CN110334879A - Power grid bus reactive load forecasting method and device - Google Patents
Power grid bus reactive load forecasting method and device Download PDFInfo
- Publication number
- CN110334879A CN110334879A CN201910625771.2A CN201910625771A CN110334879A CN 110334879 A CN110334879 A CN 110334879A CN 201910625771 A CN201910625771 A CN 201910625771A CN 110334879 A CN110334879 A CN 110334879A
- Authority
- CN
- China
- Prior art keywords
- burden
- load
- history
- power
- data sequence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of 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
- G06Q10/04—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
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
-
- 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
-
- 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
- Y04S50/00—Market activities related to the operation of systems integrating technologies related to power network operation or related to communication or information technologies
- Y04S50/16—Energy services, e.g. dispersed generation or demand or load or energy savings aggregation
Abstract
The embodiment of the present invention provides a kind of power grid bus reactive load forecasting method and device, belongs to network load electric powder prediction.It include: to obtain history burden with power data sequence and history load or burden without work data sequence；History burden with power data sequence and history load or burden without work data sequence are input in active reactive load integration prediction model, the predicted value of burden with power and load or burden without work in the following preset time period is exported.Since bus load or burden without work radix is small, non-linear strong, data inherent law excavate difficulty greatly to it is difficult to predict, and temporal aspect is significant between the active and idle sequence data of load, to be based on active reactive load integration prediction model, active reactive load can be effectively predicted, and then can realize energy-saving consumption-reducing and scheduling fine-grained management.
Description
Technical field
The present invention relates to network load electric powder prediction more particularly to a kind of power grid bus reactive load forecasting method and
Device.
Background technique
In power industry, the short-term forecast of electric load is sent out in fields such as operation of power networks, Unit Commitment and power schedulings
Increasingly important role is waved, the economical and reliability of electric system can be effectively improved by improving load prediction precision.
With the development of Process of Urbanization Construction, the power demand of user is further increased, and accessing renewable distributed generation resource on a large scale will have
There is feasibility condition.
Short-term load forecasting is the important component of user terminal microgrid energy management system, is that micro- source optimization is dispatched
Basis.Prediction result will directly affect micro-capacitance sensor operation reserve and electricity transaction.Correlative study shows that micro-grid load prediction misses
The larger operation cost that will lead to of difference is significantly increased.Load prediction prediction technique at present, mainly statistical method, mainly there is recurrence
Analytic approach, time series method etc..However, statistical method is all linear model approach, it can only and small rule few to data influence factor
Apperance is originally handled, and is just seemed helpless when encountering complexity, nonlinear problem, cannot effectively be predicted.
Summary of the invention
To solve the above-mentioned problems, the embodiment of the present invention provides one kind and overcomes the above problem or at least be partially solved
State the power grid bus reactive load forecasting method and device of problem.
According to a first aspect of the embodiments of the present invention, a kind of power grid bus reactive load forecasting method is provided, comprising:
Obtain history burden with power data sequence and history load or burden without work data sequence；
History burden with power data sequence and history load or burden without work data sequence are input to active reactive load integration
In prediction model, the predicted value of burden with power and load or burden without work in the following preset time period is exported.
According to a second aspect of the embodiments of the present invention, a kind of power grid bus reactive load forecasting device is provided, comprising:
Module is obtained, for obtaining history burden with power data sequence and history load or burden without work data sequence；
First output module, for history burden with power data sequence and history load or burden without work data sequence to be input to
In function load or burden without work integration prediction model, the predicted value of burden with power and load or burden without work in the following preset time period is exported.
According to a third aspect of the embodiments of the present invention, a kind of electronic equipment is provided, comprising:
At least one processor；And
At least one processor being connect with processor communication, in which:
Memory is stored with the program instruction that can be executed by processor, and the instruction of processor caller is able to carry out first party
Power grid bus reactive load forecasting method provided by any possible implementation in the various possible implementations in face.
According to the fourth aspect of the invention, a kind of non-transient computer readable storage medium, non-transient computer are provided
Readable storage medium storing program for executing stores computer instruction, and computer instruction makes the various possible implementations of computer execution first aspect
In power grid bus reactive load forecasting method provided by any possible implementation.
Power grid bus reactive load forecasting method and device provided in an embodiment of the present invention, due to bus load or burden without work radix
Small, non-linear strong, data inherent law excavates difficulty greatly to which it is difficult to predict and between the active and idle sequence data of load
Temporal aspect is significant, to be based on active reactive load integration prediction model, active reactive load can be effectively predicted, in turn
Energy-saving consumption-reducing and scheduling fine-grained management can be achieved.
It should be understood that above general description and following detailed description be it is exemplary and explanatory, can not
Limit the embodiment of the present invention.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is a kind of flow diagram of power grid bus reactive load forecasting method provided in an embodiment of the present invention；
Fig. 2 is a kind of RNN neural network structure schematic diagram provided in an embodiment of the present invention；
Fig. 3 is a kind of memory unit structure figure of LSTM network provided in an embodiment of the present invention；
Fig. 4 is a kind of LSTM network internal topological diagram provided in an embodiment of the present invention；
Fig. 5 is a kind of LSTM network B PTT training algorithm schematic diagram provided in an embodiment of the present invention；
Fig. 6 is that a kind of network number of plies provided in an embodiment of the present invention influences schematic diagram to precision of prediction；
Fig. 7 is a kind of structural schematic diagram of DI-LSTM network model provided in an embodiment of the present invention；
Fig. 8 is a kind of work flow diagram of DI-LSTM network model provided in an embodiment of the present invention；
Fig. 9 is the active prediction result schematic diagram in a kind of 110kV electric power village provided in an embodiment of the present invention and Jinjiang station；
Figure 10 is the active prediction result schematic diagram in a kind of 110kV electric power village provided in an embodiment of the present invention and Jinjiang station；
Figure 11 is the idle prediction result schematic diagram in a kind of 110kV electric power village provided in an embodiment of the present invention and Jinjiang station；
Figure 12 is the idle prediction result schematic diagram in a kind of 110kV electric power village provided in an embodiment of the present invention and Jinjiang station；
Figure 13 is that error map is continuously predicted in a kind of bus burden with power provided in an embodiment of the present invention；
Figure 14 is that error map is continuously predicted in a kind of bus burden with power provided in an embodiment of the present invention；
Figure 15 is that a kind of bus load or burden without work provided in an embodiment of the present invention continuously predicts error map；
Figure 16 is that a kind of bus load or burden without work provided in an embodiment of the present invention continuously predicts error map；
Figure 17 is a kind of structural schematic diagram of power grid bus reactive load forecasting device provided in an embodiment of the present invention；
Figure 18 is the block diagram of a kind of electronic equipment provided in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
It, can only be to data influence factor less and at small-scale sample since statistical method is all linear model approach
Reason just seems helpless when encountering complexity, nonlinear problem, and artificial intelligence approach can effectively solve non-linear ask
Topic.Mainly there are support vector machines, neural network, deep learning method etc. in artificial intelligence approach.Based on above description, the present invention
Embodiment provides a kind of power grid bus reactive load forecasting method.Referring to Fig. 1, this method comprises:
101, history burden with power data sequence and history load or burden without work data sequence are obtained；
102, history burden with power data sequence and history load or burden without work data sequence are input to active reactive load one
In body prediction model, the predicted value of burden with power and load or burden without work in the following preset time period is exported.
Wherein, short-term load forecasting is the important component of user terminal microgrid energy management system, is micro- source optimization
The basis of scheduling.Prediction result will directly affect micro-capacitance sensor operation reserve and electricity transaction.Correlative study shows micro-grid load
The prediction larger operation cost that will lead to of error is significantly increased.
Method provided in an embodiment of the present invention, since bus load or burden without work radix is small, non-linear strong, data inherent law is dug
Difficulty is dug greatly to which it is difficult to predict and temporal aspect is significant between the active and idle sequence data of load, thus based on active
Active reactive load can be effectively predicted in load or burden without work integration prediction model, and then can realize that energy-saving consumption-reducing and scheduling are fine
Change management.
Content based on the above embodiment, as a kind of alternative embodiment, active reactive load integration prediction model is
It is obtained based on LSTM training, LSTM is a kind of improvement RNN network model.
RNN is the deep learning neural network that a kind of pair of sequence data is modeled, and can theoretically be utilized arbitrarily long
Sequence information is carried out identical operation to element each in data, and each operates the calculated result before being dependent on.
Typical RNN network structure is as shown in Fig. 2, its structure can refer to following formula:
ht=σh(U·xt+W·ht-1+bh)；
yt=σo(V·ht+bo)；
In above-mentioned formula, xtIndicate the input of t moment network, U is the weight matrix between input layer and hidden layer, and V is
Weight matrix between hidden layer and output layer, W are to indicate hidden layer feedback weight matrix, σhWith bhRespectively the activation primitive of hidden layer with
Bias vector, htIndicate the output of moment hidden layer.σ0With b0The respectively activation primitive and bias vector of output layer, ytWhen indicating t
Carve the output of network.
The connection type of RNN hidden layer node makes its output not only related with current time input, and also and last moment
Hidden layer output valve is related, and network has recursive nature in time, but when time depth is excessive, can generate in model training
Gradient explosion or gradient disappear, and can not solve the problems, such as long-distance dependence of the data on time dimension.
And LSTM has the time recursive attribute and sequence data processing advantage of RNN, while having unique memory unit,
It can be flexibly adapted to the temporal aspect of e-learning task.Compared with traditional RNN, LSTM is by introducing control unit door machine
The internal cell structure for improving neural network hidden layer is made, the LSTM neuron with unique forgetting, memory function is formed.Such as Fig. 3
Shown, LSTM increases input gate in neuron, forgets three door, out gate units, and so-called door is exactly using mind
The operation being multiplied through network and a step-by-step, when the full Connection Neural Network layer as activation primitive can export between one 0 to 1
Numerical value, description currently input that how many information passes through door.Therefore, neural net layer output is 1, i.e., when door is opened, entirely
Portion's information can pass through, and when neural net layer output is 0, i.e., when door is closed, any information cannot all pass through.LSTM is exactly logical
The processing mode for controlling formation sequence data long-term memory or forgetting to three gate cells is crossed, is realized to sequence data over long distances
The study of dependency characteristic, solve RNN on time depth in network training gradient disappear and gradient explosion issues.
LSTM network internal detailed topologies are as shown in figure 4, wherein xtIt is the input of t moment network, htIndicate network
Hidden layer status information, ctThe status information for indicating memory unit respectively indicates and forgets door, input gate and out gate.ytIt is t moment
The output of network.It can be seen that each moment is input to the data x of LSTM networktThree control doors will be entered simultaneously, while each
Men Douhui receives the status information h of hidden layer last momentt-1.The specific work process of LSTM neuron is expressed as follows:
(1) Forgetting Mechanism: by forgeing door σfThe information that should be forgotten in processing neuron, controls status information before t moment
ct-1Input ratio.
(2) memory mechanism: determine that Current neural member needs the information of memory storage, input gate σiNetwork is received currently to input
xtIt determines more new information, candidate information is generated by tanh activation primitive, update to obtain current time memory in conjunction with Forgetting Mechanism
The state value c of unitt。
(3) network exports: the state value c of memory unit is updated by forgetting, memory mechanismtAfterwards, out gate σ0According to newest
State value ct, last moment output ht-1With current time network inputs xtCodetermine the output h of LSTM network hidden layert。
Wherein, in LSTM network operation process each variable calculation formula are as follows:
ft=σf(Wf[xt,ht-1]+bf)；
it=σi(Wi[xt,ht-1]+bi)；
ct=ft*ct-1+it*tanh(Wc[x,ht-1]+bc)；
ot=σo(Wo[xt,ht-1]+bo)；
ht=ot*tanh(ct)。
In above-mentioned formula, Wf,Wi,Wc,WoRespectively forgeing door, input gate, memory unit, with out gate to connect network defeated
Enter xtWith hidden layer status information ht-1Weight matrix, bf,bi,bc,boIt is the bias vector of various pieces respectively, is depth net
Learning training parameter in network, σf,σi,σoRespectively indicate the activation primitive for forgeing door, input gate and out gate.
In addition, the training algorithm of LSTM network mainly uses time-based back-propagation algorithm (Back
Propagation Trough Time, BPTT).BPTT can be understood as the BP algorithm of spacer step expansion on time, as shown in figure 5, will
LSTM network is unfolded according to time step, forms the LSTM memory unit of multilayer on time dimension, uses BP to the network after expansion
Algorithm is trained, and the accumulative residual error of last moment is constantly passed back to initial point, is formed to LSTM under entire time dimension
The training of network, this process is in the application that mathematics is substantially to chain type Rule for derivation.
Content based on the above embodiment is also based on burden with power data and idle as a kind of alternative embodiment
Load data excavates depth temporal characteristics therein to reflect non-linear relation, to move to Load Time Series data
State modeling, to obtain history burden with power data sequence and history load or burden without work data sequence.
Specifically, the radix of electrical system bus load is small and characteristic is different, vulnerable to the influence of service area's intra domain user,
Load or burden without work is influenced by voltage level, Line Flow and nonlinear compensation simultaneously, has stronger randomness, load or burden without work
Do not have the clearly flow direction as burden with power, it is big that data variational regularity excavates difficulty, and load is active and idle
Sequence data temporal characteristics are significant.For this purpose, the two to be used as to input, the output data of prediction model simultaneously, formation has bus
Function, load or burden without work integration prediction model, can not only Accurate Prediction burden with power, and can predict well it is non-linear it is strong,
The load or burden without work of variational regularity difference.
The power factor of bus load is not of uniform size, and load or burden without work numerically differs larger with burden with power, it is contemplated that
Activation in DI-LSTM network model (active reactive load integration prediction model is established based on DI-LSTM network model)
The sensitivity that network is fitted data nonlinear change can be improved in the input/output bound of function in [- 1,1], accelerates DI-
LSTM net training time.For this purpose, data normalization can be carried out to burden with power data sequence and load or burden without work data sequence
Processing.The embodiment of the present invention does not make specific restriction to the process of normalized, including but not limited to: being based on history burden with power
Maximum value and history burden with power minimum value, history burden with power data sequence is normalized, be based on history
The maximum value of load or burden without work and the minimum value of history load or burden without work, are normalized history load or burden without work data sequence.
Wherein, for history burden with power data sequence to be normalized, the process of normalized can refer to following public affairs
Formula:
In above-mentioned formula, x is the burden with power data in history burden with power data sequence, and min (x) indicates that history has
The minimum value of workload, max (x) indicate the maximum value of history burden with power, and x' indicates normalization result.Similarly, to history without
The process that workload data sequence is normalized can also refer to above-mentioned formula, and details are not described herein again.
In addition, different characteristics is presented on festivals or holidays and working day in bus load, thus increase in input variable working day/
This data of festivals or holidays type, to improve the accuracy that network predicts bus load.Day, categorical variable was after LSTM network
Output be regarded as festivals or holidays to work daily load correction amount.Since nonnumeric day type input data directly uses boolean
Variable indicates that is, 1 indicates working day, and 0 indicates festivals or holidays, is not necessarily to normalized.
Content based on the above embodiment, as a kind of alternative embodiment, active reactive load integration prediction model is 5
Layer structure.Specifically, it is based on DI-LSTM network model, active reactive load integration prediction model can be established.Determining model
After input layer and output layer variable tie up active and reactive load datas for 2, need to determine DI-LSTM network model by testing
Structure and other hyper parameters, i.e. the hidden layer number, data entry time step number of network, learning rate, optimizer etc..
The number of hidden layer is the direct indicator of reaction network depth in deep learning network, with the promotion of hidden layer number,
The nonlinear fitting ability of model is promoted therewith, but the complexity of model and the time cost of training also will increase simultaneously.It is deep
Layer LSTM network is multiple by LSTM structure replication at each moment, and the LSTM parameter in each layer is consistent, but different layers
In parameter can be different.For the ability to express for enhancing DI-LSTM network model, it is 1 layer, 2 layers, 3 layers, 4 that LSTM, which is set separately,
Layer and 5 layers, as shown in Figure 6, it can be seen that with the increase of the number of plies, net training time rises at double, prediction error decline after
Rise again, this is because not being able to satisfy prediction application requirement as over-fitting occurs in the increase of the network training number of plies.Base
In the content of above-described embodiment, as a kind of alternative embodiment, the stacking of 2 layers of LSTM and 1 layer of DNN is used in the embodiment of the present invention
Structure.
By above-mentioned analysis, the DI-LSTM network model that the embodiment of the present invention is established (predict by active reactive load integration
Model is established based on DI-LSTM network model) it is of five storeys altogether, wherein hidden layer is made of 2 layers of LSTM and 1 layer of DNN, network instruction
Practice optimizer and use Adam algorithm, as shown in Figure 7.According to the active and reactive load of bus in 15min accuracy prediction 1 day, LSTM
In the time step of each sample be 96, the input time step number of network is 96 points before prediction time.
Content based on the above embodiment can also be calculated as a kind of alternative embodiment by time reversal error propagation
Method completes network parameter training, the active and reactive load integration prediction of bus is formed, in terms of idle bus load precision of prediction
It is assessed.
The long short reaction of data entry time the step number periodicity of historical data time series and the completeness of knowledge, should
Determining for parameter needs while considering the difficulty of model training and the complete degree of historical data knowledge.Longer input time step
Number brings sufficient prediction history knowledge, but the training difficulty of model has been significantly greatly increased, and shorter input time step number is conducive to mould
Type training, but the missing of historical knowledge will limit the promotion of model prediction accuracy.Therefore, integrated forecasting precision and network performance,
The embodiment of the present invention select 96 points of t moment as historical data input time step number.
Learning rate refer in optimization algorithm update network weight amplitude size, learning rate can be set as preset parameter or
According to training dynamic adjust, the embodiment of the present invention set the initial learning rate of DI-LSTM network as 0.025 and pass through optimization algorithm into
Row adjust automatically.The characteristics of Adam optimization algorithm combination two kinds of optimization algorithms of AdaGrad and RMSProp, to the first moment of gradient
Estimation and second order moments estimation, which comprehensively consider, calculates update step-length, with optimization computational efficiency is high, EMS memory occupation is small, is suitble to big advise
The advantages that modulus evidence and parameter training scene, the embodiment of the present invention completes depth DI-LSTM network mould using Adam optimizer
The training of type.
The mode that the embodiment of the present invention does not obtain active reactive load integration prediction model to training specifically limits, and wraps
It includes but is not limited to: sample history burden with power data and sample history load or burden without work data are input in initial model, export
The predicted value of burden with power and load or burden without work in known time section；Based on back-propagation algorithm, according to active in known time section
Error between load and the predicted value and actual value of load or burden without work is updated the weight of layer each in initial model, obtains
Active reactive load integration prediction model.Specifically, DI-LSTM bus reactive load forecasting process can be as shown in figure 8, specific
It is expressed as follows:
(1) historical data prepares, and to the processing of network inputs data normalization, divides training set and test set；
(2) DI-LSTM network model, which is undergone training, concentrates input data, carries out data and is fitted to obtain predicted value, according to pre-
Measured value and true value calculate error；
(3) it is updated based on weight of the back-propagation algorithm to each layer in DI-LSTM network model；
(4) after setting train epochs, step 2 and step 3 are constantly repeated, is calculated using Adam optimization algorithm and updates step
It is long, until meeting loss function minimum；
(5) after the completion of DI-LSTM network training, input test collection data, active to the bus of every 15min in 1 day future,
Load or burden without work is predicted；
(6) prediction result error calculation is analyzed, model performance assessment.
In order to verify the actual effect of providing method of the embodiment of the present invention, data used in the embodiment of the present invention are somewhere
The actual measurement SCADA data in power grid on June 1st, 2011 to August 22nd, the effective bus load of metric data totally 357 in Harvest time
Item, the data of acquisition include burden with power and load or burden without work data.Training set is using June 1 to August data on the 5th, test set
Using August 6th to August data on the 22nd, data sampling frequency is set as 15min.It is analyzed according to prosthomere, bus load or burden without work
Inputoutput data is as shown in table 1.
In order to which preferably analysis and assessment prediction effect, the embodiment of the present invention use mean absolute error (MAPE, Mean
Absolute Percentage Error) and root-mean-square error (RMSE, Root Mean Square Error) as evaluation refer to
Mark.MAPE not only considers that the error of predicted value and true value also contemplates the ratio between error and true value, and intuitive reflection is pre-
Survey the accuracy of result.RMSE is the arithmetic square root of mean square error, sensitive to abnormal point in prediction result, reflects the steady of model
It is qualitative, shown in the following formula of calculation method:
In above-mentioned formula, n indicates number of samples, yrIt (t) is the load true value of t moment, ypIt (t) is then the negative of t moment
Lotus predicted value.Wherein, the composition of DI-LSTM network model inputoutput data can refer to such as the following table 1:
Table 1
In order to measure the accuracy and feasibility that propose model, using identical data sample, ARMA, single input are used respectively
LSTM network and the DI-LSTM network of proposition carry out the active and reactive load prediction of bus.In comparative experiments, ARMA uses base
Rank model is determined automatically in BIC criterion, and the hidden layer number of single input LSTM network (hereinafter referred to as LSTM) is 2 layers, hidden layer nerve
First number is 97, and training optimization algorithm is Adam algorithm.After the completion of the training of each fallout predictor, tested respectively in different bus loads
Test model on collection, will wherein typical 2 class bus load burden with power prediction result it is as shown in FIG. 9 and 10, load or burden without work
Prediction result is as shown in FIG. 11 and 12.Error condition is as shown in table 2 and table 3.Wherein, table 2 is three kinds of prediction models to difference
Bus burden with power predicts that error result, table 3 are three kinds of prediction models to different bus reactive load forecasting error results.
Table 2
Table 3
Analyzed by table 2 it is found that ARMA, LSTM and DI-LSTM can Accurate Prediction bus burden with power, DI-LSTM is pre-
Surveying the MAPE that model predicts the 1 bus burden with power of 110kV electric power village station is 1.301%, RMSE 0.379, to the Jinjiang 110kV
The MAPE that 2 female burdens with power are predicted that stands is 1.118%, RMSE 0.194.
It is analyzed by table 3 it is found that ARMA prediction model reaches the MAPE of the female reactive load forecasting in the Jinjiang 110kV station 2
240.049%, this is because the bus load or burden without work is changed significantly under the prediction day, ARMA can not based on the mode of linear fit
The situation of change of prediction data, prediction result do not have value；Compared with ARMA, LSTM prediction is female to 110kV electric power village station 1 idle
Load prediction MAPE is 15.594%, 2 mother reactive load forecasting MAPE of the Jinjiang 110kV station is that 32.026%, LSTM can be promoted
The precision of prediction of load or burden without work, but it is still larger for the reactive load forecasting error of changing rule complexity, it is not able to satisfy reality
Application demand；The active reactive load integration prediction model that the embodiment of the present invention proposes is female to 110kV electric power village station 1 idle negative
Lotus prediction MAPE is reduced to the MAPE of the female reactive load forecasting in the 8.131%, Jinjiang 110kV station 2 to be reduced to 10.148%, and bus has
Workload prediction error MAPE, which averagely reduces by 1.109%, RMSE, averagely reduces by 0.197, bus reactive load forecasting error MAPE
Averagely 62.379%, RMSE of reduction, which averagely reduces by 0.386, DI-LSTM, can be obviously improved reactive load forecasting precision, and right
The load or burden without work of complicated changing rule also has preferable precision of prediction.It follows that DI-LSTM has compared to ARMA, LSTM
Apparent bus reactive load forecasting advantage.
Figure 13 and Figure 14 shows LSTM prediction model and DI-LSTM prediction model be carried out continuously 9 days bus it is active negative
Lotus predicts error MAPE and RMSE distribution situation, and Figure 15 and Figure 16 show that LSTM prediction model and DI-LSTM prediction model connect
Continuous bus reactive load forecasting error MAPE and the RMSE distribution situation for carrying out 9 days.To simplify explanation for 110kV electric power village station 1
Matrix is shown as A-wire, and 2 matrix of the Jinjiang 110kV station is shown as second line.
As can be seen that LSTM prediction model and DI-LSTM prediction model predict accurately bus burden with power in from the graph
Stablize, LSTM can predict load or burden without work, but very big to error under non-linear prediction day strong, load variations rule is big, propose
DI-LSTM prediction model there is apparent idle prediction advantage, wherein DI-LSTM is to female (A-wire) nothing in 110kV electric power village station 1
MAPE of the workload prediction in the 8th prediction day is 5.81%, remaining prediction is below 5% day, to the female (second in the Jinjiang 110kV station 2
Line) reactive load forecasting the 1st prediction day MAPE be 13.28%, remaining prediction is below 10% day.
The above analysis, LSTM neural network have stronger nonlinear fitting performance, can satisfy general bus
Reactive load forecasting, but reactive load forecasting precision more complicated for changing rule is inadequate；What is proposed is refreshing based on DI-LSTM
Prediction model through network, can all kinds of bus load or burden without work of Accurate Prediction, and realize bus burden with power and idle negative
Lotus integration prediction, the prediction result of measured data show this method can to idle changing rule complexity bus load standard
Really prediction, with good application prospect.
Content based on the above embodiment, the embodiment of the invention provides a kind of power grid bus reactive load forecasting device,
The power grid bus reactive load forecasting device is for executing the power grid bus reactive load forecasting provided in above method embodiment
Method.Referring to Figure 17, which includes:
Module 1701 is obtained, for obtaining history burden with power data sequence and history load or burden without work data sequence；
First output module 1702, for inputting history burden with power data sequence and history load or burden without work data sequence
Into active reactive load integration prediction model, burden with power and the prediction of load or burden without work in the following preset time period are exported
Value.
As a kind of alternative embodiment, the device further include:
Normalized module is right for the minimum value of maximum value and history burden with power based on history burden with power
History burden with power data sequence is normalized, and maximum value and history load or burden without work based on history load or burden without work are most
History load or burden without work data sequence is normalized in small value.
As a kind of alternative embodiment, normalized module, for being normalized by following formula, formula
It is as follows:
Wherein, x is the burden with power data in history burden with power data sequence, and min (x) indicates history burden with power
Minimum value, max (x) indicate the maximum value of history burden with power, and x' indicates normalization result.
As a kind of alternative embodiment, active reactive load integration prediction model is 5 layers of structure.
It include hidden layer, hidden layer in 5 layers of structure of active reactive load integration prediction model as a kind of alternative embodiment
It is as composed by 2 layers of shot and long term memory network and 1 layer depth neural network.
As a kind of alternative embodiment, the device further include:
Second output module, for being input to sample history burden with power data and sample history load or burden without work data just
In beginning model, the predicted value of burden with power and load or burden without work in known time section is exported；
Update module, for being based on back-propagation algorithm, according to the pre- of burden with power in known time section and load or burden without work
Error between measured value and actual value is updated the weight of layer each in initial model, obtains active reactive load integration
Prediction model.
Device provided in an embodiment of the present invention, since bus load or burden without work radix is small, non-linear strong, data inherent law is dug
Difficulty is dug greatly to which it is difficult to predict and temporal aspect is significant between the active and idle sequence data of load, thus based on active
Active reactive load can be effectively predicted in load or burden without work integration prediction model, and then can realize that energy-saving consumption-reducing and scheduling are fine
Change management.
Figure 18 illustrates the entity structure schematic diagram of a kind of electronic equipment, and as shown in figure 18, which may include:
Processor (processor) 1810, communication interface (Communications Interface) 1820, memory (memory)
1830 and communication bus 1840, wherein processor 1810, communication interface 1820, memory 1830 are complete by communication bus 1840
At mutual communication.Processor 1810 can call the logical order in memory 1830, and to execute following method: acquisition is gone through
History burden with power data sequence and history load or burden without work data sequence；By history burden with power data sequence and history load or burden without work
Data sequence is input in active reactive load integration prediction model, exports in the following preset time period burden with power and idle
The predicted value of load.
In addition, the logical order in above-mentioned memory 1830 can be realized by way of SFU software functional unit and conduct
Independent product when selling or using, can store in a computer readable storage medium.Based on this understanding, originally
Substantially the part of the part that contributes to existing technology or the technical solution can be in other words for the technical solution of invention
The form of software product embodies, which is stored in a storage medium, including some instructions to
So that a computer equipment (can be personal computer, electronic equipment or the network equipment etc.) executes each reality of the present invention
Apply all or part of the steps of a method.And storage medium above-mentioned include: USB flash disk, mobile hard disk, read-only memory (ROM,
Read-Only Memory), random access memory (RAM, Random AccessMemory), magnetic or disk etc. it is various
It can store the medium of program code.
The embodiment of the present invention also provides a kind of non-transient computer readable storage medium, is stored thereon with computer program,
The computer program is implemented to carry out the various embodiments described above offer method when being executed by processor, for example, obtain history
Burden with power data sequence and history load or burden without work data sequence；By history burden with power data sequence and history load or burden without work number
According to sequence inputting into active reactive load integration prediction model, burden with power and idle negative is exported in the following preset time period
The predicted value of lotus.
The apparatus embodiments described above are merely exemplary, wherein unit can be as illustrated by the separation member
Or may not be and be physically separated, component shown as a unit may or may not be physical unit, i.e.,
It can be located in one place, or may be distributed over multiple network units.It can select according to the actual needs therein
Some or all of the modules achieves the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying creative labor
In the case where dynamic, it can understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be embodied in the form of software products in other words, should
Computer software product may be stored in a computer readable storage medium, such as ROM/RAM, magnetic disk, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Method described in certain parts of example or embodiment.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features；
And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (9)
1. a kind of power grid bus reactive load forecasting method characterized by comprising
Obtain history burden with power data sequence and history load or burden without work data sequence；
The history burden with power data sequence and the history load or burden without work data sequence are input to active reactive load one
In body prediction model, the predicted value of burden with power and load or burden without work in the following preset time period is exported.
2. power grid bus reactive load forecasting method according to claim 1, which is characterized in that described to have the history
Workload data sequence and the history load or burden without work data sequence are input to it in active reactive load integration prediction model
Before, further includes:
The minimum value of maximum value and history burden with power based on history burden with power, to the history burden with power data sequence
It is normalized, the minimum value of maximum value and history load or burden without work based on history load or burden without work is idle to the history
Load data sequence is normalized.
3. power grid bus reactive load forecasting method according to claim 2, which is characterized in that described to be based on burden with power
Maximum value and burden with power minimum value, the history burden with power data sequence is normalized, comprising:
Wherein, x is the burden with power data in the history burden with power data sequence, and min (x) indicates history burden with power
Minimum value, max (x) indicate the maximum value of history burden with power, and x' indicates normalization result.
4. power grid bus reactive load forecasting method according to claim 1, which is characterized in that the active reactive load
Integrated prediction model is 5 layers of structure.
5. power grid bus reactive load forecasting method according to claim 4, which is characterized in that the active reactive load
It include hidden layer in 5 layers of structure of integrated prediction model, the hidden layer is by 2 layers of shot and long term memory network and 1 layer depth nerve
Composed by network.
6. power grid bus reactive load forecasting method according to claim 1, which is characterized in that described to have the history
Workload data sequence and the history load or burden without work data sequence are input to it in active reactive load integration prediction model
Before, further includes:
Sample history burden with power data and sample history load or burden without work data are input in initial model, known time is exported
The predicted value of burden with power and load or burden without work in section；
Based on back-propagation algorithm, according to the predicted value of burden with power and load or burden without work in the known time section and actual value it
Between error, the weight of each layer in the initial model is updated, active reactive load integration prediction mould is obtained
Type.
7. a kind of power grid bus reactive load forecasting device characterized by comprising
Module is obtained, for obtaining history burden with power data sequence and history load or burden without work data sequence；
First output module, for inputting the history burden with power data sequence and the history load or burden without work data sequence
Into active reactive load integration prediction model, burden with power and the prediction of load or burden without work in the following preset time period are exported
Value.
8. a kind of electronic equipment characterized by comprising
At least one processor；And
At least one processor being connect with the processor communication, in which:
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program to instruct energy
Enough methods executed as described in claim 1 to 6 is any.
9. a kind of non-transient computer readable storage medium, which is characterized in that the non-transient computer readable storage medium is deposited
Computer instruction is stored up, the computer instruction makes the computer execute the method as described in claim 1 to 6 is any.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910625771.2A CN110334879A (en) | 2019-07-11 | 2019-07-11 | Power grid bus reactive load forecasting method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910625771.2A CN110334879A (en) | 2019-07-11 | 2019-07-11 | Power grid bus reactive load forecasting method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110334879A true CN110334879A (en) | 2019-10-15 |
Family
ID=68146503
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910625771.2A Pending CN110334879A (en) | 2019-07-11 | 2019-07-11 | Power grid bus reactive load forecasting method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110334879A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826774A (en) * | 2019-10-18 | 2020-02-21 | 广州供电局有限公司 | Bus load prediction method and device, computer equipment and storage medium |
| CN111061708A (en) * | 2019-11-11 | 2020-04-24 | 国电南瑞科技股份有限公司 | Electric energy prediction and restoration method based on LSTM neural network |
| CN111104298A (en) * | 2019-11-22 | 2020-05-05 | 国网黑龙江省电力有限公司信息通信公司 | LSTM-based power grid server running state prediction device |
| CN111355246A (en) * | 2020-03-04 | 2020-06-30 | 国网安徽省电力有限公司检修分公司 | Prediction method, system and storage medium for reactive compensation of phase modulator |
| CN111680818A (en) * | 2020-04-27 | 2020-09-18 | 中国电力科学研究院有限公司 | Short-term reactive load prediction method and system |
| CN111697560A (en) * | 2020-04-24 | 2020-09-22 | 国家电网有限公司 | Method and system for predicting load of power system based on LSTM |
| CN112134304A (en) * | 2020-09-22 | 2020-12-25 | 南方电网数字电网研究院有限公司 | Micro-grid full-automatic navigation method, system and device based on deep learning |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108304976A (en) * | 2018-03-06 | 2018-07-20 | 西安交通大学 | A kind of electric system fining load prediction and analysis method |
| CN109245099A (en) * | 2018-10-29 | 2019-01-18 | 南方电网科学研究院有限责任公司 | A kind of electric load discrimination method, device, equipment and readable storage medium storing program for executing |
| CN109376960A (en) * | 2018-12-06 | 2019-02-22 | 国网北京市电力公司 | Load Forecasting based on LSTM neural network |
| CN109861305A (en) * | 2019-01-31 | 2019-06-07 | 东南大学 | A kind of transmission & distribution collaboration economic load dispatching method of binding model PREDICTIVE CONTROL |
| US20190187682A1 (en) * | 2016-05-09 | 2019-06-20 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for data collection in production line with future status prediction and load balancing |
-
2019
- 2019-07-11 CN CN201910625771.2A patent/CN110334879A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190187682A1 (en) * | 2016-05-09 | 2019-06-20 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for data collection in production line with future status prediction and load balancing |
| CN108304976A (en) * | 2018-03-06 | 2018-07-20 | 西安交通大学 | A kind of electric system fining load prediction and analysis method |
| CN109245099A (en) * | 2018-10-29 | 2019-01-18 | 南方电网科学研究院有限责任公司 | A kind of electric load discrimination method, device, equipment and readable storage medium storing program for executing |
| CN109376960A (en) * | 2018-12-06 | 2019-02-22 | 国网北京市电力公司 | Load Forecasting based on LSTM neural network |
| CN109861305A (en) * | 2019-01-31 | 2019-06-07 | 东南大学 | A kind of transmission & distribution collaboration economic load dispatching method of binding model PREDICTIVE CONTROL |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826774A (en) * | 2019-10-18 | 2020-02-21 | 广州供电局有限公司 | Bus load prediction method and device, computer equipment and storage medium |
| CN110826774B (en) * | 2019-10-18 | 2022-03-22 | 广东电网有限责任公司广州供电局 | Bus load prediction method and device, computer equipment and storage medium |
| CN111061708A (en) * | 2019-11-11 | 2020-04-24 | 国电南瑞科技股份有限公司 | Electric energy prediction and restoration method based on LSTM neural network |
| CN111104298A (en) * | 2019-11-22 | 2020-05-05 | 国网黑龙江省电力有限公司信息通信公司 | LSTM-based power grid server running state prediction device |
| CN111355246A (en) * | 2020-03-04 | 2020-06-30 | 国网安徽省电力有限公司检修分公司 | Prediction method, system and storage medium for reactive compensation of phase modulator |
| CN111355246B (en) * | 2020-03-04 | 2023-08-18 | 国网安徽省电力有限公司超高压分公司 | Prediction method, system and storage medium for reactive compensation of camera |
| CN111697560A (en) * | 2020-04-24 | 2020-09-22 | 国家电网有限公司 | Method and system for predicting load of power system based on LSTM |
| CN111680818A (en) * | 2020-04-27 | 2020-09-18 | 中国电力科学研究院有限公司 | Short-term reactive load prediction method and system |
| CN111680818B (en) * | 2020-04-27 | 2022-11-01 | 中国电力科学研究院有限公司 | Short-term reactive load prediction method and system |
| CN112134304A (en) * | 2020-09-22 | 2020-12-25 | 南方电网数字电网研究院有限公司 | Micro-grid full-automatic navigation method, system and device based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110334879A (en) | Power grid bus reactive load forecasting method and device | |
| Zeng et al. | Multifactor-influenced energy consumption forecasting using enhanced back-propagation neural network | |
| US20190265768A1 (en) | Method, system and storage medium for predicting power load probability density based on deep learning | |
| Zhou et al. | Multi-objective artificial bee colony algorithm for short-term scheduling of hydrothermal system | |
| CN109002942A (en) | A kind of short-term load forecasting method based on stochastic neural net | |
| CN110943983B (en) | Network security prevention method based on security situation awareness and risk assessment | |
| Huang et al. | Resilient distribution networks by microgrid formation using deep reinforcement learning | |
| WO2021098384A1 (en) | Data abnormality detection method and apparatus | |
| CN108491982A (en) | A kind of short-term load forecasting method and system based on echo state network | |
| KR20200128232A (en) | Apparatus for predicting electricity demand and method thereof | |
| Wang et al. | A novel decomposition-ensemble forecasting system for dynamic dispatching of smart grid with sub-model selection and intelligent optimization | |
| CN116264388A (en) | Short-term load prediction method based on GRU-LightGBM model fusion and Bayesian optimization | |
| CN105005575A (en) | Quick developing interface method for enterprise intelligent prediction | |
| Wang et al. | Cloud computing and extreme learning machine for a distributed energy consumption forecasting in equipment-manufacturing enterprises | |
| CN111738483A (en) | Power grid loss reduction optimization method and system based on clustering and deep belief network | |
| Ming-Tao et al. | The research on stock price forecast model based on data mining of BP neural networks | |
| Kim et al. | Projection-aware deep neural network for dc optimal power flow without constraint violations | |
| Reddy et al. | An optimal neural network model for software effort estimation | |
| CN115629576A (en) | Non-invasive flexible load aggregation characteristic identification and optimization method, device and equipment | |
| Daraghmi et al. | Accurate and time‐efficient negative binomial linear model for electric load forecasting in IoE | |
| CN112288140A (en) | Keras-based short-term power load prediction method, storage medium and equipment | |
| Hapuarachchi et al. | Long term annual electricity demand forecasting in Sri Lanka by artificial neural networks | |
| CN111079996A (en) | Virtual seat system based on BP neural network power load prediction and control method thereof | |
| Deng et al. | Medium-term rolling load forecasting based on seasonal decomposition and long short-term memory neural network | |
| Sheng et al. | Research on Application of Neural Network Based on Cloud Computing in Power Load Forecasting |
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: 20191015 |