CN116340006A - Computing power resource idle prediction method based on deep learning and storage medium - Google Patents
Computing power resource idle prediction method based on deep learning and storage medium Download PDFInfo
- Publication number
- CN116340006A CN116340006A CN202310603689.6A CN202310603689A CN116340006A CN 116340006 A CN116340006 A CN 116340006A CN 202310603689 A CN202310603689 A CN 202310603689A CN 116340006 A CN116340006 A CN 116340006A
- Authority
- CN
- China
- Prior art keywords
- data
- power resource
- deep learning
- idle
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000013135 deep learning Methods 0.000 title claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000012545 processing Methods 0.000 claims abstract description 23
- 230000006399 behavior Effects 0.000 claims abstract description 20
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 230000007246 mechanism Effects 0.000 claims abstract description 7
- 238000002372 labelling Methods 0.000 claims abstract description 3
- 230000006870 function Effects 0.000 claims description 24
- 238000013528 artificial neural network Methods 0.000 claims description 22
- 238000004422 calculation algorithm Methods 0.000 claims description 22
- 238000011176 pooling Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 7
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 6
- 238000013136 deep learning model Methods 0.000 abstract description 3
- 238000004364 calculation method Methods 0.000 description 18
- 230000004913 activation Effects 0.000 description 13
- 238000010606 normalization Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 8
- 230000015654 memory Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000005284 excitation Effects 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 238000001125 extrusion Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000010355 oscillation Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
- 230000003313 weakening effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a computational power resource idle prediction method and a storage medium based on deep learning, wherein the method comprises the following steps: collecting business behavior data, classifying the data, and labeling the classified data to obtain a training set; preprocessing business behavior data, wherein the preprocessing comprises data expansion and unbalance processing; constructing a feature extraction model of computational power resource idle prediction, inputting a training set into the model for training, and extracting features of business behavior data by using the trained model; and predicting the idle computing power resources of the business behavior data through the idle computing power resource prediction model. The invention has stronger robustness and generalization capability under the conditions of insufficient service data, low universality of application scenes and the like, and simultaneously, the deep learning model provided by the invention can effectively realize the self-adaptive refinement purpose of the feature extraction stage by combining the attention mechanism.
Description
Technical Field
The invention relates to the technical field of deep learning, in particular to a computational power resource idle prediction method based on deep learning and a computer readable storage medium.
Background
The computing infrastructure has various forms, wide distribution and complex attribution, the computing power measurement has certain difficulty, the island computing power functional units are fixed, the computing power is limited, the single-point computing power supply can not meet the requirement of service diversity, and the development of the emerging industry is limited. The calculation power is used as new kinetic energy for social development, so that the calculation power is flexibly supplied as basic energy sources such as water, electricity and the like according to the need and the quantity, and the calculation power will become the development trend of the calculation power in the future. The core feature of the computing power network is that the computing power resource and the network resource are comprehensively taken over by computing power, so that the network can sense the computing power requirement of a user and the computing power state of the network in real time. After analysis, the power network can dispatch power resources of different positions and different types to serve users. Therefore, the reasonable planning is carried out on the use of the calculation force, and the reasonable distribution of the calculation force resources is realized by accurately predicting the idle condition of the calculation force resources, so that the method has very important significance.
Chinese patent publication No. CN114896070A proposes a GPU resource allocation method for deep learning tasks, which predicts the resource demand of the deep learning tasks and reasonably allocates GPU resources in a container cloud cluster according to the predicted resource demand, thereby realizing GPU resource sharing in the container cloud cluster and improving the GPU utilization rate in the container cloud cluster.
Chinese patent publication No. CN114035945A proposes a method for allocating computing resources, which comprises the following steps: under the condition that the computational power resources of target equipment in the local area network are insufficient, idle computational power resources which can be provided by support equipment in the local area network are obtained, and the service quality coefficient of the support equipment at least comprises a reliability coefficient which is used for indicating the reliability degree of the support equipment for providing computational power resource support service by utilizing the idle computational power resources in the history time; according to idle computing power resources which can be provided by support equipment in a local area network, a service quality coefficient and target computing power resources required by target equipment, at least one first support equipment for providing computing power support service for the target equipment and first idle computing power resources which are required to be provided by each first support equipment are determined, wherein the first idle computing power resources are used for processing target computing tasks corresponding to the target computing power resources. The method can avoid waste of idle computing power resources and improve the reliability of task processing.
The method can reasonably distribute the computational power resources, but can only feed back the computational power idle state in real time through a system, then manually judge whether the computational power is idle or not, and can not predict the computational power idle time according to the real-time state of the service, so that the computational power waste is easily caused in a period of time. Therefore, further improvement is required in this respect.
Disclosure of Invention
In order to solve the technical problems, the invention provides a computing power resource idle prediction method based on deep learning and a computer readable storage medium.
In order to achieve the above object, the present invention provides a method for predicting the idle computing power resource based on deep learning, comprising the following steps:
collecting business behavior data, classifying the data, and labeling the classified data to obtain a training set; preprocessing business behavior data, wherein the preprocessing comprises data expansion and unbalance processing; constructing a feature extraction model of computational power resource idle prediction, inputting a training set into the model for training, and extracting features of business behavior data by using the trained model; and predicting the idle computing power resources of the business behavior data through the idle computing power resource prediction model.
Optionally, the preprocessing further comprises redundant sample removal of the business behavior data.
Optionally, the unbalanced processing adopts a random oversampling algorithm to perform unbalanced processing on the idle prediction data of the computational power resources, samples of a minority class are sampled in a random manner, and the sampled samples are combined with the initial samples of the minority class, so that the number of the samples of the minority class is the same as that of the samples of a majority class.
Optionally, the data set processed by the random oversampling algorithm is as follows:
wherein,,
and->
Represented by majority sample and minority sample, respectively, ">
The data set obtained by carrying out data unbalance processing on the initial data set S is represented as E, and the data set obtained by carrying out random sampling on a few types of samples in the data set is represented as E.
Optionally, the feature extraction model of the computational resource free prediction comprises a residual neural network comprising a convolutional layer, a pooling layer, a residual block, dropout, and a softmax classifier.
Optionally, the computational resource idle prediction model is an extreme learning machine.
Optionally, the network structure of the extreme learning machine includes an input layer, an hidden layer, and an output layer.
Optionally, the method further comprises: optimizing a residual neural network through an evolution algorithm, taking the inverse of the square error as an fitness function, and carrying out a standard for measuring the individual fitness in the population, wherein the formula is as follows:
wherein E is an error function, P is an integral output, W is a weight vector, and X is an input vectorThe quantity F is the fitness, j is the number of selections,
is a theoretical output.
Optionally, the residual neural network utilizes an attention mechanism to improve the threshold value, and the network automatically generates a corresponding threshold value according to the data.
To achieve the above object, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform the deep learning-based power resource idling prediction method of any one of the above.
According to the technical scheme provided by the invention, the data marking is carried out by collecting the business behavior data, the data expansion and unbalance processing are carried out on the data, the feature extraction and modeling are further carried out on the data, the idle time of calculation force is predicted, and the reasonable distribution of calculation force resources is realized. Compared with the prior art, the method provided by the invention has stronger robustness and generalization capability under the conditions of insufficient service data, low application scene universality and the like, and meanwhile, the deep learning model provided by the invention can effectively realize the self-adaptive refinement purpose of the feature extraction stage by combining the attention mechanism.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions and advantages of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of an inventive computing power resource idle prediction method based on deep learning.
FIG. 2 is a flow chart of a method for unbalanced processing of computational power resource idle prediction data according to the present invention.
Fig. 3 is a schematic structural diagram of a residual neural network according to the present invention.
Fig. 4 is a flow chart of the adaptive threshold generation process of the present invention.
Detailed Description
In order to further describe the technical means and effects adopted by the present invention to achieve the preset purpose, the following detailed description refers to specific embodiments, structures, features and effects of a high-performance data caching method according to the present invention with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" means that the embodiments are not necessarily the same. Furthermore, the particular features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The invention provides a computational power resource idle prediction method based on deep learning, referring to fig. 1, the method comprises the following steps:
and S10, collecting business behavior data, classifying the data, and marking the classified data to obtain a training set.
The data adopted by the invention for carrying out idle prediction on the computing power resources is business behavior data, and comprises a series of indexes related to business. In the invention, an idle prediction feature library of computing power resources covering network flow features, data packet features, equipment operation features and the like is acquired, and data are marked as training samples for model training.
And step S20, preprocessing the business behavior data, wherein the preprocessing comprises data expansion and unbalance processing.
In one embodiment of the present invention, the preprocessing further includes redundant sample removal for the business activity data.
And redundant samples of the data are removed, so that the data quality is improved. Meanwhile, sample expansion and unbalance processing are carried out on the data. Referring to fig. 2, the present invention proposes to use a random oversampling algorithm to perform unbalanced processing on idle prediction data samples of computing resources, where the principle of the random oversampling algorithm is to sample samples of minority classes in a random manner, and combine the sampled samples with initial samples of minority classes to obtain a balanced prediction data set of computing resources, so that the number of samples of minority classes is increased to the same level as that of samples of majority classes.
Assuming that the initial data set is S, the data set obtained by randomly sampling a few types of samples in the data set is E, and the data set obtained by processing through a random oversampling algorithm is expressed as follows:
wherein the method comprises the steps of
And->
Represented by majority sample and minority sample, respectively, ">
The resulting data set that has been subjected to data imbalance processing is shown.
Furthermore, the data set E is sampled randomly from a minority class of samples
And (3) sampling. By sampling from minority classes->
Data set E obtained by middle sampling and majority sample +.>
And minority class sample->
Phase combining to obtain data unbalanceProcessed data set->
In the method, the balance of samples among different categories in the idle prediction dataset of the computing power resource is realized.
And step S30, constructing a feature extraction model of computational power resource idle prediction, inputting a training set into the model for training, and extracting features of business behavior data by using the trained model.
In an actual application scene of idle prediction of computational resources, supervised learning is needed on a large number of manually marked data sets, but the method is limited by the problems of marking cost and marking quality of marked data, and the scale of the actually available marked training data is limited, so how to fully learn the feature expression with good generalization capability from the marked data with limited scale is an important problem in idle prediction tasks of computational resources.
The invention provides a feature extraction model using a residual neural network based on an evolution algorithm and an attention mechanism as an idle prediction of computational resources.
Referring to fig. 3, the residual neural network structure includes a convolution layer, a pooling layer, a residual block, a Dropout, and a softmax classifier, and the specific structure of each layer is as follows:
(1) Convolutional layer
The convolution layer is used for extracting the characteristics of the object, and the depth characteristic extraction of the data is realized by setting the super parameters such as the number of convolution layers, the size of a convolution window and the like. A further feature of convolutional neural networks is the sharing of parameter thresholds, the magnitude of which determines the quality of the convolution. The threshold parameters are deleted correspondingly every time they are used, and then the next convolution is carried out according to the result generated as required. However, the parameter value is always unchanged in the same operation at different stages, and the convolution kernel is the number set of weight parameters. In the process of selecting the convolution kernel, the convolution kernel freely moves in fixed data and performs convolution processing on the convolution kernel and the part of the corresponding area, and then the rest part is filled. The convolution operation between the input feature and the convolution kernel (by adding a bias term) can be representedThe method comprises the following steps:
；
wherein:
is the ith channel of the input feature; />
Is the j-th channel of the output feature; k is a convolution kernel; b is a bias term; />
Is the set of jth channels used to calculate the output characteristics.
(2) Batch normalization
Bulk normalization (batch normalization, BN) is one normalization method proposed for the internal covariance offset problem. In the proposed classification algorithm, batch normalization is introduced, so that the convergence rate of the model can be increased, and more importantly, the problem of gradient dispersion in a deep network is relieved to a certain extent, so that the deep network model is easier and more stable to train.
The batch normalization selects a small batch in the deep learning training process, then calculates the mean value and variance of the small batch data, and the input of each layer of neural network is kept in the same distribution in the training process after the processing.
Unlike the general normalization method, batch normalization is an operation embedded within the deep neural network from layer to layer. The calculation process of BN is expressed as:
wherein:
and->
Respectively representing input and output characteristics of an ith observed value in batch processing; />
The number of samples of each batch in the classification task; gamma and beta are two trainable parameters, and more proper characteristic distribution can be adaptively learned; epsilon is a constant close to zero; />
For the first normalization influence factor, +.>
For the second normalization influence factor, +.>
Is a third standardized influencing factor.
(3) Activation function design
ReLU is used as the most common activation function, solves the problems of S-type local gradient explosion and gradient disappearance, and accelerates the convergence of the neural network.
The algorithm for ReLU is as follows:
wherein: x and y are the input and output, respectively, of the ReLU activation function.
The ReLU algorithm discards such vibration signals when there is oscillation in the input signal, impairing the classification predictive ability of the model. In this regard, the inventive method employs lrehu as an activation function to solve the problem that arises when there is oscillation in the input signal. The specific algorithm is as follows:
wherein: x and y are the input and output of the lrehu activation function, respectively; a is obtained according to practical experience, and a large number of experiments prove that the effect is optimal when the value of a is in the range of 0-0.5.
(4) Basic principle of residual error module
The residual modules (residual building block, RBB) are the core of Resnet, and RBB is realized by skipping the convolution layer blocks by using the shortcut connection, so that gradient explosion and disappearance are avoided, a deeper neural network structure is constructed, and the final performance of fault diagnosis is improved.
Convolution layer block
The execution path of (1) is "input>
The BN layer, the activation function ReLU, the convolution layer, the BN layer, the activation function ReLU, the convolution layer and the output->
". When the input dimension and the output dimension of the convolution layer block are the same, the output value of the shortcut connection is the input value +.>
The final output result of the residual block is shown as follows:
when the dimensions of the input and output are different, the shortcut connection requires the use of a convolution kernel of size
Matching the dimensions of the output result to obtain the output of the shortcut +.>
The final output result is shown in the following formula:
(5) Extrusion and excitation network structure
The invention adopts a squeezing and excitation network Structure (SENET), which can automatically obtain the importance of each channel and strengthen the connection among the channels, thereby achieving the purpose of improving the model performance. The core of the structure is two major operations of extrusion (Squeeze) and Excitation (specification).
The Squeeze operation is a global pooling of input features, compressing each feature into a real number with a global receptive field. The specific algorithm is shown as follows:
wherein, in the formula:
representative input is size +.>
Is the ith feature of (2).
The accounting operation mainly consists of 2 full connection layers and 2 activation functions, and can help capture channel correlation and generate weight of a corresponding channel. The algorithm is shown as follows:
wherein:
representing a first full connection layer calculation; />
Representing a second full connection layer calculation; />
Representing the output value after the squeze operation; />
Representing an activation function ReLU; sigma is a Sigmoid function specific algorithm as follows:
wherein: x represents the output value after 2 full-join calculations.
(6) Cross entropy loss function
The cross entropy loss function Softmax, which is typically an activation function of the final output layer, fixes the output value of the neural network between (0, 1) to represent the probability of different events occurring, the algorithm is shown as follows:
wherein:
representing the category to which the classification task relates, +.>
A j-th output representing the upper layer; />
Representing the jth predictor of the neural network.
(7) Pooling layer global average Pooling
Global averaging pooling (global average pooling, GAP) is an operation of averaging features, which can greatly reduce parameters during training of the neural network, accelerate the calculation speed of the neural network, and is a common deep learning pooling operation.
(8) Residual block design
In the residual block, a numerical threshold is typically set to remove redundant noise, and the threshold is applied in many noise reduction, and the formula is as follows:
where x represents the input data and τ is the threshold, i.e., the features within the interval of the threshold [ - τ, τ ] are set to 0, letting features farther from 0 tend to shrink toward 0.
Based on the depth residual error network, the threshold is improved by using the attention mechanism, so that the network automatically generates a corresponding threshold according to the data to eliminate noise, and each group of data can carry out unique characteristic channel weighting adjustment according to different importance degrees of the samples. In the generation process of the adaptive threshold, referring to fig. 4, data is subjected to global convolution processing, then is subjected to batch normalization and activation layer, and output is mapped to [0,1 ] by using Sigmoid function]Within this, the mapping scaling factor is noted α, and the final threshold can be expressed as
Different samples correspond to different thresholds. And adding the self-adaptive threshold block into a residual error network to be improved into a residual error shrinkage module, thereby achieving the purpose of eliminating or weakening noise.
Because the full-connection layer spreads the convolution layer and then classifies each feature map, the parameter calculation amount of the full-connection layer is huge, and often occupies most of the total parameter calculation amount of the network, so that the training speed of the neural network is very slow. In order to solve the problem of low training speed, global convolution is introduced into a network, wherein the global convolution is to directly carry out convolution processing on the feature map of each channel, namely, one feature map outputs one value, and then the result is input into a classifier for classification. In the identification task, the global convolution can generate a feature map for each particular class in the final convolution layer.
The GAP is added to the original full-connection layer, the parameters required to be calculated are greatly reduced, the calculation speed of the network is greatly improved, and the GAP does not need a great amount of training optimization parameters like the full-connection layer, so that the problem of over-fitting is avoided. GAP summarizes spatial information and is therefore more robust to spatial transformations of the input.
(9) Evolutionary algorithm optimized residual neural network
Based on the residual neural network, a mode of optimizing by using an evolution algorithm is provided to replace an original back propagation optimizing mode. In the evolution algorithm adopted by the invention, the inverse of the square error is adopted as the fitness function, and the standard for measuring the individual fitness in the population is measured according to the following formula:
wherein E is an error function,
for the whole output +.>
Is weight vector, x is input vector, F is fitness, j is number of selections, ++>
Is a theoretical output.
The traditional evolution algorithm often adopts a mode of 'roulette' in the working process, the probability of selecting individuals in the population is random, the optimal individuals are most likely to be lost in the selection mode, and larger errors can be generated in the actual operation process, so that the invention improves the selection operator, firstly, the individuals in the population are rearranged by using a sorting method, and the probability of selecting the individuals after the rearrangement is as follows:
wherein a is the number of populations in the evolution algorithm,
probability that the best individual is likely to be selected, s is the probability that +.>
Normalized values, b, are the positions of the nth individuals after the population has been rearranged.
And S40, predicting the idle computing power resources of the business behavior data through the idle computing power resource prediction model.
The present invention proposes to use an extreme learning machine as a predictive model. Unlike conventional single hidden layer feedforward neural networks (SLFNs), the extreme learning machine randomly assigns input weights and hidden layer biases without the need to adjust parameters as they are back-propagated to errors in the neural network. The output weight of the network model of the extreme learning machine is directly determined by solving the linear model, so that the training stage of the extreme learning machine is completed only by one iteration, and the training speed is extremely high. The network structure of the extreme learning machine comprises: input layer, hidden layer and output layer, the connection between the input layer and hidden layer is through input weight
The connection between the hidden layer and the output layer is established by the output weight beta.
Assuming that the given input data is a training dataset consisting of N arbitrary different samples
Wherein for each sample->
Comprising n features, tag->
Including m output categories. The output of a standard SLFN containing L neurons can be expressed as
Wherein the method comprises the steps of
Is the input weight of the ith hidden layer neuron, < +.>
Is the bias of the ith hidden layer neuron. />
Is the output weight of the ith neuron, the output value of the network is +.>
，/>
Is an activation function. In extreme learning machines, sigmoid functions are often regarded as activation functions
Loss function of standard SLFN
In network parameters->
With perfect adjustability, the error can be infinitely close to zero. In this case, the equation (1.10) is converted into
Thus, the N formulas can be combined together to form a matrix
Wherein,,
the matrix H is the output of the hidden layer and T is the real class label. Output weight
By solving least squares problem calculations
Wherein,,
the hidden layer outputs the MP generalized inverse of H.
The method provided by the invention has stronger robustness and generalization capability under the conditions of insufficient service data, low application scene universality and the like, and meanwhile, the deep learning model provided by the invention is combined with a attention mechanism, so that the self-adaptive refinement purpose of the feature extraction stage can be effectively realized.
Embodiments of the present invention provide a computer-readable storage medium storing computer-executable instructions for execution by one or more processors, e.g., to perform the method steps S10 through S40 of fig. 1 described above.
In particular, the computer-readable storage medium can include read-only memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM), which acts as external cache memory. By way of illustration and not limitation, RAM may be available in many forms such as Synchronous RAM (SRAM), dynamic RAM, (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The disclosed memory components or memories of the operating environment described in embodiments of the present invention are intended to comprise one or more of these and/or any other suitable types of memory.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments.
Claims (10)
1. A method for computing power resource idle prediction based on deep learning, the method comprising:
collecting business behavior data, classifying the data, and labeling the classified data to obtain a training set;
preprocessing business behavior data, wherein the preprocessing comprises data expansion and unbalance processing;
constructing a feature extraction model of computational power resource idle prediction, inputting a training set into the model for training, and extracting features of business behavior data by using the trained model;
and predicting the idle computing power resources of the business behavior data through the idle computing power resource prediction model.
2. The deep learning based computational power resource idleness prediction method according to claim 1, wherein the preprocessing further comprises redundant sample removal of business behavior data.
3. The deep learning-based idle prediction method of computing power resources according to claim 1, wherein the unbalanced processing adopts a random oversampling algorithm to perform unbalanced processing on idle prediction data of computing power resources, samples of minority classes are sampled in a random manner, and samples obtained by sampling are combined with initial samples of minority classes, so that the number of the samples of minority classes is the same as that of the samples of majority classes.
4. A deep learning based computational power resource idle prediction method according to claim 3, wherein the data set processed by the random oversampling algorithm is as follows:
wherein,,
and->
Represented by majority sample and minority sample, respectively, ">
The data set obtained by carrying out data unbalance processing on the initial data set S is represented as E, and the data set obtained by carrying out random sampling on a few types of samples in the data set is represented as E.
5. The deep learning based computational power resource free prediction method of claim 1, wherein the feature extraction model of computational power resource free prediction comprises a residual neural network comprising a convolutional layer, a pooling layer, a residual block, dropout, and a softmax classifier.
6. The deep learning-based computational power resource idle prediction method of claim 5, wherein the computational power resource idle prediction model is an extreme learning machine.
7. The deep learning based computational power resource idle prediction method of claim 6, wherein the network structure of the extreme learning machine comprises an input layer, an implicit layer, and an output layer.
8. The deep learning based computing power resource idleness prediction method of claim 5, wherein the method further comprises:
optimizing a residual neural network through an evolution algorithm, taking the inverse of the square error as an fitness function, and carrying out a standard for measuring the individual fitness in the population, wherein the formula is as follows:
wherein E is an error function, P is an integral output, W is a weight vector, X is an input vector, F is a fitness, j is a number of selections,
is a theoretical output.
9. The deep learning-based computational power resource idle prediction method of claim 5, wherein the residual neural network utilizes an attention mechanism to improve the threshold, and the network automatically generates a corresponding threshold according to the data itself.
10. A storage medium being a computer-readable storage medium storing computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform the deep learning based power resource idleness prediction method of any of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310603689.6A CN116340006B (en) | 2023-05-26 | 2023-05-26 | Computing power resource idle prediction method based on deep learning and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310603689.6A CN116340006B (en) | 2023-05-26 | 2023-05-26 | Computing power resource idle prediction method based on deep learning and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116340006A true CN116340006A (en) | 2023-06-27 |
| CN116340006B CN116340006B (en) | 2024-05-17 |
Family
ID=86884383
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310603689.6A Active CN116340006B (en) | 2023-05-26 | 2023-05-26 | Computing power resource idle prediction method based on deep learning and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116340006B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116805926A (en) * | 2023-08-21 | 2023-09-26 | 上海飞旗网络技术股份有限公司 | Network service type identification model training method and network service type identification method |
| CN116820784A (en) * | 2023-08-30 | 2023-09-29 | 杭州谐云科技有限公司 | GPU real-time scheduling method and system for reasoning task QoS |
| CN117252488A (en) * | 2023-11-16 | 2023-12-19 | 国网吉林省电力有限公司经济技术研究院 | Industrial cluster energy efficiency optimization method and system based on big data |
| CN117971511A (en) * | 2024-04-02 | 2024-05-03 | 青岛欧亚丰科技发展有限公司 | Collaborative visual simulation platform |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111740971A (en) * | 2020-06-15 | 2020-10-02 | 郑州大学 | Network intrusion detection model SGM-CNN based on class imbalance processing |
| CN113542241A (en) * | 2021-06-30 | 2021-10-22 | 杭州电子科技大学 | Intrusion detection method and device based on CNN-BiGRU mixed model |
| CN115987552A (en) * | 2022-11-18 | 2023-04-18 | 八维通科技有限公司 | Network intrusion detection method based on deep learning |
-
2023
- 2023-05-26 CN CN202310603689.6A patent/CN116340006B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111740971A (en) * | 2020-06-15 | 2020-10-02 | 郑州大学 | Network intrusion detection model SGM-CNN based on class imbalance processing |
| CN113542241A (en) * | 2021-06-30 | 2021-10-22 | 杭州电子科技大学 | Intrusion detection method and device based on CNN-BiGRU mixed model |
| CN115987552A (en) * | 2022-11-18 | 2023-04-18 | 八维通科技有限公司 | Network intrusion detection method based on deep learning |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116805926A (en) * | 2023-08-21 | 2023-09-26 | 上海飞旗网络技术股份有限公司 | Network service type identification model training method and network service type identification method |
| CN116805926B (en) * | 2023-08-21 | 2023-11-17 | 上海飞旗网络技术股份有限公司 | Network service type identification model training method and network service type identification method |
| CN116820784A (en) * | 2023-08-30 | 2023-09-29 | 杭州谐云科技有限公司 | GPU real-time scheduling method and system for reasoning task QoS |
| CN116820784B (en) * | 2023-08-30 | 2023-11-07 | 杭州谐云科技有限公司 | GPU real-time scheduling method and system for reasoning task QoS |
| CN117252488A (en) * | 2023-11-16 | 2023-12-19 | 国网吉林省电力有限公司经济技术研究院 | Industrial cluster energy efficiency optimization method and system based on big data |
| CN117252488B (en) * | 2023-11-16 | 2024-02-09 | 国网吉林省电力有限公司经济技术研究院 | Industrial cluster energy efficiency optimization method and system based on big data |
| CN117971511A (en) * | 2024-04-02 | 2024-05-03 | 青岛欧亚丰科技发展有限公司 | Collaborative visual simulation platform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116340006B (en) | 2024-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116340006B (en) | Computing power resource idle prediction method based on deep learning and storage medium | |
| Liu et al. | Stock transaction prediction modeling and analysis based on LSTM | |
| Ding et al. | Research on using genetic algorithms to optimize Elman neural networks | |
| EP3924893A1 (en) | Incremental training of machine learning tools | |
| CN110232203B (en) | Knowledge distillation optimization RNN short-term power failure prediction method, storage medium and equipment | |
| CN110738984A (en) | Artificial intelligence CNN, LSTM neural network speech recognition system | |
| CN110674636B (en) | Power consumption behavior analysis method | |
| CN113642225A (en) | CNN-LSTM short-term wind power prediction method based on attention mechanism | |
| CN111985845B (en) | Node priority optimization method of heterogeneous Spark cluster | |
| CN104199870A (en) | Method for building LS-SVM prediction model based on chaotic search | |
| CN113902116A (en) | Deep learning model-oriented reasoning batch processing optimization method and system | |
| CN113627070A (en) | Short-term photovoltaic power prediction method | |
| CN116109458A (en) | Reservoir flood discharge gate scheduling method, system, storage medium and computing equipment | |
| Ding et al. | Classification rules mining model with genetic algorithm in cloud computing | |
| Chaudhuri et al. | DSLEUTH: A distributed version of SLEUTH urban growth model | |
| Jat et al. | Applications of statistical techniques and artificial neural networks: A review | |
| CN113886454A (en) | Cloud resource prediction method based on LSTM-RBF | |
| Wei et al. | A combination forecasting method of grey neural network based on genetic algorithm | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| Wang et al. | Cloud computing and extreme learning machine for a distributed energy consumption forecasting in equipment-manufacturing enterprises | |
| Shuang et al. | Task Scheduling Based on Grey Wolf Optimizer Algorithm for Smart Meter Embedded Operating System | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| CN114820199A (en) | Method, device, storage medium and equipment for predicting price of financial derivatives | |
| Wu et al. | A Time Series Decomposition and Reinforcement Learning Ensemble Method for Short-Term Passenger Flow Prediction in Urban Rail Transit | |
| CN113132482A (en) | Distributed message system parameter adaptive optimization method based on reinforcement learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |