Optimal computing power control method based on deep learning

Abstract

The invention provides an optimal computing power control method based on deep learning. A differential equation model is based on deep learning; and constructing and solving a controller with the optimal computing power so as to realize optimal computing power control of deep learning. According to the method, a controller for solving the optimal computing power through a deep neural network model is considered to realize dynamic allocation of the computing power and control of energy consumption optimization, so that not only is the resource utilization rate improved, but also the computing cost is greatly reduced, and an intelligent solution is provided for meeting multi-task requirements and improving the platform efficiency.

Description

Technical Field

[0001] The present invention relates to technical fields such as computing power management and computing power optimization, and in particular to an optimal computing power control method based on deep learning. Background Art

[0002] With the popularity of computing-intensive tasks and the widespread application of data centers and cloud computing platforms, the control and management of computing resources has become a key issue. Traditional computing power management methods are difficult to achieve optimal cost control and are prone to waste of resources or insufficient performance. With the widespread application of deep learning, the solution to this problem has become urgent. Summary of the invention

[0003] In view of the defects and shortcomings of the prior art, the present invention considers using a deep neural network model to solve the controller of optimal computing power, so as to realize dynamic allocation of computing power and control energy consumption optimization. This not only improves resource utilization, but also greatly reduces computing costs, and provides an intelligent solution for meeting multi-task requirements and improving platform efficiency.

[0004] To this end, the purpose of the present invention is to provide an optimal computing power control method based on deep learning, firstly constructing a differential equation model of deep learning; then designing an optimal computing power control method; and finally solving computing power control parameters. The purpose is to achieve optimal computing power control of deep learning.

[0005] The present invention specifically adopts the following technical solutions:

[0006] A method for optimal computing power control based on deep learning, characterized by: a differential equation model based on deep learning; constructing and solving a controller for optimal computing power to achieve optimal computing power control of deep learning.

[0007] Furthermore, the deep learning differential equation model is constructed according to a deep feedforward neural network mathematical model framework, specifically including the following processes:

[0008] First, according to the deep feedforward neural network mathematical model framework, the modeling is as follows:

[0009]

[0010] Where l∈[1,2,…,L] is the number of hidden layers, x is the input signal, {W L …W l ...W0} is the weight of the deep feedforward neural network, {b L …b l ...b0} is the bias term of the deep feedforward neural network, z L is the output of the deep feedforward neural network, σ() is the activation function of the encoder, {h L ,…hl ,…,h1} represents the output of the hidden layer;

[0011] According to formula (1), we can further define:

[0012]

[0013] Next, the deep feedforward neural network model is transformed; after substituting formula (2) into (1) and taking the derivative of formula (1), we get:

[0014]

[0015] in, It is the derivative of the function (■) with respect to (*);

[0016] Further definition:

[0017]

[0018] The dynamics of a deep feedforward neural network model can be represented as the following large system model:

[0019]

[0020] Define a supervised learning label as follows:

[0021]

[0022] Among them, x d is the label of the learning system; the derivative of the label is defined as follows:

[0023]

[0024] On this basis, subtract the value of formula (7) from both sides of formula (5) get:

[0025]

[0026] Where E(t) = Z(t) - X d , U(t) is the optimal computing power controller,

[0027] Finally, define the following optimal computing power controller:

[0028] U(t)=U1(t)+U2(t) (9)

[0029] Among them, U1(t) is the main computing power controller, and U2(t) is the compensation controller; select U2(t) = B T (t)[B(t)B T (t)]-1 Φ(t), and after substituting formula (9) into formula (8), we get:

[0030]

[0031] Furthermore, when the matrix B(t)B T When there is no inverse matrix for (t), a scalar ρ is introduced to convert the matrix [B(t)B T (t)] -1 Replace it with the following:

[0032] [B(t)B T (t)+ρI] -1 (11)

[0033] Where I is the identity matrix.

[0034] Furthermore, the minimum computing power main controller The design process is:

[0035] First, define the computing power cost function as follows:

[0036]

[0037] Among them, J(E(t), U1(t)) is defined as the computing power cost function, The definition is as follows:

[0038]

[0039] Among them, Q and R are symmetric positive definite matrices;

[0040] In order to obtain the minimum computing power control cost, it is defined as follows:

[0041]

[0042] Where A(Ω) represents the allowed controller set. Formula (14) means to find a minimum computing power master controller in the allowed controller set A(Ω) Get the minimum computing cost J * (E(t));

[0043] Next, we take the derivative of formula (14) and set it to zero, and then consider (12) to obtain the following Hamilton-Jacobi-Bellman equation:

[0044]

[0045] in, is the derivative of E(t);

[0046] Substituting formula (13) into formula (15), we get:

[0047]

[0048] Considering formula (16) as the variable In order to obtain the minimum computing power of the main controller Solve the equation of formula (16) After taking the partial derivative of , we get:

[0049]

[0050] According to formula (17), the minimum computing power main controller is obtained for:

[0051]

[0052] The formula (18) Substituting into formula (16), we get:

[0053]

[0054] The equation listed in formula (19) is a quadratic equation, belonging to X T The form X=Y, where X is a vector and Y is a numeric value; X is solved as follows:

[0055]

[0056] Where C is a vector, subject to the constraint C T C = 1;

[0057] Further based on the result of formula (20), formula (19) is calculated as follows:

[0058]

[0059] Among them, formula (21) is further rewritten to obtain:

[0060]

[0061] Finally, after substituting formula (22) into formula (18), we get the minimum computing power main controller for:

[0062]

[0063] Furthermore, the process of solving the control parameter C of the computing power main controller is:

[0064] First, the definitions:

[0065]

[0066] After derivation, we get:

[0067]

[0068] When C is chosen as:

[0069]

[0070] Get the smallest computing power main controller for:

[0071]

[0072] Substituting formula (27) into formula (25), we obtain:

[0073]

[0074] By choosing the control parameter C in the form of formula (26), the minimum computing power main controller is ensured The form is shown in formula (27) to ensure The corresponding deep feedforward neural network converges during the learning process.

[0075] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of an optimal computing power control method based on deep learning as described above are implemented.

[0076] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of an optimal computing power control method based on deep learning as described above.

[0077] Compared with the existing technology, the present invention and its preferred solution use a deep neural network model to solve the controller of optimal computing power, so as to realize dynamic allocation of computing power and control energy consumption optimization. This not only improves resource utilization, but also greatly reduces computing costs, and provides an intelligent solution to meet multi-task requirements and improve platform efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0079] Figure 1 A schematic diagram of the process steps for constructing an optimal computing power control solution based on deep learning for the present invention. DETAILED DESCRIPTION

[0080] In order to make the features and advantages of this patent more obvious and easy to understand, the following embodiments are specifically described in detail with the accompanying drawings as follows:

[0081] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0082] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0083] like Figure 1 As shown, the embodiment of the present invention provides an analysis and construction process of the proposed optimal computing power control method based on deep learning, which generally includes the steps of first constructing a differential equation model of deep learning; then designing the optimal computing power control method; and finally solving the computing power control parameters, as shown below:

[0084] 1) Building a differential equation model for deep learning

[0085] First, according to the deep feedforward neural network mathematical model framework, the modeling is as follows:

[0086]

[0087] Where l∈[1,2,…,L] is the number of hidden layers, x is the input signal, {W L …W l ...W0} is the weight of the deep feedforward neural network, {b L …b l ...b0} is the bias term of the deep feedforward neural network, z L is the output of the deep feedforward neural network, σ( ) is the activation function of the encoder, {h L ,…h l ,…,h1} represents the output of the hidden layer;

[0088] According to formula (1), we can further define:

[0089]

[0090] Next, the deep feedforward neural network model is transformed; after substituting formula (2) into (1) and taking the derivative of formula (1), we get:

[0091]

[0092] in, It is the derivative of the function (■) with respect to (*);

[0093] Further definition:

[0094]

[0095] The dynamics of a deep feedforward neural network model can be represented as the following large system model:

[0096]

[0097] Define a supervised learning label as follows:

[0098]

[0099] Among them, x d is the label of the learning system; the derivative of the label is defined as follows:

[0100]

[0101] On this basis, subtract the value of formula (7) from both sides of formula (5) get:

[0102]

[0103] Where E(t) = Z(t) - X d , U(t) is the optimal computing power controller,

[0104] Finally, define the following optimal computing power controller:

[0105] U(t)=U1(t)+U2(t) (9)

[0106] Among them, U1(t) is the main computing power controller, and U2(t) is the compensation controller; select U2(t) = B T (t)[B(t)B T (t)] -1 Φ(t), and after substituting formula (9) into formula (8), we get:

[0107]

[0108] Note: When B(t)B T When there is no inverse matrix for (t), a scalar ρ is introduced to convert the matrix [B(t)B T (t)] -1 Replace it with the following:

[0109] [B(t)B T (t)+ρI] -1 (11)

[0110] Where I is the identity matrix.

[0111] 2) Design of optimal computing power control method

[0112] First, define the computing power cost function as follows:

[0113]

[0114] Among them, J(E(t), U1(t)) is defined as the computing power cost function, The definition is as follows:

[0115]

[0116] Among them, Q and R are symmetric positive definite matrices;

[0117] In order to obtain the minimum computing power control cost, it is defined as follows:

[0118]

[0119] Where A(Ω) represents the allowed controller set. Formula (14) means to find a minimum computing power master controller in the allowed controller set A(Ω) Get the minimum computing cost J * (E(t));

[0120] Next, we take the derivative of formula (14) and set it to zero, and then consider (12) to obtain the following Hamilton-Jacobi-Bellman equation:

[0121]

[0122] in, is the derivative of E(t);

[0123] Substituting formula (13) into formula (15), we get:

[0124]

[0125] Considering formula (16) as the variable In order to obtain the minimum computing power of the main controller Solve the equation of formula (16) After taking the partial derivative of , we get:

[0126]

[0127] According to formula (17), the minimum computing power main controller is obtained for:

[0128]

[0129] The formula (18) Substituting into formula (16), we get:

[0130]

[0131] The equation listed in formula (19) is a quadratic equation, belonging to X T The form X=Y, where X is a vector and Y is a numeric value; X is solved as follows:

[0132]

[0133] Where C is a vector, subject to the constraint C T C = 1;

[0134] Further based on the result of formula (20), formula (19) is calculated as follows:

[0135]

[0136] Among them, formula (21) is further rewritten to obtain:

[0137]

[0138] Finally, after substituting formula (22) into formula (18), we get the minimum computing power main controller for:

[0139]

[0140] 3) Solving the computing power control parameters

[0141] First, the definitions:

[0142]

[0143] After derivation, we get:

[0144]

[0145] When C is chosen as:

[0146]

[0147] Get the smallest computing power main controller for:

[0148]

[0149] Substituting formula (27) into formula (25), we obtain:

[0150]

[0151] By choosing the control parameter C in the form of formula (26), the minimum computing power main controller is ensured The form is shown in formula (27) to ensure The corresponding deep feedforward neural network converges during the learning process.

[0152] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0153] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0154] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0155] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0156] This patent is not limited to the above-mentioned optimal implementation method. Anyone can derive various other forms of an optimal computing power control method based on deep learning under the inspiration of this patent. All equal changes and modifications made according to the scope of the patent application of the present invention should be covered by this patent.

Claims

1. An optimal computing power control method based on deep learning, characterized in that: Differential equation model based on deep learning; construct and solve the optimal computing power controller to achieve optimal computing power control and convergence for deep learning.

2. The optimal computing power control method based on deep learning according to claim 1, characterized in that: The deep learning differential equation model is constructed according to the deep feedforward neural network mathematical model framework, which specifically includes the following processes: First, according to the deep feedforward neural network mathematical model framework, the modeling is as follows: Where l∈[1,2,…,L] is the number of hidden layers, x is the input signal, {W L …W l ...W0} is the weight of the deep feedforward neural network, {b L …b l ...b0} is the bias term of the deep feedforward neural network, z L is the output of the deep feedforward neural network, σ( ) is the activation function of the encoder, {h L ,…h l ,…,h1} represents the output of the hidden layer; According to formula (1), we can further define: Next, the deep feedforward neural network model is transformed; after substituting formula (2) into (1) and taking the derivative of formula (1), we get: in, It is the derivative of the function (■) with respect to (*); Further definition: The dynamics of a deep feedforward neural network model can be represented as the following large system model: Define a supervised learning label as follows: Among them, x d is the label of the learning system; the derivative of the label is defined as follows: On this basis, subtract the value of formula (7) from both sides of formula (5) get: Where E(t) = Z(t) - X d , U(t) is the optimal computing power controller, Finally, define the following optimal computing power controller: U(t)=U1(t)+U2(t) (9) Among them, U1(t) is the main computing power controller, and U2(t) is the compensation controller; select U2(t) = B T (t)[B(t)B T (t)] -1 Φ(t), and after substituting formula (9) into formula (8), we get:

3. The optimal computing power control method based on deep learning according to claim 2, characterized in that: When the matrix B(t)B T When there is no inverse matrix for (t), a scalar ρ is introduced to convert the matrix [B(t)B T (t)] -1 Replace it with the following: [B(t)B T (t)]+ρI] -1 (11) Where I is the identity matrix.

4. The optimal computing power control method based on deep learning according to claim 2, characterized in that: The smallest computing power main controller The definition of is to find a controller U1(t) within the allowed controller set Α(Ω) to obtain the minimum computing power control cost 5. The optimal computing power control method based on deep learning according to claim 4, characterized in that: The smallest computing power main controller The design process is: First, define the computing power cost function as follows: Among them, J(E(t),U1(t)) is defined as the computing power cost function, The definition is as follows: Among them, Q and R are symmetric positive definite matrices; In order to obtain the minimum computing power control cost, it is defined as follows: Where Α(Ω) represents the set of allowed controllers. Formula (14) means that a minimum computing power master controller is found in the allowed controller set Α(Ω) Get the minimum computing cost J * (E(t)); Next, we take the derivative of formula (14) and set it to zero, and then consider (12) to obtain the following Hamilton-Jacobi-Bellman equation: in, is the derivative of E(t); Substituting formula (13) into formula (15), we get: Considering formula (16) as the variable In order to obtain the minimum computing power of the main controller Solve the equation of formula (16) After taking the partial derivative of , we get: According to formula (17), the minimum computing power main controller is obtained for: The formula (18) Substituting into formula (16), we get: The equation listed in formula (19) is a quadratic equation, belonging to X T The form X=Y, where X is a vector and Y is a numeric value; X is solved as follows: Where C is a vector, subject to the constraint C T C = 1; Further based on the result of formula (20), formula (19) is calculated as follows: Among them, formula (21) is further rewritten to obtain: Finally, after substituting formula (22) into formula (18), we get the minimum computing power main controller for:

6. The optimal computing power control method based on deep learning according to claim 2, characterized in that: The computing power main controller is formed by selecting control parameters in the form of Ensure the minimum computing power main controller The form is And make the deep feedforward neural network converge during the learning process; where Q and R are symmetric positive definite matrices.

7. The optimal computing power control method based on deep learning according to claim 6, characterized in that: The process of solving the control parameter C of the computing power main controller is: First, the definitions: After derivation, we get: When C is chosen as: Get the smallest computing power main controller for: Substituting formula (27) into formula (25), we obtain: By choosing the control parameter C in the form of formula (26), the minimum computing power main controller is ensured The form is shown in formula (27) to ensure The corresponding deep feedforward neural network converges during the learning process.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of an optimal computing power control method based on deep learning are implemented as described in any one of claims 1-7.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of an optimal computing power control method based on deep learning as described in any one of claims 1 to 7 are implemented.

Images