation device are not described here.

[0069] See also Figure 3 , is a schematic diagram of the structure of a terminal device provided in an embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a cloud data center power allocation program. When the processor executes the computer program, the steps in the above-mentioned cloud data center power allocation method embodiments are implemented, such as Figure 1Steps S1 to S4 are shown. Or the processor implements the functions of each module in the above-mentioned device embodiments when executing the computer program.

[0070] Exemplarily, the computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments that can complete functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device. For example, the computer program may be divided into various modules, and the specific functions of each module are not described again.

[0071] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the schematic diagram is merely an example of a terminal device and does not constitute a limitation on the terminal device. The terminal device may include more or fewer components than shown in the diagram, or may combine certain components, or different components. For example, the terminal device may also include an input / output device, a network access device, a bus, etc.

[0072] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0073] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0074] Wherein, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0075] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A cloud data center power allocation method, characterized in that: The method comprises: The acquired electricity consumption data is clustered based on the similarity index of user attributes, and the electricity consumption data of different users are divided into different levels; Classify users with different adjustment capabilities and calculate the baseline power of the server based on the classification results; Establishing a power allocation algorithm based on a non-cooperative game model, and determining the best response function of the server according to the reference power; The optimal response function is iteratively solved based on a preset server power allocation algorithm to determine the power allocation of the server.

2. The cloud data center power allocation method according to claim 1, characterized in that: The obtained electricity consumption data is clustered based on the similarity index of user attributes, and the electricity consumption data of different users are divided into different levels, including: Performing standardization processing on the user data of the user to obtain a load standardization sequence; Fitting the probability distribution of the electric quantity amplitude in the load standardization sequence to obtain a cumulative probability distribution function; Divide the user's daily load data into several equal-length periods according to the time domain, calculate the average load of each period, and obtain the load sequence; Performing segmented amplitude division according to the quantiles of the cumulative probability distribution function, and converting the load mean in the load sequence into a corresponding integer level; The integer level corresponding to the load mean in the load sequence is output as a character string.

3. The cloud data center power allocation method according to claim 2, characterized in that: Categorize users with different adjustment abilities, including: Clustering the character strings based on the user attribute similarity index, selecting characters corresponding to the users of the data processing load at the same power usage time, and outputting a new character string; According to the new character string, a state transition probability matrix is ​​determined by using a Markov method, thereby calculating an attribute similarity index; Taking the calculated attribute similarity index as input, K-means is used for clustering to divide users into users with strong attribute similarity and weak attribute similarity.

4. The cloud data center power allocation method according to claim 3, characterized in that: The method further comprises: The demand parameter sequences of different users are used as input vectors of the self-organizing map neural network; Normalizing the weight vector of the competition layer in the self-organizing map neural network and the input vector; Comparing the similarity of all normalized weight vectors in the competition layer with each normalized input vector; The normalized weight vector with the highest similarity is determined as the winning neuron; The output of the winning neuron is determined to be 1, and the outputs of the remaining neurons are determined to be 0, and the electricity consumption data of different users are secondary clustered according to the adjusted weight vector of the winning neuron.

5. The cloud data center power allocation method according to claim 2, characterized in that: The baseline power of the server is calculated based on the classification results, including: According to the classification results, the data processing load of the electricity consumption data of users of different classifications is aggregated into different servers; Merging intermediate response states in a multi-unit state model of a server into equivalent response states; When the server operation state changes, determine the state switching probability of the server switching between different states; Determine the time characteristic distribution of state switching transfer according to the calculated state switching probability; Determine the average residence time of different states based on the calculated time characteristic distribution; Calculate the steady-state probability that the response state of the server reaches a steady state according to the calculated average stay time; The steady-state expected power of each server is calculated according to the expected value formula of instantaneous power, thereby obtaining the total expected value of steady-state response power; The reference power of the server is calculated according to the steady-state probability and the total expected value of the steady-state response power.

6. The cloud data center power allocation method according to claim 3, characterized in that: A power allocation algorithm is established based on a non-cooperative game model, and an optimal response function of the server is determined according to the reference power, including: Construct the network utility function of each server based on the non-cooperative game model; Calculating a Nash equilibrium solution of the non-cooperative game model in the cloud data center according to the constructed network utility function; The optimal response functions of different servers in the cloud data center are calculated according to the Nash equilibrium solution.

7. The cloud data center power allocation method according to claim 6, characterized in that: The best response function is iteratively solved based on a preset server power allocation algorithm to determine the power allocation of the server, including: Initialize the number of iterations and particle swarm optimization algorithm parameters; Update inertia factor; Calculating fitness values ​​for all particles using the network utility function; Update the local best position of each server in each particle and the global best position of all particles; Update the number of iterations, and use the preset particle update model to update the particle step size and particle value; Determine whether the preset convergence condition is met; If not, update the inertia factor, recalculate the fitness value, and update the local best position, global best position particle step size and particle value; If so, the global best position is output as the optimal power value for power allocation.

8. A cloud data center power distribution device, characterized in that: The device comprises: A classification module is used to cluster the acquired power consumption data based on the similarity index of user attributes, and classify the power consumption data of different users into different levels; A power calculation module is used to classify users with different adjustment capabilities and calculate the baseline power of the server according to the classification results; A response module, used to establish a power allocation algorithm based on a non-cooperative game model, and determine the best response function of the server according to the reference power; The allocation module iteratively solves the best response function based on a preset server power allocation algorithm to determine the power allocation of the server.

9. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the cloud data center power allocation method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the cloud data center power allocation method according to any one of claims 1 to 7.

Images