CN108108842A - A kind of anti-benefit deviation neural net prediction method and device based on cost association - Google Patents
A kind of anti-benefit deviation neural net prediction method and device based on cost association Download PDFInfo
- Publication number
- CN108108842A CN108108842A CN201711408089.5A CN201711408089A CN108108842A CN 108108842 A CN108108842 A CN 108108842A CN 201711408089 A CN201711408089 A CN 201711408089A CN 108108842 A CN108108842 A CN 108108842A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- mtr
- mtd
- neural network
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000001537 neural effect Effects 0.000 title claims abstract description 18
- 238000005457 optimization Methods 0.000 claims abstract description 73
- 238000003062 neural network model Methods 0.000 claims abstract description 46
- 239000011159 matrix material Substances 0.000 claims description 56
- 239000010410 layer Substances 0.000 claims description 29
- 239000011229 interlayer Substances 0.000 claims description 22
- 230000005611 electricity Effects 0.000 claims description 14
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 11
- 230000002596 correlated effect Effects 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 8
- 230000014509 gene expression Effects 0.000 claims description 6
- 238000011068 loading method Methods 0.000 claims description 5
- 230000005284 excitation Effects 0.000 claims description 4
- 239000013598 vector Substances 0.000 claims description 4
- 235000006508 Nelumbo nucifera Nutrition 0.000 claims description 2
- 240000002853 Nelumbo nucifera Species 0.000 claims description 2
- 235000006510 Nelumbo pentapetala Nutrition 0.000 claims description 2
- 238000013459 approach Methods 0.000 claims description 2
- 238000011478 gradient descent method Methods 0.000 abstract description 2
- 230000005540 biological transmission Effects 0.000 description 4
- 230000007935 neutral effect Effects 0.000 description 4
- 238000012804 iterative process Methods 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 238000009987 spinning 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06313—Resource planning in a project environment
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Game Theory and Decision Science (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Biodiversity & Conservation Biology (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The present invention provides a kind of anti-benefit deviation neural net prediction methods and device based on cost association, are associated with cost factor in the optimization training of neural network model, and are trained using the gradient descent method having modified.The present invention has taken into account accuracy and the Optimum cost value that power system load is declared, ensure that application method and apparatus provided by the invention carry out the prediction result that electric load declares prediction and both have ideal precision, it can effectively reduce cost again, Improve Efficiency, each load, which declares policymaker and formulates its market load, in auxiliary power market declares strategy.
Description
Technical field
The invention belongs to Load Prediction In Power Systems technical field more particularly to a kind of anti-benefit based on cost association are inclined
Poor neural net prediction method and device.
Background technology
The load prediction of electric system is the important component of Power System Planning and electric system economic system fortune
Capable basis, the planning and operation to electric system are all of crucial importance.It is electric system pair reasonably to carry out load forecast
The precondition that electric power resource is scheduled and plans, can be with reasonable arrangement Unit Commitment and inspection by accurate load prediction
The plan of repairing reduces spinning reserve capacity, reduces cost of electricity-generating.With going deep into for technological progress and intelligent grid, Electric Load Forecasting
It surveys theory and technology and has very big development, but existing Predicting Technique is for the purpose of improving precision of prediction, there is no simultaneous
Care for cost factor.
The content of the invention
In order to solve the above technical problem, the present invention provides a kind of anti-benefit deviation neutral net based on cost association is pre-
Survey method, including：
S1. neural network model is established；
S2. set and trained optimization aim is optimized to neural network model；The optimization aim includes precision target
It is specific as follows shown with cost objective two parts：
Wherein, Er represents precision target, and Cos represents cost objective, and term coefficient is penalized in β expressions；
S3. training is optimized to neural network model, obtains trained neural network model；
S4. the correlated characteristic data for declaring target day are inputted into trained neural network model, you can output is declared
The load prediction results of target day.
Further, the cost objective is calculated according to the following formula：
Wherein, Y is the output matrix of neural metwork training collection, and the element of matrix o rows kth row is denoted as yok；PDA is higher level
The electricity price matrix in market, the element of matrix o rows kth row are denoted as pdaok；PRT be subordinate market electricity price matrix, matrix o rows
The element of kth row is denoted as prtok；T is the true matrix of loadings of neural metwork training collection, and the element of matrix o rows kth row is denoted as
tok；Bias vectors of the Ac between Y and true load；ε (x) represents the excitation function of neuron；δ (x) approaches for a step
Function.
Further, the optimization training process in the step S3 also needs to meet constraint requirements as follows：
Wherein, C represents that load declares the error threshold of policymaker's setting.
Further, the step S3 is specifically included：
S31. the optimization thought declined according to gradient establishes optimization training method, and calculation optimization is trained according to the following formula
Independent variable：
Wherein：WihRepresent the weight matrix between input layer and interlayer；BhRepresent the bias variable matrix in interlayer；Who
Represent the weight matrix between interlayer and output layer；BoRepresent the bias variable matrix of output layer；λ is iteration step length coefficient,
Policymaker is declared by load to provide, the value of λ is bigger, and iterative convergence speed is faster, but error is also bigger simultaneously；
S32. to the independent variable W of optimization trainingho(q)、Bo(q)、Wih(q)、Bh(q) initialized；
S33. the independent variable W of calculation optimization trainingho(q+1)、Bo(q+1)、Wih(q+1)、Bh(q+1)；
S34. the numerical value of Er and Cos is calculated, judges whether to meet terminal condition；If so, iteration ends, optimize training knot
Beam；If it is not, then making q=q+1, and go to step S33.
Further, the criterion of described " whether meeting terminal condition " specifically includes：Sentenced by the numerical value of Er and Cos
Whether disconnected error between optimum results and actual value is less than given threshold；If so, meet terminal condition；If it is not, then it is discontented with
Sufficient terminal condition.
Further, the method further includes：To prevent the iterative process of step S34 from becoming endless loop, pre-set and stop
Only time or stopping cycle-index；If the error between optimum results and actual value is more than given threshold always, iteration is followed
Ring reaches dwell time or meets terminal condition automatically after stopping cycle-index.
Further, temperature of the correlated characteristic data including the object time, humidity, weather conditions and history is negative
Lotus data.
The present invention also provides a kind of anti-benefit deviation neural network prediction device based on cost association, including：
Data input module：For input data；
Neural network model module：For establishing neural network model；
Optimization aim module：Trained optimization aim is optimized to neural network model for setting；
Optimize training module：For optimizing training to neural network model, trained neutral net mould is obtained
Type；
As a result output module：For output load prediction result.
Compared with prior art, the beneficial effects of the present invention are：
Anti- benefit deviation neural net prediction method and device provided by the invention based on cost association have taken into account electric power
The accuracy and Optimum cost value that system loading is declared are associated with cost factor in the optimization training of neural network model, and
It is trained using the gradient descent method having modified, ensures that application method and apparatus provided by the invention carry out electric load and declare
The prediction result of prediction not only has ideal precision, but also can effectively reduce cost, Improve Efficiency, in auxiliary power market
Each load, which declares policymaker and formulates its market load, declares strategy.
Description of the drawings
Fig. 1 is the structure of the neural network model of the present invention and its schematic diagram for optimizing training process；
Fig. 2 is the flow chart of the neural network model optimization training of the present invention；
Fig. 3 is the structure diagram of the anti-benefit deviation neural network prediction device based on cost association of the present invention.
Specific embodiment
Embodiment 1
A kind of anti-benefit deviation neural net prediction method based on cost association, as Fig. 1 includes：
S1. neural network model is established；
The neural network model established in the present invention can select a variety of existing neural network models, such as traditional BP nerve
Network model, multilayer perceptron (MLP), adaptive neural network model etc.；The present embodiment is only with traditional BP neural network model
Exemplified by be described in detail；
The neural network model that step S1 is established is divided into three layers of input layer, interlayer and hidden layer；Wherein, input layer by
The feature space for declaring the correlated characteristic composition of load is formed；Interlayer includes multiple neurons；Output layer is neutral net
Output, for output layer equally comprising multiple neurons, neuronal structure and interlayer are completely the same；
In the practical application of method provided in this embodiment, when declaring the correlated characteristic of load generally including at least target
Between temperature, humidity, the load data of weather conditions and history；
Wherein, the output in interlayer is as follows：
Wherein, fh[] represents the transmission function in the neuron of interlayer；Q represents the step number sequence number of iteration；Wih T(q) represent
Weight matrix in the q times iteration between input layer and interlayer；X (q) represents the input of input layer during the q times iteration；Bh(q)
Represent the bias variable matrix in interlayer during the q times iteration；S (q) represents the output in interlayer during the q times iteration；
The output of output layer is as follows：
Wherein, fo[] represents the transmission function in output layer neuron；Who T(q) represent the q times iteration in interlayer with it is defeated
Go out the weight matrix between layer；Bo(q) the bias variable matrix of output layer during the q times iteration is represented；Y (q) represents the q times iteration
When output layer output；
S2. set and trained optimization aim is optimized to neural network model；
The optimization aim of optimization training includes precision target and cost objective two parts, as follows：
Wherein, Er represents the precision target of optimization training, and Cos represents the cost objective of optimization training, and term system is penalized in β expressions
Number；
In the present embodiment, the calculating of precision target is as the accuracy computation method in traditional BP neural metwork training,
Using F- norm calculations；
The calculating of cost objective considers the electricity price of higher level's electricity market and subordinate's electricity market, while is built with virtual god
It is specific as follows shown through member：
Wherein, Y is the output matrix of neural metwork training collection, and the element of matrix o rows kth row is denoted as yok；PDA is higher level
The electricity price matrix in market, the ahead market in typical higher level market such as American Electric Power market, wherein, matrix o rows kth row
Element is denoted as pdaok；PRT is the electricity price matrix in subordinate market, the real-time city in typical subordinate market such as American Electric Power market
, wherein, the element of matrix o rows kth row is denoted as prtok；T is the true matrix of loadings of neural metwork training collection, also referred to as mesh
Matrix is marked, wherein, the element of matrix o rows kth row is denoted as tok；Bias vectors of the Ac between Y and true load；ε (x) is represented
The excitation function of virtual neuron；δ (x) is a step approximating function；
It declares the permanent positivity of load in order to limit and ensures lowest accuracy, optimization training process also needs to meet following institute
The constraint requirements shown：
Wherein, C represents that load declares the error threshold of policymaker's setting, which represents averagely to miss in optimization process
Difference is no more than C；
S3. the optimization aim determined according to step S2 optimizes neural network model training, detailed process such as Fig. 2
It is shown, it specifically includes：
S31. optimization training method is established；
The optimization thought that the optimization training process of the present embodiment declines according to gradient, will based on optimization aim and optimization constraint
It asks and training is optimized to neural network model；Compared with traditional BP neural network training process, due to increasing in optimization aim
Cost objective promotes the thought that gradient declines to need to carry out cost amendment, therefore the optimization of the present embodiment in formation algorithm
The formula expression and the optimization of result of calculation and traditional BP neural network that gradient in trained backpropagation calculates are trained not
It is identical, it is specific as follows shown：
θ=PDA-PRT δ (T-Y)
-(T-Y)·PRT·δ(T-Y)′
Wherein, Y ' represents the Jacobian matrix of Y, representing matrix dot product, × representing matrix multiplication；
Therefore, optimize trained independent variable to be calculated according to equation below：
Wherein, λ is iteration step length coefficient, and declaring policymaker by load provides, and the value of λ is bigger, and iterative convergence speed is faster,
But error is also bigger simultaneously；
S32. to the independent variable W of optimization trainingho(q)、Bo(q)、Wih(q)、Bh(q) initialized, in the present embodiment,
Who(q)、Bo(q)、Wih(q)、Bh(q) initialization value can take real number value at random；
S33. the independent variable W that the optimization training method calculation optimization determined according to step S31 is trainedho(q+1)、Bo(q+1)、
Wih(q+1)、Bh(q+1)；
S34. the numerical value of Er and Cos is calculated according to the result of step S33, judges whether to meet terminal condition；If so, repeatedly
In generation, terminates, and optimization training terminates；If it is not, then making q=q+1, and go to step S33；
In the present embodiment, if the criterion for meeting terminal condition specifically includes：Judged by the numerical value of Er and Cos
Whether the error between optimum results and actual value is less than given threshold；If so, meet terminal condition；If it is not, it is then unsatisfactory for
Terminal condition；
In the present embodiment, to prevent the iterative process of step S34 from becoming endless loop, dwell time or stopping are pre-set
Cycle-index；If the error between optimum results and actual value is more than given threshold always, when iterative cycles reach stopping
Between or stop cycle-index after i.e. meet terminal condition automatically；
S4. the correlated characteristic data for declaring target day are inputted into trained neural network model, you can output is declared
The load prediction results of target day.
Embodiment 2
A kind of anti-benefit deviation neural network prediction device based on cost association, if Fig. 3 shows, including：
Data input module：For input data；In the present embodiment, the data inputted by data input module include
The optimization inputted when optimization training starts trains the initialization value of independent variable and carries out the Shen inputted when load declares prediction
Report the correlated characteristic data of target day；
Neural network model module：For establishing neural network model；
Optimization aim module：Trained optimization aim is optimized to neural network model for setting；
Optimize training module：For optimizing training to neural network model, trained neutral net mould is obtained
Type；
As a result output module：For output load prediction result.
Power train is carried out using the anti-benefit deviation neural network prediction device provided in this embodiment based on cost association
During the load prediction of system, neural network model module first establishes neural network model；Optimization aim module is set to nerve net
Network model optimizes trained optimization aim；Then optimization training independent variable W is inputted by data input moduleho(q)、Bo
(q)、Wih(q)、Bh(q) initialization value；Optimize training module and call neural network model module, neural network model is carried out
Optimization training, until optimization training terminates to obtain trained neural network model；Shen is inputted by data input module again
The correlated characteristic data of target day are reported, substitute into trained neural network model, you can export the power train of load to be declared
The load prediction results of system.
In the present embodiment, the neural network model that neural network model module is established is traditional BP neural network, is divided into
For three layers of input layer, interlayer and hidden layer；
Wherein, the output in interlayer is as follows：
Wherein, fh[] represents the transmission function in the neuron of interlayer；Q represents the step number sequence number of iteration；Wih T(q) represent
Weight matrix in the q times iteration between input layer and interlayer；X (q) represents the input of input layer during the q times iteration；Bh(q)
Represent the bias variable matrix in interlayer during the q times iteration；S (q) represents the output in interlayer during the q times iteration；
The output of output layer is as follows：
Wherein, fo[] represents the transmission function in output layer neuron；Who T(q) represent the q times iteration in interlayer with it is defeated
Go out the weight matrix between layer；Bo(q) the bias variable matrix of output layer during the q times iteration is represented；Y (q) represents the q times iteration
When output layer output.
In the present embodiment, the optimization aim of the optimization training of optimization aim module setting includes precision target Er and cost
Target Cos two parts, it is specific as follows shown：
Wherein, Er represents the precision target of optimization training, and Cos represents the cost objective of optimization training, and term system is penalized in β expressions
Number；
In the present embodiment, the calculating of precision target uses as the accuracy computation in traditional BP neural metwork training
F- norm calculations；
The calculating of cost objective considers the electricity price of higher level's electricity market and subordinate's electricity market, while is built with virtual god
It is specific as follows shown through member：
Wherein, Y is the output matrix of neural metwork training collection, and the element of matrix o rows kth row is denoted as yok；PDA is higher level
The electricity price matrix in market, the ahead market in typical higher level market such as American Electric Power market, wherein, matrix o rows kth row
Element is denoted as pdaok；PRT is the electricity price matrix in subordinate market, the real-time city in typical subordinate market such as American Electric Power market
, wherein, the element of matrix o rows kth row is denoted as prtok；T is true matrix of loadings (the also referred to as mesh of neural metwork training collection
Mark matrix), wherein, the element of matrix o rows kth row is denoted as tok；Bias vectors of the Ac between Y and true load；ε (x) tables
Show the excitation function of virtual neuron；δ (x) is a step approximating function.
It declares the permanent positivity of load in order to limit and ensures lowest accuracy, optimization training process also needs to meet as shown below
Constraint requirements：
Wherein, C represents that load declares the error threshold of policymaker's setting, which represents averagely to miss in optimization process
Difference is no more than C.
In the present embodiment, data input module input optimization training independent variable Who(q)、Bo(q)、Wih(q)、Bh(q)
It can random value during initialization value.
In the present embodiment, optimization training module, which optimizes neural network model trained detailed process, includes：
By the optimization inputted by data input module training independent variable Who(q)、Bo(q)、Wih(q)、Bh(q) initialization
Value is substituted into following equation：
Calculate Who(q+1)、Bo(q+1)、Wih(q+1)、Bh(q+1)；And then the numerical value of Er and Cos are calculated, judge to optimize
As a result whether the error between actual value is less than given threshold；If so, optimization training terminates；If it is not, then make q=q+1,
And W is calculated againho(q+1)、Bo(q+1)、Wih(q+1)、Bh(q+1), the numerical value of Er and Cos are recalculated, is iterated, until
Error between optimum results and actual value is less than given threshold, iteration ends, and optimization training terminates.
In the present embodiment, to prevent above-mentioned iterative process from becoming endless loop, dwell time or stopping cycle being pre-set
Number；If error between optimum results and actual value is more than given threshold always, iterative cycles reach dwell time or
I.e. automatic Iterative terminates after stopping cycle-index, and optimization training terminates.
In the present embodiment, the calculation formula for optimizing training independent variable is obtained by following process：
The optimization thought that optimization training process in the present embodiment declines according to gradient, is constrained based on optimization aim and optimization
It is required that training is optimized to neural network module；Compared with traditional BP neural network training process, due to increasing in optimization aim
Cost objective is added, the thought that gradient declines has been promoted to need to carry out cost amendment in formation algorithm, therefore the present embodiment is excellent
Change the formula expression of the gradient calculating in trained backpropagation and the optimization of result of calculation and traditional BP neural network is trained simultaneously
It differs, it is specific as follows shown：
θ=PDA-PRT δ (T-Y)
-(T-Y)·PRT·δ(T-Y)′
Wherein, Y ' represents the Jacobian matrix of Y, representing matrix dot product, × representing matrix multiplication；
Thus, the result of calculation of the independent variable of optimization training is as follows：
Finally it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention and it is unrestricted, although reference
The present invention is described in detail in preferred embodiment, it will be understood by those of ordinary skill in the art that, it can be to the present invention's
Technical solution is modified or replaced equivalently, and without departing from the spirit and scope of technical solution of the present invention, should all be covered
Among scope of the presently claimed invention.
Claims (8)
1. a kind of anti-benefit deviation neural net prediction method based on cost association, which is characterized in that the described method includes：
S1. neural network model is established；
S2. set and trained optimization aim is optimized to neural network model；The optimization aim include precision target and into
This target two parts, it is specific as follows shown：
<mrow>
<mi>min</mi>
<mo>&DoubleRightArrow;</mo>
<mi>G</mi>
<mo>=</mo>
<mi>E</mi>
<mi>r</mi>
<mo>+</mo>
<mi>&beta;</mi>
<mo>&CenterDot;</mo>
<mi>C</mi>
<mi>o</mi>
<mi>s</mi>
</mrow>
Wherein, Er represents precision target, and Cos represents cost objective, and term coefficient is penalized in β expressions；
S3. training is optimized to neural network model, obtains trained neural network model；
S4. the correlated characteristic data for declaring target day are inputted into trained neural network model, you can target is declared in output
The load prediction results of day.
2. according to the method described in claim 1, it is characterized in that, the cost objective is calculated according to the following formula：
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>C</mi>
<mi>o</mi>
<mi>s</mi>
<mo>=</mo>
<mi>s</mi>
<mi>u</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>Y</mi>
<mo>&CenterDot;</mo>
<mi>P</mi>
<mi>D</mi>
<mi>A</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>s</mi>
<mi>u</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>A</mi>
<mi>c</mi>
<mo>&CenterDot;</mo>
<mi>P</mi>
<mi>R</mi>
<mi>T</mi>
<mo>&CenterDot;</mo>
<mi>&epsiv;</mi>
<mo>(</mo>
<mrow>
<mi>A</mi>
<mi>c</mi>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>s</mi>
<mi>u</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>Y</mi>
<mo>&CenterDot;</mo>
<mi>P</mi>
<mi>D</mi>
<mi>A</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>o</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>O</mi>
</munderover>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>K</mi>
</munderover>
<msub>
<mi>y</mi>
<mrow>
<mi>o</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<msub>
<mi>pda</mi>
<mrow>
<mi>o</mi>
<mi>k</mi>
</mrow>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>s</mi>
<mi>u</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>A</mi>
<mi>c</mi>
<mo>&CenterDot;</mo>
<mi>P</mi>
<mi>R</mi>
<mi>T</mi>
<mo>&CenterDot;</mo>
<mi>&delta;</mi>
<mo>(</mo>
<mrow>
<mi>A</mi>
<mi>c</mi>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>o</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>O</mi>
</munderover>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>K</mi>
</munderover>
<mo>&lsqb;</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>t</mi>
<mrow>
<mi>o</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>o</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<msub>
<mi>prt</mi>
<mrow>
<mi>o</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<mi>&delta;</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>t</mi>
<mrow>
<mi>o</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>o</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>tanh</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<msup>
<mi>e</mi>
<mi>x</mi>
</msup>
<mo>-</mo>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mi>x</mi>
</mrow>
</msup>
</mrow>
<mrow>
<msup>
<mi>e</mi>
<mi>x</mi>
</msup>
<mo>+</mo>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mi>x</mi>
</mrow>
</msup>
</mrow>
</mfrac>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>&epsiv;</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>&ap;</mo>
<mi>&delta;</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<mi>tanh</mi>
<mrow>
<mo>(</mo>
<mi>&alpha;</mi>
<mo>&CenterDot;</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mn>1</mn>
</mrow>
<mn>2</mn>
</mfrac>
<mo>,</mo>
<mi>&alpha;</mi>
<mo>></mo>
<mn>1</mn>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
Wherein, Y is the output matrix of neural metwork training collection, and the element of matrix o rows kth row is denoted as yok；PDA is higher level market
Electricity price matrix, matrix o rows kth row element be denoted as pdaok；PRT be subordinate market electricity price matrix, matrix o row kth
The element of row is denoted as prtok；T is the true matrix of loadings of neural metwork training collection, and the element of matrix o rows kth row is denoted as tok；
Bias vectors of the Ac between Y and true load；ε (x) represents the excitation function of neuron；δ (x) approaches letter for a step
Number.
3. according to the method described in claim 2, it is characterized in that, the optimization training process in the step S3 also needs to meet
Constraint requirements as follows：
<mrow>
<mi>C</mi>
<mi>o</mi>
<mi>n</mi>
<mi>s</mi>
<mi>t</mi>
<mo>:</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>Y</mi>
<mo>></mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mover>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<mi>A</mi>
<mi>c</mi>
</mrow>
<mi>T</mi>
</mfrac>
<mo>)</mo>
</mrow>
<mo>&OverBar;</mo>
</mover>
<mo><</mo>
<mo><</mo>
<mi>C</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
Wherein, C represents that load declares the error threshold of policymaker's setting.
4. according to the method described in claim 3, it is characterized in that, the step S3 is specifically included：
S31. the optimization thought that foundation gradient declines establishes optimization training method, and what calculation optimization was trained according to the following formula becomes certainly
Amount：
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>h</mi>
<mi>o</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>q</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>W</mi>
<mrow>
<mi>h</mi>
<mi>o</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&lambda;</mi>
<mo>&CenterDot;</mo>
<mo>&dtri;</mo>
<msub>
<mi>G</mi>
<msub>
<mi>W</mi>
<mrow>
<mi>h</mi>
<mi>o</mi>
</mrow>
</msub>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>B</mi>
<mi>o</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>q</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>B</mi>
<mi>o</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&lambda;</mi>
<mo>&CenterDot;</mo>
<mo>&dtri;</mo>
<msub>
<mi>G</mi>
<msub>
<mi>B</mi>
<mi>o</mi>
</msub>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>h</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>q</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>h</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&lambda;</mi>
<mo>&CenterDot;</mo>
<mo>&dtri;</mo>
<msub>
<mi>G</mi>
<msub>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>h</mi>
</mrow>
</msub>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>B</mi>
<mi>h</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>q</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>B</mi>
<mi>h</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&lambda;</mi>
<mo>&CenterDot;</mo>
<mo>&dtri;</mo>
<msub>
<mi>G</mi>
<msub>
<mi>B</mi>
<mi>h</mi>
</msub>
</msub>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
Wherein：WihRepresent the weight matrix between input layer and interlayer；BhRepresent the bias variable matrix in interlayer；WhoIt represents
Weight matrix between interlayer and output layer；BoRepresent the bias variable matrix of output layer；λ is iteration step length coefficient, by bearing
Lotus declares policymaker and provides, and the value of λ is bigger, and iterative convergence speed is faster, but error is also bigger simultaneously；
S32. to the independent variable W of optimization trainingho(q)、Bo(q)、Wih(q)、Bh(q) initialized；
S33. the independent variable W of calculation optimization trainingho(q+1)、Bo(q+1)、Wih(q+1)、Bh(q+1)；
S34. the numerical value of Er and Cos is calculated, judges whether to meet terminal condition；If so, iteration ends, optimization training terminates；
If it is not, then making q=q+1, and go to step S33.
5. according to the method described in claim 4, it is characterized in that, the criterion of described " whether meeting terminal condition " is specific
Including：Judge whether the error between optimum results and actual value is less than given threshold by the numerical value of Er and Cos；If so,
Meet terminal condition；If it is not, then it is unsatisfactory for terminal condition.
6. according to the method described in claim 5, it is characterized in that, the method further includes：To prevent the iteration mistake of step S34
Cheng Chengwei endless loops pre-set dwell time or stop cycle-index；If the error one between optimum results and actual value
Directly it is more than given threshold, then iterative cycles reach dwell time or meet terminal condition automatically after stopping cycle-index.
7. according to the method described in claim 1, it is characterized in that, the temperature of correlated characteristic data including object time,
The load data of humidity, weather conditions and history.
8. a kind of anti-benefit deviation neural network prediction device based on cost association, which is characterized in that described device includes：
Data input module：For input data；
Neural network model module：For establishing neural network model；
Optimization aim module：Trained optimization aim is optimized to neural network model for setting；
Optimize training module：For optimizing training to neural network model, trained neural network model is obtained；
As a result output module：For output load prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711408089.5A CN108108842B (en) | 2017-12-22 | 2017-12-22 | Cost association-based benefit deviation resistant neural network prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711408089.5A CN108108842B (en) | 2017-12-22 | 2017-12-22 | Cost association-based benefit deviation resistant neural network prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108108842A true CN108108842A (en) | 2018-06-01 |
| CN108108842B CN108108842B (en) | 2022-04-05 |
Family
ID=62212491
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711408089.5A Active CN108108842B (en) | 2017-12-22 | 2017-12-22 | Cost association-based benefit deviation resistant neural network prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108108842B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070185823A1 (en) * | 2005-07-28 | 2007-08-09 | Dingguo Chen | Load prediction based on-line and off-line training of neural networks |
| CN103295081A (en) * | 2013-07-02 | 2013-09-11 | 上海电机学院 | Electrical power system load prediction method based on back propagation (BP) neural network |
| CN104484727A (en) * | 2015-01-12 | 2015-04-01 | 江南大学 | Short-term load prediction method based on interconnected fuzzy neural network and vortex search |
| CN107370170A (en) * | 2017-06-23 | 2017-11-21 | 浙江大学 | A kind of energy storage system capacity collocation method for considering capacity price of electricity and load prediction error |
-
2017
- 2017-12-22 CN CN201711408089.5A patent/CN108108842B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070185823A1 (en) * | 2005-07-28 | 2007-08-09 | Dingguo Chen | Load prediction based on-line and off-line training of neural networks |
| CN103295081A (en) * | 2013-07-02 | 2013-09-11 | 上海电机学院 | Electrical power system load prediction method based on back propagation (BP) neural network |
| CN104484727A (en) * | 2015-01-12 | 2015-04-01 | 江南大学 | Short-term load prediction method based on interconnected fuzzy neural network and vortex search |
| CN107370170A (en) * | 2017-06-23 | 2017-11-21 | 浙江大学 | A kind of energy storage system capacity collocation method for considering capacity price of electricity and load prediction error |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108108842B (en) | 2022-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Stochastic combined heat and power dispatch based on multi-objective particle swarm optimization | |
| CN103942461B (en) | Water quality parameter Forecasting Methodology based on online online-sequential extreme learning machine | |
| CN110458443A (en) | A kind of wisdom home energy management method and system based on deeply study | |
| CN107067121A (en) | A kind of improvement grey wolf optimized algorithm based on multiple target | |
| CN104636823B (en) | A kind of wind power forecasting method | |
| CN107578124A (en) | The Short-Term Load Forecasting Method of GRU neutral nets is improved based on multilayer | |
| CN105678407A (en) | Daily electricity consumption prediction method based on artificial neural network | |
| CN106815782A (en) | A kind of real estate estimation method and system based on neutral net statistical models | |
| CN107067190A (en) | The micro-capacitance sensor power trade method learnt based on deeply | |
| CN104636801A (en) | Transmission line audible noise prediction method based on BP neural network optimization | |
| CN103593719B (en) | A kind of rolling power-economizing method based on slab Yu contract Optimized Matching | |
| CN105631517A (en) | Photovoltaic power generation power short term prediction method based on mind evolution Elman neural network | |
| CN106203683A (en) | A kind of modeling method of power customer electro-load forecast system | |
| CN110287509A (en) | Flexibility analysis and the fault of construction diagnosis of municipal heating systems and localization method and system | |
| CN106779253A (en) | The term load forecasting for distribution and device of a kind of meter and photovoltaic | |
| CN115940294B (en) | Multi-stage power grid real-time scheduling strategy adjustment method, system, equipment and storage medium | |
| CN107590570A (en) | A kind of bearing power Forecasting Methodology and system | |
| Shang et al. | Production scheduling optimization method based on hybrid particle swarm optimization algorithm | |
| CN107145968A (en) | Photovoltaic apparatus life cycle cost Forecasting Methodology and system based on BP neural network | |
| CN106355980A (en) | Power grid regulation capability predication method based on limited memory extreme learning machine | |
| Kutschenreiter-Praszkiewicz | Application of artificial neural network for determination of standard time in machining | |
| CN108108842A (en) | A kind of anti-benefit deviation neural net prediction method and device based on cost association | |
| CN103489037B (en) | A kind of Forecasting Methodology that can power generating wind resource | |
| CN104850918A (en) | Node load prediction method taking power grid topology constraints into consideration | |
| Wu et al. | Applications of AI techniques to generation planning and investment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |