CN106529820A - Operation index prediction method and system - Google Patents
Operation index prediction method and system Download PDFInfo
- Publication number
- CN106529820A CN106529820A CN201611041426.7A CN201611041426A CN106529820A CN 106529820 A CN106529820 A CN 106529820A CN 201611041426 A CN201611041426 A CN 201611041426A CN 106529820 A CN106529820 A CN 106529820A
- Authority
- CN
- China
- Prior art keywords
- operation indicator
- data
- autoencoder network
- input
- stack
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
- G06Q10/06375—Prediction of business process outcome or impact based on a proposed change
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Educational Administration (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an operation index prediction method. The method comprises the steps of constructing a stack type autoencoder network by utilizing a layer-by-layer greed training method; forming a historical operation index data set by historical operation index data corresponding to predetermined operation indexes; and inputting the historical operation index data set to the stack type autoencoder network so as to obtain operation index prediction values corresponding to the predetermined operation indexes. According to the prediction method, the prediction values of the predetermined operation indexes at next moment can be accurately output through the stack type autoencoder network, so that a data basis is provided for enterprise operation decision-making; and the stack type autoencoder network can better capture information among different hierarchies, so that the addition of a large amount of rules is avoided, the dependency of researchers on experience is reduced, and the model training is more direct and objective. The invention discloses an operation index prediction system, which has the abovementioned beneficial effects.
Description
Technical field
The present invention relates to technical field of data processing, the Forecasting Methodology and system of more particularly to a kind of operation indicator.
Background technology
With developing for each company, company operation administration section needs following operation situation of prediction company, from
And predict development of company trend.To strengthen company operation operating capability, lifting company operation efficiency, benefit, and then take precautions against operation
Risk provides data and supports.Company operation administration section (such as Guo Wangyunjian centers in electrical network) is proposed in company operation index
On the basis of historical data, operation situation following to company is predicted, early warning, supports during company operation to following operation
The anticipation of risk and the prediction of development of company trend.The prediction to company operation index how is realized, is those skilled in the art
The technical issues that need to address.
The content of the invention
It is an object of the invention to provide the Forecasting Methodology and system of a kind of operation indicator, can by stack autoencoder network
The predictive value in subsequent time of predetermined operation indicator is accurately exported, and data foundation is provided for enterprise operation decision-making.
To solve above-mentioned technical problem, the present invention provides a kind of Forecasting Methodology of operation indicator, including：
According to the training data of input, stack autoencoder network is built using successively greedy training method；
By the corresponding history operation indicator data history of forming operation indicator data set of each predetermined operation indicator of input；
The history operation indicator data set is input into into the stack autoencoder network, each predetermined operation indicator is obtained
Corresponding operation indicator predictive value.
Optionally, the training data according to input, builds stack autoencoder network, bag using successively greedy training method
Include：
The number of plies of the hidden layer of stack autoencoder network is determined according to tuning parameter；
The sparse self-encoding encoder of each hidden layer of the stack autoencoder network is trained successively using successively greedy training method；
Output layer is carried out into logistic recurrence input training using sigmoid activation primitives and obtains Logistic recurrence
Layer.
Optionally, the training data according to input, after successively greedy training method builds stack autoencoder network,
Also include：
Renewal is finely adjusted to each layer parameter of the stack autoencoder network using back-propagation algorithm.
Optionally, the training data and the history operation indicator data are the data after being normalized.
Optionally, the history operation indicator data set is input into into the stack autoencoder network, obtains each described predetermined
The corresponding operation indicator predictive value of operation indicator, including：
The history operation indicator data set is input into into the stack autoencoder network, each predetermined operation indicator is obtained
Corresponding normalization operation indicator predictive value；
The normalization operation indicator predictive value is carried out into normalized, each predetermined operation indicator correspondence is obtained
Operation indicator predictive value.
Optionally, the Forecasting Methodology also includes：
The stack autoencoder network is updated according to predetermined period.
The present invention also provides a kind of prognoses system of operation indicator, including：
Model construction module, for the training data according to input, builds stack using successively greedy training method self-editing
Code network；
Data input module, for the corresponding history operation indicator data history of forming of each predetermined operation indicator that will be input into
Operation indicator data set；
Prediction module, for the history operation indicator data set is input into the stack autoencoder network, obtains each institute
State the corresponding operation indicator predictive value of predetermined operation indicator.
Optionally, the model construction module includes：
Number of plies determining unit, for the number of plies of the hidden layer of stack autoencoder network is determined according to tuning parameter；
Greedy training unit, for utilizing successively greedy training method to train successively, the stack autoencoder network is each to be hidden
The sparse self-encoding encoder of layer；
Output layer training unit, returns input instruction for output layer is carried out logistic using sigmoid activation primitives
Get Logistic and return layer.
Optionally, the model construction module also includes：
Fine-adjusting unit, for being finely adjusted more to each layer parameter of the stack autoencoder network using back-propagation algorithm
Newly.
Optionally, the prognoses system also includes：
Normalization unit, for the training data and the history operation indicator data are normalized；
Normalization unit is gone, for normalization operation indicator predictive value is carried out normalized, obtains each described pre-
Determine the corresponding operation indicator predictive value of operation indicator.
A kind of Forecasting Methodology of operation indicator provided by the present invention, including：According to the training data of input, using successively
Greedy training method builds stack autoencoder network；By the corresponding history operation indicator data shape of each predetermined operation indicator of input
Into history operation indicator data set；The history operation indicator data set is input into into the stack autoencoder network, each institute is obtained
State the corresponding operation indicator predictive value of predetermined operation indicator；
It can be seen that, the Forecasting Methodology by stack autoencoder network accurately can export predetermined operation indicator in lower a period of time
The predictive value at quarter, provides data foundation for enterprise operation decision-making, and stack autoencoder network more preferably can be caught between different levels
Information, while avoiding the additions of a large amount of rules, alleviate dependence of the research worker to experience, make the training of model more straight
Connect, it is objective；Present invention also offers a kind of prognoses system of operation indicator, with above-mentioned beneficial effect, will not be described here.
Description of the drawings
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
Accompanying drawing to be used needed for having technology description is briefly described, it should be apparent that, drawings in the following description are only this
Inventive embodiment, for those of ordinary skill in the art, on the premise of not paying creative work, can be with basis
The accompanying drawing of offer obtains other accompanying drawings.
The shallow-layer learning model schematic diagram without hidden layer that Fig. 1 is provided by the embodiment of the present invention；
The shallow-layer learning model schematic diagram containing single hidden layer that Fig. 2 is provided by the embodiment of the present invention；
The deep learning model schematic containing multiple hidden layers that Fig. 3 is provided by the embodiment of the present invention；
The flow chart of the Forecasting Methodology of the operation indicator that Fig. 4 is provided by the embodiment of the present invention；
Fig. 5 trains schematic diagram by the ground floor self-encoding encoder that the embodiment of the present invention is provided；
Fig. 6 trains schematic diagram by the second layer self-encoding encoder that the embodiment of the present invention is provided；
Fig. 7 trains schematic diagram by the output layer that the embodiment of the present invention is provided；
The stack autoencoder network schematic diagram of the structure that Fig. 8 is provided by the embodiment of the present invention；
The single task stack autoencoder network schematic diagram of the structure that Fig. 9 is provided by the embodiment of the present invention；
The multitask stack autoencoder network schematic diagram of the structure that Figure 10 is provided by the embodiment of the present invention；
The structured flowchart of the prognoses system of the operation indicator that Figure 11 is provided by the embodiment of the present invention.
Specific embodiment
The core of the present invention is to provide a kind of Forecasting Methodology and system of operation indicator, can by stack autoencoder network
The predictive value in subsequent time of predetermined operation indicator is accurately exported, and data foundation is provided for enterprise operation decision-making.
To make purpose, technical scheme and the advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
The a part of embodiment of the present invention, rather than the embodiment of whole.Based on the embodiment in the present invention, those of ordinary skill in the art
The every other embodiment obtained under the premise of creative work is not made, belongs to the scope of protection of the invention.
The existing Forecasting Methodology species based on mathematical modeling is various, from classical unit consumption method, elastic coefficient method, statistical
Analysis method, grey method up till now, time series forecasting, Differential Equation Model etc..Wherein, grey forecasting model is being processed
Less characteristic value data, it is not necessary to which the sample space of data is sufficiently large, can just solve that historical data is few, sequence integrity with
And the problem that reliability is low, irregular initial data can be carried out generating and obtain the stronger formation sequence of rule.But only it is suitable for
In the prediction of middle or short term, it is only suitable for being similar to the prediction of exponential increase.Time series forecasting ought run into the larger change of extraneous generation
Change, often have relatively large deviation, time series forecasting is better than the effect of long-term forecast for the effect of applying.It is micro-
Divide equation model reaction rule and its internal relation inside the matters, but as the foundation of equation is false with the independence of local rule
It is set to basis, when as long-term forecast, error is larger, and the solution of the differential equation is relatively difficult to obtain.
Therefore, the present embodiment is to improve the precision of prediction and convenience degree of operation indicator, using deep neural network
Mode carry out model construction and carry out the prediction of operation indicator.Wherein, deep learning comes from the research of artificial neural network.Contain
The multilayer perceptron of many hidden layers is exactly a kind of deep learning structure.Deep learning forms more abstract by combining low-level feature
It is high-rise to represent attribute classification or feature, to find that the distributed nature of data is represented.Deep learning gives deep neural network
Effectively pre-training algorithm allows it to play depth structure ability to express by force, represents succinct advantage.On the other hand, depth
Practise can pass through it is substantial amounts of learn most reflect the essential abstract characteristics of data without label data, it is and abstract with these high levels
Feature is classified and prediction task, improves task accuracy rate and model performance.Deep neural network can simulate complicated function, right
For the complicated level prediction of hierarchical structure, the information between different levels can be more preferably caught, while avoiding a large amount of rules
Addition, alleviate dependence of the research worker to experience, make the training of model more direct, objective.
Fig. 1, Fig. 2, Fig. 3 are the graph model schematic diagram of shallow-layer neutral net and deep-neural-network.Wherein, neutral net or
Claim artificial neural network (Artificial Neural Networks, be abbreviated as ANNs) be also referred to as neutral net (NNs) or
Referred to as link model (Connection Model), it be it is a kind of imitate animal nerve network behavior feature, carry out it is distributed simultaneously
The algorithm mathematics model of row information process.Complexity of this network by system, by adjusting between internal great deal of nodes
The relation being connected with each other, so as to reach the purpose of processing information.
The present embodiment on the basis of company operation metric history data, using deep learning method, by building stack
This deep neural network of autoencoder network, with the historical data of related operation indicator as input, output specific indexes are next
The predictive value at moment.So as to obtain accurate operation indicator predictive value, foundation is provided for company operation management.Figure is refer to specifically
The flow chart of the Forecasting Methodology of 4, Fig. 4 operation indicators provided by the embodiment of the present invention；The Forecasting Methodology can include：
S100, according to the training data of input, build stack autoencoder network using successively greedy training method；
Specifically, the training data in the present embodiment is the predetermined prediction index data of history selected.The number of training data
The species of the operation indicator that amount and species and company needs are predicted is relevant.That is training data and selected predetermined operation indicator
It is corresponding.The selected training data of every kind of index number it is relevant with the accuracy of stack autoencoder network that training is obtained.
User can be selected according to system hardware level, and required accuracy.
Specifically, due to, directly in deep neural network using back-propagation algorithm training network, fall into may parameter
Enter the problem of local extremum, or as the deficiency for having label data causes over-fitting problem.Therefore, the present embodiment sets up stack
During own coding neutral net using successively greedy training method training network successively each layer, and then pre-training entire depth
Neutral net.It is specific as follows using the principle that successively greedy training method builds stack autoencoder network (referred to as network)：
Using successively greedy coaching method, first with the ground floor that carrys out training network is originally inputted, its parameter W is obtained(1,1),W(1,2),b(1,1),b(1,2)；Then network first tier will be originally inputted and transform into the vector (vacation being made up of hidden unit activation value
If the vector is A), A is continued parameter W that training obtains the second layer as the input of the second layer then(2,1),W(2,2),b(2,1),
b(2,2)；Finally, the strategy for each layer below equally being adopted, will the output of front layer instructed as the mode of next layer of input successively
Practice.I.e. optional, building stack autoencoder network using successively greedy training method can include：
The number of plies of the hidden layer of stack autoencoder network is determined according to tuning parameter；
The sparse self-encoding encoder of each hidden layer of the stack autoencoder network is trained successively using successively greedy training method；
Output layer is carried out into logistic recurrence input training using sigmoid activation primitives and obtains Logistic recurrence
Layer.
Specifically, hidden layer hxQuantity need by tuning parameter determine, i.e., tuning parameter determines stack own coding net
The depth of network.Output layer returns (i.e. using with concealed nodes identical sigmoid activation primitive) using logistic.
Wherein, due to being employed herein sigmoid activation primitives, its form is taken for f (z)=1/1+exp (- z) ' visible its
Value scope is [0,1], therefore, the activation value of output node is between 0 to 1, so the node of input layer and output layer
Value is also limited between [0,1].Therefore, before training and using deep neural network, need to input vector X and output
Each value { x of vectorial Yi}、{yiBe normalized, afterwards using normalization afterNetwork is carried out
Training, prediction.For the predictive value of network outputGo back according to rule reduction during normalized, you can obtain true
During the predictive value, such as normalization of realityWhen then reducing
It is i.e. optional, (letter is activated using with concealed nodes identical sigmoid when output layer is returned using logistic
Number) when, the training data and the history operation indicator data are the data after being normalized.Therefore history operation
The data that achievement data is concentrated are also the data after normalization.It is i.e. optional, the history operation indicator data set is input into into institute
Stack autoencoder network is stated, the corresponding normalization operation indicator predictive value of each predetermined operation indicator is obtained；By the normalizing
Changing operation indicator predictive value carries out normalized, obtains the corresponding operation indicator predictive value of each predetermined operation indicator.
The building process of a stack autoencoder network with 2 layers of hidden layer is exemplified below：
First, with being originally inputted x(k)First self-encoding encoder of training, it can learn the single order feature for obtaining being originally inputted
Represent(as shown in Figure 5).
Then, initial data is input in the above-mentioned sparse self-encoding encoder for training, for each input x(k), all
Its corresponding single order character representation can be obtainedThen you are again with these single order features as another sparse self-encoding encoder
Input, use them to learn second order feature(as shown in Figure 6).
Equally, then single order feature it is input in the sparse self-encoding encoder of the second layer for just having trained, obtains eachCorrespondence
Second order feature activation valueNext, these second order features are returned as logistic being input into, training obtains an energy
By the model of second order Feature Mapping to serial number, (as shown in Figure 7).
Finally, this three-layered node is built into one altogether layer is returned comprising two hidden layers and a final Logistic
Stack autoencoder network, this network can export its corresponding predictive value Y, (as shown in Figure 8) according to input data X.
Further, for above-mentioned training method, when each layer parameter is trained, other each layer parameters can be fixed and is kept
It is constant.So the performance in order to lift a stack own coding neutral net, is preferably predicted the outcome.Above-mentioned pre-
After training process is completed, back-propagation algorithm can be passed through while all layers of parameter is adjusted to improve result, in stack certainly
" fine setting (fine-tuning) " can be carried out in the building process of coding network.It is i.e. preferred, also include：Using back propagation
Algorithm is finely adjusted renewal to each layer parameter of the stack autoencoder network.Fine setting is the conventional strategy in deep learning, can be with
The performance of a stack own coding neutral net is substantially improved.From for higher visual angle, finely tune stack own coding god
All layers of Jing networks are considered as a model, and so in each iteration, in network, all of weighted value can be optimised.
Fine setting to stack autoencoder network can be carried out as follows：
Once feedovered transmission, to L2Layer, L3Layer is until output layer Lnl, defined in back-propagation algorithm step
Formula calculates the activation value (exciter response) on each layer.
To output layer, order
To l=nl-1,nl- 2 ..., 2, make δ(l)=-((W(l))Tδ(l+1))f'(z(l))。
Partial derivative required for calculating：
S110, the corresponding history operation indicator data history of forming operation indicator data of each predetermined operation indicator that will be input into
Collection；
S120, the history operation indicator data set is input into into the stack autoencoder network, obtains each predetermined fortune
The corresponding operation indicator predictive value of battalion's index.
Specifically, for given history operation indicator data set { I1,I2... as the defeated of above-mentioned stack autoencoder network
Enter, the corresponding operation indicator of each described predetermined operation indicator of subsequent period can be obtained according to the output of stack autoencoder network
Predictive value.Assume that current time is T, need to predict one or more desired values at T+1 moment.Stack after wherein building is certainly
The schematic diagram of coding network may be referred to Fig. 9, Figure 10.Its rule selected, can determine according to the quantity of predetermined operation indicator
As single task or multitask, the schematic diagram of the stack autoencoder network of single task may be referred to Fig. 9, and the stack of multitask is certainly
The schematic diagram of coding network may be referred to Figure 10.Wherein, DNN (i.e. deep neural network, Deep Neural Network) this reality
It is stack own coding neutral net in applying example.
Specifically, for output layer Y, when carrying out single dependent variable prediction, the stack own coding net of Fig. 9 structures can be set up
Network, output layer only one of which node Y represent the predictive value of index to be predicted at the T+1 momentCarry out multivariate response pre-
During survey, the structure such as Figure 10 is set up, output layer there are multiple nodes, when representative needs the T+1 of multiple indexs of prediction simultaneously respectively
The predictive value at quarter
For input layer X, comprising multiple input node { x1,x2..., input value can include multi-class data, for example can be with
Including the value { y at the front k moment of the index y of predictionT-k+1,yT-k+2,…,yT, or including y and other all indexs before
The value at k momentOr other related input sources, finally need to tie by the experiment of model
Fruit determines suitable input.The present embodiment is not defined to specific input variable number and value type, as long as in instruction
It is trained using corresponding training data when practicing stack autoencoder network.
The prediction of operation indicator is carried out using stack autoencoder network, expression ratio can be carried out in the way of more compact shallow
The much bigger function set of layer network.For example, these functions can compactly be expressed with k layer networks (here be succinctly
Refer to that the number of Hidden unit only need to be with input block number in polynomial relation).But for one only have k-1 layers network and
Speech, unless it otherwise can not succinctly express these functions using the Hidden unit number having exponent relation with input block number.
The present embodiment adopts deep learning method, on the basis of company operation metric history data, by building stack
This deep neural network of autoencoder network, after which is trained and is finely tuned, with the historical data of index of correlation as input
It is predicted, exports predictive value of the specific indexes in subsequent time.User can be transported to formula according to the prediction index value of output
The business realizing comprehensive monitorings such as battalion, i.e., around company's main business activity and core resource, referred to by building monitoring model, combing
The modes such as mark system, setting metrics-thresholds, realize to company's external environment condition, comprehensive performance, operation situation, core resource, key
24 hours online dynamic comprehensive monitorings of the aspects such as flow process, the in time unusual fluctuation during discovery company operation and problem are simultaneously warned
Show.It is exactly existing to enterprise operation and management for operation diagnosis, i.e. enterprise diagnosis can also being carried out to corporate business according to prediction index value
Shape is investigated and analysed, and finds problem present in operation, with quantitatively, qualitatively analysis method finds out the original of generation problem
Cause, proposes practicable improvement project, and then instructs which to implement, to improve Business Economic Benefit.
Further, in order to keep constantly the reliability of stack autoencoder network, need to be needed in time to stack according to user
Autoencoder network is updated.User actively can be updated, it is also possible to which setting the update cycle is updated, or in detection
When having new training data to generate, it is updated automatically.It is i.e. optional, according to predetermined period to the stack autoencoder network
It is updated.For example, training etc. is updated according to the training data for newly obtaining.
Based on above-mentioned technical proposal, the Forecasting Methodology of the operation indicator that the embodiment of the present invention is carried, the Forecasting Methodology pass through stack
Formula autoencoder network can accurately export the predictive value in subsequent time of predetermined operation indicator, provide for enterprise operation decision-making
Data foundation, and stack autoencoder network can more preferably catch the information between different levels, while avoiding adding for a large amount of rules
Plus, dependence of the research worker to experience is alleviated, makes the training of model more direct, objective.
Below the prognoses system of operation indicator provided in an embodiment of the present invention is introduced, operation indicator described below
The Forecasting Methodology of prognoses system and above-described operation indicator can be mutually to should refer to.
Refer to Figure 11, the structured flowchart of the prognoses system of the operation indicator that Figure 11 is provided by the embodiment of the present invention；Should
Prognoses system can include：
Model construction module 100, for the training data according to input, builds stack certainly using successively greedy training method
Coding network；
Data input module 200, for the corresponding history operation indicator data of each predetermined operation indicator of input are formed
History operation indicator data set；
Prediction module 300, for the history operation indicator data set is input into the stack autoencoder network, obtains each
The corresponding operation indicator predictive value of the predetermined operation indicator.
Based on above-described embodiment, the model construction module 100 includes：
Number of plies determining unit, for the number of plies of the hidden layer of stack autoencoder network is determined according to tuning parameter；
Greedy training unit, for utilizing successively greedy training method to train successively, the stack autoencoder network is each to be hidden
The sparse self-encoding encoder of layer；
Output layer training unit, returns input instruction for output layer is carried out logistic using sigmoid activation primitives
Get Logistic and return layer.
Based on above-described embodiment, the model construction module 100 also includes：
Fine-adjusting unit, for being finely adjusted more to each layer parameter of the stack autoencoder network using back-propagation algorithm
Newly.
Based on above-described embodiment, the prognoses system also includes：
Normalization unit, for the training data and the history operation indicator data are normalized；
Normalization unit is gone, for normalization operation indicator predictive value is carried out normalized, obtains each described pre-
Determine the corresponding operation indicator predictive value of operation indicator.
Based on above-mentioned any embodiment, the prognoses system also includes：
Update module, for being updated to the stack autoencoder network according to predetermined period.
In description, each embodiment is described by the way of progressive, and what each embodiment was stressed is and other realities
Apply the difference of example, between each embodiment identical similar portion mutually referring to.For device disclosed in embodiment
Speech, as which corresponds to the method disclosed in Example, so description is fairly simple, related part is referring to method part illustration
.
Professional further appreciates that, with reference to the unit of each example of the embodiments described herein description
And algorithm steps, can with electronic hardware, computer software or the two be implemented in combination in, in order to clearly demonstrate hardware and
The interchangeability of software, generally describes the composition and step of each example in the above description according to function.These
Function actually with hardware or software mode performing, depending on the application-specific and design constraint of technical scheme.Specialty
Technical staff can use different methods to realize described function to each specific application, but this realization should not
Think beyond the scope of this invention.
The step of method described with reference to the embodiments described herein or algorithm, directly can be held with hardware, processor
Capable software module, or the combination of the two is implementing.Software module can be placed in random access memory (RAM), internal memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, depositor, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
Above the Forecasting Methodology and system of operation indicator provided by the present invention are described in detail.It is used herein
Specific case is set forth to the principle and embodiment of the present invention, and the explanation of above example is only intended to help and understands
The method of the present invention and its core concept.It should be pointed out that for those skilled in the art, without departing from this
On the premise of inventive principle, some improvement and modification can also be carried out to the present invention, these improve and modification also falls into the present invention
In scope of the claims.
Claims (10)
1. a kind of Forecasting Methodology of operation indicator, it is characterised in that include：
According to the training data of input, stack autoencoder network is built using successively greedy training method；
By the corresponding history operation indicator data history of forming operation indicator data set of each predetermined operation indicator of input；
The history operation indicator data set is input into into the stack autoencoder network, each predetermined operation indicator correspondence is obtained
Operation indicator predictive value.
2. the Forecasting Methodology of operation indicator according to claim 1, it is characterised in that according to the training data of input, profit
Stack autoencoder network is built with successively greedy training method, including：
The number of plies of the hidden layer of stack autoencoder network is determined according to tuning parameter；
The sparse self-encoding encoder of each hidden layer of the stack autoencoder network is trained successively using successively greedy training method；
Output layer is carried out into logistic recurrence input training using sigmoid activation primitives and obtains Logistic recurrence layers.
3. the Forecasting Methodology of operation indicator according to claim 2, it is characterised in that according to the training data of input, profit
After successively greedy training method builds stack autoencoder network, also include：
Renewal is finely adjusted to each layer parameter of the stack autoencoder network using back-propagation algorithm.
4. the Forecasting Methodology of operation indicator according to claim 3, it is characterised in that the training data and the history
Operation indicator data are the data after being normalized.
5. the Forecasting Methodology of operation indicator according to claim 4, it is characterised in that by the history operation indicator data
The collection input stack autoencoder network, obtains the corresponding operation indicator predictive value of each predetermined operation indicator, including：
The history operation indicator data set is input into into the stack autoencoder network, each predetermined operation indicator correspondence is obtained
Normalization operation indicator predictive value；
The normalization operation indicator predictive value is carried out into normalized, the corresponding fortune of each predetermined operation indicator is obtained
Battalion's index prediction value.
6. the Forecasting Methodology of the operation indicator according to any one of claim 1-5, it is characterised in that also include：
The stack autoencoder network is updated according to predetermined period.
7. a kind of prognoses system of operation indicator, it is characterised in that include：
Model construction module, for the training data according to input, builds stack own coding net using successively greedy training method
Network；
Data input module, for each predetermined operation indicator of input corresponding history operation indicator data history of forming is runed
Achievement data collection；
Prediction module, for the history operation indicator data set is input into the stack autoencoder network, obtains each described pre-
Determine the corresponding operation indicator predictive value of operation indicator.
8. the prognoses system of operation indicator according to claim 7, it is characterised in that the model construction module includes：
Number of plies determining unit, for the number of plies of the hidden layer of stack autoencoder network is determined according to tuning parameter；
Greedy training unit, for utilizing successively greedy training method to train each hidden layer of the stack autoencoder network successively
Sparse self-encoding encoder；
Output layer training unit, trains for output layer is carried out logistic recurrence inputs using sigmoid activation primitives
Layer is returned to Logistic.
9. the prognoses system of operation indicator according to claim 8, it is characterised in that the model construction module is also wrapped
Include：
Fine-adjusting unit, for being finely adjusted renewal to each layer parameter of the stack autoencoder network using back-propagation algorithm.
10. the prognoses system of operation indicator according to claim 9, it is characterised in that also include：
Normalization unit, for the training data and the history operation indicator data are normalized；
Normalization unit is gone, for normalization operation indicator predictive value is carried out normalized, each predetermined fortune is obtained
The corresponding operation indicator predictive value of battalion's index.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611041426.7A CN106529820A (en) | 2016-11-21 | 2016-11-21 | Operation index prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611041426.7A CN106529820A (en) | 2016-11-21 | 2016-11-21 | Operation index prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106529820A true CN106529820A (en) | 2017-03-22 |
Family
ID=58356423
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611041426.7A Pending CN106529820A (en) | 2016-11-21 | 2016-11-21 | Operation index prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106529820A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107580329A (en) * | 2017-10-20 | 2018-01-12 | 北京神州泰岳软件股份有限公司 | A kind of network analysis optimization method and device |
| CN108471353A (en) * | 2018-01-22 | 2018-08-31 | 浙江鹏信信息科技股份有限公司 | A method of the NE capacity analysis based on deep neural network algorithm and prediction |
| CN109313673A (en) * | 2018-04-17 | 2019-02-05 | 深圳鲲云信息科技有限公司 | The operation method and Related product of network model |
| CN109670572A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670570A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670568A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670571A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670589A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670566A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670587A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670590A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670567A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670569A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670588A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN111832993A (en) * | 2020-07-13 | 2020-10-27 | 深圳市今天国际物流技术股份有限公司 | Predictive maintenance method for warehouse logistics system and related components |
| CN111861073A (en) * | 2019-04-30 | 2020-10-30 | 北京嘀嘀无限科技发展有限公司 | Operation strategy prediction method and device, electronic equipment and readable storage medium |
| US11663679B2 (en) | 2019-10-11 | 2023-05-30 | International Business Machines Corporation | Generating mode change alerts with automatic detection from sensor data |
| US11868932B2 (en) | 2020-09-30 | 2024-01-09 | International Business Machines Corporation | Real-time opportunity discovery for productivity enhancement |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104200398A (en) * | 2014-08-28 | 2014-12-10 | 国家电网公司 | Information processing method and device of service terminal |
| CN105096041A (en) * | 2015-07-24 | 2015-11-25 | 北京中电普华信息技术有限公司 | Index change traceability and prediction method and apparatus |
-
2016
- 2016-11-21 CN CN201611041426.7A patent/CN106529820A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104200398A (en) * | 2014-08-28 | 2014-12-10 | 国家电网公司 | Information processing method and device of service terminal |
| CN105096041A (en) * | 2015-07-24 | 2015-11-25 | 北京中电普华信息技术有限公司 | Index change traceability and prediction method and apparatus |
Non-Patent Citations (2)
| Title |
|---|
| 杨逍: "机器学习中隐式因子模型及其优化算法", 《中国优秀硕士学位论文全文数据库》 * |
| 樊重俊等: "《大数据分析与应用》", 31 January 2016, 立信会计出版社 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670590A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670588A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670566A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670572A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670570A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670568A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670571A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670587A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670569A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670567A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670589A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN107580329A (en) * | 2017-10-20 | 2018-01-12 | 北京神州泰岳软件股份有限公司 | A kind of network analysis optimization method and device |
| CN108471353A (en) * | 2018-01-22 | 2018-08-31 | 浙江鹏信信息科技股份有限公司 | A method of the NE capacity analysis based on deep neural network algorithm and prediction |
| CN108471353B (en) * | 2018-01-22 | 2021-03-30 | 浙江鹏信信息科技股份有限公司 | Network element capacity analysis and prediction method based on deep neural network algorithm |
| CN109313673A (en) * | 2018-04-17 | 2019-02-05 | 深圳鲲云信息科技有限公司 | The operation method and Related product of network model |
| CN111861073B (en) * | 2019-04-30 | 2024-05-03 | 北京嘀嘀无限科技发展有限公司 | Operation policy prediction method and device, electronic equipment and readable storage medium |
| CN111861073A (en) * | 2019-04-30 | 2020-10-30 | 北京嘀嘀无限科技发展有限公司 | Operation strategy prediction method and device, electronic equipment and readable storage medium |
| US11663679B2 (en) | 2019-10-11 | 2023-05-30 | International Business Machines Corporation | Generating mode change alerts with automatic detection from sensor data |
| CN111832993A (en) * | 2020-07-13 | 2020-10-27 | 深圳市今天国际物流技术股份有限公司 | Predictive maintenance method for warehouse logistics system and related components |
| US11868932B2 (en) | 2020-09-30 | 2024-01-09 | International Business Machines Corporation | Real-time opportunity discovery for productivity enhancement |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106529820A (en) | Operation index prediction method and system | |
| Svalina et al. | An adaptive network-based fuzzy inference system (ANFIS) for the forecasting: The case of close price indices | |
| Guan et al. | An adaptive neuro-fuzzy inference system based approach to real estate property assessment | |
| CN104270281B (en) | A kind of Web Service QoS combination forecasting methods | |
| CN108921298B (en) | Multi-agent communication and decision-making method for reinforcement learning | |
| CN102622515B (en) | A kind of weather prediction method | |
| CN102622418B (en) | Prediction device and equipment based on BP (Back Propagation) nerve network | |
| CN107506590A (en) | A kind of angiocardiopathy forecast model based on improvement depth belief network | |
| CN107346464A (en) | Operational indicator Forecasting Methodology and device | |
| CN104503420B (en) | Non-linear process industry fault prediction method based on novel FDE-ELM and EFSM | |
| Babbar-Sebens et al. | A Case-Based Micro Interactive Genetic Algorithm (CBMIGA) for interactive learning and search: Methodology and application to groundwater monitoring design | |
| CN106650920A (en) | Prediction model based on optimized extreme learning machine (ELM) | |
| Yang et al. | An optimized BP neural network model for teaching management evaluation | |
| CN115099519B (en) | Oil well yield prediction method based on multi-machine learning model fusion | |
| CN111489259A (en) | Stock market risk prediction intelligent implementation method based on deep learning | |
| CN109242558A (en) | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based | |
| CN113935513A (en) | CEEMDAN-based short-term power load prediction method | |
| CN114817571B (en) | Dynamic knowledge graph-based achievement quoted quantity prediction method, medium and equipment | |
| CN114565021A (en) | Financial asset pricing method, system and storage medium based on quantum circulation neural network | |
| CN115114128A (en) | Satellite health state evaluation system and evaluation method | |
| CN108768750A (en) | Communication network failure localization method and device | |
| Robati et al. | Inflation rate modeling: Adaptive neuro-fuzzy inference system approach and particle swarm optimization algorithm (ANFIS-PSO) | |
| CN110310199B (en) | Method and system for constructing loan risk prediction model and loan risk prediction method | |
| Gong et al. | Optimization of Local Prediction Algorithm of Financial Time Series Based on Fuzzy Neural Network | |
| CN116341752A (en) | Collaborative supply chain prediction method based on graph neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170322 |
|
| RJ01 | Rejection of invention patent application after publication |