CN106126896A - The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth and system - Google Patents

The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth and system Download PDF

Info

Publication number
CN106126896A
CN106126896A CN201610444160.4A CN201610444160A CN106126896A CN 106126896 A CN106126896 A CN 106126896A CN 201610444160 A CN201610444160 A CN 201610444160A CN 106126896 A CN106126896 A CN 106126896A
Authority
CN
China
Prior art keywords
wind speed
forecasting
prediction
degree
intrinsic mode
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201610444160.4A
Other languages
Chinese (zh)
Other versions
CN106126896B (en
Inventor
陈分雄
胡凯
凌承昆
唐曜曜
毛中杰
王典洪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Geosciences
Original Assignee
China University of Geosciences
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China University of Geosciences filed Critical China University of Geosciences
Priority to CN201610444160.4A priority Critical patent/CN106126896B/en
Publication of CN106126896A publication Critical patent/CN106126896A/en
Application granted granted Critical
Publication of CN106126896B publication Critical patent/CN106126896B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
    • G16Z99/00—Subject matter not provided for in other main groups of this subclass

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a kind of mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth and system, the method comprises the following steps: original wind speed time series is decomposed by S1, rule of thumb mode decomposition, obtains multiple intrinsic mode function；S2, each intrinsic mode function is built respective training dataset and test data set；S3, training data is concentrated each intrinsic mode function training sample send into stack coding network be trained, obtain respective forecasting wind speed submodel；S4, test data set is sent to each self-corresponding forecasting wind speed submodel it is predicted, obtain the prediction output valve of each forecasting wind speed submodel；S5, the prediction output valve of each forecasting wind speed submodel is combined overlap-add procedure, obtains final overall prediction output valve.The present invention is effectively improved precision of prediction and the robustness of forecast model；Higher short-term wind speed forecasting precision can be obtained simultaneously.

Description

The mixed model wind speed forecasting method that learns based on empirical mode decomposition and the degree of depth and System
Technical field
The present invention relates to machine learning techniques field, particularly relate to a kind of based on mixing that empirical mode decomposition and the degree of depth learn Matched moulds type wind speed forecasting method and system.
Background technology
Along with going deep into of socioeconomic development and process of industrialization, China's energy and environment problem manifests day by day.One side Face, the development of China's industrialization and urbanization constantly accelerates, and entire society's economy can keep high speed to increase within the longer time Long.Meanwhile, the consumption of the energy also increases quickly, and more and more higher to the degree of dependence of electric power, and this has resulted in right The demand of electric power is increasing, and for electric power resource, its capacity is but limited.On the other hand, the economy of China by Difference between region disparate development very greatly.The past China energy resource structure mainly based on coal, and Large Copacity, Power transmission technology is not applied to certain scale at a distance, and the coal that Middle Eastern uses mostly passes through from the northwestward Highway and railway transportation, and the with serious pollution environment that destroys that fire coal is caused, owing to problem is accumulated over a long period, environment is dirty Dye constantly aggravation so that whole ecological environment is accelerated to deteriorate.In the long run, use electric energy to implement to substitute, greatly develop cleaning The energy is inexorable trend.Devoting Major Efforts To Developing and utilize clean energy resource for improve the living environment of people and realize economic society can Sustainable development has great strategic importance, is the long-range plan solving energy and environment problem.
Wind energy is a kind of reproducible clean energy resource, increasingly obtains the attention of people.Routinize to reduce in recent years The pollution to environment of the stone energy, China starts to greatly develop wind-power electricity generation.Due to wind-power electricity generation will not emission greenhouse gas, for Build low-carbon (LC) society and reply global warming, it is possible to play its important effect.Wind-power electricity generation is mainly by wind energy resources handle The kinetic energy of wind transfers electric energy to, although wind energy resources is a kind of cleaning, reproducible new forms of energy, but wind energy and wind facies close, Wind itself has certain undulatory property, and which results in wind energy also has certain undulatory property.So electricity tool produced by wind energy There is unstability, low during height when electricity there will be, situation about cutting in and out.In order to be better understood by following a period of time wind energy The situation of generating, it is simple to electrical network carries out a few thing, and wind-powered electricity generation prediction work is necessary.The prediction of wind speed and wind power is wind Establish by cable send out important step, accurately prediction can be wind-powered electricity generation generating enterprise reduce because of generated energy fluctuation or predict not The economic loss accurately caused, enterprise according to prediction case, can also preferably arrange its generation schedule.According further in advance Survey situation, it is also possible to find out wind little when, arrange some maintenance etc. of wind park.The most preferably improve wind-powered electricity generation at electricity Receiving ability in net, thus improve the development of exploitation level of wind energy resources.In order to promote the development of wind power technology, improve wind energy The utilization of resource, domestic and international research institution and wind-power electricity generation companies have developed some softwares and carry out wind speed and wind power prediction And achieve certain economic benefit.Therefore, improving forecasting wind speed accuracy is an extremely important and job highly significant.
The impact of the change natural cause such as climate, landform of wind speed, has the strongest undulatory property, in wind speed time series In jump by a relatively large margin often occurs, wind speed is the Nonlinear Time Series of a kind of high complexity.From time series angle From the point of view of, seasonal effect in time series change is affected by various factors and is presented tendency, periodicity and randomness, according to time series Complexity substantially can be divided into linear session sequence and Nonlinear Time Series.For the prediction of linear session sequence, It is to build autoregression (AR), slip average (MA), autoregressive moving average according to linear session sequence for traditional method (ARMA) or difference autoregressive moving average (ARIMA) model, for the modeling of Nonlinear Time Series, use the earliest It is the models such as Threshold Autoregressive (TAR), bilinearity (BL) and index autoregression (EAR).These traditional time series forecasting moulds Type is effective in the forecasting problem of Nonlinear Time Series, but they all lack universality, in the face of some are the most multiple During miscellaneous Nonlinear Time Series and people forecast demand the most accurately, traditional Time Series Forecasting Methods can not meet The prediction requirement of people.Degree of depth study is the dominant form of following machine learning development, and has become one new science in this century Research field.Many urgent problems are proposed in rationale and two aspects of engineering.The degree of depth study use by The greedy training algorithm of layer solves the training problem of conventional multilayer neutral net, and moreover degree of depth learning algorithm also imparts many The feature learning ability that layer neutral net is excellent, the feature that study obtains has more essential portraying to data.For processing complexity Problem of nonlinear mapping, deep neural network can show more powerful learning capacity.At present, degree of depth study exists Machine learning classification Study on Problems and actual application achieve breakthrough, and is the pre-of the Nonlinear Time Series such as wind speed Survey research and provide a new direction.Degree of depth study can improve time series forecasting essence by building multilayer neural network Degree, and the parameter amount increasing depth model can be in memory time sequence in more useful information, particularly time series Random high frequency composition characteristics.Meanwhile, how by degree of depth learning network and At All Other Times sequence prediction method to combine structure mixed Close forecast model, be also by one of degree of depth theory of learning research contents being used for time series forecasting.
Under normal conditions, due to the non-linear of wind speed and non-stationary property, it is used alone neutral net and degree of depth study Forecasting wind speed is carried out, it is difficult to improve the precision of prediction of model on model, in order to preferably learn the non-stationary feature of wind speed, right Wind speed time series carries out empirical mode decomposition (Empirical ModeDecomposition is called for short EMD), to reach to weaken Wind speed seasonal effect in time series non-stationary property.It is permissible that the mixed model using empirical mode decomposition and the degree of depth to learn carries out forecasting wind speed Strengthen the model learning capacity to plurality of target function, and be effectively improved precision of prediction and the robustness of algorithm.Meanwhile, by choosing Take the predictive value of different natural mode of vibration component to be combined optimization and can obtain higher short-term wind speed forecasting precision.
Summary of the invention
The technical problem to be solved in the present invention is the defect the highest for precision of prediction in prior art, it is provided that Yi Zhongneng Enough it is effectively improved the mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth of precision of prediction and robustness And system.
The technical solution adopted for the present invention to solve the technical problems is:
The present invention provides a kind of mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth, including with Lower step:
S1, obtain original wind speed time series, build empirical mode decomposition and the hybrid prediction model of degree of depth study, according to Original wind speed time series is decomposed by empirical mode decomposition, obtains multiple intrinsic mode function；
S2, each intrinsic mode function is built respective training dataset and test data set；
S3, training data is concentrated each intrinsic mode function training sample send into stack coding network be trained, To respective forecasting wind speed submodel；
S4, test data set is sent to each self-corresponding forecasting wind speed submodel it is predicted, obtain each wind speed pre- Survey the prediction output valve of submodel；
S5, the prediction output valve of each forecasting wind speed submodel is combined overlap-add procedure, obtains final overall pre- Survey output valve.
Further, the method for the present invention also includes the method obtaining higher precision of prediction, particularly as follows:
According to the influence degree to global error that predicts the outcome of different intrinsic mode functions, choose different forecasting wind speed The prediction output valve of model is combined optimizing, and obtains higher forecasting wind speed precision.
Further, the intrinsic mode function being decomposed out by empirical mode decomposition in step S1 of the present invention needs to meet Following two conditions:
(1) in whole time range, the zero crossing number of intrinsic mode function and Local Extremum number must be equal, Or it is one that zero crossing number and Local Extremum number at most differ.
(2) point at any time, the coenvelope line of local maximum is necessary with the average of the lower envelope line of local minimum It is zero.
Original wind speed Time Series is concretely comprised the following steps by empirical mode decomposition:
(1) primary signal x (t) all maximum within the whole time and minimum point are found out；
(2) cubic spline interpolation is used all maximum points to be linked to be coenvelope line, in like manner by all minimum points even Become lower envelope line, then by average m of upper and lower envelope1T () calculates；
(3) separating first intrinsic mode function component in primary signal, formula is as follows:
h1(t)=x (t)-m1(t)
Ideal situation primary signal and upper and lower envelope average difference once just meet two bars of intrinsic mode function Part, i.e. h1T () is exactly an intrinsic mode function.But in fact envelope there will be error in fit procedure, causes h1(t) And it is unsatisfactory for two conditions of intrinsic mode function,
At this moment need h1(t) replacement x (t) repetition three above step:
h11(t)=h1(t)-m11(t)
Iteration k time is until h1kT () meets two conditions of intrinsic mode function.
By h1kT () is designated as c1(t), then c1T () is first intrinsic mode function component of primary signal x (t), next The first intrinsic mode function component c is separated from primary signal1T (), obtains difference r1(t)。
r1(t)=x (t)-c1(t)
Again by r1T primary signal in () alternative steps one, repeats (1) (2) (3) three steps, extract under primary signal One intrinsic mode function component, then according to rn(t)=rn-1(t)-cnT () repeats above step and progressively separates primary signal Each intrinsic mode function component.
Work as cnT () is in monotonic nature or less than terminating screening process during preset value, last primary signal can be with table It is shown as multiple intrinsic mode function and survival function sum, it may be assumed that
x ( t ) = Σ i = 1 n c i ( t ) + r n ( t )
ciT () represents the i-th intrinsic mode function component of primary signal x (t), rnT () is last surplus She's signal, It is referred to as survival function, rnT () is the representative of signal moderate tone, and each rank intrinsic mode function component ciT () contains letter successively Different frequency range composition from high to low in number x (t).
Further, concrete to the method building respective training dataset and test data set in step S2 of the present invention For:
Intrinsic mode function imfnTraining dataset TrnComprise input data X of modelnWith output data Yn, input number According to output data be to intrinsic mode function imfnTime series carry out sliding window and collect, input dataWherein m is that forecast model inputs number, exports dataN takes It is worth and is determined by forecast model output number；
Intrinsic mode function imfnTest data set TenMode of choosing and training dataset TrnMode of choosing phase With.
Further, in step S3 of the present invention, stack coding network uses four-layer structure, specifically includes: an input Layer, two hidden layers and an output layer；Wherein hidden layer neuron activation function uses sigmoid function, sigmoid function Computing formula is:
f ( x ) = 1 1 + exp ( - x )
Output layer neuron activation functions uses linear function, and linear function computing formula is:
F (x)=k x+b
Wherein, k represents last hidden layer neuron unit synaptic weight parameter to output layer neuron elements, b Represent the bias term of output layer neural unit.
Further, four layers of stack coding network in step S3 of the present invention construction method particularly as follows:
Step 1: use own coding device 1 to being originally inputted xiCarrying out encoding-decoding process, the hidden layer of own coding device 1 can be given birth to Become feature 1, useRepresent；
h i ( 1 ) = f ( W ( 1 ) x i + b ( 1 ) )
W(1)For the synaptic weight parameter between input layer neural unit and first hidden layer neural unit, b(1)It is first The bias term of individual hidden layer neural unit, f () uses sigmoid activation primitive.
Step 2: use the hidden layer output of own coding device 1, i.e.As the input of own coding device 2, then to inputCarrying out encoding-decoding process, own coding device 2 hidden layer can generate feature 2, usesRepresent；
h i ( 2 ) = f ( W ( 2 ) h i ( 1 ) + b ( 2 ) )
W(2)It is the synaptic weight parameter between first hidden layer neural unit and second hidden layer neural unit, b(2) Being the bias term of second hidden layer neural unit, f () uses sigmoid activation primitive.
Step 3: neuron, the hidden layer neuron of own coding device 1, and the hidden layer of own coding device 2 will be originally inputted Neuron is stacked to together, finally the hidden layer output of own coding device 2 is delivered to predicting unit and carries out classification prediction, it was predicted that unit Use linear activation primitive.
The present invention provides a kind of mixed model forecasting wind speed system learnt based on empirical mode decomposition and the degree of depth, including:
Intrinsic mode function computing unit, is used for obtaining original wind speed time series, builds empirical mode decomposition and the degree of depth The hybrid prediction model of study, rule of thumb original wind speed time series is decomposed by mode decomposition, obtains multiple eigen mode State function；
Data set construction unit, for building respective training dataset and test data set to each intrinsic mode function；
Forecasting wind speed submodel construction unit, for concentrating each intrinsic mode function training sample to send into training data Stack coding network is trained, and obtains respective forecasting wind speed submodel；
Prediction output valve computing unit, is carried out for test data set is sent to each self-corresponding forecasting wind speed submodel Prediction, obtains the prediction output valve of each forecasting wind speed submodel；
Result output unit, for the prediction output valve of each forecasting wind speed submodel is combined overlap-add procedure, To final overall prediction output valve.
The beneficial effect comprise that: the mixed model wind learnt based on empirical mode decomposition and the degree of depth of the present invention Speed Forecasting Methodology, first against wind speed seasonal effect in time series non-stationary property, uses empirical mode decomposition method non-linear, non-flat Steady wind series adaptive decomposition becomes the basic natural mode of vibration component of different scale, then carries out through degree of depth study and neutral net Combined prediction, reduces the difference between each forecast model, strengthens the model learning capacity to plurality of target function, is effectively improved pre- Survey precision of prediction and the robustness of model；It is combined optimizing permissible by choosing the predictive value of different natural mode of vibration component simultaneously Obtain higher short-term wind speed forecasting precision.
Accompanying drawing explanation
Below in conjunction with drawings and Examples, the invention will be further described, in accompanying drawing:
Fig. 1 is the mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention Flow chart；
Fig. 2 is the mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention Four layers of stack autoencoder network building process；
Fig. 3 is the mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention Wind speed mixed model prediction flow chart；
Fig. 4 is the mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention EMD-SAE forecast result of model；
Fig. 5 is the mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention Three kinds of forecast model prediction effect contrasts；
Fig. 6 (a) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (a) of method；
Fig. 6 (b) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (b) of method；
Fig. 6 (c) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (c) of method；
Fig. 6 (d) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (d) of method；
Fig. 6 (e) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (e) of method；
Fig. 6 (f) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (f) of method；
Fig. 6 (g) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (g) of method；
Fig. 6 (h) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (h) of method；
Fig. 6 (i) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (i) of method；
Fig. 6 (j) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (j) of method；
Fig. 6 (k) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (k) of method；
Fig. 6 (l) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (l) of method；
Fig. 6 (m) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (m) of method；
Fig. 6 (n) is the mixed model forecasting wind speed side learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention The prediction effect (n) of method.
Detailed description of the invention
In order to make the purpose of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, right The present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, not For limiting the present invention.
As it is shown in figure 1, the mixed model forecasting wind speed learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention Method, comprises the following steps:
S1, obtain original wind speed time series, build empirical mode decomposition and the hybrid prediction model of degree of depth study, according to Original wind speed time series is decomposed by empirical mode decomposition, obtains multiple intrinsic mode function.Divided by empirical mode decomposition Solution intrinsic mode function out needs to meet following two conditions:
(1) in whole time range, the zero crossing number of intrinsic mode function and Local Extremum number must be equal, Or it is one that zero crossing number and Local Extremum number at most differ.
(2) point at any time, the coenvelope line of local maximum is necessary with the average of the lower envelope line of local minimum It is zero.
Original wind speed Time Series is concretely comprised the following steps by empirical mode decomposition:
(1) primary signal x (t) all maximum within the whole time and minimum point are found out；
(2) cubic spline interpolation is used all maximum points to be linked to be coenvelope line, in like manner by all minimum points even Become lower envelope line, then by average m of upper and lower envelope1T () calculates；
(3) separating first intrinsic mode function component in primary signal, formula is as follows:
h1(t)=x (t)-m1(t)
Ideal situation primary signal and upper and lower envelope average difference once just meet two bars of intrinsic mode function Part, i.e. h1T () is exactly an intrinsic mode function.But in fact envelope there will be error in fit procedure, causes h1(t) And it is unsatisfactory for two conditions of intrinsic mode function,
At this moment need h1(t) replacement x (t) repetition three above step:
h11(t)=h1(t)-m11(t)
Iteration k time is until h1kT () meets two conditions of intrinsic mode function.
By h1kT () is designated as c1(t), then c1T () is first intrinsic mode function component of primary signal x (t), next The first intrinsic mode function component c is separated from primary signal1T (), obtains difference r1(t)。
r1(t)=x (t)-c1(t)
Again by r1T primary signal in () alternative steps one, repeats (1) (2) (3) three steps, extract under primary signal One intrinsic mode function component, then according to rn(t)=rn-1(t)-cnT () repeats above step and progressively separates primary signal Each intrinsic mode function component.
Work as cnT () is in monotonic nature or less than terminating screening process during preset value, last primary signal can be with table It is shown as multiple intrinsic mode function and survival function sum, it may be assumed that
x ( t ) = Σ i = 1 n c i ( t ) + r n ( t )
ciT () represents the i-th intrinsic mode function component of primary signal x (t), rnT () is last surplus She's signal, It is referred to as survival function, rnT () is the representative of signal moderate tone, and each rank intrinsic mode function component ciT () contains letter successively Different frequency range composition from high to low in number x (t).
S2, each intrinsic mode function is built respective training dataset and test data set.
S3, training data is concentrated each intrinsic mode function training sample send into stack coding network be trained, To respective forecasting wind speed submodel；
S4, test data set is sent to each self-corresponding forecasting wind speed submodel it is predicted, obtain each wind speed pre- Survey the prediction output valve of submodel；
S5, the prediction output valve of each forecasting wind speed submodel is combined overlap-add procedure, obtains final overall pre- Survey output valve.
The method also includes the method obtaining higher precision of prediction, particularly as follows:
According to the influence degree to global error that predicts the outcome of different intrinsic mode functions, choose different forecasting wind speed The prediction output valve of model is combined optimizing, and obtains higher forecasting wind speed precision.
In another specific embodiment of the present invention, concretely comprising the following steps of the method:
First with empirical mode decomposition, original wind speed time series is decomposed, obtain several intrinsic mode functions IMF={imf1,imf2,…,imfn}。
Then each intrinsic mode function IMF is built respective training data Traindata={Tr1,Tr2,…,TrnAnd Test data Testdata={Te1,Te2,…,Ten}.Wherein, intrinsic mode function imfnTraining data TrnComprise model Input data XnWith output data Yn, input data and output data are to intrinsic mode function imfnTime series is slided Window collects, and inputs dataWherein m is that forecast model inputs number, exports dataThe value of l is determined by forecast model output number.Intrinsic mode function imfnTest data TenMode of choosing with training data TrnChoose.
Each intrinsic mode function training sample in training data Traindata is sent into stack coding network SAE training again Respective forecasting wind speed submodel IMF-SAE={IMF1-SAE1,IMF2-SAE2,…,IMFn-SAEn}.Stack coding network SAE Using four-layer structure, i.e. one input layer, two hidden layers, output layers, wherein hidden layer neuron activation function uses Sigmoid function, sigmoid function computing formula is:
f ( x ) = 1 1 + exp ( - x )
Output layer neuron activation functions uses linear function, and linear function computing formula is:
F (x)=k x+b
Wherein, k represents last hidden layer neuron unit synaptic weight parameter to output layer neuron elements, b Represent the bias term of output layer neural unit.
The structure of four layers of stack own coding model is as shown in Figure 2.
Four layers of stack autoencoder network building process can be completed by following three steps:
Step 1: use own coding device 1 to being originally inputted xiCarrying out encoding-decoding process, the hidden layer of own coding device 1 can be given birth to Become feature 1, useRepresent；
Step 2: use the hidden layer of own coding device 1 to export namelyAs the input of own coding device 2, then to defeated EnterCarrying out encoding-decoding process, own coding device 2 hidden layer can generate feature 2, usesRepresent；
h i ( 1 ) = f ( W ( 1 ) x i + b ( 1 ) )
h i ( 2 ) = f ( W ( 2 ) h i ( 1 ) + b ( 2 ) )
W(1)For the synaptic weight parameter between input layer neural unit and first hidden layer neural unit, b(1)It is first The bias term of individual hidden layer neural unit；W(2)It is between first hidden layer neural unit and second hidden layer neural unit Synaptic weight parameter, b(2)Being the bias term of second hidden layer neural unit, f () uses sigmoid activation primitive.
Step 3: neuron, the hidden layer neuron of own coding device 1, and the hidden layer of own coding device 2 will be originally inputted Neuron is stacked to together, finally the hidden layer output of own coding device 2 is delivered to predicting unit and carries out classification prediction, it was predicted that unit Use linear activation primitive.
After training respective forecasting wind speed submodel IMF-SAE, test data Testdata are sent to respective correspondence Forecasting wind speed submodel in be predicted, obtain prediction output valve pre={pre of each forecasting wind speed submodel1, pre2,…,pren}。
Then the prediction output valve of each forecasting wind speed submodel is combined overlap-add procedure, obtains final entirety pre- Survey output valve value.
Value=pre1+pre2+…+pren
The influence degree to global error that predicts the outcome according to different intrinsic mode functions chooses different forecasting wind speed The prediction output valve of submodel is combined optimizing, and obtains higher forecasting wind speed precision.The pre-flow gauge of wind speed mixed model is such as Shown in Fig. 3.
Experiment 1: the mixed model forecasting wind speed experiment of empirical mode decomposition and degree of depth study
This experiment initially sets up each component forecast model IMF-SAE, finally carries out adding up by predicting the outcome of each component To the overall predictive value that model is final, and and SAE, SVR and neural network model contrast.Fig. 4 is the pre-of EMD-SAE model Survey effect.
Neutral net (NN) model, depth model SAE and EMD-SAE mixed model are put together and are done error contrast, error Contrast is as shown in table 1.
Table 1 multi-model forecasting wind speed error contrasts
As can be seen from Table 1, from traditional BP neutral net forecasting wind speed, pre-to neutral net wind speed based on depth theory Survey, then the mixed model wind speed combined to empirical mode decomposition and degree of depth study, the estimated performance of model is improving step by step. Observe prediction effect for convenience, intercept a part of test result as shown in Figure 5.
Experiment 2:IMF component is on the impact of mixed model estimated performance and optimum organization thereof
During experiment one structure EMD-SAE forecast model, first each component is built its respective IMF-SAE pre- Survey model, shown in effect such as Fig. 6 (a)-Fig. 6 (n) of each IMF-SAE forecast model.
By observing Fig. 6 (a)-Fig. 6 (n), the prediction effect of each IMF-SAE forecast model is different, it can be seen that place Will be almost for the prediction effect of the IMF component of high frequency is in the IMF component prediction effect of low frequency relatively.Low frequency several IMF component prediction curve of output and desired output curve substantially overlap.The prediction error value of each component such as table 2, from table Numeral can see that the IMF component forecast error of high frequency is big intuitively, and the IMF component forecast error of low frequency is little.
Table 2 each component forecast error
The final output of EMD-SAE forecast model is overlapped by the output of each component forecast model IMF-SAE, Predicting the outcome of each IMF component has an impact for final predicting the outcome.By the prediction of each component forecast model by mistake Difference contrast finds, the forecast error of each component is different, thereby it is assumed that the IMF component of different frequency is in whole prediction process Played in effect be different.It is true that various noises or some other unrelated interference signal can be clipped in high frequency letter In number, finally the result of prediction can be impacted.
In order to study each component for the whole impact predicted the outcome, next go successively in whole EMD-SAE model Falling each IMF-SAE model, be i.e. added without each component in model prediction successively and be predicted, observing predicts the outcome analyzes respectively The impact on finally predicting the outcome of the individual component.Such as in order to observe the impact on overall forecast error of the IMF1 component, use The accumulated value of IMF2～RES observes the IMF1 impact on final result as final prediction output.After removing each component Forecast error as shown in table 3.
Table 3 each component forecast error
Contrasted by the forecast error of table 3, it appeared that it is different for removing each component to the impact predicted the outcome, After removing IMF1 component, the effect of model prediction there is is lifting, and has removed residual components model and cannot complete prediction.By This speculates that the wind speed effective information contained by each component is different, the weight that each component is occupied in the middle of whole model prediction Want degree the most different.Containing disturbing why information, wind series have the strongest fluctuation in a large number in random high frequency component IMF1 Property, it was predicted that difficulty, the most just it is affected by the impact of these interference information, removes the garbage in wind speed and can improve pre- Survey precision, and containing a large amount of wind speed information in residual components RES, whole prediction occupies critically important status.So to EMD The IMF component decomposited is analyzed, and uses containing the IMF component of effective information, removes some unrelated, containing a large amount of interference The component of information, finds the prediction combination of suitable IMF component, can improve the precision of prediction of model further.
In sum, the present invention proposes a kind of mixed model forecasting wind speed learnt based on empirical mode decomposition and the degree of depth Method, the method, first against wind speed seasonal effect in time series non-stationary property, uses empirical mode decomposition method non-linear, non-flat Steady wind series adaptive decomposition becomes the basic natural mode of vibration component of different scale, then carries out through degree of depth study and neutral net Combined prediction, reduces the difference between each forecast model, strengthens the model learning capacity to plurality of target function, is effectively improved pre- Survey precision of prediction and the robustness of model.It is combined optimizing permissible by choosing the predictive value of different natural mode of vibration component simultaneously Obtain higher short-term wind speed forecasting precision, provide a new method for improving the short-term forecast precision of wind speed.
The mixed model forecasting wind speed system learnt based on empirical mode decomposition and the degree of depth of the embodiment of the present invention, for real The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth of the existing embodiment of the present invention, including:
Intrinsic mode function computing unit, is used for obtaining original wind speed time series, builds empirical mode decomposition and the degree of depth The hybrid prediction model of study, rule of thumb original wind speed time series is decomposed by mode decomposition, obtains multiple eigen mode State function；
Data set construction unit, for building respective training dataset and test data set to each intrinsic mode function；
Forecasting wind speed submodel construction unit, for concentrating each intrinsic mode function training sample to send into training data Stack coding network is trained, and obtains respective forecasting wind speed submodel；
Prediction output valve computing unit, is carried out for test data set is sent to each self-corresponding forecasting wind speed submodel Prediction, obtains the prediction output valve of each forecasting wind speed submodel；
Result output unit, for the prediction output valve of each forecasting wind speed submodel is combined overlap-add procedure, To final overall prediction output valve.
It should be appreciated that for those of ordinary skills, can be improved according to the above description or be converted, And all these modifications and variations all should belong to the protection domain of claims of the present invention.

Claims (8)

1. the mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth, it is characterised in that include with Lower step:
S1, obtain original wind speed time series, build empirical mode decomposition and the hybrid prediction model of degree of depth study, rule of thumb Original wind speed time series is decomposed by mode decomposition, obtains multiple intrinsic mode function；
S2, each intrinsic mode function is built respective training dataset and test data set；
S3, training data is concentrated each intrinsic mode function training sample send into stack coding network be trained, obtain each From forecasting wind speed submodel；
S4, test data set is sent to each self-corresponding forecasting wind speed submodel it is predicted, obtain each forecasting wind speed The prediction output valve of model；
S5, the prediction output valve of each forecasting wind speed submodel is combined overlap-add procedure, obtains final entirety prediction defeated Go out value.
The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth the most according to claim 1, its Being characterised by, the method also includes the method obtaining higher precision of prediction, particularly as follows:
According to the influence degree to global error that predicts the outcome of different intrinsic mode functions, choose different forecasting wind speed submodel Prediction output valve be combined optimize, obtain higher forecasting wind speed precision.
The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth the most according to claim 1, its Being characterised by, the intrinsic mode function being decomposed out by empirical mode decomposition in step S1 needs to meet following two conditions:
(1) in whole time range, the zero crossing number of intrinsic mode function and Local Extremum number must be equal, or It is one that zero crossing number and Local Extremum number at most differ；
(2) average of point, the coenvelope line of local maximum and the lower envelope line of local minimum is necessary for zero at any time.
The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth the most according to claim 3, its Being characterised by, in step S1, original wind speed Time Series is concretely comprised the following steps by empirical mode decomposition:
(1) primary signal x (t) all maximum within the whole time and minimum point are found out；
(2) use cubic spline interpolation that all maximum points are linked to be coenvelope line, all minimum points are linked to be lower envelope Line, then by average m of upper and lower envelope1T () calculates；
(3) separating first intrinsic mode function component in primary signal, formula is as follows:
h1(t)=x (t)-m1(t)
Primary signal and upper and lower envelope carry out repeatedly average, by h1(t) replacement x (t) repetition three above step:
h11(t)=h1(t)-m11(t)
Iteration k time is until h1kT () meets two conditions of intrinsic mode function；
By h1kT () is designated as c1(t), then c1T () is first intrinsic mode function component of primary signal x (t), and from original letter The first intrinsic mode function component c is separated in number1T (), obtains difference r1(t)；
r1(t)=x (t)-c1(t)
Again by r1T primary signal in () alternative steps one, repeats step (1) (2) (3), extract the next eigen mode of primary signal State function component, then according to rn(t)=rn-1(t)-cnT () repeats above step and progressively separates each intrinsic of primary signal Mode function component；
Work as cnT () is in monotonic nature or less than terminating screening process during preset value, last primary signal can be expressed as many Individual intrinsic mode function and survival function sum, it may be assumed that
x ( t ) = Σ i = 1 n c i ( t ) + r n ( t )
ciT () represents the i-th intrinsic mode function component of primary signal x (t), rnT () is last surplus She's signal, be referred to as Survival function, rnT () is the representative of signal moderate tone, and each rank intrinsic mode function component ciT () contains signal x successively Different frequency range composition from high to low in (t).
The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth the most according to claim 1, its Be characterised by, in step S2 to build respective training dataset and test data set method particularly as follows:
Intrinsic mode function imfnTraining dataset TrnComprise input data X of modelnWith output data Yn, input data and Output data are to intrinsic mode function imfnTime series carry out sliding window and collect, input dataWherein m is that forecast model inputs number, exports dataN takes It is worth and is determined by forecast model output number；
Intrinsic mode function imfnTest data set TenMode of choosing and training dataset TrnTo choose mode identical.
The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth the most according to claim 1, its Being characterised by, in step S3, stack coding network uses four-layer structure, specifically includes: input layer, two hidden layers and Individual output layer；Wherein hidden layer neuron activation function uses sigmoid function, and sigmoid function computing formula is:
f ( x ) = 1 1 + exp ( - x )
Output layer neuron activation functions uses linear function, and linear function computing formula is:
F (x)=k x+b
Wherein, k represents last hidden layer neuron unit synaptic weight parameter to output layer neuron elements, and b represents The bias term of output layer neural unit.
The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth the most according to claim 6, its Be characterised by, the construction method of four layers of stack coding network in step S3 particularly as follows:
Step 1: use own coding device 1 to being originally inputted xiCarrying out encoding-decoding process, the hidden layer of own coding device 1 can generate spy Levy 1, useRepresent；
h i ( 1 ) = f ( W ( 1 ) x i + b ( 1 ) )
W(1)For the synaptic weight parameter between input layer neural unit and first hidden layer neural unit, b(1)Be first hidden Containing the bias term of layer neural unit, f () uses sigmoid activation primitive；
Step 2: use the hidden layer output of own coding device 1, i.e.As the input of own coding device 2, then to inputEnter Row encoding-decoding process, own coding device 2 hidden layer can generate feature 2, usesRepresent；
h i ( 2 ) = f ( W ( 2 ) h i ( 1 ) + b ( 2 ) )
W(2)It is the synaptic weight parameter between first hidden layer neural unit and second hidden layer neural unit, b(2)It is The bias term of two hidden layer neural units, f () uses sigmoid activation primitive；
Step 3: the hidden layer nerve of neuron, the hidden layer neuron of own coding device 1, and own coding device 2 will be originally inputted Unit is stacked to together, finally the hidden layer output of own coding device 2 is delivered to predicting unit and carries out classification prediction, it was predicted that unit uses Linear activation primitive.
8. the mixed model forecasting wind speed system learnt based on empirical mode decomposition and the degree of depth, it is characterised in that including:
Intrinsic mode function computing unit, is used for obtaining original wind speed time series, builds empirical mode decomposition and degree of depth study Hybrid prediction model, rule of thumb original wind speed time series is decomposed by mode decomposition, obtains multiple intrinsic mode letter Number；
Data set construction unit, for building respective training dataset and test data set to each intrinsic mode function；
Forecasting wind speed submodel construction unit, for concentrating each intrinsic mode function training sample to send into stack by training data Coding network is trained, and obtains respective forecasting wind speed submodel；
Prediction output valve computing unit, carries out pre-for test data set is sent to each self-corresponding forecasting wind speed submodel Survey, obtain the prediction output valve of each forecasting wind speed submodel；
Result output unit, for the prediction output valve of each forecasting wind speed submodel is combined overlap-add procedure, obtains Whole overall prediction output valve.
CN201610444160.4A 2016-06-20 2016-06-20 Mixed model wind speed forecasting method and system based on empirical mode decomposition and deep learning Expired - Fee Related CN106126896B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610444160.4A CN106126896B (en) 2016-06-20 2016-06-20 Mixed model wind speed forecasting method and system based on empirical mode decomposition and deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610444160.4A CN106126896B (en) 2016-06-20 2016-06-20 Mixed model wind speed forecasting method and system based on empirical mode decomposition and deep learning

Publications (2)

Publication Number Publication Date
CN106126896A true CN106126896A (en) 2016-11-16
CN106126896B CN106126896B (en) 2019-03-22

Family

ID=57471153

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610444160.4A Expired - Fee Related CN106126896B (en) 2016-06-20 2016-06-20 Mixed model wind speed forecasting method and system based on empirical mode decomposition and deep learning

Country Status (1)

Country Link
CN (1) CN106126896B (en)

Cited By (38)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106596899A (en) * 2016-11-21 2017-04-26 浙江大学 Evaluation method of random error between oil-immersed transformer on-line monitoring data and on-line detection data
CN106897794A (en) * 2017-01-12 2017-06-27 长沙理工大学 A kind of wind speed forecasting method based on complete overall experience mode decomposition and extreme learning machine
CN107092744A (en) * 2017-04-16 2017-08-25 北京工业大学 Subsidence value Forecasting Methodology based on EMD SVR
CN107330375A (en) * 2017-06-05 2017-11-07 天津大学 Lathe flutter on-line monitoring method based on empirical mode decomposition and Sample Entropy feature
CN107491723A (en) * 2017-06-27 2017-12-19 中国地质大学（武汉） A kind of Forecasting Methodology and system of fracture-pore reservoir creation data
CN107515202A (en) * 2017-08-17 2017-12-26 清华大学 Terahertz light spectral analysis method, system and equipment
CN108520269A (en) * 2018-03-10 2018-09-11 华北电力大学（保定） A kind of wind speed forecasting method and forecasting wind speed system
CN108776719A (en) * 2018-04-27 2018-11-09 南京信息工程大学 High energy electrical flux hour forecasting model method for building up based on empirical mode decomposition
CN108875706A (en) * 2018-07-18 2018-11-23 中国海洋大学 The ocean structure Time-Frequency Analysis Method collected based on sliding average and energy
CN109146194A (en) * 2018-09-05 2019-01-04 重庆工商大学 A kind of Runoff forestry method encoding interconnection vector machine certainly based on change mode convolution
CN109978275A (en) * 2019-04-03 2019-07-05 中南大学 A kind of extreme wind wind speed forecasting method and system mixing CFD and deep learning
CN110163433A (en) * 2019-05-21 2019-08-23 上海海事大学 A kind of ship method for predicting
CN110162379A (en) * 2018-04-24 2019-08-23 腾讯云计算（北京）有限责任公司 Virtual machine migration method, device and computer equipment
CN110390435A (en) * 2019-07-24 2019-10-29 广东工业大学 A kind of wind power forecasting method, device, equipment and readable storage medium storing program for executing
CN110555226A (en) * 2019-04-03 2019-12-10 太原理工大学 method for predicting residual life of lithium iron phosphate battery based on EMD and MLP
CN110598955A (en) * 2019-09-25 2019-12-20 东南大学 Maximum instantaneous wind speed probability prediction method for high-speed train
CN110763830A (en) * 2019-12-04 2020-02-07 济南大学 Method for predicting content of free calcium oxide in cement clinker
CN110852527A (en) * 2019-11-20 2020-02-28 成都理工大学 Reservoir physical property parameter prediction method combining deep learning
CN110909931A (en) * 2019-11-20 2020-03-24 成都理工大学 Logging curve prediction method based on modal decomposition reconstruction and depth LSTM-RNN model
CN111400040A (en) * 2020-03-12 2020-07-10 重庆大学 Industrial Internet system based on deep learning and edge calculation and working method
CN111428201A (en) * 2020-03-27 2020-07-17 陕西师范大学 Prediction method for time series data based on empirical mode decomposition and feedforward neural network
CN111461418A (en) * 2020-03-23 2020-07-28 上海电气风电集团股份有限公司 Wind speed prediction method, system, electronic device and storage medium
CN111832812A (en) * 2020-06-27 2020-10-27 南通大学 Wind power short-term prediction method based on deep learning
CN112100841A (en) * 2020-09-09 2020-12-18 中铁二十局集团有限公司 Shield tunneling machine attitude prediction method and device, terminal equipment and storage medium
WO2021000556A1 (en) * 2019-07-02 2021-01-07 中国科学院深圳先进技术研究院 Method and system for predicting remaining useful life of industrial equipment, and electronic device
CN112215409A (en) * 2020-09-24 2021-01-12 交控科技股份有限公司 Rail transit station passenger flow prediction method and system
CN112257894A (en) * 2020-09-08 2021-01-22 山东师范大学 Hybrid building power consumption prediction method and system based on data driving
CN112253406A (en) * 2020-09-29 2021-01-22 中国电建集团华东勘测设计研究院有限公司 Environment load prediction method and vibration pre-control system for offshore wind turbine generator
CN112398142A (en) * 2020-10-26 2021-02-23 广西大学 Power grid frequency intelligent control method based on empirical mode decomposition
CN112561180A (en) * 2020-12-21 2021-03-26 深圳大学 Short-term wind speed prediction method and device, computer equipment and storage medium
CN113030717A (en) * 2021-03-12 2021-06-25 广东电网有限责任公司 Circuit breaker fault diagnosis method and circuit based on electric signal deep learning
CN113465733A (en) * 2021-08-13 2021-10-01 重庆大学 Vibration table structure displacement response prediction method and device based on EEMD-DNN
CN113465732A (en) * 2021-08-13 2021-10-01 重庆大学 Vibration table structure displacement response prediction method and device based on EMD-DNN
CN113537638A (en) * 2021-08-13 2021-10-22 重庆大学 Short-term wind pressure prediction method and abnormal data completion method and device for high-rise building
CN113688774A (en) * 2021-09-03 2021-11-23 重庆大学 High-rise building wind-induced response prediction and training method and device based on deep learning
CN113688770A (en) * 2021-09-02 2021-11-23 重庆大学 Long-term wind pressure missing data completion method and device for high-rise building
CN114706144A (en) * 2022-02-24 2022-07-05 南方海洋科学与工程广东省实验室(湛江) Marine environment forecasting method, device, system and storage medium
CN114841417A (en) * 2022-04-18 2022-08-02 珠江水利委员会珠江水利科学研究院 High-precision salt tide forecasting method and system and readable storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103425867A (en) * 2013-06-24 2013-12-04 武汉大学 Short-term wind power combination prediction method
CN104156728A (en) * 2014-07-14 2014-11-19 西安电子科技大学 Polarized SAR image classification method based on stacked code and softmax
CN104850735A (en) * 2015-04-28 2015-08-19 浙江大学 Activity recognition method based on stack own coding

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103425867A (en) * 2013-06-24 2013-12-04 武汉大学 Short-term wind power combination prediction method
CN104156728A (en) * 2014-07-14 2014-11-19 西安电子科技大学 Polarized SAR image classification method based on stacked code and softmax
CN104850735A (en) * 2015-04-28 2015-08-19 浙江大学 Activity recognition method based on stack own coding

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
JAIMEZABALZA ET AL: "Novel segmented stacked autoencoder for effective dimensionality reduction and feature extraction in hyperspectral imaging", 《NEUROCOMPUTING》 *
尹子中等: "基于改进EMD和RBFNN的短期风速预测模型", 《广东电力》 *
张翌晖等: "基于集合经验模态分解和改进极限学习机的短期风速组合预测研究", 《电力***保护与控制》 *
王韶等: "基于经验模式分解和神经网络的短期风速组合预测", 《电力***保护与控制》 *

Cited By (53)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106596899A (en) * 2016-11-21 2017-04-26 浙江大学 Evaluation method of random error between oil-immersed transformer on-line monitoring data and on-line detection data
CN106596899B (en) * 2016-11-21 2018-11-30 浙江大学 The evaluation method of random error between oil-immersed transformer online monitoring data and live detection data
CN106897794A (en) * 2017-01-12 2017-06-27 长沙理工大学 A kind of wind speed forecasting method based on complete overall experience mode decomposition and extreme learning machine
CN107092744A (en) * 2017-04-16 2017-08-25 北京工业大学 Subsidence value Forecasting Methodology based on EMD SVR
CN107092744B (en) * 2017-04-16 2020-11-06 北京工业大学 Surface subsidence prediction method based on EMD-SVR
CN107330375A (en) * 2017-06-05 2017-11-07 天津大学 Lathe flutter on-line monitoring method based on empirical mode decomposition and Sample Entropy feature
CN107491723A (en) * 2017-06-27 2017-12-19 中国地质大学（武汉） A kind of Forecasting Methodology and system of fracture-pore reservoir creation data
CN107515202A (en) * 2017-08-17 2017-12-26 清华大学 Terahertz light spectral analysis method, system and equipment
CN108520269A (en) * 2018-03-10 2018-09-11 华北电力大学（保定） A kind of wind speed forecasting method and forecasting wind speed system
CN110162379A (en) * 2018-04-24 2019-08-23 腾讯云计算（北京）有限责任公司 Virtual machine migration method, device and computer equipment
CN108776719A (en) * 2018-04-27 2018-11-09 南京信息工程大学 High energy electrical flux hour forecasting model method for building up based on empirical mode decomposition
CN108875706A (en) * 2018-07-18 2018-11-23 中国海洋大学 The ocean structure Time-Frequency Analysis Method collected based on sliding average and energy
CN108875706B (en) * 2018-07-18 2021-08-17 中国海洋大学 Ocean structure time-frequency analysis method based on moving average and energy collection
CN109146194A (en) * 2018-09-05 2019-01-04 重庆工商大学 A kind of Runoff forestry method encoding interconnection vector machine certainly based on change mode convolution
CN109978275A (en) * 2019-04-03 2019-07-05 中南大学 A kind of extreme wind wind speed forecasting method and system mixing CFD and deep learning
CN110555226A (en) * 2019-04-03 2019-12-10 太原理工大学 method for predicting residual life of lithium iron phosphate battery based on EMD and MLP
CN110163433B (en) * 2019-05-21 2021-10-26 上海海事大学 Ship flow prediction method
CN110163433A (en) * 2019-05-21 2019-08-23 上海海事大学 A kind of ship method for predicting
WO2021000556A1 (en) * 2019-07-02 2021-01-07 中国科学院深圳先进技术研究院 Method and system for predicting remaining useful life of industrial equipment, and electronic device
CN110390435A (en) * 2019-07-24 2019-10-29 广东工业大学 A kind of wind power forecasting method, device, equipment and readable storage medium storing program for executing
CN110598955B (en) * 2019-09-25 2022-12-20 东南大学 Maximum instantaneous wind speed probability prediction method for high-speed train
CN110598955A (en) * 2019-09-25 2019-12-20 东南大学 Maximum instantaneous wind speed probability prediction method for high-speed train
CN110852527A (en) * 2019-11-20 2020-02-28 成都理工大学 Reservoir physical property parameter prediction method combining deep learning
CN110909931A (en) * 2019-11-20 2020-03-24 成都理工大学 Logging curve prediction method based on modal decomposition reconstruction and depth LSTM-RNN model
CN110763830A (en) * 2019-12-04 2020-02-07 济南大学 Method for predicting content of free calcium oxide in cement clinker
CN110763830B (en) * 2019-12-04 2022-04-05 济南大学 Method for predicting content of free calcium oxide in cement clinker
CN111400040A (en) * 2020-03-12 2020-07-10 重庆大学 Industrial Internet system based on deep learning and edge calculation and working method
CN111461418A (en) * 2020-03-23 2020-07-28 上海电气风电集团股份有限公司 Wind speed prediction method, system, electronic device and storage medium
CN111461418B (en) * 2020-03-23 2023-07-18 上海电气风电集团股份有限公司 Wind speed prediction method, system, electronic equipment and storage medium
CN111428201B (en) * 2020-03-27 2023-04-11 陕西师范大学 Prediction method for time series data based on empirical mode decomposition and feedforward neural network
CN111428201A (en) * 2020-03-27 2020-07-17 陕西师范大学 Prediction method for time series data based on empirical mode decomposition and feedforward neural network
CN111832812A (en) * 2020-06-27 2020-10-27 南通大学 Wind power short-term prediction method based on deep learning
CN112257894A (en) * 2020-09-08 2021-01-22 山东师范大学 Hybrid building power consumption prediction method and system based on data driving
CN112100841A (en) * 2020-09-09 2020-12-18 中铁二十局集团有限公司 Shield tunneling machine attitude prediction method and device, terminal equipment and storage medium
CN112100841B (en) * 2020-09-09 2024-04-19 中铁二十局集团有限公司 Method and device for predicting attitude of shield tunneling machine, terminal equipment and storage medium
CN112215409B (en) * 2020-09-24 2024-01-30 交控科技股份有限公司 Rail transit station passenger flow prediction method and system
CN112215409A (en) * 2020-09-24 2021-01-12 交控科技股份有限公司 Rail transit station passenger flow prediction method and system
CN112253406A (en) * 2020-09-29 2021-01-22 中国电建集团华东勘测设计研究院有限公司 Environment load prediction method and vibration pre-control system for offshore wind turbine generator
CN112253406B (en) * 2020-09-29 2022-05-27 中国电建集团华东勘测设计研究院有限公司 Environment load prediction method and vibration pre-control system for offshore wind turbine generator
CN112398142B (en) * 2020-10-26 2022-05-13 广西大学 Power grid frequency intelligent control method based on empirical mode decomposition
CN112398142A (en) * 2020-10-26 2021-02-23 广西大学 Power grid frequency intelligent control method based on empirical mode decomposition
CN112561180B (en) * 2020-12-21 2022-05-17 深圳大学 Short-term wind speed prediction method and device based on meta-learning, computer equipment and storage medium
CN112561180A (en) * 2020-12-21 2021-03-26 深圳大学 Short-term wind speed prediction method and device, computer equipment and storage medium
CN113030717A (en) * 2021-03-12 2021-06-25 广东电网有限责任公司 Circuit breaker fault diagnosis method and circuit based on electric signal deep learning
CN113465733A (en) * 2021-08-13 2021-10-01 重庆大学 Vibration table structure displacement response prediction method and device based on EEMD-DNN
CN113537638A (en) * 2021-08-13 2021-10-22 重庆大学 Short-term wind pressure prediction method and abnormal data completion method and device for high-rise building
CN113465732A (en) * 2021-08-13 2021-10-01 重庆大学 Vibration table structure displacement response prediction method and device based on EMD-DNN
CN113688770A (en) * 2021-09-02 2021-11-23 重庆大学 Long-term wind pressure missing data completion method and device for high-rise building
CN113688770B (en) * 2021-09-02 2024-02-06 重庆大学 Method and device for supplementing long-term wind pressure missing data of high-rise building
CN113688774A (en) * 2021-09-03 2021-11-23 重庆大学 High-rise building wind-induced response prediction and training method and device based on deep learning
CN113688774B (en) * 2021-09-03 2023-09-26 重庆大学 Advanced learning-based high-rise building wind induced response prediction and training method and device
CN114706144A (en) * 2022-02-24 2022-07-05 南方海洋科学与工程广东省实验室(湛江) Marine environment forecasting method, device, system and storage medium
CN114841417A (en) * 2022-04-18 2022-08-02 珠江水利委员会珠江水利科学研究院 High-precision salt tide forecasting method and system and readable storage medium

Also Published As

Publication number Publication date
CN106126896B (en) 2019-03-22

Similar Documents

Publication Publication Date Title
CN106126896A (en) The mixed model wind speed forecasting method learnt based on empirical mode decomposition and the degree of depth and system
Syahputra APPLICATION OF NEURO-FUZZY METHOD FOR PREDICTION OF VEHICLE FUEL CONSUMPTION.
Chang An RBF neural network combined with OLS algorithm and genetic algorithm for short-term wind power forecasting
CN106875041A (en) A kind of short-term wind speed forecasting method
Bigdeli et al. A comparative study of optimal hybrid methods for wind power prediction in wind farm of Alberta, Canada
CN107766995A (en) Power-system short-term load forecasting method based on depth recurrent neural network
CN105117593A (en) Wavelet transform and particle swarm optimized grey model-based short-term wind speed forecasting method
CN109583588A (en) A kind of short-term wind speed forecasting method and system
CN106897794A (en) A kind of wind speed forecasting method based on complete overall experience mode decomposition and extreme learning machine
CN105373856A (en) Wind electricity power short-term combined prediction method considering run detection method reconstruction
CN109614384A (en) Power-system short-term load forecasting method under Hadoop frame
CN105956252A (en) Generative deep belief network-based multi-scale forecast modeling method for ultra-short-term wind speed
CN104268698B (en) A kind of method that ranking is carried out to power grid enterprises' operation monitoring business datum
Eljazzar et al. Enhancing electric load forecasting of ARIMA and ANN using adaptive Fourier series
Sajid Machine learned artificial neural networks vs linear regression: A case of chinese carbon emissions
Wang et al. Forecasting electricity demand using Grey-Markov model
Gobu et al. Two-stage machine learning framework for simultaneous forecasting of price-load in the smart grid
Debata et al. Efficient energy management strategies for hybrid electric vehicles using shuffled frog-leaping algorithm
Haifeng et al. An adaptive BP-network approach to short term load forecasting
Ma et al. A grey forecasting model based on BP neural network for crude oil production and consumption in China
Fu A hybrid prediction of wind speed based on variational mode decomposition method and long short-term memory
Peri et al. Very short-term prediction of wind farm power: An advanced hybrid intelligent approach
Murai et al. The fundamental research on AI prediction and determination of control force for maximizing the power generation of PA-WEC in irregular waves
Niazi et al. A hybrid approach for security evaluation and preventive control of power systems
Fayziev et al. Mathematical modeling and forecasting of electricity production in enterprises of the energy system of Uzbekistan

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20190322

Termination date: 20190620