CN108932671A - A kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune - Google Patents
A kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune Download PDFInfo
- Publication number
- CN108932671A CN108932671A CN201810575699.2A CN201810575699A CN108932671A CN 108932671 A CN108932671 A CN 108932671A CN 201810575699 A CN201810575699 A CN 201810575699A CN 108932671 A CN108932671 A CN 108932671A
- Authority
- CN
- China
- Prior art keywords
- parameter
- prediction model
- lstm
- wind
- electricity generation
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 26
- 230000005611 electricity Effects 0.000 title claims abstract description 26
- 238000013277 forecasting method Methods 0.000 title claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000000034 method Methods 0.000 claims abstract description 9
- 230000007613 environmental effect Effects 0.000 claims abstract description 8
- 238000005457 optimization Methods 0.000 claims abstract description 3
- 230000006870 function Effects 0.000 claims description 15
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 239000013598 vector Substances 0.000 claims description 5
- 238000012552 review Methods 0.000 claims description 3
- 201000004409 schistosomiasis Diseases 0.000 claims description 3
- 230000003831 deregulation Effects 0.000 abstract description 4
- 238000013459 approach Methods 0.000 abstract description 2
- 238000011084 recovery Methods 0.000 abstract 1
- 230000007423 decrease Effects 0.000 description 12
- 230000015654 memory Effects 0.000 description 9
- 230000000306 recurrent effect Effects 0.000 description 4
- 230000003134 recirculating effect Effects 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 240000002853 Nelumbo nucifera Species 0.000 description 1
- 235000006508 Nelumbo nucifera Nutrition 0.000 description 1
- 235000006510 Nelumbo pentapetala Nutrition 0.000 description 1
- 206010048669 Terminal state Diseases 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- 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
-
- 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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Primary Health Care (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of LSTM wind-powered electricity generation load forecasting methods joined using depth Q neural network tune, and this approach includes the following steps：1) initial data for acquiring power system environment, chooses training set and forecast set；2) using LSTM as prediction model, the hyper parameter in prediction model is adjusted using DQN, specifically includes environmental parameter adjusting, state adjustment, movement selection, the intensified learning reward of regularized learning algorithm rate using the hyper parameter in DQN adjusting prediction model；3) training result is fed back to using experience recovery method and carries out parameter optimization in DQN by the prediction model after training set to be substituted into adjustment parameter, obtains optimal L STM prediction model；4) wind-powered electricity generation load prediction is carried out using optimal L STM prediction model.Compared with prior art, the present invention is not necessarily to need professional's de-regulation when different regions, is greatly improved forecasting efficiency.
Description
Technical field
The present invention relates to technical field of electric power, more particularly, to a kind of LSTM joined using depth Q neural network tune
Wind-powered electricity generation load forecasting method.
Background technique
Wind-powered electricity generation load prediction is the important component in power scheduling work, and prediction quality directly determines wind-powered electricity generation energy
No access network system.Wind-powered electricity generation load belongs to time series, as the variation of time is constantly updated.With LSTM (Long
Short Term Memory networks, shot and long term memory network) structure RNN (Recurrent Neural
Networks, Recognition with Recurrent Neural Network) it can effectively solve the problems, such as that the time gradient of RNN network disappears, and since RNN is special
Network structure make it have unique advantage to time series data.
Recognition with Recurrent Neural Network has special network structure, the i.e. input of hidden layer：In addition to the input layer at current time inputs,
There are also the inputs of the input layer of last moment, as shown in Figure 1.In Fig. 1, x, x1, x2 are respectively the input of different time nodes, o,
O1, o2 are then respectively the output of corresponding time, and it is shared in entire RNN that U, V, which are linear relationship matrix,.It will be born with wind-powered electricity generation
The relevant data of lotus are used as prediction model including time, the wind speed of wind field, realtime power, frequency, wind direction, outdoor temperature
Input passes through network query function and obtains output result o, o is then compared available error with corresponding wind-force load,
Obtain after error using gradient decline (Gradient Descent) and BPTT (Back-Propagation Through Time,
Time-based backpropagation) method is trained model, and BPTT solves gradient using backpropagation and updates network parameter
Weight.By the loop unrolling in RNN, upper one layer of neural network can communicate information to next layer, and here it is RNN to time series
The advantageous reason of the processing of data.Do not need to train the parameter of all neural networks, it is only necessary to one layer of training, it is therein
Parameter is shared parameter.
The problem of common RNN might have gradient disappearance or gradient explosion in face of long-time span, LSTM can retain mistake
Difference, for carrying out back transfer along time and layer.Error is maintained at more constant level by LSTM, allows recirculating network can be into
The study (more than 1000 time steps) of many a time steps of row, to open the channel for establishing remote causal relation.
Information is stored in the door control unit except recirculating network normal information stream by LSTM.These units can store,
Information is read in write-in, just as the data in calculator memory.Unit determines which information stored by the switch of door, with
And when allows to read, is written or removes information.But unlike the digital memory in computer, these doors are simulations
, comprising output area all the sigmoid function between 0~1 by element multiplication operate.Compared to digital storage, mould
The advantages of analog values is differentiable, therefore is suitble to backpropagation.Signal that these foundations receive and switch, and and nerve net
The node of network is similar, they can screen information with own weight sets, is decided whether according to its intensity and importing content
Information is allowed to pass through.These weights input as the weight of hidden state just as modulation, can pass through the study of recirculating network
Journey is adjusted.That is, memory unit can by conjecture, error back propagation, with gradient decline adjustment weight iteration
When procedural learning allows data to enter, leave or be deleted.Its structure is as shown in Figure 2.Three arrow tables of bottommost in Fig. 2
Show that information flows into memory unit (cell) from multiple points.Current input is not only admitted to memory unit sheet with past location mode
Body, while also entering three doors of unit, and these doors would decide how that processing input stain is door, by being from different
Number, which is multiplied to determine when to allow respectively newly to input, enters (yin), when remove current location modeAnd when allow list
First state has an impact (y to the network output that current time walksout)。ScIt is the current state of memory unit, and gyinIt is current
Input.Each Men Douke, which is opened, to close, and door each time step can regrouping switch state.Memory unit is each
Time step can decide whether to forget its state, if allow to be written, if allow to read.LSTM predict whether it is accurate with it is super
Parameter has direct relation, and therefore, suitable hyper parameter enables prediction model to reach or be very close to globe optimum.Existing skill
Art generallys use Q-Learning algorithm, and algorithm flow is：
Initialization Q (s, a),A ∈ A (s), arbitrary numerical value, and Q (terminal-state)=0；
It repeats (to each section episode)；
Init state S；
It repeats (to each step in episode)：
Using some policy, such as (ε-greedy) chooses a movement according to state S and executes；
After having executed movement, reward and new state S ' is observed；
Q(St,At)←Q(St, At)+a(Rt+1+λmaxaQ(St+1,a)-Q(St,At))
S←S′
Circulation is until terminating.
α in algorithm is learning rate, controls the difference degree being taken into account between previous Q value and the Q value newly proposed.
Q refers to corresponding Q value, and λ is then discount factor, and when discount factor is 0, prediction model can tend to active sheet and do decision,
The content for then tending to the trial that do not did before doing when it is 1 to expand Q table, it is however generally that discount factor takes between 0 to 1
A number come balance even if reward and explore.Rt+1+λmaxaQ(St+1, a) it is target Q value, Q-Learning algorithm is mainly
It is to allow Q (St, a) close to target Q value.Help to optimize wind-powered electricity generation prediction model, is allowed to be adapted to different geographical.However wind-powered electricity generation load
It is affected by territorial environment, the model parameter of different geographical has biggish difference, when prediction model is applied in different regions
When need professional's de-regulation, and the tune of prediction model ginseng quite takes manpower, more inconvenience.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide a kind of automatic tune ginseng, improve
Forecasting efficiency, and it is capable of the LSTM wind-powered electricity generation load forecasting method of adaptive different geographical joined using depth Q neural network tune.
The purpose of the present invention can be achieved through the following technical solutions：
A kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune, this approach includes the following steps：
S1：The initial data of power system environment is acquired, training set and forecast set are chosen；
S2：Using LSTM as prediction model, the hyper parameter in prediction model is adjusted using DQN；
It include that environmental parameter is adjusted, state adjusts, movement choosing using the particular content that DQN adjusts the parameter in prediction model
It selects and the intensified learning of regularized learning algorithm rate is rewarded.Environmental parameter, which is adjusted, combines LSTM prediction model and a series of movement, is formed
One Markovian decision model；The realization of the intensified learning reward of state adjustment, movement selection and regularized learning algorithm rate is based on
The Markovian decision model of formation.
Wherein, the particular content of environmental parameter adjusting is：
Adaptive learning rate is adjusted using learning rate adjustment function f (x), is adjusted and is adapted to using regular parameter adjustment function g (x)
Regular parameter, it is assumed that (p, y) is a training sample, and p is input, including learning rate xtWith regular parameter zt, y is desired defeated
Out, a is reality output, then has：
In formula, n is number of samples.
State adjustment particular content be：
Indicate that state, the feature vector of six state features include the phase using the feature vector comprising six state features
Hope dot product, the MI/ between hyper parameter, candidate iterative target value, past M step maximum target value, descent direction and the gradient of adjustment
MAX coding, function review number and alignment metric, then have：
IfFor the list of M obtained minimum target values of time t-1, state [St] coding determined by following formula：
In formula, as f (xt) be less thanMinimum value when, be encoded to 1, then take 0 in M F before, other situations take-
1；
To the adjustment [st] that does wellalignmentFor：
Descent directionExpression formula be：
In formula,ForAboutGradient,For the mean value of learning rate.
Act the particular content that selects for：
For given state, learning rate or regularization parameter are reset into initial value using after receiving iteration
Method carries out movement selection, and when Schistosomiasis control rate, there are two movements, keeps learning rate or half learning rate；For
Regularization coefficient is adjusted, other than two kinds of selections, it is allowed to increase a quarter.
The intensified learning of regularized learning algorithm rate rewards rid(f, xt) expression formula be：
In formula, flbFor the target lower bound of functional value, c is target floor value.
S3：It is excellent to be fed back to progress parameter in DQN by the prediction model after training set to be substituted into adjustment parameter for training result
Change, obtains optimal L STM prediction model；
The skill that experience replay is used to training part, when being updated each time to the parameter of neural network, from number
According to the inner training result randomly transferred before part, for updating DQN, and then optimal L STM prediction model is obtained.
S4：Wind-powered electricity generation load prediction is carried out using optimal L STM prediction model.
Compared with prior art, the present invention makes prediction model voluntarily learning regulation hyper parameter using DQN, is adapted to difference
The wind-powered electricity generation prediction model of region substantially increases forecasting efficiency without needing professional's de-regulation when different regions.
Detailed description of the invention
Fig. 1 is RNN structure chart；
Fig. 2 is LSTM structure chart；
Fig. 3 is the flow diagram of the method for the present invention；
Fig. 4 be in the embodiment of the present invention DQN learning rate be 0.05, e_greedy=0.01 when convergence effect picture；
Fig. 5 is to be declined and the prediction model accuracy rate pair using the decline of general gradient in the embodiment of the present invention using Q gradient
Than figure；
Fig. 6 is to decline to decline with the error prediction model using the decline of general gradient using Q gradient in the embodiment of the present invention
Restrain effect contrast figure.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.
Embodiment
As shown in figure 3, the present invention relates to a kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune,
The main contents of this method are：
1) initial data for acquiring power system environment, chooses training set and forecast set.
2) prediction model is obtained using the hyper parameter in DQN dynamically adapting prediction model as prediction model using LSTM
Output valve.Particular content using the hyper parameter in DQN dynamically adapting prediction model includes environmental parameter adjusting, state tune
Whole, movement selection, the intensified learning reward of regularized learning algorithm rate.
3) training set is substituted into the prediction model after adjustment parameter, training result is fed back to and carries out parameter optimization in DQN,
Obtain optimal L STM prediction model；
4) wind-powered electricity generation load prediction is carried out using optimal L STM prediction model.
Using LSTM as prediction model, gone in dynamic adaptive prediction model using a depth Q neural network (DQN)
Hyper parameter, whenever DQN makes the value that a movement takes a learning rate, this value, which can be modeled in prediction model, then can
There is an output, and reward valuation is carried out to it, at this moment DQN will be acted and is included in a Q table with corresponding reward valuation, and
Data volume is huge, therefore needs to use depth network to attempt before recording as a result, so as to so that DQN can learn from table
To the skill for adjusting hyper parameter.Environment, movement and reward therein are defined as follows：
1, environment
In formula, learning rate adjustment function f (xt), regular parameter adjustment function g (zt) make to join in different learning rate and canonical
It narrows the gap between the output and expectation of prediction model in the case where number, the x in two formulast、ztIt respectively indicates learning rate and canonical is joined
Number；(p, y) is a training sample, and n is number of samples, and p is input, including learning rate xt, regular parameter zt；Y is desired defeated
Out, a is reality output.Here adjustment function f (xt), regular parameter adjustment function g (zt) cross entropy cost function is used, when accidentally
Weight updating decision when poor big, error hour weight update slow.Using f (x) de-regulation adaptive learning rate, is adjusted and adapted to using g (x)
Regular parameter.Environment combines prediction model and a series of movement and other essential elements, forms a Markov
Decision model, that is, the selection acted only relies on the current state of user, not related with historical behavior before.State adjustment,
The completion of movement selection and the intensified learning reward of regularized learning algorithm rate is based on the Markovian decision model.
2, state
There are six the feature vectors of state feature for apparatus to indicate state.State feature is it is desirable that adjustment surpasses respectively
Dot product, MI/MAX coding, letter between parameter, candidate iterative target value, past M step maximum target value, descent direction and gradient
Number review number and alignment metric, first four feature can directly acquire, and for most latter two feature, have：
IfIt is the list for the M minimum target values that time t-1 is obtained, state [St] coding determined by following formula：
In formula, as f (xt) relatively before minimumAlso hour is encoded to 1, then takes 0 in M F before, other feelings
Condition takes -1.It allowsAs descent direction, expression formula is：
In formula,ForAboutGradient,For the mean value of learning rate.
To the adjustment [st] that does wellalignmentFor：
In addition, being in section [- 1,1] by all eigentransformations to make state feature independently of specific objective function
Feature.
3, it acts
For given state, movement is how to change the combination of learning rate and regularization parameter.In general, it learns
It practises rate and regularization parameter is very small.Therefore, learning rate or regularization parameter are reset to using after receiving iteration
The strategy of initial value.Therefore, when Schistosomiasis control rate, there are two movements：Keep learning rate or half learning rate.For
Regularization coefficient is adjusted, other than two kinds of selections, it is allowed to increase a quarter.
4, it rewards
For regularized learning algorithm rate, reward is defined as training the anti-distance for being lost to lower bound only from target.Regularized learning algorithm
The intensified learning of rate rewards rid(f, xt) be shown below：
In formula, flbFor the target lower bound of functional value, it in general will can be set as in fact zero, as the sum of loss function
Target.C is target lower bound.
5, experience replay
In the skill of training part application experience playback, when being updated each time to the parameter of neural network, just from number
According to it is inner randomly transfer one it is small quantities of before training result, help trains neural network.
One experience includes a (si, ai, ri+1, si+1, label)j, wherein i refers to that time step is i；J refers to that e_greed is
j.These tuples are stored in the memory of experience E.In addition to updating DQN, a subset S ∈ E quilt with most of nearest experience
The update DQN for small lot is pulled out from memory.
6, training result
LSTM prediction model is set 6 inputs (time, the wind speed of wind field, realtime power, frequencies, wind by the present embodiment
To, outdoor temperature), an output (load), Recognition with Recurrent Neural Network is set as 3 layers, and hidden unit is 128, activation primitive choosing
Softsign activation primitive is selected, the load data in experiment has 8932, wherein 80% data are as training data, 20%
Data are as test set.Discount factor λ is set as 0.99 when starting, explore probability and be set as 1, Uniform attenuation arrives in 100 steps
0.1.When it is 0.01 that the learning rate of DQN, which is 0.05, e_greed, as shown in the longitudinal axis of Fig. 4, the loss of DQN starts significantly
Convergence.The precision that DQN after 100 steps reaches about 40% is iterated to when training 8 is small, it is higher than reference line by 10%, and its loss is received
Hold back speed also fastly compared with reference line (gradient decline), if Fig. 5 horizontal axis is iterative steps, the longitudinal axis is prediction model accuracy rate, is compared
The rate of climb of the method for gradient decline, the accuracy rate of Q gradient decline model has a clear superiority with amplitude.Horizontal axis is in Fig. 6
Iterative steps, the longitudinal axis are the error prediction model of the decline of Q gradient with gradient decline, and it is more fast that Q gradient declines model error decline
Speed.It is limited to the limitation of computing capability, 100 steps of our iteration, but from wherein finding out that the learning ability of DQN still makes target
The accuracy of network rises quickly and the convergence of its error is also fine.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
The staff for being familiar with the art in the technical scope disclosed by the present invention, can readily occur in various equivalent modifications or replace
It changes, these modifications or substitutions should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with right
It is required that protection scope subject to.
Claims (8)
1. it is a kind of using depth Q neural network tune join LSTM wind-powered electricity generation load forecasting method, which is characterized in that this method include with
Lower step：
1) initial data for acquiring power system environment, chooses training set and forecast set；
2) using LSTM as prediction model, the hyper parameter in prediction model is adjusted using DQN；
3) training set is substituted into the prediction model after adjustment parameter, training result is fed back to and carries out parameter optimization in DQN, is obtained
Optimal L STM prediction model；
4) wind-powered electricity generation load prediction is carried out using optimal L STM prediction model.
2. a kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune according to claim 1, special
Sign is, includes environmental parameter adjusting, state tune using the particular content that DQN adjusts the parameter in prediction model in step 2)
Whole, movement selection and the intensified learning reward of regularized learning algorithm rate, environmental parameter, which is adjusted, combines LSTM prediction model and a series of
Movement forms a Markovian decision model, the reality of the intensified learning reward of state adjustment, movement selection and regularized learning algorithm rate
Now based on the Markovian decision model.
3. a kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune according to claim 2, special
Sign is that the particular content that environmental parameter is adjusted is：
Adaptive learning rate is adjusted using learning rate adjustment function f (x), is adjusted using regular parameter adjustment function g (x) and adapts to canonical
Parameter, it is assumed that (p, y) is a training sample, and p is input, including learning rate xtWith regular parameter zt, y is desired output, a
For reality output, then have：
In formula, n is number of samples.
4. a kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune according to claim 3, special
Sign is that the particular content of state adjustment is：
Indicate that state, the feature vector of six state features include that expectation is adjusted using the feature vector comprising six state features
Dot product, MI/MAX volume between whole hyper parameter, candidate iterative target value, past M step maximum target value, descent direction and gradient
Code, function review number and alignment metric, then have：
IfFor the list of M obtained minimum target values of time t-1, state [St] coding determined by following formula：
In formula, as f (xt) be less thanMinimum value when, be encoded to 1, then take 0 in M F before, other situations take -1；
To the adjustment [st] that does wellalignmentFor：
5. a kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune according to claim 4, special
Sign is, descent directionExpression formula be：
In formula,ForAboutGradient,For the mean value of learning rate.
6. a kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune according to claim 5, special
Sign is, act the particular content that selects for：
For given state, using the method that learning rate or regularization parameter are reset to initial value after receiving iteration
Movement selection is carried out, when Schistosomiasis control rate, there are two movements, keep learning rate or half learning rate；For adjustment
Regularization coefficient allows it to increase a quarter other than two kinds of selections.
7. a kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune according to claim 6, special
Sign is that the intensified learning of regularized learning algorithm rate rewards rid(f, xt) expression formula be：
In formula, flbFor the target lower bound of functional value, c is target floor value.
8. a kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune according to claim 1, special
Sign is that the particular content of step 3) is：
The skill that experience replay is used to training part, when being updated each time to the parameter of neural network, in data
The training result before part is randomly transferred, for updating DQN, and then obtains optimal L STM prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810575699.2A CN108932671A (en) | 2018-06-06 | 2018-06-06 | A kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810575699.2A CN108932671A (en) | 2018-06-06 | 2018-06-06 | A kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108932671A true CN108932671A (en) | 2018-12-04 |
Family
ID=64449976
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810575699.2A Pending CN108932671A (en) | 2018-06-06 | 2018-06-06 | A kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108932671A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109657874A (en) * | 2018-12-29 | 2019-04-19 | 安徽数升数据科技有限公司 | A kind of electric power Mid-long term load forecasting method based on long memory models in short-term |
| CN109882996A (en) * | 2019-01-25 | 2019-06-14 | 珠海格力电器股份有限公司 | A kind of method and apparatus of control |
| CN110245742A (en) * | 2019-05-08 | 2019-09-17 | 上海电力学院 | A kind of adaptive deep learning model optimization method based on Keras platform |
| CN110474339A (en) * | 2019-08-07 | 2019-11-19 | 国网福建省电力有限公司 | A kind of electric network reactive-load control method based on the prediction of depth generation load |
| CN110516889A (en) * | 2019-09-03 | 2019-11-29 | 广东电网有限责任公司 | A kind of load Comprehensive Prediction Method and relevant device based on Q-learning |
| CN110674993A (en) * | 2019-09-26 | 2020-01-10 | 广东电网有限责任公司 | User load short-term prediction method and device |
| CN110909941A (en) * | 2019-11-26 | 2020-03-24 | 广州供电局有限公司 | Power load prediction method, device and system based on LSTM neural network |
| CN111598721A (en) * | 2020-05-08 | 2020-08-28 | 天津大学 | Load real-time scheduling method based on reinforcement learning and LSTM network |
| CN111651220A (en) * | 2020-06-04 | 2020-09-11 | 上海电力大学 | Spark parameter automatic optimization method and system based on deep reinforcement learning |
| CN111768028A (en) * | 2020-06-05 | 2020-10-13 | 天津大学 | GWLF model parameter adjusting method based on deep reinforcement learning |
| CN111884213A (en) * | 2020-07-27 | 2020-11-03 | 国网北京市电力公司 | Power distribution network voltage adjusting method based on deep reinforcement learning algorithm |
| CN112288157A (en) * | 2020-10-27 | 2021-01-29 | 华能酒泉风电有限责任公司 | Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning |
| CN112308278A (en) * | 2019-08-02 | 2021-02-02 | 中移信息技术有限公司 | Method, device, equipment and medium for optimizing prediction model |
| CN112488452A (en) * | 2020-11-06 | 2021-03-12 | 中国电子科技集团公司第十八研究所 | Energy system management multi-time scale optimal decision method based on deep reinforcement learning |
| CN112614009A (en) * | 2020-12-07 | 2021-04-06 | 国网四川省电力公司电力科学研究院 | Power grid energy management method and system based on deep expected Q-learning |
| CN112712385A (en) * | 2019-10-25 | 2021-04-27 | 北京达佳互联信息技术有限公司 | Advertisement recommendation method and device, electronic equipment and storage medium |
| CN113361768A (en) * | 2021-06-04 | 2021-09-07 | 重庆科技学院 | Grain depot health condition prediction method, storage device and server |
| CN113988414A (en) * | 2021-10-27 | 2022-01-28 | 内蒙古工业大学 | Wind power output power prediction method based on P _ LSTNet and weighted Markov verification |
| CN114124460A (en) * | 2021-10-09 | 2022-03-01 | 广东技术师范大学 | Industrial control system intrusion detection method and device, computer equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106096729A (en) * | 2016-06-06 | 2016-11-09 | 天津科技大学 | A kind of towards the depth-size strategy learning method of complex task in extensive environment |
| CN106557462A (en) * | 2016-11-02 | 2017-04-05 | 数库（上海）科技有限公司 | Name entity recognition method and system |
| US20170213126A1 (en) * | 2016-01-27 | 2017-07-27 | Bonsai AI, Inc. | Artificial intelligence engine configured to work with a pedagogical programming language to train one or more trained artificial intelligence models |
| CN107241213A (en) * | 2017-04-28 | 2017-10-10 | 东南大学 | A kind of web service composition method learnt based on deeply |
| CN107370188A (en) * | 2017-09-11 | 2017-11-21 | 国网山东省电力公司莱芜供电公司 | A kind of power system Multiobjective Scheduling method of meter and wind power output |
| CN107909227A (en) * | 2017-12-20 | 2018-04-13 | 北京金风慧能技术有限公司 | Ultra-short term predicts the method, apparatus and wind power generating set of wind power |
-
2018
- 2018-06-06 CN CN201810575699.2A patent/CN108932671A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170213126A1 (en) * | 2016-01-27 | 2017-07-27 | Bonsai AI, Inc. | Artificial intelligence engine configured to work with a pedagogical programming language to train one or more trained artificial intelligence models |
| CN106096729A (en) * | 2016-06-06 | 2016-11-09 | 天津科技大学 | A kind of towards the depth-size strategy learning method of complex task in extensive environment |
| CN106557462A (en) * | 2016-11-02 | 2017-04-05 | 数库（上海）科技有限公司 | Name entity recognition method and system |
| CN107241213A (en) * | 2017-04-28 | 2017-10-10 | 东南大学 | A kind of web service composition method learnt based on deeply |
| CN107370188A (en) * | 2017-09-11 | 2017-11-21 | 国网山东省电力公司莱芜供电公司 | A kind of power system Multiobjective Scheduling method of meter and wind power output |
| CN107909227A (en) * | 2017-12-20 | 2018-04-13 | 北京金风慧能技术有限公司 | Ultra-short term predicts the method, apparatus and wind power generating set of wind power |
Non-Patent Citations (2)
| Title |
|---|
| SAMANTHA HANSEN: "Using Deep Q-Learning to Control Optimization Hyperparameters", 《HTTPS://ARXIV.ORG/ABS/1602.04062》 * |
| 赵冬斌等: "深度强化学习综述:兼论计算机围棋的发展 ", 《控制理论与应用》 * |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109657874A (en) * | 2018-12-29 | 2019-04-19 | 安徽数升数据科技有限公司 | A kind of electric power Mid-long term load forecasting method based on long memory models in short-term |
| CN109882996A (en) * | 2019-01-25 | 2019-06-14 | 珠海格力电器股份有限公司 | A kind of method and apparatus of control |
| CN110245742A (en) * | 2019-05-08 | 2019-09-17 | 上海电力学院 | A kind of adaptive deep learning model optimization method based on Keras platform |
| CN112308278A (en) * | 2019-08-02 | 2021-02-02 | 中移信息技术有限公司 | Method, device, equipment and medium for optimizing prediction model |
| CN110474339A (en) * | 2019-08-07 | 2019-11-19 | 国网福建省电力有限公司 | A kind of electric network reactive-load control method based on the prediction of depth generation load |
| CN110474339B (en) * | 2019-08-07 | 2022-06-03 | 国网福建省电力有限公司 | Power grid reactive power control method based on deep power generation load prediction |
| CN110516889A (en) * | 2019-09-03 | 2019-11-29 | 广东电网有限责任公司 | A kind of load Comprehensive Prediction Method and relevant device based on Q-learning |
| CN110516889B (en) * | 2019-09-03 | 2023-07-07 | 广东电网有限责任公司 | Load comprehensive prediction method based on Q-learning and related equipment |
| CN110674993A (en) * | 2019-09-26 | 2020-01-10 | 广东电网有限责任公司 | User load short-term prediction method and device |
| CN112712385B (en) * | 2019-10-25 | 2024-01-12 | 北京达佳互联信息技术有限公司 | Advertisement recommendation method and device, electronic equipment and storage medium |
| CN112712385A (en) * | 2019-10-25 | 2021-04-27 | 北京达佳互联信息技术有限公司 | Advertisement recommendation method and device, electronic equipment and storage medium |
| CN110909941A (en) * | 2019-11-26 | 2020-03-24 | 广州供电局有限公司 | Power load prediction method, device and system based on LSTM neural network |
| CN110909941B (en) * | 2019-11-26 | 2022-08-02 | 广东电网有限责任公司广州供电局 | Power load prediction method, device and system based on LSTM neural network |
| CN111598721A (en) * | 2020-05-08 | 2020-08-28 | 天津大学 | Load real-time scheduling method based on reinforcement learning and LSTM network |
| CN111651220B (en) * | 2020-06-04 | 2023-08-18 | 上海电力大学 | Spark parameter automatic optimization method and system based on deep reinforcement learning |
| CN111651220A (en) * | 2020-06-04 | 2020-09-11 | 上海电力大学 | Spark parameter automatic optimization method and system based on deep reinforcement learning |
| CN111768028B (en) * | 2020-06-05 | 2022-05-27 | 天津大学 | GWLF model parameter adjusting method based on deep reinforcement learning |
| CN111768028A (en) * | 2020-06-05 | 2020-10-13 | 天津大学 | GWLF model parameter adjusting method based on deep reinforcement learning |
| CN111884213A (en) * | 2020-07-27 | 2020-11-03 | 国网北京市电力公司 | Power distribution network voltage adjusting method based on deep reinforcement learning algorithm |
| CN112288157A (en) * | 2020-10-27 | 2021-01-29 | 华能酒泉风电有限责任公司 | Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning |
| CN112488452A (en) * | 2020-11-06 | 2021-03-12 | 中国电子科技集团公司第十八研究所 | Energy system management multi-time scale optimal decision method based on deep reinforcement learning |
| CN112614009A (en) * | 2020-12-07 | 2021-04-06 | 国网四川省电力公司电力科学研究院 | Power grid energy management method and system based on deep expected Q-learning |
| CN112614009B (en) * | 2020-12-07 | 2023-08-25 | 国网四川省电力公司电力科学研究院 | Power grid energy management method and system based on deep expectation Q-learning |
| CN113361768A (en) * | 2021-06-04 | 2021-09-07 | 重庆科技学院 | Grain depot health condition prediction method, storage device and server |
| CN114124460A (en) * | 2021-10-09 | 2022-03-01 | 广东技术师范大学 | Industrial control system intrusion detection method and device, computer equipment and storage medium |
| CN113988414A (en) * | 2021-10-27 | 2022-01-28 | 内蒙古工业大学 | Wind power output power prediction method based on P _ LSTNet and weighted Markov verification |
| CN113988414B (en) * | 2021-10-27 | 2024-05-28 | 内蒙古工业大学 | Wind power output power prediction method based on P_ LSTNet and weighted Markov verification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108932671A (en) | A kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune | |
| Liang et al. | A novel wind speed prediction strategy based on Bi-LSTM, MOOFADA and transfer learning for centralized control centers | |
| Li et al. | Deep learning based densely connected network for load forecasting | |
| Feng et al. | Deeptp: An end-to-end neural network for mobile cellular traffic prediction | |
| Donate et al. | Time series forecasting by evolving artificial neural networks with genetic algorithms, differential evolution and estimation of distribution algorithm | |
| Wang et al. | Real estate price forecasting based on SVM optimized by PSO | |
| Wei | A GA-weighted ANFIS model based on multiple stock market volatility causality for TAIEX forecasting | |
| CN106850289B (en) | Service combination method combining Gaussian process and reinforcement learning | |
| CN110381524B (en) | Bi-LSTM-based large scene mobile flow online prediction method, system and storage medium | |
| CN104636801A (en) | Transmission line audible noise prediction method based on BP neural network optimization | |
| Teng et al. | Use of group method of data handling for transport energy demand modeling | |
| Luo et al. | Fuzzy forecasting for long-term time series based on time-variant fuzzy information granules | |
| Wang et al. | A regional pretraining-classification-selection forecasting system for wind power point forecasting and interval forecasting | |
| CN114118375A (en) | Continuous dynamic network characterization learning method based on time sequence diagram Transformer | |
| CN115310782A (en) | Power consumer demand response potential evaluation method and device based on neural turing machine | |
| CN115907122A (en) | Regional electric vehicle charging load prediction method | |
| Yang et al. | [Retracted] Research on Students’ Adaptive Learning System Based on Deep Learning Model | |
| Hu et al. | [Retracted] Evaluation Method of Wushu Teaching Quality Based on Fuzzy Clustering | |
| Stankovic et al. | Forecasting ethereum price by tuned long short-term memory model | |
| Hachino et al. | Multiple Gaussian process models for direct time series forecasting | |
| Liang et al. | A wind speed combination forecasting method based on multifaceted feature fusion and transfer learning for centralized control center | |
| Guo et al. | Dynamic modification neural network model for short-term traffic prediction | |
| CN117175574A (en) | Power system scheduling method, device, computer equipment and storage medium | |
| Huiji | Big Data-assisted recommendation of personalized learning resources and teaching decision support | |
| Li et al. | Research on population development trend in Huizhou of China forecast based on optimal weighted combination method and fractional grey model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181204 |