Hybrid Learning Neural Network Architecture
Through the hybrid learning neural network of the double tower deep learning architecture, the HVAC component relationship is automatically learned, and the power efficiency prediction of the computer room is optimized, the calculation-intensive and inaccurate problems in the existing technology is solved, and more efficient energy management is achieved .
Patent Information
- Application Number
- CN201980093428.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-05-15
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2039-05-15
Smart Images

Figure CN113518962B_ABST
Abstract
Description
Background Art
[0001] In a computer room of a data center, an environmental control system, such as a heating, ventilation, and air conditioning (HVAC) system, is provided to maintain an acceptable operating environment for computing devices in the computer room, including components such as servers, power supplies, displays, routers, network and communication modules, etc. Based on the total energy consumed by the computer room and the total energy consumed by the computing devices, the power usage effectiveness (PUE) can be calculated and used to evaluate the energy efficiency of the computer room. The HVAC system can include many repeated and / or similar components, such as chillers, fans, secondary pumps, air conditioners, refrigeration units, water pumps, such as cooling water pumps (CWP), secondary cooling water pumps (SCWP), etc. For example, it is not uncommon to equip more than fifty computer room air conditioning units (CRACs) in a computer room with dozens of temperature and humidity sensors.
[0002] One method for optimizing PUE is computational fluid dynamics (CFD), which uses numerical analysis and data structures to analyze and solve problems involving fluids. CFD contains partial differential equations used to provide predictions of air flow and heat distribution in computer rooms, and CFD is widely used in the design phase. However, CFD-based methods are computationally intensive and not well suited for real-time operation. In addition, intensive verification and validation are required to ensure the accuracy of the simulation, especially when changes are involved.
[0003] Another method that can be used to predict PUE and / or control HVAC systems is deep learning neural networks. Deep learning neural networks do not depend on any physical model and do not distinguish between various input features. Classical neural networks only obtain knowledge / relationships based on historical data without any domain knowledge. General deep learning models are more difficult to apply to systems with a large number of repeated and similar devices, such as computer rooms in data centers. Although these HVAC components have complex nonlinear correlations, from the perspective of the neural network structure, the inputs from sensors are treated equally, and the information behind the data from the inputs may be biased due to repeated and / or similar inputs, which may lead to overfitting and ultimate inaccuracy, thereby causing invalidity.
[0004] In order to avoid repeated and / or similar inputs and improve the PUE or energy consumption of a computer room, a popular solution is to manually aggregate the inputs based on the domain knowledge of human experts and set the aggregated inputs as inputs to a neural network. However, this solution is room-specific and introduces additional manual work. In addition, because this solution relies on the experience and analysis of HVAC experts, it is difficult to fully understand the most reasonable correlations between various HVAC components to achieve energy-efficient computer room conditions under different operating conditions such as external temperature, external humidity, computing load, etc. BRIEF DESCRIPTION OF THE DRAWINGS
[0005] The detailed description is set forth with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The use of the same reference numbers in different figures indicates similar or identical items or features.
[0006] Figure 1 An example block diagram illustrating an environmental control system used with a hybrid learning neural network (HLNN) that can be utilized to predict power usage effectiveness (PUE) of a computer room.
[0007] Figure 2 By related level diagram Figure 1 An example detailed block diagram of an environmental control system.
[0008] Figure 3 An example block diagram illustrating an HLNN architecture.
[0009] Figure 4 An example flow chart depicting a process for predicting PUE by a HLNN is shown. DETAILED DESCRIPTION
[0010] The systems and methods discussed herein relate to predicting the energy efficiency of a computer room in a data center, and more specifically to predicting the power usage effectiveness (PUE) of a computer room according to optimized parameters using a dual-tower deep learning architecture. The dual-tower deep learning architecture can automatically learn embeddings based on data and ontology structures, and can include simultaneous training of two sub-networks, an unsupervised autoencoder network (AE-Net) that captures domain knowledge embedded in the ontology, and a supervised prediction network (P-Net) that predicts PUE based on input. The simultaneous learning of AE-Net (unsupervised) and P-Net (supervised) can be referred to as hybrid learning, and the dual-tower deep learning architecture can also be referred to as a hybrid learning neural network (HLNN) architecture.
[0011] In order to achieve PUE optimization by ensuring a reasonable and appropriate operating environment such as that of a computer room and reducing waste in setting up components of an environmental control system such as an HVAC system, a machine learning method may be used to learn from historical data to obtain the complex relationships between various HVAC components and the energy efficiency of the computer room under different operating conditions.
[0012] In the HLNN architecture, the first neural network and the second neural network, such as AE-Net and P-Net, respectively, can share a shared structure including an input layer and two concept layers. Each of AE-Net and P-Net can have its own hidden layer and output layer. AE-Net, P-Net and the shared structure can form a hybrid learning neural network (HLNN) architecture. AE-Net can be an unsupervised learning network that can be trained to make its output replicate the input with the lowest possible error, while P-Net can be a deep feed-forward neural network for predicting PUE.
[0013] For example, domain knowledge of components associated with the HVAC system and computing equipment of a computer room can be embedded into the HLNN architecture. By embedding the domain knowledge of the components, the number of inputs and the complexity of the search space can be reduced, and the accuracy of PUE prediction can be improved. The design of the dual-tower deep learning architecture of the input layer and the concept layer can be guided by a domain ontology containing multi-level nodes, where the top level can contain the root concept and the bottom level can contain multiple instances. Instances in the bottom level of the ontology can be represented by nodes in the input layer, and concepts in the intermediate levels can also have corresponding nodes in the concept layer of the shared structure. In addition, the relationships and / or connections between the various levels can be replicated in the input layer and the concept layer.
[0014] Figure 1 An example block diagram of an environmental control system 100 for use with a hybrid learning neural network (HLNN) that may be utilized to predict the power usage effectiveness (PUE) of a computer room 102 is illustrated.
[0015] The environmental control system 100 may include a plurality of components, such as a device and data collection module 104 that is communicatively coupled to the HVAC group 106 and the external device and data group 108. The device and data collection module 104 may be configured to maintain profiles of components managed by the HVAC group 106 and the external device and data group 108, receive input data from various sensors associated with those components, and send data to those components to partially control the environment of the computer room 102, and calculate a predicted PUE for the computer room 102. Some of the environmental control system components may be located in the computer room 102, while other components may be located outside the building in which the computer room 102 is located. The environmental control system 100 may monitor energy consumption of components associated with the computer room 102, the device and data collection module 104, the HVAC group 106, and the external device and data group 108. Additionally, the environmental control system 100 may be communicatively coupled to a computer 110. The computer 110 may include one or more processors 112 and a memory 114 communicatively coupled to the one or more processors 112, which may store computer-readable instructions to be executed by the computer 110 to perform the functions of the HLNN described below. The computer 110 may be located within the computer room 102 or may be located remotely from the computer room 102.
[0016] The computer room 102 may house computing devices 116 including servers, power supplies, displays, routers, network and communication modules, etc. (not shown). The computing devices 116 may be coupled to the environmental control system 100 and may provide information about the energy usage of the computing devices 116 for a predicted PUE of the computer room 102 based on historical, current, and expected energy usage and computing load.
[0017] Figure 2 Graphical representation by relevant level (levels 1-4 shown) Figure 1 Detailed block diagram of an example environmental control system 100.
[0018] The HVAC group 106 may include an HVAC control module 202, an air conditioning group 204, and a refrigeration group 206 that are communicatively coupled to the device and data collection module 104. The HVAC control module 202 may be configured to receive operational information from various sensors and controllers of the air conditioning group 204 and the refrigeration group 206. The HVAC control module 202 may forward the operational information to the device and data collection module 104 for calculation by the HLNN. The HVAC control module 202 may also be configured to send control information received from the device and data collection module 104 to the air conditioning group 204 and the refrigeration group 206 for adjusting various parameters of the air conditioning group 204 and the refrigeration group 206 to optimize the desired parameters for predicting the PUE. The HVAC group 106 may also include a secondary pump group (not shown) and may similarly transmit relevant operational information to and from the HVAC control module 202.
[0019] The air conditioner group 204 may include N air conditioners (two, AC-1 208 and AC-N 210, as shown). Although not shown, each of the N air conditioners may include several controllers and sensors, such as corresponding switches, corresponding fan speed controllers / sensors, corresponding air conditioner output air temperature sensors, and corresponding air conditioner return air temperature sensors. Each of the N air conditioners may be configured to receive AC operation information from the corresponding controllers and sensors and forward the AC operation information to the air conditioning system 204, which in turn forwards the AC operation information to the HVAC control module 202. Each of the N air conditioners may also be configured to send AC control information received from the air conditioning system 204 to the corresponding controls to optimize the desired parameters for predicting the PUE.
[0020] The refrigeration group 206 may include a plurality of refrigeration systems including a plurality of chillers (shown as chiller 1212) and a plurality of cooling towers (shown as tower 1214). Although not shown, each of the plurality of chillers may include an associated switch, cooling mode controller, outflow cooling water temperature controller / sensor, and each of the plurality of cooling towers may include an associated cooling tower fan speed controller / sensor, outflow cooling water temperature controller / sensor, and return cooling water temperature controller / sensor.
[0021] Each of the plurality of refrigeration systems may be configured to receive refrigeration operation information from corresponding controls, switches, and sensors (not shown) and forward the refrigeration operation information to the HVAC control module 202 via the refrigeration group 206. Each of the plurality of refrigeration systems may also be configured to send refrigeration control information received from the refrigeration group 206 to corresponding controls, switches, and sensors to optimize the desired parameters for predicting the PUE.
[0022] The external device and data group 108 may include an external device monitoring module 216, an external humidity module 218, an external wet bulb temperature module 220, and other modules (not shown) that are communicatively coupled to the device and data collection module 104. The external humidity module 218 may be communicatively coupled to M humidity sensors (two humidity sensors, humidity sensor-1 222 and humidity sensor-M 224, as shown). The external wet bulb temperature module 220 may be communicatively coupled to M wet bulb temperature sensors (two wet bulb temperature sensors, wet bulb temperature sensor-1 226 and wet bulb temperature sensor-M 228, as shown). The external device monitoring module 216 may receive humidity and wet bulb temperature information from the corresponding sensors and forward the information to the device and data collection module 104 for optimizing the desired parameters for predicting the PUE.
[0023] Figure 2 Each block illustrated can be associated with one of the multiple levels of the domain ontology. A domain ontology with four levels is illustrated as an example in this article, however, the number of levels of the domain ontology may not be limited to four and may also be more or less than four levels. Level 1 may include a device and data collection module 104, which may be referred to as D1. Level 2 may include q modules including an HVAC control module 202 and an external device monitoring module 216, which may be referred to as C_1, C_2, ... C_q, respectively. Level 3 may include p modules including an air conditioning group 204, a refrigeration group 206, an external humidity module 218, and an external wet bulb temperature module 220, which may be referred to as B_1, B_2, ... B_p, respectively. Level 4 may include k modules including AC-1 208, AC-N 210, cooler-1 212, tower-1 214, humidity sensor-1 222, humidity sensor-M 224, wet bulb temperature sensor-1 226, and wet bulb temperature sensor-M 228, which may be referred to as A_1, A_2, ... A_k, respectively.
[0024] Figure 3 An example block diagram of a HLNN architecture 300 is illustrated.
[0025] The HLNN structure 300 may include a domain ontology 302, a shared structure 304, a first neural network such as AE-Net 306, and a second neural network such as P-Net 308. There may be multiple levels in the ontology, and as an example, the domain ontology 302 is shown with Figure 2The illustrated blocks correspond to four levels. The top level 1 may contain the root concept D_1 310, while the bottom level 4 may contain multiple instances, of which four instances A_1 312, A_2 314, A_n 316, and A_k 318 are shown. These four instances in the 4th level of the domain ontology 302 may be represented as nodes A_1 320, A_2 322, A_n 324, and A_k 326 in the input layer 328 of the shared structure 304, respectively.
[0026] The second level, level 2, and the third level, level 3, of the domain ontology 302 may represent multiple instances, wherein two concepts, C_1 330 and C_q 332 (in level 2), and three concepts, B_1 334, B_2 336, and B_p 338 (in level 3), are shown. These instances in level 2 and level 3 of the domain ontology 302 may also have corresponding nodes, C_1 340, C_q 342, B_1 344, B_2 346, and B_p 348, respectively, in the concept layer 350 of the shared structure 304. Additionally, the relationships / connections between the levels may also be replicated in the input layer 328 and the concept layer 350. For example, in the domain ontology 302 , concept B_1 334 is shown connected to a set of instances A_1 312 , A_2 314 , and A_n 316 , and in the concept layer 350 , the corresponding node B_1 344 is also shown connected to the corresponding nodes A_1 320 , A_2 322 , and A_n 324 in the input layer 328 .
[0027] P-Net 308 may be a deep feed-forward neural network and may include hidden layers 352 and a single-node output layer 354 for outputting PUE parameters 356, plus input layer 328 and concept layer 350 sharing structure 304. An example feed-forward operation of P-Net 308 is described below. Neurons and nodes may be used interchangeably.
[0028] make represents the weight between the jth neuron or node in the (l-1)th layer and the ith neuron in the lth layer, and is the bias of the ith neuron in layer l. Using these notations, the feedforward operation can be described as
[0029]
[0030] in is the weighted input of node i, is the output of the jth node in the (l-1)th layer, and R l-1 is the number of neurons in the (l-1)th layer.
[0031] Given and Equation (1) can be simplified to:
[0032]
[0033] Using the above notation, the activation of node i is
[0034]
[0035] where f p is the activation function.
[0036] In shared structure 304, connections may be guided by domain knowledge, which may not adequately connect nodes in concept layer 350. Represents the conceptual relationship between two concept nodes, node j and node i, and then the weighted input of node i can be expressed as:
[0037]
[0038] Each layer of the concept layer 350 can be mapped from the corresponding level of the concept in the domain ontology 302. represents the number of nodes (instances and concepts) in the concept hierarchy connected to the i-th node in the l-th layer (i.e., node C_i in the domain ontology 302), and then the concept relationship weight Expressed as
[0039]
[0040] That is, if the subconcept / instance node j corresponding to the (l-1)th layer is not connected to the concept of node i, then is zero. If Then the concept relationship weight is 1, which does not affect the learning process. The loss function L PN (a,d p ) Then we can take input a and let a represent the input vector, let o p Denote the output of the calculated neural network and let d p Denotes the expected output to define d p With o p The error between .
[0041] AE-Net 306 may be an unsupervised learning model including hidden layers 358 and output layers 360 plus the input layer 328 and concept layer 350 of the shared structure 304. AE-Net 306 may be designed to minimize the difference between the input from the input layer 328 of the shared structure 304 and the output from the output layer 360. Considering the input vector a from the input layer 328, the representation vector c from the top concept layer of the concept layer 350, and the output vector r from the output layer 360 (R_1 362 and R_k 364 shown), the mapping that transforms a into c may be referred to as an encoder, while the mapping that transforms c back into r may be referred to as a decoder. The encoder may consist of the input layer 328 and the concept layer 350, whereas the decoder may consist of the hidden layer 358 and the output layer 360. The training process in AE-Net 306 may help the encoder preserve domain knowledge in the domain ontology 302.
[0042] Set the input vector a to a={a1,a2,...,a k}, indicating that the vector c can be expressed as c = {c1, c2, ..., c q}, and the output vector r can be expressed as r = {r1, r2, ..., r k}, then the encoder function f θ and the decoder function g θ Expressed as:
[0043] c=f θ (a) (6)
[0044] r = g θ (c) (7)
[0045] In the encoder function f θ and the decoder function g θ In the example, the parameter set is θ = {W, b, W', d}, where W and W' are the encoder and decoder weight matrices, and b and d are the encoder and decoder bias vectors. The encoder function f can then be θ and the decoder function g θ They are expressed as:
[0046] f θ (a) = s f (b+Wa) (8)
[0047] g θ (c) = s g (d+Wc) (9)
[0048] where s f and gare the encoder and decoder activation functions. In terms of probability, r is not the exact reconstruction of a but the parameters of the distribution p(A|R=r) that generates a with high probability. The AE-Net 306 can be trained to find the set of parameters that minimizes the reconstruction error in the following equation:
[0049] E AE (θ)=∑ a∈A L AE (a, r) = ∑ a∈A L AE (a,g θ (f θ (a))) (10)
[0050] Where A represents the training example set, L AE is the loss function or reconstruction error. The input vector may be real-valued and the loss function L AE It can be the square error L AE (a, r) = ||ar|| 2 .s f and g Both can be sigmoid functions.
[0051] The HLNN 300 can be trained in a similar manner to a standard neural network. The only difference may be the loss function L Model It can be composed of two components: the loss L of AE-Net 306 AE , and the prediction loss L of P-Net 308 PN :
[0052] L Model =L PN +αL AE (11)
[0053] Where α is the direction to L AE Provides a constant for bias or weight. Alternatively, L PN can be biased or weighted by another constant β, and L Model Expressed as L Model =βL PN +L AE .
[0054] During training, the derivative of the loss can be expressed as
[0055]
[0056] And the following substitutions can be made:
[0057]
[0058]
[0059] Where R l+1 is the number of nodes in the (l+1)th layer. Combining equations (12), (13) and (14), we obtain
[0060]
[0061] If layer l+1 is in the shared structure 304, namely the input layer 328 and the concept layer 350, then equation (15) can be transformed into:
[0062]
[0063] If layer l+1 is in the hidden layer 352 or the output layer 354 of P-Net 308, equation (15) can be transformed into:
[0064]
[0065] in represents the number of nodes in the (l+1)th layer used to calculate the output of the P-Net 308.
[0066] If layer l+1 is the hidden layer 358 or the output layer 360 of AE-Net 306, equation (15) can be transformed into:
[0067]
[0068] in represents the number of nodes in the (l+1)th layer used to calculate the output of AE-Net 306.
[0069] Equations (16), (17) and (18) show that the loss function L Model The derivatives of are back-propagated for learning both AE-Net 306 and P-Net 308. This can be achieved by passing through the loss function L as expressed by Equation (11) Model The calculated loss is minimized to optimize the PUE solution, which can be achieved by transforming the loss function L Model This is accomplished by setting the derivative of to zero and solving for the variables as in equations (16), (17), and (18). Because the solution may not always converge to zero, or may take longer than an acceptable time or number of iterations, the value of the derivative may be set to a sufficiently small and acceptable threshold.
[0070] Figure 4 An example flow chart 400 depicts a process for predicting power utilization effectiveness (PUE) by the HLNN 300 .
[0071] At block 402, HLNN 300 may create Figure 1-3The illustrated ontology having multiple levels of components associated with the environmental control system 100, which is in turn associated with the computer room 102, such as the domain ontology 302. The HLNN 300 can automatically receive information about components associated with the environmental control system 100, including corresponding relevant historical data, locations and physical connections, and information such as Figure 1-3 The hierarchy between components is shown. The computing device 116 may include servers, power supplies, displays, routers, network and communication modules (telephone, Internet, wireless devices, etc.), etc. The relationship between the components of the environmental control system 100 and the computing device 116 may be based on the load of the computing device 116, such as the workload or computational load of the server and the electrical load of the server as a function of the workload of the server.
[0072] At block 404, the HLNN 300 may receive input feature parameters of components associated with the environmental control system 100. More specifically, the input layer 328 of the shared structure 304 may receive k instances, A_1 312, A_2 314, A_n 316, and A_k 318, from the domain ontology 302, where k is an integer. Each of the k instances may have a corresponding input feature parameter (e.g., Figure 3 The illustrated A_1 320 , A_2 322 , A_n 324 , and A_k 326 ) may belong to one or more corresponding upper-level concepts of the plurality of upper-level concepts hierarchically illustrated in the concept layer 350 .
[0073] At block 406, both the first neural network, such as AE-Net 306, and the second neural network, such as P-Net 308, may be trained simultaneously. As discussed above, the input vector a or input feature parameter may be expressed as a={a1, a2, ..., ak}, and the representation vector c or concept may be expressed as c={c1, c2, ..., c q}, and the output vector r can be expressed as r = {r1, r2, ..., r k}. The mapping that transforms a into c can be called an encoder, and the mapping that transforms c back into r can be called a decoder. The encoder can be composed of an input layer 328 and a concept layer 350, and the decoder can be composed of a hidden layer 358 and an output layer 360. The training process in AE-Net 306 can help the encoder preserve domain knowledge in the domain ontology 302.
[0074] At block 408, the HLNN 300 may utilize the trained AE-Net 306 and the trained P-Net 308 to perform a loss function L Model The loss is minimized and the power usage effectiveness (PUE) of the computer room 102 is predicted at block 410. The loss function L ModelThe derivatives of, such as equations (16), (17), and (18), can be set to zero for solving for the variables. Because the solution may not always converge to zero, or may take longer than an acceptable time, the value of the derivative can be set to a sufficiently small and acceptable threshold.
[0075] The trained neural network can be automatically generated, and the training of the trained neural network can be performed by using a gradient descent algorithm to achieve learning of input feature parameters for corresponding concepts. The architecture of the trained neural network can reflect deep learning of multiple components and related concepts based on the relationships between the multiple components. The trained neural network can include hierarchical concept layers, such as concept layer 350, coupled between an input layer, such as input layer 328, and an output layer, such as output layer 354 or 360. Figure 3 As shown, a concept layer 350 is added between the input layer 328 and the hidden layers 352 and 358. The concept layer 350 can be embedded with domain knowledge from the domain ontology 302. The concept layer 350 can construct a concept structure based on the relationship between multiple components. The concept structure can be created manually or automatically with intelligent components that can communicate with each other. The training portion of the HLNN 300 and the prediction of PUE using the HLNN 300 can be performed separately and / or by different parties.
[0076] A general deep learning network may not be able to reasonably distinguish all repeated and / or similar input features, and may identify the importance of each feature based entirely on historical data. In a structure such as a computer room 102 with a large number of repeated and similar devices, if these repeated and / or similar input feature parameters are not classified, aggregated or abstracted, the complexity of the network and the space for learning and searching will be greatly increased, thereby requiring higher quality and quantity of data. Although it may be easy to obtain unreasonable overfitting, it will reduce the prediction accuracy.
[0077] Some or all of the operations of the above methods may be performed by executing computer-readable instructions stored on a computer-readable storage medium as defined below. The term "computer-readable instructions" as used in the specification and claims includes routines, applications, application modules, program modules, programs, components, data structures, algorithms, etc. The computer-readable instructions can be implemented on a variety of system configurations, including single-processor or multi-processor systems, microcomputers, mainframe computers, personal computers, handheld computing devices, microprocessor-based programmable consumer electronic devices, combinations thereof, etc.
[0078] Computer-readable storage media may include volatile memory (such as random access memory (RAM)) and / or non-volatile memory (such as read-only memory (ROM), flash memory, etc.). Computer-readable storage media may also include additional removable storage devices and / or non-removable storage devices, including but not limited to flash memory, magnetic storage devices, optical storage devices, and / or tape storage devices that can provide non-volatile storage of computer-readable instructions, data structures, program modules, etc.
[0079] Non-transient computer-readable storage media are examples of computer-readable media. Computer-readable media include at least two types of computer-readable media, namely computer-readable storage media and communication media. Computer-readable storage media include volatile and non-volatile, removable and non-removable media implemented with any process or technology to store information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage device, magnetic cassette, magnetic tape, magnetic disk storage device or other magnetic storage device, or any other non-transmission medium that can be used to store information for access by a computing device. In contrast, communication media can embody computer-readable instructions, data structures, program modules or other data in modulated data signals such as carrier waves or other transmission mechanisms. As defined herein, computer-readable storage media do not include communication media.
[0080] The computer readable instructions stored on one or more non-transitory computer readable storage media, when executed by one or more processors, may perform the above-referenced Figure 1-4 Typically, computer-readable instructions include routines, programs, objects, components, data structures, etc. that perform specific functions or implement specific abstract data types. The order in which the operations are described is not intended to be construed as a limitation, and any number of the described operations can be combined in any order and / or in parallel to implement the processes.
[0081] Sample Clauses
[0082] A. A method, comprising: receiving input feature parameters of multiple components associated with at least one computer room; training a first neural network and a second neural network based on the input feature parameters; and predicting a power usage effectiveness (PUE) of the at least one computer room based on an output of the first neural network and an output of the second neural network.
[0083] B. The method of paragraph A, wherein the first neural network is an unsupervised neural network and the second network is a supervised predictive neural network.
[0084] C. The method of paragraph A, wherein training the first neural network and the second neural network based on the input feature parameters comprises simultaneously training the first neural network and the second neural network based on the input feature parameters.
[0085] D. A method as described in paragraph A, wherein receiving input feature parameters of the multiple components associated with the at least one computer room includes: creating an ontology having multiple levels associated with the multiple components; and receiving information of the multiple components based on the ontology, including corresponding related concepts, historical data, locations, physical connections, and hierarchies between the multiple components.
[0086] E. The method of paragraph D, wherein the relationship between the plurality of components is based at least in part on the load of computing devices in the computer room.
[0087] F. The method of paragraph E, wherein the load of the computing device comprises a workload of the computing device and an electrical load used by the computing device.
[0088] G. The method of paragraph F, wherein the computing device includes a server and a power supply for the server.
[0089] H. The method of paragraph D, wherein training the first neural network and the second neural network based on the input feature parameters comprises using a gradient descent algorithm to implement learning of the input feature parameters for corresponding concepts.
[0090] I. The method of paragraph A, wherein predicting the PUE of the at least one computer room based on the output of the first neural network and the output of the second neural network includes biasing a loss associated with the first neural network with a constant value.
[0091] J. The method of paragraph I, wherein predicting the PUE of the at least one computer room based on the output of the first neural network and the output of the second neural network includes minimizing a total loss calculated based on a biased loss associated with the first neural network and an unbiased loss associated with the second neural network.
[0092] K. The method of paragraph J, wherein minimizing the loss calculated based on the loss function by utilizing the first neural network and the second neural network comprises solving a derivative of the loss function equal to zero.
[0093] L. The method of paragraph J, wherein minimizing the loss calculated based on the loss function by utilizing the trained first neural network and the trained second neural network comprises solving for a derivative of the loss function that is less than or equal to a threshold.
[0094] M. A system comprising: one or more processors; and a memory communicatively coupled to the one or more processors, the memory storing computer-readable instructions executable by the one or more processors, which, when executed by the one or more processors, cause the one or more processors to perform operations comprising the following steps: receiving input feature parameters of a plurality of components associated with at least one computer room; training a first neural network and a second neural network based on the input feature parameters; and predicting a power usage effectiveness (PUE) of the at least one computer room based on an output of the first neural network and an output of the second neural network.
[0095] N. The system of paragraph M, wherein the first neural network is an unsupervised neural network and the second network is a supervised predictive neural network.
[0096] O. The system of paragraph M, wherein training the first neural network and the second neural network based on the input feature parameters comprises simultaneously training the first neural network and the second neural network based on the input feature parameters.
[0097] P. A system as described in paragraph M, wherein receiving input feature parameters of the multiple components associated with the at least one computer room includes: creating an ontology having multiple levels associated with the multiple components; and receiving information about the multiple components based on the ontology, including corresponding related concepts, historical data, locations, physical connections, and hierarchies between the multiple components.
[0098] Q. A system as described in paragraph P, wherein the relationship between the multiple components is based at least in part on the load of the computing device.
[0099] R. The system of paragraph Q, wherein the load of the computing device comprises a workload of the computing device and an electrical load used by the computing device.
[0100] S. The system of paragraph R, wherein the computing device includes a server and a power supply for the server.
[0101] T. The system of paragraph P, wherein training the first neural network and the second neural network based on the input feature parameters comprises using a gradient descent algorithm to implement learning of the input feature parameters for corresponding concepts.
[0102] U. A system as described in paragraph M, wherein predicting the PUE of the at least one computer room based on the output of the first neural network and the output of the second neural network includes biasing the losses associated with the first neural network with a constant value.
[0103] V. A system as described in paragraph U, wherein predicting the PUE of the at least one computer room based on the output of the first neural network and the output of the second neural network includes minimizing a total loss calculated based on a biased loss associated with the first neural network and an unbiased loss associated with the second neural network.
[0104] W. The system of paragraph V, wherein minimizing the loss calculated based on the loss function by utilizing the first neural network and the second neural network comprises solving a derivative of the loss function equal to zero.
[0105] X. A system as described in paragraph V, wherein minimizing the loss calculated based on the loss function by utilizing the trained first neural network and the trained second neural network includes solving for a derivative of the loss function that is less than or equal to a threshold.
[0106] Y. A non-transitory computer-readable storage medium storing computer-readable instructions executable by one or more processors, which, when executed by the one or more processors, cause the one or more processors to perform operations including the following steps: receiving input feature parameters of multiple components associated with at least one computer room; training a first neural network and a second neural network based on the input feature parameters; and predicting the power usage effectiveness (PUE) of the at least one computer room based on an output of the first neural network and an output of the second neural network.
[0107] Z. The non-transitory computer-readable storage medium of paragraph Y, wherein the first neural network is an unsupervised neural network and the second network is a supervised predictive neural network.
[0108] AA. The non-transitory computer-readable storage medium of paragraph Y, wherein training the first neural network and the second neural network based on the input feature parameters comprises simultaneously training the first neural network and the second neural network based on the input feature parameters.
[0109] AB. A non-transitory computer-readable storage medium as described in paragraph Y, wherein receiving input feature parameters of the multiple components associated with the at least one computer room includes: creating an ontology having multiple levels associated with the multiple components; and receiving information of the multiple components based on the ontology, including corresponding related concepts, historical data, locations, physical connections, and hierarchies between the multiple components.
[0110] AC. The non-transitory computer-readable storage medium of paragraphs AB, wherein the relationship between the plurality of components is based at least in part on the load of computing devices in the computer room.
[0111] AD. The non-transitory computer-readable storage medium of paragraphs AC, wherein the load of the computing device comprises a workload of the computing device and an electrical load used by the computing device.
[0112] AE. The non-transitory computer-readable storage medium of paragraph AD, wherein the computing device comprises a server and a power source for the server.
[0113] AF. The non-transitory computer-readable storage medium of paragraphs AB, wherein training the first neural network and the second neural network based on the input feature parameters comprises using a gradient descent algorithm to implement learning of the input feature parameters for corresponding concepts.
[0114] AG. The non-transitory computer-readable storage medium of paragraph Y, wherein predicting the PUE of the at least one computer room based on the output of the first neural network and the output of the second neural network comprises biasing a loss associated with the first neural network with a constant value.
[0115] AH. The non-transitory computer-readable storage medium of paragraphs AG, wherein predicting the PUE of the at least one computer room based on the output of the first neural network and the output of the second neural network comprises minimizing a total loss calculated based on a biased loss associated with the first neural network and an unbiased loss associated with the second neural network.
[0116] AI. The non-transitory computer-readable storage medium of paragraphs AH, wherein minimizing the loss calculated based on the loss function by utilizing the first neural network and the second neural network comprises solving for a derivative of the loss function equal to zero.
[0117] AJ. The non-transitory computer-readable storage medium of paragraph AH, wherein minimizing the loss calculated based on the loss function by utilizing the trained first neural network and the trained second neural network comprises solving for a derivative of the loss function to be less than or equal to a threshold.
[0118] in conclusion
[0119] Although the subject matter has been described in language specific to structural features and / or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described. Rather, the specific features and acts are disclosed as example forms of implementing the claims.
Claims
1. A method, the method include: receiving input characteristic parameters of a plurality of components associated with at least one computer room; Training a first neural network and a second neural network based on the input feature parameters; as well as Predicting the power usage efficiency (PUE) of the at least one computer room based on the output of the first neural network and the output of the second neural network; the first neural network is trained to minimize the difference between the input and the output, and the second neural network is a deep feedforward neural network for predicting the power usage efficiency (PUE); the first neural network and the second neural network share a shared structure including an input layer and a concept layer; Wherein, the first neural network and the second neural network are guided by a domain ontology comprising multi-level nodes.
2. The method of claim 1, wherein the first neural network is an unsupervised neural network and the second neural network is a supervised predictive neural network.
3. The method of claim 1 , wherein training the first neural network and the second neural network based on the input feature parameters comprises simultaneously training the first neural network and the second neural network based on the input feature parameters.
4. The method of claim 1, wherein receiving input characteristic parameters of the plurality of components associated with the at least one computer room include: creating an ontology having a plurality of levels associated with the plurality of components; as well as Information of the plurality of components is received based on the ontology, including corresponding related concepts, historical data, locations, physical connections, and hierarchies between the plurality of components.
5. The method of claim 4, wherein the relationship between the plurality of components is based at least in part on the load of computing devices in the computer room.
6. The method of claim 5, wherein the load of the computing device comprises a workload of the computing device and an electrical load used by the computing device.
7. The method of claim 6, wherein the computing device comprises a server and a power supply for the server.
8. The method of claim 4, wherein training the first neural network and the second neural network based on the input feature parameters comprises using a gradient descent algorithm to achieve learning of the input feature parameters for corresponding concepts.
9. The method of claim 1, wherein predicting a power usage effectiveness (PUE) of the at least one computer room based on an output of the first neural network and an output of the second neural network comprises biasing a loss associated with the first neural network with a constant value.
10. The method of claim 9, wherein predicting a power usage effectiveness (PUE) of the at least one computer room based on the output of the first neural network and the output of the second neural network comprises minimizing a total loss calculated based on a biased loss associated with the first neural network and an unbiased loss associated with the second neural network.
11. The method of claim 10, wherein minimizing the total loss calculated based on a loss function by utilizing the first neural network and the second neural network comprises solving a derivative of the loss function equal to zero.
12. The method of claim 10, wherein minimizing the total loss calculated based on a loss function by utilizing the trained first neural network and the trained second neural network comprises solving a derivative of the loss function to be less than or equal to a threshold.
13. A system, wherein the system include: one or more processors; as well as a memory communicatively coupled to the one or more processors, the memory storing computer-readable instructions executable by the one or more processors, which, when executed by the one or more processors, cause the one or more processors to perform operations comprising: receiving input characteristic parameters of a plurality of components associated with at least one computer room; Training the first neural network and the second neural network based on the input feature parameters; and Predicting the power usage efficiency (PUE) of the at least one computer room based on the output of the first neural network and the output of the second neural network; the first neural network is trained to minimize the difference between the input and the output, and the second neural network is a deep feedforward neural network for predicting the power usage efficiency (PUE); the first neural network and the second neural network share a shared structure including an input layer and a concept layer; Wherein, the first neural network and the second neural network are guided by a domain ontology comprising multi-level nodes.
14. The system of claim 13, wherein the first neural network is an unsupervised neural network and the second neural network is a supervised predictive neural network.
15. The system of claim 13, wherein training the first neural network and the second neural network based on the input feature parameters comprises simultaneously training the first neural network and the second neural network based on the input feature parameters.
16. The system of claim 13, wherein receiving input characteristic parameters of the plurality of components associated with the at least one computer room include: creating an ontology having a plurality of levels associated with the plurality of components; as well as Information of the plurality of components is received based on the ontology, including corresponding related concepts, historical data, locations, physical connections, and hierarchies between the plurality of components.
17. The system of claim 16, wherein the relationship between the plurality of components is based at least in part on a load of a computing device.
18. The system of claim 17, wherein the load of the computing device comprises a workload of the computing device and an electrical load used by the computing device.
19. The system of claim 18, wherein the computing device comprises a server and a power supply for the server.
20. The system of claim 16, wherein training the first neural network and the second neural network based on the input feature parameters comprises using a gradient descent algorithm to achieve learning of the input feature parameters for corresponding concepts.
21. The system of claim 13, wherein predicting a power usage effectiveness (PUE) of the at least one computer room based on an output of the first neural network and an output of the second neural network comprises biasing a loss associated with the first neural network with a constant value.
22. The system of claim 21, wherein predicting the power usage effectiveness (PUE) of the at least one computer room based on the output of the first neural network and the output of the second neural network comprises minimizing a total loss calculated based on a biased loss associated with the first neural network and an unbiased loss associated with the second neural network.
23. The system of claim 22, wherein minimizing the total loss calculated based on a loss function by utilizing the first neural network and the second neural network comprises solving a derivative of the loss function equal to zero.
24. The system of claim 22, wherein minimizing the total loss calculated based on a loss function by utilizing the trained first neural network and the trained second neural network comprises solving for a derivative of the loss function to be less than or equal to a threshold.
25. A non-transitory computer-readable storage medium storing computer-readable instructions executable by one or more processors, which, when executed by the one or more processors, cause the one or more processors to perform operations comprising: receiving input characteristic parameters of a plurality of components associated with at least one computer room; Training the first neural network and the second neural network based on the input feature parameters; and Predicting the power usage efficiency (PUE) of the at least one computer room based on the output of the first neural network and the output of the second neural network; the first neural network is trained to minimize the difference between the input and the output, and the second neural network is a deep feedforward neural network for predicting the power usage efficiency (PUE); the first neural network and the second neural network share a shared structure including an input layer and a concept layer; in, The first neural network and the second neural network are guided by a domain ontology comprising multi-level nodes.
26. The non-transitory computer-readable storage medium of claim 25, wherein the first neural network is an unsupervised neural network and the second neural network is a supervised predictive neural network.
27. The non-transitory computer-readable storage medium of claim 25, wherein training the first neural network and the second neural network based on the input feature parameters comprises simultaneously training the first neural network and the second neural network based on the input feature parameters.
28. The non-transitory computer readable storage medium of claim 25, wherein receiving input characteristic parameters of the plurality of components associated with the at least one computer room include: creating an ontology having a plurality of levels associated with the plurality of components; as well as Information of the plurality of components is received based on the ontology, including corresponding related concepts, historical data, locations, physical connections, and hierarchies between the plurality of components.
29. The non-transitory computer-readable storage medium of claim 28, wherein the relationship between the plurality of components is based at least in part on the load of computing devices in the computer room.
30. The non-transitory computer-readable storage medium of claim 29, wherein the load of the computing device comprises a workload of the computing device and an electrical load used by the computing device.
31. The non-transitory computer-readable storage medium of claim 30, wherein the computing device comprises a server and a power source for the server.
32. The non-transitory computer-readable storage medium of claim 28, wherein training the first neural network and the second neural network based on the input feature parameters comprises using a gradient descent algorithm to implement learning of the input feature parameters for corresponding concepts.
33. The non-transitory computer-readable storage medium of claim 25, wherein predicting a power usage effectiveness (PUE) of the at least one computer room based on an output of the first neural network and an output of the second neural network comprises biasing a loss associated with the first neural network with a constant value.
34. The non-transitory computer-readable storage medium of claim 33, wherein predicting a power usage effectiveness (PUE) of the at least one computer room based on an output of the first neural network and an output of the second neural network comprises minimizing a total loss calculated based on a biased loss associated with the first neural network and an unbiased loss associated with the second neural network.
35. The non-transitory computer-readable storage medium of claim 34, wherein minimizing the total loss calculated based on a loss function by utilizing the first neural network and the second neural network comprises solving a derivative of the loss function equal to zero.
36. The non-transitory computer-readable storage medium of claim 34, wherein minimizing the total loss calculated based on a loss function by utilizing the trained first neural network and the trained second neural network comprises solving for a derivative of the loss function to be less than or equal to a threshold.
Citation Information
Cited By
AI-augmented auditing platform including techniques for automated document processing
US12705676B2
Ai-augmented auditing platform including techniques for providing ai-explainability for processing data through multiple layers
US20230004845A1
