CN115099135A - Improved artificial neural network multi-type operation power consumption prediction method - Google Patents

Improved artificial neural network multi-type operation power consumption prediction method Download PDF

Info

Publication number
CN115099135A
CN115099135A CN202210686155.XA CN202210686155A CN115099135A CN 115099135 A CN115099135 A CN 115099135A CN 202210686155 A CN202210686155 A CN 202210686155A CN 115099135 A CN115099135 A CN 115099135A
Authority
CN
China
Prior art keywords
power consumption
neural network
consumption prediction
artificial neural
data
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.)
Pending
Application number
CN202210686155.XA
Other languages
Chinese (zh)
Inventor
敬超
李佳明
邱斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guilin University of Technology
Original Assignee
Guilin University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guilin University of Technology filed Critical Guilin University of Technology
Priority to CN202210686155.XA priority Critical patent/CN115099135A/en
Publication of CN115099135A publication Critical patent/CN115099135A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/06—Power analysis or power optimisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to the technical field of high-performance calculation, in particular to an improved artificial neural network multi-type operation power consumption prediction method. The invention also introduces an attention mechanism which can focus more attention on effective and key data and weaken the influence of redundant information on a result, and the method has stronger generalization capability and nonlinear mapping capability, does not depend on a pre-established mathematical model, has self-learning and self-adapting capability, and can be suitable for various types of operation.

Description

Improved artificial neural network multi-type operation power consumption prediction method
Technical Field
The invention relates to the technical field of high-performance calculation, in particular to an improved artificial neural network multi-type operation power consumption prediction method.
Background
With the increasing development of cloud computing technology, the amount of data that computers need to process also starts to exponentially and explosively increase, and the number and the scale of server data centers also continuously increase. The increasing computing demands make data centers consume enormous amounts of power, and the resulting high cost and resource waste issues have raised considerable attention in various industries. Research has shown that servers have been identified as the most energy consuming components in a data center.
The manner in which server power consumption is monitored can generally be divided into two categories. One type is a hardware-based monitoring method that typically utilizes some intelligent device to measure power usage, which is considered the most accurate, but also the most costly method. Meanwhile, the method is only suitable for small clusters and cannot meet the low-cost and extensible monitoring requirements of large-scale data centers; the second category is software-based monitoring methods that typically rely on pre-established mathematical models to collect system performance data (e.g., CPU, memory and disk utilization, network conditions, etc.) from servers, components or applications as inputs to functional expressions and to output system power consumption as predicted estimates.
The following two problems exist for the power consumption prediction of the CPU/GPU heterogeneous server. First, jobs can be divided into different types depending on their demand for computing resources. Different types of jobs have different influences on the power consumption of the server, and wrong recognition on the job types may influence the prediction precision of the power consumption prediction model, so that the reduction of the power consumption of the system is further influenced; secondly, many previous works mainly rely on modeling and prediction by using static mathematical formulas, but the method often needs to clearly and deeply know the performance parameters and power consumption changes of the server, and the adaptability and generalization capability of the model are weak, so that the dynamic changes of the server in the actual operation process cannot be well adapted.
Disclosure of Invention
The invention aims to provide an improved artificial neural network multi-type operation power consumption prediction method, and aims to complete power consumption prediction of a CPU/GPU heterogeneous server under different types of operation environments.
In order to achieve the above object, the present invention provides an improved artificial neural network multi-type operation power consumption prediction method, which comprises the following steps:
step 1: correctly dividing the operation types;
step 2: simulating the job type and collecting data;
and step 3: preprocessing data to obtain a data set;
and 4, step 4: the data set is divided into a training set, a verification set and a test set;
and 5: establishing a power consumption prediction model based on a convolutional neural network and a bidirectional long-short term memory artificial neural network by combining an attention mechanism;
step 6: inputting the divided data set into the power consumption prediction model, and training and predicting after adjusting the network structure and parameters;
and 7: introducing a traditional regression prediction model for comparison, and evaluating a predicted result;
and 8: if the evaluation result is not accordant, returning to the step 6 to continue execution; if the evaluation combination meets the requirements, the process is ended.
The operation types comprise calculation intensive type, memory intensive type and I/O intensive type, wherein the calculation intensive type is further divided into CPU intensive type and GPU intensive type.
The collected data comprises performance data and power consumption data, the performance data is generated by a benchmark test method in a simulation mode, CPU intensive jobs are generated by a CPUGrabEX in a simulation mode, memory intensive jobs are generated by a memtest in a simulation mode, I/O intensive jobs are generated by an IOzone in a simulation mode, GPU intensive jobs are obtained by running a deep learning task on a GPU, and the power consumption data are obtained by a power meter connected with the outside.
The proportion of the training set, the proportion of the verification set and the proportion of the test set are respectively 75%, 5% and 25%, and the training set is used for power consumption prediction model training; the verification set is taken from the training set but does not participate in training, and is used for testing the effect of the power consumption prediction model after each training is finished to prevent overfitting; the test set is used to finally determine the effect of the post-test power consumption prediction model.
In the process of establishing a power consumption prediction model based on the convolutional neural network and the bidirectional long and short term memory artificial neural network by combining an attention mechanism, the learning capability of the bidirectional long and short term memory artificial neural network on sequence information is utilized to replace a pooling layer of the traditional convolutional neural network, important features are identified by using the attention mechanism, and the interference of unimportant information on results is inhibited.
The bidirectional long-short term memory artificial neural network comprises a forward feedback process and a backward feedback process, wherein the output at the current moment is not only related to the feedback at the previous moment, but also influenced by the feedback at the later moment.
In the training and predicting process after the network structure and the parameters are adjusted, the power consumption prediction model is optimized in a mode of resetting the network structure or the parameters so as to improve the prediction precision.
The method for evaluating the predicted result compares the error between the actual value and the predicted value, and the evaluation index comprises an average absolute error and an average relative error.
The invention provides an improved artificial neural network multi-type operation power consumption prediction method, which adopts a mode of combining a convolutional neural network and a bidirectional long-short term memory neural network, wherein the convolutional neural network is used for extracting characteristic information, and the bidirectional long-short term memory neural network is used for extracting relationship characteristics between the front and the back of a sequence. The invention also introduces an attention mechanism which can focus more attention on effective and key data and weaken the influence of redundant information on a result, and the method has stronger generalization capability and nonlinear mapping capability, does not need to specifically know the relation between server performance data and power consumption data, does not depend on a pre-established mathematical model, has self-learning and self-adapting capability, and can be suitable for various types of operation.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of an improved artificial neural network multi-type operation power consumption prediction method according to the invention.
FIG. 2 is a flow chart of specific implementation steps of an improved artificial neural network multi-type operation power consumption prediction method in the invention.
FIG. 3 is a process schematic of data collection of the present invention.
FIG. 4 is a schematic diagram of the cell structure of the long-term and short-term memory network of the present invention.
FIG. 5 is a schematic diagram of a bidirectional long-short term memory network according to the present invention.
FIG. 6 is a schematic of the attention mechanism of the present invention.
Fig. 7 is a schematic diagram of the structure of the power consumption prediction model of the present invention.
FIG. 8 is a schematic diagram of power consumption variation for different job types in an embodiment of the present invention.
Fig. 9 is a diagram illustrating a prediction result of a power consumption prediction model according to an embodiment of the present invention.
FIG. 10 is a plot of mean absolute error versus a conventional regression prediction model in accordance with an embodiment of the present invention.
FIG. 11 is a graph of the average relative error of an embodiment of the present invention compared to a conventional regression prediction model.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
Referring to fig. 1, the present invention provides an improved method for predicting power consumption of multiple types of operations in an artificial neural network, comprising the following steps:
s1: correctly dividing the operation types;
s2: simulating the job type and collecting data;
s3: preprocessing data to obtain a data set;
s4: the data set is divided into a training set, a verification set and a test set;
s5: establishing a power consumption prediction model based on a convolutional neural network and a bidirectional long-short term memory artificial neural network by combining an attention mechanism;
s6: inputting the divided data set into the power consumption prediction model, and training and predicting after adjusting the network structure and parameters;
s7: introducing a traditional regression prediction model for comparison, and evaluating a predicted result;
s8: if the evaluation result does not meet the requirement, returning to the step S6 to continue the execution; if the evaluation combination meets the requirements, the process is ended.
Detailed description of the flow of the steps will be performed, referring specifically to fig. 2.
And step 1, accurately dividing the operation types.
In particular, from a job performance perspective, jobs can be divided into three categories, compute intensive, memory intensive, and I/O intensive, respectively. Many previous researches mostly use CPU resources as main influence factors of intensive computation, but with the development of deep learning, GPUs are widely applied to various fields and play a great role as computation elements. Therefore, the invention further divides the computation intensive type into a CPU intensive type and a GPU intensive type, and respectively considers the influence of the CPU intensive type and the GPU intensive type on the power consumption of the server.
And 2, simulating the operation type and collecting data.
Specifically, the collected data includes two parts: performance data and power consumption data. The performance data is generated by various benchmark test methods in a simulated manner, the power consumption data is obtained by an externally connected power meter, and a schematic diagram of the data collection process is shown in fig. 3. The CPU intensive operation is generated by CPUGrabEX simulation, the memory intensive operation is generated by memtest simulation, the I/O intensive operation is generated by IOzone simulation, and the GPU intensive program is obtained by running a deep learning task on a GPU. The collected performance data mainly comprises resource utilization rates of components such as a CPU, a Memory, a GPU and the like, disk reading and writing speeds, GPU temperature and the like.
And 3, preprocessing the data to obtain a data set.
Specifically, the collected raw data is first subjected to data cleaning, and the main purpose is to remove null values and abnormal values therein. Furthermore, it is necessary to normalize the processed data prior to the experiment because it can accelerate the gradient descent and help improve the prediction accuracy. The concrete implementation is shown in a formula:
Figure BDA0003697975300000051
wherein X represents the raw data to be processed, wherein X min And X max Respectively representing the minimum and maximum values, X, in the data set k Representing the normalized value and k representing the number of input features.
Step 4, dividing the data set: 75% training set, 5% validation set, 25% test set.
Specifically, the training set is used for model training; the verification set is taken from the training set but does not participate in the training, and is used for testing the effect of the model after each training is finished to prevent overfitting; the test set is used by the model to ultimately determine the effect of the post-test model.
And 5, establishing a power consumption prediction model based on the convolutional neural network and the bidirectional long-short term memory artificial neural network by combining an attention mechanism.
Specifically, the prediction model essentially utilizes the learning capability of the bidirectional long and short term memory artificial neural network on sequence information to replace a pooling layer of the traditional convolutional neural network, combines the advantages of the convolutional neural network and the bidirectional long and short term memory artificial neural network, extracts important characteristic information, uses an attention mechanism to identify important characteristics, and inhibits the interference of unimportant information on results, thereby improving the accuracy of the model.
Long-short term memory (LSTM) networks are a variation of the Recurrent Neural Network (RNN). In recent years, LSTM has been widely used to deal with the problem of predicting time series, which can avoid the long-term dependence problem. LSTM introduces the concept of thresholds (Gates), unlike RNN which considers only recent states, the cellular state of LSTM determines which states should be left, which states should be left behind, and the output result based on the current state and input.
LSTM introduces a cell state (cell state) compared to the hidden layer of the original RNN (hidden state), and fig. 4 shows the input and output of LSTM at time t. Wherein there are three inputs, each being a cell state C t-1 Hidden layer state U t-1 And the input vector at time t
Figure BDA0003697975300000061
The output is two: cell State C t And hidden layer state U t 。
LSTM cells comprise three gates: input door i t Output gate o t And forget door f t . Wherein the forgetting gate decides what information to discard from the cell state; the input gate mainly controls what new information is to be added to the cell state; the output gate determines what information is to be output from the cell state. The update state of LSTM cell units can be expressed by the following equation:
Figure BDA0003697975300000062
Figure BDA0003697975300000063
Figure BDA0003697975300000064
Figure BDA0003697975300000065
Figure BDA0003697975300000066
u t ＝o t *tanh(C t )
wherein W f 、W i 、W c 、W o And b f 、b i 、b c 、b o Are parameters learned in LSTM units. σ and tanh are activation functions.
Figure BDA0003697975300000067
Indicating the candidate state at time t.
Convolutional neural networks were used for the earliest time in the field of image processing, and they perform well in terms of feature extraction, but CNNs have difficulty in perceiving changes in features in time series data. The bidirectional long-short term memory neural network BilSTM (Bi-Directional LSTM) comprises two processes: forward feedback and backward feedback. The output at the current time is not only related to the feedback at the previous time, but also affected by the feedback at the next time, and the structure is shown in fig. 5. Compared with the LSTM, the method can better capture the dependency between data before and after the data is captured, and is more suitable for time series prediction.
Wherein (x) 1 ,x 2 x 3 ,,,x t ) Is an input, (y) 1 ,y 2 ,y 3 ,,,y t ) Is the output of the computer system,
Figure BDA0003697975300000068
is the forward input to the hidden layer and,
Figure BDA0003697975300000069
is the reverse input of the hidden layer, and the specific calculation formula is as follows:
Figure BDA00036979753000000610
Figure BDA00036979753000000611
Figure BDA00036979753000000612
attention in the human visual system may draw them to focus on a particular object, which may effectively improve the efficiency with which people receive visual information. Attention mechanism is a technique that mimics human cognitive attention and has been a popular model for recommendation systems, machine translation and other related fields. By applying the attention mechanism to the deep neural network, key information can be quickly screened out from big data, and redundant features in the data are filtered and weakened at the same time, so that the calculation cost is reduced, and the learning efficiency is improved.
The structure of the attention mechanism is shown in fig. 6. The attention mechanism is a general idea that does not rely on a specific framework by itself, but is currently mainly used in conjunction with an encoding-decoding (Encoder-Decoder) framework. The deep learning model combined with the attention mechanism can be regarded as a weight matrix, each input data has a corresponding weight value, and the greater the weight is, the stronger the attention is. The calculation formula is as follows:
e t ＝S t-1 ·C t
Figure BDA0003697975300000071
wherein, C t Representing the hidden state of the encoder, the last hidden state S in the Decoder t-1 ，e t The correlation (alignment degree) of the two is shown, and various calculation modes are available, including point multiplication, weighted point multiplication and concat mapping, and the dot product matrix (point multiplication) is used in the invention. a is t Is a weight coefficient indicating the degree of concentration of attention.
Based on the above background, the power consumption prediction model adopted by the invention combines the advantages of CNN and BilSTM, and introduces an attention mechanism to improve the efficiency and accuracy of the prediction model. The overall structure of the model is shown in fig. 7.
Firstly, feature extraction is realized on input data through a convolutional layer and a BilSTM layer, wherein the convolutional layer is used for feature extraction, and a bidirectional long-short term memory neural network is used for extracting relationship features between the front and the back of a sequence. In the process, a Dropout method is combined to discard some neurons randomly, so that overfitting is avoided, and the generalization capability of the model is improved. Next, the Attention mechanism is utilized for extracting more important features. And finally, outputting the data through a flatten layer. The model essentially utilizes the learning capability of the LSTM on the sequence information to replace a pooling layer of the traditional convolutional neural network, combines the advantages of the CNN and the LSTM models together, retains important characteristic information, then uses an attention layer to identify the important characteristics, and inhibits the interference of unimportant information on the result, thereby improving the accuracy of the model.
And 6, inputting the divided data set into the power consumption prediction model, and training and predicting after adjusting the network structure and parameters.
Specifically, the preprocessed data are divided and respectively input into the model for training and prediction. The dimension of the input data is 8, the time step value is 3, the number of convolution kernels (output dimension) of the convolution layer is 64, and the activation functions of the convolution layer and the output layer are ReLU and sigmod, respectively. During training, the epoch size is set to 150 and the batch size to 64.
And 7, introducing a traditional regression prediction model for comparison, and evaluating a predicted result.
And (3) introducing a traditional multiple linear regression model and a non-linear regression model for comparison, and evaluating the predicted result through average absolute error and average relative error.
Specifically, the invention introduces a traditional regression prediction model for comparison, which comprises a Multiple Linear Regression (MLR) and a nonlinear regression (DTR); the predicted results are compared by comparing the error between the true and predicted values, and the regression evaluation indices include Mean Absolute Error (MAE) and Mean Relative Error (MRE).
And 8, if the result is not ideal, returning to the step 6, and improving the prediction precision by adjusting the neural network structure, parameters and other operation optimization models.
Specifically, the result is compared with a conventional regression prediction model, and if the result is not ideal, the prediction model may be optimized by adjusting a network structure or parameters (time step, activation function, batch size, etc.) to improve the prediction accuracy.
Furthermore, the invention provides a specific embodiment, and experiments prove that the method is feasible and is superior to the traditional regression prediction model.
In order to make the experiments of the present invention more clearly understood, some experimental terms in the present invention are described first:
the experiment involved four different types of work: CPU-intensive, memory-intensive, I/O-intensive, and GPU-intensive.
The evaluation indexes comprise average absolute error (MAE) and average relative error (MRE), and the specific calculation formula is as follows:
Figure BDA0003697975300000081
Figure BDA0003697975300000082
the comparative experiments include the traditional regression prediction models of Multiple Linear Regression (MLR), nonlinear regression (DTR, decision tree regression).
For convenience of plotting, the prediction models proposed by the present invention and those based on MLR and DTR were named CBLA _ PM, MLR _ PM and DTR _ PM, respectively, during the experiment.
For performance data, the invention simulates the workload data generated by different types of operation under the actual working environment by using a benchmark test method, and collects the required data through interfaces provided by Windows, NVIDIA and the like. The experiment collected a total of 1062 pieces of CPU-intensive data, 675 pieces of I/O-intensive data, 941 pieces of memory-intensive data, and 1279 pieces of GPU-intensive data. And for the power consumption data, recording the real-time change of the power consumption by using an externally connected intelligent electric meter. Fig. 8 shows the power consumption of the server over time for different job types.
Further, referring to fig. 9, fig. 9 shows a prediction situation of the power consumption prediction model according to the present invention under different operation types.
Fig. 10 and 11 show the mean absolute error and mean relative error of the proposed model and the conventional regression prediction model under different operations. As can be seen from the experimental results, the prediction accuracy of the CBLA _ PM provided by the invention is obviously better than that of the MLR _ PM and the DTR _ PM.
In summary, compared with the conventional power consumption prediction method, the method of the present invention has the following characteristics:
1. unlike many previous works which mainly rely on static mathematical formula modeling, the method often needs to have clear and profound knowledge about the relationship between related elements, and the generalization capability is weak, so that the method cannot adapt to the dynamic change of the server. The invention adopts an improved neural network method as a black box model, and as long as enough samples are provided, the relation between input and output can be established under the condition that the relation of each part cannot be solved, so that the effective prediction is realized.
2. The method provided by the invention adopts a mode of combining the convolutional neural network and the bidirectional long-short term memory neural network, and integrates the advantages of the two neural networks. The convolutional neural network is used for extracting characteristic information, and the bidirectional long-short term memory neural network is used for extracting relationship characteristics between the front and the back of the sequence. The bidirectional long-short term memory neural network makes up the defects of the convolutional neural network in the aspect of perceiving the change of the sequence data, and the combination of the two can effectively improve the feature extraction efficiency.
3. The method of the present invention also introduces a mechanism of attention. The attention mechanism may focus more on valid, critical data while attenuating the effect of redundant information on the results.
4. The proposed prediction model was validated under different types of jobs and experimental results proved to be feasible. The method provided by the invention is proved to have strong generalization capability and be suitable for various situations.
While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (8)

1. An improved artificial neural network multi-type operation power consumption prediction method is characterized by comprising the following steps:
step 1: correctly dividing the operation types;
step 2: simulating the job type and collecting data;
and step 3: preprocessing data to obtain a data set;
and 4, step 4: the data set is divided into a training set, a verification set and a test set;
and 5: establishing a power consumption prediction model based on a convolutional neural network and a bidirectional long-short term memory artificial neural network by combining an attention mechanism;
step 6: inputting the divided data set into the power consumption prediction model, and training and predicting after adjusting the network structure and parameters;
and 7: introducing a traditional regression prediction model for comparison, and evaluating a predicted result;
and 8: if the evaluation result is not accordant, returning to the step 6 for continuous execution; if the evaluation combination meets the requirements, the process is ended.
2. The improved artificial neural network multi-type job power consumption prediction method of claim 1,
the job types include compute intensive, memory intensive, and I/O intensive, wherein the compute intensive is further divided into CPU intensive and GPU intensive.
3. The improved artificial neural network multi-type job power consumption prediction method of claim 1,
the collected data comprises performance data and power consumption data, the performance data is generated by a benchmark test method in a simulation mode, CPU intensive jobs are generated by a CPUGrabEX simulation mode, memory intensive jobs are generated by a memtest simulation mode, I/O intensive jobs are generated by an IOzone simulation mode, GPU intensive jobs are obtained by running a deep learning task on a GPU, and the power consumption data are obtained by a power meter connected with the outside.
4. The improved artificial neural network multi-type job power consumption prediction method of claim 1,
the proportion of the training set, the proportion of the verification set and the proportion of the test set are respectively 75%, 5% and 25%, and the training set is used for power consumption prediction model training; the verification set is taken from the training set but does not participate in training, and is used for testing the effect of the power consumption prediction model after each training is finished, so that overfitting is prevented; the test set is used to finally determine the effect of the post-test power consumption prediction model.
5. The improved artificial neural network multi-type job power consumption prediction method of claim 1,
in the process of establishing a power consumption prediction model based on the convolutional neural network and the bidirectional long and short term memory artificial neural network by combining an attention mechanism, the learning capability of the bidirectional long and short term memory artificial neural network on sequence information is utilized to replace a pooling layer of the traditional convolutional neural network, important features are identified by using the attention mechanism, and the interference of unimportant information on results is inhibited.
6. The improved artificial neural network multi-type job power consumption prediction method of claim 5,
the bidirectional long-short term memory artificial neural network comprises a forward feedback process and a backward feedback process, wherein the output at the current moment is not only related to the feedback at the previous moment, but also influenced by the feedback at the later moment.
7. The improved artificial neural network multi-type job power consumption prediction method of claim 1,
and in the training and predicting process after the network structure and the parameters are adjusted, optimizing the power consumption prediction model in a mode of resetting the network structure or the parameters so as to improve the prediction precision.
8. The improved artificial neural network multi-type job power consumption prediction method of claim 1,
the method for evaluating the predicted result compares the error between the real value and the predicted value, and the evaluation index comprises the average absolute error and the average relative error.
CN202210686155.XA 2022-06-16 2022-06-16 Improved artificial neural network multi-type operation power consumption prediction method Pending CN115099135A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210686155.XA CN115099135A (en) 2022-06-16 2022-06-16 Improved artificial neural network multi-type operation power consumption prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210686155.XA CN115099135A (en) 2022-06-16 2022-06-16 Improved artificial neural network multi-type operation power consumption prediction method

Publications (1)

Publication Number Publication Date
CN115099135A true CN115099135A (en) 2022-09-23

Family

ID=83291400

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210686155.XA Pending CN115099135A (en) 2022-06-16 2022-06-16 Improved artificial neural network multi-type operation power consumption prediction method

Country Status (1)

Country Link
CN (1) CN115099135A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116127840A (en) * 2023-01-05 2023-05-16 无锡钊源电力电子有限公司 Data center load prediction method based on data driving
CN117620345A (en) * 2023-12-28 2024-03-01 诚联恺达科技有限公司 Data recording system of vacuum reflow oven

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116127840A (en) * 2023-01-05 2023-05-16 无锡钊源电力电子有限公司 Data center load prediction method based on data driving
CN116127840B (en) * 2023-01-05 2023-09-22 无锡钊源电力电子有限公司 Data center load prediction method based on data driving
CN117620345A (en) * 2023-12-28 2024-03-01 诚联恺达科技有限公司 Data recording system of vacuum reflow oven
CN117620345B (en) * 2023-12-28 2024-06-07 诚联恺达科技有限公司 Data recording system of vacuum reflow oven

Similar Documents

Publication Publication Date Title
CN111899510B (en) Intelligent traffic system flow short-term prediction method and system based on divergent convolution and GAT
WO2022160902A1 (en) Anomaly detection method for large-scale multivariate time series data in cloud environment
Zhang et al. [Retracted] Complex System of Vertical Baduanjin Lifting Motion Sensing Recognition under the Background of Big Data
CN110738984A (en) Artificial intelligence CNN, LSTM neural network speech recognition system
CN115099135A (en) Improved artificial neural network multi-type operation power consumption prediction method
Wu et al. A weighted deep domain adaptation method for industrial fault prognostics according to prior distribution of complex working conditions
CN111241755A (en) Power load prediction method
US20240012965A1 (en) Steady flow prediction method in plane cascade based on generative adversarial network
CN108256238B (en) Fiber grating wavelength demodulation method and device based on deep learning
CN111814956A (en) Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction
CN116340796B (en) Time sequence data analysis method, device, equipment and storage medium
CN112529071B (en) Text classification method, system, computer equipment and storage medium
Liu et al. Correntropy long short term memory soft sensor for quality prediction in industrial polyethylene process
CN114490065A (en) Load prediction method, device and equipment
CN112182954A (en) Fluid simulation data prediction model based on LSTM
CN112560981A (en) Training method, apparatus, device, program and storage medium for generating countermeasure model
Zhao et al. A neural architecture search method based on gradient descent for remaining useful life estimation
Wu et al. Quality estimation method for gear hobbing based on attention and adversarial transfer learning
Fu et al. MCA-DTCN: A novel dual-task temporal convolutional network with multi-channel attention for first prediction time detection and remaining useful life prediction
CN115062727A (en) Graph node classification method and system based on multi-order hypergraph convolutional network
CN114897138A (en) System fault diagnosis method based on attention mechanism and depth residual error network
CN114240687A (en) Energy hosting efficiency analysis method suitable for comprehensive energy system
Zeng et al. Dynamic Bayesian networks for feature learning and transfer applications in remaining useful life estimation
Zong et al. Embedded software fault prediction based on back propagation neural network
CN116885697A (en) Load prediction method based on combination of cluster analysis and intelligent algorithm

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