CN114138463B - Method for predicting load balance of spot system application layer based on deep neural network - Google Patents
Method for predicting load balance of spot system application layer based on deep neural network Download PDFInfo
- Publication number
- CN114138463B CN114138463B CN202111301688.3A CN202111301688A CN114138463B CN 114138463 B CN114138463 B CN 114138463B CN 202111301688 A CN202111301688 A CN 202111301688A CN 114138463 B CN114138463 B CN 114138463B
- Authority
- CN
- China
- Prior art keywords
- load
- server
- neural network
- deep neural
- real
- 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.)
- Active
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 43
- 238000000034 method Methods 0.000 title claims abstract description 22
- 230000008859 change Effects 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims description 21
- 238000013527 convolutional neural network Methods 0.000 claims description 14
- 230000004927 fusion Effects 0.000 claims description 10
- 238000003062 neural network model Methods 0.000 claims description 10
- 238000004891 communication Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000026683 transduction Effects 0.000 description 1
- 238000010361 transduction Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
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]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/3003—Monitoring arrangements specially adapted to the computing system or computing system component being monitored
- G06F11/3006—Monitoring arrangements specially adapted to the computing system or computing system component being monitored where the computing system is distributed, e.g. networked systems, clusters, multiprocessor systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3409—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment
- G06F11/3433—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment for load management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3447—Performance evaluation by modeling
-
- 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
- G06F9/505—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 considering the load
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L47/00—Traffic control in data switching networks
- H04L47/10—Flow control; Congestion control
- H04L47/12—Avoiding congestion; Recovering from congestion
- H04L47/125—Avoiding congestion; Recovering from congestion by balancing the load, e.g. traffic engineering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2209/00—Indexing scheme relating to G06F9/00
- G06F2209/50—Indexing scheme relating to G06F9/50
- G06F2209/508—Monitor
Abstract
The invention discloses a load balancing prediction method of an application layer of an off-the-shelf system based on a deep neural network, which comprises the steps of firstly predicting real-time information of a load server and determining a scheduling priority of the load server; then, predicting a change curve of future flow according to the real-time flow information of the proxy server, so that the flow throughput condition of the whole spot system network can be accurately analyzed; and finally, carrying out dynamic optimal load balancing scheduling on the server by utilizing the predicted scheduling priority of the load server and the change curve of the future flow, thereby ensuring the platform safety, high reliability and high availability of the whole electric power spot system.
Description
Technical Field
The invention relates to the field of deep learning, in particular to a load balancing prediction method for an application layer of a spot system based on a deep neural network.
Background
The power spot market plays an extremely critical role in power resource trading as an important component of the power market. The electric power spot transaction refers to spot electric power (electric quantity) transactions in the past and in the day, which are carried out by market subjects such as power generation enterprises, electric power selling companies, electric power users and the like meeting the admission conditions, through a centralized bidding mode and a marketized transaction mode of price clearing according to node marginal price.
The main responsibility of the electric power spot system is to ensure the normal operation of the electric power spot market business and the safe and stable operation of the electric network. The general system architecture mainly comprises an infrastructure layer, a platform service layer and an application layer. The application layer comprises a plurality of subsystem application functions, such as application functions of a spot subsystem for market pre-clearing, real-time balance market, auxiliary service market operation and the like; the contract market subsystem is responsible for transaction management, contract management, market member management and other functions; the market settlement subsystem is responsible for carrying out settlement calculation, detailed calculation bill issuing and other functions according to transaction results, metering data and the like; the data reporting and information issuing subsystem is responsible for the functions of market member registration, data reporting, information issuing and the like.
It follows that the application layer of a power off-the-shelf system is responsible for a large variety and number of application functions, and how to distribute these tasks equally to the servers is a very important and practical problem that directly affects the security and usability of the system. The existing load balancing method mainly uses software and hardware as main materials, and adopts an empirical static or dynamic load balancing algorithm, such as polling allocation, random allocation, minimum connection allocation and the like. The empirical load balancing algorithm cannot fully utilize the throughput information of the server to accurately predict the flow in the network, and load scheduling is carried out by means of the prediction information, so that the resource occupation of the server is unbalanced, and the actual load balancing cannot be achieved.
In summary, how to accurately analyze the traffic throughput of the whole spot system network and utilize the predicted data information to perform dynamic server load balancing scheduling when predicting the application layer load balancing of the power spot system is a problem to be solved at present.
Disclosure of Invention
Aiming at the problems, the invention provides a method for predicting the load balance of the application layer of the spot system based on the deep neural network, which can predict the load balance of the application layer of the power spot system, accurately analyze the traffic throughput condition of the whole spot system network, and utilize the predicted data information to dynamically perform server load balance scheduling, thereby ensuring the platform safety, high reliability and high availability of the whole power spot system.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
the method for predicting the load balance of the spot system application layer based on the deep neural network is characterized by comprising the following steps of:
step 1: real-time information of each load server of the power spot system application layer is collected in real time and summarized to a data center server;
step 2: the data center server obtains the load capacity score of each load server in real time by utilizing a deep neural network according to the acquired real-time information of the load servers, and determines the scheduling priority of each load server based on the load capacity score;
step 3: the client initiates a request to an application layer proxy server of the power spot system;
step 4: collecting flow information of the proxy server in real time and sending the flow information to the data center server;
step 5: the data center server predicts network flow for a period of time in the future by utilizing a deep neural network according to the collected flow information of the proxy server, and formulates an optimal load balancing scheduling scheme according to the predicted network flow and the scheduling priority of the load server, and returns the optimal load balancing scheduling scheme to the proxy server;
step 6: the proxy server calculates and selects an optimal load server address according to the optimal load balance scheduling scheme;
step 7: and the proxy server establishes a service communication channel between the client and the load server corresponding to the address according to the optimal load server address.
Further, the servers in the power spot system application layer comprise proxy servers, data center servers and load servers;
the proxy server is used for receiving a connection request of the client and an optimal load scheduling scheme sent by the data center server, so that an optimal load server address is calculated and selected, and a service communication channel between the client and a load server corresponding to the address is established according to the obtained optimal load server address;
the data center server is used for processing the data information transmitted by the proxy server and the load servers, predicting a future flow change curve in the network and the scheduling priority of each load server, so as to calculate an optimal load balancing scheduling scheme, and returning the optimal load balancing scheduling scheme to the proxy server;
the load server is used for realizing and executing each functional module of the power spot system application layer, receives the connection signals sent by the proxy server, and transmits the data information to the data center server and the client.
Further, the real-time information of the load server collected in the step 1 includes memory occupation condition, CPU running condition, hard disk storage space, real-time network traffic, etc.
Further, the deep neural network in the step 2 adopts a multi-mode fusion deep convolutional neural network;
the multi-modal fused deep convolutional neural network comprises a plurality of feature mapping layers and a convolutional coding layer, wherein the feature mapping layers map the input multi-modal real-time information of the load server into a unified feature space, then multi-modal feature fusion is carried out in the unified feature space, and finally the fused features are coded by the convolutional coding layer.
Further, the specific operation steps of obtaining the load capacity score of each load server by using the deep neural network in the step 2 include:
step 21: acquiring real-time information of each load server in an application layer of the electric power spot system in advance to obtain a real-time information data training set;
step 22: expert manual scoring is carried out on the real-time information in the real-time information data training set, and a load capacity scoring code of manual marking is obtained;
step 23: aiming at the real-time information of each load server obtained from the data center server, calculating the load capacity grading codes of each load server through the multi-mode fusion deep convolutional neural network, and calculating the L1 loss between the load capacity grading codes output by the deep neural network and the artificially marked load capacity grading codes; training deep neural network parameters by minimizing the L1 loss iteration, so as to obtain a deep neural network model capable of encoding real-time information of a load server into corresponding load capacity scores;
step 24: inputting the real-time information of each load server in the step 2 into the deep neural network model trained in the step 23, and outputting the load capacity scores corresponding to each load server in real time;
step 25: and sequencing the load capacity scores of the load servers to finally obtain the scheduling priority of the load servers.
Further, the traffic information of the proxy server includes traffic timing data of the proxy server in each period.
Further, the deep neural network in step 5 adopts a global and local self-atttention-based transducer network for time sequence prediction;
the global and local self-attitudes-based transform network replaces the global self-attitudes in the original transform network by the global and local self-attitudes, and can effectively capture the flow change relation of the local short distance and the global long distance.
Further, the step of predicting the network traffic for a period of time in the future by using the deep neural network and obtaining the optimal load balancing scheduling scheme in step 5 includes:
step 51: acquiring flow information of a proxy server on the proxy server in advance to obtain a flow information training set;
step 52: performing time sequence prediction training on the flow information training set through a deep neural network to obtain a time sequence prediction deep neural network model, wherein the time sequence prediction deep neural network model can predict a network flow change curve of a period of time in the future according to flow information of any time point or time period;
step 53: based on a network flow change curve, an optimal load balancing scheduling scheme is formulated according to a scheduling rule of a load server, wherein the scheduling rule is as follows: high traffic is scheduled to high priority load servers and low traffic is scheduled to low priority load servers.
The beneficial effects of the invention are as follows:
the method provided by the invention firstly predicts the real-time information of the load server to determine the dispatching priority of the load server. The method adopts the deep convolutional neural network of the multi-mode fusion to predict the real-time information of the load server, firstly utilizes the feature mapping layer to map to a unified feature space aiming at the input information of different modes, then carries out multi-mode feature fusion in the unified feature space, and finally utilizes the convolutional coding layer to code the fused features, thereby better learning the association between the information of different modes and the final prediction result and improving the prediction accuracy.
And then predicting the change curve of the future flow according to the real-time flow information of the proxy server. The method adopts the transducer network based on global and local self-saturation, and can effectively capture the flow change relation between the local short distance and the global long distance, thereby more accurately analyzing and predicting the flow throughput condition of the whole spot system network.
And finally, carrying out dynamic optimal load balancing scheduling on the server by utilizing the predicted scheduling priority of the load server and the change curve of the future flow, thereby ensuring the platform safety, high reliability and high availability of the whole electric power spot system.
Drawings
FIG. 1 is a schematic diagram of a relationship between each server and a client in an application layer of an electric power spot system in an embodiment;
FIG. 2 is a flow chart of a method for predicting load balancing of an application layer of an electric power spot system;
FIG. 3 is a schematic diagram of a multi-modal fused deep convolutional neural network.
Detailed Description
In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
As shown in fig. 2, the invention provides a method for predicting load balancing of an application layer of a spot system based on a deep neural network, which comprises the following steps:
step 1: real-time information of each load server of the power spot system application layer is collected in real time and summarized to a data center server;
step 2: the data center server obtains the load capacity score of each load server in real time by utilizing a deep neural network according to the acquired real-time information of the load servers, and determines the scheduling priority of each load server based on the load capacity score;
step 3: the client initiates a request to an application layer proxy server of the power spot system;
step 4: collecting flow information of the proxy server in real time and sending the flow information to the data center server;
step 5: the data center server predicts network flow for a period of time in the future by utilizing a deep neural network according to the collected flow information of the proxy server, and formulates an optimal load balancing scheduling scheme according to the predicted network flow and the scheduling priority of the load server, and returns the optimal load balancing scheduling scheme to the proxy server;
step 6: the proxy server calculates and selects an optimal load server address according to the optimal load balance scheduling scheme;
step 7: and the proxy server establishes a service communication channel between the client and the load server corresponding to the address according to the optimal load server address.
Examples
Fig. 1 shows a relationship between each server and a client in an application layer of a power spot system, wherein the servers can be divided into three types according to functions, namely, a proxy server, a data center server and a load server. The proxy server is mainly responsible for receiving a connection request of a client and an optimal load scheduling scheme sent by a data center server, so that an optimal load server address is calculated and selected, and then a service communication channel between the client and a load server corresponding to the optimal load server address is established; the data center server is mainly responsible for processing data information transmitted by the proxy server and the load servers, predicting a future flow change curve in the network and the scheduling priority of each load server, so as to calculate an optimal load balancing scheduling scheme, and returning the optimal load balancing scheduling scheme to the proxy server; the load server is mainly responsible for realizing and executing each functional module of the application layer of the power spot system, receives the connection signals sent by the proxy server, and transmits data information to the data center server and the client.
In the power spot system shown in fig. 1, the steps of predicting by using the power spot system application layer load balancing prediction method provided by the invention are shown in fig. 2, and include:
step S101: information of each load server of an application layer of the electric power spot system is collected in real time and summarized to a data center server;
specifically, the real-time information of each load server of the power spot system application layer may include memory occupation conditions, CPU running conditions, hard disk storage space, real-time network traffic, and the like.
Step S102, the data center server predicts the load capacity of each load server in real time by using a deep neural network according to the collected real-time information of the load servers and scores the load capacity, and determines the scheduling priority of each load server according to the predicted load capacity scores of the load servers;
specifically, the deep neural network adopts a multi-mode fusion deep convolutional neural network;
further, the structure of the multi-modal fused deep convolutional neural network is shown in fig. 3, and the deep convolutional neural network comprises a plurality of feature mapping layers and a convolutional coding layer, wherein the feature mapping layers map the input multi-modal real-time information of the load server into a unified feature space, then multi-modal feature fusion is carried out in the unified feature space, and finally the fused features are coded by using the convolutional coding layer.
Further, the multi-modal fused deep convolutional neural network firstly carries out coding task training on a real-time information data training set of a load server acquired in advance;
the real-time information data training set of the load servers collected in advance comprises real-time information of each load server of an application layer of the power spot system, and the real-time information is scored manually by a computer expert, wherein the load capacity scoring reflects the capacity of the current load server capable of loading tasks of the application layer;
the coding task training is to carry out load capacity grading coding on the real-time information of the load server after vector quantization through a depth neural network, then calculate L1 loss between the load capacity grading coding output by the network and the load capacity grading coding manually marked by an actual computer expert, and iteratively train the parameters of the depth neural network by minimizing the L1 loss, so as to train and obtain a depth neural network model for coding the real-time information of the load server into corresponding load capacity grading;
after the deep neural network model is trained, the real-time information of the load server is input, and the corresponding load capacity score can be output in real time. And according to the load capacity scores of the load servers, the scheduling priorities of the load servers are obtained.
Step S103, the client initiates a request to an application layer proxy server of the power spot system;
step S104, collecting flow information of the proxy server in real time, and summarizing and transmitting the flow information to the data center server;
step S105, the data center server predicts network flow for a period of time in the future by utilizing a deep neural network according to the collected flow information of the proxy server, and formulates an optimal load balancing scheduling scheme according to the predicted network flow and the scheduling priority of the load server, and returns the optimal load balancing scheduling scheme to the proxy server;
specifically, the deep neural network may employ a global and local self-attension based transform network for time series sequence prediction;
among them, the Transformer network is a model for the seq2seq (sequence to sequence, sequence-to-sequence) task proposed by Google in 2017 in Attention IsAllYouNeed, which has no cyclic structure of RNN (recurrentneuroalnetwork) or convolutional structure of CNN (Convolutional Neural Network ), and has improved the sequence task in machine translation and the like. The method avoids a circular model structure, and models the global dependence of input and output completely by a self-attention mechanism. Self-attention mechanisms (self-attention) have become an important component of sequence modeling and transduction models in various tasks, which allow modeling of the dependency terms of input-output sequences without regard to their distance in the sequence, so that a long range of dependencies can be better captured.
Specifically, the global and local self-attribute-based transform network replaces the global self-attribute in the original transform network by the global and local self-attribute, so that the flow change relation of the local short distance and the global long distance can be effectively and simultaneously captured, and the flow prediction accuracy of the spot system network is improved;
further, the deep neural network firstly carries out time sequence prediction task training on a flow information training set of a proxy server which is acquired in advance;
the traffic information training set of the proxy server collected in advance includes traffic time series data (traffic data sampling interval can be set according to actual requirements, such as 1 minute) of the proxy server in each time period (the time period length can be set according to actual requirements, such as 1 hour).
By training the time sequence prediction task on the traffic information training set of the proxy server, the trained deep neural network can accurately predict a network traffic change curve for a period of time in the future according to network traffic information at any time point (or time period), and then an optimal load balancing scheduling scheme is formulated by using a scheduling rule that high traffic is scheduled to a high-priority load server and low traffic is scheduled to a low-priority load server, and finally the optimal load balancing scheduling scheme is returned to the proxy server.
Step S106, the proxy server calculates and selects an optimal load server address according to the optimal load balancing scheduling scheme;
specifically, the proxy server may perform domain name and IP resolution through DNS, thereby calculating an optimal load server address.
In step S107, the proxy server establishes a service communication channel between the client and the load server corresponding to the address according to the optimal load server address.
The foregoing has shown and described the basic principles, principal features and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (8)
1. The method for predicting the load balance of the spot system application layer based on the deep neural network is characterized by comprising the following steps of:
step 1: real-time information of each load server of the power spot system application layer is collected in real time and summarized to a data center server;
step 2: the data center server obtains the load capacity score of each load server in real time by utilizing a deep neural network according to the acquired real-time information of the load servers, and determines the scheduling priority of each load server based on the load capacity score;
step 3: the client initiates a request to an application layer proxy server of the power spot system;
step 4: collecting flow information of the proxy server in real time and sending the flow information to the data center server;
step 5: the data center server predicts network flow for a period of time in the future by utilizing a deep neural network according to the collected flow information of the proxy server, and formulates an optimal load balancing scheduling scheme according to the predicted network flow and the scheduling priority of the load server, and returns the optimal load balancing scheduling scheme to the proxy server;
step 6: the proxy server calculates and selects an optimal load server address according to the optimal load balance scheduling scheme;
step 7: and the proxy server establishes a service communication channel between the client and the load server corresponding to the address according to the optimal load server address.
2. The method for predicting load balancing of spot system application layer based on deep neural network as claimed in claim 1, wherein the servers in the power spot system application layer comprise a proxy server, a data center server and a load server;
the proxy server is used for receiving a connection request of the client and an optimal load scheduling scheme sent by the data center server, so that an optimal load server address is calculated and selected, and a service communication channel between the client and a load server corresponding to the address is established according to the obtained optimal load server address;
the data center server is used for processing the data information transmitted by the proxy server and the load servers, predicting a future flow change curve in the network and the scheduling priority of each load server, so as to calculate an optimal load balancing scheduling scheme, and returning the optimal load balancing scheduling scheme to the proxy server;
the load server is used for realizing and executing each functional module of the power spot system application layer, receives the connection signals sent by the proxy server, and transmits the data information to the data center server and the client.
3. The method for predicting load balancing of spot system application layer based on deep neural network according to claim 1, wherein the real-time information of the load server collected in step 1 includes memory occupation condition, CPU running condition, hard disk storage space, real-time network traffic, etc.
4. The method for predicting the load balance of the spot system application layer based on the deep neural network according to claim 1, wherein the deep neural network in the step 2 adopts a multi-mode fusion deep convolutional neural network;
the multi-modal fused deep convolutional neural network comprises a plurality of feature mapping layers and a convolutional coding layer, wherein the feature mapping layers map the input multi-modal real-time information of the load server into a unified feature space, then multi-modal feature fusion is carried out in the unified feature space, and finally the fused features are coded by the convolutional coding layer.
5. The method for predicting load balancing of spot system application layer based on deep neural network as claimed in claim 4, wherein the specific operation step of obtaining the load capacity score of each load server by using the deep neural network in step 2 comprises:
step 21: acquiring real-time information of each load server in an application layer of the electric power spot system in advance to obtain a real-time information data training set;
step 22: expert manual scoring is carried out on the real-time information in the real-time information data training set, and a load capacity scoring code of manual marking is obtained;
step 23: aiming at the real-time information of each load server obtained from the data center server, calculating the load capacity grading codes of each load server through the multi-mode fusion deep convolutional neural network, and calculating the L1 loss between the load capacity grading codes output by the deep neural network and the artificially marked load capacity grading codes; training deep neural network parameters by minimizing the L1 loss iteration, so as to obtain a deep neural network model capable of encoding real-time information of a load server into corresponding load capacity scores;
step 24: inputting the real-time information of each load server in the step 2 into the deep neural network model trained in the step 23, and outputting the load capacity scores corresponding to each load server in real time;
step 25: and sequencing the load capacity scores of the load servers to finally obtain the scheduling priority of the load servers.
6. The method for predicting load balancing of an off-the-shelf system based on a deep neural network according to claim 1, wherein the traffic information of the proxy server includes traffic timing data of the proxy server in each time period.
7. The method for predicting the load balancing of the spot system application layer based on the deep neural network according to claim 1, wherein the deep neural network in the step 5 adopts a global and local self-attitution-based transform network for time sequence prediction;
the global and local self-attitudes-based transform network replaces the global self-attitudes in the original transform network by the global and local self-attitudes, and can effectively capture the flow change relation of the local short distance and the global long distance.
8. The method for predicting load balancing of spot system application layer based on deep neural network according to claim 1, wherein the step of predicting network traffic for a period of time in the future by using the deep neural network in step 5, and obtaining an optimal load balancing scheduling scheme comprises:
step 51: acquiring flow information of a proxy server on the proxy server in advance to obtain a flow information training set;
step 52: performing time sequence prediction training on the flow information training set through a deep neural network to obtain a time sequence prediction deep neural network model, wherein the time sequence prediction deep neural network model can predict a network flow change curve of a period of time in the future according to flow information of any time point or time period;
step 53: based on a network flow change curve, an optimal load balancing scheduling scheme is formulated according to a scheduling rule of a load server, wherein the scheduling rule is as follows: high traffic is scheduled to high priority load servers and low traffic is scheduled to low priority load servers.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111301688.3A CN114138463B (en) | 2021-11-04 | 2021-11-04 | Method for predicting load balance of spot system application layer based on deep neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111301688.3A CN114138463B (en) | 2021-11-04 | 2021-11-04 | Method for predicting load balance of spot system application layer based on deep neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114138463A CN114138463A (en) | 2022-03-04 |
CN114138463B true CN114138463B (en) | 2024-03-26 |
Family
ID=80392799
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111301688.3A Active CN114138463B (en) | 2021-11-04 | 2021-11-04 | Method for predicting load balance of spot system application layer based on deep neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114138463B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114969209A (en) * | 2022-06-15 | 2022-08-30 | 支付宝(杭州)信息技术有限公司 | Training method and device, and method and device for predicting resource consumption |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2004105355A1 (en) * | 2003-05-21 | 2004-12-02 | Nitgen Technologies Co., Ltd. | Intelligent traffic management system for networks and intelligent traffic management method using the same |
CN107864189A (en) * | 2017-10-18 | 2018-03-30 | 南京邮电大学 | A kind of application layer traffic load-balancing method based on DPI |
CN109729017A (en) * | 2019-03-14 | 2019-05-07 | 哈尔滨工程大学 | A kind of load-balancing method based on DPI prediction |
CN110784555A (en) * | 2019-11-07 | 2020-02-11 | 中电福富信息科技有限公司 | Intelligent monitoring and load scheduling method based on deep learning |
CN112949739A (en) * | 2021-03-17 | 2021-06-11 | 中国电子科技集团公司第二十九研究所 | Information transmission scheduling method and system based on intelligent traffic classification |
-
2021
- 2021-11-04 CN CN202111301688.3A patent/CN114138463B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2004105355A1 (en) * | 2003-05-21 | 2004-12-02 | Nitgen Technologies Co., Ltd. | Intelligent traffic management system for networks and intelligent traffic management method using the same |
CN107864189A (en) * | 2017-10-18 | 2018-03-30 | 南京邮电大学 | A kind of application layer traffic load-balancing method based on DPI |
CN109729017A (en) * | 2019-03-14 | 2019-05-07 | 哈尔滨工程大学 | A kind of load-balancing method based on DPI prediction |
CN110784555A (en) * | 2019-11-07 | 2020-02-11 | 中电福富信息科技有限公司 | Intelligent monitoring and load scheduling method based on deep learning |
CN112949739A (en) * | 2021-03-17 | 2021-06-11 | 中国电子科技集团公司第二十九研究所 | Information transmission scheduling method and system based on intelligent traffic classification |
Non-Patent Citations (1)
Title |
---|
基于网络探针的流量均衡VOD系统及仿真;蒋亚军;杨震伦;詹增荣;;浙江工业大学学报;20110415(第02期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN114138463A (en) | 2022-03-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10838839B2 (en) | Optimizing adaptive monitoring in resource constrained environments | |
US11048444B2 (en) | Grouping devices as a virtual device for providing better quality of device data | |
CN101421953A (en) | Control service capacity | |
US11593735B2 (en) | Automated and efficient personal transportation vehicle sharing | |
CN106803799B (en) | Performance test method and device | |
CN109842563A (en) | Content delivery network flow dispatching method, device and computer readable storage medium | |
JP2006048702A (en) | Automatic configuration of transaction-based performance model | |
Battula et al. | A generic stochastic model for resource availability in fog computing environments | |
CN114138463B (en) | Method for predicting load balance of spot system application layer based on deep neural network | |
CN117561517A (en) | Computer-implemented apparatus and method for predicting traffic conditions in a route planning application | |
Shestak et al. | A stochastic approach to measuring the robustness of resource allocations in distributed systems | |
KR101639108B1 (en) | Automatic traffic generation for a faring system | |
JP2022074133A (en) | Computing device, computer-implemented method and computer-readable storage medium for multivariate time series modeling and forecasting (probabilistic nonlinear relationships across multiple time series and external factors for improved multivariate time series modeling and forecasting) | |
CN112287503B (en) | Dynamic space network construction method for traffic demand prediction | |
CN106575241A (en) | Mobile and remote runtime integration | |
CN108021607A (en) | A kind of wireless city Audit data off-line analysis method based on big data platform | |
US10515381B2 (en) | Spending allocation in multi-channel digital marketing | |
US20220327650A1 (en) | Transportation bubbling at a ride-hailing platform and machine learning | |
US11526800B2 (en) | Determining value of corpora for machine learning using coresets | |
US20170299640A1 (en) | Identifying and Prioritizing Errors in Connectivity Models of Distribution Networks for Efficient Field Inspections | |
KR102084920B1 (en) | Apparatus and method for predicting operating hours of a neighborhood living facility | |
Chafai et al. | Towards performance evaluation of requests satisfaction system in the cloud environment | |
Gupta et al. | Fractional difference based hybrid model for resource prediction in cloud network | |
CN112163932A (en) | Malicious seat occupying order identification method and device and electronic equipment | |
LU503318B1 (en) | Edge prediction method based on deep autoregressive recurrent neural network |
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 |