Data center power supply reliability detection method and system

Abstract

The detection method and system of data center power supply reliability collects the operating status information of the power supply equipment and the operating status information of the uninterruptible power supply; establishes a prediction model, including: a first encoder, a second encoder, a decoder and a predictor; the first encoder, the second encoder and the decoder all adopt a layer structure; the first encoder and the second encoder extract the first layer feature information from the first layer feature information and input it into the decoder; the decoder outputs the first layer fusion feature information to the predictor; the predictor predicts the failure probability of the power supply equipment and the failure probability of the uninterruptible power supply according to the first layer fusion feature information. Using the constructed prediction model, the fault prediction of the IDC power supply system can be quickly and automatically realized, which is not only efficient, but also can predict the equipment that may fail in advance, and carry out operation and maintenance in advance to ensure the reliability of the IDC power supply system.

Description

Technical Field

[0001] The present invention belongs to the technical field of power supply and distribution, and in particular, relates to a detection method and system for power supply reliability of a data center. Background Art

[0002] Internet Data Center (IDC), or data center, is a physical carrier that carries massive amounts of data and can be used to store, calculate, and transmit data information. A typical data center mainly includes IT systems (such as servers, network switches, etc.), power supply systems, and cooling systems (such as precision air conditioners, etc.). In the operation of a data center, the stability and reliability of the power supply system are very important.

[0003] At present, the reliability of the data center power supply system can be regularly tested by simply monitoring the hardware and checking it regularly. However, this method is inefficient and requires a lot of manpower and material resources, making it difficult to ensure the timeliness and accuracy of reliability analysis.

[0004] The existing solution of manually analyzing the reliability of the data center power supply system is inefficient.

[0005] With the development of deep learning technology, it is possible to use deep learning models for power supply reliability analysis. The existing power supply fault detection method for data centers obtains the operation detection data and the environment detection data of the power supply system; based on the machine learning algorithm, the operation detection data and the environment detection data are input into the pre-trained fault prediction model to obtain the fault prediction probability of the power supply system, which includes the generator system and the UPS group system; the existing technology simply applies the fault prediction model to the reliability analysis scenario of the data center, and does not consider whether the traditional fault prediction model is compatible with the data center. For example, the generator system is an emergency power supply, and the UPS group system is a temporary power supply when the power supply is switched. The impact of the two on the power supply system of the data center is different. Directly using this information as the input data of the model is not easy to learn the differences in different information. Therefore, the existing fault prediction model has poor risk prediction ability and low accuracy for the data center system. Summary of the invention

[0006] In order to solve the deficiencies in the prior art, the present invention provides a method and system for detecting the reliability of power supply in a data center, which automatically realizes the fault prediction of the IDC power supply system.

[0007] The present invention adopts the following technical solution.

[0008] The present invention proposes a method for detecting the reliability of power supply in a data center. The power supply system of the data center includes power supply equipment and an uninterruptible power supply, including:

[0009] Collecting the operating status information of the power supply equipment and the operating status information of the uninterruptible power supply;

[0010] Establish a prediction model, including: a first encoder, a second encoder, a decoder and a predictor; the first encoder, the second encoder and the decoder all adopt Layer structure;

[0011] The first layer of the first encoder extracts the first-level feature information from the initial feature information of the operating status information of the power supply device. Layer from Extract the first The first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply, and the first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply. Layer from Extract the first Hierarchical feature information; among them, ;

[0012] The decoder In the first encoder layer, The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the The hierarchical concatenation feature information; the decoder In the layer, The level-by-level splicing feature information is upsampled to obtain the The first encoder upsamples feature information. The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the Hierarchical splicing feature information, The up-sampled feature information and The hierarchical splicing feature information is fused to generate the The decoder outputs the first level fusion feature information to the predictor; wherein, ;

[0013] The predictor predicts the failure probability of the power supply equipment and the failure probability of the uninterruptible power supply based on the first-level fused feature information.

[0014] Preferably, when =3, in the first layer of the first encoder, feature extraction is performed on the operating status information of the power supply equipment to generate initial feature information A0 in vector form, and convolution processing is performed on the initial feature information A0 to generate first-level feature information A1; in the second layer of the first encoder, the first-level feature information A1 is downsampled to obtain second-level intermediate feature information A1', and convolution processing is performed on the second-level intermediate feature information A1' to generate second-level feature information A2; the second-level feature information A2 is downsampled to obtain third-level intermediate feature information A2', and convolution processing is performed on the third-level intermediate feature information A2' to generate third-level feature information A3;

[0015] In the first layer of the second encoder, feature extraction is performed on the operating status information of the uninterruptible power supply to generate initial feature information B0 in vector form, and convolution processing is performed on the initial feature information B0 to generate first-level feature information B1; in the second layer of the first encoder, the first-level feature information B1 is downsampled to obtain second-level intermediate feature information B1', and convolution processing is performed on the second-level intermediate feature information B1' to generate second-level feature information B2; the second-level feature information B2 is downsampled to obtain third-level intermediate feature information B2', and convolution processing is performed on the third-level intermediate feature information B2' to generate third-level feature information B3.

[0016] Preferably, the initial feature information A0 and the initial feature information B0 are one-dimensional vectors of the same length.

[0017] Preferably, when =3, in the third layer of the decoder, the third-level feature information A3 of the first encoder and the third-level feature information B3 of the second encoder are concatenated to generate the third-level concatenated feature information C3; in the second layer of the decoder, the third-level concatenated feature information C3 is upsampled to obtain the second-level upsampled feature information D2; the second-level feature information A2 of the first encoder and the second-level feature information B2 of the second encoder are concatenated to generate the second-level concatenated feature information C2, and the second-level upsampled feature information D2 and the first layer are concatenated to generate the second-level concatenated feature information C3. The 2-level spliced ​​feature information C2 is fused to generate the 2-level fused feature information E2; in the 1st layer of the decoder, the 2-level fused feature information E2 is upsampled to obtain the 1st-level upsampled feature information D1, the 1st-level feature information A1 of the first encoder and the 1st-level feature information B1 of the second encoder are spliced ​​to generate the 1st-level spliced ​​feature information C1, and the 1st-level upsampled feature information D1 and the 1st-level spliced ​​feature information C1 are fused to generate the 1st-level fused feature information E1.

[0018] Preferably, the length of the first-level fused feature information E1 is the sum of the lengths of the initial feature information A0 and the initial feature information B0.

[0019] Preferably, the upsampled feature information of the layer and the concatenated feature information of the layer are fused, including:

[0020] 1), using a first convolutional kernel to process the upsampled feature information of the layer to generate the upsampled intermediate feature information of the layer; using a second convolutional kernel to process the concatenated feature information of the layer to generate the concatenated intermediate feature information of the layer;

[0021] wherein, the scales of the first convolutional kernel and the second convolutional kernel are different;

[0022] 2), using an attention feature unit to extract a first attention vector Va from the upsampled intermediate feature information of the layer, using an attention feature unit to extract a second attention vector Vb from the concatenated intermediate feature information of the layer; and using the Softmax function to calculate the weight wa of the first attention vector Va and the weight wb of the second attention vector Vb;

[0023] 3), using the weighted sum of the upsampled intermediate feature information of the layer and the weight wa, and the concatenated intermediate feature information of the layer and the weight wb, as the concatenated fusion feature information of the layer, satisfying the following relational expression:

[0024] E i = wa × D i '+ wb × C i '

[0025] In the formula, E i is the concatenated fusion feature information of the layer, D i ' is the upsampled intermediate feature information of the layer, C i ' is the concatenated intermediate feature information of the layer.

[0026] Preferably, the predictor adopts a one-dimensional iterative dilated convolutional neural network structure.

[0027] Preferably, compare the failure probability Pa of the predicted power supply device and the failure probability Pb of the uninterruptible power supply;

[0028] When Pa > Pb, generate an alarm message;

[0029] When Pa < Pb, the non-faulty UPS supplies power.

[0030] The present invention also proposes a data center power supply reliability detection system, the data center power supply system includes power supply equipment and an uninterruptible power supply, including:

[0031] A collection module, used to collect the operating status information of the power supply equipment and the operating status information of the uninterruptible power supply;

[0032] The prediction model module includes: a first encoder, a second encoder, a decoder and a predictor; the first encoder, the second encoder and the decoder all adopt Layer structure;

[0033] The first layer of the first encoder extracts the first-level feature information from the initial feature information of the operating status information of the power supply device. Layer from Extract the first The first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply, and the first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply. Layer from Extract the first Hierarchical feature information; among them, ;

[0034] The decoder In the first encoder layer, The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the The hierarchical concatenation feature information; the decoder In the layer, The level-by-level splicing feature information is upsampled to obtain the The first encoder upsamples feature information. The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the Hierarchical splicing feature information, The up-sampled feature information and The hierarchical splicing feature information is fused to generate the The decoder outputs the first level fusion feature information to the predictor; wherein, ;

[0035] The predictor predicts the failure probability of the power supply equipment and the failure probability of the uninterruptible power supply based on the first-level fused feature information.

[0036] The present invention is also a terminal, comprising a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method.

[0037] The present invention is also a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0038] The beneficial effects of the present invention are that, compared with the prior art, the present invention at least includes that the present invention uses a deep learning model to automatically analyze the reliability of the data center power supply system, and provides a reliability prediction model that can predict the reliability of the data center power supply system, thereby improving the timeliness and accuracy of the reliability analysis of the power supply system. By using the constructed prediction model, the fault prediction of the IDC power supply system can be quickly and automatically realized, which is not only efficient, but also can predict the equipment that may fail in advance, and carry out operation and maintenance in advance to ensure the reliability of the IDC power supply system. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a logical flow diagram of the method for detecting the reliability of power supply of a data center proposed by the present invention;

[0040] Figure 2 It is a schematic diagram of the processing flow of feature fusion in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0042] The present invention proposes a method for detecting the reliability of power supply in a data center. The power supply system of the data center includes power supply equipment and an uninterruptible power supply (UPS). The method includes:

[0043] Step 1: Collect the operating status information of the power supply equipment and the operating status information of the uninterruptible power supply.

[0044] The power supply equipment of the data center power supply system includes but is not limited to: distribution equipment and generators. Under normal circumstances, the IDC power supply system is powered by the mains. Some IDC power supply systems include distribution equipment based on new energy sources such as solar energy and wind energy. During peak hours of mains power, these new energy distribution equipment can also be used to supply power to reduce power supply costs.

[0045] Generators are usually used as emergency power sources. When the mains or other power distribution equipment is interrupted, the generators are used to provide power instead. UPS mainly provides power when the power source is switched. For example, when the power distribution equipment is switched to the generator (or when different types of power distribution equipment switch power supply), UPS can provide power briefly. UPS failure will cause the risk of data loss when the power is switched, and the failure of power supply equipment such as generators may affect the normal operation of the IDC. Therefore, the power supply equipment and UPS have different effects on the IDC power supply system, so the present invention collects the operating status information of the two separately to be able to analyze the reliability of different equipment.

[0046] like Figure 1 As shown, the operating status information A of the power supply equipment includes but is not limited to: the voltage and current of the power supply equipment; for the generator, it also includes the generator speed, temperature, etc. Similarly, the operating status information B of the UPS includes but is not limited to: the voltage, current, and load level of the UPS, which can be determined based on the data that can actually be collected.

[0047] Step 2, establish a prediction model, including: a first encoder, a second encoder, a decoder and a predictor; the first encoder, the second encoder and the decoder all adopt Layer structure;

[0048] The first layer of the first encoder extracts the first-level feature information from the initial feature information of the operating status information of the power supply device. Layer from Extract the first The first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply, and the first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply. Layer from Extract the first Hierarchical feature information; among them, , is the number of encoder layers.

[0049] Since the characteristics of the two types of running status information are different, the present invention inputs the running status information A and the running status information B into the prediction model in parallel, and the prediction model is provided with two encoders, such as Figure 1 As shown, the first encoder and the second encoder respectively extract feature information of two types of running status information.

[0050] Specifically, the first encoder and the second encoder both use Layer structure; In the embodiment, =3;

[0051] Figure 1In the first layer of the first encoder, feature extraction is performed on the operating status information of the power supply device to generate initial feature information A0 in vector form, and convolution processing is performed on the initial feature information A0 to generate first-level feature information A1; in the second layer of the first encoder, the first-level feature information A1 is downsampled to obtain second-level intermediate feature information A1', and convolution processing is performed on the second-level intermediate feature information A1' to generate second-level feature information A2; the second-level feature information A2 is downsampled to obtain third-level intermediate feature information A2', and convolution processing is performed on the third-level intermediate feature information A2' to generate third-level feature information A3;

[0052] Figure 1 In the first layer of the second encoder, feature extraction is performed on the operating status information of the uninterruptible power supply to generate initial feature information B0 in vector form, and convolution processing is performed on the initial feature information B0 to generate first-level feature information B1; in the second layer of the first encoder, the first-level feature information B1 is downsampled to obtain second-level intermediate feature information B1', and convolution processing is performed on the second-level intermediate feature information B1' to generate second-level feature information B2; the second-level feature information B2 is downsampled to obtain third-level intermediate feature information B2', and convolution processing is performed on the third-level intermediate feature information B2' to generate third-level feature information B3;

[0053] The initial feature information A0 and the initial feature information B0 are one-dimensional vectors of the same length.

[0054] In the method proposed by the present invention, two separate encoders are used to perform feature extraction respectively, so that the features of each operating status information can be extracted without affecting each other, and finally the failure probability of each device can be predicted.

[0055] Step 3, the decoder In the first encoder layer, The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the The hierarchical concatenation feature information; the decoder In the layer, The level-by-level splicing feature information is upsampled to obtain the The first encoder upsamples feature information. The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the Hierarchical splicing feature information, The up-sampled feature information and The hierarchical splicing feature information is fused to generate the The decoder outputs the first level fusion feature information to the predictor; wherein, , is the number of decoder layers.

[0056] like Figure 1 As shown, the prediction model also includes a decoder, and the decoder can be shared by the first encoder and the second encoder. The decoder adopts Layer structure;

[0057] Figure 1 In the third layer of the decoder, the third-level feature information A3 of the first encoder and the third-level feature information B3 of the second encoder are concatenated to generate the third-level concatenated feature information C3; in the second layer of the decoder, the third-level concatenated feature information C3 is upsampled to obtain the second-level upsampled feature information D2; the second-level feature information A2 of the first encoder and the second-level feature information B2 of the second encoder are concatenated to generate the second-level concatenated feature information C2, and the second-level upsampled feature information D2 and the second-level upsampled feature information D3 are upsampled to obtain the second-level upsampled feature information D2. The hierarchical splicing feature information C2 is fused to generate the second-level fused feature information E2; in the first layer of the decoder, the second-level fused feature information E2 is upsampled to obtain the first-level upsampled feature information D1, the first-level feature information A1 of the first encoder and the first-level feature information B1 of the second encoder are spliced ​​to generate the first-level splicing feature information C1, and the first-level upsampled feature information D1 and the first-level splicing feature information C1 are fused to generate the first-level fused feature information E1.

[0058] The decoder outputs the first-level fused feature information E1. The fused feature information E1 is a one-dimensional feature vector, and its length is the sum of the lengths of the initial feature information A0 and the initial feature information B0. For example, if the lengths of the initial feature information A0 and the initial feature information B0 are both L, the length of the fused feature information E1 is 2L.

[0059] The present invention uses two encoders. If the characteristic information is not fused, two prediction models are used to predict the failure probabilities of the power supply equipment and the uninterruptible power supply according to different operating status information A and B respectively. By fusing the characteristic information, not only the failure probabilities of the power supply equipment and the uninterruptible power supply can be predicted, but also the failure probability of the power supply system can be predicted.

[0060] In the present invention, in the The up-sampled feature information and When the hierarchical splicing feature information is fused, the attention weight of each feature information is determined to achieve the fusion of the two.

[0061] Specifically, no. The up-sampled feature information and The hierarchical splicing feature information is fused, including:

[0062] 1) Use the first convolution kernel to check the The upsampled feature information is processed to generate the The second convolution kernel is used to sample the intermediate feature information of the first layer. The hierarchical splicing feature information is processed to generate the Hierarchical splicing of intermediate feature information;

[0063] Among them, the scales of the first convolution kernel and the second convolution kernel are different;

[0064] 2) Use the attention feature unit from the The first attention vector Va is extracted from the intermediate feature information of the upsampled layer, and the attention feature unit is used to extract the first attention vector Va from the Extract the second attention vector Vb from the hierarchical splicing intermediate feature information; and use the Softmax function to calculate the weight wa of the first attention vector Va and the weight wb of the second attention vector Vb;

[0065] 3) Use the The intermediate feature information and weights wa are sampled on the layer. The weighted sum of the intermediate feature information and weight wb of the level splicing is used as the first The hierarchical fusion feature information satisfies the following relationship:

[0066] E i =wa×D i '+ wb×C i '

[0067] In the formula, E i For the Hierarchical fusion feature information, D i 'For the Layer upsampling intermediate feature information, C i 'For the Hierarchical splicing of intermediate feature information.

[0068] like Figure 2 As shown, the second-level upsampling feature information D2 and the second-level splicing feature information C2 are fused, including:

[0069] The 1×3 convolution kernel is used to extract the second-level upsampled feature information D2 to generate the second-level upsampled intermediate feature information D2';

[0070] A 1×5 convolution kernel is used to extract features from the second-level splicing feature information C2 to generate the second-level splicing intermediate feature information C2'.

[0071] Convolution kernels of different scales are used here to ensure that the final feature information E2 can contain features of different spatial scales, thereby enriching the feature space.

[0072] The transformer's attention mechanism is used for feature extraction. The attention mechanism is used to construct an attention feature unit. The attention feature unit is used to extract the attention vectors Va and Vb corresponding to the feature information D2' and feature information C2', respectively. Then, the attention weights wa and wb are calculated by the Softmax function. After that, the feature information D2' and feature information C2' are weighted using the attention weights wa and wb ( Figure 2 The ⊕ in denotes weighted processing), and finally generates the second-level fused feature information E2, which can be expressed as: E2=wa×D2'+wb×C2'.

[0073] In the method proposed in the present invention, feature fusion is achieved by combining the attention weight of feature information, which can generate fusion features more accurately, which is conducive to more accurate fault prediction in the subsequent process.

[0074] In the method proposed in the present invention, by utilizing processing processes such as feature splicing and feature fusion, a unified decoder can be shared, thereby reducing the complexity of the model, facilitating model training and reasoning, and improving the training efficiency and prediction efficiency of the model.

[0075] Step 4: The predictor predicts the failure probability of the power supply equipment and the failure probability of the uninterruptible power supply based on the fused feature information of the first level.

[0076] The present invention uses the fused feature information obtained by fusing the up-sampled feature information and the spliced ​​feature information as the output of the decoder, thereby achieving the fitting of the mutual influence between the operating states of the power supply equipment and the uninterruptible power supply in the power supply system, improving the prediction accuracy of the power supply equipment failure probability and the uninterruptible power supply failure probability, and making the reliability analysis result of the power supply system more accurate and reliable.

[0077] like Figure 1 As shown, the prediction model also includes a predictor, and the predictor can predict a one-dimensional feature vector to predict the corresponding fault probability. In addition, the first-level fused feature information E1 is generated by splicing the feature information of the power supply device and UPS (for example, the feature information A3 and feature information B3 of the third level, etc.). Therefore, a part of the first-level fused feature information E1 corresponds to the power supply device and feature information, and the other part corresponds to the feature information of the UPS. By using the predictor to predict the first-level fused feature information E1, the predicted fault probabilities of both the power supply device and the UPS can be predicted. Figure 1As shown, the predicted failure probability Pa of the power supply device and the predicted failure probability Pb of the UPS can be obtained.

[0078] In the present invention, the fusion feature information E1 in the form of a one-dimensional feature vector can be regarded as a sequence, so a network model capable of predicting sequences is used to construct the predictor. In the present invention, a one-dimensional IDCNN (Iterated Dilated Convolutional Neural Network) is used to construct the predictor.

[0079] Compared with recurrent neural networks such as Bi-LSTM, IDCNN can utilize the parallel processing ability of the GPU to achieve parallel computing, thereby reducing the training time and inference time and improving the prediction efficiency.

[0080] Furthermore, compare the predicted failure probability Pa of the power supply device and the failure probability Pb of the uninterruptible power supply; when Pa > Pb, generate an alarm message; when Pa < Pb, the non-faulty UPS supplies power.

[0081] In the embodiment, after determining the predicted failure probabilities of the power supply device and the UPS, reliability analysis can be carried out. For example, if the predicted failure probability Pa of the power supply device is high, an alarm message can be generated to facilitate troubleshooting by the operation and maintenance personnel; if the predicted failure probability Pb of the UPS is high, it can be switched to use other healthy UPSs to ensure stable power supply of the UPS during power supply switching.

[0082] The present invention also proposes a detection system for the power supply reliability of a data center. The data center power supply system includes a power supply device and an uninterruptible power supply, and includes:

[0083] An acquisition module for acquiring the operation status information of the power supply device and the operation status information of the uninterruptible power supply;

[0084] A prediction model module, including: a first encoder, a second encoder, a decoder, and a predictor; the first encoder, the second encoder, and the decoder all adopt layer structures;

[0085] The first layer of the first encoder extracts the first-level feature information from the initial feature information of the operation status information of the power supply device, and the layer of the first encoder extracts the -level feature information from the -level feature information; the first layer of the second encoder extracts the first-level feature information from the initial feature information of the operation status information of the uninterruptible power supply, and the layer of the second encoder extracts the -level feature information from the -level feature information; where ;

[0086] The decoder In the first encoder layer, The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the The hierarchical concatenation feature information; the decoder In the layer, The level-by-level splicing feature information is upsampled to obtain the The first encoder upsamples feature information. The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the Hierarchical splicing feature information, The up-sampled feature information and The hierarchical splicing feature information is fused to generate the The decoder outputs the first level fusion feature information to the predictor; wherein, ;

[0087] The predictor predicts the failure probability of the power supply equipment and the failure probability of the uninterruptible power supply based on the first-level fused feature information.

[0088] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0089] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0090] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0091] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for detecting the reliability of power supply in a data center, wherein the power supply system of the data center includes power supply equipment and an uninterruptible power supply, characterized in that: Including: Collect the operation status information of the power supply equipment and the operation status information of the uninterruptible power supply; Establish a prediction model, including: a first encoder, a second encoder, a decoder and a predictor; the first encoder, the second encoder and the decoder all adopt Layer structure; The first layer of the first encoder extracts the first-level feature information from the initial feature information of the operating status information of the power supply device. Layer from Extract the first The first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply, and the first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply. Layer from Extract the first Hierarchical feature information; among them, ; The decoder In the first encoder layer, The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the The hierarchical concatenation feature information; the decoder In the layer, The level-by-level splicing feature information is upsampled to obtain the The first encoder upsamples feature information. The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the Hierarchical splicing feature information, The up-sampled feature information and The hierarchical splicing feature information is fused to generate the Layer-by-layer fusion feature information includes: 1) using the first convolution kernel to The upsampled feature information is processed to generate the The second convolution kernel is used to sample the intermediate feature information of the first layer. The hierarchical splicing feature information is processed to generate the The first convolution kernel and the second convolution kernel have different scales; 2) Use the attention feature unit to obtain the intermediate feature information from the first The first attention vector is extracted from the intermediate feature information of the upsampled layer, and the attention feature unit is used to extract the first attention vector from the The second attention vector is extracted from the hierarchical splicing intermediate feature information; and the Softmax function is used to calculate the weight wa of the first attention vector and the weight wb of the second attention vector; 3) The hierarchical fusion feature information satisfies the following relationship: E j =wa×D j '+ wb×C j ', where E j For the Hierarchical fusion feature information, D j 'For the Layer upsampling intermediate feature information, C j 'For the The intermediate feature information is concatenated at different levels; the decoder outputs the first level fusion feature information to the predictor; wherein, ; The predictor predicts the failure probability of the power supply equipment and the failure probability of the uninterruptible power supply according to the first-level fusion feature information. The predictor adopts a one-dimensional iterative dilated convolutional neural network structure.

2. The detection method for the power supply reliability of a data center according to claim 1, characterized in that when =3, in the first layer of the first encoder, feature extraction is performed on the operating status information of the power supply device to generate initial feature information A0 in vector form, and convolution processing is performed on the initial feature information A0 to generate first-level feature information A1; in the second layer of the first encoder, the first-level feature information A1 is downsampled to obtain second-level intermediate feature information A1', and convolution processing is performed on the second-level intermediate feature information A1' to generate second-level feature information A2; Downsample the second-level feature information A2 to obtain the third-level intermediate feature information A2', and perform convolutional processing on the third-level intermediate feature information A2' to generate the third-level feature information A3; In the first layer of the second encoder, extract features from the operation status information of the uninterruptible power supply to generate initial feature information B0 in vector form, and perform convolutional processing on the initial feature information B0 to generate the first-level feature information B1; In the second layer of the first encoder, downsample the first-level feature information B1 to obtain the second-level intermediate feature information B1', and perform convolutional processing on the second-level intermediate feature information B1' to generate the second-level feature information B2; Downsample the second-level feature information B2 to obtain the third-level intermediate feature information B2', and perform convolutional processing on the third-level intermediate feature information B2' to generate the third-level feature information B3.

3. The method for detecting power supply reliability of a data center according to claim 2, Characterized in that The initial feature information A0 and the initial feature information B0 are one-dimensional vectors of the same length.

4. The detection method for the power supply reliability of a data center according to claim 2, characterized in that when =3, in the third layer of the decoder, the third-level feature information A3 of the first encoder and the third-level feature information B3 of the second encoder are concatenated to generate the third-level concatenated feature information C3; In the second layer of the decoder, upsample the third-level concatenated feature information C3 to obtain the second-level upsampled feature information D2; Concatenate the second-level feature information A2 of the first encoder and the second-level feature information B2 of the second encoder to generate the second-level concatenated feature information C2, and perform fusion processing on the second-level upsampled feature information D2 and the second-level concatenated feature information C2 to generate the second-level fusion feature information E2; In the first layer of the decoder, upsample the second-level fusion feature information E2 to obtain the first-level upsampled feature information D1, concatenate the first-level feature information A1 of the first encoder and the first-level feature information B1 of the second encoder to generate the first-level concatenated feature information C1, and perform fusion processing on the first-level upsampled feature information D1 and the first-level concatenated feature information C1 to generate the first-level fusion feature information E1.

5. The detection method for the power supply reliability of a data center according to claim 4, characterized in that The length of the first-level fusion feature information E1 is the sum of the lengths of the initial feature information A0 and the initial feature information B0.

6. The detection method for the power supply reliability of a data center according to claim 1, characterized in that Compare the predicted failure probability Pa of the power supply equipment and the failure probability Pb of the uninterruptible power supply; When Pa > Pb, generate an alarm message; When Pa < Pb, the non-faulty UPS supplies power.

7. A data center power supply reliability detection system, the data center power supply system includes power supply equipment and uninterruptible power supply, characterized in that: Including: A collection module for collecting the operation status information of the power supply equipment and the operation status information of the uninterruptible power supply; The prediction model module includes: a first encoder, a second encoder, a decoder and a predictor; the first encoder, the second encoder and the decoder all adopt Layer structure; The first layer of the first encoder extracts the first-level feature information from the initial feature information of the operating status information of the power supply device. Layer from Extract the first The first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply, and the first layer of the second encoder extracts the first layer feature information from the initial feature information of the operating status information of the uninterruptible power supply. Layer from Extract the first Hierarchical feature information; among them, ; The decoder In the first encoder layer, The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the The hierarchical concatenation feature information; the decoder In the layer, The level-by-level splicing feature information is upsampled to obtain the The first encoder upsamples feature information. The hierarchical feature information and the second encoder The hierarchical feature information is concatenated to generate the Hierarchical splicing feature information, The up-sampled feature information and The hierarchical splicing feature information is fused to generate the The hierarchical fusion feature information includes: 1) using the first convolution kernel to The upsampled feature information is processed to generate the The second convolution kernel is used to sample the intermediate feature information of the first layer. The hierarchical splicing feature information is processed to generate the The first convolution kernel and the second convolution kernel have different scales; 2) Use the attention feature unit to obtain the intermediate feature information from the first The first attention vector is extracted from the intermediate feature information of the upsampled layer, and the attention feature unit is used to extract the first attention vector from the The second attention vector is extracted from the hierarchical splicing intermediate feature information; and the Softmax function is used to calculate the weight wa of the first attention vector and the weight wb of the second attention vector; 3) The hierarchical fusion feature information satisfies the following relationship: E j =wa×D j '+ wb×C j ', where E j For the Hierarchical fusion feature information, D j 'For the Layer upsampling intermediate feature information, C j 'For the The intermediate feature information is concatenated at different levels; the decoder outputs the first level fusion feature information to the predictor; wherein, ; The predictor predicts the failure probability of the power supply equipment and the failure probability of the uninterruptible power supply based on the fused feature information of the first level. The predictor adopts a one-dimensional iterative expansion convolutional neural network structure.

8. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that ； When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images